Modelling and visualization behavior simulation in public open spaces based on MLP and SFM

This research is grounded in the analysis of real human activity data in public open spaces (POS) and investigates the utilization patterns of 45 spatial segments and 121 subspaces within Shanghai's waterfront public areas. Employing a multi-layer perceptron neural network model, we explore the mechanisms by which 30 space, environmental, and facility factors influence various activities, including sightseeing, leisure, and sports. These findings are then integrated into a behavior simulation visualization system. Following a fitting and adjustment process, we validate the effectiveness of this behavior simulation system. The research outcomes reveal the robust performance of the multi-layer perceptron in predicting pedestrian activity in public spaces. Translating numerical predictive models into dynamic systems of behavioral simulation driven by social force model (SFM), we observe a relative error (RE) range of 0.041–0.158, with minimal loss of fit. Finally, through practical case testing, we illustrate the application of this behavior simulation workflow. The primary objective of this research is to provide predictive and scientific foundations based on human behavioral requirements for the design and optimization of POS, supporting human-centric and refined design decisions for urban landscape space.