Bulk product source flow direction modeling method based on point source list and beidou trajectory

Abstract

The present application relates to a kind of based on point source list and big product source flow direction modeling method of big product source, belong to traffic and environmental management technical field.It includes: constructing big cargo source database, using 5G positioning and big dipper trajectory data, by time-space fusion algorithm cleaning and reconstruction trajectory, identify key transport stop point;Dynamic spatial matching model is built to optimize correlation strategy, based on graph neural network integrates production plan and demand forecast generates potential path, through double-dimensional evaluation system planning candidate route;Adopt block chain technology to collect multi-source heterogeneous data, through intelligent prediction model realizes path classification and scheme formulation, and utilize adversarial generative network and expert knowledge graph to verify and update rule;Based on digital twinning technology realizes the dynamic mapping of transport path, through time-space big data analysis and causal inference model mining logistics correlation and forecast optimization direction.The accurate modeling and decision support of source-flow-direction relationship are realized.

Description

Technical Field

[0001] This invention belongs to the field of transportation and environmental management technology, specifically involving a method for modeling the source flow of bulk products based on point source lists and BeiDou trajectories. Background Technology

[0002] In the full-chain management of coal as a bulk commodity, precise control of its source and flow is the core foundation for improving logistics efficiency, optimizing resource allocation, and strengthening environmental supervision. Currently, the industry's need for source and flow modeling of bulk commodities is increasingly urgent—on the one hand, logistics companies need to reduce transportation costs and avoid congestion risks through precise route planning; on the other hand, regulatory authorities need to rely on reliable source and flow data to trace pollution sources and verify capacity utilization rates, and consumer-end enterprises also need to ensure supply chain stability based on origin and route information.

[0003] However, existing bulk commodity source flow modeling technologies have several shortcomings that make it difficult to meet practical application needs: The accuracy of source location data collection is insufficient. Traditional methods rely heavily on manually submitted production capacity and location data from enterprises or single administrative statistical lists, which suffer from data lag and large errors. Manual submissions are prone to poor timeliness due to long statistical cycles and lack precise spatial verification. Some enterprises' geographical coordinates shift due to submission errors, making it impossible to provide reliable source flow benchmark data for source flow modeling. Although a few technologies attempt to combine satellite remote sensing data, they are not deeply integrated with point source emission inventories, nor do they employ sophisticated image recognition technology to distinguish between enterprise production areas and non-production areas, resulting in insufficient accuracy in extracting production scale and output data.

[0004] Positioning and trajectory processing technologies have limitations. Existing trajectory data mostly rely on single positioning technologies, which are prone to signal interruptions and positioning drift in complex scenarios, resulting in incomplete or inaccurate trajectory data. Even when using multi-source positioning data, there is a lack of effective spatiotemporal fusion mechanisms, making it impossible to dynamically adjust data weights based on signal strength. Furthermore, trajectory reconstruction does not fully incorporate the actual road network direction, easily leading to trajectory inflection points that do not match the real road network. This, in turn, can cause errors in subsequent stop point identification, affecting the accuracy of source flow association.

[0005] Spatial correlation and route planning have poor adaptability. Existing spatial matching methods mostly use fixed distance thresholds to associate production sites and stop points, without considering real-time traffic conditions and dynamic changes in the road network. When there is traffic congestion, road construction, or temporary control in the area, the fixed thresholds cannot adapt to the shift of stop points caused by vehicle detours, which easily leads to misassociation or omission of production sites and stop points. Route planning is mostly based on only one dimension, such as transportation cost or distance, without integrating two-dimensional data such as enterprise production plans and consumer demand forecasts. This results in a low degree of matching between the planned candidate routes and actual production capacity and demand, making it difficult to implement. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides a method for modeling the source flow of bulk commodities based on point source lists and BeiDou trajectories. The objective of this invention can be achieved through the following technical solutions:

[0007] A method for modeling the source and flow of bulk commodities based on point source lists and BeiDou trajectories includes:

[0008] S1: Based on point source emission inventories and satellite remote sensing data, artificial intelligence image recognition technology is used to extract enterprise geographical location, production scale and product output information to build a database of bulk cargo sources; 5G cellular positioning data and Beidou trajectory data of freight vehicles are obtained, and spatiotemporal fusion algorithms are used to complete data cleaning and trajectory reconstruction, and key stopping points are identified based on the trajectory pattern recognition engine;

[0009] S2: Construct a dynamic spatial matching model, combine real-time traffic conditions and road network changes, and adjust the spatial association strategy between production enterprises and the key stopping points; construct a regional logistics topology map based on graph neural network, integrate enterprise production plans and consumer demand forecast data, and generate potential transportation routes; plan candidate transportation routes based on a demand forecast-capacity matching dual-dimensional evaluation system, combined with the spatial association strategy.

[0010] S3: Based on blockchain technology, collect information on product category, transportation distance, and vehicle type corresponding to the candidate transportation routes; construct an intelligent predictive analysis model, classify the candidate transportation routes through multi-dimensional feature fusion, and formulate differentiated solutions based on the classification results; use adversarial generative networks to generate virtual paths for comparison and verification, and dynamically update the preset modeling and evaluation rules based on expert knowledge graphs;

[0011] S4: Based on digital twin technology, a three-dimensional visualization platform is built to visualize and digitally map the actual transportation path distribution of bulk products; using spatiotemporal big data analysis technology, multi-source heterogeneous data is integrated to mine the potential connections between key production enterprises, regional logistics hubs and consumer enterprises, and the source-flow-direction relationship is dynamically extrapolated using a causal inference model to predict the direction of transportation path optimization.

[0012] As a preferred embodiment of the present invention, the bulk commodity source database is constructed through a combination of point source list data preprocessing technology, satellite remote sensing data parsing technology, and dynamic data verification technology, specifically including:

[0013] The enterprise data in the point source emission inventory is processed for field standardization, removing duplicate enterprise information and invalid production capacity data, and unifying industry classification standards and data formats. Image enhancement technology is used to optimize the texture features of satellite remote sensing data, and terrain correction technology is used to eliminate the impact of terrain undulations on enterprise location positioning, extracting spatial range parameters of enterprise factory buildings and storage areas. Based on entity matching technology, the enterprise names and industry types in the point source emission inventory are correlated and compared with the enterprise spatial information extracted by remote sensing, correcting the geographical coordinate offset and production scale statistical deviation in the inventory, and outputting a geographically calibrated-production fusion dataset. A data periodic update mechanism is established to collect monthly production reports and inventory data of enterprises, and synchronously update product output and inventory balance information in the database.

[0014] Specifically, the method for using a spatiotemporal fusion algorithm to complete data cleaning and trajectory reconstruction is as follows:

[0015] Signal strength parameters of 5G cellular positioning data and BeiDou trajectory data from freight vehicles are collected, and fusion weight coefficients are paired based on a dynamic weight allocation mechanism. A three-dimensional road network constraint model is constructed, and a road network topology feature vector is generated by combining the road direction, number of lanes, and node distribution of the regional road network. The fused trajectory segments are smoothed based on semantic understanding, and trajectory inflection points that do not conform to the actual road network are identified and removed through adversarial generative networks to correct trajectory deviations. Incremental trajectory reconstruction technology is adopted, combined with edge computing, to achieve real-time cleaning and trajectory reconstruction of positioning data.

[0016] Specifically, the method for identifying the key stopping points is as follows:

[0017] The trajectory pattern recognition engine collects vehicle speed, acceleration, and direction change data to establish a driving state characteristic threshold model and capture the state transition nodes from driving to stationary. Motion parameter analysis technology is used to perform secondary verification of these state transition nodes. A quantitative index system for dwelling characteristics is constructed by calculating dwell time and statistically analyzing vehicle density per unit area. A Bayesian classification model is established by combining the layout of loading and unloading sites and parking lot location characteristics around bulk cargo origins in the geographic information spatial database. By analyzing the spatial distance between dwelling points and facilities and the matching degree of historical dwelling patterns, temporary parking and loading / unloading dwellings are distinguished, and key dwelling points related to the cargo origin are identified.

