A task allocation and scheduling method for heterogeneous unmanned systems in a dynamic marine environment

By using environmental energy consumption prediction model and non-dominant sorting genetic algorithm in dynamic marine environments, combined with Kalman filtering algorithm, the problems of low computational efficiency and unconsidered environmental energy consumption in unmanned system tasks are solved, achieving more accurate and A feasible task allocation plan.

CN118534860BActive Publication Date: 2024-11-08NANJING UNIV OF INFORMATION SCI & TECH
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202410993957.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-08
Estimated Expiration
2044-07-24

AI Technical Summary

Technical Problem

It is difficult to effectively perform task allocation and scheduling of heterogeneous unmanned systems in dynamic maritime environments, especially when facing changing environments and emergencies, the computing efficiency is low and the reasonable and effective task allocation plan cannot be made. Moreover, traditional methods have not fully considered the problems of environmental energy consumption and drone losses.

Method used

Using CNN-LSTM-SE-BO and a non-dominant sorting genetic algorithm based on the environmental energy consumption prediction model CNN-LSTM-SE-BO and a parallel optimization of multiple groups, environmental energy consumption is predicted through convolutional neural networks, long and short-term memory networks, attention mechanisms and Bayesian models. It also combines the Kalman filtering algorithm to update in real time, build a cooperative multi-task allocation model, and optimize the task allocation plan.

Benefits of technology

Improve the accuracy and feasibility of task allocation results, comprehensively consider the heterogeneity of unmanned systems, the attributes of task objectives and the instability of dynamic environments, optimize task execution time and energy consumption, and enhance the diversity of algorithms and Adaptability reduces the error between the predicted value and the real value.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118534860B_ABST
    Figure CN118534860B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment. The method predicts environmental energy consumption information by constructing an environmental energy consumption prediction model; constructs a cooperative multi-task allocation model for multiple heterogeneous unmanned systems to perform tasks on multiple stationary ground targets, uses the predicted environmental energy consumption information as model input, and matches the task execution time period and the energy consumption prediction time period in combination with the driving speed of the unmanned system in the task plan; based on the cooperative multi-task allocation model, a non-dominated sorting genetic algorithm with an elite strategy is used to determine the optimal individual as the final task pre-allocation plan; executes the task pre-allocation plan, and dynamically updates the task time period according to the current actual task execution situation, and optimizes the task and allocation plan. The present invention improves the accuracy of energy consumption prediction for task allocation results, and optimizes the task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment with task execution time and task execution energy consumption as the target.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle technology, and in particular to a method for allocating and scheduling tasks of heterogeneous unmanned aerial vehicles (UAVs) in a dynamic marine environment. Background Art

[0002] At present, algorithms used in task allocation and scheduling, such as simulated annealing algorithm, ant colony algorithm, particle swarm optimization algorithm and improved genetic algorithms, can serve the collaborative combat task allocation of heterogeneous unmanned systems to a certain extent. However, in the face of dynamic maritime mission environments, their static task allocation strategies, single-objective optimization and the limitations of ignoring the impact of the changing maritime environment on task allocation results are particularly obvious. Although traditional genetic algorithms are suitable for multi-objective optimization problems, they have low computational efficiency in large-scale and complex task allocation scenarios, and in highly dynamic scenarios, if emergencies are not considered, they may not be able to make reasonable and effective task allocation plans. When faced with multi-factor constraints and uncertainties in similar maritime mission environments, basic genetic algorithms may not be able to accurately simulate the impact of the real environment on task execution, thereby affecting the feasibility and optimality of the final task allocation plan.

