摘要
针对传统电力负荷预测算法存在模型训练速度慢、预测效果差等问题,提出基于深度学习的并行负荷预测方法。该方法基于MapReduce并行计算框架,结合深度信念网络模型,以历史负荷信息与天气信息为样本数据进行并行化训练,并通过训练模型预测负荷值。经实验验证,本文的预测方法预测的电力负荷值与实际值的平均均方根误差为2.86%,预测精度高于传统方法,且有效减少了训练与预测时间,能适应大规模电力数据的预测要求。
For the problems of slow training speed and poor prediction effect of traditional power load forecasting algorithm, a parallel load forecasting method based on depth learning is proposed. The method combines the MapReduce parallel computing framework and the depth belief network model. Parallel training is carried out by using historical load information and weather information as sample data, and the load value is forecasted by training model. Verified by the experiment, based on the deep learning in parallel with the power load forecasting method predicted value and the actual value of mean root mean square error is 2.86%, the prediction accuracy is generally higher than that of traditional method, and can effectively reduce the training and prediction of time, can adapt to the prediction of large-scale electric power data.
出处
《自动化与信息工程》
2017年第4期26-30,共5页
Automation & Information Engineering
基金
广东省科技计划项目(2017B090901041)
关键词
负荷预测
深度学习
并行计算
置信度网络
无监督学习
Load Forecasting
Deep Learning
Parallel Computing
Deep Belief Network
Unsupervised Learning
