A food delivery vehicle loading and route optimization method and system

By constructing equivalent particles and Markov random field models, combining historical and real-time data, dynamically adjusting the loading and route of food delivery vehicles, the problems of delivery delay and resource waste in the existing technology are solved, and efficient and timely Food supply and freshness guarantee.

CN118798766BInactive Publication Date: 2024-11-22UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411260731.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing food distribution technology is difficult to adjust the delivery plan in real time to deal with traffic congestion and changes in demand, resulting in delays in delivery and waste of resources, and failing to make full use of historical and real-time data for comprehensive analysis, resulting in the inability to optimize the delivery plan, affecting the timely supply of food and Freshness.

Method used

By obtaining historical food usage data, real-time inventory status, map data and traffic data, we build equivalent particle models and Markov random airport models, perform multi-objective optimization processing, dynamically adjust loading plans and routes, and use numerical simulation and statistical models to achieve numerical simulation and statistical models Optimize delivery paths and schedules.

Benefits of technology

It improves distribution efficiency, reduces fuel consumption and delays, flexibly responds to urgent needs, and ensures timely supply and freshness of ingredients.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118798766B_ABST
    Figure CN118798766B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for optimizing the loading and route of food delivery vehicles, and relates to the field of intelligent logistics technology. It includes: obtaining first information, second information and third information, the first information includes historical food usage data and real-time inventory status of each school in the area to be delivered, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area; performing prediction processing according to the first information to obtain fourth information; constructing an equivalent particle model according to the fourth information, the third information and the second information, and simulating the particle model to obtain fifth information; performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information; constructing a dynamic adjustment process according to the sixth information, real-time traffic data and real-time delivery status, and updating subsequent delivery plans. The present invention constructs an equivalent particle model, regards the delivery vehicle as a particle, optimizes the delivery path and schedule, and improves the delivery efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent logistics technology, and in particular to a method and system for optimizing the loading and route of food delivery vehicles. Background Art

[0002] In the field of food distribution, especially in distribution services for educational institutions such as schools, it is crucial to accurately and efficiently manage loading and delivery routes. Current technologies usually rely on basic route planning algorithms and simple loading strategies, such as using map software for route planning and experience-based loading decisions. Although these traditional methods can achieve delivery tasks to a certain extent, they are often inefficient due to the lack of dynamic response to real-time traffic conditions and delivery status. For example, existing technologies may not be able to adjust delivery plans in real time to cope with sudden traffic congestion or accidents, resulting in delivery delays and waste of resources. In addition, traditional distribution methods often do not make full use of historical data and real-time data for comprehensive analysis, making it impossible to predict future changes in demand, resulting in the inability to optimize the delivery plan, affecting the timely supply and freshness of food.

[0003] Based on the shortcomings of the above-mentioned existing methods, there is an urgent need for a method and system for optimizing the loading and route of food delivery vehicles. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for optimizing the loading and routing of food delivery vehicles to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a method for optimizing the loading and routing of food delivery vehicles, comprising:

[0006] Acquire first information, second information and third information, wherein the first information includes historical food usage data and real-time inventory status of schools in the delivery area, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area;

[0007] Performing prediction processing on the first information to obtain fourth information, wherein the fourth information is food demand information and urgency of each school;

[0008] constructing an equivalent particle model according to the fourth information, the third information and the second information, and performing simulation processing on the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading solution set of the delivery vehicle and a corresponding preliminary route set;

[0009] Performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, wherein the sixth information is an optimized loading scheme set and a corresponding route set;

[0010] A Markov random field model is constructed according to the sixth information, real-time traffic data and real-time delivery status, and unfinished delivery tasks are dynamically adjusted based on the Markov random field model, and subsequent delivery plans are updated.

[0011] In a second aspect, the present application also provides a food delivery vehicle loading and route optimization system, including:

[0012] an acquisition module, for acquiring first information, second information and third information, wherein the first information includes historical food usage data and real-time inventory status of schools in the area to be delivered, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area;

[0013] A prediction module, configured to perform prediction processing based on the first information to obtain fourth information, wherein the fourth information is food demand information and urgency of each school;

[0014] a construction module, configured to construct an equivalent particle model according to the fourth information, the third information and the second information, and simulate the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading scheme set of the delivery vehicle and a corresponding preliminary route set;

[0015] an optimization module, configured to perform multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, wherein the sixth information is an optimized loading scheme set and a corresponding route set;

[0016] The adjustment module is used to construct a Markov random field model according to the sixth information, real-time traffic data and real-time delivery status, dynamically adjust the unfinished delivery tasks based on the Markov random field model, and update the subsequent delivery plan.

[0017] The beneficial effects of the present invention are:

[0018] The present invention constructs an equivalent particle model, regards the delivery vehicles as particles, uses historical demand data, real-time inventory information, map data and vehicle parameters to establish dynamic equations, and uses numerical simulation methods to calculate the motion trajectory of particles. It can optimize the delivery path and schedule, improve delivery efficiency, reduce fuel consumption and delivery delays, and flexibly respond to emergency needs, ensuring that high-priority delivery tasks are met in a timely manner, and ensuring the timely supply and freshness of food. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A schematic diagram of a method for optimizing the loading and routing of food delivery vehicles according to an embodiment of the present invention;

[0021] Figure 2 Schematic diagram of the structure of the food delivery vehicle loading and route optimization system described in an embodiment of the present invention.

[0022] Markings in the figure: 1. Acquisition module; 2. Prediction module; 21. First analysis unit; 22. First prediction unit; 23. First evaluation unit; 3. Construction module; 31. First processing unit; 32. First construction unit; 33. First simulation unit; 34. Second processing unit; 4. Optimization module; 41. First conversion unit; 42. Third processing unit; 43. Fourth processing unit; 44. First optimization unit; 45. Second evaluation unit; 5. Adjustment module; 51. Second construction unit; 52. First update unit; 53. First adjustment unit. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0025] Embodiment 1:

[0026] This embodiment provides a method for optimizing the loading and routing of food delivery vehicles. Figure 1 As shown, the method includes steps S100 to S500.