[0003] Chinese patent CN117764188A discloses a method for allocating unmanned aerial vehicle (UAV) swarm tasks based on a quantum annealing algorithm model, which transforms the UAV swarm task allocation problem into a single-person traveling salesman problem, and solves the system quantum state corresponding to the optimal solution by combining the quantum annealing algorithm with the UAV swarm task allocation problem. However, this method only makes a pre-planned task allocation scheme for static scenes, and does not consider the impact of dynamic environmental factors on the task allocation results under the time scale, so it does not have the ability to adjust and re-plan in real time. Chinese patent CN115525068A discloses a method for allocating unmanned aerial vehicle (UAV) swarm collaborative tasks based on iterative optimization, which obtains the task revenue information of the effective future tasks of the UAV swarm according to the task information of the current tasks and future tasks obtained from the future task planning of the UAV swarm, and obtains the task revenue information of the effective future tasks of the UAV swarm, and uses the iterative auction mechanism to obtain the task allocation result of the UAV swarm. However, this method only considers the optimization of a single objective function of the total revenue of task execution, and does not consider the problem that the loss of UAVs will increase from the cost while increasing the total revenue of task execution. Chinese patent CN115562336A discloses a strategy based on the quantum coyote optimization mechanism to solve the problem of multi-UAV collaborative task allocation. This method can assign task objectives to each UAV in the formation under the constraints of three-dimensional scenes and time synchronization. However, when dealing with large-scale UAV formations and complex task environments, it consumes a lot of computing resources and time, and the actual application effect is not good. Summary of the invention

[0004] Purpose of the invention: The present invention aims to provide a heterogeneous unmanned system task allocation and scheduling method in a dynamic marine environment based on an environmental energy consumption prediction model and a non-dominated sorting genetic algorithm with multi-population parallel optimization.

[0005] Technical solution: The method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment described in the present invention comprises the following steps:

[0006] (1) Obtain environmental information in the mission area and construct a marine environment dataset;

[0007] (2) Construct an environmental energy consumption prediction model CNN-LSTM-SE-BO to predict the environmental energy consumption information of the time period that meets the task duration; CNN-LSTM-SE-BO includes a convolutional neural network CNN, a long short-term memory network LSTM, an attention mechanism SE module, and a Bayesian model BO;

[0008] (3) Based on the heterogeneity constraints of the unmanned systems and the task coupling constraints of the surface scenes, a cooperative multi-task allocation model CMTAP is constructed for multiple heterogeneous unmanned systems to perform tasks on multiple stationary ground targets. The environmental energy consumption information predicted in step (2) is used as the model input. Combined with the driving speed of the unmanned system in the mission plan, the task execution time period and the energy consumption prediction time period are matched.

[0009] (4) Based on the cooperative multi-task allocation model CMTAP, the non-dominated sorting genetic algorithm NSGA-II with elite strategy is used to determine the optimal individual as the final task pre-allocation solution;

[0010] (5) Execute the task pre-allocation plan determined in step (4), collect current environmental information, combine it with the predicted environmental information of the corresponding time period, calculate the rate of change of environmental variables, use the Kalman filter algorithm to update the noise covariance matrix in real time, dynamically update the task time period according to the current actual task execution status, and optimize the task and allocation plan.

[0011] Furthermore, step (2) is as follows:

[0012] After normalizing the marine environment dataset, it is divided into time series sets;

[0013] Divide the time series set into a training set and a sample set;

[0014] The environmental energy consumption prediction model CNN-LSTM-SE-BO is constructed, and the genetic algorithm is introduced to optimize the model hyperparameters and adjust the weights. The objective function is to minimize the prediction error loss function. Among them, the attention mechanism SE module is used to extract the environmental parameter characteristics that affect the energy consumption of the unmanned system and recalibrate them into energy consumption feature channels. The Bayesian model BO is used to predict the energy consumption of the unmanned system, identify the key environmental factors that affect energy consumption, and feed back to the attention mechanism SE module to re-determine the environmental parameter characteristics that affect the energy consumption of the unmanned system.

[0015] After the training model is completed, the predicted environmental energy consumption information for the time period that meets the task duration is output.

[0016] Furthermore, in step (2), the objective function of the Bayesian model BO is to minimize the mean square error of energy consumption prediction within a unit time window: :

[0017] ;

[0018] ;

[0019] ;

[0020] in, For the i The actual energy consumption in each time window; Under the parameter set θ, i The predicted energy consumption of time windows; n is the total number of time windows; Represents the expected value of the mean square error under different parameters θ; is the optimal parameter set that minimizes the mean square error obtained through iterative search of the Bayesian model; is the channel after recalibration; It is the Squeeze compression function, which is used to compress the global information of the input features; is the input feature channel; Sigmoid activation function is used to limit the value of the feature vector between 0 and 1; and is the weight matrix of the fully connected layer, which is used to reduce and increase the dimension of the input feature vector; is the ReLU activation function.

