A dynamic scheduling method and system for vehicle logistics based on deep learning prediction

Abstract

The present invention is applicable to the field of dynamic intelligent scheduling of logistics, and is specifically a dynamic scheduling method and system for vehicle logistics based on deep learning prediction. In the scheduling method, in the upper model of the scheduling model, the order set to be allocated is allocated downward using the recommended transportation path; in the middle model of the scheduling model, a loading scheme selection model is established to maximize the loading value and determine the loading scheme; in the lower model of the scheduling model, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed to maximize the satisfaction of the FLSP; the scheduling model is matched with the predicted order flow to obtain the capacity resource satisfaction of the predicted order flow, and before the time window of the predicted order flow is reached, the capacity distribution is dynamically adjusted based on the capacity resource satisfaction. The present invention can make full use of capacity resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of dynamic intelligent scheduling of logistics, and in particular to a method and system for dynamic scheduling of vehicle logistics based on deep learning prediction. Background Art

[0002] The allocation of vehicle logistics orders is one of the key decisions in the field of vehicle logistics. The quality of the decision-making level will directly affect the delivery level and cost of the vehicle. In recent years, competition in the automotive industry has continued to intensify, placing higher demands on delivery levels; highway transportation regulations have continued to tighten, and delivery costs are facing greater pressure. At present, the intelligent decision-making system for vehicle logistics is in its infancy, focusing on the problem of highway order allocation at the current point in time. It lacks a forward-looking scientific plan for full-process order allocation, and it is difficult to take into account both delivery level requirements and delivery cost targets at the same time.

[0003] The automotive logistics industry faces high complexity and technical challenges in resource scheduling, which directly affects customer satisfaction, transportation tool utilization and overall transportation costs. Formulating reasonable scheduling plans to minimize vehicle idleness, optimize transportation routes and reduce costs is a common problem faced by the industry.

[0004] Since the transportation of whole vehicles is mostly point-to-point mode, empty return trips have become the norm in the industry, seriously reducing transportation efficiency and overall profitability. High empty driving rates not only push up logistics costs, but also restrict the competitiveness of logistics companies. The current automobile logistics network is mostly composed of scattered point-to-point transportation routes, lacking an effective full-network linkage and unified scheduling mechanism. This fragmented state limits the optimal allocation of logistics resources, resulting in a low degree of transportation organization and low overall efficiency, which is an important reason for the high costs of the industry. "Idle vehicle driving" is a chronic disease of automobile logistics and even the entire transportation industry, which not only affects corporate profits, but also aggravates the waste of social resources. How to achieve efficient integration and utilization of logistics resources and reduce empty driving rates through technological innovation and model innovation while ensuring service quality is a major issue faced by the industry.

[0005] At present, the logistics scheduling of complete vehicles in the automobile industry has the following problems: First, the sales volume of complete vehicle logistics is predicted without considering the various factors that affect the sales of complete vehicles, which has caused huge challenges in the capacity scheduling of complete vehicle transportation; Second, the existing logistics transportation network in the automobile industry is composed of a large number of scattered "point-to-point" transportation routes. This transportation structure will lead to problems such as low degree of transportation organization, low overall efficiency, and high logistics costs; Third, the current complete vehicle logistics scheduling algorithm in the industry is only a match between the orders and transportation capacity of the current month, and it is still a passive and reactive static scheduling. Summary of the invention

[0006] The purpose of an embodiment of the present invention is to provide a method and system for dynamic scheduling of whole vehicle logistics based on deep learning prediction, which predicts order demand within a future time window through a deep learning prediction model, and pre-schedules transportation resources within the future time window in combination with the spatiotemporal scheduling capability of transportation resources, obtains accurate transportation resource satisfaction, and uses the offset of the transportation resource scheduling model within the time of reaching this time window to achieve a dynamic balance of transportation resource satisfaction, so that transportation resources can be fully utilized.

[0007] To achieve the above object, the present invention provides the following technical solutions.

[0008] An embodiment of the present invention provides a method for dynamic scheduling of vehicle logistics based on deep learning prediction, comprising:

[0009] Construct a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture, use the Transformer layer and the convolutional neural network CNN layer to extract features of the influencing factor data, use the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and output the predicted sales volume;

[0010] The objective function of a multi-constraint model based on transportation cost, transportation time, and carbon emission is constructed using the predicted sales volume, and the objective function of the multi-constraint model is solved using a non-dominated sorting algorithm to obtain a recommended transportation route;

[0011] A scheduling model is configured. In the upper model of the scheduling model, the order set to be assigned is assigned downward using the recommended transportation route. In the middle model of the scheduling model, the loading plan selection model is established to determine the loading plan with the goal of maximizing the loading value. In the lower model of the scheduling model, tasks are reasonably assigned to specific FLSPs based on the loading plan of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP.

[0012] The scheduling model is matched with the predicted order flow to obtain the capacity resource satisfaction of the predicted order flow. Before the time window of the predicted order flow is reached, the capacity distribution is dynamically adjusted based on the capacity resource satisfaction.

[0013] Furthermore, the influencing factor data include economic factors, market factors and policy factors; among them, the policy factors are analyzed and processed using the nlp text quantification tool to convert the policy factor text from qualitative description into quantitative indicators.

[0014] Furthermore, in the step of constructing the objective function of the multi-constraint model based on transportation cost, transportation time, and carbon emissions using the predicted sales volume:

[0015] The transportation cost includes the transportation cost between nodes, the cost of converting the transportation mode at the node, the cost of waiting for service at the node, and the penalty cost for exceeding the optimal delivery time; wherein, the transportation cost is expressed as:

[0016] ;

[0017] In the formula, represents the total transportation cost; is a 0 / 1 variable, indicating whether to choose to transport cargo h from node i to node j via transportation mode k. If yes, it is 1, otherwise it is 0; represents the transportation distance from node i to node j using transportation mode k; Indicates the unit transportation distance cost of each mode of transportation; Indicates the volume or weight of the cargo h; represents the time to reach node l; represents the time when the slave node i is triggered; TE represents the optimal delivery time, and a penalty cost is incurred when the timeout occurs; represents the waiting cost coefficient, which reflects the cost of waiting at node i; H represents the cargo set; R represents the node set, V represents the path set, and K represents the transportation mode set; , when it is 1, it means that the cargo h is converted from transportation mode k to l at node i; Cost coefficient for switching from k to l at node i; represents the total transportation time of cargo h, The critical point for optimal delivery time;

[0018] The transportation time includes the transportation time between nodes, the transportation mode conversion time of the intermediate nodes and the time waiting for service at the nodes; wherein, the transportation time is expressed as:

[0019] ;

[0020] In the formula, represents the total transportation time of cargo h; represents the transportation time from node i to node j using transportation mode k; represents the conversion time of cargo h from transport mode k to mode l at node i;