[0018] Specifically, the adjustment of the spatial association strategy includes:

[0019] Traffic flow, congestion levels, and road conditions data are collected within the region. A traffic impact weighting factor, incorporating both temporal and spatial dimensions, is constructed using a weighted average algorithm and a grey relational analysis algorithm. A three-tiered congestion early warning mechanism is established, dynamically adjusting the correlation distance threshold between production enterprises and key stopping points based on congestion levels. In response to unforeseen circumstances, the bulk product source-flow model utilizes real-time map updates to capture changes in road network topology. A hybrid path planning strategy combining Dijkstra's algorithm and genetic algorithms is employed, along with the correlation distance threshold, to reconstruct the road network and recalculate the optimal correlation path between key stopping points and production enterprises. Simultaneously, a historical traffic event database is established to predict road network change patterns, enabling dynamic adaptation of spatial correlation strategies.

[0020] Specifically, the construction of the regional logistics topology map is based on the synergy of graph neural networks, node weight assignment technology, and supply and demand data integration technology. The implementation method is as follows:

[0021] The production enterprises, logistics hubs, and consumer enterprises within the region are abstracted as nodes in a network topology. Multi-source data acquisition technology is used to collect historical transportation frequency and capacity data for these nodes. The connection strength between nodes is used as a measure of the closeness of transportation links. A node connection strength calculation model is constructed using a transportation frequency weighting algorithm. The weights of production enterprise nodes are adjusted based on capacity fluctuation data from enterprise production plans, and the weights of consumer enterprise nodes are adjusted based on changes in consumer demand. Through the iterative learning mechanism of the graph neural network, the topology structure is optimized by combining node weights and connection strength data to construct a regional logistics topology map.

[0022] Specifically, the planning of the candidate transportation routes includes:

[0023] Based on the aforementioned demand forecasting-capacity matching dual-dimensional evaluation system, this system analyzes historical consumption data and market demand trends, and uses a time-series forecasting model to intelligently predict the short-, medium-, and long-term demand scale of consumers. It employs blockchain technology to construct a shared ledger of production capacity for enterprises in production areas, synchronizing current capacity, inventory levels, and production plans in real time. A federated learning framework is used to calculate the matching degree between demand and capacity based on the intelligent forecasting. Furthermore, it introduces BeiDou spatiotemporal big data, integrating transportation costs, transportation timeliness, and road safety risk indicators to construct a dynamic risk assessment system. Digital twin technology is used to visualize and comprehensively score potential transportation routes. Finally, based on a multi-objective optimization algorithm, the candidate transportation routes are dynamically planned and prioritized using the matching degree and the comprehensive score.

[0024] Specifically, the method for classifying candidate transportation routes through multi-dimensional feature fusion is as follows:

[0025] The system integrates multi-dimensional feature information from the transportation process and normalizes these features to map them to a unified numerical range. The processed multi-dimensional features are then input into the intelligent prediction and analysis model. A decision tree algorithm is used to mine classification rules from the data, and the nonlinear fitting ability of neural networks is used to learn feature relationships. Through a multi-model fusion strategy, the prediction results are integrated and optimized. The model outputs the candidate transportation route classification results by comprehensively considering transportation time, cost, and congestion probability, and outputs differentiated solutions based on the classification results.

[0026] Specifically, the method for updating the modeling evaluation rules is as follows:

[0027] Multi-source heterogeneous data acquisition technology is used to obtain information on transportation safety standards, logistics optimization cases, and abnormal situation handling solutions in the logistics industry. A knowledge graph construction engine is used to transform the collected multi-source heterogeneous data into nodes in the graph. Based on the semantic links of temporal and causal relationships, the associations between nodes are established to form a structured expert experience knowledge base.

