Multi-vehicle path planning method, system, device and medium generated by column

Abstract

The present invention provides a column-generated multi-vehicle path planning method, system, device, and medium, relating to the field of vehicle path planning technology. The method comprises: step S1: constructing data annotations and converting the data annotation problem into a mixed integer programming problem; step S2: solving the mixed integer programming problem through each iteration, annotating the data, and performing neural network training; and step S3: calculating multi-vehicle path planning based on the trained neural network. The present invention constructs a path planning algorithm applicable to a variable number of vehicles while maintaining speed and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle path planning, and more specifically, to a method for selecting columns generated in a column-generated multi-vehicle path planning method based on machine learning, and more particularly to a column-generated multi-vehicle path planning method, system, device, and medium. Background Art

[0002] The Vehicle Routing Problem (VRP) is a combinatorial and integer programming problem. A generalization of the Traveling Salesman Problem (TSP), it aims to find the optimal vehicle allocation and route assignment to deliver goods to all customers, given a set of stops, customers, and a fleet of vehicles. VRP is widely encountered in real-world logistics and transportation scenarios, such as the delivery of parts to manufacturing facilities.

[0003] The VRP problem has been proven to be NP-hard, so it's impossible to guarantee an optimal solution without exhaustively enumerating all possible solutions. Currently, some heuristic methods are available for solving VRP, but their effectiveness is limited, especially as the computational scale increases, where efficiency decreases significantly.

[0004] The column generation algorithm is an effective method for solving optimization problems involving a large number of variables. Its core idea is to select a subset of variables as initial solutions and iteratively generate new variables that optimize the objective function. This cycle continues until no more such new variables can be found. For highly complex problems, column generation methods may generate a large number of new columns in each iteration, significantly increasing the difficulty of selecting the optimal column and thus reducing efficiency.

[0005] Machine learning is a branch of artificial intelligence. Machine learning algorithms automatically analyze patterns in data and use them to make predictions about new data. In recent years, a number of related works have attempted to use machine learning algorithms to solve combinatorial optimization problems or accelerate existing optimization methods.

[0006] Patent publication number CN113822486B discloses a vehicle route planning method, device, and system based on a column generation algorithm. The method comprises: upon receiving a route planning signal, obtaining order data to be assigned at the current logistics node; determining multiple route points to be planned based on the order data; dividing the multiple route points into multiple sector-shaped initial groups with the unified departure node of the delivery vehicle as the origin; determining the total number of subproblems based on the multiple sector-shaped initial groups; constructing multiple subproblems corresponding to the total number to construct an initial main problem; and iteratively solving the multiple subproblems and the main problem based on the column generation algorithm to determine a set of feasible vehicle delivery task routes to be assigned. In each column generation iteration, each subproblem corresponds to a subproblem construction group, and the subproblem construction group includes at least one route point from the sector-shaped initial group. This patent does not allow for filtering of the iteratively generated new columns, resulting in slow execution and a long runtime. Summary of the Invention

[0007] To address the deficiencies in the prior art, the present invention provides a column-generated multi-vehicle path planning method, system, device, and medium that can solve large-scale vehicle path planning problems. Unlike other traditional methods that use column-generated algorithms to solve path planning problems, this solution utilizes a neural network to filter the new columns generated in each iteration, thereby accelerating the calculation speed and reducing the running time.

[0008] According to the present invention, a column-generated multi-vehicle path planning method, system, device, and medium are provided, and the solution is as follows:

[0009] In a first aspect, a multi-vehicle path planning method for column generation is provided, the method comprising:

[0010] Step S1: constructing data annotation, converting the data annotation problem into a mixed integer programming problem;

[0011] Step S2: Solve the mixed integer programming problem through each iteration, label the data, and perform neural network training.

[0012] Preferably, the mixed integer programming problem includes: for existing columns, calculating the corresponding dual problem, then calculating the difference cost of each column, and selecting the corresponding column with the smallest negative value:

[0013]

[0014] Among them, c p represents the economic cost of the multi-vehicle solution corresponding to each column; θ p is the variable of the problem, indicating the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; b, a pThe expression represents all constraints that the column needs to satisfy, such as time constraints. p represents the time coefficient of the column, b represents the constraint of the time window; Ω l represents the set of variables in the original problem at the beginning of each iteration; l is the number of iterations; G t The set of columns generated for this iteration, for each generated column p∈G t , define a binary variable y p , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term ∈y is added to the objective function p , where ∈ is a small positive number;

[0015] Each iteration of column generation produces a set G consisting of generated columns t , the goal is to select G t A subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select.

