基于LTC-RNN模型的中长期电力负荷预测方法  

Medium-and Long-Term Power Load Forecasting Method Based on LTC-RNN Model

作　　者：邓斌 张楠 王江 葛磊蛟 Deng Bin;Zhang Nan;Wang Jiang;Ge Leijiao(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津300072

出　　处：《天津大学学报（自然科学与工程技术版）》2022年第10期1026-1033,共8页Journal of Tianjin University：Science and Technology

基　　金：国家自然科学基金资助项目(51807134).

摘　　要：针对中长期电力负荷预测受限于天气、温度、节假日等多重不确定性因素影响而精度不高的难题,本文采用一种基于液体时间常数的递归神经网络,实现了中长期电力负荷的精准预测,为电力系统规划设计奠定较好基础.基于液体时间常数的递归神经网络使用膜积分器改进了神经元的状态方程,并使用半隐式欧拉算法完成对神经元状态迭代更新,解决递归神经网络的自我修正能力弱和易陷入局部最优的问题.本文以天气、温度、节假日等外在影响因素作为网络输入,构建了以递归神经网络为基础架构的中长期电力负荷预测模型,并通过隐藏层的扩展计算获得输入-输出的映射关系,以通过时间的反向传播算法实现对模型的训练,完成对神经网络参数的优化.实验中,以北方某地区的实际电网数据为例,验证该预测方法的准确性与高效性.研究结果发现,基于液体时间常数的递归神经网络对中长期电力负荷的预测准确率可达到95.3%.本模型相较于长短期记忆和连续时间递归神经网络具有更稳定的训练结果与更高的预测准确度.Owing to various uncertainties,such as weather,temperature,and holidays,predicting long-term power load remains difficult.This paper introduces a liquid time constant-recurrent neural network to accurately predict medium-and long-term power load,which is the fundamental work for power system planning and design.The liquid time constant-recurrent neural network improves the equation of neural states and uses a semi-implicit Euler algorithm to complete the iterative updating of the neural states,which solves the problem that the recurrent neural network has poor self-correction ability and is prone to falling into local optimum.A recurrent neural network-based model for forecasting medium-and long-term power load is constructed using network inputs such as weather,temperature,and holidays.The mapping from input to output is then obtained using extended computation in hidden layers,and the parameters of the neural network are optimized using a time algorithm via backpropagation.Actual power grid data from a specific area in North China are used to validate the accuracy and efficiency of the proposed method.The experimental results show that the liquid time of the constant-recurrent neural network can reach 95.3%accuracy for predicting medium-and long-term power load.With better stability during training and higher accuracy in the prediction phase,this model outperforms the long short-term memory algorithm and continuous-time recurrent neural network.

关 键 词：中长期电力负荷预测 递归神经网络 电力系统 

分 类 号：TK715[动力工程及工程热物理—流体机械及工程]