Research On Gas Load Forecast Of Public Service Users Based On GRU Hybrid Model

Gas public service users such as boiler heating,catering,schools and institutions are an important part of urban gas load and play an important role in ensuring the normal operation of urban life and production.Gas load forecasting is the basic and key technology for the development of natural gas business.However,the gas load of public service users has the characteristics of transient,correlation,and nonlinearity,and the model is difficult to train,resulting in poor forecasting effect.Aiming at the above difficulties,this paper studies from three aspects:improving data quality,selecting load characteristics,and constructing prediction models,to achieve accurate prediction of gas load of public service users.The details are as follows.1.Research on the improvement of gas load data quality.According to the data situation,the gas load data is sorted and cleaned,including the processing of missing values and abnormal values.For further improve the quality of load data,this paper proposes a data quality improvement method based on the combination of empirical modal decomposition and continuous mean square error.The load data is decomposed,and modal reconstruction is performed to improve data quality.2.Research on the characteristics of gas load prediction.In this paper,the relevant features of gas load forecasting are constructed from four dimensions:continuous time features,discrete time features,time series historical features,and time series composite features.And through the three feature importance evaluation indicators based on the extreme gradient boosting algorithm,the importance of the features of different types of public service users is sorted,and finally the feature subset suitable for this type of user prediction research is determined by the sequence forward feature selection method.3.Research on gas load prediction model of public service users based on GRU.According to the characteristics of gas load data,this paper constructs a multi-factor input,adaptable GRU hybrid model that can learn the short-term and long-term time dependencies of load data,extract forward and reverse characteristic information of load data,and make stable and accurate predictions.The data quality improvement method based on empirical mode decomposition and continuous mean square error is combined with the construction model to complete the final prediction model construction.Through data quality improvement,feature research,model construction,and finally prediction experiments.The research results show that the model proposed in this paper has an average percentage error of 2.99%for the prediction results of the three types of users,compared with no data quality improvement,the overall average accuracy is increased by 0.4%.This paper studies the reasonable scheduling of regional loads,peak and voltage regulation of user loads,effectively reduce energy consumption,energy conservation and emission reduction,implement the national "dual carbon" policy,ensure safe,reliable and stable gas use in cities,and guide the planning and development of urban natural gas pipeline networks.It is of great significance to promote the application and analysis of digital technology in the gas field,and lead the industry to further explore and develop.