Big data-based dynamic route optimization system for urban joint distribution

Abstract

This invention relates to a dynamic route optimization system for urban shared delivery based on big data, comprising: a data acquisition module for collecting vehicle and driving data; a graph construction module for extracting experiential knowledge triples, establishing a preliminary universal route graph, and performing personalized corrections; an interactive feedback module for personalized adjustments during the personalized correction process; a group graph module for constructing a group universal route graph; a parameter de-emphasis module for de-emphasizing long-term inapplicable individual ability deviation coefficients during the personalized adjustment process; and a conflict verification module for multi-dimensional verification of the personalized adjustment process and decision optimization based on driving data. This system helps resolve the contradiction between system collaboration and data silos, addresses the disconnect between intelligent algorithms and human experience, and maintains the current relevance of decisions.

Description

Technical Field

[0001] This invention relates to the field of logistics and distribution technology, and in particular to a dynamic route optimization system for urban shared distribution based on big data. Background Technology

[0002] The urban shared delivery sector faces three major challenges: First, the contradiction between static planning and dynamic reality. Existing route optimization is mostly based on historical data, making it difficult to respond in real time to dynamic information such as traffic congestion and order changes. Second, the contradiction between system collaboration and data silos. Different companies cannot share data due to privacy and interest barriers, resulting in "pseudo-shared" delivery and making global optimization difficult to achieve. Third, the disconnect between intelligent algorithms and human experience. Algorithm models are mostly "black boxes," unable to absorb individual driver experience and neglecting personalization, leading to untrusted decisions. At the same time, traditional centralized data fusion methods touch on privacy and security red lines, severely restricting the development of cross-organizational intelligent collaboration.

[0003] Chinese Patent Publication No. CN120806785A discloses a method and system for analyzing cold chain food delivery based on big data. The method includes: dynamically extracting features from multi-source monitoring data of cold chain food delivery vehicles to generate transportation status parameters; correlating and matching the transportation status parameters with real-time environmental data of the cold chain food delivery vehicles to obtain environmental impact assessment results; performing delivery efficiency analysis on the vehicle route transportation data of the cold chain food delivery vehicles based on the environmental impact assessment results to form initial delivery efficiency data; and performing road condition diagnosis on the route transportation data, combined with the environmental impact assessment results. The evaluation results are used to predict the route and generate a route prediction result; the initial delivery efficiency data and the route prediction result are cross-analyzed to formulate a global delivery plan for the cold chain food delivery vehicle; it can be seen that the Chinese patent has the following problems: the contradiction between system collaboration and data silos, the inability of different enterprises to share data due to privacy and interest barriers, the disconnect between intelligent algorithms and human experience leading to the system's inability to adapt to changes in driver habits and external environment, and the system's misjudgment due to outdated knowledge, making the system unable to flexibly cope with conceptual drift problems such as urban changes and seasonal changes, and making it difficult to maintain the current relevance of decisions. Summary of the Invention

[0004] To address this, the present invention provides a dynamic route optimization system for urban shared delivery based on big data, which overcomes the contradiction between system collaboration and data silos in existing technologies. Different companies cannot share data due to privacy and interest barriers. The disconnect between intelligent algorithms and human experience leads to the system's inability to adapt to changes in driver habits and external environment, and the system's misjudgment due to outdated knowledge. This makes the system unable to flexibly cope with conceptual drift issues such as urban changes and seasonal changes, and makes it difficult to maintain the current relevance of decisions.

[0005] To achieve the above objectives, the present invention provides a dynamic route optimization system for urban shared delivery based on big data, characterized in that it includes:

[0006] The data acquisition module is used to collect vehicle data and driving data, and to perform spatiotemporal alignment on the vehicle data to obtain aligned data.

[0007] The graph construction module is used to extract empirical knowledge triples from the aligned data, establish a primary universal path graph through the empirical knowledge triples, and make personalized corrections to the primary universal path graph to obtain a personalized path graph.

[0008] The interactive feedback module is used to personalize the personalized correction process based on driving data;

[0009] The community graph module is used to optimize the initial universal path graph using federated learning methods to obtain a community universal path graph.

[0010] The parameter fade-out module is used to fade out personal ability offset coefficients that have not been applicable for a long time during the personalization process;

[0011] The conflict verification module is used to perform multi-dimensional verification of the personalized adjustment process based on the personalized route map and driving data, output the final decision, and optimize the final decision based on the driving data.

[0012] The route output module is used to output the final decision as a dynamic delivery route to the mobile terminal.

[0013] Furthermore, when the data acquisition module collects vehicle data and driving data and performs spatiotemporal alignment on the vehicle data, it collects vehicle data based on the Beidou positioning terminal and driving data based on the mobile terminal, and performs spatiotemporal alignment on the vehicle data and driving data. The spatiotemporal alignment method is as follows: the vehicle data and driving data are connected to the network protocol service for clock synchronization, and the coordinate systems of the vehicle data and driving data are identified. Spatial alignment is then performed using the Gauss-Kruger projection transformation algorithm to obtain aligned data.

[0014] Furthermore, the graph construction module extracts empirical knowledge triples based on the alignment data. When establishing a primary universal path graph using these empirical knowledge triples, it also establishes an extraction model. The specific method for establishing the extraction model is as follows:

[0015] Step A01: Define the entity type and triple relationship type, and output the definition process as the data processing layer of the preliminary extraction model. The entity type is set to include driver, vehicle, customer point, time period and route, and the triple relationship type is (head entity, connection relationship, tail entity).

[0016] Step A02: Extract the aligned data according to the entity type and triple relationship type to obtain empirical triples and obtain the context. At the same time, perform feature calculation on the graph confidence and empirical strength, and output the feature calculation process as the feature layer of the preliminary extraction model.

[0017] Step A03: Create an entity graph node for each entity type. Use the context, graph confidence, and experience strength as directed edges, pointing from the head entity to the tail entity, to obtain the experience knowledge triple graph. By combining all experience knowledge triple graphs, a primary universal path graph is obtained, which is used as the output layer of the preliminary extraction model.

[0018] Step A04: Divide the historical data into a 70% training set, a 15% validation set, and a 15% test set. Train the preliminary extraction model on the training set to obtain the trained extraction model. Input the validation set into the trained extraction model for validation to obtain the optimal parameters. Input the optimal parameters into the trained extraction model and train it on the test set to obtain the final extraction model and model accuracy. Output the final extraction model with an accuracy greater than 96%.

