基于极值优化的LSTM短期风速预测

Research on LSTM Short-Term Wind Speed Prediction Based on Extreme Value Optimization

针对风电机组组网运行存在的功率波动性和随机性的问题,提出一种基于极值优化的长短时记忆神经网络,以提高风速预测的精度和风电机组运行的稳定性。首先对数据进行标准化处理,以得到较好的拟合并防止训练发散。利用LSTM的非线性拟合能力,将数据代入LSTM网络当中进行预测。最后针对LSTM网络所预测的风速序列在局部极值点处与实际值差别较大的情况,使用极值优化模型分别对部分极大值点和极小值点进行算术平均和几何平均。实验结果表明,相比于单纯的LSTM模型,本文提出的基于极值优化的LSTM模型在风速预测上具有更好的精度。

Aiming at the problems of power fluctuation and randomness in wind turbine network operation, a long-short-term memory neural network based on extreme value optimization is proposed to improve the accuracy of wind speed prediction and the stability of wind turbine operation. The data is first normalized to get a better fit and prevent training divergence. Using the nonlinear fitting ability of LSTM, the data is substituted into the LSTM network for prediction. Finally, in view of the large difference between the wind speed sequence predicted by the LSTM network and the actual value at the local extreme points, the extreme value optimization model is used to perform arithmetic mean and geometric mean on some maximum and minimum points, respectively. The experimental results show that, compared with the pure LSTM model, the LSTM model based on extreme value optimization proposed in this paper has better accuracy in wind speed prediction.