[0021] The carbon emissions include the carbon emissions during transportation and the carbon emissions from the conversion of transportation modes at the node; wherein the carbon emissions are expressed as:

[0022] ;

[0023] In the formula, represents the total carbon emissions; The unit carbon emission coefficient represents the unit emission from node i to node j using the k-th transportation mode; is the carbon emission coefficient of switching transportation modes, which represents the carbon emission when nodes i and l switch transportation modes;

[0024] The objective function of the multi-constraint model is expressed as: ;

[0025] The transport task has a fixed start and end point represented as: ; Indicates whether to choose to use transportation mode k from node i to node i+1; Indicates the starting point node; Indicates the mode of transport; represents any one; D represents the transport destination;

[0026] Only one mode of transportation can be selected between nodes, and only one conversion can be made at a node, which can be expressed as: , is a binary decision variable used to indicate whether a mode of transport conversion occurs between node i and node l; h represents the index of the cargo;

[0027] The continuity before and after transportation is expressed as: ; Indicates whether transportation mode k is used from node i-1 to node i;

[0028] The transport capacity between nodes does not exceed the maximum allowed transport capacity, which is expressed as: ; It represents the maximum transport capacity of transport mode k from node i to node j;

[0029] The flow from the starting point to the end point is expressed as: ; Indicates whether transportation mode k is used from node g to node i; Indicates a transit node; , , They represent the destination node, activity node, and origin node respectively;

[0030] The time when the mode of transport conversion is completed at node j is equal to the sum of the departure time of the previous node, the conversion time and the section transportation, which can be expressed as: ;

[0031] The time when the conversion is completed is earlier than the departure time: ;in, It represents the time required for node l to switch the mode of transport;

[0032] The completion time of the previous segment task is the departure time of the next segment, which is expressed as: ,in, It represents the time it takes for node l to perform task g using transportation mode i to k; It represents the time for node l to carry out task i to j using transportation mode k;

[0033] The total transportation time of cargo h does not exceed the rated time and is expressed as: ,in, represents the total transportation time of cargo h, It represents the rated maximum transportation time of cargo h;

[0034] The decision variable value constraints are expressed as: .

[0035] Furthermore, the step of solving the objective function of the multi-constraint model using a non-dominated sorting algorithm includes:

[0036] Initialize the parameters of the non-dominated sorting algorithm. The initialization parameters include: population size is 30, number of iterations is 200, crossover operator is 0.6, mutation operator is 0.01;

[0037] The population is selected through fast non-dominated sorting, and the non-dominated solutions are stored in the file. After crossover and mutation operations, the first generation population is generated;

[0038] Set the initial value of the evolutionary generation to 3, and obtain the parent and child populations through evolution;

[0039] Perform fast non-dominated sorting on the parent and child populations, select the population, store the non-dominated solution in a file, calculate the crowding degree of individuals in the population, select the best individuals from the population to enter the next generation, and obtain the new parent-child population; perform selection, crossover, and mutation operations on the new parent-child population;

[0040] Until the maximum evolutionary generation is reached, the individuals in the document are non-dominated and sorted, the dominated solutions are deleted, and the remaining individuals are the solution set.

[0041] Furthermore, the optimal solution is selected from multiple solutions in the solution set by a fuzzy decision-making method; wherein the satisfaction function of each sub-goal in the solution set is expressed as:

[0042] ;

[0043] In the formula, represents the satisfaction of the objective function, is the sub-objective function value of the Pareto solution x, and are the maximum and minimum values ​​of the sub-objective function values ​​of the Pareto solution set x respectively; ;

[0044] For each Pareto optimal solution set, the satisfaction of its sub-goals is a definite value. In order to better describe it, the satisfaction is fuzzified to select a satisfactory solution that meets the decision maker's preferences from the solution set.

[0045] The triangle membership function is used to fuzzily describe the satisfaction, and the fuzzy rules are defined as: extremely unimportant, unimportant, generally important, relatively important, and extremely important.

[0046] Finally, by selecting an importance for each objective according to the decision maker's needs, the solution that meets the decision maker's preferences can be selected from the optimal solutions.

[0047] Furthermore, in the middle-level model of the scheduling model, maximizing the loading value is taken as the goal, and a loading scheme selection model is established. In the step of determining the loading scheme, maximizing the loading value is expressed as satisfying the maximum loading volume, the urgent order loading volume, and minimizing the number of carrier vehicles used and the number of destinations visited;

[0048] The objective function of the middle-level model is expressed as follows:

[0049] The maximum loading capacity is expressed as: ;

[0050] Maximizing the emergency merchandise truck loading capacity is expressed as: ;

[0051] The maximum loading capacity of large vehicles is expressed as: ;

[0052] Minimizing the use of carrier vehicles is expressed as: ;

[0053] The minimum number of visited destinations is expressed as: ;

[0054] The priority order of the objective functions is: ;

[0055] In the formula, If order i is loaded into carrier vehicle j, then =1; otherwise =0, ; If the urgency of order i is e, then ;otherwise , , ; If the size of order i is r, then ;otherwise ;

[0056] The constraints of the middle-level model are expressed as follows:

[0057] A commodity truck can be loaded into at most one carrier truck, which is expressed as: ;

[0058] For the carrier vehicle being used, the full load constraint is expressed as: ;

[0059] The parking space size constraint is expressed as: ;

[0060] The constraint on the number of visited destinations is expressed as: ;

[0061] When none of the orders loaded on carrier j go to destination t, the carrier does not go to destination t, which is expressed as: ;

[0062] When there is an order for destination t among the orders loaded on carrier j, the carrier must go to destination t, which is expressed as: ;

[0063] When the destination of order i is not t, order i will not go to t no matter which carrier vehicle it is loaded into, which can be expressed as: ,when ;

[0064] When carrier vehicle j does not go to any destination, then carrier vehicle j must not be used, which can be expressed as: ;

[0065] When carrier vehicle j goes to any destination, carrier vehicle j must be used, which is expressed as: ;

[0066] Decision variable constraints, is a 0-1 variable, expressed as:

[0067] .

[0068] Furthermore, in the middle-level model of the scheduling model, order pre-allocation constraints are introduced, where if order i represents a predicted order, then ;otherwise ;

[0069] Add the following constraints to the middle-level model: , which means that when a carrier vehicle is used, the total amount of predicted orders loaded by it shall not exceed 30% of the total parking capacity of the carrier vehicle;

[0070] Add the following to the objective function of the middle-level model: , indicating that the predicted order quantity loading is minimized, and the target priority is set as: , which means that on the basis of maximizing the total loading capacity and the loading capacity of emergency commodity vehicles, the predicted order loading capacity is minimized, and then the optimization of other sub-goals is pursued.