[0027] Step S100, obtaining first information, second information and third information, wherein the first information includes historical food usage data and real-time inventory status of schools in the delivery area, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area;

[0028] It is understandable that the first information involves the historical food usage data and real-time inventory of each school in the distribution area. These data are obtained by recording and counting the food usage of each school in the past period of time, using database query or data acquisition system, and combined with real-time inventory, real-time monitoring through IoT devices or inventory management systems to ensure the accuracy and timeliness of the data. The second information includes the parameters of the distribution vehicle, which not only covers the basic characteristics of the vehicle such as load capacity, volume, fuel efficiency, speed, but also includes detailed information such as the maintenance status of the vehicle and the driver's driving habits. These data are obtained through the vehicle management system or directly from the vehicle's sensor equipment. The third information includes map data and historical traffic data of the distribution area. These data are obtained through the geographic information system (GIS) and traffic monitoring system. The map data provides detailed geographic location information and road network structure, and the historical traffic data covers the flow, congestion and accident records of different roads in different time periods. By analyzing these data, we can understand the traffic patterns and potential problems in the distribution area. When obtaining these data, it is necessary to use big data processing and data fusion technology to standardize and uniformly process data from different sources to ensure their availability and accuracy in subsequent analysis.

[0029] Step S200: performing prediction processing based on the first information to obtain fourth information, where the fourth information is food demand information and urgency of each school;

[0030] Understandably, historical food usage data is processed through a time series analysis model to identify and capture trends and cyclical changes in the data. In this way, the food demand trend for each school over a period of time in the future can be predicted. The processing of these data usually includes stationarity tests, model parameter estimation, and demand forecasting based on historical data. Next, real-time inventory status provides instant information on current inventory levels. By combining historical demand trends and real-time inventory data, future food demand can be predicted more accurately. When assessing the urgency of demand for each school, three key factors need to be considered: demand, inventory levels, and the shelf life of food. Schools with close shelf life and high demand will be considered more urgent.

[0031] Step S300: construct an equivalent particle model according to the fourth information, the third information and the second information, and simulate the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading solution set of the delivery vehicle and a corresponding preliminary route set;

[0032] It is understandable that the particle model is a physical simulation method, which is often used to describe the motion state of particles in a force field. In the present invention, the delivery vehicles are equivalent to particles. By constructing a particle model, it is possible to simulate how the vehicles respond to the attraction of different demand points (schools) during the delivery process. The basic idea of ​​the equivalent particle model is to use the dynamic principles in physics to regard each delivery vehicle as a particle with mass and velocity, and each school exerts gravity on these vehicles according to its demand and urgency. In actual operation, the delivery vehicles are similar to the movement of physical particles in a gravitational field. The demand and urgency of the school's food can be regarded as a source of gravity, which attracts the delivery vehicles and prompts the vehicles to move to schools with high demand and urgency. Geographical distance and road conditions act as resistance, affecting the movement trajectory of the vehicle. Through this analogy, mature methods and algorithms in physical models can be used to accurately simulate and optimize the delivery process.

[0033] Step S400: performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, where the sixth information is an optimized loading scheme set and a corresponding route set;

[0034] It can be understood that the multi-objective optimization process aims to consider multiple optimization objectives at the same time, such as minimizing the delivery cost, minimizing the total driving time and maximizing customer satisfaction, so as to improve the overall delivery efficiency.

[0035] Step S500: construct a Markov random field model according to the sixth information, the real-time traffic data and the real-time delivery status, dynamically adjust the unfinished delivery tasks based on the Markov random field model, and update the subsequent delivery plan.

[0036] It can be understood that the Markov random field model is a statistical model used to represent the interdependence between multiple variables and is suitable for describing systems with local dependencies. The Markov random field model represents the conditional dependency between states through the structure of nodes and edges. Each node represents a random variable (in this embodiment, the delivery state), and the edge represents the dependency between these variables (such as path selection, traffic conditions, etc.). After a vehicle in the distribution system completes a delivery task, its next delivery path and loading state will be affected by the current state and the surrounding environment (such as traffic conditions, real-time demand). This local dependency is the strength of the Markov random field model, which can be clearly represented and calculated through the nodes and edges in the model. Therefore, the Markov random field model can comprehensively consider multiple factors, dynamically adjust the loading and routing of vehicles, maximize the use of existing resources, reduce empty loads and repeated paths, and reduce operating costs.

[0037] Further, step S200 includes step S210 to step S230.

[0038] Step S210: Performing time series analysis on historical food usage data based on an autoregressive mean model to obtain demand trend results;

[0039] Understandably, the reason for choosing the autoregressive mean model is its time series processing capabilities, especially for historical data with obvious trends and periodicity. For food usage data, there are usually periodic fluctuations (such as daily and weekly changes in usage) and long-term trends (such as seasonal demand changes). The autoregressive mean model can fully capture these characteristics through its three components: the autoregressive part, which considers the impact of data from several past moments on the current moment and identifies the internal dependencies of the data; the difference part, which is used to remove the non-stationarity of the data and make the time series more stable; the moving average part, which considers the impact of past prediction errors on the current moment and smoothes the random fluctuations of the data.

[0040] Specifically, step S210 includes step S211 to step S214.

[0041] Step S211, preprocessing the historical food usage data to obtain preprocessed data;

[0042] First, preprocess the historical food usage data. In the food delivery scenario, the accuracy and completeness of the data are crucial. Common problems include data outliers and missing data. Outliers may come from sudden large orders or mis-entered data, while missing data may be caused by recording errors or system failures. To solve these problems, statistical methods are used to remove outliers, such as using box plots to identify and remove extreme values. At the same time, interpolation or mean filling methods are used to handle missing data to ensure data integrity. Next, perform a data stationarity test, preferably using a unit root test (such as an ADF test) to test the stationarity of the data. If the data is not stationary, differential processing is required to stabilize the mean and variance of the data.

