摘要
阐述了收敛交叉映射(CCM)方法及LSTM神经网络模型在用电量预测中的具体应用。针对城市用电量时间序列的非线性特点,结合动力系统理论,采用CCM方法研究用电量和温度、风速、相对湿度、降水之间的动力学因果关系,建立LSTM神经网络模型,并将该模型在H市用电量预测中进行了初步应用。研究结果表明,LSTM神经网络模型在城市用电量预测中年度预测相对误差小于月度预测相对误差,具有较高精度;改进的引入温度因素的LSTM神经网络模型,月度、年度预测相对误差均有改进,反映了运用CCM方法研究动力学因果关系的合理性以及LSTM神经网络模型在城市用电量预测中广泛的实用性。
This paper discusses the application of convergence cross mapping （CCM） method and LSTM neural networks model in the prediction of urban electricity consumption. According to the nonlinear time series characteristics of urban power consumption, combining with the theory of dynamic system, this paper adopts CCM method to investigate dynamical casual relations between power consumption and temperature, wind speed, relative humidity, precipitation. LSTM neural networks models are established to forecast H city electricity consumption. The results show that the relative error of annual prediction is better than that of monthly prediction and the relative error of monthly and annual prediction are improved by introducing temperature factor to LSTM neural networks model, which reflects the rationality of CCM method and practical value of LSTM neural networks model in forecasting urban electricity consumption.
出处
《电力大数据》
2017年第8期25-29,14,共6页
Power Systems and Big Data
