Logistics vehicle cluster scheduling optimization method and system based on spatiotemporal data mining

Abstract

The application is suitable for the field of vehicle scheduling technology, and particularly relates to a logistics vehicle cluster scheduling optimization method and system based on space-time data mining, which comprises the following steps: positioning all logistics sites in a target area, collecting Bluetooth signals within a preset range by using Bluetooth devices pre-deployed in the logistics sites, reading unique identifiers, establishing a corresponding relationship between the unique identifiers and terminal objects, creating a rule set composed of several evaluation rules, and selecting high-load sites; finding logistics vehicles in the target area, integrating to generate multiple vehicle clusters, and setting participable factors. The application can effectively guarantee that the scheduling result tends to be locally optimal by uniformly quantifying multi-source data and eliminating subjective bias, can realize vehicle scheduling among different vehicle clusters by creating a generalization model, can realize collaborative optimal operation of the entire logistics network, and can significantly improve the overall operation efficiency and operation reliability of the logistics sites.

Description

Technical Field

[0001] This invention relates to the field of vehicle scheduling technology, and in particular to a method and system for optimizing logistics vehicle cluster scheduling based on spatiotemporal data mining. Background Technology

[0002] In the process of logistics transportation scheduling, existing technologies usually adopt a relatively simple decision-making mechanism, that is, using spatial distance as the sole basis for scheduling. For example, when a logistics station experiences a warehouse overload due to a surge in orders or abnormal fluctuations, the nearest logistics vehicle is usually selected directly to provide support. Although this method is simple and has a fast response time, it essentially ignores the mining of spatiotemporal data and real-world conditions. For example, it does not comprehensively consider the vehicle's current task load, the driver's working status (whether fatigued), the vehicle's energy consumption or battery level, the delay risk of existing delivery routes, and the real-time road congestion. As a result, the "nearest vehicle" that is scheduled may not be the optimal choice in actual execution, and may even cause new scheduling conflicts, making the overall scheduling process chaotic.

[0003] Therefore, "how to integrate spatiotemporal data mining results in the process of logistics vehicle cluster scheduling" is the technical problem that this invention needs to solve. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for optimizing logistics vehicle cluster scheduling based on spatiotemporal data mining, so as to solve the problem of "how to integrate spatiotemporal data mining results in the process of logistics vehicle cluster scheduling" mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for optimizing logistics vehicle cluster scheduling based on spatiotemporal data mining, the method comprising: Locate all logistics stations within the target area, use Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, read unique identifiers, establish the correspondence between unique identifiers and terminal objects, create a rule set consisting of several evaluation rules, and select high-load stations. Locate logistics vehicles within the target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and issue them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model. The model parameters in the decision scoring model are extracted, and the model parameters corresponding to all vehicle clusters are weighted and averaged to obtain a unified coefficient, which is then written into the initialized decision scoring model to generate a generalized model. When a new high-load site is identified, real-time values ​​of the factors that can participate are collected and input into the generalization model. The output is a score for each logistics vehicle. The logistics vehicle with the highest score is defined as the execution vehicle. It is determined whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site. If so, the scheduling task is issued to the execution vehicle. If not, the real-time value is written into the decision scoring model corresponding to the high-load site, and the execution vehicle is corrected.

[0006] Furthermore, the step of establishing the correspondence between unique identifiers and terminal objects includes: Collect attribute data of the terminal object, wherein the attribute data includes at least: type and online duration; Establish a communication link between the Bluetooth device and the terminal object, insert abnormal rules into the rule set, where each abnormal rule corresponds to an emergency response plan. When the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

[0007] Furthermore, the steps of selecting the target vehicle, creating a scheduling task for the high-load station, and issuing it to the target vehicle include: Define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and send it to the preset terminals; The available factors are written into the selection pop-up window, the selection results uploaded by the preset terminal are received, and the target vehicle is defined.

[0008] Furthermore, the step of weighted averaging of the model parameters corresponding to all vehicle clusters to obtain unified coefficients, and then writing them into the initialized decision scoring model to generate a generalized model, includes: A federated learning binary tree is constructed with the generalization model as the parent node and the decision scoring model as the child node. Establish a linkage mechanism between parent and child nodes and integrate it into the federated learning binary tree.

