Machine-learning based hybrid demand-side controller for high-rise office buildings with high energy flexibilities

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journal

6 Scopus Citations
View graph of relations

Author(s)

Detail(s)

Original languageEnglish
Article number114416
Journal / PublicationApplied Energy
Volume262
Online published17 Jan 2020
Publication statusPublished - 15 Mar 2020

Abstract

The accurate demand prediction with high efficiency and advanced demand-side controller are essential for the enhancement of energy flexibility provided by buildings, whereas the current literature fails to present the mechanism on modelling development and demand-side control. This paper aims to deal with the complexity of building demand prediction with supervised machine learning method, including the multiple linear regression, the support vector regression and the backpropagation neural network. The regularization, adding the sum of the weights to the learning function, is utilized to improve the training speed and to solve the overfitting by eliminating the unnecessary connections with small weights. The configuration of the artificial neural network was presented, and sensitivity analysis has been conducted on the learning performance regarding different training times. Energy flexibilities of sophisticated building energy systems (including renewable system, electric and thermal demands and building services systems) were quantitatively characterised with a series of quantifiable indicators. Moreover, several advanced controllers have been developed and contrasted, in regard to the flexibility utilisation of building energy systems. Results showed that, the developed hybrid controller with short-term prediction through the cross-entropy function is more technically competitive than other controllers. With the implementation of the developed hybrid controller, the peak power of the grid importation can be reduced from 500.3 to 195 kW by 61%. This study formulates a data-driven model with an advanced machine learning algorithm for the accurate building demand prediction and a hybrid advanced controller with short-term prediction for the energy management, which are critical for the promotion of energy flexible buildings.

Research Area(s)

  • Demand side management, Energy demand prediction, Energy flexible building, Hybrid energy storages, Machine learning, Renewable energy