A method and system for logistics distribution based on the internet of things

Abstract

The present application relates to the technical field of logistics distribution, in particular to a logistics distribution method and system based on the Internet of Things, comprising: calculating the predicted travel time cost of each road section in the road network in multiple future time windows through a space-time prediction model; generating a discrete signing success probability distribution for each order within the multiple time windows using a Bayesian inference model; applying a multi-objective optimization algorithm to obtain a set of Pareto optimal distribution scheme sets; selecting a final distribution route from the Pareto optimal distribution scheme set according to a preset risk preference or cost benefit strategy, and issuing it to the vehicle terminal for execution. The present application can provide decision makers with a scientific basis for selection based on different business strategies, making the final distribution decision more flexible and strategic.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics distribution. More particularly, the present application relates to a logistics distribution method and system based on the Internet of Things. BACKGROUND

[0002] With the rapid development of e-commerce and instant delivery industry, the efficiency, cost and user satisfaction of logistics distribution have become key indicators to measure the core competitiveness of enterprises. In order to cope with the increasingly complex delivery demand, the existing technology mainly focuses on the research of vehicle routing problem (VRP), aiming to plan a series of optimal customer visiting routes for multiple vehicles. Traditional logistics distribution systems usually utilize real-time GPS data and electronic map information, combined with heuristic algorithms (such as genetic algorithm, ant colony algorithm, etc.) or precise algorithms to generate delivery plans. These methods to some extent solve the basic problem of path planning, such as optimizing the total travel mileage or total travel time under the constraints of vehicle load, volume and customer time window. However, these systems still have significant limitations in dealing with the dynamics of urban traffic and the uncertainty in the delivery process.

[0003] The existing technology has the following deficiencies: First, in terms of traffic state prediction, most systems rely on real-time traffic information or simple historical data average, which is difficult to accurately predict the road passage time in a specific period of time in the future. They lack the ability to deeply model the complex spatio-temporal dependence between road network topology and traffic flow, resulting in the planned path may not be optimal due to traffic congestion in actual execution. Secondly, in terms of delivery success rate, traditional methods usually regard customer time window as a certain constraint, ignoring the probabilistic problem of successful delivery within the time window. For example, the successful delivery rate of different types of destinations (such as office buildings and residential areas) is very different in different time periods. The existing system rarely quantitatively models such delivery uncertainty, resulting in high secondary delivery rate and increasing operating costs. In terms of optimization objectives, most systems adopt single objective or fuse multiple objectives into a single objective function through linear weighting, which cannot reveal the inherent trade-off relationship between different objectives (such as time cost and delivery success rate), and it is difficult to provide flexible and diverse decision choices according to different business strategies (such as cost priority or service quality priority). SUMMARY

[0004] To solve the above technical problem of being difficult to provide flexible and diverse decision choices according to different business strategies, the present application provides the following solutions.

[0005] A logistics distribution method based on Internet of Things comprises the following steps: obtaining order information to be distributed, real-time state data of vehicles, road network topology structure of a target distribution area and historical traffic flow data; based on the road network topology structure and the historical traffic flow data, a spatio-temporal prediction model by fusing a graph convolution network and a gated recurrent unit is used to calculate predicted travel time costs of each road section in the road network in multiple future time windows; based on historical distribution data and time sensitivity, a Bayesian inference model is used to generate discrete signed success probability distributions for each order in the multiple time windows; the endurance capability, load limit and health state index in the real-time state data of the vehicles are combined to construct a feasibility constraint set for vehicle task execution; the predicted travel time costs, the signed success probability distributions and the feasibility constraint set are taken as inputs, a multi-objective optimization algorithm is applied to minimize the total predicted travel time costs of all vehicles and maximize the total signed success probability of all orders as independent optimization objectives, and a set of Pareto optimal distribution scheme sets containing multiple non-inferior solutions is obtained; according to a preset risk preference or cost-benefit strategy, a final distribution route is selected from the Pareto optimal distribution scheme set and is issued to a vehicle terminal for execution.

[0006] The present application can provide a scientific basis for decision makers to make choices based on different business strategies, and the final distribution decision is more flexible and strategic.

