CN118534860B

CN118534860B - Task allocation and scheduling method for heterogeneous unmanned system in marine dynamic environment

Info

Publication number: CN118534860B
Application number: CN202410993957.4A
Authority: CN
Inventors: 刘云平; 王富尧; 高佳宁; 陆旭春; 喻鹏鹏; 程勇; 龚毅光; 张永宏; 徐梁
Original assignee: Nanjing University of Information Science and Technology
Current assignee: Nanjing University of Information Science and Technology
Priority date: 2024-07-24
Filing date: 2024-07-24
Publication date: 2024-11-08
Anticipated expiration: 2044-07-24
Also published as: CN118534860A

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.

Description

Task allocation and scheduling method for heterogeneous unmanned system in marine dynamic environment

Technical Field

The invention relates to the technical field of unmanned aerial vehicles, in particular to a task allocation and scheduling method for a heterogeneous unmanned system in a marine dynamic environment.

Background

Currently, the method is applied to task allocation and scheduling algorithms such as simulated annealing algorithm ant colony algorithm, particle swarm optimization algorithm and improved related genetic algorithm, can serve cooperative combat task allocation of heterogeneous unmanned systems to a certain extent, but has the advantages of being particularly limited by static task allocation strategies, single target optimization and neglecting influence of offshore variable environments on task allocation results when facing dynamic offshore task environments. Although the traditional genetic algorithm is suitable for the multi-objective optimization problem, the traditional genetic algorithm has lower calculation efficiency in a large-scale and complex task allocation scene, and in a highly dynamic scene, if the emergency is not considered, a reasonable and effective task allocation scheme can not be made. The basic genetic algorithm, in the face of multi-factor constraints and uncertainties in similar offshore task environments, may not accurately simulate the impact of the real environment on task execution, thereby affecting the feasibility and optimality of the final task allocation scheme.

Chinese patent CN117764188A discloses an unmanned aerial vehicle group task allocation method based on a quantum annealing algorithm model, which converts an unmanned aerial vehicle group task allocation problem into a single travel business problem, solves a system quantum state corresponding to an optimal solution by combining a quantum annealing algorithm and the unmanned aerial vehicle group task allocation problem, but only makes a pre-planned task allocation scheme in a static scene, does not consider the influence of dynamic environment factors on a task allocation result in a time scale, and therefore, does not have the capability of real-time adjustment and re-planning. Chinese patent CN115525068a discloses a coordinated task allocation method of an unmanned aerial vehicle cluster based on iterative optimization, which obtains task profit information of an effective future task of the unmanned aerial vehicle cluster according to task information in a current task and a future task obtained by future task planning of the unmanned aerial vehicle cluster, obtains task profit information of the effective future task of the unmanned aerial vehicle cluster, and obtains a task allocation result of the unmanned aerial vehicle cluster by using an iterative auction mechanism. Chinese patent CN115562336A discloses a strategy for solving the problem of task allocation of multi-unmanned aerial vehicle collaborative execution based on quantum suburban wolf optimization mechanism, and the method can allocate task targets for each formation machine under the constraints of three-dimensional scene, time synchronization and the like, but consumes a large amount of computing resources and time when processing large-scale unmanned aerial vehicle formation and complex task environments, and has poor practical application effect.

Disclosure of Invention

The invention aims to: the invention aims to provide a heterogeneous unmanned system task allocation and scheduling method based on an environmental energy consumption prediction model and a plurality of non-dominant sequencing genetic algorithms optimized in a group-parallel manner in an offshore dynamic environment.

The technical scheme is as follows: the invention relates to a task allocation and scheduling method for a heterogeneous unmanned system in a marine dynamic environment, which comprises the following steps:

(1) Acquiring environment information in a task area, and constructing a marine environment data set;

(2) Constructing an environment energy consumption prediction model CNN-LSTM-SE-BO, which is used for predicting environment energy consumption information of a time period meeting task duration; the CNN-LSTM-SE-BO comprises a convolutional neural network CNN, a long-short-term memory network LSTM, an attention mechanism SE module and a Bayesian model BO;

