Research And Implementation Of Route Planning Strategy Learning Method

In recent years,with the rapid development of UAV technology,the mission scope of UAV extends from logistical support to tactical and strategic offensive operations.As a complement to traditional means of reconnaissance and weapons of destruction,UAV has been applied in positioning exploration,enemy reconnaissance and electromagnetic countermeasures.Great challenges to the intellectualization of route planning system is put forward because of the extensive application of UAV,the complexity of battlefield environment,the fuzziness and strong coupling of constraints.The research background of this paper is that the planning algorithm is difficult to satisfy the complex constraints and the inefficiency of planning caused by lack of experience knowledge in the interactive planning process.This paper combine machine learning related theory and technology to study the strategy learning method of route planning based on knowledge accumulation and knowledge learning.These planning strategies are used to assist planners in route planning,reduce planning difficulty and accelerate planning speed.There are three problems need to be solved in strategy learning method.First of all,how to design and collect learning samples that can reflect the knowledge and experience of planning experts.These knowledge and experience are reflected in what planning strategies experts adopt to cope with numerous track constraints and complex flight environments.Secondly,how to process sample data and extract planning strategies.Last but not least,how to implement strategy feedback and make the strategy guide the route planning effectively.In the process of collecting planning operation data,the criteria and process of sample collection are formulated,and the structure of sample is designed.The characteristics of constraints and the operation characteristics of planners are analyzed.The constraints are divided into two categories: aircraft environment constraints and aircraft characteristics related constraints.In order to further study the complex constraints and further subdivide the related constraints of aircraft characteristics,a hierarchical model of constraints and a coding method are proposed.Finally the sample classification management is realized.In the study of strategy learning method,according to the operational behavior characteristics of planners,XGBoost algorithm in ensemble learning is used to learn the planning strategy of aircraft environmental constraints,and K-prototypes algorithm in clustering algorithm is used to summarize the planning strategy of aircraft characteristic-related constraints.At the same time,for the convenience of computer calculation and planner's understanding,the learning results of aircraft characteristic constraints are expressed in the formal language of first-order predicate logic.In the process of route planning based on strategy feedback guidance,the strategy feedback is fed back to the planner according to the constraints and planning environment,when the route does not meet the constraints.Finally the planner get effective policy guidance during route planning.The experimental results show that the method can effectively extract the trajectory planning strategy.Satisfactory strategy guidance information can be given in the planning process,which reduces the labor intensity of planners and their dependence on knowledge and experience.It can also effectively improve the efficiency of route planning and reduce the difficulty of using the software.