[0019] Furthermore, the map construction module performs personalized modifications to the initial universal path map. The specific personalized modification method is as follows:

[0020] Step B01, driver correction: When a driver performs a delivery task, find a group of drivers with similarities to the driver from the primary universal route map, calculate the individual ability offset coefficient λ, calculate the personalized experience intensity based on the individual ability offset coefficient λ, and output the personalized experience intensity as the experience intensity.

[0021] Step B02, Step B02, Modify the scenario: When the context of the empirical knowledge triple does not match the actual situation, reduce the confidence of the graph and generate empirical knowledge triples that match the actual situation and add them to the primary universal path graph to obtain the updated universal graph.

[0022] Step B03: Output the updated universal map as a personalized path map.

[0023] Furthermore, the interactive feedback module adjusts the personalized correction process based on driving data, and the personalized adjustment specifically includes:

[0024] Step C01, Prior Distribution: For new drivers, the global capability offset coefficient λ0 is used as the personal capability offset coefficient λ to replace it. The likelihood function is established when the personal capability offset coefficient λ = the global capability offset coefficient λ0. The noise variance σ0 is obtained, and the prior variance σ1 and prior mean μ1 are calculated according to the distribution of the global capability offset coefficient λ0.

[0025] Step C02: Collect vehicle and driving data for the new delivery task to obtain actual personal experience Gs and general experience Tj;

[0026] Step C03, posterior distribution: Set the posterior mean μ2=[w1×μ1+w2×(Gs / Tj)] / (w1+w2), where w1 is the prior weight and w2 is the actual weight, and the posterior variance σ2=1 / (w1+w2), where: w1=1 / σ1, w2=1 / σ0.

[0027] Furthermore, the interactive feedback module adjusts the personalized correction process based on driving data. This personalized adjustment further includes: step C04, calculating the trust weight coefficient k, setting the trust weight coefficient k = μ1 × (w1 + w2), comparing the posterior variance σ2 with the preset variance σ3, judging the driver's driving situation based on the comparison result, and adjusting the personalized correction process based on the judgment result, wherein:

[0028] When σ2>σ3, the driver's driving situation is judged to be uncertain. The adjusted personal ability deviation coefficient λ1 is calculated and set as μ1×(1-k)+(Gs / Tj)×k. The adjusted personal ability deviation coefficient λ1 is then output as the personal ability deviation coefficient λ.

[0029] When σ2≤σ3, the driver's driving situation is judged to be confident, and no personalized adjustment is made to the personalized correction process.

[0030] When σ2≤σ3, the driver's driving situation is considered to be reliable, and no personalized adjustments are made to the personalized correction process.

[0031] Furthermore, when the group graph module constructs a universal group path graph based on the federated learning method and the personalized path graph, it constructs a universal group path graph. The specific method for constructing the universal group path graph is as follows:

[0032] Step D01: Process the personalized path graph on the mobile terminal: Divide the experience knowledge triple graph with the same triple relation type into triple graph sets, build an initial global model on the central server, and send the model parameters of the initial global model to the mobile terminal.

[0033] Step D02: The mobile terminal calculates the graph confidence Zx for each empirical knowledge triple graph in the triple graph set, compares the graph confidence Zx with the preset graph confidence Zx0, judges the value of the empirical knowledge triple graph based on the comparison result, and establishes a candidate triple set based on the judgment result, wherein:

[0034] When Zx≥Zx0, the value of the empirical knowledge triple graph is determined to be high value. The empirical knowledge triple graph is added to the candidate triple set, and the initial global model is trained based on the candidate triple set to obtain the trained initial global model. The model parameters of the trained initial global model are then uploaded to the central server.

[0035] When Zx < Zx0, the value of the empirical knowledge triplet graph is determined to be low, and the empirical knowledge triplet graph is not added to the candidate triplet set.

[0036] In step D03, the central server merges the model parameters of the initial global model after training using a secure aggregation algorithm to obtain a global model, and then distributes the global model to the mobile terminal for the next round of training. The central server also obtains the graph parameters through the global model and merges the graph parameters according to the clustering algorithm to obtain a universal path graph for the group.

[0037] Furthermore, when the parameter fading module fades out personal ability offset coefficients that have not been applicable for a long time during the personalized adjustment process, the fading process specifically involves:

[0038] During the personalized adjustment process, the usage of the personalized path map is monitored, and when fading is applied based on the monitoring results, a timestamp is added to each knowledge triplet graph in the personalized path map to obtain the current usage time interval T for each knowledge triplet graph. The current usage time interval T for each knowledge triplet graph is compared with a preset time interval T0. Based on the comparison results, the usage of each knowledge triplet graph is judged, and based on the judgment results, each knowledge triplet graph is faded. Specifically:

[0039] When T≥T0, the empirical knowledge triplet graph is determined to be inapplicable for a long time, and the empirical knowledge triplet graph is diluted.

[0040] When T < T0, the application of the experiential knowledge triplet map is deemed applicable, and the personal ability offset coefficient is not downplayed.

[0041] Furthermore, the fade-out process in the parameter fade-out module specifically involves calculating the usage coefficient F of the empirical knowledge triplet graph, and setting the calculation formula for the usage coefficient F as follows: ,in, It is the first The weight used for the second time It is the attenuation coefficient. This represents the total number of uses. The usage coefficient F is compared with F0. Based on the comparison results, the usage of the empirical knowledge triplet graph is judged, and the empirical knowledge triplet graph is diluted based on the judgment results. Among them:

[0042] When F≥F0, the usage of the experience knowledge triplet graph is determined to be that it is not applicable for a long time but has high activity. The personal ability offset coefficient λ of the experience knowledge triplet graph is diluted. The diluted personal ability offset coefficient λ1 is set as λ×(1-β)+μ2×β, where β is the dilution coefficient. The diluted personal ability offset coefficient λ1 is output as the personal ability offset coefficient λ.

[0043] When F < F0, the experience knowledge triplet graph is determined to be inapplicable for a long time and has low activity, and the experience knowledge triplet graph is deleted from the personalized path graph.

[0044] Furthermore, when the conflict verification module performs multi-dimensional verification of the personalization adjustment process based on the personalized path map and driving data, the multi-dimensional verification method specifically includes:

[0045] A rule set is established and matched with a personalized path graph. The matching method is as follows: Personalized triplet relationship types are obtained from the personalized path graph, and the head and tail entities in the personalized triplet relationship types are output as entities. The entities are compared with the rule set, and the personalized path graph is judged based on the comparison results. The individual ability offset coefficient λ is then diluted based on the judgment results.