[0009] Furthermore, the real-time values ​​of the collected factors are input into the generalization model, and the output is a score for each logistics vehicle. The logistics vehicle with the highest score is defined as the executing vehicle. The steps include: All logistics vehicles are sorted in descending order of scores to generate a queue, and the queue is updated at a preset frequency. The logistics vehicle ranked first in the queue is defined as the execution vehicle. A preset number of logistics vehicles are selected from the front of the queue and marked as standby vehicles.

[0010] Furthermore, the step of issuing the scheduling task to the executing vehicle includes: Obtain access to the speakers in the vehicle, read the text information in the dispatch task, and broadcast it via voice. Using pre-deployed audio pickup devices in the execution vehicles, the feedback information uploaded by the execution vehicles is collected and semantic recognition is performed to determine whether to reselect an execution vehicle.

[0011] Furthermore, the system includes: The positioning module is used to locate all logistics stations within the target area. It uses Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, read unique identifiers, establish the correspondence between unique identifiers and terminal objects, create a rule set consisting of several evaluation rules, and select high-load stations. The training module is used to locate logistics vehicles within a target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and send them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model. The writing module is used to extract the model parameters in the decision scoring model, perform a weighted average of the model parameters corresponding to all vehicle clusters to obtain a unified coefficient, and write it into the initialized decision scoring model to generate a generalized model. The judgment module is used to collect real-time values ​​of participating factors when a new high-load site is identified, input them into the generalization model, output the score of each logistics vehicle, define the logistics vehicle with the highest score as the execution vehicle, determine whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site, if so, the scheduling task is issued to the execution vehicle, if not, the real-time value is written into the decision scoring model corresponding to the high-load site, and the execution vehicle is corrected.

[0012] Furthermore, the positioning module includes: A data acquisition unit is used to collect attribute data of a terminal object, wherein the attribute data includes at least: type and online duration; The activation unit is used to establish a communication link between the Bluetooth device and the terminal object, and to insert abnormal rules into the rule set. Each abnormal rule corresponds to an emergency response plan. When the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

[0013] Furthermore, the training module includes: The distribution unit is used to define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and distribute it to the preset terminal. The receiving unit is used to write the participating factors into the selection pop-up window, receive the selection results uploaded by the preset terminal, and define the target vehicle.

[0014] Furthermore, the writing module includes: The building unit is used to construct a federated learning binary tree with the generalization model as the parent node and the decision scoring model as the child node. Integration units are used to establish a linkage mechanism between parent and child nodes and integrate them into the federated learning binary tree.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By collecting Bluetooth signals, the operational status of each logistics station can be dynamically monitored through low-cost spatiotemporal sensing, quantifying the demand for logistics vehicle scheduling and effectively enabling rapid response to sudden high-load stations. By setting participatory factors and manually identifying target vehicles, the shortcomings of single indicators can be avoided, providing a data foundation for model training and improving vehicle scheduling efficiency. By creating a decision scoring model, the results of spatiotemporal data mining can be fully integrated to quickly determine the most suitable logistics vehicle for performing scheduling tasks. By uniformly quantifying multi-source data, subjective biases can be eliminated, effectively ensuring that scheduling results tend towards local optima. By creating a generalized model, vehicle scheduling can be achieved across different vehicle clusters, realizing the coordinated optimal operation of the entire logistics network and significantly improving the overall operational efficiency and reliability of logistics stations. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention; Figure 2 This is a first sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in an embodiment of the present invention. Figure 3 This is a second sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in an embodiment of the present invention. Figure 4 This is a flowchart of the third sub-process of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in an embodiment of the present invention. Figure 5 This is a flowchart of the fourth sub-process of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in an embodiment of the present invention. Figure 6A block diagram of the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in this embodiment of the invention; Figure 7 A block diagram of the positioning module in a logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention; Figure 8 A block diagram showing the composition of the training module in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention; Figure 9 A block diagram of the writing module in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention; Figure 10 This is a block diagram of the judgment module in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In Example 1, Figure 1 The implementation flow of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention is illustrated below, and is described in detail below: S100: Locates all logistics stations within the target area, uses Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, reads unique identifiers, establishes the correspondence between unique identifiers and terminal objects, creates a rule set consisting of several evaluation rules, and selects high-load stations.