[0007] Further, the spatio-temporal prediction model by fusing the graph convolution network and the gated recurrent unit is used to calculate the predicted travel time costs of each road section in the road network in the multiple future time windows, which comprises:

[0008] The road network topology structure is abstracted as a graph, wherein intersections are nodes and road sections are edges; a graph convolution network is used to extract spatial dependency features of each road section; traffic flow data of each road section in a preset historical period is constructed as a time sequence and is input into a gated recurrent unit to extract time variation rules; the spatial dependency features and the time variation rules are fused to output predicted travel time costs of each time window in a future prediction time domain in a preset time length.

[0009] The present application fully utilizes the characteristics of the road network, constructs a graph model and uses a graph convolution network for processing.

[0010] Further, the Bayesian inference model is used to generate discrete signed success probability distributions for each order in the multiple time windows, which comprises:

[0011] Based on historical delivery data, a priori signing success probability related to order destination type and delivery time period is constructed; according to the time sensitivity level of the order, a likelihood function is established, wherein the higher the time sensitivity level of the order, the higher the signing likelihood value within the time window required by the order; the prior signing success probability and the likelihood function are integrated by applying the Bayes formula to obtain the posterior signing success probability of the order in each future time window.

[0012] The present application quantifies the signing success probability of the order by using the Bayes inference model, so that the delivery scheme can actively avoid the signing failure risk.

[0013] Further, the construction of the feasibility constraint set of vehicle task execution includes:

[0014] The cruising ability constraint is set as the total estimated energy consumption of vehicle path planning being not more than the current available energy; the load limit constraint is set as the total weight and total volume of the orders carried by the vehicle being not more than the rated load and rated volume of the vehicle; and the health state constraint is set as the health state index of the vehicle being higher than a preset threshold.

[0015] Further, the application of the multi-objective optimization algorithm to obtain a set of Pareto optimal delivery scheme sets containing multiple non-inferior solutions includes:

[0016] A multi-objective evolutionary algorithm based on non-dominated sorting is adopted to encode the potential delivery scheme; the first objective function is set as the sum of the predicted travel time cost of all vehicle paths, and is minimized; the second objective function is set as the sum of the signing success probability of all orders at their respective delivery time points, and is maximized; by iteratively performing selection, crossover and mutation operations, a set of Pareto front solutions is generated.

[0017] Instead of directly generating a scheme, the present application generates a set of Pareto optimal delivery scheme sets revealing the trade-off relationship between time cost and signing success rate, laying a foundation for subsequent steps.

[0018] Further, the selection of the final delivery route from the set of Pareto optimal delivery schemes according to the preset risk preference or cost benefit strategy includes:

[0019] For each scheme in the set of Pareto optimal delivery schemes, the total predicted travel time cost and the total signing success probability are normalized; according to the preset risk preference or cost benefit strategy, the normalized total predicted travel time cost and the total signing success probability are assigned corresponding weight coefficients; based on the weight coefficients, the weighted comprehensive score of each scheme is calculated, and the scheme with the optimal comprehensive score is selected as the final delivery route.

[0020] The application provides a final distribution route based on different service strategies, so that the decision is more flexible.

[0021] Further, the order information includes an order number, a cargo weight, a volume, a delivery address and a customer expected time.

[0022] Further, the non-dominated sorting multi-objective evolutionary algorithm is an NSGA-II algorithm.

[0023] Further, the risk preference includes a service priority strategy.

[0024] The application also provides a logistics distribution system based on the Internet of Things, which comprises a processor and a memory.

[0025] Compared with the prior art, the logistics distribution method provided by the application has achieved remarkable beneficial effects by introducing an advanced prediction and optimization model. First, the spatiotemporal prediction model fusing a graph convolution network and a gated recurrent unit can more accurately calculate the future travel time, providing a reliable data basis for path planning and improving the actual execution efficiency of the route. Second, the Bayesian inference model is used to quantify the order signing success probability, so that the distribution scheme can actively avoid the signing failure risk, thereby significantly improving the one-time delivery rate and reducing the operating cost caused by secondary distribution. By applying the multi-objective optimization algorithm, the application can generate a set of Pareto optimal distribution schemes that reveal the trade-off relationship between the time cost and the signing success rate, providing a scientific basis for decision makers to select based on different business strategies (such as cost priority or service quality priority), so that the final distribution decision is more flexible and strategic. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart schematically showing a logistics distribution method based on the Internet of Things according to an embodiment of the application;

[0027] Figure 2 is a schematic diagram schematically showing steps S5 and S6 according to an embodiment of the application;

[0028] Figure 3 is a schematic diagram schematically showing a logistics distribution system based on the Internet of Things according to an embodiment of the application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application.

