Research On Static And Dynamic Soft Sensor Method For Anaerobic Fermentation Process Based On Improved Auto-Encoder

With the increase of food waste emissions,the harmless treatment of food waste has become an important goal of environmental governance in China.Anaerobic fermentation technology has become the most environmentally,friendly and economical way to treat kitchen waste.Since there is no real-time monitoring instrument for the fermentation process,acidification of fermentation process sometimes occurs in practical application,resulting in the failure of fermentation.In this thesis,the soft sensing of volatile fatty acid concentration in anaerobic fermentation process is studied.A FC-mRMR combined feature selection algorithm considering the interaction between features is proposed to select auxiliary variables.SSAE-KELM algorithm combining stacked supervised auto-encoder(SSAE)with kernel extreme learning machine(KELM)is proposed to Improve the accuracy and efficiency of model.A dynamic soft sensing model based on SSAE-GRU-Attention is proposed to cover the dynamic characteristics of the process.The main research contents are as follows:Aiming at the problem of selecting appropriate auxiliary variables for soft sensing,a combined feature selection algorithm FC-mRMR based on correlation redundancy analysis is studied.Considering the interaction between features to mine the feature information contained in the combined features,and then the best feature subset is selected as the auxiliary variable of soft sensing through correlation and redundancy analysis.The feature selection algorithm is verified by the actual data of anaerobic fermentation of food waste and the commonly used soft sensor model.Simulation results show that the algorithm can effectively reduce the feature dimension,improve the quality of auxiliary variable feature subset,reduce the calculation cost of model training,and help to improve the accuracy and training efficiency of soft sensor model.Aiming at the problems of low prediction accuracy,poor generalization performance and low modeling efficiency of traditional soft sensing methods,a soft sensing model based on stacked supervised auto-encoder combined with kernel extreme learning machine(SSAEKELM)algorithm is studied.Firstly,the deep feature is extracted by stacked supervised autoencoder,and the deep abstract feature representation is obtained by combining tag information.And extreme learning machine algorithm is used to train the network to improve the quality and efficiency of feature extraction.Then the regression prediction model is established by kernel extreme learning machine to predict the target variables.Simulation results show that this method has higher prediction accuracy than traditional soft sensing methods.Aiming at the dynamic characteristics of anaerobic fermentation process,the SSAEGRU-Attention algorithm based on stacked supervised auto-encoder(SSAE)and gating cycle unit(GRU)was studied,and the dynamic soft sensing model was established.Firstly,deep feature information is extracted from spatial scale by stacked supervised auto-encoder.Then,the dynamic feature information is extracted from the time scale by the Gated Recurrent Unit(GRU).Finally,the predictive value is output by gated attention mechanism.The simulation results show that compared with the static soft sensing model,the dynamic soft sensing model further improves the prediction accuracy and generalization ability of the model.This thesis uses the real data of anaerobic fermentation process of food waste to verify the algorithm.The simulation results show that the soft sensor model based on the proposed and improved algorithm has good prediction accuracy and efficiency,which provides the possibility for real-time monitoring and control of anaerobic fermentation of food waste.