Neurocontroller Training with Using the Neural Network Emulating the Plant

While using the "controller modeling scheme", there are restrictions on the settings of the neural network controller. They are related to the fact that the formation of a training sample is carried out by using a reference controller. It means that the neural network controller obtained after training cannot significantly increase the quality indicators of the system's transients. The application of "emulator and neurocontrol scheme" is demonstrated in the paper. This scheme is considered as an addition to the algorithm for the design of neural network controllers with a deterministic method for choosing the architecture and weight coefficients of a neural network, using controller modeling scheme. An example on improving the qualitative characteristics of the transients of the system by additional configuration the neuro-controller for the nonlinear system "inverse pendulum on a movable base" is given. The purpose of the control was designated as stabilization of the inverse pendulum in a vertical position and moving the mobile base to a set value. To achieve these goals, a neural control scheme, which contains two neural networks: a neurocontroller (performs the function of forming a control effect on the plant) and a neuroemulator (performs the function of simulating a model of the plant and is necessary for calculating the backpropagation error skip and adjusting the weighting coefficients of the neurocontroller) is used. As a result, it is possible to obtain an automatic control system capable of controlling the specified plant.