Application Of XGBoost Algorithm In Gas Load Forecasting And Analysis Of Chengdu

Urban gas consumption is the main consumption area of natural gas in China.Urban gas load forecasting has become one of the unavoidable problems faced by the natural gas industry.From the macro point of view,accurate urban gas load forecasting is the prerequisite for the government and decision makers in the natural gas industry to formulate and implement natural gas policies rationally;from the micro point of view,accurate urban gas load forecasting is an important condition for urban gas companies to plan the production and transportation of urban gas reasonably and ensure the supply to avoid supply cuts.XGBoost algorithm has been put forward and applied in many fields.Especially in power load forecasting,urban gas load and power load have some similarities in load characteristics and influencing factors.Based on the comprehensive study of basic theory and forecasting methods of urban gas load forecasting,this paper studies the daily and quarterly load characteristics and rules of urban gas load in Chengdu.The factors of daily and quarterly gas load in Xiangcheng city are analyzed and studied.The forecast models of daily and quarterly gas load in Chengdu city based on XGBoost algorithm are established,and the forecast results are compared,analyzed and evaluated.The main research contents are as follows:(1)The characteristics and regularities of daily and quarterly load of urban gas in Chengdu are analyzed respectively.The results show that the daily load of urban gas in Chengdu is characterized by seasonality,catastrophe,trend,similarity,periodicity and randomness.The quarterly load of urban gas in Chengdu has the characteristics of time series,trend,periodicity,seasonality,fluctuation and difference.(2)Establish the influencing factors system of daily and quarterly gas load in Chengdu respectively.Qualitative factors in influencing factors are quantified by means of fuzzy quantification and weighted set value statistics.The influencing factors are screened by correlation analysis and partial correlation analysis.The final influencing factors of daily gas load in Chengdu are daily average temperature,PM2.5 and PM2.5.In terms of season,weather conditions and statutory major holidays,the influencing factors of urban gas seasonal load in Chengdu are primary industry added value,per capita disposable income of urban residents and seasonal average temperature.(3)Using grid search,the daily and quarterly load forecasting models of Chengdu city gas based on XGBoost algorithm are adjusted,and the daily and quarterly load forecasting models of Chengdu city gas are established respectively under the condition of finding the optimal parameters of the model.The results of daily load forecasting model of Chengdu city gas are compared with those of Kneighbor model,Random Forest model and Lasso Regression model.The results of daily load forecasting model of Chengdu city gas based on XGBoost algorithm fit best,and the average absolute percentage error(MAPE)is reduced to 2.31%,which is better than other models.The forecast results of Chengdu city gas quarterly load forecasting model are compared with those of support vector machine(SVM),Random Forest(Random Forest),Lasso Regression(Lasso Regression)and ARIMA(1,1,1)models.However,due to the insufficiency of quarterly load data,the quarterly load forecasting model of Chengdu city gas based on XGBoost algorithm is concluded.The fitting effect is better than other models,but the average absolute percentage is up to 9.55%,but it is acceptable.To sum up,the urban gas load forecasting model based on XGBoost algorithm can provide reference for gas forecasting in Chengdu and other cities,and has certain practical significance.