Neural network prediction has gradually become the prevailing hotspot in the field of electric load prediction in recent years,but there is still a lack of mining the dependency relationship among weather features,and no scientific guid-ance for selecting predictive time scale.The present work made a preliminary attempt to establish a multi-weather feature-based short-term load prediction model in view of the characteristics of large fluctuations and strong stochasticity of elec-tric load.The model was designed to select the optimal prediction scale of 7 days and employ a modified RNN-based LSTM deep neural network for the deep mining of the relationship between load data and various weather characteristic data.In an experimental predictive verification using actual daily data of weather and load carried out on a time scale of 7 days,the proposed model achieved a prediction accuracy as high as 94.131％.
关键词
电力负荷预测/天气特征/深度挖掘/最适预测尺度
Key words
electric load forecasting/weather characteristic/deep mining/optimal predictive scale
维普
万方数据