Short Term Load Forecasting using Multiple Linear Regression for Big Data

This paper presents short term load forecasting using multi-variable linear regression (MLR) for big data. Load forecasting is very important for planning, operation, resource scheduling and so on in power system. Total electric demand dynamically changes in a power system and mainly depends on temperature, humidity, wind speed, human nature, regular activities, events, etc. input variables. For the help of sensors and data science, enough historical and future input data with good accuracy are easily available. On the other hand, linear regression is a proven method, widely used in industries for forecasting. It is deterministic and robust. However, it is slow for big data because it needs large size matrix operations. In this paper, linear regression is formulated for small number of variables with big data and multi-core parallel processing is applied in all matrix operations that allow unlimited historical big data and unlimited scenarios in acceptable execution time limit. Mean absolute percent error is 3.99% of real field recorded data shown in Simulation and Result section.