(3) Constructing a cooperative multi-task allocation model CMTAP of the multi-heterogeneous unmanned system for executing tasks on the multi-static ground targets according to the heterogeneous constraint of the unmanned system and the task coupling constraint of the water surface scene, inputting environment energy consumption information predicted in the step (2) as a model, combining the running speed of the unmanned system in a task scheme, and matching a task execution time period and an energy consumption prediction time period;

(4) Based on the cooperative multitasking distribution model CMTAP, determining an optimal individual by adopting a non-dominant ordering genetic algorithm NSGA-II with elite strategy as a final task pre-distribution scheme;

(5) Executing the task pre-allocation scheme determined in the step (4), collecting current environment information, calculating the environment variable change rate by combining with the predicted environment information of the corresponding time period, updating the noise covariance matrix in real time by adopting a Kalman filtering algorithm, dynamically updating the task time period according to the current actual task execution condition, and optimizing the task and allocation scheme.

Further, the step (2) is specifically as follows:

after the marine environment data set is normalized, dividing the marine environment data set into a time sequence set;

Dividing the time sequence set into a training set and a sample set;

Constructing an environment energy consumption prediction model CNN-LSTM-SE-BO, introducing a genetic algorithm, performing model super-parameter optimization, adjusting weights, and enabling an objective function to be a prediction error loss minimization function; the attention mechanism SE module is used for extracting environmental parameter characteristics affecting the energy consumption of the unmanned system and recalibrating the environmental parameter characteristics into an energy consumption characteristic channel; the Bayesian model BO is used for predicting the energy consumption of the unmanned system, identifying key environmental factors influencing the energy consumption, feeding back to the attention mechanism SE module, and redefining environmental parameter characteristics influencing the energy consumption of the unmanned system;

and after the training model is completed, outputting the predicted environment energy consumption information of the time period meeting the task duration.

Further, in step (2), the objective function of the Bayesian model BO is to minimize the mean square error of the energy consumption prediction in the unit time window：

；

Wherein, The actual energy consumption of the ith time window; the predicted energy consumption of the ith time window is calculated under the parameter set theta; n is the total number of time windows; representing the expected value of the mean square error at different parameters θ; the method comprises the steps of obtaining an optimal parameter set which is obtained through Bayesian model iterative search and enables the mean square error to be minimum; A channel after recalibration; The method is a Squeeze compression function and is used for carrying out global information compression on input features; is an input feature channel; activating a function for Sigmoid for limiting the value of the feature vector to between 0 and 1; And (3) with The weight matrix is a weight matrix of the full connection layer and is used for performing dimension reduction and dimension increase operation on the input feature vector; activating the function for the ReLU.

Further, the step (3) is specifically as follows:

Establishing a task target model and defining a multi-offshore task target set; a detection task R, an attack task S and a checking task V are sequentially executed for each target;

establishing a heterogeneous unmanned system model, wherein the heterogeneous unmanned system comprises a detection unmanned plane capable of executing a detection task R and a checking task V Combat unmanned aerial vehicle capable of executing attack task SMilitary unmanned boat capable of executing all kinds of tasks；

Establishing task allocation condition constraints;

establishing coordinate point position information, and determining Euclidean distance between each point, total distance of three different tasks of a certain target point and total distance of a task allocation scheme according to starting point coordinates of an unmanned system and coordinates of a plurality of target points;

And establishing a speed model of the heterogeneous unmanned intelligent agent, and calculating the task completion time.

Further, the task allocation condition is constrained to be

；

Wherein, For the number of heterogeneous unmanned systems,The total number of configuration points and task target points for the heterogeneous unmanned system,As binary decision variables, ifThen it indicates unmanned aerial vehicleFrom the first configurationFlying to a second configurationExecution of targetTask m of (2); m is 1,2 and 3 respectively represent a investigation task R, an attack task S and a checking task V; Representing the number of tasks performed at each task target point, 。

Further, the step (4) is specifically as follows:

adopting a multi-type gene chromosome coding method, constructing an initial population by a task execution sequence, a task target point, a task type, a heterogeneous unmanned intelligent agent executor serial number and an unmanned intelligent agent movement speed, and dividing the initial population into a plurality of sub-populations by task execution time, a task time period and intelligent agent energy consumption;

Defining a non-dominant ordering optimization standard individual value as a screening standard of a migration population, adopting elite strategy to keep optimal individuals, adopting a roulette method to select self-adaptive multi-point intersection and variation, and iterating sub-populations;

Judging whether the maximum iteration times are reached:

if the maximum iteration times are reached, ending the iteration, combining all the sub population output results, and performing non-dominant sorting on the final result;

if the maximum iteration number is not reached, judging whether the maximum iteration number is dominant to the non-dominant sorting optimization standard individual value, and if so, migrating the individual into a migration population; if not, directly carrying out the next iteration, and simultaneously, before carrying out the next generation iterative optimization, each sub-population selects individuals of the non-dominant solution set from the migration population so as to enter the next generation iterative optimization of each sub-population.

