Research On Two-Dimensional Recursive Identification Methods For Batch Processes

An accurate model is dispensable for advanced control and optimization.Unlike the continuous processes,the batch processes have distinctive features such as nonstationary process operations,nonlinear dynamics,time-varying characteristics,short batch duration and repetitive nature.Consequently,directly applying the recursive identification algorithm for continuous process may not yield satisfactory result.By utilizing batch repetition nature,existing identification methods for batch processes recursively identify along batch direction.There are problems such as large fluctuations in estimated results and low data utilization.By utilizing the two-dimensional characteristics of the batch processes,this paper studies the two-dimensional recursive identification methods for time-varying ARX model of batch processes.Three Methods are carried out to improve the accuracy and convergence speed in the paper.The main contributions are summarized as follows:(1)To address the problem of large variations of parameter estimates and low data utilization when applying RLS along batch direction,by employing the repetitive nature of batch processes and the principle of local modeling,this thesis proposes a two-dimensional recursive least squares identification method for batch processes based on local polynomials,which can improve the estimation accuracy.In this method,a local polynomial is used to parameterize the time-varying characteristics along time direction And the corresponding recursive algorithm is designed by minimizing the two-dimensional loss function along both time and batch directions.The simulation results show the eficacy of the proposed method on improvement of estimation accuracy and reduction of parameter fluctuations.(2)To tackle the problems of slow tracking speed and incapability of dealing with abrupt parameter changes in the conventional forgetting factor RLS approach,by combining the two-dimensional characteristics of the batch processes and incremental identification method,this paper proposes a two-dimensional incremental recursive least squares method for batch processes.In this method,parameter increments along time direction is introduced to describe the time-varying parameters of ARX model.And recursive least square method is adopted to estimate the parameter increments along the batch direction.By extending batch-invariant parameter assumption to the batch-invariant parameter increment assumption,the algorithm can update the parameters from the two-dimensional estimation results of the time t-1 and batch k-1,which can effectively accelerate the convergence speed.(3)Finally,by integrating the local polynomial modeling and incremental identification method,a two-dimensional incremental recursive least squares identification method for batch processes based on local polynomial modeling is proposed.In this method,the time-varying parameter increments along time direction is parameterized by local polynomials,and the recursive algorithm is designed by recursively estimate the local polynomial coefficients along batch direction.Simulation results show that this method can improve the identification accuracy and accelerate the convergence speed.