[0043] Step S212, performing parameter estimation based on the preprocessed data to obtain a parameter combination;

[0044] The autoregressive average model consists of three parts: autoregression (AR), difference (I) and moving average (MA), and the corresponding parameters are p, d, and q. These parameters represent different time series characteristics in historical food usage data and can help capture trends and patterns in food demand.

[0045] Specifically, the autoregressive (AR) parameter p represents the use of data from the past p moments to predict the demand at the current moment. In historical food usage data, the current food demand is not only related to the most recent data point, but also affected by earlier data points. For example, a larger p value indicates that the demand has a longer memory, reflecting the impact of demand patterns over a longer period of time in the past on the current demand. The differencing (I) parameter d indicates that d differencing operations are required to make the time series data stationary. In food usage data, stationarity means that the statistical properties of the data, such as the mean and variance, do not change over time. The differencing operation helps eliminate trends and seasonality by subtracting the data from the previous moment, resulting in a more stable data series that is more suitable for modeling. The number of differencing operations d usually reflects the long-term trend changes in the data. The moving average (MA) parameter q represents the use of the forecast errors of the past q moments to correct the current forecast. In historical food usage data, this means that the current demand is not only affected by the actual demand in the past, but also by the forecast errors in the past. A larger q value indicates a more complex error correction mechanism to capture short-term fluctuations and irregular changes in the data.

[0046] This method ensures that the selected autoregressive mean model parameters p, d, and q can fully reflect the time series characteristics of historical food usage data, thereby improving the model's predictive ability and reliability. The accuracy of parameter estimation directly affects the model's fitting effect and the accuracy of subsequent demand trend analysis.

[0047] Step S213: Use the determined parameter combination to perform model fitting on the historical food usage data to construct a trend analysis model;

[0048] It should be noted that after determining the parameters of the autoregressive mean model, these parameters are used to fit the model to the historical food usage data. The model fitting process optimizes the model parameters by minimizing the sum of squares of the residuals so that it can best fit the fluctuations and changes in the historical data. In this embodiment, the purpose of the fitting model is to accurately capture the trends and cyclical changes in the data and reflect the actual situation of food demand. The specific process includes: using the determined p, d, and q parameters to establish a trend analysis model, using historical food usage data for fitting, and adjusting the model parameters by minimizing the sum of squares of the prediction residuals so that the model can capture the characteristics of the historical data to the greatest extent. The result of model fitting is a trend analysis model that can reflect the trends and fluctuations of historical data.

[0049] Step S214: Generate demand trend results based on the trend analysis model.

[0050] It is understandable that through model fitting, the trend analysis model generates trend results for food demand, showing the long-term trend and cyclical changes in food use. The demand trend results include changes in demand over a period of time in the future, but do not specifically predict future demand. This result provides the overall direction and pattern of food use, helps understand the changing laws of food demand, and provides important basic data for subsequent demand forecasting and loading optimization.

[0051] Step S220: Based on the demand trend results and the real-time inventory status, a support vector machine algorithm is used to predict the immediate demand for each ingredient to obtain a prediction result, wherein the prediction result includes the ingredient demand information within a preset time period in the future;

[0052] It should be noted that the support vector machine is a supervised learning model suitable for classification and regression tasks. By constructing an optimal hyperplane to classify data points, it can handle nonlinear relationships and map data to high-dimensional space through kernel functions, thus achieving good results on complex data sets. Specifically, the demand trend results and real-time inventory data are integrated to form the input feature set of the prediction model. The specific features include historical demand trends, current inventory levels, shelf life, etc. Since the input features may have different dimensions and ranges, the features are standardized to have similar scales. Then the model is trained using historical data. During the training process, the model optimizes the hyperplane so that different demand states in the training data are best separated. For regression problems, the support vector machine minimizes the prediction error by optimizing the error function. Finally, the real-time inventory situation and current features are input into the trained support vector machine model for demand prediction. The prediction results output by the model include the food demand information in the future preset time period, reflecting the demand changes under different inventory levels and trends.

[0053] Step S230: perform evaluation processing based on the prediction results, and use a decision tree classification model to calculate the urgency of each demand based on the demand for food, inventory level, and proximity to the expiration date.

[0054] Understandably, the demand reflects the demand for each ingredient in each school in the future. High demand usually means that the ingredient will be in high demand in the future and needs to be delivered first. Real-time inventory data provides details of the current inventory. Low inventory levels indicate that the existing inventory of the ingredient may not be sufficient to meet the upcoming demand and needs to be given priority for replenishment. The closeness to the expiration date refers to the urgency of the expiration date of the ingredient. Ingredients that are about to expire need to be delivered first to avoid waste and loss.

[0055] Specifically, step S230 includes step S231 to step S233.

[0056] Step S231, integrating food usage demand, inventory level and shelf life proximity to form an input feature set;

[0057] It should be noted that this step first obtains future demand data from the demand forecasting system, obtains real-time inventory level data from the inventory management system, and obtains food shelf life information from the shelf life record. These data are cleaned and standardized to ensure consistency and integrity. The real-time inventory level is combined with the future demand data to generate a feature that represents the number of ingredients that need to be replenished in the future. At the same time, the proximity of the shelf life of each ingredient is calculated and quantified into a feature to represent the urgency of the ingredient approaching its shelf life. Finally, these processed features are integrated into a new feature set, specifically including: future demand, current inventory level, quantity to be replenished, and shelf life urgency. These features fully reflect the immediate demand and urgency of each school for food, providing an accurate and reliable data basis for the decision tree classification model.