[0021] Furthermore, step (3) is as follows:

[0022] Establish a mission target model and define a set of multiple maritime mission targets; perform reconnaissance mission R, attack mission S and inspection mission V on each target in turn;

[0023] Establish a heterogeneous unmanned system model, which includes reconnaissance drones that can perform reconnaissance missions R and inspection missions V , combat drones capable of performing attack missions S and military unmanned boats that can perform all kinds of missions ;

[0024] Establish task allocation constraints;

[0025] Establish the coordinate point location information, determine the Euclidean distance between each point, the total distance of three different tasks of a certain target point, and the total distance of the task allocation plan based on the coordinates of the unmanned system starting point and the coordinates of multiple target points;

[0026] Establish a speed model for heterogeneous unmanned agents and calculate the task completion time.

[0027] Furthermore, the task allocation condition is constrained as

[0028] ;

[0029] ;

[0030] in, is the number of heterogeneous unmanned systems, Configure the total number of configuration points and mission target points for heterogeneous unmanned systems, is a binary decision variable, if , then it means that the drone From the first configuration Fly to the second configuration Execution against target The value of m is 1, 2, and 3, which represent the reconnaissance task R, the attack task S, and the inspection task V, respectively. Indicates the number of tasks executed at each task target point, .

[0031] Furthermore, step (4) is as follows:

[0032] Using a multi-type gene chromosome encoding method, the task execution order, task target point, task type, heterogeneous unmanned intelligent agent executor sequence number, unmanned intelligent agent movement speed, task execution time, task time period and intelligent agent energy consumption are used to construct the initial population, which is divided into multiple sub-populations;

[0033] Define the individual value of non-dominated sorting optimization standard as the screening standard of the migration population, use the elite strategy to retain the best individuals, use the roulette method to select adaptive multi-point crossover and mutation, and iterate the sub-population;

[0034] Determine whether the maximum number of iterations has been reached:

[0035] If the maximum number of iterations is reached, the iteration ends, all sub-population output results are merged, and the final results are non-dominated sorted;

[0036] If the maximum number of iterations has not been reached, it is determined whether it is dominated by the standard individual value of the non-dominated sorting optimization. If so, the individual is migrated into the migration population; if not, the next iteration is carried out directly. At the same time, before each sub-population performs the next generation of iterative optimization, it selects individuals from the non-dominated solution set from the migration population to enter the next generation of iterative optimization of each sub-population.

[0037] Furthermore, in the non-dominated sorting genetic algorithm NSGA-II with elite strategy, the fitness of each subpopulation is calculated as

[0038] ;

[0039] ;

[0040] in, is the fitness function of the task execution time, The fitness function for the total energy consumption of the task allocation scheme; , , , is the weight; is the total task execution time, is the total energy consumption of the task; is the total distance of the task allocation solution; Q represents the constraint violation evaluation coefficient.

[0041] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. The present invention constructs an environmental energy consumption prediction model CNN-LSTM-SE-BO through a convolutional neural network, a long short-term memory network, an attention mechanism SE module and a Bayesian model BO, extracts environmental features from the collected marine environmental data set, and predicts environmental energy consumption information in the future. The attention mechanism SE module focuses on extracting the features of environmental parameters that affect the energy consumption of the unmanned system. The Bayesian model BO predicts the energy consumption of the unmanned system, identifies key environmental factors that affect energy consumption, and feeds back to the attention mechanism SE module to re-determine the features of environmental parameters that affect the energy consumption of the unmanned system, thereby improving the accuracy of energy consumption prediction for task allocation results; 2. The present invention adopts a multi-type gene chromosome encoding method to construct an initial population that meets the heterogeneity and task coupling constraints of the unmanned system, taking into account the The heterogeneity between unmanned systems, the attributes of various task objectives, the instability of the dynamic environment and the dynamic adjustment of the deployment information of the unmanned system are used to optimize the two objectives of task execution time and task execution energy consumption; 3. The present invention introduces multi-subpopulation parallel computing and subpopulation migration optimization strategies to the NSGA-II algorithm, which enhances the diversity and retention ability of the algorithm, helps to attract excellent individuals, enrich the diversity of the population, and solves the problem of insufficient robustness of the NSGA-II algorithm when facing complex, multi-peak or highly nonlinear multi-objective optimization problems, while improving the algorithm's exploration ability and adaptability in different solution space areas; 4. The present invention calculates the rate of change of environmental variables and uses the Kalman filter algorithm to update the noise covariance matrix in real time to achieve adaptive recognition of temperature, air pressure, wind speed and humidity in complex marine environments, thereby reducing the error between the predicted value and the true value. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flowchart for training the environmental energy consumption prediction model CNN-LSTM-SE-BO;

