Probabilistic Prediction of Wind Power Based on Improved BNN-LSTM
Aiming at the problem that deterministic wind power prediction is difficult to provide fluctuation intervals of predic-tion 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 fea-tures 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 dimensional-ity 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 classifi-cation 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.
万方数据
维普