[0058] Step S232: Use historical data to train a decision tree model. During the training process, the model continuously segments the data and selects the optimal segmentation point based on the feature value to generate a decision tree structure, wherein the leaf nodes of the decision tree represent different urgency classification labels;

[0059] First, the feature set and the corresponding urgency labels (such as high, medium, and low) are extracted from the historical records. Then the CART decision tree model is initialized and the historical data is randomly divided into a training set and a validation set. During the training process, the CART algorithm recursively splits the data set and selects the optimal split point for splitting. Specifically, for each feature, the Gini coefficient at different split points is calculated, and the split point with the lowest Gini coefficient is selected as the optimal split point to maximize the subset purity. The model splits at the optimal split point, generates new child nodes, and repeats this process until the stopping condition (such as the maximum tree depth or the minimum number of leaf node samples) is met. The leaf nodes of the generated decision tree structure represent the classification results and are marked with different urgency levels by the labels in the historical data. Subsequently, the validation set is used to evaluate the classification accuracy of the model, and the decision tree is optimized by pruning technology to remove nodes that contribute less to the prediction and prevent overfitting. Finally, the decision tree model trained by the CART algorithm can accurately and intuitively evaluate the urgency of food demand, providing a reliable basis for optimizing distribution decisions.

[0060] Step S233: input the integrated feature set into the trained decision tree model to calculate the urgency of each demand.

[0061] Further, step S300 includes step S310 to step S340.

[0062] Step S310, performing model parameter initialization processing according to the fourth information, the third information and the second information to obtain an initial parameter set, wherein the initial parameter set includes school gravity parameters, dynamic properties of the vehicle and an initial motion state;

[0063] It can be understood that this step is equivalent to specific parameters in the particle model by integrating demand, urgency, geographic location, traffic data and vehicle parameters. These parameters have specific physical meanings in the particle model and are suitable for simulating the dynamic behavior of delivery vehicles.

[0064] First, based on the demand and urgency in the fourth information, the demand of each school is equivalent to the gravity parameter in the particle model. Schools with large demand and high urgency have greater gravity in the particle model, attracting delivery vehicles to go there first. This conversion generates a comprehensive gravity value by weighted summing the demand and urgency, so that these schools have a strong attraction to vehicles in the simulation, similar to the gravitational effect in physics.

[0065] Secondly, the geographical locations of schools and distribution centers are determined based on the map data and historical traffic data in the third information, and these locations are converted into spatial coordinates in the particle model. Traffic data is used to set the travel time and congestion probability of the road as a resistance factor in the particle model. Through the geographic information system (GIS), the map data and traffic data are integrated to generate the movement resistance of vehicles on different roads. These resistances affect the vehicle's travel speed and path selection, similar to the friction of particles when moving in a medium.

[0066] Finally, the distribution vehicle parameters in the second information are integrated, including load capacity, volume, speed, acceleration and fuel efficiency. The load capacity and volume limit of the vehicle are equivalent to the mass and volume parameters in the particle model, and the speed and acceleration parameters reflect the dynamic properties of the vehicle. The specific processing methods include: inputting the physical parameters of the vehicle into the simulation system, setting the initial position of the vehicle to the distribution center, the initial speed to zero, and setting the initial load according to the current loading situation. These initial states are similar to the initial position and initial velocity in physics, providing a starting point for the simulation of the particle model.

[0067] Through these conversion methods, the initial parameter set generated includes the school's gravitational parameters, the vehicle's dynamic properties and the initial motion state. These parameters reflect the attraction of each school to the vehicle, the resistance of the road and the physical characteristics of the vehicle in the particle model, providing accurate and reliable basic conditions for the simulation of the equivalent particle model.

[0068] Step S320: constructing a dynamic equation according to the initial parameter set to obtain a group of dynamic equations for describing the force state of the delivery vehicle in the force field;

[0069] In this embodiment, the gravity parameters are converted into a path-dependent gravity model, taking into account the geographical location of the school, the road network, and the maximum speed and loading capacity of the vehicle. The gravity parameters of each school will be determined based on the demand and urgency of the school. Considering the map path, the distance is replaced by the path length, and the complexity factor of the map path is introduced to reflect the impact of road conditions on gravity. In addition, the maximum speed and loading capacity of the vehicle will also affect the calculation of gravity. The gravity equation is:

[0070] ;

[0071] in, Indicates the serial number of the school; Indicates the serial number of the delivery vehicle; Indicates school The gravitational vector of represents the gravitational constant; Indicates the vehicle's loading capacity; Indicates school gravitational parameters; Indicates vehicle To school The path length; Indicates the road complexity factor, which reflects the road conditions (such as slope, road conditions, etc.), Indicates the maximum speed of the vehicle; Indicates vehicle To school The unit direction vector of .

[0072] Furthermore, resistance mainly comes from road conditions and traffic conditions, which is equivalent to friction in physics. Combined with the actual map path and traffic data, the resistance coefficient is refined to take into account the real-time traffic congestion and road type. The resistance equation is:

[0073] ;

[0074] in, represents the drag vector acting on the vehicle; represents the velocity vector of the vehicle; represents the basic resistance coefficient; represents the speed influence coefficient; Indicates the traffic condition impact coefficient; Indicates the current speed of the vehicle; Indicates the maximum speed of the vehicle; Indicates vehicle To school The path length; represents the free path length in the absence of congestion; Indicates the traffic density of the current road; Indicates the maximum traffic density of the road.

[0075] Through the set of dynamic equations, the force state and motion trajectory of the delivery vehicle in the force field can be simulated more accurately, and the delivery route and strategy can be dynamically adjusted based on the school's needs, road conditions and real-time traffic information.

[0076] Step S330: Perform numerical simulation based on the dynamic equations, use the Runge-Kutta method to calculate the position, speed, path and interaction force between each delivery vehicle and each school during the simulation process to obtain simulation results, and convert the simulation results into potential energy diagrams and dynamic field diagrams;

[0077] Specifically, during the numerical simulation, the fourth-order Runge-Kutta method is used to calculate the vehicle position and velocity at each time step to obtain the simulation results.

[0078] Compute the median value at each time step:

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] Update position and velocity:

[0084] ;

[0085] ;

[0086] in, , , and represents the intermediate value in the fourth-order Runge–Kutta method; represents the system of dynamic equations; Indicates the time point of the current time step; express The vehicle position vector at a given point in time; express The vehicle velocity vector at a point in time; represents the time step, represents the vehicle position vector at the next time step; Represents the vehicle velocity vector at the next time step.