[0028] When new transportation scenarios, policy adjustments, or technological updates occur, an associative reasoning process is triggered. A graph neural network algorithm is used to analyze the logical relationship between the new situation and existing rules. Combined with real-time feedback data from actual application scenarios, a reinforcement learning algorithm is used to iteratively correct the parameters of existing rules, thereby achieving adaptive dynamic updates of the modeling and evaluation rules.

[0029] Specifically, the method for implementing the visualized digital mapping is as follows:

[0030] Based on satellite remote sensing technology, geospatial data in the transportation network is collected. Combined with computer vision and 3D reconstruction technology, the terrain and building structure are reconstructed to build a 3D entity model. The transportation route data of bulk products is preprocessed through edge computing and federated learning technology. The dynamic data is mapped to the 3D entity model, and spatiotemporal heat maps and topological relationship maps are introduced. Paths with different transportation flow or transportation status are distinguished by dynamic color coding.

[0031] Specifically, the mining of the potential associations includes:

[0032] This process involves acquiring dynamic production data, cargo transportation information, and core operational data indicators from enterprises to construct a multi-source heterogeneous dataset. Data cleaning techniques are used to remove duplicate records and logically erroneous data. Unstructured text data is converted into structured data storage through format transformation. Outlier detection algorithms are used to identify and correct abnormal data. Association rule mining algorithms are employed to establish a quantitative relationship model between production capacity fluctuations of enterprises in production areas and container throughput and railway freight volume at logistics hubs. Route analysis algorithms are used to analyze the impact of seasonal changes in consumer demand on transportation route selection preferences. Finally, spatiotemporal clustering algorithms are used to identify core hub nodes in the logistics network within a specific time period.

[0033] Specifically, the dynamic deduction of the source-flow-direction relationship includes:

[0034] Based on data mining technology, multi-dimensional data on production capacity fluctuations, regional market demand changes, industry policy adjustments, and natural disaster frequency are extracted from historical source-flow-direction databases. Granger causality tests are used to quantitatively analyze the impact of each factor on changes in transportation routes. A visualized causal relationship network is constructed using graph neural network algorithms, and dynamic impact weights are assigned to factors such as capacity utilization rate, demand gap rate, and policy implementation intensity through node weight matrices.

[0035] An elastic response mechanism to external factors is constructed, taking sudden policy adjustments and natural disasters as exogenous variables; a probability distribution map of transportation routes is generated using the Monte Carlo simulation method, and combined with a time series prediction model, the evolution trend of the source-flow-direction relationship is dynamically predicted, identifying key nodes and transportation corridors of route changes.

[0036] The beneficial effects of this invention are as follows:

[0037] By collaboratively constructing a source location database through point source inventory preprocessing, satellite remote sensing analysis, and dynamic verification technologies: field standardization removes invalid data, and unified data format avoids statistical bias; image enhancement and terrain correction technologies optimize satellite remote sensing data to accurately extract spatial parameters of enterprise factory buildings and storage areas; entity matching technology corrects coordinate offset and scale errors, and combined with monthly production reports to dynamically update inventory and output, compared with traditional manual reporting or single lists, it effectively solves the problems of data lag, coordinate bias, and inaccurate scale statistics, improving the spatial accuracy of source location data to closely match the actual factory area.

[0038] To address the signal deficiencies of single positioning technologies, this method integrates 5G cellular and BeiDou positioning data, using a dynamic weight allocation mechanism to offset fluctuations caused by single technologies. A 3D road network constraint model and adversarial generative networks eliminate trajectory inflection point deviations, and incremental reconstruction ensures that the trajectory closely matches the actual road network. Key stop point identification employs a Bayesian classification model, combining vehicle motion parameters and surrounding facility features to distinguish between temporary stops and loading / unloading stops, effectively solving the problems of positioning drift, trajectory incompleteness, and misjudgment of stop points in complex scenarios. This significantly improves the consistency between the reconstructed trajectory and the real road network, and enhances the accuracy of key stop point identification compared to traditional methods, providing effective data support for subsequent spatial correlation.