[0016] Preferably, the numerical features selected by the neural network in step S2 include: the difference cost of each column, the number of stations visited by the reroute, the amount of cargo taken away by each station, the arrival time window, a zero-one variable indicating whether the column is newly generated, and vehicle type and cargo volume data;

[0017] After the selected numerical features are normalized, they are processed using a convolutional neural network (CNN). The learned features are then passed through several fully connected layers and finally a unit layer. The activation function is sigmoid and the loss function is binary cross entropy.

[0018] Preferably, the neural training step includes:

[0019] 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results;

[0020] 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

[0021] In a second aspect, a multi-vehicle path planning system with column generation is provided, the system comprising:

[0022] Module M1: Constructing data annotation, converting the data annotation problem into a mixed integer programming problem;

[0023] Module M2: solves the mixed integer programming problem through each iteration, labels the data, and performs neural network training.

[0024] Preferably, the mixed integer programming problem includes: for existing columns, calculating the corresponding dual problem, then calculating the difference cost of each column, and selecting the corresponding column with the smallest negative value:

[0025]

[0026] Among them, c p represents the economic cost of the multi-vehicle solution corresponding to each column; θ p is the variable of the problem, indicating the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; b, a p The expression represents all constraints that the column needs to satisfy, such as time constraints. p represents the time coefficient of the column, b represents the constraint of the time window; Ω l represents the set of variables in the original problem at the beginning of each iteration; l is the number of iterations; G t The set of columns generated for this iteration, for each generated column p∈G t , define a binary variable y p , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term ∈y is added to the objective function p , where ∈ is a small positive number;

[0027] Each iteration of column generation produces a set G consisting of generated columns t , the goal is to select G t A subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select.

[0028] Preferably, the numerical features selected by the neural network in the module M2 include: the difference cost of each column, the number of stations visited by the reroute, the amount of cargo taken away by each station, the arrival time window, a zero-one variable indicating whether the column is newly generated, and the vehicle type and cargo volume data;

[0029] After the selected numerical features are normalized, they are processed using a convolutional neural network (CNN). The learned features are then passed through several fully connected layers and finally a unit layer. The activation function is sigmoid and the loss function is binary cross entropy.

[0030] Preferably, the neural training step includes:

[0031] 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results;

[0032] 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

[0033] According to a third aspect, a device is provided, comprising:

[0034] one or more processors;

[0035] a storage device for storing one or more programs,

[0036] When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the steps in the method.

[0037] In a fourth aspect, a computer-readable storage medium storing a computer program is provided, wherein the computer program implements the steps in the method when executed by a processor.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention constructs a path planning algorithm that can be applied to a variable number of vehicles while maintaining speed and accuracy;

[0040] 2. The present invention uses a neural network to filter the columns generated by the column generation algorithm, thereby reducing the running time.

[0041] Other beneficial effects of the present invention will be explained through the introduction of specific technical features and technical solutions in the specific implementation methods. Those skilled in the art should be able to understand the beneficial technical effects brought about by the introduction of these technical features and technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0043] Figure 1 It is the overall flow chart of the present invention;

[0044] Figure 2 A neural network diagram. DETAILED DESCRIPTION

[0045] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0046] The embodiment of the present invention provides a multi-vehicle path planning method based on column generation, which is mainly aimed at the multi-vehicle path planning problem. The overall optimal path is calculated through a column generation algorithm, and a neural network is used to filter the new columns generated in each iteration, thereby speeding up the calculation speed. Figure 1 As shown, the method specifically includes the following contents:

[0047] Step S1: First, we need to construct data annotations. For the existing columns, we need to calculate the reduced cost of each column (in actual business, this can be regarded as the economic cost of a trip) and select the corresponding column with the smallest value. This cost problem can be transformed into a mixed integer programming problem:

