Research on Multi-UAV Collaborative Search in Dynamic Environment

. Combine the uncertainty of dynamic targets, a multi-UAV reconnaissance scheduling problem model was constructed under dynamic environment, take advantage of the characteristics of dual evolution, the insertion point operator and reverse sequence operator are improved, and the problem is solved by the artificial bee colony algorithm based on the semi-random search strategy. Finally, the simulation experiment was done in the background of South China Sea, and the experimental result shows the effectiveness and feasibility of the proposed algorithm for solving the multi-UAV reconnaissance scheduling problem.


Introduction
The unmanned aerial vehicle (UAV) reconnaissance scheduling problem is the core problem of multiple UAV cooperative reconnaissance, due to the uncertainty and dynamics of some targets, the multiple UAV reconnaissance scheduling problem become very complicated.
Literature [1] took the cost and benefit of task allocation as an overall objective function and assigns tasks based on the bidding process. Literature [2] built a single objective optimization model based on the maximum value of attack task, and utilized the sub-populations of ant colony algorithm solved the problem. Literature [3] completed the task assignment process based on the binary particle swarm algorithm. The Artificial Bee Colony (ABC) algorithm is based on the rationale of imitating the intelligent behavior of the swarm foraging behavior [5] . The solution quality of ABC is relatively good, and it has the characteristic of less control parameters, simple operation, strong robustness and higher search precision [6][7] . However, the basic ABC algorithm has a defect of " convergence precocity", which has shortcomings such as insufficient of development capability and slow convergence speed [8][9] . The literature [10] enhanced the convergence speed and global searching ability of the algorithm by improving the traditional artificial swarm algorithm. Literature [11] combined the ABC algorithm with genetic operator, and the search capability of algorithm was improved. In literature [12], an ABC algorithm based on particle swarm was designed to find optimal solution by combining the global search and local search. Literature [14] combined various of search strategies, and the local search capability of algorithm was enhanced. Literature [15] combined the global search capability of ABC algorithm with the local search capability of ACO algorithm to obtain an ACO-ABC hybrid algorithm to find optimal solution. Literature [16] set the optimal neighborhood solution to replace the current solution only after many times of not updated, so that the potential feasible solution can be preserved.
Contrapose the uncertainty about the emergence time of dynamic targets, this paper designed an improved double evolution artificial swarm algorithm. The algorithm was guided by the objective function of model, and the local optimal solution was processed in global search. The algorithm utilized the advantages of different local search operator, accelerated the search speed of algorithm, and enriched the diversity of feasible solution. Finally, through simulation experiment showed that the algorithm can effectively solve the UAV reconnaissance scheduling problem. (1) ' t N for static targets, the location of these targets and the moment of emergence are known.
(2) ' ' t N N for dynamic targets, the location is known, but the moment of emergence is unknown.
For any moment τ , use t i M  represent the reliability of the dynamic target i appears at the moment τ . So, , assume that the reliability distribution t i M  is known.

Profit calculation
For each moment tT  , and for each goal  (1): . C as the normal number of large enough, in this paper, the reliability estimation of dynamic target j appears in the previous of i L is estimated as equation (2): That is, the sum value of the reliability estimation of dynamic target j appears at the time

Objective function
The objective function as equation (3): ; the first part of objective function is the total reconnaissance profit, the second is the UAV's total flight distance, and the third is the inherent flight cost of UAV. If all the targets can be detected by UAV, the total reconnaissance profit is maximum; if complete the reconnaissance mission calls fewest UAV and the flight distance is minimum, inherent cost and the total distance of flight is minimum, so the overall goal is to complete the reconnaissance mission with a minimum flight mileage and a minimum quantity of UAV.

2.4The constraint
In the process of reconnaissance, UAV carries out a reconnaissance of a single target at most, it can reach target area before the start of the first reconnaissance moments, but must begin before the end of the reconnaissance time window, in addition to ensure the continuity of time. To sum up, the reconnaissance scheduling model of each decision stage is as follows: , ;

Scheduling solution 3.1 Initialize
Given the current scheme X and unallocated target set ' t N , the initial population construction process is as follows (assuming the current scenario is empty): 1) The base is set as ( ) 1 1 v , the minimum cost LC set to infinity, and the maximum quantity of UAV is set to # veh , initialize the best sequence of programs as . d) Update the target set to be assigned and update the current plan.
3) Output the parameter set and it's corresponding best solution set.

