Time Series Forecasting Based Day-Ahead Energy Trading in Microgrids: Mathematical Analysis and Simulation

Gyohun Jeong,Sangdon Park,Ganguk Hwang

doi:10.1109/access.2020.2985258

Abstract

In this paper, we propose a periodic energy trading system in microgrids based on day-ahead forecasting of energy generation and consumption. In the proposed model, each noncooperative prosumer calculates her reward function under her energy change forecasting based on Gaussian process regression and determines her optimal action. Then, the system establishes the equilibrium trading price when all prosumers execute their optimal actions simultaneously. We prove the existence of the equilibrium trading price and establish an algorithm that leads to the equilibrium. Our numerical example shows that the proposed system outperforms its previous model.

Highlights

As power consumption continues to increase globally, power grids have begun to transform into a new form
The existing power grids consist of large-scale power generations based on fossil fuels and nuclear power as main energy sources and unidirectional energy transmission from centralized power plants to energy consumers
We propose a new energy trading system among prosumers with the help of time series analysis and prediction of prosumer energy level changes

Summary

INTRODUCTION

As power consumption continues to increase globally, power grids have begun to transform into a new form. A noncooperative energy trading model based on the distributional forecasting is suggested in [36] In this model, each prosumer splits each period into much shorter time durations and stochastically predict the energy level changes during all time durations. Note that, when the energy level goes below 0 or goes beyond Emax , an immediate trading occurs with the macrogrid to prevent a loss from blackout or a waste of overflowing energy In this algorithm, buyk (a|h) and sellk (a|h) are defined by the amount of energy to buy/sell from/to the macrogrid in the kth duration in the period for action a and sample energy change h. If the difference becomes 0, the two energy level processes coincide until the end of the period

REWARD FUNCTION

EQUILIBRIUM TRADING PRICE

NUMERICAL RESULTS

GAUSSIAN PROCESS REGRESSION

CONCLUSIONS