| Agriculture is the source of human clothing and food,the foundation of survival,and the foundation of national economic development.Effective agricultural plant protection work can be of great help in improving crop yields.For a long time,our country’s agricultural plant protection work is mainly done by manpower,which is inefficient and may endanger health.President Xi once pointed out during his inspection that“it is necessary to strengthen the integration of agriculture and science and technology,and strengthen the innovation of agricultural science and technology”.Unmanned Aerial Vehicle(UAV),as an important agent,has been widely used in key fields such as surveying and mapping,rescue,and geological exploration.Although it is believed that in the field of agricultural plant protection,UAVs should play an increasingly important role,such as detecting the growth of crops,watering crops,helping crops sow,and spraying pesticides on crops.But in fact,it has not been applied to agricultural production on a large scale.The main reason is that the current agricultural plant protection UAV work still relies on professionals to manually control the UAV for operation,and the shortage of qualified operators leads to higher labor costs.Therefore,it has become an urgent need to develop a system that enables UAVs to automate operations in agricultural plant protection environments.Multi-agent task planning is an important research direction in the field of multi-agent systems,and it is also one of the core research work of establishing an automatic operation system for agricultural plant protection UAVs.Its work goal is to plan tasks to agents in complex environments.However,when solving the problem,it is easy to fall into local optimum,the exploration space is too small,and the convergence speed is slow.In order to improve the performance of the algorithm,many researchers have proposed a large number of improvements to traditional algorithms to obtain better performance solutions.In this paper,a systematic study is made on several basic models of multi-agent non-cooperative task planning and multi-agent cooperative task planning,and the corresponding solution methods are provided,and the proposed algorithm is applied in the field of agriculture plant protection UAV agriculture.The main research work and innovations of this paper include the following aspects:(1)Aiming at the key problem that the agent may have emergencies in the work process,which leads to the reduction of work efficiency,a Non-Cooperative Task Planning in Range Allocation Subadditive Value model(RASV-NCTP)is established.In order to solve the model,this paper proposes the Conditional Gaussian Walk Dragonfly Optimization Algorithm(CGWDO).By introducing the CGW strategy,the CGWDO algorithm speeds up the chance of the population jumping out of the local optimum,improves the diversity and exploration of the population,and improves the shortcomings of the traditional dragonfly algorithm.The experimental results on 19sets of data and 6 benchmark functions show that the CGWDO algorithm has obvious improvements in usability and stability.In this paper,the CGWDO algorithm is applied to the task planning problem of agricultural plant protection UAVs.The experimental results show that the CGWDO algorithm has a good performance in solving the task planning problem of agricultural plant protection UAVs,and its performance in evaluation indicators is higher than other methods,and it has better planning results.(2)Aiming at the key problem that the state of the agent and the task information cannot be completely obtained in some environments,a Non-Cooperative Task Planning in Partial Ordinal Allocation model(POA-NCTP)is established.Robin Algorithm Based on Importance Evaluation(IERR)is proposed.This paper theoretically proves that the r-MMS(Min Max Share)index of the IERR algorithm does not exceed 2+[m· α_i·(1+n)-n]/n~2 even in the worst case.In this paper,the IERR algorithm is compared with the 8 population intelligent optimization algorithms and the 2 task allocation algorithms.The experimental results show that the IERR algorithm has outstanding performance in both usability and operational efficiency.(3)Some complex tasks require different types of agents to work together.In this regard,this paper establishes a model of Cooperative Task Planning in Prepare Separately for Centralized Execution Environment(PSCE-CTP).In order to solve the model,this paper proposes a Approximate State Matching Q-Learning Algorithm(ASMQL).This paper proves from both theoretical and experimental aspects that the ASMQL algorithm is significantly higher than other methods in terms of usability and operational efficiency.In this paper,the ASMQL algorithm is applied to the strategic planning problem of agricultural plant protection UAVs.The experimental results show that the ASMQL algorithm has a good effect in solving the strategic planning problem of agricultural plant protection UAVs,and at the same time,the performance of the evaluation index is higher than other methods,and the optimal strategy can be obtained. |
