Research On Control And Filtering Strategy For Non-Gaussian System

With the large-scale and complex development of industrial systems,the random disturbances in stochastic systems play more significant role in system control and identification.The previous design method of control system is usually based on the interferences obeying gaussian distribution,while in the actual industrial process,the system model is usually non-linear,and the disturbance is not necessarily obeying gaussian distribution.In recent decades,the stochastic control theory has been continuously improved and updated,which brings hope to solve problems such as control of stochastic system.On the account of the probability density function of the system variables,it is not limited to the mean value and variance of the variables,and takes into account the high-order moment information of the variables,which has a broader significance.In this paper,considering that the Organic Rankine Cycle system(ORC)is subjected to non-gaussian disturbances,a recurrent neural network with special structure is adopted to identify the system.The neural network use the survival information potential(SIP)as the training criterion,meanwhile,signal flow graph is used to calculate the sensitivity,and the simulation results verify the effectiveness of the identification algorithm.Because the rotating speeds of the expand and pump cannot change too much,a predictive control algorithm based on the neural network model is designed,and the control law is obtained by using the differential evolution(DE)algorithm,the test of the set-point tracking ability shows that the designed controller is effective.Secondly,a controller based on quantized minimum entropy is proposed for ORC system.The controller adopts quantized minimum entropy as performance index,which greatly reduces the computational complexity of entropy compared with the traditional Renyi entropy.Due to some disadvantages of the traditional gradient descent method,such as slow convergence speed,easy to fall in local optimum,Particle Swarm Optimization algorithm(PSO)is used to solve the optimal control law,and the simulation results show that the designed controller can quickly track the change of the set-point,meanwhile,the probability density function of tracking error becomes more sharp and narrow.Finally,a mixture correntropy filter is scheduled for non-gaussian system.Compared with the traditional correntropy,the mixture correntropy utilized the gaussian kernel function with different kernel widths as the Merce kernel,which effectively solves the selection of kernel width.The proposed filter is applied in the wind energy conversion system and PSO is adopted to obtain the gain matrix of the filter.The simulation results show that the designed filter can estimate the system state effectively.