Further, in non-dominant ranking genetic algorithm NSGA-II with elite strategy, each sub-population computes fitness as

；

Wherein, As a fitness function of the task execution time,Distributing a fitness function of the total energy consumption of the scheme for the task;、、、 is the weight; for the total time that the task is to be executed, Total energy consumption for a task; Assigning a total distance of the scheme to the task; q represents a constraint violation evaluation coefficient.

The beneficial effects are that: compared with the prior art, the invention has the remarkable advantages that: 1. according to the invention, an environmental energy consumption prediction model CNN-LSTM-SE-BO is constructed through a convolutional neural network, a long-term and short-term memory network, an attention mechanism SE module and a Bayesian model BO, environmental characteristic extraction is carried out on an acquired marine environmental data set, environmental energy consumption information in a period of time in the future is predicted, wherein the attention mechanism SE module is used for extracting the characteristic of environmental parameters affecting the energy consumption of an unmanned system in an emphasized manner, the Bayesian model BO predicts the energy consumption of the unmanned system, identifies key environmental factors affecting the energy consumption, feeds back the key environmental factors to the attention mechanism SE module, redetermines the environmental parameter characteristics affecting the energy consumption of the unmanned system, and improves the accuracy of the task allocation result of the energy consumption prediction; 2. according to the invention, an initial population meeting the isomerism and task coupling constraint of the unmanned systems is constructed by adopting a multi-type gene chromosome coding method, and the isomerism among the unmanned systems, the attribute of each task target, the instability of a dynamic environment and the dynamic adjustment of the deployment information of the unmanned systems are comprehensively considered to optimize the task execution time and the task execution energy consumption; 3. according to the invention, a multi-sub-population parallel calculation and sub-population migration optimization strategy is introduced into the NSGA-II algorithm, so that the diversity and the holding capacity of the algorithm are enhanced, good individuals are attracted, the diversity of the population is enriched, the problem of insufficient robustness of the NSGA-II algorithm when the NSGA-II algorithm faces the complex, multimodal or highly nonlinear multi-objective optimization problem is solved, and the exploration capacity and the adaptability of the algorithm in different solution space areas are improved; 4. according to the invention, the environment variable change rate is calculated, and the noise covariance matrix is updated in real time by using a Kalman filtering algorithm, so that the self-adaptive identification of air temperature, air pressure, air speed and humidity in a complex offshore environment is realized, and the error between a predicted value and a true value is reduced.

Drawings

FIG. 1 is a flow chart of training of an environmental energy consumption prediction model CNN-LSTM-SE-BO;

FIG. 2 is a flow chart of the NSGA-II algorithm of the present invention.

Detailed Description

The invention is further described below with reference to fig. 1 to 2.

The invention relates to a task allocation and scheduling method for a heterogeneous unmanned system in a marine dynamic environment, which comprises the following steps:

(1) Environmental information in the task area is acquired, and a marine environmental data set is constructed.

And (3) monitoring the temperature, the air pressure, the wind speed and the humidity data of the task area or the nearby area at intervals of 3min by using the mobile monitoring platform, wherein the monitoring time period is 10 to 20 days nearest to the current area. Since the task execution area is a small area within a short time, the environmental information of different target points at the same time can be regarded as the same. The environmental data sample monitored by the mobile monitoring platform is large enough, and the environmental data sequence has periodicity and volatility, so that the model of the invention can be satisfied for environmental prediction. Wherein the selection of the monitoring month depends on the month in which the time period of the task was performed. And meanwhile, all environmental data are normalized, so that all characteristic values are ensured to be on the same time scale, and the training of a subsequent model is facilitated.