[0071] Furthermore, in the lower model of the scheduling model, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle model and the logistics service capability of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP, where:

[0072] The objective function of the lower model is expressed as follows: , used to represent the maximum global FLSP satisfaction;

[0073] The constraints of the lower model are expressed as follows:

[0074] Each task must be assigned a representation as: , where the task set ;

[0075] The number of model u carriers assigned to FLSPv shall not exceed the total number of model u carriers that the FLSP can provide, expressed as: ;

[0076] Satisfaction is expressed as:

[0077] ;

[0078] When FLSPv is assigned a task Less than the expected lower limit When the satisfaction is equal to , when the assigned task volume is greater than the expected lower limit When the satisfaction is equal to ;

[0079] Decision variables is a 0-1 variable, expressed as: .

[0080] Furthermore, the step of dynamically adjusting the capacity distribution based on the capacity resource satisfaction before the time window for predicting the order flow is reached includes:

[0081] If the capacity resource satisfaction is less than 100%, the capacity weight for dispatching to the region will be increased when allocating orders;

[0082] If the capacity resource satisfaction is greater than 100%, the capacity weight for dispatching to the region will be reduced when allocating orders.

[0083] Another embodiment of the present invention provides a vehicle logistics dynamic scheduling system, including:

[0084] The sales volume forecasting module is used to build a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture. It uses the Transformer layer and the convolutional neural network (CNN) layer to extract features from the influencing factor data, uses the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and outputs the predicted sales volume.

[0085] A transportation route generation module is used to construct an objective function of a multi-constraint model based on transportation cost, transportation time, and carbon emissions using the predicted sales volume, and solve the objective function of the multi-constraint model using a non-dominated sorting algorithm to obtain a recommended transportation route;

[0086] The transport resource scheduling module is used to configure a scheduling model. In the upper model of the scheduling model, the order set to be assigned is assigned downward using the recommended transportation route. In the middle model of the scheduling model, a loading scheme selection model is established to determine the loading scheme with the goal of maximizing the loading value. In the lower model of the scheduling model, tasks are reasonably assigned to specific FLSPs based on the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP.

[0087] The capacity resource allocation module is used to match the scheduling model with the predicted order flow, obtain the capacity resource satisfaction of the predicted order flow, and dynamically adjust the capacity distribution based on the capacity resource satisfaction before the time window of the predicted order flow is reached.

[0088] Compared with the prior art, the method and system for dynamic scheduling of vehicle logistics based on deep learning prediction provided by the embodiment of the present invention have the following beneficial effects:

[0089] First, the present invention analyzes and clarifies the multiple influencing factors of policy, economy, and market, proposes a composite deep learning model architecture based on Transformer-LSTM-CNN through the analysis of multimodal data, quantifies and processes the data and inputs it into the composite model of Transformer-LSTM-CNN, divides the data set, and performs hyperparameter tuning of the model architecture. Finally, the prediction results are obtained, and the sales volume of vehicle logistics is predicted;

[0090] Second, the present invention considers a vehicle logistics order allocation model with hierarchical decision-making. The model divides the order allocation problem into different levels and makes corresponding decisions at each level to maximize the overall transportation efficiency and service quality. In addition, based on the hierarchical decision allocation model, the present invention also introduces the concept of order pre-allocation in the middle-level model of the hierarchical decision model, identifies the orders that can fill the current gap in the predicted next-day orders, and pre-allocates them to the corresponding transportation resources, thereby optimizing the vehicle logistics order allocation model.

[0091] Third, the non-dominated sorting algorithm II (NSGA-II) of the present invention calculates the crowding distance between different individuals, divides the levels, and finds the optimal non-inferior solution, thereby effectively solving the objective function of the multi-constraint model; and the present invention solves the model with the NSGA-II multi-objective evolutionary algorithm with global archive. The NSGA-II algorithm with global archive saves the non-dominated solution during the iteration process and then performs crossover and mutation operations. On the one hand, it can retain good solutions, improve computing efficiency, and reduce complexity; on the other hand, it can overcome the problem of elite solution loss and greatly improve computing power;

[0092] Fourth, the present invention predicts the order demand in the future time window through the prediction model of deep learning, and realizes the pre-scheduling of the capacity resources in the future time window in combination with the scheduling model of the capacity resources, obtains the accurate capacity resource satisfaction, and uses the offset of the capacity scheduling model within the time of reaching this time window to complete the dynamic balance of the capacity resource satisfaction, so that the capacity resources are fully utilized. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0094] Figure 1 It is a schematic diagram of the system architecture of the present invention;

[0095] Figure 2 A schematic diagram of a process flow of a vehicle logistics dynamic scheduling method based on deep learning prediction according to the present invention;

[0096] Figure 3 A schematic diagram of a flow chart of sales volume prediction provided by an embodiment of the present invention;

[0097] Figure 4 A schematic diagram of a flow chart of a non-dominated sorting algorithm provided by an embodiment of the present invention;

[0098] Figure 5 A schematic diagram of a triangle membership function provided by an embodiment of the present invention;

[0099] Figure 6 A structural block diagram of a vehicle logistics dynamic scheduling system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0100] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0101] Finished vehicle logistics refers to the process of distributing vehicles from the OEM to finished vehicle distribution and then to various dealers.

[0102] At present, the logistics scheduling of complete vehicles in the automobile industry has the following problems: First, the sales volume of complete vehicle logistics is predicted without considering the various factors that affect the sales of complete vehicles, which has caused huge challenges in the capacity scheduling of complete vehicle transportation; Second, the existing logistics transportation network in the automobile industry is composed of a large number of scattered "point-to-point" transportation routes. This transportation structure will lead to problems such as low degree of transportation organization, low overall efficiency, and high logistics costs; Third, the current complete vehicle logistics scheduling algorithm in the industry is only a match between the orders and transportation capacity of the current month, and it is still a passive and reactive static scheduling.

[0103] To solve the above problems, the present invention predicts the order demand in the future time window through a deep learning prediction model, and realizes the pre-scheduling of the capacity resources in the future time window in combination with the spatiotemporal scheduling capability of the capacity resources, so as to obtain accurate capacity resource satisfaction, and use the offset of the capacity scheduling model within the time of reaching this time window to complete the dynamic balance of the capacity resource satisfaction and make full use of the capacity resources.

[0104] The following is a detailed description of the specific implementation of a method and system for dynamic scheduling of vehicle logistics based on deep learning prediction in conjunction with specific embodiments of the present invention.