[0030] As Figure 1As shown, a logistics distribution method based on Internet of Things includes the following steps:

[0031] S1, obtaining order information to be distributed, real-time state data of the vehicle, road network topology structure of the target distribution area and historical traffic flow data;

[0032] Specifically, the order information includes order number, cargo weight, volume, delivery address and customer expected time; through the vehicle-mounted Internet of Things terminal such as GPS locator and on-board diagnostic system OBD interface, the latitude and longitude coordinates, remaining range, current load and device health diagnosis code of the vehicle are periodically collected as the real-time state data of the vehicle; by calling the open platform API of Gaode Map or Baidu Map, the road network topology structure (including the digital road network map composed of roads and intersections) and the historical traffic flow data (the traffic flow data is very rich, including flow, speed and density, etc., and in this embodiment, mainly includes the historical average speed or passing time data of the corresponding road section aggregated by hour or minute in the past months) are obtained.

[0033] S2, based on the road network topology structure and historical traffic flow data, a spatio-temporal prediction model by fusing graph convolution network and gate recurrent unit is used to calculate the predicted passing time cost of each road section in the road network in multiple future time windows;

[0034] Specifically, the road network topology structure is abstracted into a graph, in which the intersections are nodes and the road sections are edges. The spatial dependency features of each road section are extracted by using the graph convolution network. The traffic flow data of each road section in the preset historical period is constructed into a time series and input into the gate recurrent unit to extract the time variation law; the spatial dependency features and the time variation law are fused to output the predicted passing time of each time window in the future prediction time domain with a preset time length, i.e. the predicted passing time cost.

[0035] For example, taking the road network of a certain area in the city center as an example, the area contains hundreds of intersections and road sections connecting them. First, the road network is constructed into a graph structure, in which each intersection such as People's Square or Century Avenue intersection is defined as a node of the graph, and the road connecting these intersections such as a section of Nanjing East Road is defined as an edge of the graph. The graph convolution network model learns this graph structure and understands how the congestion of People's Square will affect other road sections several kilometers away through Yan'an Elevated Road, which is the spatial dependency feature.

[0036] For example, for the section of Nanjing Road, the average speed and traffic volume data every five minutes in the past thirty days are collected to form a long time series. The time series is input into the gated recurrent unit model, and the model learns that there is a periodic traffic time peak on this section from 8:00 to 9:00 in the morning and from 6:00 to 7:00 in the evening, i.e., the time variation rule (i.e., the time periodicity rule feature). Finally, the model fuses the spatial dependency features extracted by the graph convolution network with the time periodicity rule features extracted by the gated recurrent unit, so as to output the accurate travel time prediction value for each five-minute time window in the next two hours, for example, it is predicted that it takes 300 seconds to pass through this section at 8:05, and only 120 seconds at 10:10.

[0037] S3, based on historical delivery data and time sensitivity, using a Bayesian inference model to generate a discrete signed success probability distribution for each order in the plurality of time windows;

[0038] Specifically, based on a large amount of historical delivery data (including destination location area, time period, and prior signed success probability of different destination types), the prior signed success probability of different areas, different time periods, and different destination types such as office buildings, residential areas, and shops is counted.

[0039] For an order to be delivered, the prior signed success probability related to the destination type and the delivery time period of the order is constructed; according to the time sensitivity level of the order, a likelihood function is established to measure how likely the observed evidence (the time sensitivity of the order) is under a given hypothesis (such as the order can be signed successfully in a certain time period), wherein the higher the time sensitivity level of the order, the higher the signed likelihood value of the order within the required time window; the prior signed success probability and the likelihood function are integrated by applying the Bayes formula to obtain the posterior signed success probability of the order in future time windows.

[0040] Specifically, assuming that the system needs to evaluate an order sent to a residential area. Based on the analysis of massive historical delivery data, the system knows that the address type of the residential area has a prior probability of 70% of signed success rate from 9:00 to 11:00 in the morning on weekdays, and a prior probability of 40% of signed success rate from 14:00 to 16:00 in the afternoon. This constitutes the prior signed success probability of the order.

