A heterogeneous helicopter fleet scheduling scheme hierarchical optimization method

Abstract

The application discloses a heterogeneous helicopter fleet scheduling scheme hierarchical optimization method, comprising the following steps: S1, important information in a disaster is summarized, and structured input task scenario information is input; S2, a scheduling problem model of the heterogeneous helicopter fleet is constructed, including two levels of a task division optimization model and a path planning problem model; S3, a data structure of a solution is designed, a task division matrix is taken as the data structure corresponding to the task division optimization model, and a helicopter task trajectory is taken as the data structure corresponding to the path planning problem model; S4, a path planning solver is designed in combination with the data structure of the solution and the path planning problem model; and S5, a task division matrix optimization solver is designed in combination with the data structure of the solution, the task division optimization problem model and the path planning solver. The method changes the scheduling scheme from a text type into a data type, and changes real-time scheduling into scheme making, and can effectively improve the scheduling scheme efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aviation emergency rescue and operations research (dispatch scheme optimization, path planning), and particularly relates to a method for hierarchical optimization of a macroscopic dispatch scheme for a heterogeneous helicopter fleet under large-scale natural disasters. BACKGROUND

[0002] Natural disasters often cause great economic losses, and post-disaster rescue is of high difficulty and complexity due to the vulnerability of traffic infrastructure to disasters. Aviation emergency rescue is an important part of natural disaster rescue, and helicopters are widely used in the rescue process of natural disasters due to their hovering, maneuvering flexibility, and easy deployment, and are usually used to perform tasks such as material dropping, personnel transfer, hoisting equipment, reconnaissance, and fire extinguishing. After a disaster occurs, the emergency management department needs to command emergency forces to coordinate rescue within the jurisdiction. For an air station, how to quickly develop a task dispatch scheme according to the rescue demand and resource information, combined with the characteristics of the helicopter fleet, and report a flight plan for the helicopter fleet is the most important problem, and is also an application problem of operations research optimization technology.

[0003] The dispatch scheme optimization problem is a branch of operations research optimization, and "parallel machine scheduling" is a type of problem that manufacturing plants often face in real life: a job is divided into several processes to complete. It allows the process to be processed by any one of a set of machine tools, and the purpose of scheduling is to allocate the process to each machine tool and sort the processes on each machine tool to minimize the time to complete all processes. Focusing on the optimization problem of the fleet dispatch scheme, internationally, Ozdamar L has studied the post-disaster helicopter logistics coordination system and proposed an interactive method for hierarchical analysis of post-disaster helicopter logistics transportation. This method can generate the most suitable crew / fleet configuration and flight route, and solve the personnel allocation, route and transportation problems encountered in the initial response stage of disaster management. However, the six levels proposed by this method are very complex, resulting in high computational complexity; at the same time, since Ozdamar L did not design a corresponding solution program, this theoretical method lacks a fast and efficient optimization algorithm, and it is difficult for a general commercial software to complete the calculation within a reasonable calculation time.

[0004] In the case of the task to be performed, the place to be visited by the helicopter is known, so how to make the scheduling scheme is simplified as how to design the task trajectory of the helicopter, which essentially belongs to the vehicle routing problem. In the forefront of this field, Ozkan based on simulated annealing and local search metaheuristic method, puts forward an algorithm using multi-base multi-unmanned aerial vehicle (UAV) to detect forest fire, which is the application of vehicle routing problem in the field of aviation emergency rescue, but the algorithm is only for the case of using unmanned aerial vehicle for search, and for more complex material transportation, personnel transfer problem is not involved.

[0005] In summary, the prior art mainly has the following deficiencies: first, the helicopter fleet macro-scheduling scheme includes demand splitting, resource and demand matching, task division of the helicopter fleet, and path planning of the helicopter. There is no overall optimization method for the whole macro-scheduling scheme. Second, the existing scheduling theory is rarely applied in the field of helicopters, and some of the simplified assumptions are unreasonable; some do not consider personnel transfer, material transfer, equipment hoisting and other complex operation modes; some only have theoretical basis without corresponding optimization algorithm; some do not consider the different task capabilities of different helicopters in the fleet. Third, the existing path planning theory does not simultaneously consider the characteristics of the helicopter "with endurance time limit" and "transferring from A to B in a fixed operation mode". SUMMARY

[0006] Currently, the models and optimization methods in the field of scheduling are more focused on single model, single task type, or fixed task (transferring from A to B) problems; while the scheduling problem of heterogeneous helicopter fleet addressed by the present application has the following difficulties:

[0007] (1) A demand can be divided into multiple parts, each of which can be provided by different resources; transferring 100 people from A to B can be assigned to multiple helicopters, or each helicopter may perform multiple batches. Therefore, the task allocation problem under complex task decomposition needs to be solved.

[0008] (2) Different helicopters have different cruising speeds and different task execution capabilities (capacity, speed), so the problem of how to arrange tasks for heterogeneous helicopter fleet according to the attributes of the helicopters needs to be solved.