[0043] Figure 2 It is a flow chart of the NSGA-II algorithm of the present invention. DETAILED DESCRIPTION

[0044] The following is combined with Figures 1 to 2 The present invention is further described.

[0045] The method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment of the present invention comprises the following steps:

[0046] (1) Obtain environmental information in the mission area and construct a marine environment dataset.

[0047] The mobile monitoring platform is used to monitor the temperature, air pressure, wind speed, and humidity data for the mission area or nearby areas every 3 minutes, and the monitoring period is the latest 10 to 20 days in the current area. Since the mission execution area is a small area within a short period of time, the environmental information of different target points at the same time can be regarded as the same. The environmental data samples monitored by the mobile monitoring platform are large enough, and the environmental data sequence has periodicity and volatility, which can meet the environmental prediction of the model of the present invention. The choice of monitoring month depends on the month in which the mission execution time period is located. At the same time, each environmental data is normalized to ensure that all characteristic values ​​are on the same time scale, which is convenient for the training of subsequent models.

[0048] (2) Construct an environmental energy consumption prediction model CNN-LSTM-SE-BO to predict the environmental energy consumption information of the time period that meets the task duration; wherein, CNN-LSTM-SE-BO includes a convolutional neural network CNN, a long short-term memory network LSTM, an attention mechanism SE module and a Bayesian model BO; the attention mechanism SE module is used to extract the environmental parameter characteristics that affect the energy consumption of the unmanned system and recalibrate them into energy consumption feature channels; the Bayesian model BO is used to predict the energy consumption of the unmanned system, identify the key environmental factors that affect the energy consumption, and feed back to the attention mechanism SE module to re-determine the environmental parameter characteristics that affect the energy consumption of the unmanned system.

[0049] (21) Perform normalization preprocessing on the environmental data set, split the continuous environmental data into input sequence X and target output y in a time series manner, and set the time window size to , and obtain a time series set.

[0050] (22) The time series set is divided into a training set and a test set in a ratio of 8:2, and the data order is not disrupted by default to maintain the continuity of the time series.

[0051] (23) Construct an environmental energy consumption prediction model CNN-LSTM-SE-BO. First, apply a one-dimensional convolutional layer to capture local features and potential time series patterns in environmental information data. Then, construct an LSTM layer to process long-term dependencies in time series data. Set the hidden layer size and use a single-layer LSTM structure. Then construct the SE module to dynamically adjust the importance of features. Introduce the Bayesian model BO to predict the energy consumption of unmanned systems, identify key environmental factors that affect energy consumption, and feed back to the attention mechanism SE module to re-determine the environmental parameter characteristics that affect the energy consumption of unmanned systems. Finally, a fully connected layer is used to output the predicted values ​​of various environmental parameters.

[0052] Among them, the Bayesian model BO is constructed as follows: in scenarios with different temperatures, air pressures, wind speeds, humidity, and different driving speeds, the energy consumption of unmanned intelligent agents of the same model, power supply, and load conditions as the task-performing drones and unmanned boats is tested. The collected data include temperature, air pressure, wind speed, humidity, drone driving speed, and drone energy consumption within a unit time window of 3 minutes. The data is preprocessed to remove outliers, complete data normalization, and divide the training set and test set. Define the Bayesian regression model, and use the training set data for training. After the training is completed, use the test set to predict energy consumption, and use the mean square error to evaluate the model performance. Finally, obtain the feature weights in the Bayesian regression model, analyze the impact of each feature on the energy consumption prediction, feed back the features with the largest impact to the SE module, adjust the SE module in the LSTM, and recalibrate the corresponding feature channels, thereby improving the accuracy of energy consumption prediction for task allocation results. The objective function of the energy consumption model is defined as minimizing the mean square error of energy consumption prediction within the unit time window. :

