Study On Multi-step Energy Consumption And Carbon Emissions Forecasting Model Based On Machine Learning Method Of The Papermaking Processes

With the increasing demand for production capacity in paper mills,its energy requirements are rising year on year and its carbon emissions base is expanding.Under the carbon peaking and carbon neutrality goals,paper mills are under dual pressure to improve energy efficiency and reduce pollution and carbon emissions.The carbon emissions of paper mills are indirectly derived from electricity and steam energy consumption.The accurate forecasting results of the multi-step energy consumption and carbon emission forecasting models can be used to help paper mills optimize their production scheduling,improve energy efficiency,save energy and reduce costs.It can also provide accurate demand information to public works providers to rationalize production scheduling and additional consumption and emissions caused by excess or insufficient energy.Finally,it helps to form a virtuous cycle of reducing pollution,carbon emissions reduction and cost reduction,and improving energy efficiency on the supply and demand sides.Paper mills have the characteristic of the process industry and the discrete industry,resulting in numerous factors that influence energy consumption and carbon emissions having complex coupling relationships and strong irregularities.In terms of influencing factors,there are hundreds of collection points with complex and variable correlations and many key process parameters that cannot be measured online.In terms of process characteristics,the electricity consumption,steam consumption and carbon emissions of the papermaking process have unique characteristics,including electricity consumption with high volatility and interference,steam consumption currently with weak robustness and uncertainty,and carbon emissions with sloppy accounting and unclear formation mechanisms.The above problems make it difficult to clearly explore the key factors affecting energy consumption and carbon emissions,and geometrically difficult to propose accurate multi-step forecasting models for energy consumption and carbon emissions in the papermaking process with the forecasting step increasing.To address those problems,the following works are undertaken in this thesis:(1)Statistical methods were used to uncover the key factors affecting energy consumption and carbon emissions in the papermaking process.First,this study analyzed the features affecting the energy consumption and carbon emission of the paper production process from the time domain and the frequency domain.On this basis,this study used grey correlation analysis,correlation analysis and autocorrelation analysis methods to initially screen those analyzed features and then used the GBRT-SHAP method to filter the key parameters more precisely and find reasonable input variables for the subsequent modeling of energy consumption and carbon emissions.(2)A new multi-step forecasting strategy was developed.Firstly,this study analyzed the advantages and disadvantages of five classical multi-step forecasting strategies(iterative strategy,direct strategy,multi-input multi-output strategy,direct and iterative strategy,and direct and multiple-output strategy)and proposed a new multi-step forecasting strategy based on different acquisition frequencies.The six forecasting strategies were used to build multi-step forecasting models.And the proposed multi-step forecasting strategy was verified by the comparison with those class strategies.(3)A multi-step electricity consumption forecasting model based on a deep learning method for the papermaking process was proposed.Firstly,this study proposed a single-step electricity consumption forecasting model based on PSO-LSTM.After that,the eleven forecasting models based on the combination of WOA,PSO with BPNN,Elman,LSSVM and LSTM were used as contrasting cases to verify the performance of the proposed electricity consumption forecasting model.The results show that the proposed forecasting model has the highest accuracy.Compared to other models,the MAPE and RMSE of the proposed model are reduced by 9.37% and 10.06% on average,respectively.Next,this study proposed a multi-step forecasting model based on the single forecasting model and the proposed strategy.The collected data was used to verify the proposed multi-step forecasting model with different forecasting steps.The results show that when the number of forecasting steps is within 24,and the proposed multi-step forecasting model for electricity consumption in the papermaking process has the advantages of high accuracy and stability,with a pre-MAPE of less than 2%.(4)A multi-step interval forecasting model for steam consumption in the papermaking process based on the key parameter identification was proposed.First,a key parameter identification model based on Light GBM for the papermaking process was proposed.On this basis,a probabilistic forecasting model for steam consumption of the papermaking process was proposed based on BO-LSTM.The forecasting model based on PSO-LSTM and BNN was used as a contrasting case to verify the performance of the proposed model.The results show that the proposed forecasting model has higher accuracy and better stability than the other two contracting models,and its MAPE and RMSE decreased by 23.2% and 6.78%,34.4% and27.1%,respectively.The interval forecasting model for steam energy consumption in the papermaking process was developed based on the KDE method.The actual data was used to validate the proposed interval forecasting model.The results showed that 99% of the actual values fell within the forecasting interval at a confidence level of 90%.Finally,a multi-step interval forecasting model for steam consumption in the papermaking process was developed.By comparing the forecasting results for different numbers of forecasting steps,the proposed multi-step forecasting model for steam energy consumption could achieve high accuracy with98% of the actual values falling within the forecasting interval at a confidence level of 99%when the forecasting steps are below 24.(5)A multi-step carbon emissions forecasting model for the papermaking process based on a combination method was proposed.First,the study proposed a carbon emission combination forecasting model based on the ARMA,grey method and LSSVM with the proposed key parameters identification model.Second,the weights of the combination model were solved using the IOWA operator and MCST method.The carbon emissions forecasting models based on ARMA,grey method and LSSVM were used as contracting cases to verify the proposed model.The results showed that the proposed carbon emissions forecasting model combined the advantages of the three contacting models to a large extent,and its forecasting accuracy and stability were the best.At the same time,this study compared the errors in the direct forecasting of carbon emissions and the indirect calculation of carbon emissions through the forecasting results of the proposed electricity consumption forecasting model and the proposed steam energy forecasting model.The results show that the direct forecasting model of carbon emissions has higher accuracy.Finally,the multi-step carbon emissions forecasting model was proposed.The collected data were used to validate the proposed model.The results show that the proposed multi-step forecasting model could accurately forecast carbon emissions for paper mills within 12 forecasting steps,and its 90% relative error percentage is less than10%.Through the above research,this study analyzed the key factors affecting the energy consumption and carbon emissions of the papermaking process,proposed a novel multi-step forecasting strategy,and on this basis developed three high-precision and highly stable multistep forecasting models for energy consumption and carbon emissions,which provides data and technical support for the subsequent pollution and carbon reduction,cost reduction and efficiency improvement of the papermaking process.