[0048]

[0049] Among them, c p represents the economic cost of the multi-vehicle solution corresponding to each column; θ p is the variable of the problem, indicating the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; b, a p The expression represents all constraints that the column needs to satisfy, such as time constraints. p represents the time coefficient of the column, b represents the constraint of the time window; Ω l represents the set of variables in the original problem at the beginning of each iteration; l is the number of iterations; G t The set of columns generated for this iteration, for each generated column p∈G t , define a binary variable y p , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term ∈y is added to the objective function p , where ∈ is a small positive number;

[0050] Each iteration of column generation produces a set G consisting of generated columns t , the goal is to select G t A subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select.

[0051] Step S2: Solve the mixed integer programming problem through each iteration, label the data, and perform neural network training.

[0052] Step S3: After the neural network training is completed, the multi-vehicle path planning is calculated based on it. The specific steps are as follows:

[0053] 1) First, generate a set of mediocre solutions, such as sending a train to each station, each train only goes to one station and then returns to the depot. Each set of such solutions is called a "train".

[0054] 2) Based on the existing columns, new "columns" are continuously generated iteratively.

[0055] a) Select any two columns from the existing columns and merge them into one column (one train number).

[0056] b) Check whether the cargo volume exceeds the vehicle load limit.

[0057] c) Solve the Traveling Salesman Problem (TSP) for all sites in the two columns. You can use the LKH algorithm or some heuristic algorithms, such as genetic algorithm, simulated annealing, particle swarm algorithm, ant colony algorithm, etc. to find the optimal site visiting order.

[0058] d) Calculate the time window for the obtained access sequence and check whether the time window constraints of all sites are met. If not, return to step c) and take the suboptimal solution.

[0059] e) If the newly generated column satisfies the constraints of b, c, and d, add the column to the newly generated column and use the load mileage as its evaluation score (reduced cost).

[0060] f) Continue looping ae until no new columns can be generated.

[0061] 3) All generated columns and features are processed and input into the neural network for prediction, and all generated new columns are filtered based on the results.

[0062] 4) Add the filtered new columns to the original problem and return to step 2) to iterate again until the upper limit of the number of iterations is reached or no new columns are filtered out.

[0063] The numerical features selected by the neural network include: the reduced cost of each column, the number of stations visited by the reroute, the amount of cargo taken away at each station, the arrival time window, a zero-one variable indicating whether to select the newly generated column, and vehicle model and cargo volume data.

[0064] Reference Figure 2 As shown in the figure, the selected numerical features are normalized and processed using a convolutional neural network (CNN). The learned features are then passed through several fully-connected-dense-layers and finally through a unit layer. The activation function is sigmoid and the loss function is binary-cross-entropy.

[0065] The neural training steps include:

[0066] 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results;

[0067] 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

[0068] The present invention also provides a multi-vehicle path planning system generated in columns. This system can be implemented by executing the process steps of the multi-vehicle path planning method described in columns. Those skilled in the art can understand the multi-vehicle path planning method described in columns as a preferred embodiment of the multi-vehicle path planning system described in columns. Specifically, the system includes:

[0069] Module M1: First, we need to construct data annotation. For the existing columns, we need to solve the corresponding dual problem to calculate the reduced cost of each column and select the corresponding column with the smallest negative value. This problem can be transformed into a mixed integer programming problem:

[0070]

[0071] Among them, c p represents the economic cost of the multi-vehicle solution corresponding to each column; θ p is the variable of the problem, indicating the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; b, a p The expression represents all constraints that the column needs to satisfy, such as time constraints. p represents the time coefficient of the column, b represents the constraint of the time window; Ω l represents the set of variables in the original problem at the beginning of each iteration; l is the number of iterations; G t The set of columns generated for this iteration, for each generated column p∈G t , define a binary variable y p , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term ∈y is added to the objective function p , where ∈ is a small positive number;

[0072] Each iteration of column generation produces a set G consisting of generated columns t , the goal is to select G tA subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select.

[0073] Module M2: solves the mixed integer programming problem through each iteration, labels the data, and performs neural network training.

