Complexity learning-driven unmanned electric card collection and distribution path optimization method

Abstract

The invention provides a complexity learning-driven unmanned electric card collection and distribution path optimization method, which comprises the steps of comprehensively considering static road attributes, dynamic traffic flow, weather factors and traffic flow and road network interaction characteristics, and constructing a natural driving data set according to n road scene samples; calculating the objective weight of each feature of the road scene sample by adopting an independence weight method MCC, and representing the difference of the contribution of different features to the complexity; carrying out quantitative scoring on the complexity of the road scene sample by using a TOPSIS (approximation ideal solution); building a neural network model to adaptively learn the internal law of a road scene sample, and estimating the complexity of a dynamic uncertain scene; training a neural network model by using the natural driving data set; and introducing an ODD boundary and a vehicle endurance constraint, and constructing a complex constraint vehicle path optimization model to realize path optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned electric card deployment, and in particular to a complexity learning-driven unmanned electric card collection and distribution path optimization method. Background Art

[0002] Autonomous driving technology has great potential in improving the safety of logistics and transportation, reducing labor costs and improving efficiency. In recent years, governments around the world have successively introduced support policies to encourage the application and testing of autonomous driving vehicles in controlled environments such as closed parks. Ports are relatively closed and have stable road conditions, making them ideal application scenarios for autonomous driving technology. At present, some large ports at home and abroad have tried to apply unmanned electric container trucks (Automated Electric Container Trucks) to internal port logistics operations, and coordinated with automated terminal loading and unloading equipment, initially realizing unmanned horizontal transportation in ports.

[0003] However, the existing deployment methods of unmanned electric trucks still have many shortcomings. The battery life of unmanned electric trucks is limited, making it difficult to support long-distance transportation operations; the road environment is complex and changeable, and the autonomous driving system is difficult to adapt to situations that exceed the autonomous driving system design conditions (ODD). Therefore, it is necessary to evaluate the complexity of the roads outside the port to determine the applicability of unmanned electric trucks, and use neural network models to optimize the path. Summary of the invention

[0004] According to the technical problems raised above, a complexity learning-driven unmanned electric truck collection and distribution path optimization method is provided. The present invention considers static road properties, dynamic traffic flow and weather factors, and vehicle flow-road network interaction characteristics, evaluates road complexity, and uses a neural network model to construct and optimize a complex constraint vehicle path optimization model to achieve path optimization.

[0005] The technical means adopted by the present invention are as follows:

[0006] A complexity learning-driven unmanned electric truck collection and distribution path optimization method, comprising:

[0007] Taking into account static road properties, dynamic traffic flow, weather factors, and the interaction characteristics between traffic flow and road network, a natural driving dataset is constructed based on n road scene samples;

[0008] The independent weight method MCC is used to calculate the objective weight of each feature of the road scene sample to characterize the difference in the contribution of different features to complexity;

[0009] The approximate ideal solution TOPSIS is used to quantify the complexity of road scene samples;

[0010] Build a neural network model to adaptively learn the inherent laws of road scene samples and estimate the complexity of dynamic uncertain scenes;

[0011] Use natural driving datasets to train neural network models;

[0012] ODD boundaries and vehicle endurance constraints are introduced to build a complex constraint vehicle path optimization model to achieve path optimization.

[0013] Furthermore, n road scene samples of port distribution roads are extracted from the natural driving data set, each sample contains m feature attributes, r = (r1, r2, ..., r m ) represents any road scene sample, r j is the j characteristic attribute value of the sample, r j is the value vector of the jth feature in all samples.

[0014] Furthermore, the independence weight method MCC is performed by j Perform weighted regression to construct a linear relationship with other features; normalize the complex correlation coefficient c j Get feature r j The weight w j :

[0015]

[0016] Multiple correlation coefficient c j Used to measure the independence between the jth feature and other features.