[0046] When an entity belongs to the rule set, the personalized path graph is determined to be a match. The matched personalized path graph is marked as invalid, and the personal ability offset coefficient is weakened.

[0047] When an entity does not belong to the rule set, the personalized path graph is determined to be a mismatch. The mismatched personalized path graph is marked as valid, and the marked valid personalized path graph is output as the final decision.

[0048] Furthermore, when the conflict verification module optimizes the final decision based on driving data, it obtains the delivery time Th and calculates the delivery time difference Tx based on the delivery time Th. The delivery time difference Tx is set to Th - Tj. The delivery time difference Tx is compared with a preset time difference Tx0. Based on the comparison result, the module judges the situation of the final decision and optimizes the final decision based on the judgment result, wherein:

[0049] When Tx≥Tx0, the final decision is determined to be that no optimization is needed, and no optimization is performed on the final decision.

[0050] When Tx < Tx0, the final decision is determined to require optimization, and driving data is added to the final decision.

[0051] Compared with existing technologies, the beneficial effects of this invention are that it resolves the contradiction between system collaboration and data silos, avoids the inability of different enterprises to share data due to privacy and interest barriers, solves the problem of the disconnect between intelligent algorithms and human experience leading to the system's inability to adapt to changes in driver habits, and avoids the problem of misjudgments caused by changes in the external environment and outdated knowledge, preventing the system from flexibly responding to conceptual drift such as urban changes and seasonal changes, and making it difficult to maintain the current relevance of decisions.

[0052] In particular, the data acquisition module performs spatiotemporal alignment of vehicle data, which helps transform messy raw data into reliable facts, fundamentally avoiding decision-making errors caused by data misalignment. Furthermore, the graph construction module establishes a primary universal path graph based on experiential knowledge triplets and performs personalized corrections to this primary universal path graph. This helps transform drivers' implicit, intuitive experiences into explicit, computable, and storable data, and also improves the feasibility of routes and drivers' trust. The interactive feedback module adjusts the personalized correction process based on driving data, automatically adjusting the logic and parameters of personalized corrections, enabling the system to continuously evolve and adapt to changes in driver habits and the external environment. Finally, the group graph module constructs a group universal path graph and performs personalized corrections to the primary universal path graph. The process of establishing a path graph and optimizing the graph helps to aggregate scattered individual experiences into a collective knowledge base, which in turn benefits all participants and solves the problem of data silos. Through the parameter de-emphasis module, the personal ability deviation coefficient that has been inapplicable for a long time is de-emphasized during the personalized adjustment process. This helps to automatically decay outdated parameters that have not been applicable for a long time, prevent the system from making misjudgments due to outdated knowledge, and enable the system to flexibly respond to conceptual drift problems such as urban changes and seasonal changes, maintaining the current relevance of decisions. Furthermore, the conflict verification module performs multi-dimensional verification of the personalized adjustment process and optimizes the final decision. This helps to conduct real-time quantitative trade-offs and risk simulations among multiple objectives such as efficiency, safety, compliance and personal habits, outputting intelligent and safe final decisions, and transforming conflict cases into nutrients for optimizing the overall system. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the urban shared delivery dynamic route optimization system based on big data in this embodiment. Detailed Implementation

[0054] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0055] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0056] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0057] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0058] Please see Figure 1 As shown, this is a schematic diagram of the urban shared delivery dynamic route optimization system based on big data in this embodiment. The system includes:

[0059] The data acquisition module is used to collect vehicle data and driving data, and to perform spatiotemporal alignment on the vehicle data to obtain aligned data.

[0060] The graph construction module is used to extract empirical knowledge triples from the aligned data, establish a primary universal path graph through the empirical knowledge triples, and perform personalized corrections on the primary universal path graph to obtain a personalized path graph. The graph construction module is connected to the data acquisition module.

[0061] An interactive feedback module is used to personalize the personalized correction process based on driving data. The interactive feedback module is connected to the map construction module.

[0062] The group graph module is used to optimize the initial universal path graph using a federated learning method to obtain a group universal path graph. The group graph module is connected to the interactive feedback module.

[0063] The parameter fade-out module is used to fade out personal ability offset coefficients that have not been applicable for a long time during the personalized adjustment process. The parameter fade-out module is connected to the population map module.

[0064] The conflict verification module is used to perform multi-dimensional verification of the personalized adjustment process based on the personalized path map and driving data, output the final decision, and optimize the final decision based on the driving data. The conflict verification module is connected to the parameter fading module.

[0065] The path output module is used to output the final decision as a dynamic delivery route to the mobile terminal. The path output module is connected to the conflict verification module.

[0066] Specifically, the big data-based urban collaborative delivery dynamic route optimization system is applied to the urban logistics and delivery industry, such as cold chain delivery route planning terminals. The system collects vehicle and driving data, aligns the data to obtain aligned data, extracts experiential knowledge triples from the aligned data, establishes a preliminary universal route map, and then performs personalized corrections to obtain a personalized route map. The personalized correction process involves individual adjustments and the construction of a group universal map using federated learning methods. Further, the personalized adjustments are processed by de-emphasis and multi-dimensional verification to address the contradiction between system collaboration and data silos. This system avoids data sharing barriers between different companies due to privacy and interests, thus resolving the problem of the disconnect between intelligent algorithms and human experience, which prevents the system from adapting to changes in driver habits. It also avoids misjudgments caused by changes in the external environment and outdated knowledge, preventing the system from flexibly responding to conceptual drift issues such as urban changes and seasonal shifts, and maintaining the relevance of decisions in the present moment. In particular, the system uses a data acquisition module to perform spatiotemporal alignment of vehicle data, transforming messy raw data into reliable facts, thereby fundamentally avoiding decision-making errors caused by data misalignment. Furthermore, a graph construction module establishes primary universal paths based on experiential knowledge triplets. The system constructs a generalized path graph and personalizes the initial universal path graph to transform drivers' implicit, intuitive experiences into explicit, computable, and storable data. This improves path feasibility and driver trust. An interactive feedback module adjusts the personalization process based on driving data, automatically adjusting the logic and parameters to allow the system to continuously evolve and adapt to changes in driver habits and the external environment. Furthermore, a group path graph module constructs a generalized group path graph and optimizes the initial universal path graph construction process. This aggregates scattered individual experiences into a group knowledge base, feeding back into all participants. To address the data silo problem, a parameter fading module is used to fade outdated personal ability bias coefficients that have been inapplicable for a long time during the personalized adjustment process. This allows for the automatic decay of outdated parameters that are no longer applicable, preventing misjudgments caused by obsolete knowledge. The system can flexibly respond to conceptual drift issues such as urban changes and seasonal changes, maintaining the current relevance of decisions. Furthermore, a conflict verification module performs multi-dimensional verification of the personalized adjustment process and optimizes the final decision. This allows for real-time quantitative trade-offs and risk simulations among multiple objectives such as efficiency, safety, compliance, and personal habits, outputting intelligent and safe final decisions, and transforming conflict cases into nutrients for optimizing the overall system.