(2) Constructing an environment energy consumption prediction model CNN-LSTM-SE-BO, which is used for predicting environment energy consumption information of a time period meeting task duration; the CNN-LSTM-SE-BO comprises 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 for extracting environmental parameter characteristics affecting the energy consumption of the unmanned system and recalibrating the environmental parameter characteristics into an energy consumption characteristic channel; the Bayesian model BO is used for predicting the energy consumption of the unmanned system, identifying key environmental factors influencing the energy consumption, feeding back to the attention mechanism SE module, and redefining environmental parameter characteristics influencing the energy consumption of the unmanned system.

(21) Carrying out normalization pretreatment on the environment data set, dividing continuous environment data into an input sequence X and a target output y in a time sequence mode, and setting the size of a time window as followsA time series set is obtained.

(22) The time series set was set as per 8:2 are divided into training and testing sets and do not disrupt data order by default to preserve the continuity of the time series.

(23) And constructing an environment energy consumption prediction model CNN-LSTM-SE-BO. Firstly, a one-dimensional convolution layer is applied for capturing local features and potential time sequence modes in environment information data, secondly, an LSTM layer is constructed for processing long-term dependency relationship of the time sequence data, and the size of a hidden layer is set, and the structure is a single-layer LSTM structure. And a SE module is built again and is used for dynamically adjusting the importance of the features, a Bayesian model BO is introduced and is used for predicting the energy consumption of the unmanned system, identifying key environmental factors influencing the energy consumption, feeding back to the attention mechanism SE module and redefining the environmental parameter features influencing the energy consumption of the unmanned system. And finally outputting predicted environment parameter values through a full connection layer.

Wherein, a Bayesian model BO is constructed, and the method concretely comprises the following steps: in the scenes of different air temperatures, air pressures, wind speeds, humidity and different self running speeds, the unmanned intelligent agent which is the same as the unmanned aerial vehicle and the unmanned aerial vehicle for task execution and has the same model, power supply and load conditions is subjected to energy consumption test, and the collected data comprise the air temperatures, the air pressures, the wind speeds, the humidity, the running speeds of the unmanned aerial vehicle and the energy consumption of the unmanned aerial vehicle in a unit time window of 3 minutes. Preprocessing the data, removing abnormal values, completing data normalization, and dividing a training set and a testing set. Defining a Bayesian regression model, training by using training set data, and performing energy consumption prediction by using a test set after training, and evaluating the performance of the model by using a mean square error. And finally, acquiring the feature weight in the Bayesian regression model, analyzing the influence of each feature on the energy consumption prediction, feeding back the feature with large influence to the SE module, and adjusting the SE module in the LSTM to recalibrate the corresponding feature channel, thereby improving the accuracy of the energy consumption prediction on the task allocation result. Wherein the objective function defining the energy consumption model is minimizing the mean square error of the energy consumption prediction within a unit time window：

；

(24) Introducing a genetic algorithm to optimize super parameters, setting parameters such as population size, mutation rate, crossover rate and the like of the genetic algorithm, encoding learning rate, LSTM hidden state dimension, LSTM layer number and reduction ratio of SE layer into individual genes, defining a fitness function, and optimizing a minimized prediction mean square error loss function MSE as an objective function. And executing genetic algorithm iteration, and selecting an individual with highest fitness as an optimal super-parameter configuration.

(25) Defining training loops, the numbers include defining optimizers and loss functions. The optimizer chooses Adam optimizer to adjust model parameters, sets learning rate according to the optimized result in step 2-4, LSTM hidden state dimension, LSTM layer number and reduction ratio of SE layer, and also chooses MSE as loss function for measuring difference between predicted value and actual value. In each cycle training, model generalization capability is increased by disturbing training data sequence, and models are trained in small batches, and updating weights are reversely propagated through forward propagation calculation loss. Every pass byThe individual epochs record one training and test loss and evaluate model performance.

(3) According to the heterogeneous constraint of the unmanned system and the task coupling constraint of the water surface scene, a cooperative multi-task distribution model CMTAP of the multi-heterogeneous unmanned system for executing tasks on the multi-static ground target is constructed, the environment energy consumption information predicted in the step (2) is taken as a model input, the running speed of the unmanned system in the task scheme is combined, and the task execution time period and the energy consumption prediction time are matched.

