A HEN-PPO strategy for home energy management system with reduce EV anxieties

This article proposes a novel method for scheduling residential building loads, including thermal and EV (Electric Vehicle) loads, to maximise demand response potential in smart home environment. In this approach, a load scheduling solution based on deep reinforcement learning (DRL) has been proposed. A policy search method based on Home Energy Management with Noise-Adaptive Proximal Policy Optimization (HEN-PPO) has been used to train the network. HEN-PPO can mitigate the influence of noise on temperature and solar radiation sensor data. The modification of the reward function for the system has been taken into account to address the distinct human behavior of randomly distributed cases of EV anxiety about range and time in conjunction with the optimal utilization of photovoltaic (PV) generation. This approach effectively manages both discrete and continuous activities of electrical loads at the same time, categorized as shiftable and regulatable, throughout the entire building model. Additionally, to replicate human behavior, real switching data from different household appliances has been used to determine the parameters for the probability distribution of residential load usage. Dynamic and unpredictable factors such as user behavior, ambient conditions, and real-time electricity pricing challenge scheduling residential building loads which are efficiently managed in this approach. To validate the effectiveness of the proposed method, simulation tests were carried out using real-time data distribution of household loads and surrounding temperature. The simulation results demonstrate promising energy savings, achieving 24%–29% reductions with the proposed approach. • A reinforcement learning technique based HEN-PPO method investigates the current issue of energy cost mini- mization and develops an effective method to mitigate the same for res- idential building. • To increase the robustness of the system both anxiety factor (range anxiety and time anxiety) of EV is considered in used reward function to include the dynamic behavior of EV owner. • This method demonstrates proficiency in mitigating the in- fluence of noise in temperature and solar radiation sensor data. • To address the real-time DR scheduling problem, a model-free method with multi distributed policy architecture based on DRL approach op- timized is used.