PERFORMANCE AND ANALYSIS OF INTELLIGENT HVAC HYBRID CONTROL FOR HIGH ENERGY EFFICIENCY IN BUILDINGS

Abstract

HVAC systems, including heating, ventilation and air-conditioning, have a big impact on the energy use of buildings. In commercial buildings these systems can represent almost 40-60% of the total electricity consumption of the facility. Buildings in general also account for almost a third of the world’s energy consumption. This high demand has made the improvement of HVAC operations an important area of research for reducing energy use, controlling operating expenses and supporting environmental sustainability. In this paper, we propose a predictive HVAC control method based on the combination of GRU and Random Forest. The main objective was to evaluate the feasibility of integrating a deep-learning model with a traditional machine learning algorithm to provide more dependable control in various operating conditions. The proposed Hybrid GRU+RF model was evaluated under two cases. Renewable-energy information was not included in the first case, which is a normal operating environment. The model in this case reduced energy consumption by 24.29%. For the second case, the solar irradiance was another input. The existence of this variable complicated the prediction problem. However, the model proposed still achieved an energy saving of 15.43%. Results from LSTM and GRU individually were significantly different. These individual deep-learning models performed relatively well under the more straightforward condition, but the performance became worse when renewable-energy features were added. The resulting energy-saving values were negative, being between −15.85% and −16.27%. This leads to an approximate 36-37 % performance loss compared to the case without renewable-energy inputs. The Hybrid GRU+RF model was not deteriorated to the same extent. It even presented positive results in terms of energy saving. This means that the combination of GRU and Random Forest was better to deal with the additional variations related to solar irradiance. The proposed model was also tested not only with simulated or laboratory data . The system was also studied with real operational data collected over a period of one full year from an industrial plant in Jalgaon, Maharashtra, India. The dataset included 10,510 observations. Additional validation was done on another data set of a multi-zone building (2,016 samples) from the ORNL FRP-2 facility. The Hybrid GRU+RF model achieved a R² score of 0.9335 for the prediction of temperature and 0.9205 for the prediction of energy. The results show that the model was able to predict both the energy requirements and the temperature of the building with a good degree of accuracy. The testing also showed that the hybrid approach could perform well when applied to building conditions different from those used during the model development. The standalone Random Forest model, however, was showing signs of overfitting. The results of the study indicate that the Hybrid GRU+RF model can be utilized as a practical approach for predictive HVAC energy management. One of its main advantages was that its performance was useful even after the renewable energy information was included into the prediction process. This is especially true of existing buildings where solar power and other renewables are becoming more prevalent. Based on the obtained results, the proposed method could estimate an annual saving of about $1,600–$2,000 per building and could reduce carbon emissions by approximately 11.5 tons of CO2 per facility per year. Therefore, the study suggests that the combination of GRU with Random Forest may improve the reliability of intelligent HVAC control. It can contribute to the creation of more energy-efficient and sustainable smart buildings.