[0039] The dynamic spatial matching model overcomes the limitations of fixed distance thresholds: it dynamically adjusts the correlation threshold through traffic impact weight factors, and the three-level congestion early warning mechanism adapts to different traffic conditions; when the road network changes suddenly, the hybrid path planning strategy quickly reconstructs the road network, and the historical event database enables the prediction of correlation strategies, avoiding false or missed correlations; the regional logistics topology map combines graph neural networks and supply and demand data to dynamically adjust node weights, and path planning incorporates a two-dimensional evaluation of production planning and demand forecasting. Compared with traditional single-dimensional planning, it reduces the empty running rate of transportation caused by poor path adaptability, lowers logistics costs, and avoids traffic congestion risks.

[0040] The 3D visualization platform built with digital twin technology intuitively displays the spatial distribution and flow changes of transportation routes through spatiotemporal heat maps and dynamic color coding, making it easier to capture congested nodes compared to traditional two-dimensional tables. Spatiotemporal big data analysis uncovers potential connections between production areas, hubs, and consumption ends. The causal inference model, combined with Monte Carlo simulation and time series forecasting, can dynamically deduce the evolution trend of source-flow-direction relationships, providing forward-looking decision support for supply chain adjustments, reducing the impact of sudden risks on transportation efficiency, and ensuring the stability of the bulk commodity supply chain. Attached Figure Description

[0041] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0042] Figure 1 This is a flowchart illustrating a method for modeling the source flow of bulk products based on a point source list and BeiDou trajectories according to the present invention.

[0043] Figure 2 This is a diagram illustrating the candidate transportation route planning architecture of the present invention;

[0044] Figure 3 This is a schematic diagram of the process for updating the modeling and evaluation rules of the present invention. Detailed Implementation

[0045] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0046] Please see Figure 1 A method for modeling the source flow of bulk commodities based on point source lists and BeiDou trajectories, including:

[0047] S1: Based on point source emission inventories and satellite remote sensing data, artificial intelligence image recognition technology is used to extract enterprise geographical location, production scale and product output information to build a database of bulk cargo sources; 5G cellular positioning data and Beidou trajectory data of freight vehicles are obtained, and spatiotemporal fusion algorithms are used to complete data cleaning and trajectory reconstruction, and key stopping points are identified based on the trajectory pattern recognition engine;

[0048] S2: Construct a dynamic spatial matching model, combine real-time traffic conditions and road network changes, and adjust the spatial association strategy between production enterprises and the key stopping points; construct a regional logistics topology map based on graph neural network, integrate enterprise production plans and consumer demand forecast data, and generate potential transportation routes; plan candidate transportation routes based on a demand forecast-capacity matching dual-dimensional evaluation system, combined with the spatial association strategy.

[0049] S3: Based on blockchain technology, collect information on product category, transportation distance, and vehicle type corresponding to the candidate transportation routes; construct an intelligent predictive analysis model, classify the candidate transportation routes through multi-dimensional feature fusion, and formulate differentiated solutions based on the classification results; use adversarial generative networks to generate virtual paths for comparison and verification, and dynamically update the preset modeling and evaluation rules based on expert knowledge graphs;

[0050] S4: Based on digital twin technology, a three-dimensional visualization platform is built to visualize and digitally map the actual transportation path distribution of bulk products; using spatiotemporal big data analysis technology, multi-source heterogeneous data is integrated to mine the potential connections between key production enterprises, regional logistics hubs and consumer enterprises, and the source-flow-direction relationship is dynamically extrapolated using a causal inference model to predict the direction of transportation path optimization.