[0017] Furthermore, the ideal most complex scenario is introduced into the approximate ideal solution TOPSIS And the simplest ideal scenario in

[0018] The road scene sample r is compared with the ideal most complex scene r + And the ideal simplest scenario - The weighted Euclidean distance d + and d - It is expressed as:

[0019]

[0020] The weighted Euclidean distance d + and d - Transformed into the complexity score C of the road scene sample r d ′:

[0021]

[0022] Among them, the complexity score Cd The value range of ′ is [0,1].

[0023] Furthermore, the neural network model includes an input layer, a hidden layer and an output layer;

[0024] The input layer receives the m-dimensional feature vector r=(r1, r2, ..., r m ), the output layer generates the corresponding complexity estimate C d ; The forward propagation process of the neural network is expressed as:

[0025]

[0026] Where N(·) represents the Sigmoid activation function, I represents the number of neurons in the input layer, H represents the number of neurons in the hidden layer, and O represents the number of neurons in the output layer; ω hi represents the connection weight from the input layer to the hidden layer, ω oh represents the connection weight from the hidden layer to the output layer, θ h represents the hidden layer neuron threshold, θ o represents the neuron threshold of the output layer.

[0027] Furthermore, a particle swarm optimization algorithm PSO is used to optimize the connection weights in the neural network model, the optimal solution obtained by the particle swarm optimization algorithm is used as the initial connection weights of the neural network model, and the error back propagation algorithm BP and the gradient descent algorithm are used to fine-tune the neural network model;

[0028] Define the mean square error loss function:

[0029]

[0030] Among them, C d ′ represents the complexity score, C d represents the complexity estimate, and e represents the error between the estimate and the actual value;

[0031] The gradient descent method is used to iteratively update the neuron connection weights and thresholds:

[0032]

[0033] Among them, η is the learning step size, δ∈{ω hi ,ω oh ,θ h ,θ o} is the neuron connection weight or threshold.

[0034] Furthermore, the training of the neural network model using the natural driving data set specifically includes:

[0035] Representative road scene fragments are randomly selected from the natural driving dataset, road scene samples are selected, and the characteristic attributes of the samples are extracted; the independence weight method and the approximate ideal solution are used to generate the complexity label of each scene, and the labels are randomly divided into training set and test set in proportion; the training set is input into the neural network model for model training, and the particle swarm optimization algorithm is used to optimize the parameters of the neural network model.

[0036] Furthermore, the complex constraint vehicle path optimization model is based on the following assumptions:

[0037] Assume that the port collection and distribution network is represented by a complete graph G = (V o ,E), where V o is a node set, including port node 0 and customer node set V, and E is a directed arc set between nodes;

[0038] Assume that the import and export box requirements of the customer node are known and can be split, that is, each node can be served multiple times by multiple unmanned electric trucks;

[0039] Assume that the unmanned electric truck fleet consists of pure electric trucks with the same TEU rating and range, and all support autonomous driving within a specific ODD boundary;

[0040] Assume that random factors are not considered and the energy consumption per unit distance is constant;

[0041] Assume that the container size of each customer node is unified as standard TEU, and empty containers can be shared between nodes;

[0042] It is assumed that customer demand is known and fixed during the planning period, and the dynamic changes in demand are not considered.

[0043] Furthermore, the ODD boundary and vehicle range constraints are introduced to construct a complex constraint vehicle path optimization model, which specifically includes:

[0044] Considering the multiple pick-up and delivery characteristics of container transportation and the empty container sharing mode; the complex constraint vehicle path optimization model is modeled as a mixed integer programming model, which is expressed as:

[0045]

