Prediction Analysis And Comparative Evaluation Based On Daily Gas Consumption Data Of Urban Residents

With the promotion of clean energy policy and the continuous development of natural gas market,the demand and consumption of urban gas has increased significantly,and the infrastructure of gas industry has improved greatly.However,the supply capacity of gas and the dispatching level of supply and demand system are still insufficient.Therefore,in order to improve the operation efficiency of gas enterprises and promote the management and control planning of relevant departments,the accurate prediction of gas consumption is particularly necessary.In the existing related research,there are still some problems such as low accuracy of data,single evaluation dimension and weak universality of methods.Based on the above deficiencies,this thesis obtains a wealth of urban residents' daily gas consumption data with the help of the Internet of things technology,and carries out prediction analysis and comparative evaluation of gas consumption combined with the related influencing factors to improve the prediction accuracy and compare the advantages and disadvantages of various methods.The main contents of this thesis are as follows.(1)The existing research and related methods in the field of gas prediction are sorted out,and the appropriate prediction model is selected based on the characteristics of the data,and many data preprocessing measures are taken to increase the reliability of prediction.On the one hand,the gas data is cleaned,on the other hand,the influence features such as history,environment,date and policy are constructed by means of web crawler and data transformation,and the features are processed and selected,which greatly improves the overall data quality.(2)The classical method and the mainstream machine learning method are used to predict the gas consumption,and some improvement measures are taken to improve the accuracy of prediction.The classical methods include time series method and regression analysis method.The machine learning methods include integrated learning XGBoost model,deep learning LSTM model and improved XGBoost-LSTM combination method.By performing corresponding processing on different models and combining hyper parameter selection methods such as grid search,we have obtained good prediction results.Among them,the improved method has the highest prediction accuracy.Through measures such as custom loss function and weighted combination,the relative error rate of prediction is 2.19%,which improves the accuracy compared with the single model.(3)The prediction results of the two types of methods are compared from the perspective of a single model,and the selection criteria of gas consumption prediction methods under different characteristics and prediction periods are clarified.The results show that if the amount of data and features are rich and effective,the mainstream machine learning methods is better.Among them,the accuracy of the LSTM model is relatively high,the accuracy and performance of the XGBoost model are also excellent.With the change of the number of features and the prediction period,the classic method can sometimes get a good result.By analyzing the applicability,advantages and disadvantages of different methods under different conditions,this thesis provides some guidance and reference for relevant enterprises and researchers.