基于ISSD-GRU模型的台区售电量预测方法

ISSD-GRU Model-based Electricity Sale Prediction

针对台区售电量不确定影响因素多、预测精度不高的问题,提出了一种基于改进ISSD优化GRU神经网络的台区售电量预测方法,利用反向学习提高SSD算法对最优参数的搜索效率。以某地台区历史售电量、温度、工作日类型和节假日类型作为影响因素对GRU模型进行训练,利用ISSD算法实现对GRU隐藏层神经元个数和学习率超参数的寻优,构建用于台区售电量预测的ISSD-GRU模型。算例分析表明,ISSD-GRU模型在台区售电量预测结果上精度更高。

The present work made an attempt on addressing problems of excessive uncertainties-induced low accuracy in station area electricity sale prediction,and proposed a prediction method based on modified ISSD optimized GRU neural network which uses reverse learning to improve searching efficiency of the SSD algorithm for optimal parameters.The GRU model was trained with influences data such as historical electricity sales,temperature,working day type and holiday type of a certain distribution station,and ISSD optimization algorithm was used to realize optimal searching of number of hidden layer neurons and hyperparameters of learning rate,thereby an ISSD-GRU model for electricity sale prediction was established.The proposed ISSD-GRU model was indicated by case analysis to have improved accuracy for predicting station area electricity sales.