A smart heating set point scheduler using an artificial neural network and genetic algorithm

Abstract

This paper introduces a novel, adaptive, heating set point scheduler that aims to minimise the heating energy consumption in a building while maintaining thermal comfort levels. The presented control strategy couples two computational intelligence techniques, an Artificial Neural Network, ANN and a Genetic Algorithm, GA. The ANN surrogate model was trained using data from multiple Energy Plus building simulations with varied heating set points. This allows quick, computationally cheap, calculation of a fitness function, opposed to time consuming Energy Plus simulations. The ANN's inputs include weather and occupancy predictions as well as taking account of the buildings thermal inertia to predict energy consumption, predicted percentage dissatisfied, PPD, and indoor temperature. A multi-objective GA uses the ANN to calculate the energy sum over 24 hours and the average occupied PPD as its two objectives. The optimisation strategy was applied on a week in January over which it reduced energy consumption by 4.93% and improved PPD by 0.76%. The great advantage of this controller is that it could be simply adapted to take account of the changing energy environment. This could include an addition of renewable resources or demand response controls as part of a district heating network.