Residential Gas Demand Forecasting Based On Clustering And Combined Forecasting Model

Implementing the clean and low-carbon development strategy is an important task of China’s modernization development.As a clean and efficient energy,natural gas has become a bridge to the new energy economy.Realizing the accurate prediction of gas demand on the user side is the basic work to ensure the safe operation of gas system and the stability of natural gas supply chain.Taking residential users as the basic unit of gas consumption has important practical significance in residential gas pricing,classified management of different users,provision of personalized services and so on.However,due to the higher complexity and instability of users’ gas demand data,many studies on gas demand forecasting still have limitations in accuracy.Therefore,this thesis puts forward a method that combines fuzzy clustering method with combined forecasting model,clustering analysis of residential users,and forecasting the gas demand load of various users.First of all,collect and sort out the historical gas load data of residential users,analyze and sort out the law of user gas load changes and influencing factors.Then,the fuzzy c-means clustering model is constructed to analyze the characteristics of users’ gas consumption behavior.For each type of residential users,the differential Autoregressive Integrated Moving Average(ARIMA)model is used to process the linear features in the user gas demand data,and the Conditional Generative Adversarial Nets(CGAN)model is used to process the non-linear features in the user gas demand data,and In the internal network structure of the generative model and the discriminant model,the Long-short-term Memory Recurrent Neural Network(LSTM)is introduced;the e Xtreme Gradient Boosting algorithm(XGBoost)is used to combine the prediction results of the two single models to construct the ARIMA+CGAN+XGBoost combined prediction model,To obtain more accurate user gas demand load forecast results.Finally,based on the first mock exam of the actual gas data in Tianjin K District,the classification of users in different gas consumption modes is realized.1488 resident users in the district are divided into two categories.The gas demand combination forecasting model is established for each type of user.Compared with the single model,the ARIMA+ CGAN+XGBoost combined forecasting model has higher prediction accuracy.The method used in this thesis takes into account the linear and nonlinear characteristics of user gas demand data,and realizes the theoretical innovation of user gas demand prediction method to a certain extent.