基于MPA-LSTM模型和Bootstrap方法的短期光伏功率区间预测

A short-term PV power interval forecasting based on MPA-LSTM network model and Bootstrap method

光伏发电功率的波动性和间歇性为电力系统调度管理带来巨大的挑战,精确的光伏功率区间预测是解决上述问题的一种有效途径。为此,本文提出了一种基于LSTM网络的新型短期光伏功率区间预测模型。采用MPA对LSTM网络的隐含层神经元数和训练批次数等超参数进行自动寻优,以克服随机选取LSTM模型参数过程中存在的盲目性、费时等问题;并将MPA-LSTM模型用于光伏功率点预测。然后,采用Bootstrap方法分析模型预测结果的误差分布,确定模型预测输出的区间范围。最后,通过对比仿真验证所提模型的有效性。结果表明:本文所提的MPA-LSTM模型均方误差的平均值为1.09%,优于SVM、LSTM、PSO-LSTM和MPA-SVM模型;Bootstrap方法能够准确地描述给定置信度水平下的光伏功率波动范围。

The fluctuation and intermittence of photovoltaic(PV) power bring great challenges to the power system scheduling and management. Accurate PV power interval prediction is an effective way to address these challenges. Therefore, this paper proposed a new short-term PV power interval prediction model based on LSTM network. In order to avoid the blindness and time wasting during random selecting LSTM model parameters, MPA was used to optimize the number of hidden layer neuron and batch size of LSTM network, and MPA-LSTM model was used for PV power point prediction. Then, Bootstrap method was used to analyze the error distribution and determine the interval prediction range of proposed model. Finally, the effectiveness of the proposed method was verified by comparison simulations. The results show that the proposed MPA-LSTM model has an average root mean square error of 1.09%, which is better than models established by SVM, LSTM, PSO-LSTM and MPA-SVM, respectively. Moreover, the Bootstrap method accurately describes the fluctuation range of PV power at given confidence levels.