Research And Development Based LSTM Method On City Gas Load Forecasting

Natural gas with absolute advantages in safety,efficiency and economy,as a leader in the field of energy consumption,has been extensively penetrated into all aspects of production and life in recent years and plays an indispensable role.However,there are still urgent problems that need to be solved in city gas pipeline dispatching and operation,such as unstable gas supply and irregular gas consumption by users.In order to guarantee the safe and stable gas consumption of users better,this paper researched the city gas load forecasting technology based on the Long Short-Term Memory(LSTM)model,in order to provide a forward-looking guidance for the city gas consumption forecast.First of all,the paper analyzed the influence mechanism of daily gas load characteristics and influencing factors on gas load changes.Second,the paper pre-processed the historical gas data.In the prediction process,the integrity of historical data is particularly important for the prediction results,but there are often missing values and outliers.For this problem,the SPSS software was used to detect and process the gas historical load and the missing values and abnormal data of influencing factors in the paper,and the accuracy of the load forecasting model is further improved by mining the historical outlier data.Next,the Pearson correlation coefficient was used to verify the significance relationship between the influencing factors and historical load data,to determine the input variable characteristics of the forecasting model.Finally,used the emerging deep learning category LSTM model and GM(1,1),support vector machine,BP neural network,combined with 2-year historical data of city A to forecast the daily gas load,the results show that the forecasting relative errors of SVM and BP neural network are similar,the maximum relative error is about 8.86% and 7.66%,of which the GM(1,1)relative error is the largest,the LSTM model is superior to other models,and the relative error is about 4.49%.At the same time,the paper implemented the calculation of four prediction models based on Python language,and used PyQt5 for software development,the forecast results data and graphics are displayed in the natural gas daily load forecasting system,which provides a powerful guidance platform and auxiliary means for city gas load forecasting.