基于分布式图计算的台区负荷预测技术研究

Research on Load Forecasting Technology of Transformer Areas Based on Distributed Graph Computing

随着智能电网数据采集技术的发展,电网公司积累的海量数据为台区负荷预测提供了数据基础。该文重点研究基于Apache Spark计算框架的台区短期负荷预测技术。首先,建立了包含温度、湿度、日类型等因素及负荷时间序列的动态贝叶斯网络,作为台区负荷预测模型。其次,为解决基于海量数据的模型参数计算及大规模配变负荷预测的并行化问题,搭建了基于Spark GraphX并行图计算组件的分布式计算平台,利用GraphX的Vertex Cut和并行边检索快速计算概率矩阵,以Pregel计算模型并行处理前向-后向算法,实现预测方法的分布式集群计算,保证全局预测任务的并行化。最后,通过实验结果表明,基于分布式图计算的台区负荷预测技术具有较高的预测精度和较快的计算速度,是一种高效可行的电力大数据分析技术。

With the development of smart grid and data acquisition and monitoring technology, the massive data accumulated by the power company provides the data base for load forecasting. This paper focused on the short-term load technology of transformer district based on Apache Spark computing framework. At first, a dynamic Bayesian network was established as a load forecasting model of transformer areas in this paper, which included the causal relationship between temperature, humidity, day type and other factors with load and load time series. Next, in order to solve the problem of model parameter calculation based on massive data and the parallelization of lots of transformers＇ load forecasting, a distributed graph computing platform based on Spark GraphX parallel computing component was built. Then combining Vertex cut and parallel path search of GraphX to extracting probabilistic information fast. Pregel computing model used to parallelizing forward backward algorithm to realize the distributed cluster computing of the prediction method and ensure the parallelization of global prediction tasks. The experimental results show that load forecasting technology of transformer district based on distributed graph computing has high prediction accuracy and fast calculation speed and is an efficient and feasible technique for large data analysis.