基于改进LSTM的电力设备状态融合预测模型

针对电力大数据存在数据随机缺失进而降低长短期记忆模型（Long Short-term Memory, LSTM）预测准确率的问题，本文提出了一种基于改进LSTM的电力设备状态融合预测模型。该模型首先对状态数据进行缺值检测和平稳分析，根据历史数据利用差分整合移动平均自回归模型（Autoregressive Integrated Moving Average model, ARIMA）对缺失的数值进行预测，并将预测的数值补充至相应的缺失位置；然后将新的完整数据输入到ARIAM模型和改进LSTM模型中以获取两种预测值；最后根据改进LSTM模型的学习准确率和ARIAM模型的拟合度对预测值进行权重分配，并在此基础上进行状态趋势融合预测。为了验证本文模型的普适性和预估准确性，选择电力负荷数据开展实验，结果表明：基于改进LSTM的电力设备状态融合预测模型在数据完整情况下的预测准确率比ARIAM和LSTM分别提高了52%和25% ，在数据缺失情况下的预测准确率分别提高了44%和57%。

State fusion prediction model of power equipment based on improved LSTM

Aiming at the problem of random missing data of electric power big data, which reduces the accuracy of Long Short-term Memory prediction, this paper proposes a power state fusion prediction model based on improved LSTM. The model first performs missing value detection and stationary analysis of state data, and uses the differential integrated moving average autoregressive model to predict missing values based on historical data, and supplements the predicted values to the corresponding ones. The missing position; then the new complete data is input into the ARIAM model and the improved LSTM model to obtain two predicted values; finally, the predicted values are weighted according to the learning accuracy of the improved LSTM model and the fit of the ARIAM model, and On this basis, state and trend fusion is performed. In order to verify the universality and accuracy of the model in this paper, electric load data was selected to conduct experiments. The results show that the prediction accuracy rate of the power equipment state fusion prediction model based on improved LSTM is improved compared with ARIAM and LSTM, respectively. 52% and 25%, the prediction accuracy in the absence of data has improved by 44% and 57%, respectively.