北方农村燃气日负荷预测的BP神经网络优化  

Optimization of BP Neural Network for Daily Gas Load Forecasting in Northern Rural Areas

作　　者：徐鹏[1,2] 杜景勃 张飞龙 刘伟 XU Peng;DU Jingbo;ZHANG Feilong;LIU Wei

机构地区：[1]北京建筑大学供热,供燃气,通风及空调工程北京市重点实验室,北京100044 [2]北京建筑大学燃气研究中心,北京100044 [3]北京市煤气热力工程设计院有限公司,北京100032

出　　处：《煤气与热力》2022年第8期1-8,共8页Gas & Heat

基　　金：2022年度北京建筑大学研究生创新项目(PG2022055);内蒙古自治区科技重大专项(2021ZD0038)。

摘　　要：在调研部分农村地区煤改气居民用户实际用气情况的基础上,建立了一种基于小波阈值去噪和采用遗传算法优化BP神经网络的短期燃气负荷预测模型(称为GA-BP神经网络预测模型)。以华北地区农村煤改气居民用户作为研究对象,对974户管道天然气居民用户2018年1月—2021年12月的日用气量进行采集。对采集数据进行小波阈值去噪处理,进行日负荷预测影响因素的选择及量化。将负荷预测影响因素和日负荷组成的数据集划分为训练集和测试集,对BP神经网络预测模型、GA-BP神经网络预测模型进行训练和测试。将两种模型的日负荷预测值与真实值进行对比,并将两种模型的评价指标进行对比,验证两种预测模型的准确性。研究结论如下。小波阈值去噪处理去噪效果良好,可用于燃气日负荷预测数据预处理。日平均温度、天气类型、节假日情况、前一日用气量、供暖情况是影响燃气日负荷预测的5个主要影响因素。有必要关注供暖过渡期的日负荷变化。这段时期温差变化大,用气情况复杂多变,对供气不确定性影响较大。对这部分的合理处理可以有效减小预测误差。日平均温度是影响农村居民用气非常重要的因素。遗传算法对BP神经网络的优化,可以很好地为网络初始权值和阈值的确定提供依据,优化了网络参数。相较于BP神经网络,GA-BP神经网络的稳定性提高,预测误差减小,预测精度提高。GA-BP神经网络预测模型应用于燃气日负荷预测是可行的。On the basis of the investigation of the actual gas consumption of residential users of coal-to-gas conversion in some rural areas,a short-term gas load forecasting model(called GA-BP neural network forecasting model)based on wavelet threshold denoising and using genetic algorithm to optimize BP neural network was established.Taking rural residential users of coal-to-gas conversion in North China as the research object,the daily gas consumption of 974 residential users of pipeline natural gas from January 2018 to December 2021 was collected.The wavelet threshold denoising was performed on the collected data to select and quantify the influencing factors of daily load forecasting.The data set composed of influencing factors of load forecasting and daily load was divided into the training set and the test set,and the BP neural network forecasting model and GA-BP neural network forecasting model were trained and tested.The daily load forecasting values of the two models were compared with the real value,and the evaluation indicators of the two models were compared to verify the accuracy of the two forecasting models.The research conclusions are as follows.Wavelet threshold denoising process has good denoising effect,and can be used for data preprocessing of gas daily load forecasting.Daily average temperature,weather type,holiday situation,gas consumption on the previous day and heating situation are the five main factors affecting the daily gas load forecasting.It is necessary to pay attention to the daily load change during the heating transition period.During this period,the temperature difference varies greatly,and the gas consumption is complex and changeable,which has a great impact on the uncertainty of gas supply.Reasonable treatment of this part can effectively reduce the forecasting error.The daily average temperature is a very important factor affecting the gas consumption of rural residents.The optimization of the BP neural network by the genetic algorithm can provide a good basis for the determination o

关 键 词：燃气负荷预测 短期预测 小波阈值去噪 遗传算法 BP神经网络 

分 类 号：TU996.8[建筑科学—供热、供燃气、通风及空调工程]