Assistant Decision-Making For Multi-Operator-Multi-UAV Command And Control

In battle fields the gun-launched unmanned aerial vehicle(GLUAV)command and control has to confront the uncertainty of information,the complexity of mathematical model,and the pressure of computing time.Assistant decision-making for Multi-Operator-Multi-UAV(MOMU)command and control is one of the focuses to exert the advantages of cooperation of multi-UAV in network war.This paper focuses on assistant decision-making model,operator tasks assignment,attention allocation and adaptive sorting of operator tasks,builds mathematical model,researches optimization theory,designs algorithm,carries out simulation and experiments under the background of MOMU command and control including multi-operator,multiple tasks and multiple targets.(1)The typical tasks of GLUAV's are selected,which are decomposed based on the task execution process.Then we get the sub-task set of GLUAV.According to the automation level of UAV,the sub-tasks which need operator to process are extracted and the selected sub-tasks are the task set of operator's.The attributes of operator's tasks are studied,and the attributes of typical targets are set.According to the process MOMU command and control,the logical structure assistant decision-making is designed and a system of assistant decision-making for MOMU command and control is established.Key questions of the assistant decision-making are divided into three questions: multi-operator task assignment,single-operator attention allocation and adaptive sorting of operator tasks.(2)According to the characteristics of targets in battlefield,the operator task assignment is divided into centralized structure and distributed structure.Centralized task allocation combines with Model Prediction Control(MPC),and the task allocation model based on maximizing global reword is established.The solving algorithm of centralized task model based on improved artificial immune algorithm is designed.Distributed task allocation combines with operator assignment characteristics,based on the contract network protocol a distributed task allocation model is established.The hybrid task assignment model is automatically adopted according to the characteristics of tasks,maximizing the global reword of non-time-sensitive task assignment and dynamically assigning time-sensitive tasks in real-time.(3)Three task processing strategies which are first-in-first-out(FIFO),priority-first and optimized task sorting are considered.According to the characteristics of the task processing strategies single-operator attention allocation models are established.The solving algorithm for FIFO and priority-first is designed using.The solving algorithm for optimized task sorting is carried out by Combining dynamic program and artificial immune algorithm.(4)As operator selects task from the task queue manually will destroy system settings,it needs to predict the sorting of tasks according to the habit of operator.An operator task processing model is designed first based on the attributes of operator's tasks.And tasks are sorted according to the operator task processing model,which are treated as training data.Then the training data are used by an improved linear regression method,the adaptive sorting model according to the operator habit is obtained.(5)Based on the shared structure of MOMU command and control,a MOMU command and control test platform is designed.Using this test platform,multi-operator task assignment algorithm,single-operator attention allocation algorithm and adaptive task sorting algorithm proposed in this paper are verified.The results are as follows.The algorithms proposed in this paper reduce the time of operators to fulfill all tasks.The average time each task consumed in the task assignment is about 0.53 seconds,and this is longer than the result of control group.However,under the framework of MPC,the algorithm meets the real-time requirements of MOMU command and control.The integrated reward of all the tasks is bigger than that of control group by 26%.The adaptive task sorting model can learn from the operator task processing data and predict the sorting of tasks.And the correct rate of prediction is between 88% and 90%.The correct rate is higher than the control group by 33%.It can reduce operator's operations and workload,improving the efficiency of the operator.