[0041] Specifically, if the order is a medicine that is urgently needed, its time sensitivity level is set to the highest level. The system will create a likelihood function for this order, which has a very high likelihood value, such as 0.95, within the time window of 3-4 pm specified by the customer, and a likelihood value close to zero at other time windows. The Bayesian inference model will then combine this high likelihood value with the low prior probability of 40% of successful delivery in the afternoon, to calculate an updated posterior probability of successful delivery. The final result can be that the probability of successful delivery in the 3-4 pm window is greatly increased to 90%, and the probability of other time windows is correspondingly reduced, forming a probability distribution (i.e. generating a discrete probability distribution of successful delivery within the plurality of time windows), guiding the delivery personnel to deliver at the optimal time. That is, when the Bayesian formula combines the "low prior probability of 40% of successful delivery in the afternoon" with the "very high likelihood value of 0.95 within the time window required by the order", the "probability of successful delivery in the 3-4 pm window is greatly increased to 90%" is obtained, and this 90% is the posterior probability of successful delivery of the order in the 3-4 pm window.

[0042] S4, combining the endurance, load limit and health status index in the real-time state data of the vehicle, to construct a feasibility constraint set for vehicle task execution;

[0043] Specifically, the constraint set is composed of a series of mathematical inequalities or logical judgments. For example, for any vehicle, the total weight of all the goods of the orders allocated to it must not exceed the rated load upper limit of the vehicle; the total driving distance of its planned route must be less than the maximum endurance distance supported by the current remaining fuel or power; the health status index of the vehicle, a numerical value that comprehensively evaluates the state of components such as the engine and tires, must be higher than a preset safety threshold, for example 80 points, otherwise the vehicle does not participate in this task allocation. These conditions together define the boundaries of each vehicle that can effectively execute the task.

[0044] In an optional embodiment, the constructing a feasibility constraint set for vehicle task execution comprises: setting the endurance constraint as the total estimated energy consumption of the vehicle's planned path not exceeding the currently available energy; setting the load limit constraint as the total weight and volume of the orders carried by the vehicle not exceeding the rated load and rated volume of the vehicle; and setting the health status constraint as the health status index of the vehicle being higher than a preset threshold.

[0045] Specifically, take an electric delivery vehicle as an example, its current battery remaining capacity is fifty kilowatt-hours, which is its current available energy. The system estimates the total estimated energy consumption of the path to be fifty-three kilowatt-hours according to the total length of the path to be planned and the weight of the vehicle cargo. Since the estimated energy consumption of fifty-three kilowatt-hours exceeds the available energy of fifty kilowatt-hours, the path scheme is determined to be infeasible due to not meeting the endurance constraint.

[0046] Specifically, the rated load of the vehicle is five hundred kilograms, and the rated volume is four cubic meters. An order combination to be allocated has a total weight of four hundred and eighty kilograms, which meets the rated load requirement, but its total volume is calculated to be four point two cubic meters, which exceeds the rated volume of the vehicle. Therefore, this order combination scheme cannot be allocated to the vehicle due to not meeting the load limit constraint. In addition, each vehicle has a dynamically updated health status index ranging from zero to one hundred. The system sets a threshold of eighty, if a vehicle's health index drops to seventy-five due to severe tire wear or battery aging, any delivery scheme involving the vehicle will be automatically rejected by the system due to not meeting the health status constraint. The health status index can be obtained by predefining the evaluation rules by technicians.

[0047] S5, taking the predicted travel time cost, the probability distribution of successful signing, and the set of feasibility constraints as inputs, applying a multi-objective optimization algorithm to minimize the predicted travel time cost of all vehicles and maximize the total probability of successful signing of all orders as independent optimization objectives, to obtain a set of Pareto optimal delivery scheme sets containing multiple non-inferior solutions.

[0048] In an optional embodiment, the application of a multi-objective optimization algorithm to obtain a set of Pareto optimal delivery scheme sets containing multiple non-inferior solutions includes: using a non-dominated sorting genetic algorithm (such as the NSGA-II algorithm) to encode the potential delivery scheme; setting the first objective function as the sum of the predicted travel time costs of all vehicle travel paths, and minimizing it; setting the second objective function as the sum of the total probability of successful signing of all orders at their respective delivery time points, and maximizing it; generating a set of Pareto optimal solutions, i.e. a set of Pareto optimal delivery schemes, by iteratively performing selection, crossover and mutation operations.