And (2) combining the environment energy consumption information predicted in the step (2) with the energy consumption conditions predicted by the running speeds of the unmanned aerial vehicle and the unmanned ship in the task scheme, and correspondingly setting the time period of the energy consumption conditions with each task execution time period. The method comprises the following specific steps:

(31) Establishing a task target model, defining a plurality of offshore task targets, wherein a target set T is defined as

；

Wherein, Is the total number of task targets.

Wherein the task type of the same task object must be executed after the execution of S is completed, and the task type V must be executed after the execution of S is completed.

The number of tasks performed on each target isAnd each.

。

(32) And establishing a heterogeneous unmanned system model. The heterogeneous unmanned systems can be divided into three types, including a detection unmanned aerial vehicle that can perform a detection task R and a verification task VCombat unmanned aerial vehicle capable of executing attack task SMilitary unmanned aerial vehicle capable of executing all kinds of tasks R, S, V. Thus, the overall unmanned system set may be represented as

；

Wherein, Representing the total number of heterogeneous unmanned system configurations. According to different types of unmanned systems, further constructing a heterogeneous unmanned system:

；

wherein, ，，The configuration quantity of the investigation unmanned aerial vehicle, the combat unmanned aerial vehicle and the military unmanned aerial vehicle is respectively represented. AggregationA heterogeneous collection of unmanned systems capable of performing both a scout task and a ping task is defined.An attack task S task heterogeneous unmanned system set is defined. Meanwhile, the number of the heterogeneous unmanned systems is as follows:

；

At the same time The total number of unmanned system configuration points and task target point configurations is represented. Establishing task allocation condition constraints:

；

any single task representing any target point should be assigned only once. Wherein the method comprises the steps of Is a binary decision variable ifThen it indicates unmanned aerial vehicleFrom the first configurationFlying to a second configurationExecution of targetTask m of (2); m takes the value of 1,2 and 3 to respectively represent three tasks of a investigation task R, an attack task S and a checking task V.

Representing the number of tasks performed on each task target pointShould be exactly equal to 3.

The execution of the investigation task R, the attack task S and the checking task V is required to follow strict task sequence constraint.

(33) And establishing coordinate point position information. And calculating Euclidean distances between the points according to the starting point coordinates of the unmanned system and the coordinates of the multiple target points. And selecting the sum of the calculated distances among the three tasks under the reference of the target point as the distance of the target point. The Euclidean distance between two points can be expressed as

；

Wherein, And (3) withRepresenting two different location points, wherein the location points comprise a starting point and a target point of the heterogeneous unmanned system.

The total distance for three different tasks for a target point can be expressed as

；

Wherein, The distances from the heterogeneous unmanned agent to the current task target point position when the heterogeneous unmanned agent starts from the last position to execute the R, S, V task under the ith task target point are respectively expressed, and the sum of the total distances of the tasks for all the target points is the total distance of the task allocation scheme. The total distance cost of the task allocation scheme can be expressed as

；

Where NT is the total number of target points.

(34) And calculating the task completion time. Establishing speed models of three heterogeneous unmanned intelligent agents:

；

wherein, And respectively representing the running speeds of the investigation unmanned aerial vehicle, the combat unmanned aerial vehicle and the military unmanned aerial vehicle.In order to detect the speed range of the unmanned aerial vehicle,In order to combat the speed range of the drone,Is a speed range of military unmanned boats.

Establishing a task completion time model:

；

wherein T is the total time for completing the task, i is the task of executing the ith target point, The total distance travelled by the drone is detected for the ith frame,For the total distance traveled by the ith combat unmanned aerial vehicle,Is the total distance travelled by the ith military unmanned ship.The total time of the task allocation scheme is represented as the total time of all tasks minus the time of executing the tasks in parallel in the same time. The time when the unmanned agent executor arrives at the target point is regarded as the completion of the task, so that the total completion time only considers the running time of the agent.

(4) Based on the cooperative multitasking assignment CMTAP model, an optimal individual is determined by adopting a non-dominant ranking genetic algorithm NSGA-II with elite strategy and is used as a final task pre-assignment scheme.