[0053] ;

[0054] ;

[0055] ;

[0056] in, For the i The actual energy consumption in each time window; Under the parameter set θ, i The predicted energy consumption of time windows; n is the total number of time windows; Represents the expected value of the mean square error under different parameters θ; is the optimal parameter set that minimizes the mean square error obtained through iterative search of the Bayesian model; is the channel after recalibration; It is the Squeeze compression function, which is used to compress the global information of the input features; is the input feature channel; Sigmoid activation function is used to limit the value of the feature vector between 0 and 1; and is the weight matrix of the fully connected layer, which is used to reduce and increase the dimension of the input feature vector; is the ReLU activation function.

[0057] (24) Genetic algorithm is introduced to optimize hyperparameters. The population size, mutation rate, crossover rate and other parameters of the genetic algorithm are set. The learning rate, LSTM hidden state dimension, number of LSTM layers and reduction ratio of SE layer are encoded as individual genes. Finally, the fitness function is defined and the minimization of the prediction mean square error loss function MSE is optimized as the objective function. The genetic algorithm is iterated and the individual with the highest fitness is selected as the optimal hyperparameter configuration.

[0058] (25) Define the training loop, including the optimizer and loss function. The optimizer uses the Adam optimizer to adjust the model parameters, and sets the learning rate, LSTM hidden state dimension, number of LSTM layers, and reduction ratio of the SE layer according to the optimization results of steps 2-4. The mean square error (MSE) is also used as the loss function to measure the difference between the predicted value and the actual value. In each training cycle, the generalization ability of the model is increased by shuffling the order of the training data, and the model is trained in small batches. The loss is calculated by forward propagation and the weights are updated by back propagation. indivual epoch Record the training and test losses once to evaluate the model performance.

[0059] (3) According to the heterogeneity constraints of the unmanned systems and the task coupling constraints of the surface scenes, a cooperative multi-task allocation model CMTAP is constructed for multiple heterogeneous unmanned systems to perform tasks on multiple stationary ground targets. The environmental energy consumption information predicted in step (2) is used as the model input. Combined with the driving speed of the unmanned system in the mission plan, the task execution time period and the energy consumption prediction time are matched.

[0060] Secondly, the environmental energy consumption information jointly predicted in step (2) is combined with the energy consumption predicted by the speed of the drone and unmanned boat in the mission plan to match the time period of each task execution time. The specific steps are as follows:

[0061] (31) Establish a mission objective model and define multiple maritime mission objectives. The objective set T is defined as

[0062] ;

[0063] in, is the total number of mission objectives.

[0064] The task type R of the same task target can only be executed after S is completed, and the task type V can only be executed after S is completed.

[0065] The number of tasks executed on each target is indivual.

[0066] .

[0067] (32) Establish a heterogeneous unmanned system model. Heterogeneous unmanned systems can be divided into three types, including reconnaissance drones that can perform reconnaissance missions R and inspection missions V. , combat drones capable of performing attack missions S And military unmanned boats that can perform all types of missions R, S, V Therefore, the overall unmanned system set can be expressed as

[0068] ;

[0069] in, Indicates the total number of heterogeneous unmanned system configurations. According to different types of unmanned systems, heterogeneous unmanned systems are further constructed:

[0070] ;

[0071] ;

[0072] in, , , Respectively represent the number of reconnaissance drones, combat drones and military unmanned boats. A collection of heterogeneous unmanned systems that can perform reconnaissance and inspection missions is defined. A set of heterogeneous unmanned systems with attack mission S is defined. At the same time, the number of heterogeneous unmanned systems in the set satisfies:

[0073] ;

[0074] at the same time Indicates the total number of unmanned system configuration points and mission target point configurations. Establish task allocation condition constraints:

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] Any single task representing any target point should be assigned only once. is a binary decision variable. If , then it means that the drone From the first configuration Fly to the second configuration Execution against target The value of m is 1, 2, and 3, which represent the three tasks of reconnaissance task R, attack task S, and inspection task V respectively.

