DQN-PACG: load regulation method based on DQN and multivariate prediction model

Demand response plays a pivotal role in modern smart grid systems, aiding in balancing energy consumption. However, the increasing energy demands of contemporary society have placed a significant burden on power systems. To simulate the interaction between electricity supply and demand, this paper introduces the concept of Deep Q-Network (DQN) to the domain of demand response. Additionally, a novel multivariate forecasting model, referred to as PreAttention-CNN-GRU (PACG), is proposed to predict in real time the impact of electricity prices on consumer electricity usage behavior. Finally, a load control method, denoted as DQN-PreAttention-CNN-GRU (DQN-PACG), is presented to achieve price-based demand response. The performance of PACG was tested on a real-world German dataset, demonstrating superior predictive accuracy compared to traditional forecasting models such as Long Short-Term Memory Networks. Furthermore, the test results of DQN-PACG on the same dataset contribute to alleviating the load and stress on the power grid. This paper also includes a case study of southern provinces in China, where the model was able to reduce electricity consumption by 1.64