[0019] The process involves identifying all logistics stations within a target area, which can be a city or a district. Logistics stations refer to delivery stations or cargo transfer stations, etc. Bluetooth devices are installed within each logistics station, or at key entrances / exits and loading / unloading areas. These devices periodically scan for Bluetooth signals within a preset range. When a device captures a Bluetooth signal from a nearby terminal device, it parses the unique identifier (e.g., MAC address) of that terminal device, identifying the terminal object corresponding to each unique identifier. This terminal object can be a driver's mobile phone, an in-vehicle Bluetooth device, or a handheld terminal. By tracking and analyzing the appearance, persistence, and disappearance of these unique identifiers, the activity trajectory of the terminal objects within the logistics station is determined, enabling dynamic perception of data such as personnel flow, vehicle entry / exit behavior, and dwell time. Based on the perception results, evaluation rules are created. These rules primarily characterize the load status of the logistics stations. The set of evaluation rules (the rule set) maps the Bluetooth device's perception results to various operational indicators and compares them with the evaluation rules within the rule set to identify high-load stations. The definition and evaluation rules for high-load sites are formulated by the managers of the logistics sites. Since the parcel handling capacity of each logistics site varies, different logistics sites correspond to different evaluation rules.

[0020] For example, one evaluation rule states that if the Bluetooth signals of all terminal devices within a logistics station exhibit characteristics such as continuous location changes and significant signal strength fluctuations, and the duration of continuous location changes exceeds a threshold, it is determined that the staff and vehicles within the logistics station are in a high-frequency movement state, indicating frequent personnel flow or vehicle traffic and a high load within the station. Another evaluation rule states that if the number of currently scanned active terminals is 80 (approximately 50 personnel terminals and 30 vehicle terminals), while the historical average for the same period is 70, it indicates that the station is under normal load.

[0021] S200: Locate logistics vehicles within the target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and issue them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model.

[0022] All logistics vehicles within the target area are located and divided into multiple vehicle clusters based on logistics planning and spatial distribution characteristics. In practice, each vehicle cluster is primarily responsible for the daily collection, delivery, and transfer tasks of a corresponding express delivery station. The factors determining the participation criteria for each logistics vehicle are identified, including: the distance between the logistics vehicle and the high-load station, the driver's fatigue level, and the risk of delays in the original task (i.e., the impact on existing transportation tasks if the logistics vehicle is temporarily reassigned). Each logistics vehicle and its corresponding participation criteria are integrated, and the integrated information is sent to the logistics station management personnel. The management personnel then select the target vehicle, which is the most suitable vehicle to go to the high-load station to perform support tasks.

[0023] Based on the actual operating status and the scale of pending parcels at high-load stations, a dispatch task is created. This task should include target station information, cargo size, priority, and timeliness requirements. The task is then sent to the target vehicle. After the task is completed, feedback from the target vehicle driver is obtained. Combined with the operating status of potentially affected stations, the task impact is determined. Task impact refers to the adjustment of the original service plan caused by the target vehicle's dispatch, thereby affecting the operational progress and transportation timeliness of other related logistics stations. Task impact includes: the degree of deviation from the original route, estimated delay time, and the cascading impact on the service capacity of other stations. Logistics station managers score the dispatch task based on its impact, resulting in a task score. The task score primarily measures the disturbance of the dispatch task to related stations; a higher score indicates less disturbance to the overall logistics distribution caused by the target vehicle's dispatch.

[0024] By utilizing deep learning algorithms, a decision-scoring model is constructed. This model quantifies the global impact of scheduling tasks based on available factors. After completing multiple scheduling tasks, the available factors and task scores from each task are integrated to generate a training set. This training set is then used to train the decision-scoring model, which possesses self-learning capabilities and can be continuously trained and optimized, gradually approaching the optimal decision. This ensures that while the target vehicle completes the parcel processing tasks at high-load stations, the impact on overall logistics is minimized. Each logistics station corresponds to one decision-scoring model, and each logistics station also corresponds to one vehicle cluster; in other words, each vehicle cluster corresponds to one decision-scoring model.

[0025] S300: Extract the model parameters from the decision scoring model, perform a weighted average of the model parameters corresponding to all vehicle clusters to obtain a unified coefficient, and write it into the initialized decision scoring model to generate a generalized model.

