Day-ahead Strategic Marketing of Energy Prosumption: A Machine Learning Approach Based on Neural Networks

Abstract

In this paper, we propose a day-ahead strategic marketing method for multi-period energy markets using a machine learning approach based on neural networks. An aggregator, which has renewable energy generation devices, needs to schedule the energy production and consumption (prosumption) in a situation where the renewable power generation amount is not exactly predicted in day-ahead scheduling. If imbalance, defined as the difference between a day-ahead schedule and an actual prosumption profile, occurs, the aggregator is required to pay imbalance penalty costs. As a scheduling method to avoid paying imbalance penalty costs, we propose a scheduling model by machine learning based on the results of past transactions. In particular, the scheduling model is given as a neural network, which has an advantage in terms of computational costs compared to the kernel method. For developing a training algorithm, we show that the gradient of the profit function with respect to design parameters can be calculated as a solution to linear programming. Finally, we show the efficiency of the proposed method through a numerical example.