摘要
集成学习作为一类组合优化方法,通过组合多个或多类基学习器以获得性能更优的组合模型,达到提高模型泛化能力和输出精度的目的。针对短时电力负荷曲线预测需求,构建一种基于集成思想的深度神经网络预测模型。首先,利用Bootstrap方法采样生成多个样本子集,并行训练相应的深度神经网络基学习器。在此基础上,将多个模型的预测输出进行平均加权作为最终预测输出。最后,利用某电厂采集的短时负荷数据及其影响因素数据对模型的预测效果进行仿真验证。仿真结果表明,基于集成思想的深度神经网络预测模型,其预测效果优于单一的深度神经网络模型,可用于对短时电力负荷的预测。
Ensemble learning,as a kind of combination optimization method,can obtain a combined model with better generalization performance and output accuracy by combining multiple or multi-class base learners.In order to meet the requirements of the short-term power load curve forecasting,a deep neural network forecasting model based on ensemble learning was established.First,the Bootstrap method was used to generate multiple sample subsets and train corresponding base learners of the deep neural network model in parallel.On this basis,the prediction outputs of multiple models were averagely weighted as the final prediction output.Finally,the short-term data collected from a power plant and its influencing factors were used to verify the effect of prediction of the model.The simulation results show that the prediction model based on the ensemble deep neural network is better than the single deep neural network model,which can be used to predict short-term power load.
作者
邓真平
DENG Zhenping(Chongqing Keyuan Energy Technology Development Co.,Ltd.,Chongqing 401147,P.R.China)
出处
《重庆电力高等专科学校学报》
2023年第1期5-10,共6页
Journal of Chongqing Electric Power College
关键词
集成学习
组合优化
基学习器
负荷曲线预测
深度神经网络
ensemble learning
combination optimization
base learner
load curve forecasting
deep neural network