[0067] Specifically, when the data acquisition module collects vehicle data and driving data and performs spatiotemporal alignment on the vehicle data, it collects vehicle data based on the Beidou positioning terminal and driving data based on the mobile terminal, and performs spatiotemporal alignment on the vehicle data and driving data. The spatiotemporal alignment method is as follows: the vehicle data and driving data are connected to the network protocol service for clock synchronization, and the coordinate systems of the vehicle data and driving data are identified. Spatial alignment is then performed using the Gauss-Kruger projection transformation algorithm to obtain aligned data.

[0068] Specifically, the vehicle data refers to vehicle data during vehicle operation, including path trajectory, real-time speed, and abnormal driving behavior. The path trajectory refers to the vehicle's driving trajectory, the real-time speed refers to the vehicle's speed, and the abnormal driving behavior refers to sudden braking and abrupt driving. The driving data refers to the driver's manual deviation from the system-recommended path, textual annotations on specific road segments, and subjective evaluations after task completion. The network protocol service refers to a protocol service used to achieve high-precision clock synchronization in a computer network. Clock synchronization refers to determining a unique time reference and unifying time through the network protocol service. The Gauss-Kruger projection transformation algorithm refers to a precise mathematical method for converting Earth's spherical coordinates into planar map coordinates. Spatial alignment refers to determining a unique reference coordinate system through the Gauss-Kruger projection transformation algorithm. The aligned data refers to vehicle data and driving data after spatiotemporal alignment.

[0069] Specifically, the data acquisition module, by performing spatiotemporal alignment on vehicle data, helps to transform messy raw data into reliable facts, fundamentally avoiding decision-making errors caused by data misalignment.

[0070] Specifically, the graph construction module extracts empirical knowledge triples based on alignment data. When establishing a primary universal path graph using these empirical knowledge triples, the extraction model is built using an extraction model building method. This extraction model building method specifically includes:

[0071] Step A01: Define the entity type and triple relationship type, and output the definition process as the data processing layer of the preliminary extraction model. The entity type is set to include driver, vehicle, customer point, time period and route, and the triple relationship type is (head entity, connection relationship, tail entity).

[0072] Step A02: Extract the aligned data according to the entity type and triple relationship type to obtain empirical triples and obtain the context. At the same time, perform feature calculation on the graph confidence and empirical strength, and output the feature calculation process as the feature layer of the preliminary extraction model.

[0073] Step A03: Create an entity graph node for each entity type. Use the context, graph confidence, and experience strength as directed edges, pointing from the head entity to the tail entity, to obtain the experience knowledge triple graph. By combining all experience knowledge triple graphs, a primary universal path graph is obtained, which is used as the output layer of the preliminary extraction model.

[0074] Step A04: Divide the historical data into a 70% training set, a 15% validation set, and a 15% test set. Train the preliminary extraction model on the training set to obtain the trained extraction model. Input the validation set into the trained extraction model to validate it and obtain the optimal parameters. Input the optimal parameters into the trained extraction model and train it on the test set to obtain the final extraction model and model accuracy. Output the final extraction model with an accuracy greater than 96%.

[0075] When the map construction module performs personalized modifications to the initial universal path map, the personalized modification method is as follows:

[0076] Step B01, driver correction: When a driver performs a delivery task, find a group of drivers with similarities to the driver from the primary universal route map, calculate the individual ability offset coefficient λ, calculate the personalized experience intensity based on the individual ability offset coefficient λ, and output the personalized experience intensity as the experience intensity.

[0077] Step B02, modify the scenario: when the context of the empirical knowledge triple does not match the actual situation, reduce the confidence of the graph and generate empirical knowledge triples that match the actual situation and add them to the primary universal path graph to obtain the updated universal graph.

[0078] Step B03: Output the updated universal map as a personalized path map.

