Method for distributed monitoring and adaptive control of multilevel system and apparatus for realising said method

FIELD: information technology.

SUBSTANCE: method for distributed monitoring and adaptive control of a multilevel system is based on the adaptive control system principle. During normal operation, a distributed structure is viewed by a wide on-line field having low-resolution which is sufficient for detecting a local disturbance. Further, the on-line field is narrowed in the vicinity of the disturbance which is viewed in more detail and through fine analysis, the nature of what is happening in is determined (recognition) in a specific decision block. The apparatus, which is in form of a decision unit for determining the state of the multilevel system, operates in three modes: current monitoring, monitoring error estimation and training.

EFFECT: reduced probability of blocking the system by reducing the transmitted volume of control information through redistribution thereof on control levels.

6 cl, 9 dwg

Technical solutions combined to form a single inventive concept relate to electrical engineering, namely to control and manage multi-tier distributed systems, and can be used, for example, when designing a global management systems geographically distributed networks.

Known methods of control and management of complex technical systems, described in the work of Levin BYR, Schwartz C. Probabilistic models and methods in communication and control systems. - M.: Radio and communication, 1985. - S-269. So is described in this paper, the method of stochastic control is provided by the following sequence of actions:

- given the known characteristics of the system and the probabilistic characteristics of the interference are the control algorithm, providing the extremum of the selected quality criterion;

- assess the state of the system;

- identify the state of the system;

- decide on the control action.

The disadvantage of this method is that its implementation requires information about the correlation functions of the error signal, which makes it almost impossible to use this method for managing a distributed system, since in this case such information is not available.

Also there is a method adapting the telecommunication management network (see Budko P.A., V.V. Fedorenko Management in communication networks. Mathematical models and optimization methods. - Moscow: Publishing house of physical and mathematical literature, 2003. - S - 209., RES; Budko P.A. Fomin L.A., gakhova NN. Informational aspects of the internal organization of telecommunication systems. // Biomedical technology and electronics. 2003, №6, p.10-19). The sequence of actions during the implementation of this method consists in the following:

- find the extremum in the field of space adaptation of some indicators of the quality of functioning of the network, such as the load factor of the channels according to the criterion of minimum-time delivery of information;

- identify (recognize) the parameters of the network status and sources of disturbances;

- change the structure and parameters with the aim of bringing the network to the optimal state, which leads to the conflation of the roles of management and training.

In this way the process of adaptive management is achieved through a compromise satisfy two conflicting requirements: achievement of performance without sufficient state information and the current network properties; excessive accumulation of information, resulting in the delay of received management actions.

The disadvantage of this analog for distributed systems is the occurrence of so-called Blo is irewoc combination, since increasing degradation of the system, the number of control information increases dramatically. This system state is the most dangerous even in networks with dedicated channels for the transmission of service information. Need to solve the problem of reduction of the volume management information.

The closest to the technical nature of the claimed method (prototype), is a method of control and management of intelligent network (see Steklov VK, Berkman L. Estimation of the volume management information in the information networks. // Telecommunications. - 2000. No. 6. - P.34-36), consisting of the following sequence of actions:

- pre-set thresholds on the monitored parameters for each node in the system;

measure generic measure of quality in the form of a vector of the state variables at each of the N nodes in the systemwhere z₁(t); z₂(t);...z_n(t) - values of measured parameters over time (t), n=1, 2, ..., N;

- compare the measured parameters with the set threshold values;

- assess the state of the system by comparing the results;

- form a control action on the system evaluation.

The disadvantage of the closest analogue is a high probability of blocking network, due to the presence of a large amount of control information, pascalc is in the process of developing a control action is constantly used all available measurement information, in a distributed system (global control loop) parencymal channels of communication and control of the transit flows of information. This is because in the process of functioning of the network, all measuring tools are in an active state (under load). In addition, no system control circuit associated with exposure to the source of the disturbances, which can also lead to permanent system locks (the locks are combined).

A device adaptive communication system that implements the above method (Levin BYR, Schwartz C. Probabilistic models and methods in communication and control systems. - M.: Radio and communication, 1985. - S, Fig.7.1.), which consists of a system of information transmission source of the disturbance and interference, device identification, recording devices and controls.

The disadvantage of this device is the relatively low reliability of the control, due to the fact that the method does not take into account errors of the first and second kind ("the risk of the customer" and "consumer risk").

Also if there is only one control circuit for a distributed system requires the measurement and processing of a huge number of parameters and, consequently, increase control time and volume of transmitted control information.

Also known device auto is adicheskogo control object (Fomin L.A., Chernoskutov A.I. Optimization errors with a two-stage procedure control /Automatic control and computer engineering. - 1975. No. 3. - P.34-37), containing the registration unit, two elements, And the unit of comparison, the first output of which is connected to the first input of the first element And the second output with the first input of the second element And whose output is connected to the first input of the recording unit, the unit summation, the output of which is connected to the first input of the comparison, units conversion, each connected with its output to one input of summation block and the inputs to the corresponding information input device, the sensor of random numbers, the first output of which is connected to the second input of the first element And the second output - the second input of the second element And the control unit, the output of which is connected to the sensor input random number and the second input of the recording unit.

The disadvantage of this device is that it is in the control state of the object assigning tolerances for each parameter leads to large classification errors and, moreover, does not account for the overlap of the distributions of parameter values for the healthy and faulty objects, when accurate classification is impossible. Also for control of complex engineering systems and of revealing the fact of their refusal must be made, u is the the conversion and processing of a large number of parameters that are often associated with the shutdown and its non-working.

