Short Term Distributed Load Forecasting Method Based On Big Data

Short Term Load Forecasting (STLF) is an essential part of Energy Management System. The forecasting error of STLF has an immediate impact on the secure economic operation of power system. In the context of smart grid, large-scale power grids are being formed and the cost of data collection is gradually reduced. Data types that can be obtained are increasingly abundant and the trend of big data is increasingly evident. In a power system covering large-scale geographical area, there are many differences between different regions for certain time periods in load characteristics that are affected by different external factors. A centralized forecasting model is difficult to accurately grasp load variation and weather diversity throughout the region, and it has a limited capacity to deal with big data for big analysis. Thus, a distributed load forecasting method based on big data is proposed in this paper. For the comprehensive utilization of big data and machine learning methods, the idea is to discuss the specific influence of exogenous factors on load patterns over a large region for improving the forecasting accuracy of STLF. Focusing on this, research work can be described as follows in this paper.First, load patterns analysis based on big data is proposed in this section. All kinds of information for different types of power users and local environment over the forecasting area should be collected as fully as possible within the bounds of cost in order to accurately grasp load variation and weather diversity throughout the region. This paper analyzes the year’s, month’s and day’s load patterns based on actual load data of Electric Power Company A and B in southern of china, respectively, and focuses on the analysis of the city’s summer load patterns. It follows that load patterns in different regions affected by different external factors are different and it is necessary to establish targeted load forecast models for different regions.Second, a distributed load forecasting method based on big data is proposed to improve the forecasting accuracy in this paper. In accordance with administrative division and the meteorological features, each220kV node is taken as a basic element to pre-divide the forecast area into subnets. Separate subnets are required to be re-partitioned/combined by similarity evaluation method for load curves. Based on this, regional forecasting models (ARIMA and BP neural network) are established point by point. According to the load forecasting result and the predicted load scale factor at various points, a system load forecasting model is established to forecast the aggregate system load. Finally, a case study on load forecasting is presented and compared with centralized load forecasting methods, which are based on actual load data of Electric Power Company in southern China in this paper. The proposed method is used for both weekday and weekend load forecasts to verify the validity and accuracy of the proposed method. Example analysis shows that the proposed method yields better performance in bulk power systems with big data. Load forecasting models are running in parallel greatly reducing the time required to forecast short term load of sepereate subnets. Thus the proposed method has feasibility in practical applications.