基于CNN-LSTM-XGBoost模型的超短期光伏功率预测

Ultra-short-term photovoltaic power prediction based on CNN-LSTM-XGBoost model

超短期光伏功率预测对电网的调度与运行具有重要意义。针对传统单一预测模型难以有效分析历史数据波动规律导致预测精度不高的问题,提出了一种CNN-LSTM-XGBoost的混合预测模型。剔除历史数据中的异常值后对数据进行归一化处理,并采用Pearson相关系数分析光伏发电功率与各气象因素的相关关系,选择相关系数较高的因素作为预测模型的输入特征。使用卷积神经网络(CNN)提取数据的空间特征,再经过长短期记忆(LSTM)网络提取时间特征,并采用误差倒数法将其与XGBoost模型并行拼接后完成对光伏发电功率的预测。以澳洲爱丽丝泉光伏系统为例进行仿真分析,仿真结果表明,相比于单一模型,融合模型在不同天气类型下具有更高的预测精度,验证了所提模型的有效性。

Ultra-short-term photovoltaic power prediction is of great significance to the dispatch and operation of the power grid.Aiming at the problem that the traditional single prediction model is difficult to effectively analyze the fluctuation rules of historical data,which leads to the poor prediction accuracy,a hybrid prediction model of CNNLSTM-XGBoost is proposed.After eliminating the abnormal values in the historical data,the data is normalized,and the Pearson correlation coefficient is adopted to analyze the correlation between photovoltaic power generation and each meteorological factors,and the factors with higher correlation coefficients are selected as input features of the prediction model.The CNN module is used to extract the spatial features of the data,then the LSTM module is used to extract the temporal features,and the reciprocal error method is used to splice it in parallel with the XGBoost model to complete the prediction of photovoltaic power generation.Finally,the Australian Alice Springs photovoltaic system is taken as an example for simulation analysis,the simulative results show that compared with a single model,the fusion model has higher prediction accuracy under different weather types,which verifies the effectiveness of the proposed model.