Closest to the claimed device multi-level distributed control system is a device decision-making (see Fomin L.A., Budko P.A. Efficiency and quality of ICT systems. Optimization methods. - M.: Fizmatlit, 2008. - P.146-157, RES)operating in accordance with the count recognition of the health system that implements the condition for nding the optimal thresholds classification that provides the minimum error identification system state, while, in comparison with the above described device, it introduced additional conversion unit, the two units forming the threshold values, the second block of comparison, the third element And the element OR. The sensor of random numbers is replaced by a generator of artificial traffic. Element OR is connected with its inputs to the outputs of the first and third elements, And output to the first inputs of the blocks forming the threshold values and to the third input of the recording unit, a second input connected to the first output of the first unit of comparison and the first input of the third element And. the Second output unit of comparison is connected to the input blocks of transform and third inputs of the first and second elements And. the First input p is pout Comparer is connected to the output of the additional conversion unit, the inputs of which are connected with the corresponding outputs of the generator of artificial traffic and entrance systems, second input units of comparison is connected to the outputs of the respective blocks the formation of the threshold values, the second inputs of which are connected with the output of the control unit, the third input from the output of the second element And the second input of the third element And is connected to the first generator output artificial traffic.

The disadvantage of the prototype is that the shaping unit thresholds purpose tolerances on the system parameters is carried out without taking into account its technical condition, the load of the communication channels and the buffer device of the switching nodes.

The technical result achieved by using the inventive method and device is to reduce the blocking probability of the system by reducing the transmitted volume management information due to its redistribution levels (contours) of the control.

In the claimed technical result is achieved by the fact that in the known method for distributed monitoring and adaptive management telecommunication system under the influence of random perturbations, based on the fact that the operation of the system in terms of exposure to destabilizing factors pre-set threshold value to strairway parameters of each of the nodes, forming the system, measured at each node of the system, its parameters, compare the values of measured parameters with threshold, by comparing the results to assess the condition of the system and when the deviation of the system state on a valid form control action on the system. However≥3-tier system consisting of N(k) of nodes in the k-th level, where k=1, 2, ... K, and n(k)-th node, where n(k)=1, 2..., N(k), characterized by M(k, n) parameters, the system state control on To the levels for which the threshold values of the parameters z^nop(k, n, m), where m=1, 2, ..., M(k, n) is pre-set for each of the nodes To level, and then measure the values of z(k, n, m) parameters of the nodes of each level and compare them with the preset corresponding threshold values z^then(k, n, m), and assessment of the state of the system is performed in stages To stages, the first stage threshold and compare the measured values of the parameters of the system nodes belonging to the first level of control, and if their values are within the threshold, then the system is considered to be workable, otherwise generate a control action for the serial assessment of system nodes belonging to k=2, 3, ..., levels of management, which compare the measured and the threshold values of the parameters of the system nodes of the k-th level, if the measured pairs of the points of the nodes in the system the k-th level of control do not extend beyond the threshold, the state of the nodes of the k-th level considered healthy, otherwise record the failure of the nodes of the k-th level and allocate its parameters do not satisfy the threshold, in line with what produce the control action for the recovery of parameters of the nodes of the k-th level of control, and threshold values of the parameters z^nop(k, n, m) set on the basis of a priori information about the current state of the system at all levels of management, forming a feedback loop system.

With the introduction of multi-stage principle of state control of a multilevel system can significantly reduce the exchange of information circulating, since the first stage is mainly used local information about the state of the subsystem. The amount of control information increases with the introduction of new levels of control.

The control circuit associated with exposure to the source of the disturbance, based on the analysis of interference necessary to prevent the internal lock, and the extent of this impact is determined in the second solution of the inverse problem - identification.

In the inventive device the technical result is achieved by the fact that in the known device distributed control and adaptive control of multi-level system that provides a method of distributed control and hell is the efficient management of multi-level system, contains the unit of measurement of the generalized parameter, the input bus "a generalized index of which is connected to an output bus "a generalized index of" multi-level system (ICC), the inputs "interference" which is connected to the outputs of the source of disturbances, the ICC is equipped with an information input and output and control output unit registration and control connected to the control inputs of the ICC and the source of the disturbance. Moreover, it introduced additional unit of decision-making and To units of measure, where K≥3, the total number of levels of control and management of the ICC. The input bus of the system parameters of the k-th level of the k-th unit of measurement, where k=1, 2, ..., K, is connected to an output bus "system settings" appropriate level of control and management of the ICC, the output of the measuring unit of the generalized indicator connected to the input "a generalized index" block decision, N(k)×M(n, k) outputs the "parameters" of the k-th unit of measurement, where N(k) is the number of nodes belonging to the k-th level of the ICC, and M(n,k) is the number of parameters of the n-th node belonging to k-th level of the ICC, is connected to the appropriate inputs settings" block decision, provided additional input bus "a priori", outputs "training" and "normal" block decision-making connected to the same inputs of the block registration and management, K+1 inputs "error" under which is turned off to the respective outputs of the error block decision-making.

The unit of decision-making consists of a detector, resolver and identifier. The input bus "a priori" and the input "a generalized index" of the detector are the same input bus and entry generalized metric" unit, direct access and exit "training" of the detector are respectively the outputs of the "norm" and "training" block, N(k×M)(n, k) inputs settings" k-ID is the corresponding group of inputs settings" block, direct output and output status detector connected respectively to the second and first inputs of the resolver, in addition, the exit status the detector is connected to the first digits of the three-digit dual To the group of inputs resolver, entrance "signs" of the k-th identifier, starting with k=2, is connected with the inverted yield k-1-th identifier, and the second digits of a three-digit dual k-th group of inputs of the Recognizer, and the entrance "signs" of the first identifier (k=1) is also connected to an inverted output of the detector, while the direct output of the k-th identifier is connected with the third digit of the first built-in input of the k-th group of inputs of the Recognizer, and the inverted output of the k-th ID - the third digit of the second built-in input of the k-th group of inputs resolver, the first and subsequent outputs are outputs "oshi the SC block.

