Analysis of Cooling Energy Reduction through PMV-Based Machine Learning Control under Hot and Dry Climate Conditions in Saudi Arabia

This study introduces a cooling control strategy focused on thermal comfort, aimed at reducing energy consumption while maintaining or improving indoor comfort in the extremely hot and dry climates of Saudi Arabia. Experiments were conducted in a full-scale climatic chamber using three operating modes: a constant-speed air conditioner set to 24°C (Case 1), an inverter air conditioner also set to 24°C (Case 2), and an intelligent controller (Case 3) that adjusts cooling based on real-time Predicted Mean Vote (PMV) values. In Case 3, an artificial neural network (ANN) model predicted the mean radiant temperature using easily measurable inputs—indoor air temperature, outdoor air temperature, air conditioner setpoint, and time-related variables. This prediction was then used to calculate PMV and dynamically adjust the cooling setpoint. Over a 24-hour test period, energy consumption was recorded at 16.1 kWh for Case 1, 8.7 kWh for Case 2, and 8.2 kWh for Case 3. This represents a 46% reduction in energy use with the inverter system compared to the constant-speed system, along with an additional 6% saving with the PMV-based control over the inverter baseline. In terms of thermal comfort, both Case 1 and Case 2 showed a cool-side bias, with mean PMV values of −0.9 and −0.7, and PMV unmet rates of 100% and 56%, respectively. In contrast, Case 3 achieved a mean PMV of 0.1 and a PMV unmet rate of only 6%, indicating that approximately 94% of occupied hours fell within the ASHRAE-55 comfort range. These results highlight that PMV-targeted, machine learning-enabled setpoint modulation can correct cool bias, enhance thermal comfort, and deliver additional energy savings beyond inverter-only operation in hot and dry climates.