[0079] Specifically, the experiential knowledge triple refers to a data unit that represents and stores implicit experiential knowledge of human drivers in a structured form. The driver refers to an individual delivery person, the vehicle refers to a specific delivery vehicle including vehicle type attributes, the road segment refers to a directed road unit in a road network, the customer point refers to a specific delivery address, the time period refers to a time interval, and the path refers to a sequence composed of multiple road segments connected sequentially. The triple relationship types include (driver, proficient, path), (driver, usually avoids, path), (driver, familiar, customer point), (vehicle type, unsuitable, road segment), (customer point, has characteristics, time period), and (road segment, under conditions, time period). The (driver, proficient, path) relationship refers to the driver's experience on that path. The above demonstrates a stable efficiency advantage. (Driver, usually avoids, route) refers to the route drivers tend to avoid. (Driver, familiar, customer point) refers to the driver's familiarity with the entry, exit, and parking details of the customer point. (Vehicle type, unsuitable, road segment) refers to a road segment that is unsuitable due to factors such as vehicle size and weight. (Customer point, characteristic, time period) refers to the specific attributes of the customer point during a particular time period, such as placing packages in a parcel locker late. (Road segment, under certain conditions, time period) refers to the typical state of the road segment during a particular time period, such as weekday morning rush hour congestion. The context refers to the preconditions that trigger the triplet's effectiveness, such as (weather: sunny). The graph confidence level refers to the reliability and stability of the empirical knowledge triplet, calculated as: Graph confidence level Zx = Cz 路 / Cz 总 Cz 路 Cz represents the number of times a driver selects a particular route. 总The total number of times selected for the driver; the experience intensity refers to the time saved due to manual deviation operations; one entity type refers to one of the following: driver, vehicle, customer point, time period, and route; the entity graph node refers to the most basic and core data unit that constitutes the graph; the driver group with similarity to the driver refers to a group of drivers whose experience knowledge triples are similar to the driver's; the context does not match the actual situation refers to a context that is different from the actual situation, such as (driver B, avoid, road segment R) where the context is (weather: sunny) but the actual situation is (weather: heavy rain); the personal ability deviation coefficient is a coefficient used to quantify the stable deviation between the individual driver's traffic efficiency and the system's baseline prediction value in a specific road segment or scenario, and its calculation method is: personal ability deviation coefficient λ=Gj / Tj, where Gj is personal experience and Tj is general experience; personal experience refers to the average delivery time of the individual driver, and general experience is the average time calculated by the system based on the delivery times of all drivers; the personalized experience intensity refers to the time saved by the individual driver due to manual deviation operations, and its calculation method is: personalized experience intensity Tg=(1-λ) / Tj ... The method for reducing the confidence level of the graph is as follows: the confidence level of the graph Zx is reduced according to the reduction coefficient jd to obtain the reduced confidence level Zx1. Zx1 is set to jd × Zx. In order to ensure that the confidence level of the graph decreases gradually and to retain error margin for the system, the reduction coefficient jd is set to 0.3. The reduced confidence level Zx1 is output as the graph confidence level Zx. The head entity refers to the entity node in the empirical knowledge triple that is the relation issuer, such as the driver in (driver, skilled at travel, path). The tail entity refers to the entity node in the empirical knowledge triple that is the relation issuer. The entity nodes of the relationship recipient, such as the path in (driver, skilled at travel, path), the historical parameters refer to parameters obtained in history, the training set refers to historical data used to extract the model, the validation set refers to historical data used to validate the initial extraction model, the test set refers to historical data used to test the initial extraction model after inputting the best data, the model accuracy refers to the value used to judge the accuracy of the final extraction model, the best parameters refer to the model parameters obtained by inputting the test set into the initial extraction model, and the final extraction model refers to the model obtained by training on the test set.

[0080] Specifically, the map construction module establishes a primary universal route map through experience knowledge triples and makes personalized corrections to the primary universal route map. This is beneficial for transforming the driver's implicit and intuitive experience into explicit, computable, and storable data, and also helps to improve the feasibility of the route and the driver's trust.

[0081] Specifically, when the interactive feedback module makes personalized adjustments to the personalized correction process based on driving data, the personalized adjustments are as follows:

[0082] Step C01, Prior Distribution: For new drivers, the global capability offset coefficient λ0 is used as a substitute for the personal capability offset coefficient λ. The likelihood function is established when the personal capability offset coefficient λ is equal to the global capability offset coefficient λ0. The noise variance σ0 is obtained, and the prior variance σ1 and prior mean μ1 are calculated based on the distribution of the global capability offset coefficient λ0.

[0083] Step C02: Collect vehicle and driving data for the new delivery task to obtain actual personal experience Gs and general experience Tj;

[0084] Step C03, posterior distribution: Set the posterior mean μ2 = [w1 × μ1 + w2 × (Gs / Tj)] / (w1 + w2), where w1 is the prior weight and w2 is the actual weight, and the posterior variance σ2 = 1 / (w1 + w2), where: w1 = 1 / σ1, w2 = 1 / σ0;

[0085] Step C04: Calculate the trust weight coefficient k, setting it to k = μ1 × (w1 + w2). Compare the posterior variance σ2 with the preset variance σ3. Based on the comparison result, judge the driver's driving performance and adjust the personalized correction process accordingly.

[0086] When σ2>σ3, the driver's driving situation is judged to be uncertain. The adjusted personal ability deviation coefficient λ1 is calculated and set as μ1×(1-k)+(Gs / Tj)×k. The adjusted personal ability deviation coefficient λ1 is then output as the personal ability deviation coefficient λ.

[0087] When σ2≤σ3, the driver's driving situation is judged to be confident, and no personalized adjustment is made to the personalized correction process.

[0088] When σ2≤σ3, the driver's driving situation is considered to be reliable, and no personalized adjustments are made to the personalized correction process.

[0089] Specifically, the "new driver" refers to a driver undertaking a delivery task for the first time, and the "global capability offset coefficient λ0" refers to the statistical description of the entirety of the individual capability offset coefficients λ of all drivers in the system. The calculation method is: Global capability offset coefficient λ0 = λ 总 / Nj, where λ 总 The total capacity offset coefficient λ is the total capacity offset coefficient, where Nj is the number of drivers. 总This refers to the sum of the individual capability deviation coefficients λ of all drivers; the number of drivers Nj refers to the number of drivers participating in the delivery; the prior variance σ1 refers to the variance calculated based on the distribution of the global capability deviation coefficients λ0; the prior mean μ1 refers to the average value calculated based on the distribution of the global capability deviation coefficients λ0; the vehicle data and driving data for the new delivery task refer to the vehicle data and driving data for the first delivery task; the actual personal experience Gs refers to the personal experience obtained based on the vehicle data and driving data for the new delivery task; the noise variance σ0 refers to the value representing the distribution error of the global capability deviation coefficients λ0; the prior weight w1 refers to the weight of the prior mean μ1 when calculating the posterior mean μ2; the actual weight w2 refers to the weight of the actual situation when calculating the posterior mean μ2; the trust weight coefficient k refers to the trust weight of the new driving data; and the preset variance σ3 refers to the preset value used to judge the driver's driving situation. According to historical data, when the variance σ3 > 0.09, the driver's cognitive uncertainty about the road during the driving process exceeds 90%, so the variance σ3 is set to 0.09.

[0090] Specifically, the interactive feedback module uses driving data to personalize the personalized correction process, which helps to automatically adjust the logic and parameters of personalized correction, enabling the system to continuously evolve and adapt to changes in driver habits and the external environment.

[0091] Specifically, when the group graph module constructs a universal group path graph based on the federated learning method and the personalized path graph, it constructs a universal group path graph. The specific method for constructing the universal group path graph is as follows:

[0092] Step D01: Process the personalized path graph on the mobile terminal: Divide the experience knowledge triple graph with the same triple relation type into triple graph sets, build an initial global model on the central server, and send the model parameters of the initial global model to the mobile terminal.