[0009] (3) The helicopter has endurance limit and needs to return to the base for support within the specified time window; meanwhile, the transfer task cannot be interrupted. Therefore, the overall task trajectory planning problem of the helicopter under the condition of endurance limit and fixed task path needs to be solved.

[0010] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0011] A heterogeneous helicopter fleet scheduling scheme hierarchical optimization method, comprising the following steps:

[0012] S1, induct important information in disaster, structure input task assumption information, including the composition of the helicopter fleet and the performance parameters of each helicopter, the rescue demand of the aviation emergency rescue task, the location of the aviation emergency rescue task resource provider, and the geographical location information involved in the aviation emergency rescue task;

[0013] S2, according to the input task assumption information, construct the scheduling problem model of heterogeneous helicopter fleet, the model includes two levels of task division optimization model and path planning problem model;

[0014] S3, corresponding to the task division optimization model and the path planning problem model, design the data structure of the solution, take the task division matrix as the data structure corresponding to the task division optimization model, and take the helicopter task trajectory as the data structure corresponding to the path planning problem model;

[0015] S4, combining the data structure of the solution and the path planning problem model, designing the path planning solver: the solver uses branch and bound algorithm and adaptive ant colony algorithm to complete the calculation, which can determine the task trajectory of each helicopter according to the division determined by the task division matrix, so that the total flight trajectory of the helicopter to complete all tasks is the shortest;

[0016] S5, combining the data structure of the solution, the task division optimization problem model and the path planning solver, designing the solver of task division matrix optimization: the solver uses the heuristic operator of pseudo particle swarm algorithm to complete the calculation, realizing the optimization of task division matrix.

[0017] Further, in the step S2, the optimization objective function and constraint condition formula of the path planning problem model are:

[0018]

[0019]

[0020]

[0021]

[0022] wherein, t h is the time for helicopter h to complete all the tasks allocated to it; is the time for helicopter h to work during the task process; is the time for helicopter h to be on the road during the task process; N is the demand point set; K nh is the number of tasks in the task list that helicopter h needs to perform for demand n; mtnhk t nhk is the operation time of m is the time of helicopter h on the road during the task; v h is the cruising speed of the helicopter; H is the set of helicopters; is the remaining endurance time of helicopter h at l; d 0，1 is the distance between location l and the base (airport) (where the location number of the base is 0); p h is the task trajectory of helicopter h, which is the set of locations passed through; P h [i] represents the i-th location on the task trajectory of helicopter h; is the distance between the i-th location and the i+1-th location on the task trajectory of helicopter h.

[0023] Further, in the step S2, the optimization objective function and the constraint condition formula of the task division optimization model are:

[0024] Minimize t = max(t1, t2,..., t h ,...) (5)

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] where t h is the time for helicopter h to complete all the tasks assigned to it; m nhk is the task type of m nhk ; C h is the set of task types that helicopter h can perform; K nh is the number of tasks in the task list that helicopter h needs to perform for demand n; q n is the task amount of demand n; m nhk is the task load of m nhk ; m nhkr r nhkr = 0 / 1, if m nhk is provided by r, m nhkr = 1; c nh is the maximum single task capacity of helicopter h when performing the task corresponding to demand n; is the maximum endurance of helicopter h; m nhkfor m nhk the operation time; mpf nhk for m nhk the first location in the mission path; mpl nhk for m nhk the second location in the mission path.

[0031] Further, in the step S3, each row of the mission assignment matrix corresponds to a helicopter, and each column corresponds to a demand; the element in the mission assignment matrix is a mission list, which represents the mission list of the i-th helicopter assigned by the command center to execute the j-th demand; each mission in the mission list has four attributes: path, load, type, and execution time;

[0032] The helicopter mission trajectory is represented by a string of numbers, where each number represents a city.

[0033] Further, in the step S4, the path planning solver uses an adaptive ant colony algorithm in solving and uses a branch and bound algorithm in verifying the optimal solution.

[0034] Further, the adaptive ant colony algorithm converts the helicopter path planning into an ant colony path search optimization problem; Ant is a class, which has attributes including the current city l h , the remaining endurance time at the current city , the current trajectory p h , the total distance D h , and the remaining mission list ML Ant ; the final mission path of different Ants in Ants is regarded as a search of the best mission trajectory by the helicopter h, and is a feasible helicopter mission trajectory;

[0035] In the process of trajectory search by Ants, Ant first calculates the task selection probability Ant of m nhk in ML , and then selects the next task according to the principle of roulette; in each round of loop, all Ants in Ants are first initialized, and then each Ant finds a feasible mission trajectory according to the heuristic rule, and finally the individual with the shortest mission trajectory in the round of loop is found as the optimal Ant in the iteration process of the round.

