As urbanization accelerates, megacities are facing challenges such as inefficient land use and traffic congestion, particularly in the context of rail transit-oriented development, where land use optimization remains a significant research gap. Current urban planning still relies heavily on the experience and intuition of government planning departments, without achieving quantitative, intelligent, and scientific decision making. This study takes Panda Avenue Subway Station as a case study to analyze the evolution of land use patterns around subway stations and explore optimization strategies to enhance land development efficiency and spatial utilizationTo fill this research gap, this paper proposes a CNN-AIMatch model based on machine learning algorithm and an enhanced PLUS-Markov prediction model using the increase and decrease of floor area ratio as a control measure, which adopts an increase in plot ratio as a control measure to improve the accuracy of the Kappa coefficient in different plot ratio scenarios and the prediction of 3D urban spatial growth trends. The model effectively overcomes the limitations of the conventional 2D perspective in predicting urban expansion. By simulating urban renewal and ecological preservation scenarios, it provides an innovative solution for land use pattern optimization and plot ratio control at the block level in subway station areas. The goal of this study is to optimize land use and floor area ratio control strategies through the application of this model, intelligently respond to the challenges of high-density development and quality of life assurance, achieve the best use of land, and promote sustainable urban development and the construction of smart cities.

Inefficient land reuse has emerged as a critical pathway for the sustainable development of urban spaces. Efficient land development in megacities’ industrial heritage areas is heavily influenced by the influx of mass passenger flows from new subway stations. To address this issue, a dynamic passenger flow-oriented land use prediction model for subway stations was developed. This model iterates a simulation model for dynamic passenger flow based on tourists and residents with an artificial neural network for land use prediction. By enhancing the kappa coefficient to 0.86, the model accurately simulated pedestrian flow density from stations to streets. Experiments were conducted to predict inefficient land use scenarios, which were then compared with the current state in national industrial heritage areas. The results demonstrated that the AnyLogic-Markov-FLUS Coupled Model outperformed expert experience in objectively assessing dynamic passenger flow impacts on the carrying capacity of old city neighborhoods during peak and off-peak periods at subway stations. This model can assist in resilient urban space planning and decision-making regarding mixed land use.