Online local learning soft-sensing method for ball mill load of power plant

Based on the fact that the ball mill load in power plant is hard to detect effectively, an online local learning improved weighted least square support vector machine (WLSSVM) soft-sensing method is proposed form improving the prediction accuracy and adaptive ability of the soft-sensing model. Firstly, a similarity measurement criterion of data samples based on the samples distance and trend information is designed, and the modeling neighborhood dataset for local model is obtained by using two-step search strategy. Secondly, for the purpose of improving the prediction performance of the local model, an improved WLSSVM local model which can reflect the similarity and the error information of the modeling data is proposed, and the parameters of the local model is optimized by using improved differential evolution algorithm with mutation operation. Thirdly, in order to control the data size of the historical database, the active update strategy is adopted to realize the selective preservation of the new data. Finally, the simulation experiments are carried out based on the actual operation data of the coal pulverizing system. Simulation results show that compared with the standard LSSVM model, the proposed soft-sensing model has better prediction accurate and can predict the power plant ball mill load effectively.