针对目前光伏出力确定性预测缺乏考虑不确定性信息导致预测精度受限的问题,提出了一种计及组合区间不确定性分析的光伏出力动静态赋权搜索预测模型.首先分别构建卷积神经网络(convolutional neural network,CNN)-注意力机制(attention)-双向长短期记忆网络(bidirectional long-short term memory,BiLSTM)(CNN-Attention-BiLSTM)和自然梯度提升(natural gradient boosting,NGBoost)模型进行确定性预测;然后通过分析2个模型的确定性预测值和真实值的分布揭示了组合区间相对可靠性规律,基于该规律提出一种动静态赋权搜索预测模型,利用NGBoost模型进行不确定性预测,并根据所提供的概率分布信息,动态区分确定性预测结果的重要性,结合模型历史静态的指标评价信息,实现组合区间内搜索预测结果;最后采用澳大利亚沙漠知识太阳能中心和宁夏某光伏电站的数据集进行仿真研究,验证了该文方法的有效性和适用性,可为光伏出力确定性组合预测提供新的研究思路.
Dynamic and Static Weighted Searching Prediction Model for PV Outputs With Combined Interval Uncertainty Analysis
To address the problem that the prediction accuracy has been limited due to the lack of uncertainty information in the current deterministic prediction of photovoltaic(PV)outputs,a dynamic and static weighted searching prediction model for PV outputs that takes into account the uncertainty analysis of the combined interval is proposed.Firstly,a convolutional neural network(CNN)-attention-bidirectional long-short term memory(BiLSTM)(CNN-Attention-BiLSTM)model and a natural gradient boosting(NGBoost)model are constructed for the deterministic prediction.The distributions of the deterministic predicted values and the actual values of the two models are then analysed to reveal the law of relative reliability of the combined intervals.Based on this analysis,a dynamic and static weighted searching prediction model is proposed.The NGBoost model is used to perform the uncertainty prediction and dynamically differentiates the importance of the deterministic prediction results based on the information provided in the probability distribution.Combined with the model's historical static information on the evaluation of indicators,the prediction results within the combined interval are searching.Finally,a simulation study,using the datasets from the Desert Knowledge Solar Centre in Australia and a PV power plant in Ningxia Hui Autonomous Region,verifies the effectiveness and applicability of the methodology of this paper,which provides a new research idea for the deterministic portfolio prediction of the PV outputs.
PV output predictionCNN-Attention-BiLSTMNGBoostrelative reliability law of the combined intervaldynamic and static weighted searching prediction
万方数据
维普
