Path and track collaborative optimization method and device based on neural network enhancement, medium and product

Abstract

The invention discloses a path and trajectory collaborative optimization method and device based on neural network enhancement, a medium and a product, and the method comprises the steps: obtaining starting and ending nodes of a moving target and a graph structure of a to-be-solved problem, and obtaining a robust Pareto frontier of the to-be-solved problem based on a collaborative optimization model; comprising the following steps: training a variational self-encoding network model to generate a candidate path set; obtaining an ideal speed curve based on the candidate path set, and generating a trajectory set by using a trajectory generation model; comprising the steps of performing variation on a hidden space feature vector of a variational self-encoding network model to obtain a new hidden space feature vector; generating a new candidate path set according to the new hidden space feature vector; obtaining an ideal speed curve of each candidate path in the new candidate path set to generate a new trajectory set; and screening according to the trajectory set and the new trajectory set to obtain a robust effective trajectory set as a robust Pareto frontier. According to the method, the defect of current path and trajectory multi-target robust collaborative optimization is effectively solved.

Description

Technical Field

[0001] The present invention relates to a path and trajectory collaborative optimization method, device, medium and product based on neural network enhancement, belonging to the technical field of intersection of transportation and artificial intelligence. Background Art

[0002] In planned operating spaces such as airports, in order to improve the safety and efficiency of operations, the coordinated optimization of paths and trajectories has become a key technical requirement. In the prior art, the optimization process of paths and trajectories is usually divided into two relatively independent steps. First, the shortest path algorithm is used to assign the optimal path to the aircraft or ground vehicle. Subsequently, based on the determined path, detailed trajectory planning is performed to ensure that the operation process meets the established safety standards and efficiency requirements. This step-by-step optimization method simplifies the calculation process to a certain extent, so that the optimization of paths and trajectories can be achieved with limited computing resources.

[0003] However, this step-by-step optimization method of the prior art has obvious shortcomings. First, the optimization process of the path and trajectory lacks sufficient synergy, resulting in the optimization result not being globally optimal. Since path planning and trajectory planning are carried out separately, it is difficult to find the best balance between the two, which affects the overall operation effect. Secondly, there are often more than one optimization objectives that need to be considered in the actual operation process, which may include time, energy consumption, pollutant emissions and other aspects. There are often conflicts between these objectives, which significantly increases the complexity of the path and trajectory optimization problem. In addition, due to the inevitable existence of trajectory tracking errors in the actual operation process, higher requirements are also placed on the robustness of path and trajectory optimization. However, the prior art lacks an effective multi-objective robust collaborative optimization method, and it is difficult to effectively deal with the uncertainties caused by multiple conflicting optimization objectives and trajectory tracking errors, and ensure the robust optimality of the path and trajectory. Therefore, it is particularly important to develop an effective multi-objective robust collaborative optimization method for paths and trajectories.

[0004] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgement or any form of suggestion that the information constitutes the prior art already known to ordinary technicians in this field. Summary of the invention

[0005] The technical problem to be solved by the present invention is to solve the shortcomings of the current multi-objective robust collaborative optimization of paths and trajectories.

[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0007] In a first aspect, the present invention provides a path and trajectory collaborative optimization method based on neural network enhancement, characterized in that it includes:

[0008] Get the starting and ending nodes of the moving target and the graph structure of the problem to be solved;

[0009] According to the start and end nodes of the moving target and the graph structure of the problem to be solved, based on a pre-trained collaborative optimization model, an optimization is performed to obtain a robust Pareto frontier of the problem to be solved;

[0010] The training method and optimization method of the collaborative optimization model include:

[0011] Obtain the historical start and end nodes of the moving target and the graph structure of the historical problems to be solved as training samples;

[0012] According to the training samples, a supervised learning method is used for training to determine the variational autoencoder network model parameters and the latent space of the encoder to generate the latent space feature vector;

[0013] When searching for the optimal solution, obtain the starting and ending nodes of the moving target and the graph structure of the problem to be solved;

[0014] Based on the pre-trained variational autoencoder network model, the latent space feature vector is randomly selected, and a candidate path set is generated according to the starting and ending nodes of the moving target and the graph structure of the problem to be solved;

[0015] Obtain the ideal speed curve of each candidate path in the candidate path set, and generate a trajectory set using a trajectory generation model based on a deep neural network;

[0016] Repeat the following steps until the optimization calculation time exceeds the threshold:

[0017] Perform mutation and crossover calculations on the latent space feature vector to obtain a new latent space feature vector;

[0018] According to the new latent space feature vector, the starting and ending nodes of the moving target and the graph structure of the problem to be solved, a new set of candidate paths is generated based on the pre-trained variational autoencoder network model;

[0019] Obtain the ideal speed curve of each candidate path in the new candidate path set, and generate a new trajectory set using a trajectory generation model based on a deep neural network;

[0020] Screening robust and effective solutions from the trajectory set and the new trajectory set to obtain a robust and effective trajectory set;

[0021] The latent space feature vectors corresponding to the dominant individuals in the robust and effective trajectory set are retained, and the candidate path set is updated to the candidate paths corresponding to the dominant individuals;

[0022] After the optimization calculation is completed, the final set of robust and effective trajectories is used as the robust Pareto frontier for the coordinated optimization of paths and trajectories.