[0046] Where V represents the set of demand nodes, V o Represents a node set, where V o = {0}∪V, 0 represents the port, K ​​represents the set of unmanned electric cards, and T represents the set of unmanned electric card trip times; r ij The road scene vector representing the transportation route between nodes i and j, d ij represents the distance between nodes i and j, p i represents the export box demand of node i, di represents the demand for imported containers at node i, Q represents the carrying capacity of the unmanned electric truck, L represents the range of the unmanned electric truck, represents the ODD boundary complexity of the unmanned electric card, is a 0-1 variable, which is 1 if the k-th unmanned electric card drives from i to j during the tth trip, otherwise it is 0; represents the volume of export containers loaded by the k-th unmanned electric truck during its t-th trip from i to j, It represents the quantity of imported containers loaded by unmanned electric truck No. k during its t-th trip from i to j.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] The complexity learning-driven unmanned electric truck collection and distribution path optimization method provided by the present invention comprehensively considers static road attributes, dynamic traffic flow, weather factors, and the interaction characteristics of traffic flow and road network, and provides a system framework for comprehensively evaluating the complexity of the road network in this scenario. A natural driving data set is constructed based on n road scene samples; the independent weight method MCC is used to calculate the objective weights of each feature of the road scene sample to characterize the difference in the contribution of different features to complexity; the approximate ideal solution TOPSIS is used to quantify the complexity of the road scene sample; a nonlinear mapping relationship between generalized scene features and complexity is established using a machine learning model to achieve adaptive evaluation of dynamic and nonlinear road scenes. A neural network model is built to adaptively learn the inherent laws of road scene samples and estimate the complexity of dynamic uncertain scenes; the neural network model is trained using a natural driving data set; ODD boundaries and vehicle endurance constraints are introduced to construct a complex constraint vehicle path optimization model to achieve path optimization; the multiple pick-up and delivery characteristics of container transportation and the empty box sharing mode are considered, which can effectively optimize the energy consumption cost of unmanned electric trucks under different levels of autonomous driving.

[0049] Based on the above reasons, the present invention can be widely promoted in the field of unmanned electric card deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0051] Figure 1 This is a flow chart of the unmanned electric truck collection and distribution path optimization method driven by complexity learning in the present invention.

[0052] Figure 2It is a technical roadmap in the embodiment of the present invention.

[0053] Figure 3 It is a schematic diagram of the unmanned electric truck collection and distribution route planning scheme in an embodiment of the present invention.

[0054] Figure 4 This is the neural network model structure for evaluating the complexity of road scenes in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0057] like Figure 1 As shown, the present invention provides a complexity learning driven unmanned electric truck collection and distribution path optimization method, comprising:

[0058] Taking into account static road properties, dynamic traffic flow, weather factors, and the interaction characteristics between traffic flow and road network, a natural driving dataset is constructed based on n road scene samples;

[0059] In the specific implementation, as a preferred embodiment of the present invention, n road scene samples of port collection and distribution roads are extracted from the natural driving data set, each sample contains m feature attributes, r = (r1, r2, ..., r m ) represents any road scene sample, r j is the j characteristic attribute value of the sample, r j is the value vector of the jth feature in all samples.

[0060] The independent weight method MCC is used to calculate the objective weight of each feature of the road scene sample to characterize the difference in the contribution of different features to complexity;

[0061] In specific implementation, as a preferred embodiment of the present invention, the independence weight method MCC is performed by j Perform weighted regression to build a linear relationship with other features; normalize the complex correlation coefficient c j Get feature r j The weight w j :

[0062]

[0063] Multiple correlation coefficient c j It is used to measure the independence between the jth feature and other features. In implementation, c j The smaller the value, the higher the r j The weaker the correlation with other features, the more independent information it carries.

[0064] The approximate ideal solution TOPSIS is used to quantify the complexity of road scene samples;

[0065] In specific implementation, as a preferred embodiment of the present invention, the ideal most complex scenario is introduced into the approximate ideal solution TOPSIS And the simplest ideal scenario in

[0066] The road scene sample r is compared with the ideal most complex scene r + And the ideal simplest scenario - The weighted Euclidean distance d + and d - It is expressed as:

[0067]

[0068] The weighted Euclidean distance d + and d - Transformed into the complexity score C of the road scene sample r d ′:

[0069]

[0070] Among them, the complexity score C d The value range of ′ is [0,1].

