基于天然气用户画像的需求量预测方法研究

Research on demand forecasting method based on natural gas user portrait

随着“双碳”目标的提出和实施,中国的天然气需求量逐年上升。天然气用户日益增多,用户的用气特征和用气规律不断丰富,掌握用户的共性特征、准确预测短期用气需求量是天然气管道调度调峰的关键问题。采用用户画像的方法对天然气用户的用气特征进行提取,建立天然气用户画像体系,并且根据用户画像结果结合神经网络模型、时间序列模型、机器学习模型对天然气用户的用气需求量进行预测。研究表明,天然气用户画像体系可以由直观型、比值型、推导型三大类指标体系作为框架建立,根据用气规律的不同可以划分为居民用气、非居民用气等用户属性;针对不同种类的用户采用神经网络方法和现代机器学习算法进行用户的小时需求量预测,结果误差在10%以下;确定了用户适用的模型——ARIMA模型更适用于用气波动较小的非居民用户,BP网络模型及GA-BP模型更适用于用气波动较大的居民或公服用户。

With the proposal and implementation of the “dual carbon” goals,the demand for natural gas in China increases year by year.The number of natural gas users is increasing,and their gas consumption characteristics and patterns are constantly enriched.Mastering the common characteristics of users and accurately predicting short-term gas demand are the key issues in natural gas pipeline scheduling and peaking.The user portrait method is used to extract the gas consumption characteristics of natural gas users,establish a natural gas user portrait system,and predict the gas demand of natural gas users based on the results of user portrait combined with a time series model,machine learning model,and neural network model.The study shows that the natural gas user portrait system can be established by such three main types of index system as a framework,namely intuitive,ratio,and derivation,and can be divided into user attributes such as residential gas consumption and non-residential gas consumption according to different gas consumption patterns.The hourly demand prediction of users is carried out by using neural network method and modern machine learning algorithms for different types of users,and the result is less than 10% error.It found that the user-applicable model is ARIMA model that is more suitable for non-residential users with smaller fluctuations in gas consumption and BP network model and GA-BP model are more suitable for residential or public service users with larger fluctuations in gas consumption.