Reinforcement Learning with Dual Safety Policies for Energy Savings in Building Energy Systems

Reinforcement learning (RL) is being gradually applied in the control of heating, ventilation and air-conditioning (HVAC) systems to learn optimal sequences for energy savings. However, due “trial error” issue, output RL may cause potential operational safety issues when real systems. To solve those problems, an algorithm with dual policies savings HVAC proposed. In proposed policies, implicit policy a part model, which integrates into optimization target RL, by adding penalties reward actions that exceed constraints. explicit policy, online classifier built filter outputted RL; thus, only are classified as safe have highest benefits will be finally selected. this way, controlled running algorithms can effectively satisfied while reducing consumptions. verify algorithm, we implemented existing commercial building. After certain period self-studying, consumption had been reduced more than 15.02% compared proportional–integral–derivative (PID) control. Meanwhile, independent application without proportion indoor temperature not meeting demand 25.06%.