{"title":"基于深度强化学习的时间窗约束下农机调度","authors":"Yuemei Wu , Hui Fang","doi":"10.1016/j.biosystemseng.2025.104188","DOIUrl":null,"url":null,"abstract":"<div><div>This study proposes a deep reinforcement learning (DRL)-based approach to address key limitations of traditional methods, such as low efficiency, poor accuracy, and unstable solution quality, in solving agricultural machinery scheduling with time window constraints. This study first analyses the agricultural machinery scheduling problem as a Markov decision process (MDP), constructs an attention-based network utilising the Transformer architecture, and trains the model with the actor-critic algorithm. A local search algorithm is then integrated to further refine the scheduling solution. This study uses real-world datasets of field with point scales of 20, 50, and 100 to compare the performance of different scheduling strategies, including DRL, ant colony optimisation (ACO) and genetic algorithm (GA). Experimental results show that, at a 20-point scale, DRL reduces scheduling costs by an average of 5.96 % and 6.57 % compared to ACO and GA, while decreasing runtime by 85.1 % and 90.8 %, respectively. At a 50-point scale, DRL reduces scheduling costs by an average of 9.91 % and 16.7 % compared to ACO and GA, while decreasing runtime by 96.0 % and 98.1 %. At the 100-point scale, DRL reduces scheduling costs by an average of 13.8 % and 21.9 % compared to ACO and GA, while decreasing runtime by 98.5 % and 99.1 %. Overall, the DRL algorithm consistently minimises scheduling costs and substantially decreases runtime relative to ACO and GA, providing an efficient and scientifically grounded solution for agricultural machinery scheduling.</div></div>","PeriodicalId":9173,"journal":{"name":"Biosystems Engineering","volume":"257 ","pages":"Article 104188"},"PeriodicalIF":5.3000,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Agricultural machinery scheduling under time window constraints using deep reinforcement learning\",\"authors\":\"Yuemei Wu , Hui Fang\",\"doi\":\"10.1016/j.biosystemseng.2025.104188\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>This study proposes a deep reinforcement learning (DRL)-based approach to address key limitations of traditional methods, such as low efficiency, poor accuracy, and unstable solution quality, in solving agricultural machinery scheduling with time window constraints. This study first analyses the agricultural machinery scheduling problem as a Markov decision process (MDP), constructs an attention-based network utilising the Transformer architecture, and trains the model with the actor-critic algorithm. A local search algorithm is then integrated to further refine the scheduling solution. This study uses real-world datasets of field with point scales of 20, 50, and 100 to compare the performance of different scheduling strategies, including DRL, ant colony optimisation (ACO) and genetic algorithm (GA). Experimental results show that, at a 20-point scale, DRL reduces scheduling costs by an average of 5.96 % and 6.57 % compared to ACO and GA, while decreasing runtime by 85.1 % and 90.8 %, respectively. At a 50-point scale, DRL reduces scheduling costs by an average of 9.91 % and 16.7 % compared to ACO and GA, while decreasing runtime by 96.0 % and 98.1 %. At the 100-point scale, DRL reduces scheduling costs by an average of 13.8 % and 21.9 % compared to ACO and GA, while decreasing runtime by 98.5 % and 99.1 %. Overall, the DRL algorithm consistently minimises scheduling costs and substantially decreases runtime relative to ACO and GA, providing an efficient and scientifically grounded solution for agricultural machinery scheduling.</div></div>\",\"PeriodicalId\":9173,\"journal\":{\"name\":\"Biosystems Engineering\",\"volume\":\"257 \",\"pages\":\"Article 104188\"},\"PeriodicalIF\":5.3000,\"publicationDate\":\"2025-07-21\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Biosystems Engineering\",\"FirstCategoryId\":\"97\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S1537511025001242\",\"RegionNum\":1,\"RegionCategory\":\"农林科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"AGRICULTURAL ENGINEERING\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Biosystems Engineering","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1537511025001242","RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"AGRICULTURAL ENGINEERING","Score":null,"Total":0}
Agricultural machinery scheduling under time window constraints using deep reinforcement learning
This study proposes a deep reinforcement learning (DRL)-based approach to address key limitations of traditional methods, such as low efficiency, poor accuracy, and unstable solution quality, in solving agricultural machinery scheduling with time window constraints. This study first analyses the agricultural machinery scheduling problem as a Markov decision process (MDP), constructs an attention-based network utilising the Transformer architecture, and trains the model with the actor-critic algorithm. A local search algorithm is then integrated to further refine the scheduling solution. This study uses real-world datasets of field with point scales of 20, 50, and 100 to compare the performance of different scheduling strategies, including DRL, ant colony optimisation (ACO) and genetic algorithm (GA). Experimental results show that, at a 20-point scale, DRL reduces scheduling costs by an average of 5.96 % and 6.57 % compared to ACO and GA, while decreasing runtime by 85.1 % and 90.8 %, respectively. At a 50-point scale, DRL reduces scheduling costs by an average of 9.91 % and 16.7 % compared to ACO and GA, while decreasing runtime by 96.0 % and 98.1 %. At the 100-point scale, DRL reduces scheduling costs by an average of 13.8 % and 21.9 % compared to ACO and GA, while decreasing runtime by 98.5 % and 99.1 %. Overall, the DRL algorithm consistently minimises scheduling costs and substantially decreases runtime relative to ACO and GA, providing an efficient and scientifically grounded solution for agricultural machinery scheduling.
期刊介绍:
Biosystems Engineering publishes research in engineering and the physical sciences that represent advances in understanding or modelling of the performance of biological systems for sustainable developments in land use and the environment, agriculture and amenity, bioproduction processes and the food chain. The subject matter of the journal reflects the wide range and interdisciplinary nature of research in engineering for biological systems.