Research On Load Forecasting And Peak Shaving Strategies Of Short Term Natural Gas Based On GA-LSTM And Transfer Learning

The rapid development of urbanization and the continuous increase of new urban areas and residential areas have brought about an increasing demand for urban natural gas.Due to the lack of sufficient historical load data in the new urban area,effective short-term natural gas load forecasting cannot be carried out,which brings many inconveniences to the construction of urban natural gas facilities and gas supply planning arrangements.If effective natural gas facility construction and gas supply planning cannot be carried out,it will be difficult to effectively regulate the peak period of urban gas consumption.Therefore,studying how to predict short-term natural gas load in newly built urban areas lacking historical data has important theoretical value and practical significance.In order to solve the problem of short-term natural gas load forecasting and peak shaving in new urban areas,this paper proposes a combined forecasting method of GA-LSTM and Transfer learning on the basis of analyzing the characteristics of natural gas load and basic forecasting methods.First,the new urban area is selected as the target area,and the GA-LSTM prediction model is constructed using a small amount of historical natural gas load data obtained from the target area urban area.Then,the old urban area in the same city is selected as the source area.Based on this model,a large number of historical load data in the selected source area are evaluated for the correlation of Mutual information and divided into levels.Then,based on the correlation from low to high,the hierarchical data will be sequentially trained into a deep GA-LSTM prediction model,thereby completing the effective migration of source domain urban data to the target new urban area,and achieving short-term natural gas load prediction for new urban areas lacking historical data.Finally,the Feature selection method is used to optimize the migration model to ensure the effective migration of strongly related features.According to the prediction model,the natural gas load in the next year’s peak period is predicted.Combined with the peak shaving capacity of existing peak shaving facilities,the cost and reliability of peak shaving facilities and other factors,the appropriate peak shaving method is selected,and then the peak shaving strategy suitable for the city is formulated.The empirical results show that the accuracy of the combined prediction model of GA-LSTM and deep migration is 21.6% higher than that of the general migration prediction model.Adding Feature selection optimization on the basis of deep migration can improve the accuracy of the prediction model by 6.8%,which proves the effectiveness of Transfer learning in solving the prediction problem of lack of data.By using this model to predict the natural gas load during the peak period of the next year,the daily peak coefficient and pre peak shaving volume are calculated,and compared with the peak shaving capacity of existing facilities.Taking into account economic and reliability factors,appropriate peak shaving methods are proposed,and peak shaving strategies suitable for the city are formulated.By constructing a short-term natural gas load forecasting model for newly built urban areas lacking historical data,selecting case studies for empirical research,and combining the economic and reliability factors of peak shaving facilities on the basis of load forecasting,selecting urban peak shaving methods and formulating peak shaving strategies that are suitable for the city.The proposed method enriches the theoretical foundation of natural gas load forecasting and provides certain guidance for urban natural gas load forecasting and peak shaving strategy formulation.