[0023] Furthermore, the collaborative optimization model includes:

[0024] Variational autoencoder network model, including encoder and decoder;

[0025] The encoder is used to encode each path of the moving target into a latent space feature vector conditioned by the graph structure and the start and end nodes;

[0026] The decoder is used to gradually generate a node sequence of a candidate path using a latent space feature vector generated by the encoder with the graph structure and the start and end nodes as conditions;

[0027] A trajectory generation model based on a deep neural network is used to receive the ideal speed curve of each candidate path and generate a trajectory set that conforms to the motion characteristics of the moving target.

[0028] Furthermore, based on the pre-trained variational autoencoder network model, the latent space feature vector is randomly selected, and a candidate path set is generated according to the start and end nodes of the moving target and the graph structure of the problem to be solved, including:

[0029] The selected latent space feature vector is input into the decoder, and a candidate path is generated based on the start and end nodes of the moving target and the graph structure of the problem to be solved;

[0030] Using multiple latent space feature vectors, multiple candidate paths are generated to form a candidate path set.

[0031] Furthermore, mutation and crossover calculations are performed on the latent space feature vector to obtain a new latent space feature vector, including:

[0032] Randomly select three different latent space feature vectors from the current latent space feature vector as the first individual, the second individual, and the third individual;

[0033] Calculate the difference between the first individual and the second individual;

[0034] Add the difference between the first individual and the second individual to the third individual to generate a mutant individual;

[0035] Use the mutant individual and the third individual to perform a crossover operation, randomly select crossover points to exchange gene fragments in the mutant individual and the third individual, and generate offspring individuals;

[0036] The offspring individuals are used as new latent space feature vectors.

[0037] Furthermore, the calculation expression for calculating the difference between the first individual and the second individual is:

[0038] (1);

[0039] In the formula, represents the difference between the first individual and the second individual, Represents the first individual, Represents the second individual.

[0040] Furthermore, the calculation expression for generating variant individuals is:

[0041] (2);

[0042] In the formula, represents a variant individual, Represents the third individual.

[0043] Furthermore, the definition of the robust and effective solution includes:

[0044] For any trajectory x in the trajectory set:

[0045] When there are other trajectories y that satisfy the following conditions, other trajectories y dominate any trajectory x, and any trajectory x is a non-robust valid solution:

[0046] Other trajectories y belong to the trajectory set;

[0047] The cost of other trajectories y in probability space U is not greater than the cost of any trajectory x in probability space U;

[0048] There are specific variables Belong to the probability space U, so that other trajectories y in a specific variable The cost under the condition is less than any trajectory x in a specific variable The cost value of the following;

[0049] Any trajectory x is considered a robust valid solution when it is not dominated by any other trajectory in the trajectory set and satisfies the time window constraints on the candidate paths.

[0050] In a second aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0051] In a third aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, wherein the computer program / instruction, when executed by a processor, implements the steps of the method described in the first aspect.

[0052] In a fourth aspect, the present invention provides a computer program product, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention can fully understand the operation requirements and constraints of the mobile target by obtaining the start and end nodes of the mobile target and the graph structure of the problem to be solved. On this basis, using the pre-trained collaborative optimization model, the present invention realizes the collaborative optimization of the path and trajectory, and successfully obtains the robust Pareto frontier of the collaborative optimization of the path and trajectory. The method provided by the present invention is innovative and effectively overcomes the shortcomings of the prior art that the path and trajectory optimization are relatively independent and lack of synergy, and significantly improves the global optimality of the optimization results. By comprehensively considering multiple optimization objectives, such as time, energy consumption, and pollutant emissions, the method of the present invention can ensure that the operation of the mobile target is both safe and efficient while meeting these objectives. In addition, due to the introduction of the concept of robust Pareto frontier, the method of the present invention can still maintain high robustness and stability when facing uncertain factors such as trajectory tracking errors, thereby ensuring the stable operation of the mobile target.

