基于超参数优化的电力负荷预测模型研究  

Research on power load forecasting model based on optimization of hyper-parameter

作　　者：张宜祥 张玲华[1,2] ZHANG Yixiang;ZHANG Linghua(College of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Jiangsu Engineering Research Center of Communication and Network Technology,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]南京邮电大学江苏省通信与网络技术工程研究中心,江苏南京210003

出　　处：《电子设计工程》2024年第4期37-42,共6页Electronic Design Engineering

基　　金：太阳能高效利用及储能运行控制湖北省重点实验室开放基金(HBSEES202113)。

摘　　要：电力负荷数据的多样性与复杂性,会导致负荷预测过程中出现超参数难以确定、拟合效果较差和预测精度不高等问题。针对以上问题,提出一种基于樽海鞘群算法的融入注意力机制的双向长短期记忆神经网络模型——SSA-AM-BiLSTM模型。该模型使用BiLSTM学习特征的内部变化规律,引入注意力机制为特征进行权重分配,并且利用樽海鞘群算法优化网络超参数。基于具体数据集进行的负荷预测仿真实验表明,相较于GRU、LSTM、AM-BiLSTM和PSO-AM-BiLSTM模型,所提出的SSA-AM-BiLSTM模型的MAPE分别减少了2.15%、1.93%、1.42%和0.45%,并且优化了拟合效果,显著提高了预测精度。Due to the diversity and complexity of power load data,there will be problems in the process of load forecasting,such as difficulty in determining hyper-parameters,poor fitting effect and low forecasting accuracy.To solve the above problems,a Bi-directional Long Short-Term Memory model incorporating attention mechanism based on Salp Swarm Algorithm(SSA-AM-BiLSTM model) is proposed.The model uses BiLSTM to learn the internal change rule of features,introduces the attention mechanism to distribute the weight of features,and uses Salp Swarm Algorithm to optimize the network hyper-parameters.The load forecasting simulation experiment based on specific data sets shows that,compared with GRU,LSTM,AM-BiLSTM and PSO-AM-BiLSTM models,the MAPE of the SSA-AM-BiLSTM model proposed in this paper is reduced by 2.15%,1.93%,1.42% and 0.45% respectively,and the fitting effect is optimized,which significantly improves the forecasting accuracy.

关 键 词：负荷预测 超参数 双向长短期记忆网络 注意力机制 樽海鞘群算法 

分 类 号：TN919.5[电子电信—通信与信息系统]