基于SSA-LERNN的光伏出力超短期预测研究

Ultra-short-term Photovoltaic Power Prediction Method Based on SSA-LERNN

针对光伏发电功率模型预测准确度依赖数据质量的问题,提出一种结合奇异谱分析和局域情绪重构神经网络的超短期光伏发电功率组合预测方法。首先,利用奇异谱分析对实测光伏发电功率进行降噪处理,从复杂干扰信号中提取出平稳性好、可预测性强的有用信号。为解决奇异谱分析中参数选择主观性强、方法不系统的问题,采用基于搜索机制优化奇异谱分析参数的选取方法,以进一步提升降噪效果;然后,利用改进C-C法对降噪后的光伏发电功率时间序列进行混沌相空间重构,以深度挖掘数据隐含波动信息;最后,建立局域情绪重构神经网络预测模型捕捉相空间轨迹规律,超短期预测光伏出力。仿真结果表明,与局域情绪重构神经网络预测法以及边缘型人工情绪神经网络预测法相比,所提预测方法的预测准确性更高。

Aiming at the problem that the prediction accuracy of photovoltaic power model is depended on the quality of data, an ultra-short-term prediction method of the photovoltaic(PV) power based on SSA-LERNN is proposed. Firstly, the singular spectrum analysis(SSA) is used to reduce the noise of the measured PV power to extract useful signals with good stability and predictability from complex interference signals. Aiming at the shortcomings that the parameters selection in the SSA is too subjective and the method is not systematic, the selection of SSA parameters is optimized based on the search mechanism to further improve the effect of noise reduction. Secondly, the improved C-C method is used to reconstruct the chaotic phase space of the denoised PV power time series to deeply mine the implied fluctuated information. Finally, a prediction model based on the localized emotion reconstruction neural network(LERNN) is established to capture the law of phase space trajectory and perform ultra-short-term prediction. The simulation results show that the proposed prediction method has higher prediction accuracy than the LERNN prediction method and the limbic-based artificial emotional neural network(LiAENN) prediction method.