This paper proposes a method for modelling and optimising spatial influences in agentbased models by combining procedural generation with a genetic algorithm. Applied to an urban growth model, this method enables agents, representing residents, to make decisions based on their environment (e.g., proximity to roads encouraging construction). Procedural generation aids in modeling these influences, but the complexity of manually adjusting the parameters necessitates the use of a genetic algorithm for automatic optimization. The approach examines three spatial measures—Chamfer distance, kernel density, and density grid—to train the model and simulate the location of new housing. Experimental results demonstrate the method's effectiveness, generality, and the critical role of fitness functions. - 74