基于时空信息组合的分布式光伏功率预测方法研究

Research on distributed photovoltaic power prediction method based on combination of spatiotemporal information

为保障电网安全稳定运行,提高光伏电站经济效益,需要进一步提升光伏功率预测的准确性。为此提出一种基于时空信息组合的分布式光伏功率预测方法。首先,基于极度梯度提升-长短时记忆神经网络(XGBoost-LSTM)集成模型,利用光伏历史数据时间序列进行预测;然后,基于最小二乘支持向量机(LSSVM)模型,利用光伏电站间的空间相关性进行预测;最后,基于信息熵的基本原理,结合多误差评价角度改进的信息熵确定2种单一预测方法的权值,构建时空信息组合预测模型。研究结果表明:基于多误差评价标准的信息熵权的XGBoost-LSTM+LSSVM模型的平均绝对误差与基于交叉熵的组合模型和基于信息熵的组合模型相比,分别下降1.6%、8.3%;结合单一预测模型的优势,组合模型可降低预测误差,基于多误差评价标准的信息熵组合模型具有更高的鲁棒性与综合性能,可提升分布式光伏功率预测的准确性。

To ensure the stability of power grid operation and improve the economic benefits of photovoltaic power plants,the forecast accuracy of distributed photovoltaic power output needs to be further improved.Thus,a distributed photovoltaic power prediction method based on the combination of spatiotemporal information is proposed.Firstly,on the basis of XGBoost-LSTM integrated model,the time series of photovoltaic historical data areused to carry out theprediction.Then,by using theLSSVM model,the spatial correlation between photovoltaic power plants is used to conduct theprediction.Finally,according to the basisc principle of information entropy,and combing with the information entropy improved by multi-error evaluationcriteria,the weight of the two prediction methods is determined,and the combined prediction model is constructedto realize spatiotemporal information combination prediction.The research result shows that,the mean absolute error of the XGBoost-LSTM+LSSVM model based on the information entropy improved by multi-error evaluation criteria decreases by 1.6% and 8.3% respectively,compared with the combined model based on cross entropy and the model based on information entropy.Combining with the advantages of a single forecasting model,the combined models can reduce the forecasting error.The information entropy combination model based on multi-error evaluation criteria has higher robustness and comprehensive performance,and canimprove the accuracy of distributed photovoltaic power prediction.