[0036] Further, each node of the branch and bound algorithm is equivalent to an intermediate state of a helicopter during task execution, having four attributes of current node trajectory, current trajectory time consumption, uncompleted task list and remaining endurance time of the helicopter; starting from the root node, a new generation of nodes is generated by continuously branching until a branch completes all tasks, and the search tree generates a first end node; thereafter, the search tree will start pruning operation to stop the growth of intermediate nodes that do not meet the conditions until all the newly generated child nodes are end nodes, and the algorithm ends.

[0037] Further, in the step S5, the solver uses the heuristic operation operator of the pseudo particle swarm algorithm to complete the calculation, and realizes optimization of the task division matrix, including the following steps.

[0038] (1) generating P size initial task division matrices, each of which corresponds to a particle;

[0039] (2) evaluating all particles to solve the optimal task trajectory of all helicopters corresponding to the task division matrix;

[0040] (3) finding the particle P best with the shortest task completion time among all particles, and moving the particle swarm;

[0041] (4) repeating steps (2) and (3) I max times, outputting the optimal particle in the history record, and taking the task division matrix corresponding to the optimal particle as the final solution of the task division solver.

[0042] Compared with the prior art, the present application has the beneficial effects of:

[0043] 1. The present application models the post-disaster aerial emergency rescue scene, and defines the scheduling problem (MHFSP) of the heterogeneous helicopter fleet. Further, a hierarchical optimization method (HOMHFS) of the heterogeneous fleet scheduling scheme is proposed. The scheduling problem of the heterogeneous helicopter fleet is divided into two levels of task division and path planning, and the optimization result of the path planning level is used as the evaluation of the task division scheme. This establishes the model of the transfer task and the heterogeneous helicopter fleet under large-scale natural disasters, and through the splitting of the levels, the scheduling scheme of the fleet can be analyzed and optimized. It can be said that the scheduling scheme changes from text type to data type, and from real-time scheduling to scheme development.

[0044] 2. The encoding form of the task division matrix is designed for helicopters with different capability attributes and various types of task requirements, which can meet the complexity of demand decomposition and task allocation. Further, the invention designs the operation operator of the pseudo particle swarm algorithm in the process of solving the task division matrix optimization problem, so as to obtain the optimized task division and improve the efficiency of the scheduling scheme.

[0045] 3. The loading-unloading path planning problem of the helicopter under the endurance constraint is proposed, and the branch and bound method and the adaptive ant colony algorithm are applied to solve it, so as to obtain the optimized helicopter task trajectory and reduce the time of the helicopter to complete the task.

[0046] 4. The benefits of using the optimization method proposed in this paper are obvious: in the case of high complexity and high time pressure after natural disasters, the scheduling scheme optimization algorithm can provide decision support for the relevant leaders of the emergency management department and the regional airport, and assist in formulating the scheduling scheme of the helicopter fleet. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a hierarchical optimization method framework of the heterogeneous helicopter fleet macro-scheduling scheme.

[0048] Figure 2 is the data structure of the task division matrix.

[0049] Figure 3 is the average calculation time curve of the branch and bound algorithm and the adaptive ant colony algorithm when inputting different numbers of tasks.

[0050] Figure 4 is the absolute error and relative error performance of the adaptive ant colony algorithm when inputting different numbers of tasks.

[0051] Figure 5 is the strategy diagram of the non-optimal particle moving to the optimal particle in the pseudo particle swarm algorithm.

[0052] Figure 6 is the strategy diagram of the non-optimal particle disorder mutation in the pseudo particle swarm algorithm.

[0053] Figure 7 is the final helicopter task trajectory diagram corresponding to the calculation instance. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme of the invention more clear and clear, combined with the drawings. It should be understood that the specific embodiments described herein are only used to explain the invention, and are not used to limit the invention.

[0055] The invention provides a hierarchical optimization method for a heterogeneous helicopter fleet macro-scheduling scheme, the architecture and process of which are as followsFigure 1 The specific flow can be divided into the following steps as shown:

[0056] S1, important information in the disaster situation is summarized, and structured input task assumption is determined, specifically including the composition of the helicopter fleet and the performance parameters of each aircraft, the rescue demand of the aviation emergency rescue task, the location where resources for the aviation emergency rescue task can be provided, and the geographic location information involved in the aviation emergency rescue task.

[0057] S2, according to the input information, a scheduling problem model of the heterogeneous helicopter fleet is constructed. The model is divided into two levels of "path planning problem model" and "task division optimization model", and when it is specifically formulated into mathematical formula, it contains optimization objective function, constraint condition formula and all variable parameters used. The parameters include the set of helicopters, demand points, resource points and their attributes.

[0058] S3, the data structure of the solution is designed according to the problem model. Since the problem model is divided into two levels of "task division" and "path planning", the data structure needs to be designed accordingly. In this invention, in order to express the complex situation of heterogeneous fleet division, we propose "task division matrix" as the optimization object (data structure) corresponding to the task division level, and "helicopter trajectory" as the optimization object (data structure) corresponding to the "path planning".

