针对现有风力发电预测精度低的问题,提出一种改进的注意序列到序列门控循环单元(attention sequence-to-sequence gated recurrent unit,ASSGRU)的风力发电预测模型.该模型为典型的多输入多输出(multiple input multiple output,MIMO)模型,并基于注意力机制选择重要特征,从而提高风力发电预测的精度和稳定性.通过中国某电力公司发布的风电数据集对提出的预测模型进行验证.与自适应小波神经网络(adaptive wavelet neural network,AWNN)、K均值聚类的前馈神经网络(k-means-feedforward neural network,K-FNN)和长短时记忆(long short-term memory,LSTM)等模型相比,所提模型均方根误差变异系数(coefficient variation of root mean square error,CV-RMSE)、平均绝对百分比误差(mean absolute percent error,MAPE)以及归一化均方根误差(normalized root mean square error,NRMSE)等指标更优,从而验证了所提模型的可行性和有效性.该模型对混合智能电网智能化服务及新能源调度规划具有一定借鉴作用.
A Wind Power Forecasting Model Based on Attention Sequence-to-sequence Gated Recurrent Unit
Aiming at the problem of low accuracy in existing wind power generation prediction,an improved attention sequence to sequence gated cycle unit wind power generation prediction model is proposed.This model is a typical multiple input multiple output(MIMO)model,which selects important features based on attention mechanism to improve the accuracy and stability of wind power generation prediction.The proposed predictive model is validated using a wind power dataset released by a Chinese power company.Compared with the models such as adaptive wavelet neural network(AWNN),K-means feedforward neural network(K-FNN),and long short term memory(LSTM),the proposed model has superior coefficients of root mean square error(CV-RMSE),mean absolute percentage error(MAPE),and normalized root mean square error(NRMSE),thereafter the feasibility and effectiveness of the proposed model is verified.This model has certain reference for intelligent services of hybrid smart grids and new energy scheduling planning.
smart gridwind power forecastgated cycle unitfeature extractionattention mechanism
NETL
NSTL
万方数据