[0105] Please refer to Figure 1-Figure 3 In one embodiment of the present invention, a method for dynamic scheduling of vehicle logistics based on deep learning prediction is provided, comprising the following steps:

[0106] S101: Build a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture, use the Transformer layer and the convolutional neural network (CNN) layer to extract features from the influencing factor data, use the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and output the predicted sales volume;

[0107] Specifically, the influencing factor data mentioned in the above embodiment include economic factors, market factors and policy factors;

[0108] Among them, the policy factors are analyzed and processed by NLP text quantification tools to convert the policy factor text from qualitative description to quantitative indicators: first, the policy text is cleaned, including removing stop words, punctuation marks and numbers to extract key information; then, through word frequency statistics, the keywords that appear frequently in the policy text are identified to reflect the theme and focus of the policy; sentiment analysis tools are used to evaluate the sentiment tendency of the policy text to help understand the overall tendency of the policy; topic modeling techniques, such as Latent Dirichlet Allocation (LDA), are used to extract themes in the policy text; the construction of a keyword co-occurrence network helps to analyze the relationship between keywords; TF-IDF analysis determines the importance of keywords in the text; clustering algorithms are used to classify similar texts into one category to reveal the similarities and differences in policy documents; semantic similarity technology is used to compare the similarities between policy texts; finally, the quantitative indicators obtained through these steps can be used to establish numerical representations of policy features, which will become the input of the prediction model and help to more comprehensively understand the relationship between policy and sales.

[0109] When analyzing the impact of fluctuations in automobile sales on the scheduling of transportation resources, the present invention proposes a prediction model for automobile order flow; among them, in view of the fact that the impact of various factors affecting automobile sales is relatively vague, the present invention proposes and constructs a composite model that integrates Transformer, LSTM and CNN. This model shows significant efficiency when processing complex sequence data, and is particularly suitable for scenarios that need to process long-term and short-term dependencies at the same time.

[0110] Furthermore, in step S101 of the embodiment of the present disclosure, firstly, the multiple influencing factors such as policies, economy, and market are analyzed and clarified, and relevant data of automobiles in the industry are collected respectively, including historical policy changes, macroeconomic indicators, industry policies, and internal enterprise data;

[0111] The present invention proposes a composite deep learning model architecture based on Transformer-LSTM-CNN through the analysis of multimodal data. The data is quantized and processed and then input into the composite model of Transformer-LSTM-CNN. After dividing the data set, the hyperparameters of the model architecture are tuned to finally obtain the prediction results.

[0112] Among them, the Transformer architecture of the embodiment of the present disclosure uses its self-attention mechanism as the core to process long-distance dependencies. This mechanism enables the model to consider the relationship between all elements in the sequence at the same time, thereby effectively capturing global context information; LSTM processes short-term dependencies in time series, effectively avoiding the gradient vanishing problem that occurs when processing long sequences; the present invention combines the Transformer and LSTM models, and the LSTM layer is used to process the initial stage of sequence data to capture subtle dynamics in time;

[0113] Subsequently, the Transformer layer takes over the processing, using its self-attention mechanism to capture a wider range of contextual relationships; this not only improves the model’s ability to handle long-distance dependencies, but also retains sensitivity to short-term dynamics;

[0114] At the same time, the present invention further introduces the convolutional neural network CNN into the model to form a composite structure of Transformer-LSTM-CNN. By adding CNN to the model, the local features in the sequence data can be better processed, thereby further improving the performance of the model, and having significant advantages in processing application scenarios that require accurate capture of long-term and short-term dependency characteristics. In the time series prediction task, the model can take into account both short-term fluctuations and long-term trends in time, thereby improving the accuracy and robustness of the prediction.

[0115] like Figure 3 As shown, the Transformer-LSTM-CNN composite model architecture of the present invention is designed to perform refined time series automobile sales volume forecasting. The Transformer layer and the CNN layer can capture the long-term dependencies and complex pattern relationships in the input data through their self-attention mechanism; the LSTM layer that follows strengthens the model's ability to process sequential information of the time series, and provides the model with the ability to retain past information and grasp time dynamics; the output end passes through a densely connected layer, which maps the internal representation to the continuous output of the predicted sales volume; preferably, the mean square error (MSE) is used as the loss function during the training process of the Transformer-LSTM-CNN composite model of the present invention, and the accuracy of the predicted output results is ensured by iteratively adjusting the network weights.

[0116] Preferably, in the training process of the Transformer-LSTM-CNN composite model of the present invention, the model is first compiled. In this step, the mean square error (MSE) is selected as the loss function to measure the gap between the model's prediction and the actual sales data; the Adam optimizer is used to optimize the parameters of the model. The Adam optimizer has the characteristics of adaptive learning rate, which helps the model converge to the optimal solution faster; then the model is trained for 100 iterations, and the batch size is selected as 1 in each iteration to ensure that the temporal nature of the time series is retained; in addition, in order to improve the performance of the model, the ReduceLROnPlateau callback function is introduced to automatically reduce the learning rate when the verification loss stagnates, which helps the model to better adapt to the data.

[0117] Therefore, the prediction model architecture of the present invention has the ability to deeply understand the complex time series predictions that integrate multiple influencing factors, and is also advanced in processing multi-dimensional input data with complex time dependencies.

[0118] Please continue to refer to Figure 2 The vehicle logistics dynamic scheduling method based on deep learning prediction provided by the embodiment of the present disclosure also includes the following steps:

[0119] S102: constructing an objective function of a multi-constraint model based on transportation cost, transportation time, and carbon emission using the predicted sales volume, and solving the objective function of the multi-constraint model using a non-dominated sorting algorithm to obtain a recommended transportation route;

[0120] Among them, in the step of using the predicted sales volume to construct the objective function of the multi-constraint model of transportation cost, transportation time, and carbon emissions:

[0121] The transportation cost includes the transportation cost between nodes, the cost of converting the transportation mode at the node, the cost of waiting for service at the node, and the penalty cost for exceeding the optimal delivery time; wherein, the transportation cost is expressed as:

[0122] ;

[0123] In the formula, represents the total transportation cost; is a 0 / 1 variable, indicating whether to choose to transport cargo h from node i to node j via transportation mode k. If yes, it is 1, otherwise it is 0; represents the transportation distance from node i to node j using transportation mode k; Indicates the unit transportation distance cost of each mode of transportation; Indicates the volume or weight of the cargo h; represents the time to reach node l; represents the time when the slave node i is triggered; TE represents the optimal delivery time, and a penalty cost is incurred when the timeout occurs; represents the waiting cost coefficient, which reflects the cost of waiting at node i; H represents the cargo set; R represents the node set, V represents the path set, and K represents the transportation mode set; , when it is 1, it means that the cargo h is converted from transportation mode k to l at node i; Cost coefficient for switching from k to l at node i; represents the total transportation time of cargo h, The critical point for optimal delivery time;

[0124] The transportation time includes the transportation time between nodes, the transportation mode conversion time of the intermediate nodes and the time waiting for service at the nodes; wherein, the transportation time is expressed as:

[0125] ;

[0126] In the formula, represents the total transportation time of cargo h; represents the transportation time from node i to node j using transportation mode k; represents the conversion time of cargo h from transport mode k to mode l at node i;

[0127] The carbon emissions include the carbon emissions during transportation and the carbon emissions from the conversion of transportation modes at the node; wherein the carbon emissions are expressed as:

[0128] ;