[0087] Furthermore, numerical simulation is used to obtain the position, speed, path and interaction force between each delivery vehicle and each school at different time points. In order to better understand and analyze these data, the simulation results are converted into potential energy diagrams and dynamic field diagrams. First, collect simulation data, record the position and speed of the vehicle at each time step, and calculate the potential energy value of each position based on gravity and resistance. Map these potential energy values ​​to various positions on the map to generate a potential energy diagram, which shows the potential energy distribution formed by the gravity and resistance exerted on the vehicle during the delivery process, which helps to identify areas of high gravity and high resistance and optimize delivery paths and strategies. At the same time, calculate the velocity vector and generate a dynamic field diagram based on the position and velocity in the simulation data. Map the velocity vector to various positions on the map to show the speed and direction distribution of the vehicle during the delivery process, which helps to intuitively observe the vehicle's motion path and dynamic behavior. Through the conversion of potential energy diagrams and dynamic field diagrams, the force situation and motion trajectory of the vehicle can be more intuitively displayed. Combining these two diagrams, key issues in distribution and optimized paths can be effectively identified.

[0088] Step S340: Perform path planning and loading design processing according to the potential energy diagram and the power field diagram to obtain fifth information.

[0089] Preferably, this step uses an image recognition algorithm for analysis and processing to obtain an optimized preliminary loading scheme set and a corresponding preliminary route set. The specific process includes: first, grayscale processing and edge detection are performed on the potential energy map and the power field map to extract key features in the image. Then, a convolutional neural network is used to extract features in the image, identify low potential energy areas in the potential energy map and high speed areas in the power field map, and use the YOLO algorithm to locate the location of each school and the connection relationship between the roads. Based on the image recognition results, a road network diagram is established, and a graph search algorithm is applied to find the optimal path from the warehouse to each school. In terms of loading design, the demand map of each school is grayscale processed and edge detected by an image recognition algorithm, key features are extracted, and different demand areas are segmented. According to the image recognition results, the food demand information of each school is extracted, and the maximum amount of food that each vehicle can transport is calculated in combination with the vehicle's loading capacity and maximum speed.

[0090] Further, step S400 includes step S410 to step S450.

[0091] Step S410: extract the position and speed data of the vehicle at each time according to the fifth information and perform delivery task coding processing, and obtain the decision variable representation by converting the delivery order, target school and load of each vehicle into an integer sequence;

[0092] Specifically, first, the position and speed data of each vehicle at different time points are extracted from the fifth information, and its position and speed at each time step are recorded. Next, the delivery order, target school and loading capacity of each vehicle are converted into integer sequences. The delivery order is achieved by numbering each delivery point and recording the delivery order of the vehicle, the target school is achieved by assigning a unique integer code to each school, and the loading capacity is achieved by converting the vehicle's loading capacity into an integer representation. Finally, the extracted position, speed data and the encoded delivery task are combined to form a decision variable representation for subsequent optimization processing. In this way, the complex delivery path and loading data are converted into a unified integer sequence representation, which provides basic data for multi-objective optimization and ensures that the delivery task can be completed more efficiently.

[0093] Step S420: extract the vehicle travel time and distance according to the fifth information and calculate the cost and time efficiency, and obtain the target function set by setting a specific cost function and time target;

[0094] It should be noted that the cost function takes into account fuel costs, vehicle maintenance costs, labor costs, and food freshness costs. In addition, delivery time targets and time window constraints are set to ensure that the vehicle completes the delivery task within the specified time. In addition, delivery time targets and time window constraints are set to ensure that the vehicle completes the delivery task within the specified time.

[0095] The total cost minimization objective function is:

[0096] ;

[0097] in, represents the minimization of the total cost function; represents the fuel cost per unit distance of the delivery vehicle; Indicates the total distance traveled by the vehicle during the entire delivery process; Indicates the maintenance cost required for the vehicle per unit time; Indicates the total time the vehicle spends in the entire delivery process; Indicates the maintenance cost of the vehicle per unit distance; It represents the driver’s labor cost per unit time; It represents the cost of food due to the loss of freshness per unit time; It represents the total time it takes for vehicles to arrive at each delivery point; Indicates the school's serial number.

[0098] Delivery time minimization objective function formula:

[0099] ;

[0100] in, represents the objective function of minimizing delivery time; and Indicates the serial number of the school; represents the total number of schools; Departure from warehouse to the time required for school; Indicates from School to driving time to school; Indicates from School to The waiting time weight of the school.

[0101] Step S430: Obtain constraint conditions according to the second information and the preset delivery rules, where the constraint conditions include vehicle load limit and delivery time window;

[0102] It is understandable that the second information includes parameters such as the vehicle's load capacity, maximum speed, fuel consumption, etc. The preset delivery rules include the school's receiving time window, the maximum distance for each delivery, the maximum time for each delivery, etc. These information and rules together constitute constraints, which are used in the optimization algorithm to ensure that the generated delivery plan is feasible and efficient in actual operation.

[0103] Step S440: According to the objective function set and the constraint conditions, a multi-objective evolutionary algorithm is used to perform genetic operations on the decision variable representation to obtain an optimized solution set;

[0104] Specifically, the initial population is first generated to represent a set of possible delivery solutions. When generating the initial population, the geographical distribution of schools, the initial location of vehicles, and the shelf life of food ingredients need to be considered. Next, through the selection operation, the non-dominated sorting and crowding calculation methods are used to select individuals with excellent performance from the current population. In the selection process, individuals that can still efficiently complete the delivery task under the load and time window constraints are given priority. Then, new sub-individuals are generated through crossover and mutation operations. In the crossover operation, it is necessary to ensure that the new sub-individuals can meet the basic constraints, including load restrictions and time windows. The path-based crossover operation is used to ensure that the new solution after the crossover is reasonable in terms of path and loading. In the mutation operation, the problem-based mutation strategy is used to ensure that the mutated solution will not increase significantly in total cost and delivery time. Through multiple generations of iterations, the fitness of individuals in the population is gradually improved, and finally converges to a set of optimized solutions. In the genetic operation process, the objective function and constraints are integrated through the fitness function, and the penalty value is added to individuals that violate the load limit and delivery time window to ensure that the generated delivery plan is feasible and efficient in actual operation. In this way, a balance can be effectively found between multiple objective functions, and a set of optimized solutions can be obtained to improve the overall efficiency and feasibility of the delivery system.