[0093] Step D02: The mobile terminal calculates the graph confidence Zx for each empirical knowledge triple graph in the triple graph set, compares the graph confidence Zx with the preset graph confidence Zx0, judges the value of the empirical knowledge triple graph based on the comparison result, and establishes a candidate triple set based on the judgment result, wherein:

[0094] When Zx≥Zx0, the value of the empirical knowledge triple graph is determined to be high value. The empirical knowledge triple graph is added to the candidate triple set, and the initial global model is trained based on the candidate triple set to obtain the trained initial global model. The model parameters of the trained initial global model are then uploaded to the central server.

[0095] When Zx < Zx0, the value of the empirical knowledge triplet graph is determined to be low, and the empirical knowledge triplet graph is not added to the candidate triplet set.

[0096] In step D03, the central server merges the model parameters of the initial global model after training using a secure aggregation algorithm to obtain a global model, and then distributes the global model to the mobile terminal for the next round of training. The central server also obtains the graph parameters through the global model and merges the graph parameters according to the clustering algorithm to obtain a universal path graph for the group.

[0097] Specifically, the empirical knowledge triplet graphs with the same triplet relation type refer to empirical knowledge triplet graphs with the same content composition structure. Each empirical knowledge triplet graph in the triplet graph set refers to each empirical knowledge triplet graph in the triplet graph set with the same triplet relation type. The preset graph confidence level Zx0 is a preset value used to judge the value of the empirical knowledge triplet graph. Experimental data shows that when the preset graph confidence level Zx0 ≥ 0.75, the empirical knowledge triplet graph has value; therefore, Zx0 = 0.75 is set. The candidate triplet set refers to a set of valuable empirical knowledge triplet graphs. The clustering algorithm refers to an algorithm that automatically sorts a large number of data points according to similarity. The machine learning method for grouping and classification, wherein the graph parameters refer to the graph feature parameters and empirical knowledge triple graph obtained through the global model, the central server refers to the logical central server that coordinates the joint training of multiple parties but does not directly access or store any original privacy data, the initial global model refers to the extraction model built on the central server, the model parameters of the initial global model refer to the initial model parameters before the initial global model is trained, the parameters of the trained initial global model refer to the model parameters obtained after training through a mobile terminal, and the secure aggregation algorithm refers to the cryptographic protocol that allows multiple participants to jointly calculate the aggregate value of these inputs without disclosing their respective private inputs.

[0098] Specifically, the group graph module, by constructing a universal path graph for the group and optimizing the process of establishing the initial universal path graph, helps to aggregate scattered individual experiences into a group knowledge base and feed back to all participants, thus solving the problem of data silos.

[0099] Specifically, when the parameter fading module fades out personal ability offset coefficients that have not been applicable for a long time during the personalized adjustment process, the fading process is as follows:

[0100] During the personalized adjustment process, the usage of the personalized path map is monitored, and when fading is applied based on the monitoring results, a timestamp is added to each knowledge triplet graph in the personalized path map to obtain the current usage time interval T for each knowledge triplet graph. The current usage time interval T for each knowledge triplet graph is compared with a preset time interval T0. Based on the comparison results, the usage of each knowledge triplet graph is judged, and based on the judgment results, each knowledge triplet graph is faded. Specifically:

[0101] When T≥T0, the empirical knowledge triplet graph is determined to be inapplicable for a long time, and the empirical knowledge triplet graph is diluted.

[0102] When T < T0, the application of the experiential knowledge triplet map is deemed applicable, and the personal ability offset coefficient is not downplayed.

[0103] The fade-out process in the parameter fade-out module specifically involves calculating the usage coefficient F of the empirical knowledge triplet graph, and setting the calculation formula for the usage coefficient F as follows: ,in, It is the first The weight used for the second time It is the attenuation coefficient. This represents the total number of uses. The usage coefficient F is compared with F0. Based on the comparison results, the usage of the empirical knowledge triplet graph is judged, and the empirical knowledge triplet graph is diluted based on the judgment results. Among them:

[0104] When F≥F0, the usage of the experience knowledge triplet graph is determined to be that it is not applicable for a long time but has high activity. The personal ability offset coefficient λ of the experience knowledge triplet graph is diluted. The diluted personal ability offset coefficient λ1 is set as λ×(1-β)+μ2×β, where β is the dilution coefficient. The diluted personal ability offset coefficient λ1 is output as the personal ability offset coefficient λ.

[0105] When F < F0, the experience knowledge triplet graph is determined to be inapplicable for a long time and has low activity, and the experience knowledge triplet graph is deleted from the personalized path graph.

[0106] Specifically, the timestamp refers to the time data used to record the usage time of the experience knowledge triplet graph; the current usage time interval T of each experience knowledge triplet graph refers to the time interval between the current usage and the last usage of each experience knowledge triplet graph; the preset time T0 is a preset value used to determine the usage status of the experience knowledge triplet graph, set to T0 = 30 days; and the usage coefficient F of the experience knowledge triplet graph is a coefficient used to represent the activity level of the experience knowledge triplet graph. This embodiment does not include the... Weight used for the second time The specific settings are limited, and those skilled in the art can choose according to the actual situation, such as setting them during the first use. =1, the attenuation coefficient It refers to the first The attenuation factor is used to adjust the weight of subsequent uses. For example, if the weight used more than 30 days ago is set to be halved, then the attenuation factor is set. The diluted personal ability offset coefficient λ1 refers to the personal ability offset coefficient after dilution processing. The dilution coefficient β refers to the parameter used for dilution. To ensure the accuracy of the empirical knowledge triplet map, the attenuation rate is set to 50%, so the dilution coefficient β = 0.05 is set. The global mean μ2 refers to the average efficiency of the entire driver group in a specific road segment described in historical data.

[0107] Specifically, the parameter de-emphasis module de-emphasizes personal ability offset coefficients that have been inapplicable for a long time during the personalized adjustment process. This helps to automatically decay outdated parameters that have been inapplicable for a long time, prevents the system from making misjudgments due to outdated knowledge, and enables the system to flexibly cope with conceptual drift problems such as urban changes and seasonal changes, maintaining the current relevance of decisions.

[0108] Specifically, when the conflict verification module performs multi-dimensional verification on the personalization adjustment process based on the personalized path map and driving data, the multi-dimensional verification method is as follows:

[0109] A rule set is established and matched with a personalized path graph. The matching method is as follows: Personalized triplet relationship types are obtained from the personalized path graph, and the head and tail entities in the personalized triplet relationship types are output as entities. The entities are compared with the rule set, and the personalized path graph is judged based on the comparison results. The individual ability offset coefficient λ is then diluted based on the judgment results.

