Integrated harvest and distribution scheduling of fresh agricultural products for multiple farms using a Q-learning-based artificial bee colony algorithm with problem knowledge

Abstract

Nowadays, a great many of farmers sell fresh agricultural products through direct online sales. In this context, the farm-to-door supply mode has emerged, playing a crucial role in reducing transportation cost and quality deterioration. This work addresses an integrated harvest and farm-to-door distribution scheduling problem involving multiple farms. First, a mixed integer programming model is formulated to minimize total operation cost and maximize customer satisfaction regarding product quality. Second, a Q-learning-based artificial bee colony algorithm with problem knowledge (Q-ABC-K) is developed in particular. The algorithm is featured with the following strategies: (i) a hybrid initialization method with two rules to generate a high-quality population; (ii) a crossover operation to prompt a collaborative search between the population and external archive at the employed bee phase; (iii) a Q-learning method to favorably select premium neighborhood structures at the onlooker bee phase; and (iv) a knowledge-based local search method to refine the nondominated solutions. Finally, a large number of comparison experiments are conducted on a set of test instances. Through observing and analyzing the experimental results, three conclusions are acquired as follows: (i) The design of Q-learning and knowledge-based local search methods plays a significant role in enhancing the performance of Q-ABC-K; (ii) Q-ABC-K performs better than four state-of-the-art approaches in dealing with the considered problem; and (iii) Q-ABC-K has an advantage over an exact solver CPLEX in solving small-scale cases.