[0105] Step S450: Evaluate and screen the set of optimized solutions, compare the performance of each solution in terms of cost, delivery time and food freshness, and use Pareto front analysis to screen out the optimal solution, thereby obtaining the sixth information.

[0106] This step evaluates each optimization solution in terms of cost, delivery time, and food freshness. Using the Pareto frontier analysis method, by comparing all solutions in the optimization solution set, the non-dominated solution set, namely the Pareto frontier, is found. The optimal solution in the non-dominated solution set is determined by judging the Pareto dominance relationship. Finally, the most feasible solution in practical application is further selected from the non-dominated solution set to form the sixth information.

[0107] Further, step S500 includes step S510 to step S530.

[0108] Step S510: define states according to the sixth information, each state represents a delivery state of a delivery node, and the transition probability between states is based on the logical relationship between the route set and the loading plan. By setting the dependency and interaction between each delivery route and each loading, a preliminary Markov random field model reflecting the delivery operation is obtained;

[0109] It should be noted that the Markov random field model is a probabilistic graphical model suitable for describing the joint distribution of a set of random variables. It represents variables and their dependencies through nodes and edges, where each node represents a variable and the edge represents the mutual influence between variables. The Markov random field model assumes that the state of each variable depends only on the state of its neighbor variables and is independent of the state of other variables. In this embodiment, the Markov random field model is suitable for describing the relationship between various distribution nodes (such as warehouses and schools) in the distribution system. The state of the distribution node (such as inventory, demand, expected arrival time, etc.) is affected by the state of the adjacent nodes. For example, when a distribution vehicle departs from a node, the state of its next distribution node will affect the vehicle's path selection and loading strategy. Through the Markov random field model, the dynamic changes of the state of each node in the distribution process can be effectively simulated and predicted, thereby optimizing the distribution path and strategy. In order to construct a Markov random field model suitable for the application scenario of the present invention, it is necessary to define the state of each distribution node and establish the transition probability between each state. In this embodiment, the state of the distribution node includes the current inventory, demand and expected arrival time. Specifically, the transition probability between states can be set based on the optimized route and loading plan, reflecting the vehicle's driving path and loading situation between different nodes. By setting the dependencies and interactions between each delivery route and each loading, the Markov random field model can capture the dynamic changes in the distribution system and provide support for real-time adjustment and optimization.

[0110] Step S520: Based on the real-time traffic data and the real-time delivery status, the preliminary Markov random field model is updated in real time for the state transition probability, and the updated Markov random field model is obtained by adjusting the route feasibility caused by the traffic changes and the delivery progress changes;

[0111] Specifically, this process includes obtaining real-time traffic data from traffic monitoring systems and navigation applications, as well as obtaining the current location, speed, remaining delivery tasks, and current status of each delivery node from the distribution system. Based on these data, the state transition probabilities of each road and delivery node in the Markov random field model are adjusted. For example, if a road is congested, the transition probability of the road is reduced; if a road is unobstructed, the transition probability of the road is increased. If the demand for a distribution node increases, the transition probability to the node is increased. In this way, the real-time updated Markov random field model can accurately reflect the current distribution environment and status, thereby optimizing the distribution strategy and ensuring the efficient completion of the distribution task.

[0112] Step S530: According to the updated Markov random field model, the path reorganization processing from the current delivery state to all subsequent unfinished delivery points is performed, and the optimized delivery plan is obtained by optimizing the calculation and selecting the successful delivery path with the highest probability and the best delivery time window. The optimized delivery plan includes the dynamically adjusted delivery sequence and route.

[0113] It should be noted that, starting from the current distribution status, the updated Markov random field model is used to calculate the possible paths to all unfinished distribution points, and the success probability of each path is determined by probability calculation. Then, among all possible paths, the path with the highest probability of success is selected to ensure that more distribution points are covered in the shortest time and that an efficient distribution order is maintained. At the same time, for each unfinished distribution point, its optimal distribution time window is calculated, taking into account the traffic conditions, the needs and priorities of the distribution points, to ensure that the vehicle can arrive at the optimal time. Finally, an optimized distribution plan containing dynamically adjusted distribution sequences and routes is generated. In this way, the distribution efficiency can be maximized and the distribution tasks can be efficiently completed in a dynamically changing environment.

[0114] Embodiment 2:

[0115] like Figure 2 As shown, this embodiment provides a food delivery vehicle loading and route optimization system, the system comprising:

[0116] Acquisition module 1, used to acquire first information, second information and third information, the first information includes historical food usage data and real-time inventory status of each school in the delivery area, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area;

[0117] Prediction module 2, used for performing prediction processing based on the first information to obtain fourth information, where the fourth information is food demand information and urgency of each school;

[0118] A construction module 3 is used to construct an equivalent particle model according to the fourth information, the third information and the second information, and simulate the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading solution set of the delivery vehicle and a corresponding preliminary route set;

[0119] Optimization module 4, used for performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, where the sixth information is an optimized loading scheme set and a corresponding route set;

[0120] The adjustment module 5 is used to construct a Markov random field model according to the sixth information, real-time traffic data and real-time delivery status, dynamically adjust the unfinished delivery tasks based on the Markov random field model, and update the subsequent delivery plan.

[0121] In a specific implementation disclosed in the present application, the prediction module 2 includes:

[0122] The first analysis unit 21 performs time series analysis on the historical food usage data based on an autoregressive mean model to obtain a demand trend result;

[0123] The first prediction unit 22 is used to predict the immediate demand of each food material using a support vector machine algorithm according to the demand trend result and the real-time inventory status to obtain a prediction result, wherein the prediction result includes the food material demand information within a preset time period in the future;

[0124] The first evaluation unit 23 is used to perform evaluation processing according to the prediction result, and use a decision tree classification model to calculate the urgency of each demand according to the demand for food usage, inventory level and proximity to the expiration date.

