为了解决单个神经网络预测的局限性和时间序列的波动性,提出了一种奇异谱分析(singular spectrum analy-sis,SSA)和Stacking框架相结合的短期负荷预测方法.利用随机森林筛选出与历史负荷相关性强烈的特征因素,采用SSA为负荷数据降噪,简化模型计算过程;基于Stacking框架,结合长短期记忆(long and short-term memory,LSTM)-自注意力机制(self-attention mechanism,SA)、径向基(radial base functions,RBF)神经网络和线性回归方法集成新的组合模型,同时利用交叉验证方法避免模型过拟合;选取PJM和澳大利亚电力负荷数据集进行验证.仿真结果表明,与其他模型比较,所提模型预测精度高.
Stacking fusion based LSTM-SA-RBF short-term load forecasting
To avoid the limitations of individual neural network forecasting and the volatility of time series, this paper proposes a short-term load forecasting model combining singular spectrum analysis (SSA) and stacking framework.First, the strong correlation characteristic factors with historical load are screened by random forest and SSA to reduce noise for load data and simplify the model calculation process.Second, based on the stacking framework, a new combined model is integrated with long- and short-term memory (LSTM) self-attention mechanism(SA) , radial base functions (RBF) neural network and linear regression methods, and cross-validation is employed to avoid model over-fitting.Finally, the PJM and Australian electricity load datasets are adopted for validation.Our simulation results show the proposed model achieves higher prediction accuracy compared with other models.
国家科技期刊平台
NSTL
万方数据
