Study On Data-driven Based Short Term Electric Load Forecasting Model For Papermaking Process

Papermaking process consumes tremendous electricity for production,the electricity cost is also increasing year by year.Forecasting future electric load could be conducive to manage the electricity consumption,determine the optimal production scheduling,which could improve energy efficiency and reduce the production cost.Therefore,this paper collects real-time data from different papermaking enterprises to analyze the electricity consumption characteristics of papermaking enterprises.According to different forecasting models,this paper analyzes the influence of input variables on the accuracy of forecasting models.At the same time,according to the electricity consumption characteristics in papermaking process,this paper has proposed different short-term load forecasting models,and analyzes the performance of different forecasting models in different production processes.Firstly,this paper collects real-time electric data from two different papermaking enterprises and analyzes the electricity consumption characteristics of papermaking enterprises.According to the analysis of the total effective power,there is no periodicity and instability in the electricity consumption of papermaking enterprises.Using three working conditions(normal,scheduled and unscheduled paper-machine downtime)of the papermaking process to classify the real-time electric data,and then preprocessing(including removing outliers,data re-filled and data filtering)the classified data,and collecting the reprocessed data into data sets before modeling and analysis.Finally,this paper uses the correlation function and lag autocorrelation function respectively to obtain the impact factors of the total effective electric power.According to the electricity consumption characteristics of the papermaking process,this paper proposed a short term electric load forecasting model based on GA-PSO-BPNN hybrid algorithm.According to the correlation function,the electric load of the electric equipment is recognized as input variable if the absolute correlation coefficient does beyond 0.6.The GAPSO algorithm is used in the proposed model to optimize the parameters of BPNN.And this paper proposed an electric load forecasting model to forecast half-hourly electric load for the papermaking process.Compared with the PSO-BPNN model and GA-BPNN model,the proposed model has the highest accuracy,and the MAPE is only 0.77%.This paper analyzes the accuracy of the forecasting model using different input variables.According to the correlation function and the lag autocorrelation function,the impact factors with high correlation coefficients are selected as input variables of the forecasting model.In order to ensure that the algorithm would not affect the comparison of the forecasting results,two hybrid algorithms(PSO-BPNN and PSO-LSSVM)are selected to develop the forecasting models for the papermaking process.In order to ensure that the data is not specific,this paper also uses real-time data from two different papermaking enterprises for modeling.The verification results reveal that the forecasting model with the input vasriables selected by lag autocorrelation function has higher accuracy.In order to solve the problem of deviation in forecasting electric load for the papermaking process,this paper proposed forecasting models based on signal decomposition hybrid algorithm.Firstly,this paper analyzed the characteristics of EMD algorithm and VMD algorithm separately.Combined with the advantages of the EMD algorithm and VMD algorithm,this paper proposed the EMD-VMD decomposition model.In order to ensure that the forecasting model has no effect on the verification results,this paper uses two hybrid algorithms(PSO-BPNN and PSO-LSSVM)to build forecasting models and analyzes the advantages and disadvantages of forecasting models based on different signal decomposition algorithm.The results revealed that building the forecasting model based on EMD-PSO-BPNN hybrid algorithm,VMD-PSO-BPNN hybrid algorithm and EMD-VMD-PSO-LSSVM hybrid algorithm for the papermaking process with stable production have universal application.For industrial processes with enormous fluctuations,building the forecasting model based on EMDPSO-LSSVM hybrid algorithm and VMD-PSO-BPNN hybrid algorithm have universal application.Finally,in order to realize the industrialization for electric load forecasting model,the paper uses the real-time electric data of a papermaking enterprise to analyze its power consumption characteristics.According to the requirements of the papermaking enterprise,this paper selects two algorithms(EMD-PSO-LSSVM algorithm and VMD-PSO-BPNN)to build forecasting model.Compared with the forecasting model based on EMD-PSO-LSSVM algorithm,the forecasting model based on VMD-PSO-BPNN algorithm has higher accuracy.In order to ensure the stable operation of the selected model,this paper compared the performance of the selected model with 5,10,15,20,and 30 as the update period.The results revealed that when the update period is 10,the stability of the model is the best.In order to ensure the long-term effectiveness of the selected model in this paper,the selected model was validated offline.The forecasting results proved that the forecast results of the selected model is consistent with the real-time load trend.The selected forecasting has been successfully applied to the MES system of papermaking enterprises.