In recent years, there has been a growing interest in smart agriculture, aimed at enhancing efficiency in farming through the use of technology. Farmers, working with multiple machines across various fields, necessitate the development of a versatile task assignment and route generation system (task allocation system) to cater to a diverse range of conditions. Previous studies have proposed various methods to address this problem. However, it remains unclear which method is most suitable when specific conditions tailored to individual farms are considered. Additionally, the performance of these methods is not fully utilized as hyperparameters are often set empirically. Consequently, the development of a truly versatile task allocation system for agriculture remains unachieved. Therefore, this study aims to develop a versatile task allocation system for agriculture. Specifically, we created “the field nodes divided graph” by dividing fields into the maximum number of vehicles operating in that field, and by formalizing it as a split delivery vehicle routing problem (SDVRP), we modeled the farmland considering the operation of multiple vehicles. Using computer simulations, this study identifies the most effective optimization methods and hyperparameter combinations under varying farm sizes and maximum calculation times. The results of computer simulations applying a local search method, a simulated annealing (SA), a genetic algorithm (GA), and an ant colony optimization (ACO) to five types of hypothetical grid-shaped farmlands revealed that the SA algorithm is superior for smaller farmlands, while the local search method excel in larger farmlands with shorter calculation times. Furthermore, the study derives the best hyperparameter values corresponding to different farm sizes.