[0129] In the formula, represents the total carbon emissions; The unit carbon emission coefficient represents the unit emission from node i to node j using the k-th transportation mode; is the carbon emission coefficient of switching transportation modes, which represents the carbon emission when nodes i and l switch transportation modes;

[0130] The objective function of the multi-constraint model is expressed as: ;

[0131] The transport task has a fixed start and end point represented as: ; Indicates whether to choose to use transportation mode k from node i to node i+1; Indicates the starting point node; Indicates the mode of transport; Indicates any one; D indicates the destination of transportation;

[0132] Only one mode of transportation can be selected between nodes, and only one conversion can be made at a node, which can be expressed as: , is a binary decision variable used to indicate whether a mode of transport conversion occurs between node i and node l; h represents the index of the cargo;

[0133] The continuity before and after transportation is expressed as: ; Indicates whether transportation mode k is used from node i-1 to node i;

[0134] The transport capacity between nodes does not exceed the maximum allowed transport capacity, which is expressed as: ; It represents the maximum transport capacity of transport mode k from node i to node j;

[0135] The flow from the starting point to the end point is expressed as: ; Indicates whether transportation mode k is used from node g to node i; Indicates a transit node; , , They represent the destination node, activity node, and origin node respectively;

[0136] The time when the mode of transport conversion is completed at node j is equal to the sum of the departure time of the previous node, the conversion time and the section transportation, which can be expressed as: ;in, It represents the time when the cargo is converted from transport mode k to mode l at node i;

[0137] The time when the conversion is completed is earlier than the departure time: ; It represents the time required for node l to switch the mode of transport;

[0138] The completion time of the previous segment task is the departure time of the next segment, which is expressed as: ; It represents the time it takes for node l to perform task g using transportation mode i to k; It represents the time for node l to carry out task i to j using transportation mode k;

[0139] The total transportation time of cargo h does not exceed the rated time and is expressed as: ,in, represents the total transportation time of cargo h, It represents the rated maximum transportation time of cargo h;

[0140] The decision variable value constraints are expressed as: .

[0141] Further, such as Figure 4As shown, the step of solving the objective function of the multi-constraint model by using the non-dominated sorting algorithm includes: initializing the parameters of the non-dominated sorting algorithm, and the initialized parameters include: the population size is 30, the number of iterations is 200 times, the crossover operator is 0.6, and the mutation operator is 0.01; selecting the population through fast non-dominated sorting, storing the non-dominated solution in the document, and generating the first generation population through crossover and mutation operations; setting the initial value of the evolutionary generation to 3, and obtaining the parent and child populations through evolution; performing fast non-dominated sorting on the parent and child populations, selecting the population, storing the non-dominated solution in the document, calculating the crowding degree of individuals in the population, and selecting the population individuals to enter the next generation to obtain a new parent-child generation population; performing selection, crossover, and mutation operations on the new parent-child generation population; until the maximum evolutionary generation is reached, the individuals in the document are non-dominated sorted, the dominated solutions are deleted, and the remaining individuals are the solution set.

[0142] Furthermore, the optimal solution is selected from multiple solutions in the solution set by fuzzy decision-making; the satisfaction function of each sub-goal in the solution set is expressed as: ;

[0143] In the formula, represents the satisfaction of the objective function, is the sub-objective function value of the Pareto solution x, and are the maximum and minimum values ​​of the sub-objective function values ​​of the Pareto solution set x respectively; ;

[0144] For each Pareto optimal solution set, the satisfaction of its sub-goals is a definite value. In order to better describe it, the satisfaction is fuzzified to select a satisfactory solution that meets the decision maker's preferences from the solution set.

[0145] like Figure 5 As shown, the present invention adopts a triangular membership function to fuzzily describe the satisfaction level, and defines the fuzzy rules as: extremely unimportant, unimportant, generally important, relatively important, and extremely important; finally, an importance is selected for each goal according to the needs of the decision maker, and the solution that meets the decision maker's preferences can be screened out from the optimal solution.

[0146] In the field of vehicle logistics, logistics companies need to effectively allocate vehicle transport orders to different transport vehicles to achieve optimal transportation efficiency and cost control. To solve this problem, we consider a vehicle logistics order allocation model with hierarchical decision-making. This model divides the order allocation problem into different levels and makes corresponding decisions at each level to maximize the overall transportation efficiency and service quality.

[0147] Please continue to refer to Figure 2The vehicle logistics dynamic scheduling method based on deep learning prediction provided by the embodiment of the present disclosure also includes the following steps:

[0148] S103: configuring a scheduling model. In the upper model of the scheduling model, the order set to be assigned is assigned downward using the recommended transportation route. In the middle model of the scheduling model, a loading scheme selection model is established to determine the loading scheme with the goal of maximizing the loading value. In the lower model of the scheduling model, tasks are reasonably assigned to specific FLSPs based on the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP.

[0149] S104: Match the scheduling model with the predicted order flow to obtain the capacity resource satisfaction of the predicted order flow, and dynamically adjust the capacity distribution based on the capacity resource satisfaction before the time window of the predicted order flow is reached, wherein the step of dynamically adjusting the capacity distribution based on the capacity resource satisfaction before the time window of the predicted order flow is reached includes: if the capacity resource satisfaction is less than 100%, increase the capacity weight for scheduling to the region when allocating orders; if the capacity resource satisfaction is greater than 100%, reduce the capacity weight for scheduling to the region when allocating orders.

[0150] Among them, in the upper-level decision-making stage, according to the customer's delivery requirements and the time and economic constraints of the transportation mode, it is necessary to determine the transportation volume under different transportation modes (road, water and rail); in the middle-level decision-making stage, for orders placed under road transportation, LSI (loading scheme design) calculates the order loading combination according to the loading constraints, including maximizing the loading volume, the loading volume of urgent orders, and minimizing the use of carrier vehicles and the number of destinations visited; in the lower-level decision-making stage, according to the order loading combination and the logistics service capabilities of each FLSP (logistics service provider), it is necessary to reasonably allocate tasks to specific FLSPs.

[0151] The optimization goal is to maximize the satisfaction of FLSP while completing the order allocation decisions for all vehicles.

[0152] Furthermore, based on the actual transportation situation, the present invention makes the following assumptions: (1) Order set ; Carrier vehicle collection ; Destination Collection ; One order contains only one vehicle, and each order is independent of each other; (2) Transportation method , k=1 represents waterway, k=2 represents railway, and k=3 represents highway. When three modes of transportation coexist in a certain area, they are ranked by transportation time: waterway, railway, highway; and by transportation cost: highway, railway, waterway. (3) Order urgency , e=1 represents emergency, e=2 represents non-emergency; (4) City orders in the same dispatching area can be carried on the same carrier vehicle, and a single carrier vehicle can visit no more than 2 destinations per round; different models of carrier vehicles have different specifications, and the number of large and small parking spaces is different. According to the backward compatibility rule, large parking spaces can accommodate large and small commercial vehicles; (5) Task set , Highway FLSP Collection , carrier vehicle model collection FLSP has an expected value for the task volume, and the upper limit of the expected value is equal to the maximum service capacity of the FLSP. In view of the fact that the same-city order sharing is more economical and efficient, the present invention introduces a task satisfaction coefficient, which decreases as the number of destinations increases.

