ICEEMDAN-ISSA-LSTM短期电力负荷预测

Short-term electric load forecasting based on ICEEMDAN-ISSA-LSTM

针对传统短期负荷预测方法误差大的问题,提出一种基于改进型自适应白噪声的完全集合经验模态分解(ICEEMDAN)和改进的麻雀搜索算法(ISSA)优化长短期记忆网络(LSTM)的短期负荷预测方法。首先,针对负荷序列波动性大,导致直接使用负荷数据进行预测难以获取内在特征的问题,运用ICEEMDAN方法将原始负荷序列进行分解,得到时间尺度各异的IMF分量;其次,针对LSTM模型参数较难选取的问题,采用ISSA对LSTM的超参数寻优,利用Fuch混沌映射、反向学习策略和自适应t变异改进麻雀算法,减小SSA陷入局部最优的风险,提高麻雀算法的寻优能力和收敛速度;最后,依据分解得到的各组数据特征,建立ISSA-LSTM模型并进行预测,再将各组分量的预测值进行叠加,得到最终的电力负荷预测结果。仿真结果表明:与其他预测模型相比,ICEEMDAN-ISSALSTM模型具有更高的短期电力负荷预测精度,其预测平均绝对误差为9.39 kW,均方根误差为11.47 kW,平均绝对百分比误差为0.19%。

Aiming at the problem of large errors in traditional short-term load forecasting methods,this paper proposes a combined prediction method based on an improved complete ensemble empirical mode decomposition with adaptive noise(ICEEMDAN)and an improved sparrow search algorithm(ISSA)optimizes long short-term memory(LSTM).First of all,in view of the problem that it is difficult to obtain intrinsic characteristics by directly using the load data for prediction due to the large fluctuation of the load sequence,the ICEEMDAN method is used to decompose the original load sequence to obtain IMF components with different time scales.Secondly,for the problem that the parameters of LSTM model are difficult to select,ISSA is used to optimize the hyperparameters of LSTM,and the sparrow algorithm is improved by using Fuch chaotic mapping,reverse learning strategy and adaptive t mutation to reduce the risk of SSA falling into local optimum.The optimization ability and convergence speed of the sparrow algorithm are improved;finally,according to the characteristics of each group of data obtained by decomposition,the ISSA-LSTM model is established and predicted,and the predicted values of each group are superimposed to obtain the final power load forecasting result.The simulation results show that,compared with other forecasting models,the ICEEMDAN-ISSA-LSTM model has higher short-term power load forecasting accuracy,its forecast mean absolute error is 9.39 kW,the root mean square error is 11.47 kW,and the mean absolute percentage error is 0.19%.