[0080] Indicates the number of tasks executed at each task target point Should be exactly equal to 3.

[0081] It means that the execution of reconnaissance task R, attack task S, and inspection task V should follow strict task sequence constraints.

[0082] (33) Establish the coordinate point location information. Calculate the Euclidean distance between each point based on the coordinates of the unmanned system starting point and the coordinates of multiple target points. Select the sum of the distances between the three task points calculated under the target point benchmark as the distance of the target point. The Euclidean distance between two points can be expressed as

[0083] ;

[0084] in, and Represents two different location points, including the starting point and target point of the heterogeneous unmanned system.

[0085] The total distance of three different tasks for a certain target point can be expressed as

[0086] ;

[0087] in, Respectively represent the execution i The R, S, and V tasks under the task target point are the distances from the previous position of the heterogeneous unmanned intelligent agent to the current task target point. The sum of the total distances of all target points is the total distance of this task allocation plan. The total distance cost of the task allocation plan can be expressed as

[0088] ;

[0089] Among them, NT is the total number of target points.

[0090] (34) Calculate the task completion time. Establish speed models for three heterogeneous unmanned intelligent agents:

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] in, They respectively represent the speeds of reconnaissance drones, combat drones and military unmanned boats. To detect the speed range of drones, is the speed range of combat drones, This is the speed range of military unmanned boats.

[0096] Modeling task completion time:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] Among them, T is the total time to complete the task, i is the execution of the i-th target point task, is the total distance traveled by the i-th reconnaissance drone, is the total distance traveled by the i-th combat drone, is the total distance traveled by the i-th military unmanned boat. The total time of this task allocation plan is the total time of all tasks minus the time of parallel tasks in the same time. This time, the unmanned agent executor is considered to have completed the task when it reaches the target point, so the total completion time only considers the driving time of the agent.

[0103] (4) Based on the cooperative multi-task allocation CMTAP model, the non-dominated sorting genetic algorithm NSGA-II with elite strategy is used to determine the optimal individual as the final task pre-allocation scheme.

[0104] (41) Construct the chromosome according to the model and constraints constructed in step (3) and initialize the parameters. Each chromosome individual in the population has three benchmark parameters, namely, the task execution order, the task target point, and the heterogeneous unmanned intelligent agent executor serial number. Construct the remaining parameter objects according to the three benchmark parameters: construct three types of tasks based on each target point according to the task target point; construct the movement speed of each heterogeneous unmanned intelligent agent executor according to the heterogeneous unmanned intelligent agent executor serial number, and calculate the execution time of each task according to the distance between the execution task points assigned by the task, and assign the time period of the environmental information to correspond to the time period of the executed task; finally, calculate the energy consumption of the intelligent agent according to the movement speed of the intelligent agent and the environmental information in the time period. In this way, a complete chromosome individual is constructed and the parameters are initialized.

[0105] (42) Construct and initialize the population, and divide the initial population into multiple sub-populations. Calculate the fitness value for each sub-population:

[0106] ;

[0107] ;

[0108] in, is the fitness function of the task execution time, The fitness function for the total energy consumption of the task allocation scheme; , , , is the weight; is the total task execution time, is the total energy consumption of the task; is the total distance of the task allocation plan; Q represents the constraint violation evaluation coefficient, and its specific values ​​are as follows:

[0109] ;

[0110] Among them, C means satisfying the constraint condition, is a conditional function. If the constraint C is satisfied, that is, there is no violation of the constraint, then , , otherwise .

[0111] Define the non-dominated sorting optimization standard individual value as the screening standard for the migration population, then use the elite strategy to retain the best individual, and use the roulette method to select the three benchmark parameters, multi-point adaptive crossover and mutation (that is, individuals with high fitness values ​​have a low probability of crossover and mutation), and then determine whether the maximum number of iterations has been reached. If not, determine whether it is dominated by the non-dominated sorting optimization standard individual value. If so, migrate the individual into the migration population, if not, directly proceed to the next iteration; at the same time, each sub-population will select individuals from the non-dominated solution set from the migration population before the next generation of iterative optimization to enter the next generation of iterative optimization of each sub-population. When the maximum number of iterations is reached, the iteration ends, all sub-populations are merged to output the results, and the final results are non-dominated sorted.

