Customer Energy Flexibility Forecasting with Different Machine Learning Models

Abstract

Most concerns about the sustainability of modern society are centred around energy issues. To address these, the power system is transitioning towards a more intelligent, flexible, and interactive system with higher penetration of renewable energy generation. A key enabler of this transition is energy flexibility, which focuses on the demand response of consumers at different times. It is the traceability of the energy flow among individual energy consumers. This master’s thesis focuses on the demand response-based flexibility of the consumers at different time scales. Accurate load forecasting of individual customers is increasingly vital for future grid planning and operation. The main objective of this thesis was to identify the most flexible consumer among seven consumers and to identify the most effective machine learning (ML) model for forecasting energy flexibility across time horizons and time scales. This study investigates how different machine learning models—such as recurrent neural networks (RNN), gradient boosting, linear regression, and long short-term memory (LSTM)—can be used in the energy forecasting process. This study utilised a total of 11 distinct machine-learning models. The experimental findings clearly demonstrate that the proposed model is superior to the others in terms of accuracy in predicting consumption, as measured by the root mean square error (RMSE).

In addition, the model also evaluates the degree of flexibility of these households to adjust or reduce energy consumption in response to price fluctuations. This study proposes the implementation of a three-data model strategy to effectively manage load flexibility forecasting (Customer profit maximization). This technique aims to provide flexibility services, like congestion management, peak shaving to the local distribution system operator (DSO) and integrate renewable energy sources by leveraging several features like advance forecasting and analytics, demand response.