考虑并行时序卷积的短期风电功率预测OA

To further improve the accuracy of wind power prediction,a short-term wind power prediction method considering parallel temporal convolution was proposed by effectively extracting and fusing features from power sequences and external variable sequences.Firstly,the external variables with high correlation with wind power were selected as external sequences with strong correlation with wind power using maximum mutual information coefficient(MIC),and the wind power sequences were used as internal sequences,which were inputted into two encoder modules in parallel for sequence coding and feature extraction.Then,the internal sequence was parallelly input into the global TCN and local TCN.The global TCN effectively extracted the long-term temporal dependencies of the power sequence by expanding the temporal receptive field,while the local TCN perceived the local temporal relationships through dilated causal convolution.Two encoders were used in parallel to extract features from external and internal sequences,while two TCN modules performed dual time scale temporal perception of internal sequences.Based on the fusion of dual cross attention layers,the temporal relationship between external and internal sequences was correlated,as well as the global dependencies and local temporal features of the internal sequence of wind power.Finally,based on actual wind farm data,experiments were conducted to demonstrate that the proposed method effectively improves the accuracy of wind power prediction.