[0055] The collaborative optimization model training method provided by the present invention is unique and efficient. By obtaining the historical start and end nodes of the mobile target and the graph structure of the historical problems to be solved as training samples, the information in the historical data can be fully utilized to improve the training effect and generalization ability of the model. The present invention generates a candidate path set using the latent space feature vector of the pre-trained variational autoencoder network model, which not only reduces the computational complexity, but also improves the diversity and accuracy of candidate path generation. The present invention further realizes the refined planning of the trajectory by obtaining the ideal speed curve of the candidate path and generating a trajectory set using a trajectory generation model based on a deep neural network. In the iterative optimization process, the present invention continuously explores new optimization spaces by mutating and cross-calculating the latent space feature vectors until the time threshold is reached. Finally, by screening the dominant individuals that meet the robust effectiveness, the robust Pareto frontier of collaborative optimization is obtained. The method of the present invention can continuously output high-quality and robust path and trajectory optimization solutions, and also provides strong support for subsequent dynamic operation decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flowchart of a method for obtaining a robust and effective trajectory set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0058] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0059] Example 1

[0060] This embodiment introduces a path and trajectory collaborative optimization method based on neural network enhancement, including the following steps:

[0061] Step 1: Train the collaborative optimization model, including:

[0062] Step 1.1: Obtain the historical start and end nodes of the moving target and the graph structure of the historical problem to be solved as training samples.

[0063] Wherein, the collaborative optimization model includes:

[0064] Variational autoencoder network model, including encoder and decoder;

[0065] The encoder is used to encode each path of the moving target into a latent space feature vector conditioned by the graph structure and the start and end nodes;

[0066] The decoder is used to gradually generate a node sequence of a candidate path using a latent space feature vector generated by the encoder with the graph structure and the start and end nodes as conditions;

[0067] A trajectory generation model based on a deep neural network is used to receive the ideal speed curve of each candidate path and generate a trajectory set that conforms to the motion characteristics of the moving target.

[0068] Step 1.2: According to the training sample, training a variational autoencoder network model includes:

[0069] According to the historical starting position node and the historical ending position node contained in the training sample i, the encoder is used to convert the historical starting and ending nodes The corresponding optimal path is encoded as a graph structure corresponding to the training sample And historical start and end nodes The latent space eigenvectors are conditioned on ,in, , N is the number of samples, Represents the starting node in the historical start and end nodes, Indicates the end node in the history start and end nodes.

[0070] The decoder is used to calculate the latent space feature vector 、Historical start and end nodes And the graph structure , generating a sequence of nodes containing Candidate paths for ,in, Indicates the starting node included in the candidate path, which is the starting position node of the problem to be solved. Indicates the second node included in the candidate path. Indicates the terminal node included in the candidate path, which is the destination node of the problem to be solved.

[0071] Taking the optimal path calculated offline as the true value, the variational autoencoder network model is trained using a supervised learning method to determine the latent space Z to which the variational autoencoder network model parameters and the latent space feature vector generated by the encoder belong.

[0072] Step 2: Obtain the start and end nodes of the moving target and the graph structure of the problem to be solved, and perform optimization based on the pre-trained collaborative optimization model according to the start and end nodes of the moving target and the graph structure of the problem to be solved to obtain the robust Pareto frontier of the problem to be solved, including:

[0073] Step 2.1: Based on the pre-trained variational autoencoder network model, randomly select latent space feature vectors from the latent space Z, and generate a set of candidate paths according to the start and end nodes of the moving target and the graph structure of the problem to be solved, including:

[0074] The latent space feature vector randomly selected Input decoder to move the start and end nodes of the target And the graph structure of the problem to be solved Generate candidate paths as conditions ,in , n is the number of randomly selected latent space features;

[0075] Using n latent space feature vectors , generate n candidate paths and form a candidate path set .

[0076] Step 2.2: Get the candidate path set Each candidate path The ideal speed curve is: , is the total number of paths in the candidate path set. The trajectory generation model based on deep neural network is used to generate the trajectory set ,in, represents the first trajectory in the trajectory set, represents the last trajectory in the trajectory set, and K represents the total number of ideal speed curves.

[0077] The probability distribution of each trajectory in the trajectory set at time t is expressed as .

