基于BPNN-EMD-LSTM组合模型的城市短期燃气负荷预测

Short-term Forecast of Urban Natural Gas Load Based on BPNN-EMD-LSTM Combined Model

城市短期燃气负荷具有高随机性和复杂性特征,利用单一的模型难以做出准确预测。以某城市民用类燃气日负荷为研究对象,在分析该市两年多燃气日负荷特征的基础上,建立了基于BP神经网络(BPNN)-经验模态分解(EMD)-长短期记忆(LSTM)神经网络的组合预测模型,对该市短期燃气日负荷进行了预测。首先通过BPNN模型学习温度、日期属性影响下燃气负荷的主要特征,增长趋势等次要特征则体现在BPNN模型预测产生的残差中;然后采用EMD算法分解残差得到有限个本征模函数(IMF),并利用LSTM模型学习各IMF分量的短期时序规律,将各IMF分量的预测值相加得到残差预测值;最后将两部分预测值代数相加得到最终的预测结果。实证结果表明:与单一的LSTM模型和BPNN-LSTM模型相比,该组合预测模型半月步长的平均绝对误差为3.4%,预测精度更高,是一种更为有效的城市短期燃气负荷预测方法。

Because of the strong randomness and complexity of short-term natural gas load,it is difficult for a single model to make an accurate forecast.To address this issue,this paper proposes a combine model based on BP Neural Network(BPNN),Empirical Mode Decomposition(EMD)and Long-Short-Term Memory(LSTM)neural network based on the analysis of the natural gas load characteristics of a city during two yeares.Firstly,the paper uses BPNN to learn the main characteristics of natural gas load under the influence of temperature and date properties,while the secondary features such as growth trends are reflected in the residuals produced by BP neural network forecast.Then,the paper adopts the the EMD algorithm to decompose the residual to obtain a finite number of IMF components,uses the LSTM model to learn the short-term timing regular of each IMF component,and adds the forecasted results of every IMF to get the real residuals’forecasted results.Finally,the paper adds the two-part forecasted algebra to get the final forecasted values.The experimental results show that compared with a single LSTM or BPNN-LSTM model respectively,the combined model has a higher accuracy with the MAPE of the 15-day forecast reaching 3.4%,which is more effective for short-term natural gas load forecast.