Dynamic Multi-Model Soft Sensing Modeling Method Based On Clustering

On account of the complicated operation conditions, quick process changes in the actual production, often bringing about the measurement difficulty of the key variable. As an effective method to realizing the online estimation of the key variable, soft sensor technology could not only overcome the shortage of the online analytical instrument, realize the measurement of key variable, but also help with the optimized operation and control to the production process. So it received extensive attention by many experts and scholars, also had successful application. Soft sensing modeling method based on single model cannot describe global features of complicated systems, and the common multi-model soft sensing modeling methods had a deep dependent on the accuracy of clustering, in addition, they had the shortages of low integration ability and convergence performance ability. Because of these issues, this article do some work from the data clustering and multi-model fusion, combining with the process characteristics of industry, main work are as following:On the optimization of clustering algorithm:traditional fuzzy kernel clustering algorithm has a depend on the initial cluster centers and was easy to be trapped into local optimum, on account of these problems, the adaptive optimal clustering number of fuzzy kernel clustering method based on Gaussian kernel was proposed. It used the method that combined the density and distance to select the initial cluster centers and use the kernel validity index to evaluate the clustering results. The proposed algorithm could classify the data set successfully, determine the optimal number of categories, solve the problem of sensitivity to initial value, improve the efficiency of clustering and realize the algorithm without supervision. The simulation and application verified the effectiveness of the proposed method.On the problem of data clustering of the dynamic multi-model soft sensing modeling method:on account of the disadvantages of traditional model algorithm for the soft sensor, such as low predictive ability and poor accuracy, a multi-model soft sensor method is proposed based on D-S rule. Firstly, the adaptive fuzzy kernel clustering method (AFKCM) was used to establish multiple sub-models. Then the D-S rule was introduced into soft sensor, and the output of the soft sensor was obtained through the fusion of the sub-models based on the weight factor calculated by D-S rule. The ARMA model was used to realize the dynamic correction to the static multi-model output. Simulation results indicated that, the proposed method has better predictive performance.On the problem of soft sensing modeling method to the key variable in the actual chemical process:on account of the disadvantages of traditional soft sensing modeling method for the actual application, such as low fusion ability and poor adaptation ability, a multi-model soft sensor algorithm is proposed based on D-S rule and difference autoregressive moving average (ARIMA)model. On account of the predictive accuracy of the output of multi model soft sensor based on the affinity propagation clustering method, the weight factors were obtained by the D-S rule, then, on account of the precision of the weight factors, they were corrected by the discount factor. Then the multi-output of the soft sensor was obtained through the fusion of the sub-models based on the weight factor. In consideration of the non-stationarity of multi-output, the ARIMA model was used to realize the dynamic correction to the multi-output. The proposed method was used to estimate the ester rate and the result indicated that, the proposed method was an effective method.