Spark云平台电力负载预测系统的设计与实现

Design and Implementation of Spark Cloud Platform Power Load Forecasting System

基于Spark云计算大数据处理平台,设计了一种新的电力负荷预测方法。该方法以电力负载的一阶梯梯度特征表示电力负载的变化特征;采用非监督机器学习方法避免了人工划分数据为上升与下降两个类别产生的人工成本问腿;并用所划分的两类数据集训练分类神经网络和回归神经网络模型,使得该神经网络模型可以自适的将当前的电力负荷数据调度到相应的回归神经网络模型中进行分类预测。通过实际的电力负荷数据进行验证,实验结果表明,所提出的方法不仅能有效地将电力负载数据自适应进行分类,同时在多种评价指标上均优于其他传统的电力负载预测模型。

Based on Spark cloud computing big data processing platform,a new power load forecasting method is designed.The method uses the first gradient feature of power load to represent the changing feature of power load.Then,the unsupervised machine learning method is used to avoid the problem of labor cost caused by manually dividing data into ascending and descending categories.The neural network model can automatically dispatch the current power load data to the corresponding recurrent neural network model for classification prediction.The experimental results show that the proposed method can not only effectively classify the power load data adaptively,but also outperform other traditional power load forecasting models in a variety of evaluation indicators.