Dual evolutionary process
The algorithm performs local search and adopts dual evolutionary method in the search process, and the specific operation process is as follows: (1) Semi-random optimal insertion point operator, as shown in figure 1: a) The insertion point 1 r is randomly selected, and the insertion location is denoted as 2 r ; b) 2 r iterate through all feasible insertion locations, generate new feasible schemes, and calculate the adaptive value of each feasible scheme; c) Find a feasible solution that minimizes the adaptive value and insert the target 1 r into the position.    This search process is relatively fast and efficient, with a small amount of computation. In this paper, it is named as a local search method based on the reconstruction plan.

Termination conditions
In general, if the set iteration number has completed in algorithm, the termination can be determined. In addition, it is also can be determined whether the solution generated by the algorithm in the global search process is rapidly convergent as the condition of the termination.

Experimental and results analysis 4.1 Apply scenarios
This article will set the task background as UAV spy on the south China sea reef monitoring, assuming the base is located in the midpoint of reef cluster (longitude 114.60 °; latitude 9.47 °). The reef position as shown in figure 3, the distance between each target (base) is determined by Euclidean space distance. Assume that there will be illegal vessels appearing near 5 islands, as dynamic uncertainty targets, the basic setting is shown in table 1. The uncertainty distribution in table 1 is based on the known information and battlefield experience, and the specific content is the undetermined distribution of its occurrence time, duration is the length of time when other's ships stay in disputed waters. Set the speed of UAV is 144 km/h.

The simulation test
In the process of UAV's mission, combine the local search strategy of HB-ABC and RC-ABC, a dynamic multi-objective reconnaissance scheduling solution was performed. In the experiment, the maximum number of cycles is set to 1000, the value of  is 3, and the value of Limit is set to 20. The experiment result gained by STK software shown in figure 4. As can be seen from the figure, the UAVs have conducted reconnaissance on all targets, and its total reconnaissance revenue reaches the maximum. To test the effectiveness of combining the local search strategy of RC-ABC algorithm and HB-ABC algorithm, with these two kinds of ABC algorithm to track separately, and compared with the eventually tracking results from RC-HB-ABC algorithm. In RC-HB-ABC algorithm, the parameters  related to the termination conditions is taken as 3, and the weight of dynamic target is greater than the known targets. The experiment design 2 different dynamic attitudes to test, 10% and 30% respectively. That is, 10 percent and 30 percent of the targets only have the experts' uncertain knowledge before the UAV begins reconnaissance. The test sets of C101, C106, R101, R106, RC101 and RC106 were randomly selected for modification and experiment. In the experiment, the points in the original test data set were randomly selected as the dynamic target points, and the appear moment was from the time window   0, i L . For each group of dynamic test data, the best results were selected through 30 operations, as shown in table 2 and  table 3.  It can be seen from table 2 and 3 that, from the perspective of optimal distance, the combination of two strategies is the best. The optimization results of RC-ABC algorithm and HB-ABC algorithm are different according to the test set dynamic attitude. On the whole, RC-HB-ABC algorithm has the best results in dynamic attitude test.
In order to evaluate the results, in 10% dynamic attitude, the convergence curve of the three tracking strategies to solve the R103 test set is shown in figure5.  Figure 5 shows that the search strategy of RC-HB-ABC algorithm combines the advantages of RC-ABC and HB-ABC, so in the case of time is higher pressing degree, the convergence speed is almost the same as the local search strategy of HB-ABC, in the case of long search time, it is possible to find a better solution than RC-ABC. The experimental results verify the validity of proposed algorithm and the feasibility of solving the multi-UAV scheduling problem.

Conclusion
This paper based on the model of multi-UAV scheduling in dynamic environment, an improved double evolutionary artificial bee colony is designed to solve this model. Through the simulation experiment, proves that the algorithm can completely detect all targets in the dynamic conditions to gain the maximum reconnaissance profit, and the UAV's quantity and flight distance are the minimum, minimize the flight cost. Finally, the algorithm was compared with RC-ABC and HB-ABC algorithm, and the experimental results showed that the proposed algorithm can improve the convergence speed and global search performance observably.