(41) Constructing a chromosome according to the model and the constraint constructed in the step (3) and initializing parameters. Each chromosome individual in the population has three reference parameters, which are respectively a task execution sequence, a task target point and a serial number of the heterogeneous unmanned agent executor. Constructing the rest parameter objects according to three reference parameters: constructing three task types based on each target point according to the task target points; constructing a motion speed based on each heterogeneous unmanned aerial vehicle executor according to the serial numbers of the heterogeneous unmanned aerial vehicle executors, calculating execution time of each task according to the distance between task points of task allocation, and allocating a time period of environmental information to correspond to the executed task time period; and finally, calculating the energy consumption of the intelligent body according to the environmental information of the movement speed of the intelligent body in the time period of the intelligent body. Thus, a complete chromosome individual is constructed, and parameters are initialized.

(42) The population is constructed and initialized, and the initial population is divided into a plurality of sub-populations. Calculating fitness values for each sub-population:

；

wherein, As a fitness function of the task execution time,Distributing a fitness function of the total energy consumption of the scheme for the task;、、、 is the weight; for the total time that the task is to be executed, Total energy consumption for a task; assigning a total distance of the scheme to the task; q represents a constraint violation evaluation coefficient, and the specific values are as follows:

；

wherein C represents that the constraint condition is satisfied, As a conditional function, if constraint C is satisfied, i.e., no constraint is violated, then，Otherwise。

Defining a non-dominant ranking optimization standard individual value as a screening standard of a migration population, reserving an optimal individual by adopting an elite strategy, selecting three basic parameters by adopting a roulette method, judging whether the maximum iteration number is reached or not (namely, the probability of the individual cross mutation with high fitness value is low), judging whether the non-dominant ranking optimization standard individual value is dominant or not if the maximum iteration number is not reached, if the non-dominant ranking optimization standard individual value is not reached, migrating the individual into the migration population, and if the non-dominant ranking optimization standard individual value is not dominant, directly carrying out the next iteration; and before the next generation iterative optimization is carried out on each sub-population, selecting individuals with non-dominant solution sets from the migration population so as to enter the next generation iterative optimization of each sub-population. And finishing iteration when the maximum iteration times are reached, combining all the sub population output results, and performing non-dominant sorting on the final results.

(43) And selecting a proper optimal individual as a final task allocation pre-allocation scheme and outputting according to the chromosome corresponding to the final non-dominant solution set and the actual situation requirement.

The ground station control system comprises a host end of an industrial personal computer or a personal host, the heterogeneous unmanned aerial vehicle system comprises an unmanned aerial vehicle, an unmanned ship, a control main board, a data transmission module, a data storage module, a GPS positioning module, an IMU attitude module and an environment monitoring module, and the heterogeneous unmanned aerial vehicle system further comprises a barometer module. The host side contains control software which supports a plurality of communication protocols. The system is communicated with the heterogeneous unmanned system by using software, is used for acquiring information such as the gesture, the position and the like of the unmanned intelligent agent, can simultaneously send task allocation requirements to each unmanned intelligent agent task executor in a command mode, and the executor receives signals through the data transmission module and controls the unmanned intelligent agent to execute tasks through the control main board. In the process of executing the task, an unmanned system executor monitors current environment information once every 3min at intervals when the task starts by utilizing a self-carried monitoring platform, calculates the change rate of environment variables by combining the current information predicted in the past, and updates a noise covariance matrix in real time by utilizing a Kalman filtering algorithm so as to realize self-adaptive identification of air temperature, air pressure, air speed and humidity in a complex offshore environment, thereby reducing the error between a predicted value and a true value. And dynamically updating the task time period in the pre-planned task allocation scheme according to the execution time period allocated by the current actual task.

Claims

1. A method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment, characterized in that it comprises the following steps:

(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 as described in 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 a unit time window. :

;

in, is the actual energy consumption of the i - th time window; is the predicted energy consumption of the i -th time window under the parameter set θ; 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 specifically 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 the method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment according to claim 3, it is characterized in that the task allocation condition constraint is

;

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 on 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. According to the method for task allocation and scheduling of heterogeneous unmanned systems in a dynamic marine environment as described in claim 4, it is characterized in that step (4) is specifically 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. According to the task allocation and scheduling method of heterogeneous unmanned systems in a dynamic marine environment according to claim 5, it 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.