Data-driven Process Performance Evaluation Method

Process performance evaluation system plays a very important role in industrial production and economic construction,and undertakes the important social responsibility of ensuring the safety and stable operation of enterprises.The performance of the industrial process system may deviate from the optimal performance due to the industrial process,personnel's operating errors,and the production of the industrial process system,which may lead to the decline of economic benefits and even production accidents.In order to detect the change of industrial process performance in time,it is necessary to evaluate the performance of the industrial process system on line to adjust the industrial process system before the performance drops.In order to solve the problem that the physical or chemical model of industrial process is difficult to get and the performance test result of industrial process is difficult to get online,an online evaluation method of industrial process performance based on data driving is proposed in this paper.Firstly,the background,application value and characteristics of data-driven methods are described,and the relevant algorithms of data-driven methods are analyzed and studied.Then,based on the application of PCA in performance evaluation in multivariate statistical analysis,analysis the principle,calculation process and characteristics of PCA.Since the principal component analysis method lacks the investigation on the relationship between process variables and quality variables,partial least squares method is adopted in this paper,and the basic principle,calculation process,advantages and disadvantages of this method are analyzed.The principal component analysis method and partial least square method are simulated and verified.In view of the complexity of partial least squares algorithm and the lack of orthogonal decomposition of data space,this paper fuses the idea of autoregression with partial least squares to design an autoregressive latent structure projection algorithm.The algorithm can carry out orthogonal decomposition of the sample data space and simplify the modeling process.Finally,the algorithm is simulated and verified.In order to obtain the evaluation results of process performance,this paper combines the fuzzy c-means clustering algorithm with the autoregressive latent structure projection algorithm.The sample data used in modeling are divided into categories,and the membership functions of each category are calculated.Using membership functions to calculate the online calculation results of membership degrees of each category,and the final evaluation results are obtained according to the fuzzy comprehensive evaluation method.Finally,the simulation of blast furnace temperature state evaluation is carried out to verify the effectiveness of the method,which provides an effective scheme for complex and nonlinear process performance evaluation.