The detector unit of decision-making consists of a transducer characteristic "a generalized index", item comparison, shaper threshold, the RS flip-flop generator artificial traffic and control. The input bus "a priori", the input "a generalized index of" in and out "training" are the same bus, input and output unit decision-making, and direct its output is "normal", "status" connected with artificial traffic generator and transducer characteristic "a generalized index", the first input connected to the input "a generalized index", this Converter sign of its other output connected to the first input of the comparison element, the second input is connected to the first output of the shaper threshold, connected to the input bus "a priori condition system, the second output of the shaper threshold is the first output of the detector, and the third is his second release, he is connected with the control, the R-input of the trigger and is the output of "training" block decision, also the S-input of the trigger is connected to the comparison element and its direct and inverted outputs are the same outputs of the detector, control its output connected to the generator artificial trafi the A.

IDs block decision consist of M transducers signs "settings", adder element of comparison and RS-flip-flop. K-th group of M input parameters is the same group of inputs of the k-th identifier, and each m-th input, m=1, 2,..., M, of this group and the output identifier "signs" are connected respectively with the first and second inputs of the m-th transducer signs "settings", the outputs are transmitted to the corresponding inputs of the adder, the output of which in turn is connected to the first input of the comparison element, a second input connected to the first input and output of the ID, the output of the comparison element is a S-input trigger, R-input of which is connected to the second input and output identifier, and direct and inverted outputs of the trigger are the same outputs ID, the ID of the top level control (k=K) second input of the comparison element is connected only to the first input and the R input of flip - flop-only with the second input of the identifier.

The Recognizer unit of decision-making consists of the first element And dual groups of elements And To elements OR. While the first and second inputs of the resolver connected to respective inputs of the first element And whose output is the first output "error", first, second, and Tr is the part of the bits of the three-digit dual To the group of inputs resolver connected respectively with the first, the second and third inputs To the dual groups of elements And the outputs are pairwise connected to both inputs To elements OR outputs which are To output error of the resolver.

Since the implementation of the method is carried out using the statistical theory of pattern recognition and statistical theory of decision making, the quality criterion is the total classification error (the sum of the errors of the first and second kind), which occur on the first and subsequent stages of control. This fact is used to build the block a decision, in which when forming the thresholds used a priori information about the state of the system in normal operation and in case of anomalies. These conditions are simulated in the formation thresholds x₀, y₀, ..., y₀unit dimensions using artificial traffic generator in the learning process.

Technical solutions are illustrated by the drawings on which is shown:

figure 1 - structural diagram of the device of the distributed control and adaptive control of multi-level system;

figure 2 - functional block circuit decision;

figure 3 - schematic diagram of the detector;

figure 4 - schematic diagram of the IDs of the lower levels of management;

in Fig. - schematic diagram of the ID of the top level management;

figure 6 - schematic diagram of the resolver;

figure 7 - results of the density distribution characteristics in accordance with the criterion of Neyman-Pearson (for simplicity - law Rayleigh distribution);

on Fig - graph recognition health of multilevel system that implements a phased principle of decision-making;

figure 9 - graphs of the reduction of the volume of information through the use of multi-stage control procedure.

The implementation of the inventive method is explained as follows. Multi-tiered system consisting of N(k) of nodes in the k-th level, in terms of exposure to destabilizing factors, controlling for the levels where To≥3, k=1, 2, ...,K. In this case, each n(k)-th node, where n(k)=1, 2, ..., N(k), characterized by M(k, n) is controlled by parameters that are pre-set threshold value z^then(k, n, m), where m=1, 2, ..., M(k, n). During the control operation of the system measured values z(k, n, m) parameters of the nodes of each level and compare them with the preset corresponding threshold values z^then(k, n, m). This assessment of the state of the system is performed in stages To stages. At the first stage threshold and compare the measured values of the parameters of the system nodes belonging to the first level control the population. If their values are within the threshold, then the system is considered to be workable, otherwise generate a control action for the serial assessment of system nodes belonging to k=2, 3, ..., levels of management, which compare the measured and the threshold values of the parameters of the system nodes of the k-th level. If the measured parameters of the system nodes of the k-th level of control do not extend beyond the threshold, the state of the nodes of the k-th level is considered healthy. Otherwise, record the failure of the nodes of the k-th level and allocate its parameters do not satisfy the threshold, in line with what produce the control action for the recovery of parameters of the nodes of the k-th level of management. Moreover, the threshold values of the parameters z^then(k, n, m) set on the basis of a priori information about the current state of the system at all levels of management, forming a feedback loop system.

When the collective use of resources in a distributed on a huge global multi-level system cannot be distributed on request without extra effort, because competing for the resource requirements cannot self-organize in a consistent place. Occurs independent distributed management task that requires overcoming Tr is desta when you try to create a centralized system of control due to inevitable delays. The problems associated with routing in the area of distributed control, are overcome by using a distributed adaptive routing. The routes are formed as necessary in accordance with the current state of the system. The system receives control packets to adjust routing on current queue lengths and the resulting overload. These packets are given a low priority in the allocation of computing power that is not always justified. Actually these packet streams adjustments use the same expensive system resources and additionally its overload. In the proposed method, instead of periodic adjustments proposed a number of methods aperiodic correction at which these packets are sent only in the case when the parameters of the system will exceed certain thresholds, as well as a number of procedures that produce routing decisions based on local information about the lengths of the queues in a specific node, given knowledge of the current topology and automatic tracking changes to the system configuration, i.e. creating local control loops.

In fact, this procedure implements a multi-principle decision, moving to a decentralized control method and control distributed global INR is urovnevye systems.

