基于云计算和极限学习机的分布式电力负荷预测算法

A Distributed Load Forecasting Algorithm Based on Cloud Computing and Extreme Learning Machine

为了提高电力负荷预测精度,应对电力系统智能化所带来的数据海量化高维化带来的单机计算资源不足的挑战,提出了一种在线序列优化的极限学习机短期电力负荷预测模型。针对电力负荷数据特性,对极限学习机预测算法进行在线序列优化;引入分布式和multi-agent思想,提升负荷预测算法预测准确率;采用云计算的MapReduce编程框架对提出的算法模型进行并行化改进,提高其处理海量高维数据的能力。选用EUNITE提供的真实电力负荷数据进行算例分析,在32节点云计算集群上进行实验,结果表明基于该模型的负荷预测精度均优于传统支持向量回归预测算法和泛化神经网络预测算法,且提出的算法具有优异的并行性能。

To improve the accuracy of load forecasting and cope with the challenge of single computer＇s insufficient computing resource under massive and high-dimension data due to power grid intellectualization, a short-term distributed load forecasting model based on cloud computing and extreme learning machine is proposed. According to the features of load data the online sequential optimization of load forecasting algorithm based on extreme learning machine is performed; leading in the distributed and multi-agent thinking the accuracy of load forecasting algorithm is improved; adopting the MapReduce programming framework of cloud computing the parallelization of the model of the proposed algorithm is carried out to enhance its ability of dealing with massive and high-dimension data. The analysis of example based on real load data provided by EUNITE is conducted and corresponding experiments are done by 32-node cloud computing cluster, and experimental results show that the load forecasting accuracy by the proposed model is higher than the accuracy by traditional vector regression forecasting algorithm and the accuracy by generalized neural network based load forecasting algorithm, besides, the proposed forecasting algorithm possesses excellent parallel performance.