摘要
Abstract
Taking the daily gas consumption in City Z as the research object,a neural network-based predic-tion model was developed.The model was trained us-ing existing data,and the prediction results were evalu-ated.Before load prediction,the original data need to be preprocessed,including filling in missing values,detecting outliers,filling in outliers,standardizing data,and quantifying influencing factors.Five types of neural networks were selected,namely BP neural net-work,LSTM neural network,GRU neural network,IPSO-GRU neural network,and KPCA-IPSO-GRU neu-ral network.The IPSO-GRU neural network has four in-ertia weight change strategies,including the parablic method,random method,Sigmoid method,and sine method.The daily gas consumption data of City Z from January 1,2018 to December 31,2020 were collected,and the gas consumption data were divided into a train-ing set and a test set,which were used to train and evaluate the neural network prediction model respec-tively.After quantifying the influencing factors of daily gas consumption,they were used as input feature pa-rameters of the neural network prediction model,mainly including temperature,wind force level,weather type,weekday type,holiday type,and daily gas consumption in the previous 3 consecutive days.The prediction results of the five neural network prediction models were compared.The results show that the KPCA-IPSO-GRU neural network prediction model has the smallest mean absolute percentage error and root mean square error,demonstrating the highest pre-diction accuracy.关键词
燃气负荷预测/供暖期/神经网络/优化算法Key words
gas load prediction/heating period/neural networks/optimization algorithm分类
土木建筑