[0051] Specifically, the bulk commodity source database is constructed through a combination of point source list data preprocessing technology, satellite remote sensing data parsing technology, and dynamic data verification technology, including:

[0052] The enterprise data in the point source emission inventory is processed for field standardization, removing duplicate enterprise information and invalid production capacity data, and unifying industry classification standards and data formats. Image enhancement technology is used to optimize the texture features of satellite remote sensing data, and terrain correction technology is used to eliminate the impact of terrain undulations on enterprise location positioning, extracting spatial range parameters of enterprise factory buildings and storage areas. Based on entity matching technology, the enterprise names and industry types in the point source emission inventory are correlated and compared with the enterprise spatial information extracted by remote sensing, correcting the geographical coordinate offset and production scale statistical deviation in the inventory, and outputting a geographically calibrated-production fusion dataset. A data periodic update mechanism is established to collect monthly production reports and inventory data of enterprises, and synchronously update product output and inventory balance information in the database.

[0053] Specifically, the method for using a spatiotemporal fusion algorithm to complete data cleaning and trajectory reconstruction is as follows:

[0054] Signal strength parameters of 5G cellular positioning data and BeiDou trajectory data from freight vehicles are collected, and fusion weight coefficients are paired based on a dynamic weight allocation mechanism. A three-dimensional road network constraint model is constructed, and a road network topology feature vector is generated by combining the road direction, number of lanes, and node distribution of the regional road network. The fused trajectory segments are smoothed based on semantic understanding, and trajectory inflection points that do not conform to the actual road network are identified and removed through adversarial generative networks to correct trajectory deviations. Incremental trajectory reconstruction technology is adopted, combined with edge computing, to achieve real-time cleaning and trajectory reconstruction of positioning data.

[0055] In this embodiment, a cement transportation route from a cement production plant in the suburbs of City A in a certain province to a cement mixing plant in the urban area is taken as an example. The route is about L1 kilometers long and includes: ① a two-lane rural road in both directions in the suburbs (unobstructed but with weak signal), ② a four-lane expressway in both directions in the urban area (dense high-rise buildings make signal obstruction easy), and ③ a two-lane one-way ramp from the expressway to the mixing plant (with many curves). The transportation vehicles are 10 heavy-duty cement tanker trucks, each equipped with a vehicle-mounted terminal with 5G positioning function (supporting the collection of 5G cellular positioning data) and a Beidou dual-mode positioning terminal (supporting the output of Beidou trajectory data). It is necessary to realize the real-time and accurate reconstruction of the transportation trajectory to provide data support for the subsequent cement source flow association.

[0056] 1. Signal strength acquisition and dynamic weighting coefficient pairing: Deploy a "real-time signal monitoring module" in the on-board terminal of each cement truck. This module synchronously collects the signal strength parameters of 5G cellular positioning data and Beidou trajectory data every t1 seconds - specifically, the signal-to-noise ratio (SNR) of 5G signal and the carrier-to-noise ratio (CNR) of Beidou signal.

[0057] The dynamic weight allocation rules are set as follows: When the 5G signal SNR is ≥35dB (stable signal, such as a section of urban expressway without tall buildings blocking the view) and the BeiDou signal CNR is ≥42dB, the 5G positioning data is assigned a weight coefficient of 0.6 and the BeiDou data is assigned a weight coefficient of 0.4; when the 5G signal SNR is <30dB (weak signal, such as a rural road in the suburbs) and the BeiDou signal CNR is ≥40dB, the weight coefficient of the 5G data is reduced to 0.3 and the weight coefficient of the BeiDou data is increased to 0.7; for overlapping positioning points with the same timestamp (e.g., 14:00:02), the fused coordinates are calculated according to the above weights.

[0058] 2. Construction of a 3D Road Network Constraint Model: Obtain the "Road Network Vector Database" of the area to which the transport route belongs, provided by the Transportation Bureau of City A, and extract the following core data: ① Rural road section (K0-K15): Road orientation is north-south, two-way two-lane, node spacing 500 meters (including 3 intersection nodes); ② Expressway section (K15-K40): Road orientation is east-west, two-way four-lane, node spacing L2 meters (including 2 interchange nodes); ③ Ramp section (K40-K45): Road orientation is northwest-southeast, one-way two-lane, node spacing L3 meters (including 1 curve node). Using "Road ID - Orientation Angle - Number of Lanes - Node Spacing - Node Type" as feature dimensions, generate the feature vector of the road network topology structure for this area.

