Real-time occupancy prediction in a large exhibition hall using deep learning approach

Abstract Intelligent control systems for optimizing the energy management of ordinary buildings and houses have been commonly studied for decades, but the development of such management systems has not been studied much in large exhibition halls. While occupancy prediction is considered as a key element of such intelligent control systems, it is not easy in a large exhibition hall due to its spatial volume and irregular movements of visitors. In this paper, we propose spatial partitioning of the hall and an occupancy prediction model based on recurrent neural network (RNN) with long short-term memory units (LSTM) to solve the mentioned problems. We test the feasibility of our RNN approaches to predict short-term and long-term occupancy using the sequence patterns for hall occupancy changes in separated multiple zones until a current time point. We demonstrate that the proposed RNN model achieves superior performance by comparing with other prediction models. Then we apply our software toolset for predicting real-time occupancy in actual exhibition events in a large exhibition hall. Our prediction software pipeline is integrated into energy management systems in the exhibition hall.