[0153] As a preferred embodiment of the present disclosure, in the middle-level model of the scheduling model, the embodiment of the present invention establishes a loading scheme selection model with the goal of maximizing the loading value. In the step of determining the loading scheme, maximizing the loading value is expressed as satisfying the maximum loading volume, the urgent order loading volume, and minimizing the number of carrier vehicles used and the number of destinations visited;

[0154] The objective function of the middle-level model is expressed as follows:

[0155] The maximum loading capacity is expressed as: ;

[0156] Maximizing the emergency merchandise truck loading capacity is expressed as: ;

[0157] The maximum loading capacity of large vehicles is expressed as: ;

[0158] Minimizing the use of carrier vehicles is expressed as: ;

[0159] The minimum number of visited destinations is expressed as: ;

[0160] The priority order of the objective functions is: ;

[0161] In the formula, If order i is loaded into carrier vehicle j, then =1; otherwise =0, ; If the urgency of order i is e, then ;otherwise , , ; If the size of order i is r, then ;otherwise ;

[0162] The constraints of the middle-level model are expressed as follows:

[0163] A commodity vehicle can be loaded into at most one carrier vehicle as: ;

[0164] For the carrier vehicle being used, the full load constraint is expressed as: ;in, represents the total loading capacity of carrier vehicle j; If carrier vehicle j is used, then =1, otherwise = 0, j∈V;

[0165] The parking space size constraint is expressed as: ;

[0166] The constraint on the number of visited destinations is expressed as: ; indicates that the carrier vehicle j will pass through the destination t, then =1; otherwise =0, j∈V, t∈D, V represents the set of carrier vehicles, V={1, 2, ..., n};

[0167] When none of the orders loaded on carrier j go to destination t, the carrier does not go to destination t, which is expressed as: ; If order i is loaded on carrier vehicle j and sent to destination t, then =1, otherwise =0, j∈V, t∈D, i∈I;

[0168] When there is an order for destination t among the orders loaded on carrier j, the carrier must go to destination t, which is expressed as: ;

[0169] When the destination of order i is not t, order i will not go to t no matter which carrier vehicle it is loaded into, which can be expressed as: ,when ; Indicates that order i goes to destination t. If =1, otherwise =0;

[0170] When carrier vehicle j does not go to any destination, then carrier vehicle j must not be used, which can be expressed as: ;

[0171] When carrier vehicle j goes to any destination, carrier vehicle j must be used, which is expressed as: ;

[0172] Decision variable constraints, is a 0-1 variable, expressed as:

[0173] .

[0174] Furthermore, the purpose of adding the forecast of the next day's order demand is to find out whether there are orders in the next day's demand that can fill the current gap based on the identification of the gap in the current order loading plan. If there is an opportunity, the next day's orders can be pre-allocated, that is, the orders that can fill the current gap are identified in the predicted next day's orders and pre-allocated to the corresponding transportation resources. Therefore, in the preferred embodiment provided by the present invention, the order pre-allocation constraint is introduced in the middle-level model of the scheduling model, where if order i represents a predicted order, then ;otherwise ;

[0175] Add the following constraints to the middle-level model: , which means that when a carrier vehicle is used, the total amount of predicted orders loaded by it shall not exceed 30% of the total parking capacity of the carrier vehicle;

[0176] Add the following to the objective function of the middle-level model: , indicating that the predicted order quantity loading is minimized, and the target priority is set as: , which means that on the basis of maximizing the total loading capacity and the loading capacity of emergency commodity vehicles, the predicted order loading capacity is minimized, and then the optimization of other sub-goals is pursued.

[0177] Furthermore, in the lower-level model of the scheduling model provided in the embodiment of the present invention, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle-level model and the logistics service capability of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP, wherein:

[0178] The objective function of the lower model is expressed as follows: , which is used to represent the maximum global FLSP satisfaction; where the highway FLSP set F = {1,2,...,f}, express The importance weight of LSI satisfies >0, , express Satisfaction with the task, satisfying: ;

[0179] The constraints of the lower model are expressed as follows:

[0180] Each task must be assigned a representation as: ; Among them, the task set ;

[0181] The number of model u carriers assigned to FLSPv shall not exceed the total number of model u carriers that the FLSP can provide, expressed as: ; express The upper limit of the available number of carrier vehicles of model u, the carrier vehicle model set TR = {1, 2, ..., tr}, F represents the highway FLSP set;

[0182] Satisfaction is expressed as:

[0183] ;

[0184] in, represents the satisfaction coefficient of task w, which satisfies ;

[0185] When FLSPv is assigned a task Less than the expected lower limit When the satisfaction is equal to , when the assigned task volume is greater than the expected lower limit When the satisfaction is equal to ;

[0186] Decision variables is a 0-1 variable, expressed as: , where w represents the task and V represents the highway FLSP.

[0187] Please refer to Figure 6 In another preferred embodiment of the present invention, a vehicle logistics dynamic scheduling system is also provided, comprising:

[0188] The sales volume forecasting module 201 is used to construct a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture, use the Transformer layer and the convolutional neural network CNN layer to extract features of the influencing factor data, use the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and output the predicted sales volume;

[0189] The transport path generation module 202 is used to construct an objective function of a multi-constraint model based on transport cost, transport time, and carbon emission using the predicted sales volume, and solve the objective function of the multi-constraint model using a non-dominated sorting algorithm to obtain a recommended transport path;

[0190] The transport resource scheduling module 203 is used to configure a scheduling model. In the upper model of the scheduling model, the order set to be allocated is allocated downward by using the recommended transportation route; in the middle model of the scheduling model, a loading scheme selection model is established to determine the loading scheme with the goal of maximizing the loading value; in the lower model of the scheduling model, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP;

[0191] The capacity resource allocation module 204 is used to match the scheduling model with the predicted order flow, obtain the capacity resource satisfaction of the predicted order flow, and dynamically adjust the capacity distribution based on the capacity resource satisfaction before the time window of the predicted order flow is reached.

[0192] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0193] Technicians in the field can clearly understand that for the sake of convenience and conciseness, the specific working process of the system described above can refer to the corresponding process in the aforementioned embodiment of the dynamic scheduling method for whole vehicle logistics based on deep learning prediction, and will not be repeated here.

[0194] Another preferred embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a method for dynamic scheduling of vehicle logistics based on deep learning prediction as described in any one of the above embodiments is implemented.

[0195] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc.; the identification device may include but is not limited to a processor and a memory.

[0196] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0197] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0198] Another preferred embodiment of the present invention provides a storage medium, which is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned embodiments of the dynamic scheduling method for whole vehicle logistics based on deep learning prediction.