[0078] Step 2.3: Repeat the following steps until the latent space feature vector The time for mutation and crossover calculation exceeds the threshold, such as Figure 1 As shown:

[0079] Step 2.3.1: Latent space feature vector Perform mutation and crossover calculations to obtain n new feature vectors ,include:

[0080] From the current latent space feature vector Three different latent space feature vectors are randomly selected as the first entity , the second individual and the third individual ;

[0081] Calculate the first volume and the second individual The difference between ;

[0082] The first body and the second individual The difference between Adding a third person , generate variant individuals ;

[0083] Using mutant individuals With the third person Perform a crossover operation and randomly select crossover points to exchange mutation individuals and the third individual Gene fragments in the gene to generate offspring individuals ;

[0084] The calculation expression for calculating the difference between the first individual and the second individual is:

[0085] (1);

[0086] In the formula, represents the difference between the first individual and the second individual, Represents the first individual, represents the second individual;

[0087] The calculation expression for generating variant individuals is:

[0088] (2);

[0089] In the formula, represents a variant individual, Represents the third individual.

[0090] The offspring individuals As the new latent space feature vector .

[0091] Step 2.3.2: Based on the new latent space feature vector As well as the graph structure of the problem to be solved, a new set of candidate paths is generated based on the pre-trained variational autoencoder network model .

[0092] Step 2.3.3: Get a new set of candidate paths The ideal speed curve of each candidate path in is used to generate a new trajectory set using a trajectory generation model based on a deep neural network.

[0093] Step 2.3.4: According to the definition of robust and effective solution, filter the trajectory set and the new trajectory set to obtain a robust and effective trajectory set .

[0094] The definition of the robust solution includes:

[0095] For any trajectory x in the trajectory set:

[0096] When there are other trajectories y that satisfy the following conditions, other trajectories y dominate any trajectory x, and any trajectory x is a non-robust valid solution:

[0097] Other trajectories y belong to the trajectory set;

[0098] The cost of other trajectories y in probability space U is not greater than the cost of any trajectory x in probability space U;

[0099] There are specific variables Belong to the probability space U, so that other trajectories y in a specific variable The cost under the condition is less than any trajectory x in a specific variable The cost value of the following;

[0100] Any trajectory x is considered a robust valid solution when it is not dominated by any other trajectory in the trajectory set and satisfies the time window constraints on the candidate paths.

[0101] Step 2.3.5: Keep the latent space feature vector corresponding to the dominant individual in the robust and effective trajectory set, and update the candidate path set to the candidate path corresponding to the dominant individual.

[0102] Step 2.4: After the optimization calculation is completed, the final set of robust and effective trajectories is used as the robust Pareto frontier for the coordinated optimization of paths and trajectories.

[0103] Example 2

[0104] Based on the same inventive concept as Example 1, this embodiment introduces a computer device, including a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the method described in Example 1.

[0105] Example 3

[0106] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium on which a computer program / instruction is stored, characterized in that the computer program / instruction implements the steps of the method described in Embodiment 1 when executed by a processor.

[0107] Example 4

[0108] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including a computer program / instruction, characterized in that the computer program / instruction implements the steps of the method described in Embodiment 1 when executed by a processor.

[0109] In summary, the present invention can fully understand the operation requirements and constraints of the mobile target by obtaining the start and end nodes of the mobile target and the problem to be solved. On this basis, using the pre-trained collaborative optimization model, the present invention realizes the collaborative optimization of the path and trajectory, and successfully obtains the robust Pareto frontier of the collaborative optimization of the path and trajectory. The method provided by the present invention is innovative and effectively overcomes the shortcomings of the prior art that the path and trajectory optimization are relatively independent and lack of synergy, and significantly improves the global optimality of the optimization results. By comprehensively considering multiple optimization objectives, such as time, energy consumption, and pollutant emissions, the method of the present invention can ensure that the operation of the mobile target is both safe and efficient while meeting these objectives. In addition, due to the introduction of the concept of robust Pareto frontier, the method of the present invention can still maintain high robustness and stability when facing uncertain factors such as trajectory tracking errors, thereby ensuring the stable operation of the mobile target.

[0110] The collaborative optimization model training method provided by the present invention is unique and efficient. By obtaining the historical start and end nodes of the mobile target and the graph structure of the historical problems to be solved as training samples, the information in the historical data can be fully utilized to improve the training effect and generalization ability of the model. The present invention generates a candidate path set using the latent space feature vector of the pre-trained variational autoencoder network model, which not only reduces the computational complexity, but also improves the diversity and accuracy of candidate path generation. The present invention further realizes the refined planning of the trajectory by obtaining the ideal speed curve of the candidate path and generating a trajectory set using a trajectory generation model based on a deep neural network. In the iterative optimization process, the present invention continuously explores new optimization spaces by mutating and cross-calculating the latent space feature vectors until the time threshold is reached. Finally, by screening the dominant individuals that meet the robust effectiveness, the robust Pareto frontier of collaborative optimization is obtained. The method of the present invention can continuously output high-quality and robust path and trajectory optimization solutions, and also provides strong support for subsequent dynamic operation decisions.

