Short-term building occupancy prediction based on deep forest with multi-order transition probability

Occupancy plays a vital role in optimizing the operation of building service systems. This study proposed a novel model for predicting short-term building occupancy. In the model, the autocorrelation function (ACF) and partial autocorrelation function (PACF) are used to determine the most relevant occupancy status. Then the multi-order transition probability (MOTP) is established and integrated with the deep forest (DF) for occupancy prediction. To assess the effectiveness of the proposed MOTP-DF model, a validation experiment was conducted in an office room to compare its prediction performance with conventional methods, including Markov chain, decision tree (DT), and support vector machine (SVM) models. The results show that the proposed model can track occupancy change with higher accuracy and fewer fluctuations. Moreover, it improves the prediction accuracy by 6.3–10.0%, 4.6–8.3%, and 4.8–8.3% over the Markov chain, DT, and SVM models, respectively. A further evaluation indicates that MOTP can quantitatively incorporate occupancy's status interdependency and stochastic characteristics into its prediction. Compared with the DT and SVM models, the DF model with deep and ensemble structures could benefit more from the integration with MOTP due to its higher robustness. This study also found that a proper selection of transition probability orders, forest algorithms, and corresponding maximum tree depths can further enhance the prediction accuracy of the proposed model.