[0125] In a specific implementation disclosed in the present application, building block 3 includes:

[0126] A first processing unit 31 is used to perform model parameter initialization processing according to the fourth information, the third information and the second information to obtain an initial parameter set, wherein the initial parameter set includes a school gravity parameter, a dynamic property of the vehicle and an initial motion state;

[0127] A first construction unit 32 is used to construct a dynamic equation according to the initial parameter set to obtain a group of dynamic equations for describing the force state of the delivery vehicle in the force field;

[0128] The first simulation unit 33 performs numerical simulation based on the dynamic equations, uses the Runge-Kutta method to calculate the position, speed, path and interaction force between each delivery vehicle and each school during the simulation process to obtain simulation results, and converts the simulation results into a potential energy diagram and a dynamic field diagram;

[0129] The second processing unit 34 is used to perform path planning and loading design processing according to the potential energy diagram and the power field diagram to obtain fifth information.

[0130] In a specific implementation disclosed in the present application, the optimization module 4 includes:

[0131] The first conversion unit 41 is used to extract the position and speed data of the vehicle at each time according to the fifth information and perform delivery task coding processing, and obtain the decision variable representation by converting the delivery order, target school and load of each vehicle into an integer sequence;

[0132] The third processing unit 42 is used to extract the vehicle travel time and distance according to the fifth information and perform cost and time efficiency calculation, and obtain an objective function set by setting a specific cost function and time target;

[0133] A fourth processing unit 43, configured to obtain a constraint condition according to the second information and a preset delivery rule, wherein the constraint condition includes a vehicle load limit and a delivery time window;

[0134] A first optimization unit 44 is used to perform genetic operations on the decision variable representation using a multi-objective evolutionary algorithm according to the objective function set and the constraint conditions to obtain an optimization solution set;

[0135] The second evaluation unit 45 is used to evaluate and screen the set of optimized solutions, compare the performance of each solution in three dimensions of cost, delivery time and food freshness, and use Pareto frontier analysis to screen out the optimal solution to obtain the sixth information.

[0136] In a specific implementation disclosed in the present application, the adjustment module 5 includes:

[0137] A second construction unit 51 is used to define states according to the sixth information, each state represents a delivery state of a delivery node, and the transition probability between states is based on the logical relationship between the route set and the loading plan. By setting the dependency and interaction between each delivery route and each loading, a preliminary Markov random field model reflecting the delivery operation is obtained;

[0138] A first updating unit 52 performs real-time updating of the state transition probability of the preliminary Markov random field model based on the real-time traffic data and the real-time delivery status, and obtains an updated Markov random field model by adjusting the route feasibility caused by traffic changes and delivery progress changes;

[0139] The first adjustment unit 53 is used to reorganize the path from the current delivery state to all subsequent unfinished delivery points according to the updated Markov random field model, and obtain the optimized delivery plan by optimizing the calculation and selecting the successful delivery path with the highest probability and the best delivery time window. The optimized delivery plan includes the dynamically adjusted delivery sequence and route.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for optimizing the loading and routing of food delivery vehicles, characterized in that: include: Acquire first information, second information and third information, wherein the first information includes historical food usage data and real-time inventory status of schools in the delivery area, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area; Performing prediction processing on the first information to obtain fourth information, wherein the fourth information is food demand information and urgency of each school; constructing an equivalent particle model according to the fourth information, the third information and the second information, and performing simulation processing on the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading solution set of the delivery vehicle and a corresponding preliminary route set; Performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, wherein the sixth information is an optimized loading scheme set and a corresponding route set; A Markov random field model is constructed according to the sixth information, the real-time traffic data and the real-time delivery status, and unfinished delivery tasks are dynamically adjusted based on the Markov random field model, and subsequent delivery plans are updated; Constructing an equivalent particle model according to the fourth information, the third information and the second information, and simulating the equivalent particle model to obtain fifth information, including: Performing model parameter initialization processing according to the fourth information, the third information and the second information to obtain an initial parameter set, wherein the initial parameter set includes school gravity parameters, dynamic properties of the vehicle and an initial motion state; Performing a dynamic equation construction process according to the initial parameter set to obtain a group of dynamic equations for describing the force state of the delivery vehicle in the force field; Numerical simulation is performed based on the dynamic equations, and the Runge-Kutta method is used to calculate the position, speed, path and interaction force between each delivery vehicle and each school during the simulation process to obtain simulation results, and the simulation results are converted into potential energy diagrams and dynamic field diagrams; The fifth information is obtained by performing path planning and loading design processing according to the potential energy diagram and the power field diagram.

2. The food delivery vehicle loading and route optimization method according to claim 1 is characterized in that: Performing prediction processing on the first information to obtain fourth information includes: Performing time series analysis on the historical food usage data based on an autoregressive mean model to obtain demand trend results; According to the demand trend result and the real-time inventory situation, a support vector machine algorithm is used to predict the immediate demand for each food to obtain a prediction result, wherein the prediction result includes food demand information within a preset time period in the future; An evaluation process is performed based on the prediction results, and a decision tree classification model is used to calculate the urgency of each demand based on the demand for food ingredients, inventory levels, and proximity to the expiration date.

3. The food delivery vehicle loading and route optimization method according to claim 1 is characterized in that: Performing multi-objective optimization processing according to the fifth information and the second information to obtain sixth information includes: According to the fifth information, the position and speed data of the vehicle at each time are extracted and the delivery task coding process is performed, and the delivery order, target school and load of each vehicle are converted into an integer sequence to obtain a decision variable representation; Extracting the vehicle travel time and distance according to the fifth information and calculating the cost and time efficiency, and obtaining the target function set by setting a specific cost function and time target; Obtaining constraint conditions according to the second information and preset delivery rules, wherein the constraint conditions include vehicle load limit and delivery time window; According to the objective function set and the constraint conditions, a multi-objective evolutionary algorithm is used to perform genetic operations on the decision variable representation to obtain an optimized solution set; Evaluation and screening are performed based on the set of optimized solutions, and the performance of each solution in three dimensions, namely, cost, delivery time and food freshness, is compared, and the optimal solution is screened out using Pareto front analysis to obtain the sixth information.

