Research on Rural Landscape Space Optimization Algorithm Based on Machine Learning

This paper proposes a rural landscape space optimization algorithm based on machine learning, aiming to achieve efficient optimization of rural landscape through deep reinforcement learning and graph neural network modules. The algorithm framework comprehensively considers multidimensional indicators such as the rationality of landscape space layout, ecosystem service functions and residents' satisfaction. By constructing accurate state space and action space, combined with reward functions, it guides the intelligent agent to make optimization decisions. The deep reinforcement learning module uses the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\epsilon$</tex>-greedy strategy to balance exploration and utilization, and the graph neural network module updates node features by aggregating neighbor node information to effectively handle the complex relationship between landscape elements. The experiment selected three typical rural landscape areas, namely the North China Plain, Jiangnan Water Village and Southwest Mountain Area, for simulation verification. The results show that in terms of compactness and connectivity, the proposed algorithm is significantly better than the genetic algorithm and simulated annealing algorithm; in terms of ecosystem service functions, the proposed algorithm performs better in carbon sequestration capacity and biodiversity improvement; in terms of residents' satisfaction, the proposed algorithm obtains higher scores in landscape beauty, living convenience and environmental comfort. This algorithm provides an efficient and intelligent solution for rural landscape optimization, with good universality and application prospects.