[0111] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0115] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A path and trajectory collaborative optimization method based on neural network enhancement, characterized in that: include: Get the starting and ending nodes of the moving target and the graph structure of the problem to be solved; According to the start and end nodes of the moving target and the graph structure of the problem to be solved, based on a pre-trained collaborative optimization model, an optimization is performed to obtain a robust Pareto frontier of the problem to be solved; The training method and optimization method of the collaborative optimization model include: Obtain the historical start and end nodes of the moving target and the graph structure of the historical problems to be solved as training samples; According to the training samples, a supervised learning method is used for training to determine the variational autoencoder network model parameters and the latent space of the encoder to generate the latent space feature vector; When searching for the optimal solution, obtain the starting and ending nodes of the moving target and the graph structure of the problem to be solved; Based on the pre-trained variational autoencoder network model, the latent space feature vector is randomly selected, and a candidate path set is generated according to the starting and ending nodes of the moving target and the graph structure of the problem to be solved; Obtain the ideal speed curve of each candidate path in the candidate path set, and generate a trajectory set using a trajectory generation model based on a deep neural network; Repeat the following steps until the optimization calculation time exceeds the threshold: Perform mutation and crossover calculations on the latent space feature vector to obtain a new latent space feature vector; According to the new latent space feature vector, the starting and ending nodes of the moving target and the graph structure of the problem to be solved, a new set of candidate paths is generated based on the pre-trained variational autoencoder network model; Obtain the ideal speed curve of each candidate path in the new candidate path set, and generate a new trajectory set using a trajectory generation model based on a deep neural network; Screening robust and effective solutions from the trajectory set and the new trajectory set to obtain a robust and effective trajectory set; The latent space feature vectors corresponding to the dominant individuals in the robust and effective trajectory set are retained, and the candidate path set is updated to the candidate paths corresponding to the dominant individuals; After the optimization calculation is completed, the final set of robust and effective trajectories is used as the robust Pareto frontier for the coordinated optimization of paths and trajectories.

2. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 1 is characterized in that: The collaborative optimization model includes: Variational autoencoder network model, including encoder and decoder; The encoder is used to encode each path of the moving target into a latent space feature vector conditioned by the graph structure and the start and end nodes; The decoder is used to gradually generate a node sequence of a candidate path using a latent space feature vector generated by the encoder with the graph structure and the start and end nodes as conditions; A trajectory generation model based on a deep neural network is used to receive the ideal speed curve of each candidate path and generate a trajectory set that conforms to the motion characteristics of the moving target.

3. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 1 is characterized in that: Based on the pre-trained variational autoencoder network model, the latent space feature vector is randomly selected, and a candidate path set is generated according to the start and end nodes of the moving target and the graph structure of the problem to be solved, including: The selected latent space feature vector is input into the decoder, and a candidate path is generated based on the start and end nodes of the moving target and the graph structure of the problem to be solved; Using multiple latent space feature vectors, multiple candidate paths are generated to form a candidate path set.

4. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 1 is characterized in that: Perform mutation and crossover calculations on the latent space feature vector to obtain a new latent space feature vector, including: Randomly select three different latent space feature vectors from the current latent space feature vector as the first individual, the second individual, and the third individual; Calculate the difference between the first individual and the second individual; Add the difference between the first individual and the second individual to the third individual to generate a mutant individual; Use the mutant individual and the third individual to perform a crossover operation, randomly select crossover points to exchange gene fragments in the mutant individual and the third individual, and generate offspring individuals; The offspring individuals are used as new latent space feature vectors.

5. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 4 is characterized in that: The calculation expression for calculating the difference between the first individual and the second individual is: （1）； In the formula, represents the difference between the first individual and the second individual, Represents the first individual, Represents the second individual.

6. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 4 is characterized in that: The calculation expression for generating variant individuals is: （2）； In the formula, represents a variant individual, Represents the third individual.

7. The path and trajectory collaborative optimization method based on neural network enhancement according to claim 1 is characterized in that: The definition of the robust and effective solution includes: For any trajectory x in the trajectory set: When there are other trajectories y that satisfy the following conditions, other trajectories y dominate any trajectory x, and any trajectory x is a non-robust valid solution: Other trajectories y belong to the trajectory set; The cost of other trajectories y in probability space U is not greater than the cost of any trajectory x in probability space U; There are specific variables Belong to the probability space U, so that other trajectories y in a specific variable The cost under the condition is less than any trajectory x in a specific variable The cost value of Any trajectory x is considered a robust valid solution when it is not dominated by any other trajectory in the trajectory set and satisfies the time window constraints on the candidate paths.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images