[0026] Model parameters are extracted from each decision scoring model, including weight coefficients, bias terms, or feature mapping coefficients. Each vehicle cluster corresponds to a set of model coefficients. Based on the data scale, number of vehicles, and task completion quality of each vehicle cluster, a weight value is assigned to each cluster. The model coefficients of all vehicle clusters are then weighted and averaged to obtain a unique set of unified coefficients. These unified coefficients are deployed into the initialized decision scoring model to obtain a generalized model. The unified coefficients comprehensively reflect the decision-making experience of different vehicle clusters under different scenarios and load conditions. The advantage of this approach is that by effectively integrating the local models (the decision scoring models corresponding to each vehicle cluster), it can not only eliminate the bias caused by the limited samples or local environmental differences of a single vehicle cluster, but also demonstrate global stability through the generalized model. Simply put, the decision scoring model can select who is most suitable to perform the scheduling task, while the generalized model can not only identify the optimal execution object, but also determine who, after performing the scheduling task, causes the least interference to other logistics stations. By constructing a generalized model, it is possible not only to select execution vehicles from the vehicle clusters corresponding to high-load sites, but also from other vehicle clusters, thus expanding the scheduling scope from "local optimum" to "global optimum" and greatly improving scheduling efficiency.

[0027] S400: When a new high-load site is identified, collect the real-time values ​​of the factors that can participate and input them into the generalization model. Output the score of each logistics vehicle and define the logistics vehicle with the highest score as the execution vehicle. Determine whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site. If so, send the scheduling task to the execution vehicle. If not, write the real-time value into the decision scoring model corresponding to the high-load site and correct the execution vehicle.

[0028] After identifying high-load sites, real-time values ​​of the factors that can be used by logistics vehicles in the corresponding vehicle cluster are collected. These real-time values ​​are input into a generalization model, which outputs a score representing the suitability of each logistics vehicle. Each logistics vehicle has a corresponding score, and they are sorted in descending order of score. The logistics vehicle with the highest score is selected as the execution vehicle. It is then determined whether the execution vehicle is within the vehicle cluster corresponding to the high-load site. If it is, it means that the execution vehicle is already in the service system of the high-load site, and its handling of scheduling tasks will not cause significant fluctuations in logistics services. In this case, the scheduling task is directly sent to the execution vehicle. Conversely, if the execution vehicle is not in the service system of the high-load site, the real-time values ​​are re-input into the decision scoring model corresponding to the high-load site to determine the execution vehicle again. The execution vehicles determined by the generalization model and the decision scoring model are integrated and sent to the management personnel, who then determine the final execution vehicle to complete the scheduling task.

[0029] In Example 2, Figure 2 The first sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided by an embodiment of the present invention is shown. The following details the step of establishing the correspondence between unique identifiers and terminal objects: S101: Collect attribute data of the terminal object, wherein the attribute data includes at least: type and online duration.

[0030] Collect attribute data of the terminal object, including: the type of the terminal object (driver's mobile phone, vehicle Bluetooth device and handheld terminal), the online duration of the terminal object, etc. The online duration refers to the continuous duration for which the terminal object is within the scanning range of the Bluetooth device.

[0031] S102: Establish a communication link between the Bluetooth device and the terminal object, insert abnormal rules into the rule set, wherein each abnormal rule corresponds to an emergency response plan, and when the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

[0032] Establish a communication link between the Bluetooth device and the terminal object, create exception rules, where each exception rule corresponds to an emergency response plan, and integrate the exception rules into a rule set. For example, an exception rule might be: if the driver's mobile phone is online for more than 5 hours, activate the alarm mechanism and send a mandatory rest SMS reminder to the driver's mobile phone.

[0033] In Example 3, Figure 3 The second sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention is shown. The steps of selecting target vehicles, creating scheduling tasks for high-load stations, and issuing them to target vehicles are described in detail below: S201: Define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and send it to the preset terminal.

[0034] The vehicle clusters corresponding to high-load sites are defined as candidate clusters. The candidate clusters and corresponding participating factors are integrated to generate a selection pop-up window, which is then sent to the preset terminal, where the preset terminal refers to the mobile terminal of the logistics site manager.

[0035] S202: Write the participating factors into the selection pop-up window, receive the selection results uploaded by the preset terminal, and define the target vehicle.

[0036] In addition to listing each logistics vehicle in the candidate cluster, the selection pop-up also includes the factors that allow each logistics vehicle to participate. Managers can select the logistics vehicle to perform the scheduling task through the selection pop-up to obtain the target vehicle.

[0037] In Example 4, Figure 4 The third sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention is shown. The following details the step of weighted averaging of the model parameters corresponding to all vehicle clusters to obtain unified coefficients, and writing them into the initialized decision scoring model to generate a generalized model: S301: Construct a federated learning binary tree with the generalization model as the parent node and the decision scoring model as the child node.

[0038] Create a parent node corresponding to the generalization model and child nodes corresponding one-to-one with the decision scoring model. Both parent and child nodes are logical nodes, not actual processing devices; they are only used to represent the relationship between the generalization and decision scoring models. Attach child nodes to parent nodes to construct a federated learning binary tree. A parent node can have multiple child nodes. This can be simply understood as multiple vehicle clusters sharing a single generalization model, while each vehicle cluster corresponds to a decision scoring model. The generalization model has stronger global capabilities and uses its output to execute scheduling tasks, having less impact on the overall logistics. The decision scoring model, on the other hand, uses its output to execute scheduling tasks, resulting in higher processing efficiency.

[0039] S302: Establish a linkage mechanism between parent and child nodes and integrate it into the federated learning binary tree.

[0040] Federated learning binary trees are similar to binary trees in existing technologies, but with the difference that there is a linkage mechanism between parent and child nodes. The linkage mechanism is that when the model parameters of a child node are updated, the model parameters of the parent node will be adjusted synchronously through federated learning.

[0041] In Example 5, Figure 5 The fourth sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention is shown below. The following details the steps of collecting real-time values ​​of participating factors, inputting them into the generalization model, outputting the score of each logistics vehicle, and defining the logistics vehicle with the highest score as the executing vehicle: S401: Sort all logistics vehicles in descending order of scores, generate a queue, and update the queue at a preset frequency.

[0042] After determining the score for each logistics vehicle, the vehicles are sorted in descending order of their scores to create a queue. The queue is then updated at a preset frequency, which can be once per hour or once per day.

[0043] S402: Define the logistics vehicle ranked first in the queue as the execution vehicle, select a preset number of logistics vehicles from the front of the queue, and mark them as standby vehicles.

[0044] The logistics vehicle at the top of the queue is defined as the executing vehicle. From the front of the queue, a predetermined number of logistics vehicles, excluding the executing vehicle, are selected as backup vehicles. For example, the logistics vehicle in the first position of the queue is defined as the executing vehicle, and the logistics vehicles in positions 2-5 are defined as backup vehicles. When the executing vehicles cannot meet the processing needs of a high-load site, scheduling tasks can be issued to the backup vehicles as appropriate.

[0045] In Example 6, Figure 5 The diagram illustrates the fourth sub-process flowchart of the logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining provided in this embodiment of the invention. The following details the step of issuing the scheduling task to the executing vehicle: S403: Obtain access to the speakers in the vehicle, read text information from the dispatch task, and broadcast it via voice.

[0046] The system obtains access to the loudspeakers in the execution vehicle. When the execution vehicle receives a dispatch task, it parses the text information and inputs the parsed text content into the speech synthesis module. The text information is then processed into speech to generate corresponding speech data, which is then played through the loudspeakers.

[0047] S404: Using pre-deployed audio pickup devices in the execution vehicle, collect feedback information uploaded by the execution vehicle, perform semantic recognition, and determine whether to reselect the execution vehicle.

[0048] After the speaker finishes playing, the voice information provided by the driver is continuously collected using a pre-deployed sound pickup device (such as an in-vehicle microphone or voice interaction terminal) inside the vehicle. The feedback information is then semantically recognized to determine whether the driver is willing to accept the dispatch task. If not, a new vehicle is selected, and the logistics vehicle ranked second in the queue is defined as the vehicle to be dispatched.

[0049] Figure 6 This diagram illustrates the structural block diagram of a logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention. The logistics vehicle cluster scheduling optimization system 1 based on spatiotemporal data mining includes: The positioning module 11 is used to locate all logistics stations within the target area. It uses Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, read unique identifiers, establish a correspondence between unique identifiers and terminal objects, create a rule set consisting of several evaluation rules, and select high-load stations. Training module 12 is used to find logistics vehicles in the target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and send them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model. The writing module 13 is used to extract the model parameters in the decision scoring model, perform a weighted average of the model parameters corresponding to all vehicle clusters to obtain a unified coefficient, and write it into the initialized decision scoring model to generate a generalized model. The judgment module 14 is used to collect real-time values ​​of the factors that can participate when a new high-load site is identified, input them into the generalization model, output the score of each logistics vehicle, define the logistics vehicle with the highest score as the execution vehicle, determine whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site, if so, send the scheduling task to the execution vehicle, if not, write the real-time value into the decision scoring model corresponding to the high-load site, and correct the execution vehicle.

[0050] Figure 7 This diagram illustrates the composition of the positioning module 11 in a logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining, as provided in an embodiment of the present invention. The positioning module 11 includes: The acquisition unit 111 is used to acquire attribute data of the terminal object, wherein the attribute data includes at least: type and online duration; The activation unit 112 is used to establish a communication link between the Bluetooth device and the terminal object, and to insert abnormal rules into the rule set. Each abnormal rule corresponds to an emergency response plan. When the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

[0051] Figure 8 This diagram illustrates the structural composition of the training module 12 in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention. The training module 12 includes: The distribution unit 121 is used to define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and distribute it to the preset terminal. The receiving unit 122 is used to write the participating factors into the selection pop-up window, receive the selection results uploaded by the preset terminal, and define the target vehicle.

[0052] Figure 9This diagram illustrates the structural block diagram of the writing module 13 in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention. The writing module 13 includes: Building unit 131 is used to construct a federated learning binary tree with the generalization model as the parent node and the decision scoring model as the child node. Integration unit 132 is used to establish a linkage mechanism between parent nodes and child nodes and integrate it into the federated learning binary tree.

[0053] Figure 10 This diagram illustrates the structural composition of the judgment module 14 in the logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining provided in an embodiment of the present invention. The judgment module 14 includes: The update unit 141 is used to sort all logistics vehicles in descending order of scores, generate a queue, and update the queue at a preset frequency. The marking unit 142 is used to define the logistics vehicle ranked first in the queue as the execution vehicle, select a preset number of logistics vehicles from the front of the queue, and mark them as spare vehicles. The broadcasting unit 143 is used to obtain the access permission of the loudspeakers in the vehicle, read the text information in the scheduling task, and broadcast it by voice. The identification unit 144 is used to collect feedback information uploaded by the execution vehicle using a pickup device pre-deployed in the execution vehicle, and to perform semantic recognition to determine whether to reselect the execution vehicle.

[0054] The positioning module 11 is mainly used to complete step S100, the training module 12 is mainly used to complete step S200, the writing module 13 is mainly used to complete step S300, and the judgment module 14 is mainly used to complete step S400. The acquisition unit 111 is mainly used to complete step S101, and the activation unit 112 is mainly used to complete step S102. The sending unit 121 is mainly used to complete step S201, and the receiving unit 122 is mainly used to complete step S202; The building unit 131 is mainly used to complete step S301, and the integration unit 132 is mainly used to complete step S302. The update unit 141 is mainly used to complete step S401, the marking unit 142 is mainly used to complete step S402, the broadcasting unit 143 is mainly used to complete step S403, and the identification unit 144 is mainly used to complete step S404.

[0055] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0056] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing logistics vehicle cluster scheduling based on spatiotemporal data mining, characterized in that, The method includes: Locate all logistics stations within the target area, use Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, read unique identifiers, establish the correspondence between unique identifiers and terminal objects, create a rule set consisting of several evaluation rules, and select high-load stations. Locate logistics vehicles within the target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and issue them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model. The model parameters in the decision scoring model are extracted, and the model parameters corresponding to all vehicle clusters are weighted and averaged to obtain a unified coefficient, which is then written into the initialized decision scoring model to generate a generalized model. When a new high-load site is identified, real-time values ​​of the factors that can participate are collected and input into the generalization model. The output is a score for each logistics vehicle. The logistics vehicle with the highest score is defined as the execution vehicle. It is determined whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site. If so, the scheduling task is issued to the execution vehicle. If not, the real-time value is written into the decision scoring model corresponding to the high-load site, and the execution vehicle is corrected.

2. The logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining according to claim 1, characterized in that, The steps for establishing the correspondence between unique identifiers and terminal objects include: Collect attribute data of the terminal object, wherein the attribute data includes at least: type and online duration; Establish a communication link between the Bluetooth device and the terminal object, insert abnormal rules into the rule set, where each abnormal rule corresponds to an emergency response plan. When the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

3. The logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining according to claim 1, characterized in that, The steps of selecting the target vehicle, creating a scheduling task for the high-load station, and sending it to the target vehicle include: Define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and send it to the preset terminals; The available factors are written into the selection pop-up window, the selection results uploaded by the preset terminal are received, and the target vehicle is defined.

4. The logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining according to claim 1, characterized in that, The step of generating a generalized model by weighted averaging of the model parameters corresponding to all vehicle clusters to obtain uniform coefficients and writing them into the initialized decision scoring model includes: A federated learning binary tree is constructed with the generalization model as the parent node and the decision scoring model as the child node. Establish a linkage mechanism between parent and child nodes and integrate it into the federated learning binary tree.

5. The logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining according to claim 3, characterized in that, The real-time values ​​of the collected factors are input into the generalization model, and the output is a score for each logistics vehicle. The logistics vehicle with the highest score is defined as the executing vehicle. The steps include: All logistics vehicles are sorted in descending order of scores to generate a queue, and the queue is updated at a preset frequency. The logistics vehicle ranked first in the queue is defined as the execution vehicle. A preset number of logistics vehicles are selected from the front of the queue and marked as standby vehicles.

6. The logistics vehicle cluster scheduling optimization method based on spatiotemporal data mining according to claim 5, characterized in that, The step of sending the scheduling task to the vehicle to perform it includes: Obtain access to the speakers in the vehicle, read the text information in the dispatch task, and broadcast it via voice. Using pre-deployed audio pickup devices in the execution vehicles, the feedback information uploaded by the execution vehicles is collected and semantic recognition is performed to determine whether to reselect an execution vehicle.

7. A logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining, characterized in that, The system includes: The positioning module is used to locate all logistics stations within the target area. It uses Bluetooth devices pre-deployed in the logistics stations to collect Bluetooth signals within a preset range, read unique identifiers, establish the correspondence between unique identifiers and terminal objects, create a rule set consisting of several evaluation rules, and select high-load stations. The training module is used to locate logistics vehicles within a target area, integrate and generate multiple vehicle clusters, set participation factors, wherein the participation factors include at least: distance, personnel status and original task delay risk, select target vehicles, create scheduling tasks for high-load stations and send them to target vehicles, obtain task impact, receive task scores uploaded by preset terminals, create a decision scoring model corresponding to each vehicle cluster, integrate the participation factors and task scores of target vehicles, generate a training set, and train the decision scoring model. The writing module is used to extract the model parameters in the decision scoring model, perform a weighted average of the model parameters corresponding to all vehicle clusters to obtain a unified coefficient, and write it into the initialized decision scoring model to generate a generalized model. The judgment module is used to collect real-time values ​​of participating factors when a new high-load site is identified, input them into the generalization model, output the score of each logistics vehicle, define the logistics vehicle with the highest score as the execution vehicle, determine whether the execution vehicle is located in the vehicle cluster corresponding to the high-load site, if so, the scheduling task is issued to the execution vehicle, if not, the real-time value is written into the decision scoring model corresponding to the high-load site, and the execution vehicle is corrected.

8. The logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining according to claim 7, characterized in that, The positioning module includes: A data acquisition unit is used to collect attribute data of a terminal object, wherein the attribute data includes at least: type and online duration; The activation unit is used to establish a communication link between the Bluetooth device and the terminal object, and to insert abnormal rules into the rule set. Each abnormal rule corresponds to an emergency response plan. When the attribute data meets the abnormal rule, the corresponding emergency response plan is activated.

9. The logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining according to claim 7, characterized in that, The training module includes: The distribution unit is used to define the vehicle clusters corresponding to high-load sites as candidate clusters, create a selection pop-up window consisting of logistics vehicles in the candidate clusters, and distribute it to the preset terminal. The receiving unit is used to write the participating factors into the selection pop-up window, receive the selection results uploaded by the preset terminal, and define the target vehicle.

10. The logistics vehicle cluster scheduling optimization system based on spatiotemporal data mining according to claim 7, characterized in that, The writing module includes: The building unit is used to construct a federated learning binary tree with the generalization model as the parent node and the decision scoring model as the child node. Integration units are used to establish a linkage mechanism between parent and child nodes and integrate them into the federated learning binary tree.

Images