LSTM模型在电力系统负荷预测中的应用

Application for LSTM model in power system load forecasting

为解决数据随时间变化的电力系统短期负荷预测问题,本文阐述、采用和总结线性回归,普通BP神经网络,GRU神经网络,LSTM神经网络在电力系统短期负荷预测中应用的理论基础和方法,分别针对神经网络的多种预测方法进行归纳总结。LSTM即长短期记忆网络,拥有门机制,可以选择性的遗忘和记忆过去的信息,特别能记忆一个序列时间段的信息。本文以某城市2019年全年的电力负荷数据为基础,该数据集包含城市350天的每天48个时间点的16800条数据,通过python绘图,解决对比不同算法之间的性能差异问题,最终得到LSTM在电力系统短期负荷预测上性能最优的结果。通过结果分析,得出利用电力负荷时间序列具有历史依赖性的特点,使用LSTM深度神经网络与传统神经网络进行了对比预测,证明了深度学习在电力负荷预测领域的适用性和精确性。

In order to solve the problem of power system short-term load forecasting with data varying with time,this paper states,sorts out and summarizes the linear regression method,common BP neural network,GRU neural network,LSTM neural network in theoretical basis and methods of application to short-term load forecasting for power system,various prediction methods of neural network are summarized respectively.LSTM is the long and short term memory network,and it is one of the deep learning methods.LSTM can own the gate mechanism,and selectively forget and remember information from the past,especially remember information about a sequence of time periods.There are a large number of power system load data based on one city in 2019,and the data set contains 16,800 pieces of data at 48 time points per day over 350 days in the city.Through python,the performance differences between different algorithms are compared,and the optimal performance of LSTM is obtained in power system short-term load forecasting finally.By analyzing the results,It is concluded that the use of power load time series is historically dependent,the LSTM deep neural network is compared with the traditional neural network,Which proves the applicability and accuracy of deep learning in power load forecasting.