4. The food delivery vehicle loading and route optimization method according to claim 1 is characterized in that: A Markov random field model is constructed according to the sixth information, the real-time traffic data and the real-time delivery status, and unfinished delivery tasks are dynamically adjusted based on the Markov random field model, and subsequent delivery plans are updated, including: Defining states according to the sixth information, each state represents a delivery state of a delivery node, and the transition probability between states is based on the logical relationship between the route set and the loading plan. By setting the dependency and interaction between each delivery route and each loading, a preliminary Markov random field model reflecting the delivery operation is obtained; Based on the real-time traffic data and the real-time delivery status, the preliminary Markov random field model is updated in real time with respect to the state transition probability, and the updated Markov random field model is obtained by adjusting the route feasibility caused by the traffic changes and the delivery progress changes; According to the updated Markov random field model, the path reorganization processing is carried out from the current delivery status to all subsequent unfinished delivery points, and the optimized delivery plan is obtained by optimizing the calculation and selecting the successful delivery path with the highest probability and the best delivery time window. The optimized delivery plan includes the dynamically adjusted delivery sequence and route.

5. A food delivery vehicle loading and route optimization system, characterized in that: include: an acquisition module, for acquiring first information, second information and third information, wherein the first information includes historical food usage data and real-time inventory status of schools in the area to be delivered, the second information includes delivery vehicle parameters, and the third information includes map data and historical traffic data of the delivery area; A prediction module, configured to perform prediction processing based on the first information to obtain fourth information, wherein the fourth information is food demand information and urgency of each school; a construction module, configured to construct an equivalent particle model according to the fourth information, the third information and the second information, and simulate the equivalent particle model to obtain fifth information, wherein the fifth information includes a preliminary loading scheme set of the delivery vehicle and a corresponding preliminary route set; an optimization module, configured to perform multi-objective optimization processing according to the fifth information and the second information to obtain sixth information, wherein the sixth information is an optimized loading scheme set and a corresponding route set; An adjustment module, configured to construct a Markov random field model according to the sixth information, real-time traffic data and real-time delivery status, dynamically adjust unfinished delivery tasks based on the Markov random field model, and update subsequent delivery plans; The building blocks include: a first processing unit, configured to perform model parameter initialization processing according to the fourth information, the third information and the second information to obtain an initial parameter set, wherein the initial parameter set includes a school gravity parameter, a dynamic property of the vehicle and an initial motion state; A first construction unit is used to perform a dynamic equation construction process according to the initial parameter set to obtain a group of dynamic equations for describing the force state of the delivery vehicle in the force field; The first simulation unit performs numerical simulation based on the dynamic equations, uses the Runge-Kutta method to calculate the position, speed, path and interaction force between each delivery vehicle and each school during the simulation process to obtain simulation results, and converts the simulation results into a potential energy diagram and a dynamic field diagram; The second processing unit is used to perform path planning and loading design processing according to the potential energy diagram and the power field diagram to obtain fifth information.

6. The food delivery vehicle loading and route optimization system according to claim 5, characterized in that: The prediction module comprises: A first analysis unit performs time series analysis on the historical food usage data based on an autoregressive mean model to obtain a demand trend result; A first prediction unit, configured to use a support vector machine algorithm to predict the immediate demand for each food material according to the demand trend result and the real-time inventory status to obtain a prediction result, wherein the prediction result includes food material demand information within a preset time period in the future; The first evaluation unit is used to perform evaluation processing according to the prediction result, and use a decision tree classification model to calculate the urgency of each demand according to the demand for food usage, inventory level and proximity to the expiration date.

7. The food delivery vehicle loading and route optimization system according to claim 5, characterized in that: The optimization module includes: A first conversion unit is used to extract the position and speed data of the vehicle at each time according to the fifth information and perform delivery task coding processing, and obtain a decision variable representation by converting the delivery order, target school and load of each vehicle into an integer sequence; A third processing unit is used to extract the vehicle travel time and distance according to the fifth information and perform cost and time efficiency calculation, and obtain an objective function set by setting a specific cost function and time target; a fourth processing unit, configured to obtain a constraint condition according to the second information and a preset delivery rule, wherein the constraint condition includes a vehicle load limit and a delivery time window; A first optimization unit, configured to perform genetic operations on the decision variable representation using a multi-objective evolutionary algorithm according to the objective function set and the constraint conditions to obtain an optimization solution set; The second evaluation unit is used to evaluate and screen the set of optimized solutions, compare the performance of each solution in three dimensions of cost, delivery time and food freshness, and use Pareto frontier analysis to screen out the optimal solution to obtain the sixth information.

8. The food delivery vehicle loading and route optimization system according to claim 5, characterized in that: The adjustment module comprises: A second construction unit is used to define states according to the sixth information, each state represents a delivery state of a delivery node, and the transition probability between states is based on the logical relationship between the route set and the loading plan, and a preliminary Markov random field model reflecting the delivery operation is obtained by setting the dependency and interaction between each delivery route and each loading; A first updating unit performs real-time updating processing of the state transition probability of the preliminary Markov random field model based on real-time traffic data and real-time delivery status, and obtains an updated Markov random field model by adjusting the route feasibility caused by traffic changes and delivery progress changes; The first adjustment unit is used to reorganize the path from the current delivery state to all subsequent unfinished delivery points according to the updated Markov random field model, and obtain the optimized delivery plan by optimizing the calculation and selecting the successful delivery path with the highest probability and the best delivery time window. The optimized delivery plan includes the dynamically adjusted delivery sequence and route.