An Optimization Method for the Land-Use Index in a Site Area: Improving the BP Neural Network Model Based on a Genetic Algorithm

The land-use layout in the core area of a station under the transit-oriented development (TOD) model plays a crucial role in shaping urban development. Although many rail transit stations have been constructed, most do not exhibit the intensive, economical, efficient, and orderly land-use characteristics typical of TOD stations. The traditional backpropagation (BP) neural network model offers significant advantages in addressing the nonlinear aspects of land-use evaluation and its influencing factors. However, due to the limited number of mature TOD stations in operation, available training data are insufficient, leading to suboptimal model performance. To address the issue of model underfitting caused by limited training samples, a genetic algorithm (GA) is employed to optimize the BP neural network, forming a GA-BP model (GA-BP). In this study, we construct a GA-optimized BP neural network for land-use evaluation and a rail transit ridership prediction model. These models are then integrated into a multiobjective planning framework, and the optimal land-use scheme within a 500-m radius of a TOD station is identified using nondominated sorting genetic algorithm II. The empirical results indicate that (1) the GA-BP model is more capable of finding global optima and demonstrates low error, high efficiency, and strong stability, effectively mitigating the problem of underfitting due to limited data; and (2) the model shows good applicability and can offer valuable optimization recommendations. We apply this framework to predict and optimize TOD land use in a target area, providing strategic guidance for the coordinated development of land use and urban form in areas surrounding TOD stations.