Forecasting of 10-Second Power Demand of Highly Variable Loads for Microgrid Operation Control

This paper addresses very short-term (10 s) forecasting of power demand highly variable loads. The main purpose this study is to develop methods useful for type forecast. We have completed a comprehensive using two different time series, which are difficult access in practice, 10 s characterized by big dynamics load changes. an emerging and promising research topic, yet be more widely recognized the community. problem particularly important microgrids, i.e., small energy micro-systems. Power forecasting, like renewable generation, key importance, especially island mode operation microgrids. due necessity ensuring reliable supplies consumers. Inaccurate forecasts can cause improper microgrids or increase costs/decrease profits electricity market. presents detailed statistical analysis data sample low voltage loads large variability, located sewage treatment plant. experience authors that such Special attention has been paid methods, applied forecast, selection explanatory variables these methods. Some ensemble models (eight selected belonging following classes methods: random forest regression, gradient boosted trees, weighted averaging ensemble, machine learning) proposed scope choice sets constituting set unique developed study. obtained presented analyzed detail. Moreover, qualitative carried out. analyze various measures quality. think some practical applications, microgrid control, because their accuracy stability. usefulness independent series essential, valuable element Thanks this, reliability conclusions concerning preferred considerably increased.