[0074] The numerical features selected by the neural network include: the reduced cost of each column, the number of stations visited by the reroute, the amount of cargo taken away by each station, the arrival time window, a zero-one variable indicating whether the column is newly generated, and the vehicle model and cargo volume data.

[0075] Reference Figure 2 As shown in the figure, the selected numerical features are normalized and processed using a convolutional neural network (CNN). The learned features are then passed through several fully-connected-dense-layers and finally through a unit layer. The activation function is sigmoid and the loss function is binary-cross-entropy.

[0076] The neural training steps include:

[0077] 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results;

[0078] 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

[0079] Next, the present invention will be described in more detail through examples.

[0080] In a specific vehicle routing problem, given a fleet of N vehicles, each with a payload volume V, a series of n stops, and the cargo volume at each stop, we need to determine the order in which each vehicle should depart, the stops it passes through, the cargo volume it picks up, and ultimately return to its starting point. The algorithm steps for column generation are as follows:

[0081] (1) First, a set of mediocre solutions is generated. For example, a train is dispatched to each station, and each train only goes to one station, and then there is a parking lot. Each set of such solutions is called a "train";

[0082] (2) Based on the existing columns, new "columns" are continuously generated iteratively:

[0083] a) Select any two columns from the existing columns and merge them into one column (one train number);

[0084] b) Check whether the cargo volume exceeds the vehicle load limit;

[0085] c) Solve the Traveling Salesman Problem (TSP) for all sites in the two columns. You can use the LKH algorithm or some heuristic algorithms, such as genetic algorithm, simulated annealing, particle swarm optimization, ant colony algorithm, etc. to find the optimal order of site visits;

[0086] d) Calculate the time window for the obtained access sequence and check whether the time window constraints of all sites are met. If not, return to step c) and take the suboptimal solution;

[0087] e) If the newly generated column satisfies the constraints of b, c, and d, add the column to the newly generated column and use the load mileage as its evaluation score (reduced cost);

[0088] f) Continue looping ae until no new columns can be generated.

[0089] (3) All generated columns and features are processed and input into the neural network for prediction, and all generated new columns are filtered based on the results.

[0090] (4) Add the filtered new columns to the original problem and return to step (2) to iterate again until the upper limit of the number of iterations is reached or no new columns are filtered out.

[0091] Embodiments of the present invention provide a column-generated multi-vehicle path planning method, system, device, and medium, which construct a path planning problem algorithm applicable to a variable number of vehicles while maintaining speed and accuracy, thereby improving operational efficiency.

[0092] Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, and units provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same functions of the system and its various devices, modules, and units provided by the present invention in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; the devices, modules, and units for implementing various functions can also be considered as both software modules implementing the method and structures within the hardware component.

[0093] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A multi-vehicle path planning method based on column generation, characterized in that: include: Step S1: constructing data annotation, converting the data annotation problem into a mixed integer programming problem; Step S2: solving the mixed integer programming problem through each iteration, labeling the data, and performing neural network training; Step S3: After the neural network training is completed, the multi-vehicle path planning is calculated based on it; The specific steps of calculating the multi-vehicle path planning are as follows: 1) First, generate a set of mediocre solutions. For each station, send a train. Each train goes to only one station and then returns to the depot. Each set of such solutions is called a train. 2) Based on the existing columns, continuously iteratively generate new columns: a) Select any two columns from the existing columns and merge them into one column; b) Check whether the cargo volume exceeds the vehicle load limit; c) Solve the traveling salesman problem for all the sites in the two columns and find the optimal sequence for visiting the sites; d) Calculate the time window for the obtained access sequence and check whether it meets the time window constraints of all sites. If not, return to step c) and take the suboptimal solution; e) If the newly generated column satisfies the constraints of b), c), and d), add the column to the newly generated column and use the load mileage as its evaluation score; f) Repeat steps a) to e) until no new columns can be generated; 3) All generated columns and features are processed and input into the neural network for prediction, and all generated new columns are filtered based on the results; 4) Add the filtered new columns to the original problem and return to step 2) to iterate again until the upper limit of the number of iterations is reached or no new columns are filtered out.

