Research On Intelligent Forecasting Technology Of City Gas Short-term Load

With the construction of many natural gas transmission pipelines such as West to East gas transmission,Sichuan gas transmission to East,Shaanxi Gas transmission to Beijing and so on,China has achieved rapid development in the construction of natural gas pipe network,which makes the gas supply capacity of Chinese cities improve further.However,at the same time,problems such as unbalanced supply and demand of urban gas,unscientific gas dispatching,inadequate gas supply during peak hours of gas consumption and imperfect construction of gas storage and distribution facilities are becoming increasingly prominent.Therefore,it is of great significance to study the variation law of urban gas load and accurately forecast the gas load for realizing the efficient operation and reasonable dispatching of gas supply system and the construction of gas storage and distribution facilities.By summarizing the current research status of urban short-term gas load in recent years,it is found that the current methods used for gas load prediction are mainly computer intelligent prediction models based on machine learning and intelligent optimization algorithm.Therefore,on the basis of studying the change rule and influencing factors of gas load,this paper studies the common single machine learning model and its optimization prediction model.The main research contents are as follows:Through the analysis of the historical gas load,it is found that the city gas load has three characteristics: periodicity,volatility and randomness.Factors affecting urban short-term gas load are explored from four aspects: historical load,meteorological condition,date type and heating condition.The influencing factors are screened by grey correlation analysis method,and 10 factors affecting urban short-term gas load are finally determined.In this process,the non-numerical data is numerically encoded to facilitate computer reading and recognition,and then the min-max normalization method is adopted to standardize the research data.By summarizing the research status of intelligent gas load prediction at home and abroad,BP neural network,least square support vector machine and extreme learning machine were selected as the basic prediction models.Particle swarm optimization algorithm and Seagull optimization algorithm were used to optimize the relevant parameters of the three models.Six optimized short-term gas load intelligent forecasting models of PSO-BPNN,SOA-BPNN,PSO-LSSVM,SOA-LSSVM,PSO-ELM and SOA-ELM are formed.Nine gas load prediction models,including three single models,were used to make simulation prediction combined with the historical data of gas load of Weinan City in2019-2020.Through the prediction of gas load for 15 consecutive days,the results showed that: Among the three single prediction models,the prediction accuracy of LSSVM model and ELM model is similar,and the average absolute percentage error is 21.5% and 25.7% lower than that of BPNN model respectively;The prediction accuracy of the six prediction models optimized by swarm intelligent algorithm has been improved,among which the prediction accuracy of PSO-BPNN and SOA-BPNN models has been improved by nearly 60% compared with BPNN,and the prediction accuracy of PSO-LSSVM and SOA-LSSVM models has been improved by nearly 60% compared with LSSVM.The prediction accuracy of PSO-ELM and SOA-ELM models is nearly 30% higher than that of ELM;In terms of average absolute error,root-mean-square error and average absolute percentage error,PSO-LSSVM and SOA-LSSVM models perform better than other models in forecasting accuracy and stability.According to the results of this paper,the optimized LSSVM model is more suitable for short-term urban gas load prediction.