基于改进特征模态分解的NRBO-LSTM短期电力负荷预测

Short-Term Power Load Forecasting Based on Improved Feature Mode Decomposition and NRBO-Optimized LSTM

摘要:

针对传统方法在电力负荷预测中存在的多时间尺度无法精准捕捉与深度模型超参数难以自适应优化等问题,构建了一种基于改进特征模态分解(FMD)与牛顿拉夫逊算法(NRBO)优化长短期记忆网络(LSTM)的短期负荷预测模型；该模型在特征模态分解模块中引入辅助白噪声以抑制模态混叠,并将传统峰度模态筛选改进为基于模态能量的筛选方法,以剔除低能量噪声模态并增强分解结果的稳定性,基于上述改进将原始负荷序列分解为一系列具有不同中心频率的本征模态函数,同时引入 NRBO 算法对 LSTM 的超参数进行全局寻优,提高模型的收敛速度与泛化性能；设置多个预测模型进行对比实验,实验结果表明,所提的FMD-NRBO-LSTM 模型的MAE、RMSE、MAPE分别为5.1949KW,6.2923KW和1.2747%,所有指标均优于对比模型,验证了所提模型的预测精度与泛化性能。

Abstract:

To address the limitations of traditional short-term load forecasting methods in accurately capturing multi-time-scale features and the difficulty of adaptively optimizing hyperparameters in deep learning models, a short-term load forecasting model based on improved Feature Mode Decomposition (FMD) and Newton–Raphson Based Optimization (NRBO) for Long Short-Term Memory (LSTM) networks is proposed. In the FMD module, auxiliary white noise is introduced to suppress mode mixing, and the traditional kurtosis-based mode selection is replaced with an energy-based selection mechanism to remove low-energy noise modes and enhance the stability of the decomposition. With these improvements, the original load sequence is decomposed into a set of intrinsic mode functions with different central frequencies. Meanwhile, NRBO is employed to globally optimize the LSTM hyperparameters, thereby improving the model’s convergence speed and generalization capability. Several forecasting models are constructed for comparison, and the experimental results show that the proposed FMD-NRBO-LSTM model achieves MAE, RMSE, and MAPE values of 5.1949 kW, 6.2923 kW, and 1.2747%, respectively, all outperforming the benchmark models and verifying the superior forecasting accuracy and generalization performance of the proposed approach.