[0199] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0200] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A dynamic scheduling method for vehicle logistics based on deep learning prediction, characterized in that: include: Construct a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture, use the Transformer layer and the convolutional neural network CNN layer to extract features of the influencing factor data, use the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and output the predicted sales volume; The objective function of a multi-constraint model based on transportation cost, transportation time, and carbon emission is constructed using the predicted sales volume, and the objective function of the multi-constraint model is solved using a non-dominated sorting algorithm to obtain a recommended transportation route; A scheduling model is configured, and in an upper model of the scheduling model, the order set to be allocated is allocated downwardly by using the recommended transportation route; In the middle-level model of the scheduling model, with the goal of maximizing the loading value, a loading scheme selection model is established to determine the loading scheme; In the lower model of the scheduling model, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle model and the logistics service capabilities of each FLSP, with the goal of maximizing the satisfaction of the FLSP, and a vehicle FLSP selection model is constructed; The scheduling model is matched with the predicted order flow to obtain the capacity resource satisfaction of the predicted order flow. Before the time window of the predicted order flow is reached, the capacity distribution is dynamically adjusted based on the capacity resource satisfaction. In the middle-level model of the scheduling model, a loading scheme selection model is established with the goal of maximizing the loading value. In the step of determining the loading scheme, maximizing the loading value is expressed as satisfying the maximum loading volume, the urgent order loading volume, and minimizing the number of carrier vehicles used and the number of destinations visited; wherein: The objective function of the middle-level model is expressed as follows: The maximum loading capacity is expressed as: ; Maximizing the emergency merchandise truck loading capacity is expressed as: ; The maximum loading capacity of large vehicles is expressed as: ; Minimizing the use of carrier vehicles is expressed as: ; The minimum number of visited destinations is expressed as: ; The priority of the objective functions is: Z3>Z4>Z5>Z6>Z7; In the formula, u ij If order i is loaded into carrier vehicle j, then u ij =1; otherwise u ij =0, i∈I, j∈V; If the urgency of order i is e, then ;otherwise , i∈I, e∈{1,2}; If the size of order i is r, then ;otherwise , i∈I, r∈{1,2}; The constraints of the middle-level model are expressed as follows: A commodity truck can be loaded into at most one carrier truck, which is expressed as: ; For the carrier vehicle being used, the full load constraint is expressed as: Among them, C j A represents the total loading capacity of carrier vehicle j; j If carrier vehicle j is used, then A j =1, otherwise A j = 0, j∈V; The parking space size constraint is expressed as: ; The constraint on the number of visited destinations is expressed as: ;y jt indicates that the carrier vehicle j will pass through the destination t, then y jt =1; otherwise y jt =0, j∈V, t∈D, V represents the set of carrier vehicles, V={1, 2, ..., n}; When none of the orders loaded on carrier j go to destination t, the carrier does not go to destination t, which is expressed as: ; b ijt If order i is loaded on carrier vehicle j and sent to destination t, then b ijt =1, otherwise b ijt =0, j∈V, t∈D, i∈I; When there is an order for destination t among the orders loaded on carrier j, the carrier must go to destination t, which is expressed as: ; When the destination of order i is not t, order i will not go to t no matter which carrier vehicle it is loaded into, which is expressed as: b ijt =0, when =0, , j∈V, t∈D; Indicates that order i goes to destination t. If =1, otherwise =0; When carrier vehicle j does not go to any destination, then carrier vehicle j must not be used, which can be expressed as: ； When carrier vehicle j goes to any destination, carrier vehicle j must be used, which is expressed as: ； Decision variable constraints, u ij ,y jt 、b ijt , A j is a 0-1 variable, expressed as: ； In the middle-level model of the scheduling model, order pre-allocation constraints are introduced, where if order i represents a predicted order, then fur i =1; otherwise i =0, i∈I; Add the following constraints to the middle-level model: , which means that when a carrier vehicle is used, the total amount of predicted orders loaded by it shall not exceed 30% of the total parking capacity of the carrier vehicle; Add the following to the objective function of the middle-level model: , indicating the minimization of the predicted order quantity loading; The target priority is set as: Z3>Z4>Z9>Z5>Z6>Z7, which means that the predicted order loading is minimized on the basis of maximizing the total loading volume and the loading volume of the emergency commodity vehicle; In the lower model of the scheduling model, tasks are reasonably allocated to specific FLSPs according to the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP, where: The objective function of the lower model is expressed as follows: , which is used to represent the maximization of global FLSP satisfaction, where the highway FLSP set F = {1,2,...,f}, β v Indicates FLSP v The importance weight of LSI satisfies β v >0, , S v Indicates FLSP v Satisfaction with the task, satisfied: S v ≥0; The constraints of the lower model are expressed as follows: Each task must be assigned a representation as: , where the task set H = {1, 2, ..., h}; Allocation to FLSP v The number of Model U carriers shall not exceed the total number of Model U carriers that the FLSP can provide, expressed as: , ,v∈F; Indicates FLSP v The upper limit of the available number of carrier vehicles of model u, the carrier vehicle model set TR = {1, 2, ..., tr}, F represents the highway FLSP set; Satisfaction is expressed as: ； Among them, δ w represents the satisfaction coefficient of task w, which satisfies ; When FLSP v The amount of tasks assigned Less than the expected lower limit When the satisfaction is equal to ; When FLSP v The amount of assigned tasks is greater than the expected lower limit When the satisfaction is equal to ; Decision variable z wv is a 0-1 variable, expressed as: , where w represents the task and V represents the highway FLSP.

2. According to claim 1, a dynamic scheduling method for vehicle logistics based on deep learning prediction is characterized in that: The influencing factor data include economic factors, market factors and policy factors; the policy factors are analyzed and processed using nlp text quantification tools to convert the policy factor text from qualitative descriptions into quantitative indicators.