The claimed device is a distributed monitoring and adaptive management multi-level system, shown in figure 1, consists of a tiered system 1, the inputs "interference" which is connected to the outputs of the perturbation source 2, the unit of measurement of the generalized indicator 3, the input bus "a generalized index of which is connected to an output bus "a generalized index of" multi-level system 1, while the ICC is equipped with an information input and output, and output buses, the system parameters of the k-th level, where k=1, 2, ..., K; K≥3, associated with the same tire units of measurement 4.1, 4.2, ..., 4. To each level of government, which through m of its inputs and parameters are linked To the same groups of inputs of the block decision 5, is provided in addition to the input bus "a priori", input "a generalized index" associated with the output unit of measurement of the generalized parameter, outputs "training", "normal" and "error", connected respectively to the same unit registration and management 6, having a control output connected to the control inputs of a multi-level system and the source of the disturbance.

The decision block 5, as shown in figure 2, the device of the distributed control and adaptive control of multi-level system designed to detect neribotas the vulnerable state of the system, identification of the level of disaggregation of the system (node), (b) where the fault has occurred, as well as recognition of the class of the technical condition of the system. It consists of a detector 5.1, IDs 5.2 (the number of levels of the management system) and resolver 5.3. The purpose of the block elements is as follows:

5.1 detector, designed to detect an inoperable state of the system by comparing its overall index with a threshold value, the transfer function identification of failure on the levels of disaggregation system, the establishment of threshold values for the parameters and the generalized metric system based on the analysis of the prior system state, the formation of the artificial traffic to ensure the training system when defining classes of its technical status;

5.2₁; 5.2₂; 5.2_Tothe identifiers are used to identify the system failure of control levels by comparing the parameters of the nodes in a particular level with their threshold values;

5.3 resolver is designed to determine the error control to match the class of the technical state of the system in the event of its failure and this information can be transferred to the registration unit and control system.

The input bus "a priori" and the input "a generalized index of the detector 5.1 are the same input bus and the input "a generalized index" block, direct access and exit "training" of the detector are respectively the outputs of the "norm" and "training" block, N(k)×M(n,k) inputs settings of the k-th ID 5.2_kare the respective group of inputs settings" block, direct output and output status detector connected respectively to the second and first inputs resolver 5.3, in addition, the exit status of the detector is connected to the first digits of the three-digit dual To the group of inputs resolver, entrance "signs" of the k-th identifier, starting with k=2, is connected to an inverted yield k-1-th identifier, and the second digits of a three-digit dual k-th group of inputs of the Recognizer, and the entrance "signs" of the first identifier (k=1) is also connected to the inverse the output of the detector, while the direct output of the k-th identifier is connected with the third digit of the first built-in input of the k-th group of inputs of the Recognizer, and the inverted output of the k-th ID - with the third digit of the second built-in input of the k-th group of inputs resolver, the first and subsequent outputs are outputs "error" block.

The detector 5.1, shown in figure 3 block a decision, is designed to detect an inoperable state of the system by comparing its overall index with a threshold value, the transfer function identification of failure on ur the attention of disaggregation system, setting thresholds on the parameters and the generalized metric system based on the analysis of the prior system state, the formation of the artificial traffic to ensure the training system when defining classes of its technical condition. It consists of a transducer characteristic "generalized indicator 5.1.3, item 5.1.4 comparison, shaper threshold 5.1.5, RS-flip-flop 5.1.6, generator artificial traffic 5.1.7 and control 5.1.8.

The purpose of the detector elements is as follows:

5.1.1 output 1 of the detector, is designed to transfer the values specified thresholds on the parameters of the nodes of the system identifiers of each level of disaggregation;

5.1.2 output 2 detector, designed for identity management unit of decision making in the training mode;

5.1.3 Converter characteristic "composite index" is designed to convert measured in the unit of measurement of the generalized parameter 3 values of the generalized indicator of the quality and parameters of the system into an electrical signal of a certain amplitude;

5.1.4 - element comparison is to compare the electrical signals from the different inputs and production control according to the comparison results;

5.1.5 - shaper threshold is designed for four is investing threshold values of indicators of the quality of telecommunication systems at various levels of disaggregation with regard to the information coming from the tires "a priori" about the current state of the system at all levels of government (during normal operation and in case of anomalies);

5.1.6 - RS-trigger;

5.1.7 generator artificial traffic is designed to simulate different load conditions of the system, necessary in the learning process and device settings;

5.1.8 - control is used to activate current control, bug control, and training mode;

The input bus "a priori", the input "a generalized index of" in and out "training" are the same bus, input and output unit decision-making, and direct its output "norm". The output of the detector status is ' connected with the generator of artificial traffic 5.1.7 and transducer characteristic "generalized indicator 5.1.3, the first input connected to the input "a generalized index". This Converter sign of its other output connected to the first input of the comparison element 5.1.4, the second input is connected to the first output of the shaper threshold 5.1.5 connected to the input bus "a priori". The second output of the shaper threshold is the first output of the detector, and the third is his second release. He connected with elements stored the control 5.1.8, R-input trigger 5.1.6 and is the output of "training" block decision, also the S-input of the trigger is connected to the comparison element and its direct and inverted outputs are the same outputs of the detector. Control its output connected to the generator of artificial traffic.

IDs 5.2_kblock a decision, shown in figure 4 and 5, are intended to identify system failure To control levels by comparing the parameters of the nodes in a particular level with their threshold values. Each of them consists of M transducers signs "settings" 5.2.3.m_kitem 5.2.4 comparison_k, adder 5.2.5_kand RS-flip-flop 5.2.6_k.

The purpose of the elements of the identifier consists of the following:

5.2₁; 5.2₂; 5.2_K-1the IDs of the lower management levels from the first to the (K-1)-th, shown in figure 4, similar in construction and purpose to all lower levels of management, except for the top K-th;

5.2_To- the ID of the top level management, shown in figure 5, similar in construction and purpose to the identifiers of the lower levels of management, in addition to the lack of outputs 1 and 2;

5.2.1₁; 5.2.1₂; ...; 5.2.1_K-1- input 1 identifiers of the lower levels of management, designed to supply values specified thresholds on parameters Oslo the system on the comparison element;

5.2.2₁; 5.2.2₂; ...; 5.2.2_K-1- input 2 IDs lower levels of management, is designed to transmit control signals RS-trigger mode "training";

5.2.1₁*; 5.2.1₂*; ...; 5.2.1_K-1* output 1 identifiers of the lower levels of management, is designed to transfer the values specified thresholds on the parameters of the nodes of the system identifiers of the higher levels of disaggregation;

5.2.2₁*; 5.2.2₂*; ...; 5.2.2_K-1* output 2 IDs lower levels of management, is designed to transmit control signals RS-trigger IDs higher levels of disaggregation;

5.2.3.1_k; 5.2.3.2_k; ...; 5.2.3.M_k- converters signs "parameters" k-level control system designed to convert measured in the measurement unit 4.k values of system parameters into an electrical signal of a certain amplitude;

5.2.4_k- item comparisons at each level of management is intended for comparing the electrical signals from the different inputs and production control according to the comparison results;

5.2.5_kthe adder at each level of management is intended to summarize arriving at its inputs signals and outputting the total signal amplitude on the elements of comparison;

5.2.6_k- RS-trigger on each level of the s control.

K-th group of M input parameters is the same group of inputs of the k-th identifier, and each m-th input, m=1, 2, ..., M, of this group and the output identifier "signs" are connected respectively with the first and second inputs of the m-th transducer characteristics options 5.2.3.1-M, the outputs are transmitted to the corresponding inputs of the adder 5.2.5, the output of which in turn is connected to the first input of the comparison element 5.2.4, a second input connected to the first 5.2.1 input and output 5.2.1* ID. The output of the comparison element is the S-input of the trigger 5.2.6, R-input of which is connected to the second 5.2.2 input and output 5.2.2* ID, and direct and inverted outputs of the trigger are the same outputs identifier. In the ID 5.2_Kupper control level (k=K) second input of the comparison element is connected only to the first input and the R input of flip - flop-only with the second input of the identifier.

The Recognizer 5.3 block a decision, shown in Fig.6 is intended to determine the error control to match the class of the technical state of the system in the event of its failure and this information can be transferred to the registration unit and control system.

The controls are the Recognizer is as follows:

5.3.1 - the first item;

5.3.1₁, 5.3.1₁*; 5.31 ₂, 5.3.1₂*; ...,5.3.1_K, 5.3.1_K* - dual group of items And;

5.3.2₁; 5.3.2₂; ...; 5.3.2_Toelements OR.

While the first and second inputs of the resolver connected to respective inputs of the first element And 5.3.1, the output of which is his first release of "error", first, second and third digit three-digit dual To the group of inputs resolver connected respectively with the first, second and third inputs To the dual groups of elements And 5.3.1₁, 5.3.1₁*-5.3.1_To, 5.3.1_To*the outputs are pairwise connected to both inputs To elements OR 5.3.2₁-5.3.2_kthe outputs which are To output error of the resolver.

The structure of the device of the distributed control and adaptive control of multi-level system is such that the block 1, indicated in the diagram of figure 1 as a geographically dispersed multi-level system can be a separate subsystem (switching node, local, regional, or global system), in this case organized control loops in each subsystem. The unifying element is indicated in the scheme of bus "a priori", which allows you to exchange information necessary to make sufficiently informed decisions with neighboring nodes p and distributed adaptive management or control center - with centralized management. This justifies the title of the invention that is associated with the territorial dispersal of resources multi-level system.

The block decision of the state system works in three modes: monitoring, evaluation of error control and learning.

An example of the operation block decision-making mode of the current control.

In control mode, the system state control 5.1.8 detector 5.1 disables the generator of artificial traffic 5.1.7. At the first stage of checking the correct functioning of the system in the unit of measurement of the generalized parameter 3 is the dimension of the generalized indicatorand the measured value is entered into the inverter characteristic "a generalized index" 5.1.3 detector 5.1. It is, conversion in accordance with the expression

where Λ(x) is the likelihood ratio.

Item 5.1.4 comparison of the detector is to compare the value of Λ(x) with a threshold value of x₀formed in the shaper threshold 5.1.5. If Λ(x)>x_oi.e. the malfunction is not detected, the signal from the output of the comparison element 5.1.4 via RS-trigger 5.1.6 and its direct output (output "norm" unit decision is passed to block registration and management 6, fiksirovannoi (N) the working state of the system. Otherwise, if Λ(x)<x₀), fixed abnormal () the state of the system and is a more accurate assessment of its condition by examining the set of features y_i, ..., γ_icoming from appropriate units of measurement 4.1, 4.2, ..., 4. the control loops of different hierarchy levels of the system through the input options, where. With the inverted output of the RS-flip-flop detector has a high potential is supplied to the second inputs of elements And 5.3.1₁and 5.3.1₁* the first dual-band Recognizer 5.3 and through the entrance "signs" includes converters features "settings" 5.2.3.1₁, ..., M₁the ID of the first level, allowing for processing values measured in block 4.1 system settings.

Values of measured characteristics in the k-th control loop, where k=1, 2, ..., K, with M inputs settings" enter M converters features "settings" 5.2.3.1_k, 5.2.3.2_k, ..., 5.2.3.M_kin which are formedand so on up tofurther summed in adders 5.2.5_kidentity of the appropriate levels of management.

In the elements of comparison 5.2.4_kidentifiers of the received amount is compared with the thresholds y₀...γ₀produced in f is Miravalle thresholds 5.1.5 of the detector and input through its output 5.1.1 inputs all identifiers 5.2.1 _k.

Whenthe signal on the performance of the system is fed with direct access RS-trigger identifier to the third input of the first element And 5.3.1_kk-th dual-band Recognizer 5.3. Ifthe signal from the inverted output of the RS-flip-flop of k-ID is supplied to the third input of the second element And 5.3.1k* k-th dual-band Recognizer, the second inputs of the first and second elements And (k+1)-th dual-band Recognizer and to the input of "signs" (k+1)th identifier, which comes to the second inputs of converters features "settings". As with the inverted output of the RS-trigger (k-1)-th identifier (and for the first level control - inverted output of the RS-trigger detector) on the second inputs of elements and k-th dual band resolver goes to a high potential signal, the locking extreme situation, with the second element And the k-th dual of the group goes through the element OR 5.3.2_kk-th group resolver 5.2 block registration and management 6. Similarly the identification of violations of the health system and beyond the contours of the control system. When this inverted output of the ID of the top level management 5.2_To(k=K) is connected only to the third input of the second element And 5.3.1_ToTo nd the dual group of the resolver.

Example a unit of decision making in evaluation mode error control.

In the evaluation of error control control 5.1.8 detector includes a generator of artificial traffic 5.1.7, which simulates normal N and abnormalthe state of a layered system in accordance with a priori probability P₁=P(N) and R₂=1-P₁=P(). Depending on the value of the threshold x₀set in the shaper threshold 5.1.5 joint and realization of the random variable Λ(x), decisions are made about the state of the system. If the value of Λ(x)>x₀with the output of the comparison element 5.1.4 via RS-trigger detector signal is supplied to the second input of the first element And 5.3.1 resolver. If the source is an abnormal system conditionthen from the first generator output artificial traffic 5.1.7 of the detector signal is fed to the first inputs of all the elements And resolver. However, due to the fact that both inputs of the first element And resolver serves a high potential on its output and the first output of the resolver appears a signal corresponding to "undetected" breach of system status -(failure is not detected) and, due to the statistical properties trafika the error detector, is passed to block registration and management 6.

In the case of fixing the detector violations mode system - On (failure detected signal is transmitted to the third input of the second element And 5.3.1k* k-th dual-band Recognizer 5.3 (via converters features "settings" 5.2.3.1_k, ..., 5.2.3.M_kthe k-th identifier adder 5.2.5_kthe element of comparison 5.2.4_kRS-trigger 5.2.4_kand its inverted output), the output of which appears the momentum corresponding to the event, when the detector detects the violation Of the normal operation of the system is N, but because of the error control Recognizer took him to the lock state. Because in this situation all the inputs of the second element And 5.3.1_k* the k-th dual band resolver receives a high potential on its output pulse appears, corresponding to the event when the detector is properly recorded violation -and the Recognizer classifies the normal state of the system N.

Similarly commit undetected failure" and "false rejection rate" and the subsequent hierarchy levels of the control system. The pulses from the outputs of all elements And each k-th dual group 5.3.1_k, 5.3.1_k* always transmitted to the recording unit and control 6, which when dostatochnomolodyh the number of tests generated the probability of a "false" condition R _l=P₁α₀α_pand the total, due to the detector and Recognizer "undetectable" extreme system state:.

Example a unit of decision making in the learning mode.

In the learning mode in connection with a multi-stage principle, the inclusion of resolver signal from the detector, there is a reduction R_lby increasing R_n. When optimizing the total value of P_∑=λ_l+P_nwhere 0≤λ≤1, undetermined Lagrange multiplier, the reduction of the second term can be achieved by optimal choice of thresholds. Thus the decrease of the values of x₀,₀, ..., γ₀made for m trials by clarifying the (m-1) test x₀(m-1), y₀(m-1), ..., y₀(m-1) due toknown methods, for example by the method of stochastic approximation performed by the driver threshold 5.1.5. The increment signis determined by the presence of a signal at the outputs of the first element And 5.3.1 or from the outputs of the elements OR 5.3.2_kresolver.

Appendix a shows an example of the calculation of the optimal thresholds classification that provides the minimum error making the Oia decisions (identification) system state.

An example of reducing the amount of information due to the phased principle of distributed control of a multilevel system.

Since the solution of the normal functioning of the system at the first stage can be made based on local information about the state of this node (for example, buffer memory, the status of the outbound communication channels and others), there is no need of exchanging information with other network nodes.

On the second and last stages together with (A.10) the analysis is made on the part of information, which causes the appearance probability of false alarm:P₁α₁P₁α₁α₂, ..., P₁α₁α₂, ..., α_nand which should be subjected to further analysis.

A value of P_LT=P₁α₁(1+α₂+α₂α₃+...+α₂α₃...α_n) the probability of "false alarm" gives the fraction of the total flow of information, adopted 1 that is subject to analysis on the second and subsequent stages. It determines the degree of reduction η_withthe amount of information that must be transmitted between nodes on a network to clarify the type of problem:

Thus, the degree of reduction of the volume of circulating in the system control information depends on the magnitude of errors of the first kind arising at each stage.

Phased about edora control allows precise detection of abnormal situations in the system, because it uses at each stage independent of the signs of recognition and, consequently, is not worse than Bayesian. With staged control the application of the decision on the state of the system is carried out with the involvement of additional signs as needed. The number of switching stages decreases as the growth rates of the stage in the x_itime. Control ends if the decision on the normal functioning of the system. The last stages are rarely used, while the total number of measurement data in the limit reaches the maximum value, almost using all available measurement information supplied by the system netmeasure, for a final decision on the status of the controlled system.

Analysis of the simulation results (figure 9) shows that the gain in terms of reduction of the volume of control information depends on the information content of signs of recognition on the second and subsequent stages, because the informative sign at the first stage is fixed and is determined by the amount of free buffer space, the value of which can be controlled strictly local information of each specific site. However, increasing the informative signs at subsequent stages associated with measurement systems is, amount of which determines the quality of a decision-making step-by-step control. These measurements are to improve the information content associated with the need for additional measuring natural resources and increasing time of analysis.

Building a multi-tiered system based on multi-stage decision-making process in comparison with the prototype and other known technical solutions allows you to validate the choice of the threshold values x₀, y₀, ... γ₀solving this problem optimally in the sense of minimum error classification of abnormal States of the system. While the classes of the system state, denoted by Fig mean:

"1" - the system is locked, the failure is detected and the detected;

"2" - the system works, a false detection and recognition;

"3" - the system is locked, the failure is detected but not detected;

"4" - the system works, a false detection is not detected;

"5" - the system is locked, the failure is not detected;

"6" - good health is recognized as healthy.

* at each of the stages of the control

In technical systems, it is preferable to have an error of the rst kind α₀(false rejection)than the error of the second kind β₀(undetected failure). Therefore preferred to Fig will be a system state "6*" and "1". The error is of ntrolle the proposed method can be reduced through the learning management system by analyzing current information, accumulated during the operation of the systems, methods, statistical theory of pattern recognition.

Annex a

An example of the calculation of the optimal thresholds classification that provides the minimum error decision (identification) status multilevel system

During the operation of complex, multi-tiered systems, performance which is characterized by a large number of parameters, the monitoring of their technical condition, it is advisable to carry out in several stages. At the first stage of any generalized metric tests the system performance and, in case of detection of abnormal situations at later stages as a result of more careful monitoring using information local, regional or global control loop is judged on its actual condition. This control procedure leads to a significant reduction in the time of the inspection and volumes circulating in the measurement data.

However, in the case when the detector IDs and the resolver on the first and subsequent stages make mistakes first (α₀α_p) and the second (β₀>β_p) sort of becomes an urgent question of the choice of thresholds in the state classification system for extreme situations and about the absence.

theory of statistical decisions allows you to specify the method based on the results of the analysis, which gives the minimum probability of error

the answer to the question which of the two sets N oris this particular state of the system and the corresponding vector S=S(x₁, x₂,..., x_n). In the process of applying the classifier decisions are errors of the first and second birth. A type I error α occurs when the hypothesis U₁(S=) is rejected although it is true, and the error of the second kind β - accepted hypothesis U₂(S-N), when it turns out a fair hypothesis U₁(S-).

Need to find a rule that would minimize the average risk of a W, or an average cost of decision-making (Bayes rule) about the errors of the first and second kind:

W=δ_a·α+δ_bβ,

where δ_a- weight errors of the first kind; δ_b- weight errors of the second kind. Decision rule can be formulated as follows:

where- a priori probability of occurrence of an abnormal situation in the system; P(x₁, ..., x_n/N) - conditional probability density of the normal functioning of the system; P(x₁,_...x_n/) is the conditional density is the terrain of the probability of occurrence of anomalies.

Thus, the decision rule that minimizes the average risk, compares the ratio of the probabilities with some threshold θ, which is a constant for certain values of the weights δ_αand δ_b.

This decision rule is called the criterion Bayes, and the relationis called the likelihood ratio. The conditional probability density P(x/N) and P(x/) formed in the process of learning the system, it should be assumed that they are close to true.

Criterion Bayes provides the highest accuracy of the solution walternative task recognition (identification).

A significant limitation of the methods of theory of statistical solutions is the complexity of their implementation, especially in distributed systems, in which the measurement information is distributed in the system using the implementation of the functions of coordination and control of expensive network resources.

In the process of constructing a multi-dimensional dividing line has to use all available measurement information in centralized and decentralized management. In the first case, all of the measurement information, based on which a decision is made on the operation of all nodes in the system, is collected in the main control center and on the developing countries to meet distributed on the periphery. In the second case, each node must have enough information to make an informed decision about the condition of not only this site, but, in General, and about the status of all nodes in the system to prevent global overloads.

During operation of the distributed multi-level systems, which are characterized by a large number of parameters, control of their health it is advisable to use a phased principle of classification. At the first stage of any generalized metric tests the state of the telecommunications system and, in case of detection of abnormal situations, on the second and subsequent stages conduct more thorough testing to determine its true state.

Since at each step the control system makes mistakes the first α(x_oiand the second β(x_oi) kind, becomes relevant to the choice of thresholds x_oiat each stage (Fig.7).

The error of the first kind

corresponds to the case of a decision about the occurrence of the abnormal situation on the network, while the system is operating normally, and is called the error of false alarm.

Error of the second kind

means the decision about the absence of abnormal situations, when there is Nara is giving mode of the system, called crossing violations.

Directly from the probabilistic state graph multilevel system (Fig) to obtain the total error skip violations for the entire system

and errors of false alarm for the system

where R₁=1-P₂- a priori probability of occurrence of the abnormal situation and P2 is the a priori probability of its absence.

In accordance with the criterion of Neyman-Pearson fix the probability of false alarm at a preset level

and minimize the likelihood of missing a

The problem of minimising the skip function (A.4)where the variables x_oiare linked to the functional dependence of (A.3)is a constraint optimization problem. Prepared functional optimization:

where λ is an undetermined Lagrange multiplier.

Calculated partial derivativeswe obtain a system of n

Equations

which together with equation (A.3) allow us to find an undetermined multiplier λ and n variables x_oi.

Equations (A.8) and (A.5) subject to (A1) and (A.2) after differentiation of the upper and lower limits, a modification of the t form:

In equations (A.9) are searched optimal thresholds

classification at each stagethat gives the minimum of the function (A.6), i.e. the minimum of the probability of missing in the system that is associated with the lowest probability of error occurrence of the abnormal situation. The solution enables to determine the probability of applying the right decisions about the absence of a violation of the system operating mode:

In each of the following stages of the analysis is the information about making the right decisions, that is about the normal functioning of the system.

1. The method of distributed control and adaptive control of multi-level system (ICC), which consists in the fact that the operation of the system in terms of exposure to destabilizing factors pre-set threshold values of monitored parameters for each of the nodes that form the system, measured at each node of the system, its parameters, compare the values of measured parameters with threshold, by comparing the results to assess the condition of the system and when the deviation of the system state on a valid form control action on the system, wherein the≥3 - tier system consisting of N(k) of nodes in the k-th level, where k=1, 2, ..., K, and n(k)-th node, where n(k)=1, 2, ..., N(k), characterized by M(k, n) parameters, the system state control on To the levels for which the threshold values of the parameters z^then(k, n, m), where m=1, 2, ..., M(k, n), pre-set for each of the nodes To level, and then measure the values of z(k, n, m) parameters of the nodes of each level and compare them with the preset corresponding threshold values z^nop(k, n, m), and assessment of the state of the system is performed in stages To stages, the first stage threshold and compare the measured values of the parameters of the system nodes belonging to the first level of control, and if their values are within the threshold, then the system is considered to be workable, otherwise generate a control action for the serial assessment of system nodes belonging to k=2, 3, ..., levels of management, which compare the measured and the threshold values of the parameters of the system nodes of the k-th level, if the measured parameters of the nodes in the system k-level of control do not extend beyond the threshold, the state of the nodes of the k-th level considered healthy, otherwise record the failure of the nodes of the k-th level and allocate its parameters satisfy the threshold, in line with what produce the control action for the recovery of parameters of the units of the k-th level control, moreover, the threshold values of the parameters z^then(k, n, m) set on the basis of a priori information about the current state of the system at all levels of management, forming a feedback loop system.

2. Device control and management of multi-level system that contains the unit of measurement of the generalized parameter, the input bus "a generalized index of which is connected to an output bus "a generalized index of" multi-level system (ICC), the inputs "interference" which is connected to the outputs of the source of disturbances, the ICC is equipped with an information input and output and control output unit registration and control connected to the control inputs of the ICC and the source of disturbances, characterized in that it further introduced the unit of decision-making and To units of measure, where K≥3, the total number of levels of control and management of the ICC, the input bus of the system parameters k level k-th unit of measurement, where k=1, 2, ..., K, is connected to an output bus "system settings" appropriate level of control and management of the ICC, the output of the measuring unit of the generalized indicator connected to the input "a generalized index" block decision, N(k)×M(n, k) outputs the "parameters" of the k-th unit of measurement, where N(k) is the number of nodes belonging to the k-th level of the ICC, and M(n, k) is the number of parameters n-th node belonging to k-th level of the ICC, is connected to the appropriate inputs "pairs the meters" block decision-making, equipped with advanced input bus "a priori", outputs "training" and "normal" block decision-making connected to the same inputs of the block registration and management, K+1 inputs "error" which is connected to the corresponding outputs of the error block decision-making.

3. The device control according to claim 2, characterized in that the unit of decision-making consists of a detector, resolver and identifier, the input bus "a priori" and the input "a generalized index" of the detector are the same input bus and entry generalized metric" unit, direct access and exit "training" of the detector are respectively the outputs of the "norm" and "training" block, N(k)×M(n, k) inputs settings" k-ID is the corresponding group of inputs settings" block, direct and the exit status of the detector are connected respectively to the second and first inputs of the resolver, in addition, the exit status of the detector is connected to the first digits of the three-digit dual To the group of inputs resolver, entrance "signs" of the k-th identifier, starting with k=2, is connected to the inverse output (k-1)-th identifier and the second digits of a three-digit dual k-th group of inputs of the Recognizer, and the entrance "signs" of the first identifier is also connected in ernim the output of the detector, when this occurs, the direct output of the k-th identifier is connected with the third digit of the first built-in input of the k-th group of inputs of the Recognizer, and the inverted output of the k-th ID - with the third digit of the second built-in input of the k-th group of inputs resolver, the first and subsequent outputs are outputs "error" block.

4. The device control according to claim 3, characterized in that the detector unit of decision-making consists of a transducer characteristic "a generalized index", item comparison, shaper threshold, the RS flip-flop generator artificial traffic and control input bus "a priori", the input "a generalized index of" in and out "training" are the same bus, input and output unit decision-making, and direct its output is "normal", "status" connected with artificial traffic generator and transducer characteristic "a generalized index", the first input which is connected to the input "a generalized index", this Converter sign of its other output connected to the first input of the comparison element, the second input is connected to the first output of the shaper threshold, connected to the input bus "a priori", the second output of the shaper threshold is the first you is Odom's joint, and the third is his second release, he is connected with the control, the R-input of the trigger and is the output of "training" block decision, also the S-input of the trigger is connected to the comparison element and its direct and inverted outputs are the same outputs of the detector, control its output connected to the generator of artificial traffic.

5. The device control according to claim 3, characterized in that each of the IDs unit of decision-making consists of M transducers signs "settings", adder element of comparison and RS-flip-flop, the k-th group of M input parameters is the same group of inputs of the k-th identifier, and each m-th input, m=1, 2, ..., M, of this group and the output identifier "signs" are connected respectively with the first and second inputs of the m-th transducer signs "settings", the outputs of which are received in corresponding the inputs of the adder, the output of which, in turn, connected to the first input of the comparison element, a second input connected to the first input and output ID output element of comparison is the S-input of the trigger, R-input of which is connected to the second input and output identifier, and direct and inverted outputs of the trigger are the same outputs ID, K-m the identifier of the second input element of the comparison is connected only to the first input, and the R-input of flip - flop-only with the second input of the identifier.

6. The device control according to claim 3, characterized in that the Recognizer unit of decision-making consists of the first element And dual groups of elements And To elements OR the first and second inputs of the resolver connected to respective inputs of the first element And whose output is the first output "error", first, second and third digit three-digit dual To the group of inputs resolver connected respectively with the first, second and third inputs To the dual groups of elements And the outputs are pairwise connected to both inputs To elements OR outputs which are To outputs "error" resolver.