基于麻雀搜索优化的Attention-GRU短期负荷预测方法

Short-term Load Forecasting Method Based on Sparrow Search Optimized Attention-GRU

针对短期电力负荷预测目前存在的难点与问题,提出了一种基于麻雀搜索优化的注意力门控循环单元预测方法。首先,应用注意力机制对输入序列进行权重分配;然后,输入门控循环单元组合网络对内部特征进行学习,并输出预测时间负荷值;最后,使用麻雀搜索算法对网络超参数进行组合优化,以验证集损失最小为目标函数获取最优化网络结构超参数。该方法实现了原始输入序列结构权重分配与组合网络超参数的最优化。算例分析表明,所提方法比传统预测模型精确度更高。

Aimed at the existing difficulties and problems in short-term power load forecasting,a forecasting method based on sparrow search algorithm(SSA)optimized Attention-gated recurrent unit(GRU)is proposed.First,the Attention mechanism is applied to assign weights to input sequences.Then,the updated sequences are imported into the combined GRU network to learn the internal characteristics,and the forecasted load values in the corresponding time are output.Finally,SSA is used to optimize the combination of network hyperparameters,and the optimal network structure hyperparameters are obtained through minimizing the loss of the validation set.The proposed method realizes structural weights distribution of the original input sequences and the optimization of combined network hyperparameters.The analysis of a numerical example shows that compared with the traditional forecasting models,the proposed method has a higher accuracy.