[0059] S4, combined with the data structure of the solution and the path planning problem model, the path planning solver is developed: the solver can determine the task trajectory of each helicopter according to the division determined by the task division matrix, so that the total flight trajectory of the helicopter to complete all tasks is the shortest. This scheme adopts branch and bound algorithm and adaptive ant colony algorithm to complete the calculation.

[0060] S5, combined with the data structure of the solution, the task division optimization problem model and the path planning solver, the solver for task division matrix optimization is developed: the heuristic operator of the pseudo particle swarm optimization (PPSO) is designed in this invention. The task division matrix is optimized through this heuristic algorithm.

[0061] Further details of the ideas, difficulties and differences in the technical scheme:

[0062] The scheduling problem model of the heterogeneous helicopter fleet in S2 includes two levels of "path planning problem model" and "task division optimization model", and the parameter table required to be used is shown as follows.

[0063] S2.1, the variable parameter table used for modeling:

[0064] Table 1 Variable parameter table

[0065]

[0066]

[0067]

[0068] S2.2, the optimization objective function of the path planning problem model, the constraint condition formula is shown as follows:

[0069]

[0070]

[0071]

[0072]

[0073] Formula (1) is the objective function of the path planning problem, which represents the minimization of the time for the helicopter to complete all tasks, which is the time for the helicopter to perform the work during the task, plus the time on the road in its task trajectory. Formula (2) specifies that the work time of the helicopter h during the task is equal to the sum of the work time of all tasks m nhk assigned to it. Formula (3) shows that the time of the helicopter h on the road during the task is equal to the length of the task trajectory divided by the speed of the helicopter. In addition, formula (4) ensures that the helicopter has sufficient remaining fuel at any point in its task trajectory to return to the base for refueling and support.

[0074] S2.3, the optimization objective function of the task allocation optimization model, the constraint condition formula is shown as follows:

[0075] Minimize t = max (t1, t2,..., t h ,...) (15)

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] Formula (5) is the optimization objective of the task allocation matrix, which is to minimize the maximum value of the time for all helicopters to complete the task, where t h is the optimal value of the path planning problem in the previous subsection. Formula (6) ensures that the command center assigns the corresponding type of task to the helicopter h according to the task capacity of the helicopter. Formula (7) shows that the total demand of demand n is equal to the relevant task mnhk The sum of the task loads of all helicopters is less than or equal to the sum of the demands, so that the demands are satisfied. Equation (8) ensures that the command center does not require the resource point r to provide more resources than it can provide. Equation (9) requires that the task load of each helicopter for each task is less than or equal to the maximum task capacity of the helicopter. Equation (10) is a feasibility constraint for task assignment, which ensures that any task assigned to the helicopter h can be completed in a single mission of the helicopter.

[0082] In S3, the data structure of the task assignment matrix and the helicopter mission trajectory are mainly designed.

[0083] S3.1, as shown in FIG. 3, each row of the task assignment matrix corresponds to a helicopter, and each column corresponds to a demand. The elements in the task assignment matrix are task lists, which represent the task list of the command center for the jth demand assigned to the ith helicopter (hj). Each task in the task list has four attributes: "path", "load", "type", and "execution time". The structure of the task assignment matrix can adapt to mixed fleets, different task types, and highly flexible demand decomposition. Figure 2

[0084] S3.2, the helicopter mission trajectory can be represented by a string of numbers, such as "013065120", where each number represents a city.

[0085] S4 needs to determine the task trajectory that makes the total flight path of the helicopter to complete all tasks the shortest. For a helicopter, if its task list to be completed is determined, then its optimal path to complete all tasks is also determined. Then, due to the task background of the helicopter rescue, new constraints need to be added:

[0086] 1. The helicopter has a range and safety fuel limit, which is equivalent to a vehicle routing problem with time windows (VRPTW).

[0087] 2. The helicopter needs to pass through two locations when completing a task, rather than reaching a location in the traditional VRP. For example, when doing a material transfer task, the helicopter needs to first go to a warehouse to load materials, and then transport them to the disaster-stricken point with demand.

[0088] The following describes the technical solutions of two algorithms for solving the path planning problem and how to choose the two algorithms in view of the above difficulties (two special constraints):

[0089] ​S4.1, design improved branch and bound algorithm, obtain the precise solution of path planning. Branch and bound algorithm is a traversal algorithm, in the specific implementation, the whole feasible solution space is continuously divided into smaller and smaller subsets, that is, branches, and a lower bound or an upper bound (called bound) is calculated for the value of the solution in each branch. Further, the algorithm prunes according to the existing optimal value and the upper and lower bounds of the branch during exploration, thereby reducing the solution space that needs to be explored and accelerating the completion of traversal.

[0090] The parameters used in the application are shown in the following table:

[0091] Table 2 Parameters used in branch and bound algorithm

[0092]

[0093]

[0094] Among them, each node of the search tree corresponds to an intermediate state of a helicopter during task execution, and has four attributes of "current node trajectory", "current trajectory time consumption", "uncompleted task list", and "remaining endurance time of the helicopter". Starting from the root node, branching operation is continuously performed to generate a new generation of nodes, until a branch completes all tasks, and the search tree generates a first end node. Then, the search tree will start pruning operation to stop the growth of intermediate nodes that do not meet the conditions. Until all the newly generated child nodes are end nodes, the algorithm ends.

