基于PEEEMD-BiLSTM-XGboost光伏发电功率预测方法研究

Research on Power Prediction Method for Photovoltaic Power Generation Based on PEEEMD-BiLSTM-XGboost

摘要: 由于光伏功率数据的强不确定性，单一模型预测精度受到限制，提出多融合信号模态分解和双向长短期记忆网络(BiLSTM)、极端梯度提升(XGboost)组合模型的短期光伏发电功率预测方法。首先，为降低光伏功率信号的复杂性，通过自适应噪声完备集合经验模态分解(EEMD)、排列熵(PE)对光伏发电功率数据进行预处理，得到各模态分量；其次分析选取重要相关影响因素，构建BiLSTM-XGboost组合模型对光伏发电历史功率数据进行预测。最后，以某地光伏电站数据进行测试，仿真结果表明所提出的集成预测模型能够有效提高短期光伏功率预测精度，具有更少计算时间、较高的估计精度、算法稳定性高、鲁棒性强，并带来较强的实用价值。

Abstract: Due to the strong uncertainty of photovoltaic power data, the prediction accuracy of a single model is limited. A short-term photovoltaic power prediction method based on the combination model of multi fusion signal mode decomposition and bidirectional short-term memory network (BiLSTM) and extreme gradient boost (XGboost) is proposed. Firstly, in order to reduce the complexity of photovoltaic power signals, the photovoltaic power data is preprocessed using adaptive noise com-plete set empirical mode decomposition (EEMD) and permutation entropy (PE) to obtain various modal components; secondly, analyze and select important relevant influencing factors, and con-struct a BiLSTM-XGboost combination model to predict the historical power data of photovoltaic power generation. Finally, testing was conducted using data from a photovoltaic power plant in a certain area. The simulation results showed that the proposed integrated prediction model can ef-fectively improve the accuracy of short-term photovoltaic power prediction, with less computational time, higher estimation accuracy, high algorithm stability, strong robustness, and strong practical value.