Predictive modelling and optimization of HVAC systems using neural network and particle swarm optimization algorithm

The concept of maintaining indoor environmental quality comprising building indoor temperature, relative humidity, CO2, and volatile organic compound (VOC) level poses new challenges to the optimal operation of heating, ventilation and air-conditioning (HVAC) systems. While existing case studies demonstrate the energy saving potentials for efficient HVAC operation, there is a lack of studies quantifying energy savings whilst considering indoor environmental conditions. This study proposes a state-of-the-art modelling and optimization approach to minimize the energy consumption of the HVAC systems without compromising indoor environmental quality. While the primary objective of ensuring optimal operation of HVAC systems is to minimize energy consumption, controlling indoor environmental parameters to remain within the acceptable range imposes excess energy use. These two conflicting objectives constitute a multi-variable constrained optimization problem that has been solved using a particle swarm optimization (PSO) algorithm. Real-time predictive models are developed for the individual indoor environmental parameters and HVAC energy consumption using a Nonlinear Autoregressive Exogenous (NARX) neural network (NN). During model development, models' performance is optimized in terms of complexity, predictive accuracy, and ease of application to a real system. The proposed predictive models are then optimized to provide an optimal control setting for the HVAC systems considering seasonal variations. The results indicate that it is possible to reduce 7.8% of total energy, without negotiating indoor environmental conditions, e.g., air temperature 19.60-28.20C and relative humidity 30-65% as per ASHRAE Standard 55, and CO2 <= 800 ppm and VOC <= 1000 ppm as per AS 1668.2.

Automated summary

This study proposes an advanced modeling and optimization approach to minimize the energy consumption of HVAC systems without compromising indoor environmental quality. Real-time predictive models are developed using a nonlinear autoregressive exogenous neural network, and optimized to consider seasonal variations. The results indicate that it is possible to reduce total energy consumption while maintaining indoor environmental conditions within acceptable ranges.