基于电力计量大数据的区域性短期负荷预测算法设计

Design of regional short-term load forecasting algorithm based on power metering big data

针对单一算法在对电力负荷进行预测时存在的局限性,同时为了提高短期负荷的预测精度,文中提出了一种基于CEEMDAN分解的门控循环单元和小波神经网络相结合的短期负荷预测算法,并构建了SSA-GRU&WNN预测模型。该模型采用CEEMDAN算法分解负荷数据,以降低数据的波动性与不确定性,利用样本熵算法对分解得到的分量进行评估及分组。同时分别利用GRU和WNN对两组分量加以预测,且引入麻雀搜索算法实现对二者超参数的优化。实验结果表明,所提算法的MAE、RMSE和MAPE分别为66.54 MW、58.62 MW及67.8%,相比传统单一负荷预测算法的误差更小、时间成本也更低。

In view of the limitations of a single algorithm in power load forecasting,and in order to improve the accuracy of short-term load forecasting,this paper proposes a short-term load forecasting algorithm based on CEEMDAN decomposition of gated cycle unit and wavelet neural network,and constructs SSA-GRU&WNN forecasting model.The CEEMDAN algorithm is used to decompose the load data to reduce the volatility and uncertainty of the load data,and the sample entropy algorithm is used to evaluate and group the decomposed components.At the same time,GRU and WNN are respectively used to predict the two groups of components,and sparrow search algorithm is introduced to optimize their super parameters.The experimental results show that the MAE,RMSE and MAPE of the proposed algorithm are 66.54 MW,58.62 MW and 67.8%respectively,which is less error and lower time cost than the traditional single load forecasting algorithm.