[0110] When an entity belongs to the rule set, the personalized path graph is determined to be a match. The matched personalized path graph is marked as invalid, and the personal ability offset coefficient is weakened.

[0111] When an entity does not belong to the rule set, the personalized path graph is determined to be a mismatch. The mismatched personalized path graph is marked as valid, and the marked valid personalized path graph is output as the final decision.

[0112] When the conflict verification module optimizes the final decision based on driving data, it obtains the delivery time Th and calculates the delivery time difference Tx based on Th. The delivery time difference Tx is set to Th - Tj. It then compares the delivery time difference Tx with a preset time difference Tx0. Based on the comparison result, it judges the situation of the final decision and optimizes the final decision based on the judgment result. Wherein:

[0113] When Tx≥Tx0, the final decision is determined to be that no optimization is needed, and no optimization is performed on the final decision.

[0114] When Tx < Tx0, the final decision is determined to require optimization, and driving data is added to the final decision.

[0115] Specifically, the rule set refers to the set of rules that the final decision must follow, such as: prohibiting trucks from entering, prohibiting vehicles with current heights exceeding the road height limit, and including a reported dangerous bridge in the planned route. The entity includes head and tail entities. The personalized triplet relationship type in the personalized path graph refers to the relationship type of experiential knowledge triples in the personalized path graph. An entity belonging to the rule set means that its specific state meets the conditions of the rule set; an entity not belonging to the rule set means that its specific state does not meet the conditions of the rule set. The matched personalized path map refers to a personalized path map that matches the content of the rule set. The marked valid personalized path map refers to a personalized path map that does not match the content of the rule set. The delivery time Th refers to the time consumed in the delivery process, which is obtained through the mobile terminal. The preset time difference Tx0 is a preset value used to determine the final decision. Because when the preset time difference delivery time difference Tx > 0, it means that the final decision delivery time is faster than the actual delivery time, and vice versa, and to ensure that external influencing factors are not affected, it is set to -0.5min≤Tx0≤0.5min.

[0116] Specifically, the conflict verification module performs multi-dimensional verification of the personalized adjustment process and optimizes the final decision. This is beneficial for real-time quantitative trade-offs and risk simulation among multiple objectives such as efficiency, safety, compliance and personal habits, outputting a final decision that is both intelligent and safe, and transforming conflict cases into nutrients for optimizing the overall system.

[0117] Specifically, the path output module outputs the final decision as a dynamic delivery route and pushes the dynamic delivery route to the mobile terminal via the WebSocket protocol.

[0118] Specifically, the WebSocket protocol, also known as the WebSocket protocol, is a network transmission protocol that enables full-duplex communication over a single TCP connection.

[0119] Specifically, the route output module outputs the final decision as a dynamic delivery route to the mobile terminal, which helps provide drivers with a delivery route reference that is beneficial to road conditions and improves delivery efficiency.

[0120] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A dynamic route optimization system for urban shared delivery based on big data, characterized in that, include: The data acquisition module is used to collect vehicle data and driving data, and to perform spatiotemporal alignment on the vehicle data to obtain aligned data. The graph construction module is used to extract empirical knowledge triples from the aligned data, establish a primary universal path graph through the empirical knowledge triples, and make personalized corrections to the primary universal path graph to obtain a personalized path graph. The interactive feedback module is used to personalize the personalized correction process based on driving data; The community graph module is used to optimize the initial universal path graph using federated learning methods to obtain a community universal path graph. The parameter fade-out module is used to fade out personal ability offset coefficients that have not been applicable for a long time during the personalization process; The conflict verification module is used to perform multi-dimensional verification of the personalized adjustment process based on the personalized route map and driving data, output the final decision, and optimize the final decision based on the driving data. The route output module is used to output the final decision as a dynamic delivery route to the mobile terminal.

2. The urban shared delivery dynamic route optimization system based on big data according to claim 1, characterized in that, The graph construction module extracts empirical knowledge triples based on the alignment data. When building a primary universal path graph using these empirical knowledge triples, it establishes an extraction model. The specific method for establishing the extraction model is as follows: Step A01: Define the entity type and triple relationship type, and output the definition process as the data processing layer of the preliminary extraction model. The entity type is set to include driver, vehicle, customer point, time period and route, and the triple relationship type is (head entity, connection relationship, tail entity). Step A02: Extract the aligned data according to the entity type and triple relationship type to obtain empirical triples and obtain the context. At the same time, perform feature calculation on the graph confidence and empirical strength, and output the feature calculation process as the feature layer of the preliminary extraction model. Step A03: Create an entity graph node for each entity type. Use the context, graph confidence, and experience strength as directed edges, pointing from the head entity to the tail entity, to obtain the experience knowledge triple graph. By combining all experience knowledge triple graphs, a primary universal path graph is obtained, which is used as the output layer of the preliminary extraction model. Step A04: Divide the historical data into a 70% training set, a 15% validation set, and a 15% test set. Train the preliminary extraction model on the training set to obtain the trained extraction model. Input the validation set into the trained extraction model for validation to obtain the optimal parameters. Input the optimal parameters into the trained extraction model and train it on the test set to obtain the final extraction model and model accuracy. Output the final extraction model with an accuracy greater than 96%.

3. The urban shared delivery dynamic route optimization system based on big data according to claim 2, characterized in that, The map construction module performs personalized modifications on the initial universal path map. The specific personalized modification method is as follows: Step B01, driver correction: When a driver performs a delivery task, find a group of drivers with similarities to the driver from the primary universal route map, calculate the individual ability offset coefficient λ, calculate the personalized experience intensity based on the individual ability offset coefficient λ, and output the personalized experience intensity as the experience intensity. Step B02, modify the scenario: when the context of the empirical knowledge triple does not match the actual situation, reduce the confidence of the graph and generate empirical knowledge triples that match the actual situation and add them to the primary universal path graph to obtain the updated universal graph. Step B03: Output the updated universal map as a personalized path map.