[0095] The branch and bound algorithm has the advantage of being able to ensure the optimal solution, but due to its traversal algorithm idea, the branch and bound algorithm has some defects in calculation time. When the number of tasks exceeds 9, it is impossible to complete the calculation on a computer with a memory of 16GB.

[0096] S4.2, design improved adaptive ant colony algorithm, obtain the approximate solution of path planning under small calculation amount. The algorithm can be mutually matched with the branch and bound algorithm.

[0097] The application designs an ant object and parameterized modeling: the helicopter path planning is converted into an ant colony path search optimization problem. Ant is a class, which has attributes including the current city l h , the remaining endurance time in the current city , the trajectory p walked so far h , the total distance D walked h , the remaining task list ML Ant . The final task path of different ants in Ants can be regarded as a search of the best task trajectory of the helicopter h, which is a feasible helicopter task trajectory. All the parameters used by the adaptive ant colony algorithm and their explanations are shown in the following table:

[0098] Table 3 Parameters used in adaptive ant colony algorithm

[0099]

[0100]

[0101] In the process of Ants exploring the trajectory and the branch and bound method, the next node is constantly searched for, which is similar to the difference that the process of Ant selecting the next task. Ants will first calculate the task selection probability of m Ant in the model nhk Then the next task is selected according to the principle of roulette. In each round of the algorithm, all Ants in the Ants are first initialized, and then each Ant finds a feasible task trajectory according to the heuristic rule, and finally finds the individual with the shortest task trajectory in the round as the optimal Ant in the iteration process.

[0102] Compared with the traditional ant colony algorithm, the main adjustments of the improved ant colony algorithm used in the application are as follows:

[0103] According to the constraint of the remaining endurance time of the helicopter in the model, a time window is set at all places in the Ant task trajectory. The time window requires Ants to ensure that they can return to the base within the remaining endurance time.

[0104] The min-max method is used to set dynamic upper and lower limits for the pheromone on the path, so as to guarantee the search ability of the heuristic algorithm and avoid the algorithm converging to a local optimal solution too early.

[0105] S4.3, determine the different time of using algorithm: combined Figure 3 It can be seen that when the number of tasks input into the algorithm increases, the calculation time of the branch and bound algorithm (BBA) rises exponentially, and when the number of input tasks is 10, the test device cannot complete the calculation due to memory limitation (RAM = 16GB). In comparison, the adaptive ant colony algorithm (AACA) changes little, and the calculation time increases linearly from 0.003s when there is 1 task to 0.014s when there are 9 tasks. Combined Figure 4 It can be seen that since BBA is a traversal algorithm, its output must be a global optimal solution, and on this basis, the calculation error of AACA is observed. The calculation error of 100 test inputs under each number of tasks is averaged to observe the error of AACA. The absolute error and relative error of AACA will increase with the increase of the number of tasks, but from the example, the order of magnitude of the relative error always remains around 0.01, which is within an acceptable range.

[0106] ​Combining the accuracy and calculation time cost of the two algorithms, it can be considered that AACA has a calculation time advantage, and the loss of accuracy is within an acceptable error range. Therefore, AACA can be used in model solving to speed up the model, and BBA with accuracy guarantee is used when verifying the optimal solution (when the number of tasks is not more than 9)

[0107] S5, finally, the data structure of the solution, the task division optimization problem model and the path planning solver need to be combined to develop a solver for task division matrix optimization. The following explains the heuristic operator of the pseudo particle swarm algorithm and the technical scheme of task division matrix optimization through this heuristic algorithm: Compared with the standard particle swarm algorithm (PSO), the pseudo particle swarm algorithm used in the present application can simplify many parameters such as individual extremum and particle speed. The main reference is the idea of "moving to the optimal particle" in PSO: we assume that the current optimal particle P best is optimal because its task division structure is reasonable and relatively plays the optimal performance of the unit system. Therefore, the task division matrix of the non-optimal particle P not best is adjusted by referring to the task division matrix of P best . The specific operation mode is "replace the task division scheme of an air emergency rescue demand as a whole".

[0108] The following will describe the specific implementation of the above-mentioned large-scale natural disaster oriented heterogeneous helicopter fleet scheduling scheme hierarchical optimization method provided by the present application through a specific embodiment.

[0109] Embodiment 1:

[0110] Step 1: input task scenario information for modeling. The attributes of the four types of objects included in the task scenario information must be determined, including:

[0111] Step 1.1: input the performance parameters of each aircraft, including: cruising speed, fuel consumption rate, maximum fuel capacity, maximum endurance time, whether it can hang large equipment, maximum cargo load, and maximum number of passengers.