[0071] Build a neural network model to adaptively learn the inherent laws of road scene samples and estimate the complexity of dynamic uncertain scenes;

[0072] In specific implementation, as a preferred embodiment of the present invention, the neural network model includes an input layer, a hidden layer and an output layer. Figure 4 As shown;

[0073] The input layer receives the m-dimensional feature vector r=(r1, r2, ..., r m ), the output layer generates the corresponding complexity estimate C d ; The forward propagation process of the neural network is expressed as:

[0074]

[0075] Where N(·) represents the Sigmoid activation function, I represents the number of neurons in the input layer, H represents the number of neurons in the hidden layer, and O represents the number of neurons in the output layer; ω hi represents the connection weight from the input layer to the hidden layer, ω oh represents the connection weight from the hidden layer to the output layer, θ h represents the hidden layer neuron threshold, θ o represents the neuron threshold of the output layer.

[0076] In the specific implementation, as a preferred embodiment of the present invention, a particle swarm optimization algorithm PSO is used to optimize the connection weights in the neural network model, the optimal solution obtained by the particle swarm optimization algorithm is used as the initial connection weights of the neural network model, and the error back propagation algorithm BP and the gradient descent algorithm are used to fine-tune the neural network model;

[0077] Define the mean square error loss function:

[0078]

[0079] Among them, C d ′ represents the complexity score, C d represents the complexity estimate, and e represents the error between the estimate and the actual value;

[0080] The gradient descent method is used to iteratively update the neuron connection weights and thresholds:

[0081]

[0082] Among them, η is the learning step size, δ∈{ω hi ,ω oh ,θ h ,θ o} is the neuron connection weight or threshold.

[0083] Use natural driving datasets to train neural network models;

[0084] In specific implementation, as a preferred embodiment of the present invention, the use of natural driving data sets to train the neural network model specifically includes:

[0085] Representative road scene fragments are randomly selected from the natural driving dataset, road scene samples are selected, and the characteristic attributes of the samples are extracted; the independence weight method and the approximate ideal solution are used to generate the complexity label of each scene, and the labels are randomly divided into training set and test set in proportion; the training set is input into the neural network model for model training, and the particle swarm optimization algorithm is used to optimize the parameters of the neural network model.

[0086] ODD boundaries and vehicle endurance constraints are introduced to build a complex constraint vehicle path optimization model to achieve path optimization.

[0087] In specific implementation, as a preferred embodiment of the present invention, the complex constraint vehicle path optimization model is based on the following assumptions:

[0088] Assume that the port collection and distribution network is represented by a complete graph G = (V o ,E), where V o is a node set, including port node 0 and customer node set V, and E is a directed arc set between nodes;

[0089] Assume that the import and export box requirements of the customer node are known and can be split, that is, each node can be served multiple times by multiple unmanned electric trucks;

[0090] Assume that the unmanned electric truck fleet consists of pure electric trucks with the same TEU rating and range, and all support autonomous driving within a specific ODD boundary;

[0091] Assume that random factors are not considered and the energy consumption per unit distance is constant;

[0092] Assume that the container size of each customer node is unified as standard TEU, and empty containers can be shared between nodes;

[0093] It is assumed that customer demand is known and fixed during the planning period, and the dynamic changes in demand are not considered.

[0094] In specific implementation, as a preferred embodiment of the present invention, the introduction of ODD boundaries and vehicle endurance constraints to construct a complex constraint vehicle path optimization model specifically includes:

[0095] Considering the multiple pick-up and delivery characteristics of container transportation and the empty container sharing mode; the complex constraint vehicle path optimization model is modeled as a mixed integer programming model, which is expressed as:

[0096] Minimize the total driving distance of the unmanned electric truck fleet:

[0097]

[0098] Limit the unmanned electric card to only be used when the complexity does not exceed the ODD boundary value Execute autonomous driving on the path:

[0099]

[0100] N(r ij ) represents the road attribute vector r ij A function that is mapped to a complexity score; the function can be implemented by a trained neural network model;

[0101] Ensure that each client node is visited at least once and that there are no sub-circuits in a single trip:

[0102]

[0103] Vehicle range limit:

[0104]

[0105] The traffic balance equation ensures that the number of unmanned electric vehicles entering and leaving each node is equal:

[0106]

[0107] The cargo flow balance equation ensures that the loading and unloading volume matches the node demand:

[0108]

[0109] The cargo volume on the restriction arc does not exceed the vehicle capacity:

[0110]

[0111] Define the value range of the decision variable:

[0112]

[0113] Where V represents the set of demand nodes, V o Represents a node set, where V o = {0}∪V, 0 represents the port, K ​​represents the set of unmanned electric cards, and T represents the set of unmanned electric card trip times; r ij The road scene vector representing the transportation route between nodes i and j, d ij represents the distance between nodes i and j (km), p i represents the export box demand (TEU) of node i, d i represents the import container demand of node i (TEU), Q represents the carrying capacity of the unmanned electric truck (ft), L represents the cruising range of the unmanned electric truck (km), represents the ODD boundary complexity of the unmanned electric card, is a 0-1 variable, which is 1 if the k-th unmanned electric card drives from i to j during the tth trip, otherwise it is 0; represents the volume of export containers (TEU) loaded by the k-th unmanned electric truck during its t-th trip from i to j, It represents the quantity of imported containers (TEU) loaded by unmanned electric truck No. k during its t-th trip from i to j.

[0114] Example

[0115] like Figure 1 As shown, the present invention provides a complexity learning driven unmanned electric truck collection and distribution path optimization method. Figure 3 A simplified diagram of port collection and distribution operations is given. There are multiple container shippers distributed in the hinterland of a port, and the unmanned electric truck needs to perform container transportation tasks between the port and the shippers. The unmanned electric truck departs from the port and delivers two imported heavy containers to customer A via route r1. After loading one exported heavy container and one empty container, it was originally planned to return to the port directly via r2. However, after optimization, it can go to customer B via r3 to load the second exported heavy container, and then return via r4.

[0116] The transportation process includes two stages: transporting imported heavy containers from the port to each consignor, and transporting exported heavy containers from each consignor back to the port. In order to improve the overall transportation efficiency and make full use of the return load capacity, the unmanned electric truck can load and transport the generated exported heavy containers or idle empty containers to other consignors in need after visiting the import consignor, and then pull the exported heavy containers back to the terminal for loading, completing a business cycle.

[0117]

[0118]

[0119] A quantitative method integrating multi-attribute decision-making and machine learning is proposed. This method labels the complexity of representative scenes through multi-attribute decision-making, and uses machine learning models to establish a nonlinear mapping relationship between generalized scene features and complexity, thereby realizing adaptive evaluation of dynamic and nonlinear road scenes.

[0120] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0121] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0123] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0124] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0125] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A complexity learning driven unmanned electric truck collection and distribution path optimization method, characterized in that: include: Taking into account static road properties, dynamic traffic flow, weather factors, and the interaction characteristics between traffic flow and road network, a natural driving dataset is constructed based on n road scene samples; The independent weight method MCC is used to calculate the objective weight of each feature of the road scene sample to characterize the difference in the contribution of different features to complexity; The approximate ideal solution TOPSIS is used to quantify the complexity of road scene samples; Build a neural network model to adaptively learn the inherent laws of road scene samples and estimate the complexity of dynamic uncertain scenes; Use natural driving datasets to train neural network models; ODD boundaries and vehicle endurance constraints are introduced to build a complex constraint vehicle path optimization model to achieve path optimization.

2. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The natural driving dataset extracts n road scene samples of port distribution roads, each of which contains m feature attributes, r = (r1, r2, ..., r m ) represents any road scene sample, r j is the j characteristic attribute value of the sample, r j is the value vector of the jth feature in all samples.

3. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The independence weight method MCC is to calculate the jth feature r j Perform weighted regression to build linear relationships with other features; Normalized multiple correlation coefficient c j Get feature r j The weight w j : Multiple correlation coefficient c j Used to measure the independence of the jth feature from other features.

4. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: Introducing the ideal most complex scenario into the approximate ideal solution TOPSIS And the simplest ideal scenario in The road scene sample r is compared with the ideal most complex scene r + And the ideal simplest scenario - The weighted Euclidean distance d + and d - It is expressed as: The weighted Euclidean distance d + and d - Transformed into the complexity score C of the road scene sample r d ′: Among them, the complexity score C d The value range of ′ is [0,1].

5. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The neural network model includes an input layer, a hidden layer and an output layer; The input layer receives the m-dimensional feature vector r=(r1, r2, ..., r m ), the output layer generates the corresponding complexity estimate C d ; The forward propagation process of the neural network is expressed as: Where N(·) represents the Sigmoid activation function, I represents the number of neurons in the input layer, H represents the number of neurons in the hidden layer, and O represents the number of neurons in the output layer; ω hi represents the connection weight from the input layer to the hidden layer, ω oh represents the connection weight from the hidden layer to the output layer, θ h represents the hidden layer neuron threshold, θ o represents the neuron threshold of the output layer.

6. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 5 is characterized in that: The particle swarm optimization algorithm PSO is used to optimize the connection weights in the neural network model, the optimal solution obtained by the particle swarm optimization algorithm is used as the initial connection weights of the neural network model, and the error back propagation algorithm BP and the gradient descent algorithm are used to fine-tune the neural network model; Define the mean square error loss function: Among them, C d ′ represents the complexity score, C d represents the complexity estimate, and e represents the error between the estimate and the actual value; The gradient descent method is used to iteratively update the neuron connection weights and thresholds: Among them, η is the learning step size, δ∈{ω hi ,ω oh ,θ h ,θ o } is the neuron connection weight or threshold.

7. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The use of the natural driving data set to train the neural network model specifically includes: Representative road scene fragments are randomly selected from the natural driving dataset, road scene samples are selected, and the characteristic attributes of the samples are extracted; the independence weight method and the approximate ideal solution are used to generate the complexity label of each scene, and the labels are randomly divided into training set and test set in proportion; the training set is input into the neural network model for model training, and the particle swarm optimization algorithm is used to optimize the parameters of the neural network model.

8. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The complex constraint vehicle routing optimization model is based on the following assumptions: Assume that the port collection and distribution network is represented by a complete graph G = (V o ,E), where V o is a node set, including port node 0 and customer node set V, and E is a directed arc set between nodes; Assume that the import and export box requirements of the customer node are known and can be split, that is, each node can be served multiple times by multiple unmanned electric trucks; Assume that the unmanned electric truck fleet consists of pure electric trucks with the same TEU rating and range, and all support autonomous driving within a specific ODD boundary; Assume that random factors are not considered and the energy consumption per unit distance is constant; Assume that the container size of each customer node is unified as standard TEU, and empty containers can be shared between nodes; It is assumed that customer demand is known and fixed during the planning period, and the dynamic changes in demand are not considered.

9. The complexity learning driven unmanned electric truck collection and distribution path optimization method according to claim 1 is characterized in that: The introduction of ODD boundaries and vehicle endurance constraints to construct a complex constraint vehicle path optimization model specifically includes: Considering the multiple pick-up and delivery characteristics of container transportation and the empty container sharing mode; the complex constraint vehicle path optimization model is modeled as a mixed integer programming model, which is expressed as: Where V represents the set of demand nodes, V o Represents a node set, where V o = {0}∪V, 0 represents the port, K ​​represents the set of unmanned electric cards, and T represents the set of unmanned electric card trip times; r ij The road scene vector representing the transportation route between nodes i and j, d ij represents the distance between nodes i and j, p i represents the export box demand of node i, d i represents the demand for imported containers at node i, Q represents the carrying capacity of the unmanned electric truck, L represents the range of the unmanned electric truck, represents the ODD boundary complexity of the unmanned electric card, is a 0-1 variable, which is 1 if the k-th unmanned electric card drives from i to j during the tth trip, otherwise it is 0; represents the volume of export containers loaded by the k-th unmanned electric truck during its t-th trip from i to j, It represents the quantity of imported containers loaded by unmanned electric truck No. k during its t-th trip from i to j.

Images