Research On Soft Sensing Of SO₂ Emission In A Power Plant

Based on the desulfurization system of A power plant,this paper traces and studies the components and working principle of the wet limestone-gypsum desulfurization system.But there are many problems with field sensors,such as:the high cost of sensor equipment installation,drift caused by humidity and ambient temperature,long measuring time,etc.The main purpose of this study is to find a soft-sensing method to measure for real-time monitoring of SO₂ concentration in the system,so that it can be kept within a certain range,so as to overcome the problem that the sensor can not meet the real-time on-line detection due to the time delay.Soft sensor technology is one of the hotspots of industrial process control research.The problem of non-linearity and high data dimension in process modeling under large data environment,a fast and accurate non-linear variable selection algorithm is also the necessary work of modeling.In this study,a soft sensing algorithm based on sequential backward selection（SBS）and multi-layer perception（MLP）is designed for desulfurization system.The details are as follows:1.A brief overview of the current desulfurization technology types is given.The principle and system structure of wet desulfurization technology and its influencing factors in the reaction process are emphatically analyzed.At the same time,the desulfurization technology of A power plant is introduced and studied in detail.2.The established algorithm model introduces akaike information criterion（AIC）criterion into SBS,so that the mean square error（MSE）value of the system stopping criterion is determined not only by the training MLP network,but also by the number of remaining variables removed and the training network.At the same time,cross-validation is added to obtain a more reliable and stable model.3.The proposed algorithm is applied to the desulfurization system to establish a model and compare with other algorithms.The results show that the proposed algorithm can achieve accurate measurement of SO₂ concentration in the system,and is superior to MLP neural network and traditional SBS-MLP algorithm in model accuracy and stability.4.Combining process principle and field operation experience,the results of variable selection are analyzed,and the importance of key variables in process and the correlation with target variables are discussed.The research results established in this study can provide reference for the optimization of wet limestone-gypsum desulfurization system.