[0112] Table 4 Aircraft performance parameters

[0113] Type 1 Type 2 Cruise speed 255 251 Fuel consumption 3000 1065 Maximum fuel 12000 3500 Maximum range 4 3.3 Whether can hang large equipment 1 0 Maximum material load 20000 4000 Maximum number of people 82 27

[0114] Step 1.2: input air emergency rescue demand information, including: demand point location (geographical location number), demand type, and demand quantity.

[0115] Table 5 Air emergency rescue demand information

[0116]

[0117] Step 1.3: Input aviation emergency rescue resource information, including: resource point location (geographical location number), resource type, and resource reserves.

[0118] Table 6 Information on Aviation Emergency Rescue Resources

[0119]

[0120] Step 1.4: Enter the geographic location information, including the location name, location number, location longitude, and location latitude.

[0121] Table 7 Geographic Location Information

[0122] name Code longitude latitude Yiwu Airport 0 120.03 29.34 Ninghai 1 121.41 29.28 Pan'an 2 120.45 29.05 Jinyun 3 120.09 28.65 Yueqing 4 120.97 28.11 Tiantai 5 120.98 29.17 Xianju 6 120.71 28.85 Taizhou 7 121.42 28.65 Wenzhou 8 120.70 28.00

[0123] Step 2: Use the pseudo-particle swarm optimization algorithm for the first level, i.e., optimization of the task allocation scheme. The initialization system parameters (obtained based on debugging experience during the development process according to this invention) are shown in the table below:

[0124] Table 8 Initialization system parameters for the pseudo-particle swarm algorithm

[0125] Parameter Value Meaning P size ]]> 50 Particle swarm size I max ]] 80 Maximum iteration number σ 0.9 Learning rate, non-optimal particles in particle swarm move to optimal particles with a probability of σ

[0126] The specific steps include:

[0127] Step 2.1: Generate P size There are two initial task allocation matrices, each corresponding to a particle. The generation of the task allocation matrix includes two levels. The first level is demand decomposition, which finds resource supply points for demand points. The second level is task allocation, which assigns helicopters to complete the transfer tasks of the "demand-resource" pairs generated in the first level.

[0128] Considering the different capacity constraints of different helicopters performing different tasks, this invention employs a simplified task allocation matrix generation algorithm. For each requirement, a helicopter with relevant operational capabilities is randomly selected, and a resource point with relevant resources is randomly selected. The task load for this mission is defined as the minimum value among "helicopter single-operation load," "remaining requirements," and "remaining resources."

[0129] Step 2.2: Evaluate all particles (corresponding to a task allocation matrix). This evaluation process is actually to solve for the optimal mission trajectory of all helicopters corresponding to the task allocation matrix, so as to obtain the maximum value of the mission completion time of all helicopters. This time value will be used as the evaluation of the task allocation matrix (the timeliness of the entire scheduling scheme).

[0130] Therefore, a second-level optimization solver for path planning is needed:

[0131] Step 2.2.1: Call adaptive ant colony algorithm, initialize system parameters (obtained according to debugging experience in the development process of the application) as shown in the table below, generate Ants according to the number of ant colonies:

[0132] Table 9 initialization system parameters of adaptive ant colony algorithm

[0133]

[0134]

[0135] Step 2.2.2: update all Ant objects to the initial state, its l h is the base (city code is 0)

[0136] Step 2.2.3: for each Ant, select the next task node and execute, update the state until its ML Ant is empty. First, let Ant calculate the probability of selecting the next remaining task according to the probability selection formula, note that before using the formula, first verify the endurance constraint of the helicopter, that is, ensure that the helicopter still has remaining fuel to return to the base after selecting the next task, at this time we call the task is accessible. The task selection probability of the task that cannot be accessed is Generally, the task selection probability is calculated as follows:

[0137]

[0138] Second, let Ant draw lots from all accessible tasks according to the probability of selecting the task, obtain the next task node and execute, update its own state, including ML Ant , l h and D h . Repeat the above task selection process until all Ants complete all tasks.

[0139] Step 2.2.4: update the pheromone matrix P m . Select the optimal ant with the shortest path from the ant colony Ants Put it in G B , and update the pheromone matrix P m of the algorithm according to , which is based on simulating the pheromone left by the ant. Mark the trajectory of , increase the probability of selecting the shortest trajectory in the future. P m is the accumulation of iteration experience in the entire algorithm process. The formula for updating the pheromone is as follows:

[0140]

[0141] Step 2.2.5: Dynamically and adaptively adjust the pheromone matrix P m The value of P. m The maximum pheromone value 'aa' in the matrix is ​​used to reduce pheromone values ​​greater than 0.8*aa to 0.8*aa, and to increase pheromone values ​​less than 0.2*aa to 0.2*aa. This ensures that the pheromone matrix does not converge prematurely, thereby optimizing the exploration range and improving the accuracy of the algorithm.

[0142] Step 2.2.6: Loop through the process of the ant colony exploring and updating the pheromone matrix. max Next, record the best ant in this process and output its trajectory as the final solution. This concludes the process of calling the adaptive ant colony algorithm.