2. The multi-vehicle path planning method based on column generation according to claim 1, characterized in that: The mixed integer programming problem involves calculating the corresponding dual problem for the existing columns, calculating the difference cost for each column, and selecting the corresponding column with the smallest negative value: in, Indicates the economic cost of the multi-vehicle solution corresponding to each column; is the problem variable, representing the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; , b represents the time window constraint; represents the set of variables in the original problem at the beginning of each iteration; is the number of iterations; The set of columns generated for this iteration, for each generated column , define a binary variable , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term is added to the objective function , in is a small positive number; Each iteration of column generation produces a set of generated columns , the goal is to select A subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select.

3. The multi-vehicle path planning method based on column generation according to claim 2, characterized in that: The numerical features selected by the neural network in step S2 include: the difference cost of each column, the number of stations visited by the path, the amount of cargo taken by each station, the arrival time window, a zero-one variable indicating whether to select the newly generated column, and vehicle type and cargo volume data; After the selected numerical features are normalized, they are processed using a convolutional neural network (CNN). The learned features are then passed through several fully connected layers and finally a unit layer. The activation function is sigmoid and the loss function is binary cross entropy.

4. The multi-vehicle path planning method based on column generation according to claim 2, characterized in that: The neural network training step includes: 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results; 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

5. A multi-vehicle path planning system generated by columns, characterized in that include: Module M1: Constructing data annotation, converting the data annotation problem into a mixed integer programming problem; Module M2: Solve the mixed integer programming problem through each iteration, label the data, and perform neural network training. The mixed integer programming problem includes: for each existing column, calculate the corresponding dual problem, calculate the difference cost of each column, and select the corresponding column with the smallest negative value: in, Indicates the economic cost of the multi-vehicle solution corresponding to each column; is the problem variable, representing the vehicle solution corresponding to each column, and the subscript p refers to one of the columns generated in each iteration; , b represents the time window constraint; represents the set of variables in the original problem at the beginning of each iteration; is the number of iterations; The set of columns generated for this iteration, for each generated column , define a binary variable , if the column is selected, it takes 1, otherwise it takes 0; in order to minimize the number of selected columns, a penalty term is added to the objective function , in is a small positive number; Each iteration of column generation produces a set of generated columns , the goal is to select A subset g is generated and added to the original problem; the original problem is regarded as a binary classification problem. For each column generated, it is classified into two categories y(0, 1), 0 means discard and 1 means select; The specific steps for calculating multi-vehicle path planning are as follows: 1) First, generate a set of mediocre solutions. For each station, send a train. Each train goes to only one station and then returns to the depot. Each set of such solutions is called a train. 2) Based on the existing columns, continuously iteratively generate new columns: a) Select any two columns from the existing columns and merge them into one column; b) Check whether the cargo volume exceeds the vehicle load limit; c) Solve the traveling salesman problem for all the sites in the two columns and find the optimal sequence for visiting the sites; d) Calculate the time window for the obtained access sequence and check whether it meets the time window constraints of all sites. If not, return to step c) and take the suboptimal solution; e) If the newly generated column satisfies the constraints of b), c), and d), add the column to the newly generated column and use the load mileage as its evaluation score; f) Repeat steps a) to e) until no new columns can be generated; 3) All generated columns and features are processed and input into the neural network for prediction, and all generated new columns are filtered based on the results; 4) Add the filtered new columns to the original problem and return to step 2) to iterate again until the upper limit of the number of iterations is reached or no new columns are filtered out.

6. The multi-vehicle path planning system based on column generation according to claim 5, characterized in that: The numerical features selected by the neural network in the module M2 include: the difference cost of each column, the number of stations visited by the path, the amount of cargo taken away by each station, the arrival time window, a zero-one variable indicating whether the column is newly generated, and the vehicle type and cargo volume data; After the selected numerical features are normalized, they are processed using a convolutional neural network (CNN). The learned features are then passed through several fully connected layers and finally a unit layer. The activation function is sigmoid and the loss function is binary cross entropy. The neural network training step includes: 1) Construct a data set, solve the integer programming problem for each column after each iteration, and label the generated columns according to the results; 2) The columns and numerical features after each iteration are processed and input into the neural network to perform feature learning based on the annotations.

7. A multi-vehicle path planning device generated by columns, characterized in that The device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the method according to any one of claims 1 to 4.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images