基于改进BNN-LSTM的风电功率概率预测

Probabilistic Prediction of Wind Power Based on Improved BNN-LSTM

针对确定性的风电功率预测难以提供预测结果的波动区间和支撑风险决策的问题,以贝叶斯网络为基础,通过将先验分布置于LSTM网络层权重参数之上,构建了贝叶斯LSTM神经网络(BNN-LSTM)。以时间卷积神经网络(TCNN)处理风电功率预测的历史时序数据,提取时序数据的关联特征。使用互信息熵方法分析了风电功率的气象数据集,剔除关联性小的变量,对气象数据集进行降维处理。并采用嵌入(embedding)结构学习风电功率时间分类特征。随后将TCNN处理后的时序数据、降维后的气象数据以及时间分类特征数据一起送入BNN-LSTM预测模型,通过在某风电数据集不同算法的概率预测指标pinball损失和Winkler评分的对比验证,可知,本文所提方法能从可对风电功率波动做出较为准确的响应,预测效果更好。

Aiming at the problem that deterministic wind power prediction is difficult to provide fluctuation intervals of prediction results and support risky decision-making.Based on Bayesian network,Bayesian LSTM neural network(BNN-LSTM)is constructed by placing the prior distribution on top of the LSTM network layer weight parameters.The historical time series data of wind power prediction was processed by Time Convolutional Neural Network(TCNN)to extract the correlation features of the time series data.The meteorological dataset of wind power was analyzed using the mutual information entropy method,and the variables with small correlation were eliminated,and the meteorological dataset was processed by dimensionality reduction.And the embedding(Embedding)structure was used to learn the temporal classification features of wind power.Subsequently,the TCNN-processed time series data,the dimensionality reduced meteorological data and the temporal classification feature data are fed into the BNN-LSTM prediction model together,and through the comparison and verification of the probability prediction indexes of different algorithms of pinball loss and Winkler scores in a certain wind power dataset,it can be seen that the method proposed in this paper can respond to the fluctuation of wind power accurately from the wind power fluctuations,and the prediction effect can be better.