The Adjusting Of The Neural Network PID Controllers Parameters Based On Memristive Synapse

With the development of modern control technology, the novel intelligent controller is also improved constantly. However, as the result of the cost of intensive lithography increasing and the size of transistor tending to extremity, the areas based on the electronic circuit technology are limited greatly. Artificial neural network is a kind of nonlinear network, which has the properties of adaptivity, self- organization and learning. The synapses in neural network, connecting the neurons and completing the information transmissions, are crucial. Due to complex feature of synapse, such as nonlinear and correlation with time, it is difficult to model electric synapse with the existing electronic devices. Fortunately, the emergence of memristor provides an opportunity to solve these problems, which has nanoscale size and self-storage function. It is conducive to the further development of various hot research areas, including artificial brain, nonvolatile memory, chaotic system, the computer which could be powered on immediately and keep all the data in the memory. Similarly, the natural advantages of memristor also provide energy for the study of intelligent PID controller. PID controller is the main control mode in industry on account of the simple structure as well as the ease of operation and good stability. Yet, it is not easy to adjust the parameters of conventional PID controller once values of the parameters set. To settle this problem, the neural network PID controller is proposed. Using neural network to identify the parameters of system and continually adjusting it in the working process of the system, the aim of self-adaptation control is reached.Firstly, this paper introduces briefly the memristor and neural network principle, and discusses the similarity between them, providing the theory basis for the application of memristor as electric synapse. Moreover,Radial basis function neural network is a local approximation network, characterized by the features of rapid convergence, strong generalization and simple structure. It can approximate continuous function with arbitrary precision. In this paper,memristor is applied to radial basis function neural network PID controller. The mathematic theory of memristive synapse and the update formula of synaptic weights are deduced, and the structure of novel intelligent controller is build. The proposed system is tested by simulation with MATLAB and Simulink. What is more, the electric synapse based on Spintronic memristor is more authenticity and operability with its threshold characteristics, fast resistance changing and low production cost. PID neural network combining the advantages of conventional PID controller and neural network, can greatly control the multivariable time-varying systems. In this article, CPSO is used to initialize the weights of the network by using iteration optimization, to accelerate the convergence speed of the system. To build the circuit of PID neural network, the spintronic memristor is applied to PID neural network to simulate the feature of nature synapse, by taking simple pulse control to update weights. On the basis of detailed theoretical derivation, the effectiveness of the proposed scheme in the paper is verified by decoupling the strong coupling system.