[0143] Step 2.3: After obtaining the evaluations of all particles, find the particle P with the highest score (shortest task completion time, best timeliness). best Then, the particle swarm moves. The overall movement strategy of the particle swarm is:

[0144] Under a given learning rate σ, the non-optimal particle P notbest Reference P best The task allocation matrix needs to be adjusted, and the specific operation process is as follows: Figure 5 As shown, its macroscopic meaning is that the non-optimal particle P notbest From P best In the task allocation matrix, randomly select a task allocation scheme that corresponds to a specific requirement (corresponding to a column of the task allocation matrix) and replace itself (P). not best The task allocation matrix shows the task allocation scheme for this requirement.

[0145] Otherwise P not best Unordered mutations will be performed, with mutation operators such as... Figure 6 As shown. Its macroscopic meaning is to regenerate a new task allocation matrix (MAM). new Non-optimal particle P not best From MAM new In the task allocation matrix, randomly select a task allocation scheme that corresponds to a specific requirement (corresponding to a column of the task allocation matrix) and replace itself (P). notbest The task allocation matrix shows the task allocation scheme for this requirement.

[0146] Step 2.4: Repeat steps 2.2-2.3. max Next, the best particle in the historical record is output, and its corresponding task division matrix is ​​used as the final solution of the task division solver.

[0147] Step 3: According to the optimized task assignment matrix, the scheduling scheme of the corresponding helicopter fleet is solved, that is, the task order (trajectory) of each helicopter is obtained. At this point, the branch and bound algorithm can be used to obtain the exact solution.

[0148] Step 3.1: Initialize B t = 1000, F is an empty set, generate an initial node, and nodes = {iniNode}, nodes is a set of nodes that need to be grown in each generation.

[0149] Step 3.2: Determine whether nodes is empty, if it is empty, go to step 3.5. Otherwise, go to step 3.3.

[0150] Step 3.3: For each node node in nodes, generate a new generation of nodes tempNodes for node. First, determine the set of tasks that have not been completed in the state of the node, and then determine whether these tasks can be accessed in the current state of the node. The standard for accessibility is that after completing the task, it can still return to the base for support. It should be understood that in each node, it can always generate a node corresponding to "return to base support". Finally, tempNodes is added to newNodes as a new generation of nodes.

[0151] Step 3.4: Determine whether an end node endNote is generated in the newly generated generation of nodes newNodes, if it is, go to step 3.4.1; otherwise, go to step 3.2.

[0152] Step 3.4.1: Then put the task trajectory corresponding to these end nodes endNote into F, obtain the optimal task trajectory (corresponding to the shortest time) in F, and the corresponding time B t . For the task trajectory time corresponding to all nodes in the current new generation of nodes newNodes, if there is an intermediate node that has not completed all tasks, but the corresponding task trajectory time is greater than B t , then it can be inferred that the end node grown from this node cannot obtain a better solution than B t , so it is pruned and deleted. Assign newNodes after removing such nodes that do not meet the requirements to nodes, and go to step 3.2.

[0153] Step 3.5: Output the optimal trajectory B r in F and its corresponding time B t as the final solution of the branch and bound solver, and the algorithm ends.

[0154] Step 4: Output the current optimal task allocation matrix and the optimal task trajectory (sequence) for each helicopter. Optimization process ends.

[0155] The optimization results corresponding to the task assumption are given as follows:

[0156] The final task allocation matrix obtained is converted into a table form as shown below:

[0157] Table 10 Optimal task allocation table

[0158]

[0159]

[0160] The final task trajectory of each helicopter is shown in Table 11: Figure 7

[0161] The total task time for each helicopter is:

[0162] Table 11 Total task time for each aircraft

[0163] h1(Type 2) h2(Type 2) h3(Type 1) Task trajectory 0633501523034870 0688606363015150 028860320350 Total time (h) 9.688506906 9.463470399 9.321607074 ​

