基于LSTM-RNN的电网雷电智能预警模型设计

Design of power grid lightning intelligent early warning model based on LSTM-RNN

为了提升电力网络的安全运行水平,对电网的雷电智能预警方法进行了研究。对历史雷电数据加以提取和分析,从而建立起完备的历史数据集。基于历史数据与数学建模的思想,将雷电预警问题抽象为时间序列的处理问题,并引入循环神经网络(RNN)完成对问题的求解。同时在RNN的应用过程中,采用长短期记忆模型(LSTM)以及Dropout机制解决了传统模型中的过拟合现象。LSTM模型使用输入门、遗忘门和输出门对信号的传递流程进行了修正,且替代了RNN隐藏层中的普通节点,确保了误差函数梯度在训练过程中能够严格遵循时间步骤。而Dropout机制则将部分神经元置为失效状态,避免了其在训练中的相互适应。在山东地区的雷电数据集上进行的仿真实验结果表明,RNN网络在引入LSTM与Dropout机制后,模型的预测精度提升了4.61%。

ed as the processing problem of time series,and the Recurrent Neural Network(RNN)is introduced to solve the problem.At the same time,in the application of RNN,the Long Short-Term Memory(LSTM)and Dropout mechanism are used to solve the over fitting phenomenon in the traditional model.LSTM model uses input gate,forgetting gate and output gate to modify the signal transmission process,and replaces the ordinary nodes in the RNN hidden layer to ensure that the error function gradient can strictly follow the time steps in the training process.The Dropout mechanism puts some neurons in the failure state to avoid their mutual adaptation in training.The simulation results on the lightning data set in Shandong Province show that the prediction accuracy of the model is improved by 4.61% after introducing LSTM and Dropout mechanism into RNN network.