Research on Apache Spark Based Transformer Areas Load Forecasting

The massive data accumulated by the power company provides the basic data profile for load forecasting. In this paper, a dynamic Bayesian network is built as a load forecasting model of transformer areas. The parallel computing platform Apache Spark is used to calculate the parameters of the model based on large volume of transformers' historical data in parallel. Meanwhile, the Pregel computing model is used to parallelize the forward backward algorithm to realize the forecasting tasks. The experimental results show that the proposed transformer areas load forecasting technology based on distributed graph computing has high prediction accuracy and fast calculation speed.

I. Introduction

Short term load forecasting (STLF) is a crucial component of power generation schedule determination, gird operation modes adjustment and demand side management of the power system[1]–[2]. In particular, with the requirement of refined operation and management of distribution network, transformer areas load forecasting (TALF) is essential to evaluate the load development trend and state of the distribution network comprehensively.