[0112] (43) Combining the chromosome corresponding to the final non-dominated solution set with the needs of the actual situation, the appropriate optimal individual is selected as the final task allocation pre-allocation plan and output.

[0113] (5) Execute the task pre-allocation plan determined in step (4), collect current environmental information, combine it with the predicted environmental information of the corresponding time period, calculate the rate of change of environmental variables, use the Kalman filter algorithm to update the noise covariance matrix in real time, dynamically update the task time period according to the current actual task execution status, and optimize the task and allocation plan.

[0114] The ground station control system includes the host side of an industrial computer or a personal host. The heterogeneous unmanned system includes a drone, an unmanned boat, a control mainboard, a data transmission module, a data storage module, a GPS positioning module, an IMU attitude module, an environmental monitoring module, and a barometer module for the drone system. The host side contains control software, and the software supports multiple communication protocols. The software is used to communicate with the heterogeneous unmanned system to obtain information such as the attitude and position of the unmanned intelligent body. At the same time, the task allocation requirements can be sent to each unmanned intelligent body task executor in the form of instructions. The executor receives the signal through the data transmission module and controls the unmanned intelligent body to perform the task through the control mainboard. In the process of executing the task, the unmanned system executor uses its own monitoring platform to monitor the current environmental information every 3 minutes at the beginning of the task, and calculates the rate of change of environmental variables combined with the current information predicted in the past. The Kalman filter algorithm is used to update the noise covariance matrix in real time to achieve adaptive recognition of temperature, air pressure, wind speed, and humidity in a complex marine environment, thereby reducing the error between the predicted value and the true value. At the same time, according to the execution time period of the current actual task allocation, the task time period in the pre-planned task allocation plan is dynamically updated.

Claims

1. A method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment, characterized in that: The following steps are involved: (1) Obtain environmental information in the mission area and construct a marine environment dataset; (2) Construct an environmental energy consumption prediction model CNN-LSTM-SE-BO to predict the environmental energy consumption information of the time period that meets the task duration; CNN-LSTM-SE-BO includes a convolutional neural network CNN, a long short-term memory network LSTM, an attention mechanism SE module, and a Bayesian model BO; the convolutional neural network CNN is used to capture the local features and potential time series patterns in the environmental information data; the long short-term memory network LSTM is used to process the long-term dependencies of the time series data; (3) Based on the heterogeneity constraints of the unmanned systems and the task coupling constraints of the surface scenes, a cooperative multi-task allocation model CMTAP is constructed for multiple heterogeneous unmanned systems to perform tasks on multiple stationary ground targets. The environmental energy consumption information predicted in step (2) is used as the model input. Combined with the driving speed of the unmanned system in the mission plan, the task execution time period and the energy consumption prediction time period are matched. (4) Based on the cooperative multi-task allocation model CMTAP, the non-dominated sorting genetic algorithm NSGA-II with elite strategy is used to determine the optimal individual as the final task pre-allocation solution; (5) Execute the task pre-allocation plan determined in step (4), collect current environmental information, combine it with the predicted environmental information of the corresponding time period, calculate the rate of change of environmental variables, use the Kalman filter algorithm to update the noise covariance matrix in real time, dynamically update the task time period according to the current actual task execution status, and optimize the task and allocation plan; Among them, step (2) is specifically as follows: After normalizing the marine environment dataset, it is divided into time series sets; Divide the time series set into a training set and a sample set; The environmental energy consumption prediction model CNN-LSTM-SE-BO is constructed, and the genetic algorithm is introduced to optimize the model hyperparameters and adjust the weights. The objective function is to minimize the prediction error loss function. Among them, the attention mechanism SE module is used to extract the environmental parameter characteristics that affect the energy consumption of the unmanned system and recalibrate them into energy consumption feature channels. The Bayesian model BO is used to predict the energy consumption of the unmanned system, identify the key environmental factors that affect energy consumption, and feed back to the attention mechanism SE module to re-determine the environmental parameter characteristics that affect the energy consumption of the unmanned system. After the training model is completed, the predicted environmental energy consumption information for the time period that meets the task duration is output.