[0059] 3. Smoothing adjustment of trajectory segments and elimination of inflection points: When a cement truck travels to the K25 section of the expressway (a densely populated area of ​​high-rise buildings), the 5G signal is blocked and momentarily drifts. The merged positioning point deviates to a residential area 50 meters east of the expressway (not within the road network range).

[0060] Generative Adversarial Network (GAN) is activated for inflection point identification: The location point is compared with the road network topology feature vector of the K25 section of the expressway. It is found that the "spatial location-road network matching degree" of the point is lower than the preset threshold (0.3), and it is judged as a "trajectory inflection point that does not match the actual road network" and automatically removed. Then, based on the two valid location points before and after the inflection point, combined with the orientation characteristics of the "east-west straight road segment" of the expressway, a linear fitting algorithm is used to generate a corrected location point to ensure that the trajectory segment fits the actual orientation of the expressway.

[0061] 4. Incremental trajectory reconstruction and real-time edge computing processing: An edge gateway (supporting edge computing function) is deployed next to the on-board terminal of each cement truck. The gateway is connected to the positioning terminal via CAN bus and receives the fused positioning data every 5 seconds.

[0062] Real-time cleaning: The edge gateway has a built-in "rapid screening rule for outliers". When the vehicle speed corresponding to the received location data (calculated by combining the time difference between adjacent location points) exceeds the reasonable range of heavy tank trucks, it is immediately marked as outlier data and removed without uploading to the cloud.

[0063] Incremental Reconstruction: The gateway reconstructs the trajectory according to the "segmented storage" logic. When a vehicle departs from the factory area (starting point A) and travels to the K10 section of the suburban highway (point B), the gateway only stores the AB segment trajectory. When the vehicle continues to travel to the K20 section of the expressway (point C), the gateway does not need to recalculate the AB segment, but only incrementally supplements the BC segment trajectory. Finally, when the vehicle arrives at the mixing plant (end point D), the gateway generates the complete ABCD trajectory. The entire process does not involve repeated calculation of trajectory data, ensuring the real-time nature of trajectory reconstruction.

[0064] Specifically, the method for identifying the key stopping points is as follows:

[0065] The trajectory pattern recognition engine collects vehicle speed, acceleration, and direction change data to establish a driving state characteristic threshold model and capture the state transition nodes from driving to stationary. Motion parameter analysis technology is used to perform secondary verification of these state transition nodes. A quantitative index system for dwelling characteristics is constructed by calculating dwell time and statistically analyzing vehicle density per unit area. A Bayesian classification model is established by combining the layout of loading and unloading sites and parking lot location characteristics around bulk cargo origins in the geographic information spatial database. By analyzing the spatial distance between dwelling points and facilities and the matching degree of historical dwelling patterns, temporary parking and loading / unloading dwellings are distinguished, and key dwelling points related to the cargo origin are identified.

[0066] Specifically, the adjustment of the spatial association strategy includes:

[0067] Traffic flow, congestion levels, and road conditions data are collected within the region. A traffic impact weighting factor, incorporating both temporal and spatial dimensions, is constructed using a weighted average algorithm and a grey relational analysis algorithm. A three-tiered congestion early warning mechanism is established, dynamically adjusting the correlation distance threshold between production enterprises and key stopping points based on congestion levels. In response to unforeseen circumstances, the bulk product source-flow model utilizes real-time map updates to capture changes in road network topology. A hybrid path planning strategy combining Dijkstra's algorithm and genetic algorithms is employed, along with the correlation distance threshold, to reconstruct the road network and recalculate the optimal correlation path between key stopping points and production enterprises. