Modified fuzzy rule intelligent controller. RU patent 2504002.
 Method of registering unit element using fuzzy logic methods / 2473958In the method of recording a unit element, readings obtained after signal discretisation are associated with membership functions which characterise "good" reception of logic "1" or logic "0". Further, a procedure for resolving a fuzzy integral is carried out, after which a decision on receiving a binary unit "1" or zero "0" is made. To increase reliability of receiving a unit element on the decision boundary, presence of a zone of uncertain registration, wherein a decision on reception of a "delete" signal is made, is taken into account. |
 Optoelectronic defuzzification apparatus / 2439652Apparatus has a source of coherent radiation, an optical n-output splitter, two transparency filters, n optical Y-splitters, two optical integrators with a definite integration function, two photodiodes, a field-effect transistor with a controlled p-n-junction connected in a common-source circuit. |
 Optical boundary disjunctor for fuzzy sets / 2437139Device has a radiation source, two linear transparency filters, three optical n-output splitters, an optical three-output splitter, a group of n units for calculating the result, each having two optical Y-couplers, two pairs of optically connected waveguides, a photodetector and a piezoelectric element in which a second pair of optically connected waveguide is integrated. |
 Optical disjunctor for continuous (fuzzy) sets / 2432600Device contains m groups of k units for spatial distribution of optical flux, each consisting of a photodetector, a radiation source, an electrooptic deflector, a group of n optical waveguides, a linear transparency filter, a group of n optical j-output splitters and a group of n optical (n-j+1)-input couplers, k groups of n optical m-input couplers, k groups of n intensity normalisation units, each consisting of m-1 pairs of optically connected waveguides, m-1 transparency filters and an optical m-input coupler and k optical n-input couplers. |
 Optical or gate for fuzzy sets / 2422876Device has m groups of k photodetectors, m coherent radiation sources, m 2k-output optical splitters, m groups of k optical amplitude modulators, m groups of k optical phase modulators, m groups of k optical Y-couplers, k minimum signal selectors, k square-root extractors and k subtracters. |
 Optical and gate for continuous sets / 2419128Device has a radiation source, an optical Y-splitter, two optical k×n output splitters, two matrix transparency filters, k groups on n optical Y-couplers, k groups on n pairs of optically connected waveguides, k groups on n transparency filters, k optical n-input couplers. |
 Optical or gate for continuous sets / 2419127Device has a radiation source, an optical Y-splitter, two optical k×n output splitters, two matrix transparency filters, k groups on n optical Y-couplers, k groups on n intensity normalisation units, k optical n-input couplers. |
 Method of constructing fuzzy logic systems and device for implementing said method / 2417442Method of constructing fuzzy logic systems, where a sequence of fuzzy logic rules is created first. From each of these rules, a numerical characteristic is assigned - control quality index, where the fuzzy logic rules are enforced based on a trained neural network. Information signals or signals from the control object are transmitted to the inputs of the neural network. A sequence of output signals or a sequence of instructions and recommendations is formed at the output, where the trained neural network is a large trained artificial neural network. Each of the fuzzy logic rules is enforced by a separate fragment of the large trained artificial neural network (domain), where the number of domains corresponds to the number of fuzzy logic rules and also contains a certain excess number of backup domains. One of the domains functions of an arbitrator and switches outputs of the domains with outputs of the neural network based on the control quality index. |
 Multipurpose module / 2348976Invention concerns field of computer facilities and application in computing systems with parallel processing of the information and high speed can find. The device contains blocks of devices "AND", (n+1)-input adders, the logic block consisting from r of chains from consistently included frequency-modulated generator of start, the high-frequency self-oscillator with the plan of self-clearing which is delivery-radiating systems of the high-frequency self-oscillator. |
 Neural fuzzy net recognition device / 2340940Invention refers to data processing of special application, specifically for nonweighted digital codes driven signal and image transformation to weighed codes, and can be used for signal and image processing and recognition. Device contains two membership function drivers, overall similarity index driver and two fuzzy value shapers. |
 Behaviour recognition system / 2475853Objects represented in a stream are determined based on video frame analysis. Each object may have a corresponding search model used to track movement of the object from frame to frame. Object classes are determined and semantic representations of the objects are formed. The semantic representations are used to determine object behaviour and study behaviour arising in an environment represented by the obtained video streams. |
 Dynamic expert system / 2462752Dynamic expert system comprises a database memory unit, a rule base (RB) memory unit, an interference engine, an initialisation unit, a unit for definition of rules and consultations, registers of a list of updated rules, a control variable, a control rule, a number of rules conflict, a criterion of active rules, a status vector, a list of variables, a derivative rule, a list of justified rules, a local status vector, a limited variable and a true variable, a unit of rule selection, a unit of condition update, a unit of RB scanning, an input unit, a processor, a unit of derivability entirety, an explanation unit, a training unit, a communication unit, a decision generation unit and a user terminal. |
 Modified intelligent controller / 2458390Modified intelligent controller includes a control object, a solver, where the output of the control object is connected to the first input of the solver, a support computing unit, an action unit, a Kalman filter, a Kalman filter memory and an action selection unit. |
 Intelligent controller with self-modification rules of educational and control neural networks / 2447494Intelligent controller with self-modification rules of educational and control neural networks contains control neural network, educational neural network, control object. Additionally following units has been introduced in this controller: efficiency coefficient formula unit, educational neural network history unit, educational neural network control unit, control neural network control unit, control neural network history unit. In the second version, the educational neural network control unit consists of efficiency coefficient analysis unit, educational neural network rules unit, educational neural network operation results unit, and random-number generator of educational neural network. In the third version, the educational neural network control unit additionally consists of the control neural network analysis unit, control neural network history generation, control neural network rules unit, control neural network operation results unit, and random-number generator of control neural network. |
 Training convolutional neural network on graphics processing units / 2424561Method comprises the following steps: receiving graphics data representing a state of the convolutional neural network and comprising one or more textures representing one or more neural network variables, wherein said textures comprise a texture with two-dimensional addressing, and at least one of the textures represents a neural network variable with addressing of more than two dimensions which has been flattened into two dimensional addressing, the convolutional neural network comprising at least one layer comprising a plurality of patches; executing one or more programs on the graphics processing unit (GPU) in order to perform a forward pass in the convolutional neural network, executing one or more programs to perform a backward pass in the convolutional neural network, the executing including performing convolution operations on the patches; executing one or more programs in order to modify the patches in the convolutional neural network by changing the graphics data based on results of the backward pass; and repeating execution of one or more programs to perform forward passes, backward passes, and to modify the graphics data until the convolutional neural network is trained. |
 Cooled substrate and self-optimising supercomputer / 2367125Invention relates to connection of input-output devices or central processor devices or transfer of information or other signals between these devices. Provision for autoshaping manipulators in supercomputers in accordance with a given program with creation of an optimum architecture for the solved task is achieved due to that, the cooled substrate is a body with an inbuilt cooler, on which are attached (glued) microcircuit chips, containing multiple contact fibre optic connectors. On the end surface of the substrate there is a device for automatic attachment and connection of the substrate to a support. The autoshaping supercomputer contains a computer-builder, manipulators, controlled by this computer, as well as library for substrates with microcircuit chips and a fibre optic bus library. The supercomputer is automatically formed by manipulators by mounting the substrates onto a support and executing inter-substrate fibre optic connections through instructions from the computer builder. Manipulators are in the top part of the supports and guides above the supports. |
 System architecture and connected with it methods for dynamic addition of program components for system processes functionality enhancement / 2353968Invention contains description of systems and methods for automatic installation and using of filters for multimedia data processing. Installed filters later are called by other system processes such as device drivers, applications and software for data collection and processing. Filters are objects which can be used by multiple processes at any current moment. Search and numbering of installed filters may be performed by filter control service according to their categories. |
 Method of optimising controlled machine parameters / 2350999Present invention pertains to the method of optimising controlled parameters of a machine. The controlled parameters are optimised through executing at least, one algorithm included in a data processing system. Since the controlled parameters are optimised through execution of at least, one algorithm, included in a data processing system, optimisation is largely independent of the experience and knowledge of the operator. Thus, the optimisation process yields optimum adjustments of the machine much faster. This considerable independency of the optimisation process from experience and knowledge of the operator becomes significant when the machines are being used by service contractor organisations, where the operators are frequently not specialists in the given field, and usually have little knowledge or experience or none at all. |
 Proactive user interface containing evolving agent / 2331918Proactive user interface makes active proposals to the user based on the previous experience in relation to the current user and/or different preliminary programmed templates allowing the system to choose different options depending on user's behaviour. These proposals can be made using changes in appearance, at least, some part of the display, for example, by changing the menu or its part, providing different menus for display image, and/or changing functional abilities of the sensor display. Proposal can be also made in the form of sound. |
| Method for distributing keys in large system distributed over significant space / 2273877 For realization of method to key distribution center biometric images of user are delivered once and recorded, source artificial neural network is generated and user key as well, aforementioned neural network is taught so, that during its input to biometric image of user it spawned user key at outputs, after that neural network taught in such a way is delivered to user object via open communication channels, where it is recorded and utilized for forming key by feeding to its input by user of his biometric image, and if it is necessary to change the key new key is generated in center, source artificial neural network is taught again on new key and transformed neural network is delivered along open communication channel to user object, where it is recorded and utilized for receiving new key. |
 Adaptive terminal control system / 2500009Adaptive terminal control system further includes series-connected second unit for conversion from the system status function, second unit for calculating the system fundamental matrix, second matrix multiplier, vector adder which is connected by the output to actuating devices and by the second input to the output of an adder accumulator; the input of the second unit for converting to the partial derivative of the system status function is connected to the output of the unit of the free movement model of the control object, the input of the first unit for conversion to the partial derivative of the system status function and the input of the unit for converting to the partial derivative of the target function and series-connected unit of the weighting matrix of the terminal member of the optimised functional and third matrix multiplier, wherein the output of the third matrix multiplier is connected to the second input of the second matrix multiplier, the final state coefficient vector unit of the control object, connected by the output to the second input of the third matrix multiplier, as well as a storage element, a switch and a launch unit, the output of which is connected to the second control input of the switch and the second scalar input of a discrete Kalman filter; the first input of the switch is connected to the output of the storage element, and the output is connected to the scalar input of the unit of the free movement model of the control object and the scalar input of the adder accumulator. |
FIELD: information technology.
SUBSTANCE: device has a control object, an efficiency coefficient unit, a control neural network self-training rule unit, a system operation history unit, a control neural network, a fuzzification unit, a fuzzy output unit and a defuzzification unit.
EFFECT: improved adaptive properties of the control system.
5 dwg, 1 tbl
The invention relates to intellectual controllers, using the principle of learning reinforcement learning, fuzzy logic, and can be used for control of complex systems in non-deterministic environment.
Known intelligent controller-based networks adaptive critics, for example, U.S. patent 5448681, IPC G06F 15/18. This device consists of a control object, network criticism and the management of the neural network. Output control object is connected to the first input of the network critics and the first input of the neural network management, yield management neural network is associated with the input of object of management and a second entrance network critics, exit network criticism is connected to the second input of the management of the neural network.
Principle of operation of device for U.S. patent 5448681, IPC G06F 15/18 the following: control object produces a signal about its status, network critics calculates reinforcements to the current time iteration and the state of the object, the management neural network calculates the control action with regard to the reinforcement.
A General lack of devices based on networks of adaptive critics is that this algorithm is generalized and sufficient for the construction of a universal adaptive control system object, applicable in non-deterministic environment to ensure this, just look at the number of methods for implementing and structures (DHP, HDP, YOU GDHP, FACL, GlFACL and others). The disadvantage, as the system must be controlled in real time, a greater number of calculations. Also the disadvantages include the fact that the control system built on the basis of adaptive critic, cannot radically change their behavior and to develop new reactions when brand new, unknown to the environmental data and the control object D.Prokhorov, D.Wanch. Adaptive critic designs. IEEE transactions on Neural Networks, September 1997, pp.997-1007).
The closest technical solution is «intelligent controller with the neural network and the rules of self-modulation» patent RF №2266558, IPC G06F 9/00. The controller under this patent consists of the object of management, training of the neural network, neural network management, block coefficient of efficiency, block of rules of self management neural network, block history of the system work. Outputs status and operation of the control object associated with the first and second management inputs of the neural network output management neural network is associated with the input of the control object. The outputs of the operation and status of the object of management is also linked with the first and second inputs block coefficient of efficiency. Block coefficient of efficiency is connected to the first input of the training of the neural network, the second input of the training of a neural network is associated with the first output of the driver historical operating system, which is also connected with a second entrance of the block of the rules of self-study of the management of the neural network. Unit history of the system work in the first and second inputs connected with the object of management, and the third entrance of the block of the rules of self-study of the management of the neural network. The second output of the unit history of the system work associated with the release of the management of the neural network, third and fourth outputs are associated with the first and second management inputs of the neural network. Output of the training of the neural network is connected to the first input of the block of the rules of self-study of the management of the neural network. For more detailed schemes are also present blocks: the block of the function of the coefficient of efficiency, stack coefficient of efficiency, stack parameter adaptation and stack operation of the object of management, and also relations between them and above the blocks. As the management and training of neural networks is used multilayered perceptron. As a teaching method learning and the management of neural networks, the authors recommended method of reverse distribution of errors.
The operation principle of the patent RF №2266558, IPC G06F 9/00 the following: control object sends signals to the States and actions for the management of the neural network, block coefficient of efficiency and block the history of the system. Unit efficiency ratio calculates the coefficient of efficiency of the current cycle management and the dynamics of changes of the coefficient of efficiency compared to previous cycles. Managing neural network, if the coefficient of efficiency and more equal to the specified coefficient of efficiency, calculates the control action according to the current rules of management. Training neural network receives data about the change of the coefficient of efficiency and applies the rules of training the training of the neural network in the case when the coefficient of efficiency is less than specified. The output from the new values for the parameters adaptation. Block rules self-learning management neural network on the basis of changes in the parameters adaptation, if the coefficient of efficiency is less, changes the rules of the self-study of the management of the neural network. Managing neural network, if the coefficient of efficiency is less, according to the changed rules. After teaching management neural network generates a control signal. The necessary parameters are entered in the stacks.
The disadvantages of this controller is the insufficient adaptation properties, the impossibility of changing the rules of self-training training the neural network, limited change the rules of the self-study of the management of the neural network.
Task - the development of a modified intelligent controller with fuzzy rules.
The technical result of the proposed device is to increase the adaptation properties of the control system based on intelligent controller.
The technical result is achieved by a smart controller with fuzzy rules containing the control unit coefficient of efficiency, block rules for self-study of the management of a neural network, a block of historical operating system that controls a neural network, the output of which is connected with the input of the control object, the first and second outputs of the control object associated respectively with the first and second inputs block coefficient of efficiency, the first and the second block entrances history of the system work, the first and the second inputs of the management of a neural network, the first output of the unit's work history system is connected with a second entrance block of rules of self management of a neural network, third and fourth outputs are associated with the second and the first inputs the management of a neural network, a block of rules of self management neural networks associated with the fourth entrance of the block of historical operating system, introduced block , block fuzzy inference block , and the output of the unit coefficient of efficiency linked to the first entrance of the block , the first output of the unit's work history, the system is also connected with a second entrance of the block , the output of the unit connected with the entrance of the block fuzzy inference, the output of the unit fuzzy inference is connected with the entrance of the block , the output of the unit connected to the first input of the block of the rules of self-learning the management of the neural network and the third entrance of the block of historical operating system, the second output of the unit's work history system is connected with the third entrance of the block of the rules of self-study of the management of a neural network, the fifth time the block of historical operating system is connected with the third entrance of the management of a neural network, and the output of the neural network management is also associated with the fifth entrance of the block of the history of the system.
The task of enhancing the adaptive properties are achieved due to the introduction of blocks , block fuzzy inference and block that implement fuzzy conclusion of the adaptation option. Due to the data blocks are rising adaptation properties intelligent controller.
Thus, the set of essential characteristics included in the formula of the invention permits to achieve the desired result.
Figure 1 shows a General scheme of an intelligent controller with fuzzy rules, figure 2 shows the block coefficient of efficiency, figure 3 shows a block diagram of the interaction of the rules of self-learning management neural network, and block the history of the operation of the system, figure 4 shows the scheme of fuzzy inference coefficient degree of confidence, on the figure 5 shows the management neural network in the context of neurons.
The system consists of several structural components: control object 1, block efficiency coefficient 2, block 3, block fuzzy inference 4, block 5, block of rules of self management neural networks 6, block historical operating system 7, the management of a neural network 8.
Block coefficient of efficiency consists of the block of the formula of the coefficient of efficiency of 9 and stack efficiency coefficient 10.
Unit history of the operation of the system consists of a stack parameters adaptation 11 and stack operation of the object of management 12.
The system also contains the following communication: from the object management are signals condition 13 and 14, respectively block a formula coefficient of efficiency and 14.1 13.1, the stack object management 13.2 and 14.2 and control neural network 13.3 and 14.3.
From the block of the formula of the coefficient of efficiency is a signal 15 respectively 15.1 on the block and 15.2 per stack of the coefficient of efficiency of the unit go signal 16 on the block fuzzy inference from the block of fuzzy inference is a signal 17 on the block . From the block go signal 18, respectively on the stack parameters adaptation 18.1 and block rules self-learning neural network management 18.2. From the stack parameters adaptation is a signal 19 respectively on the block rules self-learning neural network management 19.1 and block 19.2. From the stack of the management of a neural network are signals 20.1 and 20.2 of the block of the rules of self-study of the management of a neural network, and from the stack of the management of a neural network is a signal 21, respectively - 21.1 per stack parameters adaptation and 21.2 per stack operation of the facility management. From the exit of the management of a neural network is a signal 22 on the stack object management. From the block history of the system work are signals - 23 on the management neural network, 24 and 25 of the inputs of the neural network management. From the control of a neural network is a signal 26 on the control object.
Managing a neural network consists of a layer of input neurons 34, hidden layers of neurons 35, the output layer neurons 36. The input layer neurons is connected to the first hidden layer connections 37.1, hidden layers are connected by ties 37.2...37.F+1, and the last hidden layer connected to the output layer of neurons connections 37.F+2.
The number of neurons in the input layer 34 management of the neural network is equal to 8 R+N, where R is the dimension of the state vector, and N is the dimension of the vector of steps of the control object. Neurons in figure 5 are marked as 34.1-34.R - input, is the state vector, 34.R+1-34 .R+N - inputs on the vector actions. From a management of a neural network can be F hidden layers, the number of which is specified by the developer (the recommended value F=2), respectively hidden layers are marked as 35.1-35.F, and the number of neurons in the hidden layers indicated 35.1.1-35.1.K in the first hidden layer and 35.F.1-35.F.L - in the last hidden layer. The number of neurons in the output layer 36 is equal to the dimension of the vector control management object and they are marked 36.1-36.. The number of R, N, F, K, L, M, Z are selected from the condition of a solved problem intelligent controller.
Block formula coefficient of efficiency 9 is intended for the calculation of the effectiveness of the system. Coefficient of efficiency is a dimensionless value, which shows how good or bad the whole system works, or by a particular parameter. The formula of calculation of the coefficient of efficiency is defined by the developer.
Stack efficiency coefficient 10 is intended for storage of values of the coefficient of efficiency in previous iterations management.
Block 3 is designed for calculation of the functions of the facilities.
Block fuzzy inference 4 is intended for calculation of forces fuzzy inference rules.
Block 5 is designed to calculate the precise values of the coefficient degree of confidence.
Block rules self-learning management 6 neural network is used for storage and adjustments sets of rules for training the neural network management 8 for various variants of behaviour of the control object. The essence of the formulas self-learning management neural network, depending on the value of the coefficient of efficiency, parameters adaptation, the status and actions of the object of management or the dynamics of their change, to prepare the training data for the management of the neural network and control the neural network.
Unit history of the system work 7 is intended to store information about the operation of the object of management, the management of the neural network and the parameters adaptation. Unit history of the system work is divided into two components - stack parameters adaptation and stack operation of the facility management.
Managing neural network 8 is designed to provide a control signal to the control object. This network is realized on the basis of multilayer perceptron. For training of the management of the neural network is used the method of reverse distribution of errors.
Stack parameters adaptation 11 is designed for storage of the parameters adaptation in previous iterations management. Parameters adaptation are the system settings, changing that to change the behavior of the system.
Stack operation of the facility management 12 is designed for storing the history of the object management 1 (on signals 13.2 and 14.2).
The claimed device operates as follows. Object management 1 generates signals condition 13 and 14. On these signals block formula coefficient of efficiency of 9 counts on a given formula resulting value of the coefficient of efficiency at the current iteration of control. The resulting value of the coefficient of efficiency at the signal 15.1 goes on the block 3 and is pushed onto the stack coefficient of efficiency at the signal 15.2. Block 3 gets the value of the current coefficient of efficiency, as well as synchronously from the stack, the coefficient of efficiency of the 10 previous values of the coefficient of efficiency at the signal 15.3 and values adaptation parameters in previous iterations on a signal from 19.2 stack parameters adaptation 11. Immersion depth by the coefficient of efficiency and the parameters adaptation-that is, how many of the previous iterations management take into account the data signals are specified by the developer. Immersion depth by the coefficient of efficiency and the parameters adaptation may not be equal to each other. Principle of operation ligament - block fuzzy inference - the following (see also N.G. «Fundamentals of theory of fuzzy and hybrid systems». M: Finance and statistics, 2004). The nodes of the input terms of the 27 are served crisp values of coefficient of efficiency for the latest and previous iteration management and AP for the previous iteration management. Output nodes 27 degrees, which preset inputs satisfy the functions of the facilities associated with the hosts. Each T-node 28 calculates the strength of the rules.
The output of top T-node 31.1 equal to:
α 1 =EC(t)_ formula TBE(t-1)_ formula AP (t-1)_
Output lower T-node is:
α z =KE(t)-_ formula TBE(t-1)_ formula AP(t-1)_
For other similar (see Table 1 below). N-sites 29 normalize force of the regulations from the outputs of the T-28 knots. Output for each N-node is equal to
Outputs of nodes normalization force of the rules of 30 are equal to the product of the normalized power of the rules and the individual output of the relevant rules. For example, for the first node of the normalization of force rules the output is b 1 /AP(t)_ (α 1 ), and for the last b 2 /AP(t)_ (α z ). Block 5 calculates the total value of the parameter adaptation of 18 by the formula:
y 0 =b 1 ·Y 1 +...+b Z ·Y Z
Further signals parameter adaptation of 18 go on the stack parameters adaptation at the signal 18.1 and block rules self-learning neural network management 6 on a signal 18.2. If the value of the signal 18 lies within modify, retraining process management neural network 8 does not start, in cases of «increased» and «decreased» start retraining process management neural network 8. Retraining of the management of a neural network 8 is as follows. Block rules self-learning neural network management 6 on a signal 21.2 initiates the operation of a stack object management 12 for the retraining of neural network management 8. The stack of the management of a neural network 12 activates outputs 24 and 25, who go to the inputs of the neural network management 8. When activated, 24 and 25 it is breaking inputs 13.3 and 14.3 and exit 26 of the management of a neural network from the object management 1. According to the signals of 24 and 25 on the control neural network 8 starting to get input combinations of examples, and at the signal 22 in the stack object management 12 begins to receive the output of the neural network management 8, which is compared with what had to happen when submitting a certain combination of inputs 24 and 25. The learning process takes place on the algorithm back propagation of error for a multilayer perceptron (see D.E. Rumelhart, Hinton G.E., R.J. Williams, "Learning representations by back-propagating errors," Nature, vol.323, pp.533-536, 1986). The weights 37 neural network management 8 goes on signals 23. After an error is managing the neural network was less than or equal to a specified, retraining process stops, outputs 24 and 25 of the stack object control is disabled and the device starts working in the mode of working off of signals of control object 1 and supply control signals 26 at him.
Calculation of signals neural network management follows a standard pattern.
The effect of this controller is that the proposed device through the use of fuzzy logic blocks - block , block fuzzy inference and block adaptively adjusts to changes the state of the environment and the control object. Thus, when formulating the values of the adaptation option that uses fuzzy logic, which allows not hard and gently adjust the adaptation option and accordingly rules for self-study of the management of the neural network.
TableThe inputs and outputs of training of neural network
Entrance 15.1 TBE(t)
Entrance 19.2 AP(t-1)
Entrance 15.3 TBE(t-1)
Exit 18 AP(t)
-1 -1 -1 1 -1 -1 0 1 -1 -1 1 -1 -1 0 -1 1 -1 0 0 1 -1 0 1 -1 -1 1 -1 1 -1 1 0 1 -1 1 1 -1 -1 -1 -1 1 0 -1 0 0 0 -1 1 -1 0 0 -1 1 0 0 0 1 0 0 1 -1 0 1 -1 0 0 1 0 0 0 1 1 0 1 1 -1 -1 1 1 0 0 1 1 1 1 1 0 -1 -1 1 0 0 0 1 0 1 1 1 1 -1 -1 1 1 0 0 1 1 1 1Modified intelligent controller with fuzzy rules, contains the control unit coefficient of efficiency, block rules for self-study of the management of a neural network, a block of historical operating system, managing the network, the output of which is connected with the input of the control object, the first and second outputs of the control object associated respectively with the first and second inputs block coefficient of efficiency, the first and the second block entrances history of the system work, the first and the second inputs of the management of a neural network, the first output of the unit's work history system is connected with a second entrance block of rules of self management of a neural network, third and fourth outputs are associated with the second and the first inputs of the management of a neural network, a block of rules of self management neural networks associated with the fourth entrance of the block's work history system, characterized in that it introduced the block , block fuzzy inference block , and the output of the unit coefficient of efficiency linked to the first entrance of the block , the first output of the unit's work history, the system is also connected with a second entrance of the block , the output of the unit connected with the entrance of the block fuzzy inference, the output of the unit fuzzy inference is connected with the entrance of the block , the output of the unit connected to the first input block of rules of self management neural network, and the third entrance of the block of historical operating system, the second output of the unit's work history system is connected with the third entrance of the block of the rules of self-study of the management of a neural network, the fifth time the block of historical operating system is connected with the third entrance of the management of a neural network, and the output of the neural network management is also associated with the fifth entrance of the block of the history of the system.