Research And Application Of Data-driven Energy Consumption Prediction Model For Reduction Furnace

Silicon material is an important basic material in photovoltaic industry and electronic industry.In a sense,silicon material is a strategic resource in the new energy field.As the main equipment of polycrystalline silicon production,the energy consumption of reduction process accounts for 70%.Under the background of green manufacturing and low carbon manufacturing,how to reduce the energy consumption of reduction furnace has become the focus of attention,and the reduction process is a complex and dynamic process.Therefore,it is necessary to explore the relationship between energy consumption and influencing factors,establish energy consumption prediction model,ensure supply and demand balance,reduce resource waste,and provide support for enterprise decisionmaking.Taking the reduction process in polycrystalline silicon production as the object of study,this paper analyzed the factors affecting energy consumption and energy consumption data.Aiming at the problems of complex factors affecting energy consumption,non-linearity and low accuracy of traditional prediction methods,based on clustering of energy consumption data and eliminating abnormal data points,an energy consumption prediction model based on improved LSTM long short-term memory neural network was proposed.Secondly,the energy consumption prediction module based on micro-service is developed,and the prediction module is integrated into the energy management system.The main contents of this paper are as follows:(1)In view of the complex structure of the reduction furnace and energy consumption in the reduction process,this paper analyses the structure of the reduction furnace,working principle and energy consumption characteristics of the reduction furnace,and reveals the influencing factors and periodic characteristics of energy consumption of the reduction furnace,and forms the characteristic set of influencing factors of energy consumption in polycrystalline silicon reduction process.(2)Data analysis of energy consumption in reduction process.Because the actual production situation is complex,equipment failure and fluctuation of acquisition program transmission will cause redundancy and abnormal value of energy consumption data in reduction process.On the one hand,using reasonable methods to detect abnormal data can help decision makers find abnormal situation,on the other hand,improve the accuracy of energy consumption prediction model in reduction process.Based on the analysis of the influence factors of energy consumption by GRA-PCA algorithm,this paper proposes a clustering analysis method for energy consumption data based on the combination of neighbor propagation clustering and local anomaly factors.(3)Establish a prediction model.In this paper,an improved LSTM neural network energy consumption prediction model is proposed.By analyzing different optimization methods,Adam algorithm is selected to train the neural network model,and Adaboost lifting algorithm is introduced to continuously adjust the weight of the model to obtain an enhanced network model.Finally,a prediction model is constructed to realize energy consumption prediction in the reduction process.In order to verify the validity of the prediction results,two modeling methods,BP neural network and long-term memory neural network,are compared in this paper.The results show that the improved LSTM model has the highest prediction accuracy.(4)Establishment of energy demand forecasting model base.An energy consumption prediction module based on micro-service is constructed in the energy management system.With the help of Java API,the trained prediction model is embedded into the system,and preliminarily applied in enterprise production.