The Research Of Stochastic Distribution Theory In Non-Gaussian Systems

Due to the external disturbance and nonlinear factors,system variables usually obey non-Gaussian distribution in industry processes,and this is a challenge for system identification and control.With the continuous development of stochastic theories,many effective and feasible algorithms can be applied to the non-Gaussian systems.Neural networks feature in approximation and classification,so neural networks are widely used in system identification and control.As a novel neural network,extreme learning machine(ELM)can achieve minimum training error and minimum weight norm without iteratively adjusting weights in networks,so ELM possesses the advantages of fast learning and good generalization performance.In this paper,ELM method is utilized for the identification between pump speed and superheat degree in ORC(Organic Rankine Cycle)system,at the same time,inlet temperature and flow rate of flue gas are taken into account as non-Gaussian disturbances.Compared with the LSSVM(Least Square Support Vector Machine)method in training time and identification precision,ELM shows the better results.In addition,shape control problems in non-Gaussian system are studied in this paper.Firstly,in a nonlinear multi-variable system,output PDF(Probability Density Function)is approximated using RBF(Radical Basis Function)neural networks,further,relationship between weight vectors and control inputs is obtained applying dynamic neural networks.Based on the two step neural networks,control purpose of shape tracking is realized.During the controller design,SIP(Survival Information Potential)as an important part in performance index can address the influence suffered from non-Gaussian noise.Secondly,under the background of styrene polymerization,subspace identification method is used to obtain a state space model which describes the relationship between weight vectors and control inputs,finally,the shape tracking of molecular weight distribution(MWD)is realized.