[0049] As Figure 2As shown, specifically, the algorithm aims to balance two conflicting objectives, namely total delivery time and total successful delivery rate. A potential delivery plan, i.e. which driver delivers which package and the delivery order, is encoded into a gene sequence. The algorithm first randomly generates one hundred such plans as the initial population. In the iteration process, the algorithm evaluates the total predicted travel time of each plan, e.g. plan A takes four hundred and twenty minutes in total, plan B takes five hundred minutes in total. Meanwhile, the algorithm also calculates the overall successful delivery rate of each plan, e.g. plan A has an overall success rate of eighty-five percent, plan B has a success rate of ninety-five percent.

[0050] Specifically, through repeated selection, crossover and mutation operations, the algorithm continuously evolves these plans. After hundreds of generations of evolution, the algorithm outputs a Pareto front solution set. This set may contain plan A, which takes three hundred and eighty minutes in total and has an eighty percent success rate; plan B, which takes four hundred and thirty minutes in total and has a ninety-two percent success rate; and plan C, which takes five hundred and ten minutes in total and has a ninety-eight percent success rate. Here, no plan is superior to another in all objectives, e.g. plan B has a higher success rate than A but is more time-consuming, and is less time-consuming than C but has a lower success rate. These plans collectively constitute a set of high-quality alternative plans representing different trade-off strategies.

[0051] S6, according to the preset risk preference or cost-benefit strategy, select the final delivery route from the set of Pareto optimal delivery plans, and issue it to the vehicle terminal for execution.

[0052] As Figure 2 shown, specifically, the decision maker can set the selection strategy according to the current business focus. For example, during a promotional campaign, in pursuit of customer experience, a service-first strategy can be adopted, i.e. choose the plan with the highest overall successful delivery rate from the Pareto optimal solution set. In daily operations, to control costs, the plan with the lowest total travel time cost can be selected. Alternatively, the TOPSIS decision analysis method can be applied, by calculating the relative closeness of each non-inferior solution to the ideal point (i.e. the shortest time and the highest probability) and the negative ideal point (i.e. the longest time and the lowest probability), a compromise plan that performs most balanced in both objectives can be quantitatively selected. After selecting the plan, the system packages the specific travel path of each vehicle, the customer visit order and the navigation instructions, and sends them to the driver's mobile application or the vehicle-mounted navigation device of the corresponding vehicle through the wireless network.

[0053] In an optional embodiment, the selecting the final delivery route from the set of Pareto optimal delivery schemes according to a preset risk preference or cost benefit strategy comprises: normalizing the total time cost (i.e. the sum of predicted travel time costs) and the integrated successful delivery probability (i.e. the sum of successful delivery probabilities) of each scheme in the set of Pareto optimal delivery schemes; assigning corresponding weight coefficients to the normalized total time cost and the integrated successful delivery probability according to the preset strategy; calculating a weighted integrated score of each scheme based on the weight coefficients, and selecting the scheme with the optimal integrated score as the final delivery route.

[0054] Specifically, assuming that there are three schemes in the set of Pareto optimal schemes, the time costs of which are three hundred minutes, three hundred and fifty minutes and four hundred minutes respectively, and the corresponding integrated successful delivery probabilities are eighty-five percent, ninety-two percent and ninety-six percent respectively. The system first normalizes these values, mapping both the time cost and the successful probability to the interval of zero to one, so as to facilitate comparison.

[0055] Specifically, if the current business strategy is cost priority, for example, ordinary package delivery in off-peak hours, the operator can set the weight of the time cost to be zero point seven and the weight of the successful delivery probability to be zero point three. The system will calculate the weighted integrated score of each scheme with this weight combination, at this time, the scheme with a cost of three hundred minutes is likely to get the highest score because of its huge advantage in time. Conversely, if it is a batch of high-value same-day delivery urgent items, the strategy will be service quality priority, and the time cost weight may be set to zero point two and the successful probability weight to zero point eight. In this case, the scheme with a successful probability of ninety-six percent is likely to get the highest integrated score even if it takes longer, and is selected as the final delivery route.

[0056] As shown in Figure 3 The present application also relates to a system for logistics distribution based on Internet of Things, comprising a processor and a memory, the memory storing a computer program, the processor being capable of interacting with the memory, calling the computer program (for example, through a bus), and then the processor executing the computer program, when the computer program is executed by the processor, realizing the method for logistics distribution based on Internet of Things of the above-mentioned embodiments.

[0057] Those skilled in the art will think of many changes, alterations and alternatives without departing from the spirit and principles of the present application. It should be understood that various alternatives to the embodiments of the present application described herein can be employed in practicing the application.