3. According to claim 2, a method for dynamic scheduling of vehicle logistics based on deep learning prediction is characterized in that: In the step of using the predicted sales volume to construct the objective function of the multi-constraint model with transportation cost, transportation time, and carbon emissions: The transportation cost includes the transportation cost between nodes, the cost of converting the transportation mode at the node, the cost of waiting for service at the node, and the penalty cost for exceeding the optimal delivery time; wherein, the transportation cost is expressed as: ； In the formula, Z c represents the total transportation cost; is a 0 / 1 variable, indicating whether to choose to transport cargo h from node i to node j via transportation mode k. If yes, it is 1, otherwise it is 0; S ijk represents the transportation distance from node i to node j using transportation mode k; cl k represents the unit transportation distance cost of each mode of transportation; q h Indicates the volume or weight of the cargo h; tl ijl Indicates the time to reach node l; t3 ikl represents the departure time from node i; TE represents the optimal delivery time, and overtime generates penalty cost; μ i represents the waiting cost coefficient, which reflects the cost of waiting at node i; H represents the cargo set; R represents the node set, V represents the path set, and K represents the transportation mode set; , when it is 1, it means that the cargo h is converted from transportation mode k to l at node i; Cost coefficient for switching from k to l at node i; represents the total transportation time of cargo h, E h The critical point for optimal delivery time; The transportation time includes the transportation time between nodes, the transportation mode conversion time of the intermediate nodes and the time waiting for service at the nodes; wherein, the transportation time is expressed as: ； In the formula, represents the total transportation time of cargo h; represents the transportation time from node i to node j using transportation mode k; represents the conversion time of cargo h from transport mode k to mode l at node i; The carbon emissions include the carbon emissions during transportation and the carbon emissions from the conversion of transportation modes at the node; wherein the carbon emissions are expressed as: ； In the formula, Z t Indicates total carbon emissions; el ijk The unit carbon emission coefficient represents the unit emission from node i to node j using the k-th transportation mode; is the carbon emission coefficient of switching transportation modes, which represents the carbon emission when nodes i and l switch transportation modes; The objective function of the multi-constraint model is expressed as: ; The transport task has a fixed start and end point represented as: ; Indicates whether to choose to use transportation mode k from node i to node i+1; i∈O represents the starting node; k∈K represents the transportation mode; represents any one; D represents the transport destination; Only one mode of transportation can be selected between nodes, and only one conversion can be made at a node, which can be expressed as: ;in, is a binary decision variable used to indicate whether a mode of transport conversion occurs between node i and node l; h represents the index of the cargo; The continuity before and after transportation is expressed as: ; Indicates whether transportation mode k is used from node i-1 to node i; The transport capacity between nodes does not exceed the maximum allowed transport capacity, which is expressed as: ;Q ijk It represents the maximum transport capacity of transport mode k from node i to node j; The flow from the starting point to the end point is expressed as: ; Indicates whether transportation mode k is used from node g to node i; g∈R represents the transit node; i∈D, i∈A, i∈O represent the destination node, activity node, and starting node respectively; The time when the mode of transport conversion is completed at node j is equal to the sum of the departure time of the previous node, the conversion time and the section transportation, which can be expressed as: ; Among them, t3 jkl It represents the time when the cargo is converted from transport mode k to mode l at node i; The time when the conversion is completed is earlier than the departure time: tl ijl -t3 ikl ≥0; where tl ijl It represents the time required for node l to switch the mode of transport; The completion time of the previous segment task is the departure time of the next segment, which is expressed as: tl1 gik =tl ijk , where tl1 gik represents the time it takes for node l to perform task g using transportation mode i to k; tl ijk It represents the time for node l to carry out task i to j using transportation mode k; The total transportation time of cargo h does not exceed the rated time and is expressed as: ,in, represents the total transportation time of cargo h, F h It represents the rated maximum transportation time of cargo h; The decision variable value constraints are expressed as: .

4. A method for dynamic scheduling of vehicle logistics based on deep learning prediction according to claim 3, characterized in that: The step of solving the objective function of the multi-constraint condition model by using the non-dominated sorting algorithm includes: initializing the parameters of the non-dominated sorting algorithm, wherein the initialized parameters include: a population size of 30, an iteration number of 200 times, a crossover operator of 0.6, and a mutation operator of 0.01; selecting the population through fast non-dominated sorting, storing the non-dominated solution in a document, and generating the first generation population through crossover and mutation operations; setting the initial value of the evolutionary generation to 3, and obtaining the parent and child populations through evolution; performing fast non-dominated sorting on the parent and child populations, selecting the population, storing the non-dominated solution in a document, calculating the crowding degree of individuals in the population, and selecting the population individuals to enter the next generation to obtain a new parent-child generation population; performing selection, crossover, and mutation operations on the new parent-child generation population; until the maximum evolutionary generation is reached, performing non-dominated sorting on the individuals in the document, deleting the dominated solution, and the remaining individuals are the solution set.

5. A method for dynamic scheduling of vehicle logistics based on deep learning prediction according to claim 4, characterized in that: The optimal solution is selected from multiple solutions in the solution set by fuzzy decision-making; wherein the satisfaction function of each sub-goal in the solution set is expressed as: ; In the formula, ω i (x) represents the satisfaction of the objective function, f i (x) is the sub-objective function value of the Pareto solution set x, f imax and f imin are the maximum and minimum values ​​of the sub-objective function values ​​of the Pareto solution set x; ω i (x)∈[0,1]; For each Pareto optimal solution set, the satisfaction of its sub-goal is a certain value, the satisfaction is fuzzified, and a satisfactory solution that meets the decision maker's preference is selected from the solution set; The triangle membership function is used to fuzzily describe the satisfaction, and the fuzzy rules are defined as: extremely unimportant, unimportant, generally important, relatively important, and extremely important. An importance degree is selected for each goal according to the needs of the decision maker, and the solution that meets the decision maker's preferences is selected from the optimal solution.

6. A method for dynamic scheduling of vehicle logistics based on deep learning prediction according to claim 5, characterized in that: The step of dynamically adjusting the capacity distribution based on the capacity resource satisfaction before the time window for predicting the order flow is reached includes: If the capacity resource satisfaction is less than 100%, the capacity weight for dispatching to the region will be increased when allocating orders; If the capacity resource satisfaction is greater than 100%, the capacity weight for dispatching to the region will be reduced when allocating orders.

7. A vehicle logistics dynamic scheduling system that implements a vehicle logistics dynamic scheduling method based on deep learning prediction as described in any one of claims 1 to 6, characterized in that: The vehicle logistics dynamic scheduling system includes: The sales volume forecasting module is used to build a demand forecasting model based on the Transformer-LSTM-CNN composite model architecture. It uses the Transformer layer and the convolutional neural network (CNN) layer to extract features from the influencing factor data, uses the time series features and the extracted features of the influencing factor data as the input of the LSTM layer, and outputs the predicted sales volume. A transportation route generation module is used to construct an objective function of a multi-constraint model based on transportation cost, transportation time, and carbon emissions using the predicted sales volume, and solve the objective function of the multi-constraint model using a non-dominated sorting algorithm to obtain a recommended transportation route; The transport resource scheduling module is used to configure a scheduling model. In the upper model of the scheduling model, the order set to be assigned is assigned downward using the recommended transportation route. In the middle model of the scheduling model, a loading scheme selection model is established to determine the loading scheme with the goal of maximizing the loading value. In the lower model of the scheduling model, tasks are reasonably assigned to specific FLSPs based on the loading scheme of the middle model and the logistics service capabilities of each FLSP, and a vehicle FLSP selection model is constructed with the goal of maximizing the satisfaction of the FLSP. The capacity resource allocation module is used to match the scheduling model with the predicted order flow, obtain the capacity resource satisfaction of the predicted order flow, and dynamically adjust the capacity distribution based on the capacity resource satisfaction before the time window of the predicted order flow is reached.

Images