4. The urban shared delivery dynamic route optimization system based on big data according to claim 3, characterized in that, The interactive feedback module personalizes the personalized correction process based on driving data, and the personalized adjustments include: Step C01, Prior Distribution: For new drivers, the global capability offset coefficient λ0 is used as the personal capability offset coefficient λ to replace it. The likelihood function is established when the personal capability offset coefficient λ = the global capability offset coefficient λ0. The noise variance σ0 is obtained, and the prior variance σ1 and prior mean μ1 are calculated according to the distribution of the global capability offset coefficient λ0. Step C02: Collect vehicle and driving data for the new delivery task to obtain actual personal experience Gs and general experience Tj; Step C03, posterior distribution: Set the posterior mean μ2=[w1×μ1+w2×(Gs / Tj)] / (w1+w2), where w1 is the prior weight and w2 is the actual weight, and the posterior variance σ2=1 / (w1+w2), where: w1=1 / σ1, w2=1 / σ0.

5. The urban shared delivery dynamic route optimization system based on big data according to claim 4, characterized in that, The interactive feedback module personalizes the personalized correction process based on driving data, and the personalized adjustment further includes: Step C04: Calculate the trust weight coefficient k, setting it to k = μ1 × (w1 + w2). Compare the posterior variance σ2 with the preset variance σ3. Based on the comparison result, judge the driver's driving performance and adjust the personalized correction process accordingly. When σ2>σ3, the driver's driving situation is judged to be uncertain. The adjusted personal ability deviation coefficient λ1 is calculated and set as μ1×(1-k)+(Gs / Tj)×k. The adjusted personal ability deviation coefficient λ1 is then output as the personal ability deviation coefficient λ. When σ2≤σ3, the driver's driving situation is considered to be reliable, and no personalized adjustments are made to the personalized correction process.

6. The urban shared delivery dynamic route optimization system based on big data according to claim 5, characterized in that, When the group graph module constructs a universal group path graph based on the federated learning method and the personalized path graph, it constructs a universal group path graph. The specific method for constructing the universal group path graph is as follows: Step D01: Process the personalized path graph on the mobile terminal: Divide the experience knowledge triple graph with the same triple relation type into triple graph sets, build an initial global model on the central server, and send the model parameters of the initial global model to the mobile terminal. Step D02: The mobile terminal calculates the graph confidence Zx for each empirical knowledge triple graph in the triple graph set, compares the graph confidence Zx with the preset graph confidence Zx0, judges the value of the empirical knowledge triple graph based on the comparison result, and establishes a candidate triple set based on the judgment result, wherein: When Zx≥Zx0, the value of the empirical knowledge triple graph is determined to be high value. The empirical knowledge triple graph is added to the candidate triple set, and the initial global model is trained based on the candidate triple set to obtain the trained initial global model. The model parameters of the trained initial global model are then uploaded to the central server. When Zx < Zx0, the value of the empirical knowledge triplet graph is determined to be low, and the empirical knowledge triplet graph is not added to the candidate triplet set. In step D03, the central server merges the model parameters of the initial global model after training using a secure aggregation algorithm to obtain a global model, and then distributes the global model to the mobile terminal for the next round of training. The central server also obtains the graph parameters through the global model and merges the graph parameters according to the clustering algorithm to obtain a universal path graph for the group.

7. The urban shared delivery dynamic route optimization system based on big data according to claim 6, characterized in that, The parameter fading module monitors the usage of the personalized path map during the personalization adjustment process and performs fading processing based on the monitoring results. It adds timestamps to each knowledge triplet graph in the personalized path map to obtain the current usage time interval T for each knowledge triplet graph. The current usage time interval T for each knowledge triplet graph is compared with a preset time interval T0. Based on the comparison results, the usage of each knowledge triplet graph is judged, and based on the judgment results, each knowledge triplet graph is faded. Wherein: When T≥T0, the empirical knowledge triplet graph is determined to be inapplicable for a long time, and the empirical knowledge triplet graph is diluted. When T < T0, the application of the empirical knowledge triplet map is deemed applicable, and the personal ability offset coefficient is not diluted.

8. The urban shared delivery dynamic route optimization system based on big data according to claim 7, characterized in that, The fade-out process in the parameter fade-out module specifically involves calculating the usage coefficient F of the empirical knowledge triplet graph, and setting the calculation formula for the usage coefficient F as follows: ,in, It is the first The weight used for the second time It is the attenuation coefficient. This represents the total number of uses, and the usage coefficient F is compared with F0. Based on the comparison results, the usage of the empirical knowledge triplet graph is judged, and based on the judgment results, the empirical knowledge triplet graph is diluted. Specifically: When F≥F0, the usage of the experience knowledge triplet graph is determined to be that it is not applicable for a long time but has high activity. The personal ability offset coefficient λ of the experience knowledge triplet graph is diluted. The diluted personal ability offset coefficient λ1 is set as λ×(1-β)+μ2×β, where β is the dilution coefficient. The diluted personal ability offset coefficient λ1 is output as the personal ability offset coefficient λ. When F < F0, the experience knowledge triplet graph is determined to be inapplicable for a long time and has low activity, and the experience knowledge triplet graph is deleted from the personalized path graph.

9. The urban shared delivery dynamic route optimization system based on big data according to claim 8, characterized in that, When the conflict verification module performs multi-dimensional verification on the personalization adjustment process based on the personalized path map and driving data, the multi-dimensional verification method is as follows: A rule set is established and matched with a personalized path graph. The matching method is as follows: Personalized triplet relationship types are obtained from the personalized path graph, and the head and tail entities in the personalized triplet relationship types are output as entities. The entities are compared with the rule set, and the personalized path graph is judged based on the comparison results. The individual ability offset coefficient λ is then diluted based on the judgment results. When an entity belongs to the rule set, the personalized path graph is determined to be a match. The matched personalized path graph is marked as invalid, and the personal ability offset coefficient is weakened. When an entity does not belong to the rule set, the personalized path graph is determined to be a mismatch. The mismatched personalized path graph is marked as valid, and the marked valid personalized path graph is output as the final decision.

10. The urban shared delivery dynamic route optimization system based on big data according to claim 9, characterized in that, When the conflict verification module optimizes the final decision based on driving data, it obtains the delivery time Th and calculates the delivery time difference Tx based on Th. The delivery time difference Tx is set to Th - Tj. It then compares the delivery time difference Tx with a preset time difference Tx0. Based on the comparison result, it judges the situation of the final decision and optimizes the final decision based on the judgment result. Wherein: When Tx≥Tx0, the final decision is determined to be that no optimization is needed, and no optimization is performed on the final decision. When Tx < Tx0, the final decision is determined to require optimization, and driving data is added to the final decision.

Images