Claims

1. A method of logistics distribution based on the Internet of Things, characterized in that: The following steps are involved: Obtain information about orders to be delivered, real-time status data of vehicles, road network topology of the target delivery area, and historical traffic flow data; Based on the road network topology and historical traffic flow data, a spatiotemporal prediction model integrating a graph convolutional network and a gated recurrent unit is used to calculate the predicted travel time cost of each road segment in the road network in multiple future time windows. Based on historical delivery data and time sensitivity, a Bayesian inference model is used to generate a discrete probability distribution of successful delivery for each order within the multiple time windows, including: Based on historical delivery data, a priori probability of successful receipt is constructed, which is related to the order's destination type and delivery time. A likelihood function is established based on the order's time sensitivity level, where the higher the time sensitivity level, the higher the likelihood of receipt within the order's required time window. The Bayesian formula is used to integrate the prior probability of successful receipt and the likelihood function to obtain the posterior probability of successful receipt for the order in each future time window. Combining the vehicle's real-time status data with the vehicle's endurance, load limit, and health status index, a feasibility constraint set for the vehicle's mission execution is constructed; Taking the predicted travel time cost, the probability distribution of successful receipt, and the feasibility constraint set as input, a multi-objective optimization algorithm is applied, with minimizing the total predicted travel time cost of all vehicles and maximizing the total successful receipt probability of all orders as independent optimization objectives, to obtain a set of Pareto optimal delivery solutions containing multiple non-inferior solutions; According to the preset risk preference or cost-effectiveness strategy, the final delivery route is selected from the Pareto optimal delivery solution set and sent to the vehicle terminal for execution.

2. The method according to claim 1, characterized in that The spatiotemporal prediction model that integrates graph convolutional networks and gated recurrent units calculates the predicted travel time cost of each road segment in the road network in multiple future time windows, including: The road network topology is abstracted into a graph, with intersections as nodes and road sections as edges. The spatial dependency features of each road section are extracted using a graph convolutional network. The traffic flow data of each road section within a preset historical period is constructed into a time series and input into a gated recurrent unit to extract the temporal variation pattern. The spatial dependency features are integrated with the temporal variation pattern to output the predicted travel time cost for each time window in the future prediction domain with a preset duration as the unit.

3. The method according to claim 1, characterized in that The feasibility constraint set for constructing the vehicle mission execution includes: The cruising range constraint is set to ensure that the total estimated energy consumption of the vehicle's planned route does not exceed the currently available energy; the load limit constraint is set to ensure that the total weight and total volume of the orders carried by the vehicle do not exceed the vehicle's rated load and rated capacity; and the health status constraint is set to ensure that the vehicle's health status index must be higher than the preset threshold.

4. The method according to claim 1, wherein The multi-objective optimization algorithm is applied to obtain a set of Pareto optimal distribution solutions containing multiple non-inferior solutions, including: A multi-objective evolutionary algorithm based on non-dominated sorting is used to encode potential delivery plans. The first objective function is set as the sum of the predicted travel time costs of all vehicle routes and is minimized. The second objective function is set as the sum of the successful sign-in probabilities of all orders at their respective delivery time points and is maximized. The Pareto front solution set is generated by iteratively performing selection, crossover and mutation operations.

5. The method according to claim 1, wherein The selecting of the final delivery route from the Pareto optimal delivery solution set according to the preset risk preference or cost-effectiveness strategy includes: For each plan in the Pareto optimal delivery plan set, its total predicted travel time cost and total successful sign-off probability are normalized; according to the preset risk preference or cost-effectiveness strategy, corresponding weight coefficients are assigned to the normalized total predicted travel time cost and total successful sign-off probability; based on the weight coefficients, a weighted comprehensive score is calculated for each plan, and the plan with the best comprehensive score is selected as the final delivery route.

6. The method according to claim 1, characterized in that The order information includes order number, cargo weight, volume, delivery address and customer expected time.

7. The method according to claim 4, characterized in that The multi-objective evolutionary algorithm based on non-dominated sorting is the NSGA-II algorithm.

8. The method according to claim 5, characterized in that The risk appetite includes a service priority strategy.

9. A logistics distribution system based on the Internet of Things, characterized in that: The system comprises a processor and a memory, wherein the memory stores a computer program and the processor executes the computer program to implement the method for logistics distribution based on the Internet of Things as described in any one of claims 1 to 8.

Images