Building occupancy modeling using generative adversarial network

Due to the energy crisis and the awareness of sustainable development, the research on energy-efficient buildings has increasingly attracted attention. To achieve this objective, one important factor is to capture occupancy properties for building control systems, which refers to occupancy modeling in buildings. Due to the complexity of building occupancy, previous works try to simplify the modeling with some specific assumptions which may not always hold. In this paper, we propose a Generative Adversarial Network (GAN) framework for building occupancy modeling without any prior assumptions. The GAN approach contains two key components, i.e. a generative network and a discriminative network, which are designed as two powerful neural networks. Owing to the strong generalization capacity of neural networks and the adversarial mechanism in the GAN approach, it is able to accurately model building occupancy. We perform real experiments to verify the effectiveness of the proposed GAN approach and compare it with two state-of-the-art approaches for building occupancy modeling. To quantify the performance of all the models, we define five variables with two evaluation criteria. Results show that our proposed GAN approach can achieve a superior performance.