2. According to the method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment according to claim 1, it is characterized in that: In step (2), the objective function of the Bayesian model BO is to minimize the mean square error of energy consumption prediction within the unit time window: : ; ; ; in, For the i The actual energy consumption in each time window; Under the parameter set θ, i The predicted energy consumption of time windows; n is the total number of time windows; Represents the expected value of the mean square error under different parameters θ; is the optimal parameter set that minimizes the mean square error obtained through iterative search of the Bayesian model; is the channel after recalibration; It is the Squeeze compression function, which is used to compress the global information of the input features; is the input feature channel; Sigmoid activation function is used to limit the value of the feature vector between 0 and 1; and is the weight matrix of the fully connected layer, which is used to reduce and increase the dimension of the input feature vector; is the ReLU activation function.

3. According to claim 2, the method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment is characterized in that: Step (3) is as follows: Establish a mission target model and define a set of multiple maritime mission targets; perform reconnaissance mission R, attack mission S and inspection mission V on each target in turn; Establish a heterogeneous unmanned system model, which includes reconnaissance drones that can perform reconnaissance missions R and inspection missions V , combat drones capable of performing attack missions S and military unmanned boats that can perform all kinds of missions ; Establish task allocation constraints; Establish the coordinate point location information, determine the Euclidean distance between each point, the total distance of three different tasks of a certain target point, and the total distance of the task allocation plan based on the coordinates of the unmanned system starting point and the coordinates of multiple target points; Establish a speed model for heterogeneous unmanned agents and calculate the task completion time.

4. According to claim 3, the method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment is characterized in that: The task allocation constraints are ; ; in, is the number of heterogeneous unmanned systems, Configure the total number of configuration points and mission target points for heterogeneous unmanned systems, is a binary decision variable, if , then it means that the drone From the first configuration Fly to the second configuration Execution against target The value of m is 1, 2, and 3, which represent the reconnaissance task R, the attack task S, and the inspection task V, respectively. Indicates the number of tasks executed at each task target point, .

5. The method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment according to claim 4 is characterized in that: Step (4) is as follows: Using a multi-type gene chromosome encoding method, the task execution order, task target point, task type, heterogeneous unmanned intelligent agent executor sequence number, unmanned intelligent agent movement speed, task execution time, task time period and intelligent agent energy consumption are used to construct the initial population, which is divided into multiple sub-populations; Define the individual value of non-dominated sorting optimization standard as the screening standard of the migration population, use the elite strategy to retain the best individuals, use the roulette method to select adaptive multi-point crossover and mutation, and iterate the sub-population; Determine whether the maximum number of iterations has been reached: If the maximum number of iterations is reached, the iteration ends, all sub-population output results are merged, and the final results are non-dominated sorted; If the maximum number of iterations has not been reached, it is determined whether it is dominated by the standard individual value of the non-dominated sorting optimization. If so, the individual is migrated into the migration population; if not, the next iteration is carried out directly. At the same time, before each sub-population performs the next generation of iterative optimization, it selects individuals from the non-dominated solution set from the migration population to enter the next generation of iterative optimization of each sub-population.

6. The method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment according to claim 5 is characterized in that: In the non-dominated sorting genetic algorithm NSGA-II with elite strategy, the fitness of each subpopulation is calculated as ; ; in, is the fitness function of the task execution time, The fitness function for the total energy consumption of the task allocation scheme; , , , is the weight; is the total task execution time, is the total energy consumption of the task; is the total distance of the task allocation solution; Q represents the constraint violation evaluation coefficient.

Citation Information

Patent Citations

  • Unmanned aerial vehicle cluster cooperative task allocation method based on iterative optimization

    CN115525068A

  • Multi-unmanned aerial vehicle cooperative task allocation method based on quantum suburb wolf optimization mechanism

    CN115562336A

  • Unmanned aerial vehicle group task allocation method based on quantum annealing algorithm model

    CN117764188A

  • Abnormal value detection model training method and abnormal value detection method and system

    CN115587335A

  • Intelligent bus scheduling method and system

    CN117671992A