Claims

1. A hierarchical optimization method for heterogeneous helicopter fleet scheduling schemes, characterized in that, The method comprises the following steps: S1, important information in the disaster situation is summarized, structured input task assumption information is input, including the composition of the helicopter fleet and the performance parameters of each helicopter, the rescue demand of the aviation emergency rescue task, the location of the aviation emergency rescue task, and the geographical location information involved in the aviation emergency rescue task; S2, according to the input task assumption information, a scheduling problem model of the heterogeneous helicopter fleet is constructed, the model comprises a task division optimization model and a path planning problem model at two levels; S3, corresponding to the task division optimization model and the path planning problem model, the data structure of the solution is designed, the task division matrix is taken as the data structure corresponding to the task division optimization model, and the helicopter task trajectory is taken as the data structure corresponding to the path planning problem model; S4, combining the data structure of the solution and the path planning problem model, a path planning solver is designed: the solver completes the calculation by using a branch and bound algorithm and an adaptive ant colony algorithm, can determine the task trajectory of each helicopter according to the division determined by the task division matrix, and makes the total flight trajectory of the helicopters to complete all tasks shortest; S5, combining the data structure of the solution, the task division optimization problem model and the path planning solver, a task division matrix optimization solver is designed: the solver completes the calculation by using a heuristic operator of a pseudo particle swarm algorithm, and realizes the optimization of the task division matrix; In the step S2, the optimization objective function and the constraint condition formula of the path planning problem model are as follows: where t h is the time for the helicopter h to complete all the tasks it is assigned; is the time for the helicopter h to work on the tasks; is the time for the helicopter h to be on the road during the tasks; N is the set of demand points; K nh is the number of tasks in the list of tasks that the helicopter h needs to perform for the demand n; m nhk is the number of tasks in the list of tasks that the helicopter h needs to perform for the demand n; m nhk is the working time of m nhk ; is the time for the helicopter h to be on the road during the tasks; v h is the cruising speed of the helicopter; H is the set of helicopters; is the remaining endurance time of the helicopter h at time l; d 0,l is the distance between the location l and the base, where the location number of the base is 0; p h is the task trajectory of the helicopter h, which is the set of locations it passes through; P h [i] represents the i-th location on the task trajectory of the helicopter h; is the distance between the i-th location and the i+1-th location on the task trajectory of the helicopter h; In the step S2, the optimization objective function and the constraint condition formula of the task division optimization model are as follows: Minimize t = max(t1, t2,..., t h ,...) (5) where t h is the time for the helicopter h to complete all the tasks it is assigned; mc nhk is the category of the task m nhk ; C h is the set of task categories that the helicopter h can perform; K nh is the number of tasks in the list of tasks that the helicopter h needs to perform for the demand n; q n is the amount of tasks for the demand n; mq nhk is the task load for m nhk ; mr nhkr , mr nhkr = 0 / 1, if m nhk is provided by r, mr nhkr = 1; c nh is the maximum single task capacity of the helicopter h when performing the task corresponding to the demand n; is the maximum endurance of the helicopter h; mt nhk is the operation time for m nhk ; mpf nhk is the first location in the task path for m nhk ; mpl nhk is the second location in the task path for m nhk ; In the step S3, each row of the task division matrix corresponds to a helicopter, and each column corresponds to a demand; the elements in the task division matrix are task lists, which represent the task list of the i-th helicopter arranged by the command center to execute the j-th demand; each task in the task list has four attributes of path, load, type and execution time; The helicopter task trajectory is expressed by a string of numbers, wherein each number represents a city.

2. The heterogeneous helicopter fleet dispatch scheme hierarchical optimization method of claim 1, wherein, In the step S4, the path planning solver uses the adaptive ant colony algorithm in solving, and uses the branch and bound algorithm in verifying the optimal solution.

3. The heterogeneous helicopter fleet dispatch scheme hierarchical optimization method of claim 2, wherein, The adaptive ant colony algorithm converts the helicopter path planning into an ant colony path search optimization problem; Ant is a class, which has attributes including the current city l h The remaining endurance time in the current city The current trajectory p h The total distance D h The remaining task list ML Ant The final task path of different ants in Ants is regarded as a search of the helicopter h for the best task trajectory, which is a feasible helicopter task trajectory; During the trajectory exploration process, Ant first calculates the ML. Ant m nhk Task selection probability Then, the next task is selected according to the principle of roulette; in each round of the loop, all Ants in Ants are initialized first, and then each Ant searches for a feasible task trajectory according to the heuristic rule. Finally, the individual with the shortest task trajectory in this round of the loop is found as the optimal Ant in this round of iteration.

4. The heterogeneous helicopter fleet dispatch scheme hierarchical optimization method of claim 2, wherein, The branch and bound algorithm, each node of the search tree is equivalent to an intermediate state of a helicopter in the process of executing a task, has four attributes of current node trajectory, current trajectory time consumption, uncompleted task list and helicopter remaining endurance time; starting from the root node, a new generation of nodes is generated by continuously performing branching operation, until a branch completes all tasks, and the first end node of the search tree is generated; then, the search tree starts pruning operation, stops the growth of the intermediate nodes that do not meet the conditions, until all the generated child nodes are end nodes, and the algorithm ends.

5. The heterogeneous helicopter fleet dispatch scheme hierarchical optimization method of claim 2, wherein, In the step S5, the solver completes the calculation by using the heuristic operator of the pseudo particle swarm algorithm, realizes the optimization of the task division matrix, and comprises the following steps: (1) generating P size initial task division matrices, each task division matrix corresponding to a particle; (2) all particles are evaluated, and the optimal task trajectory of all helicopters corresponding to the task division matrix is solved; (3) find the particle P with the shortest time to complete the task among all particles best , move the particle swarm; (4) repeat steps (2) and (3) for i max Finally, the optimal particle in the output history is output, and the corresponding task division matrix is taken as the final solution of the task division solver.

Images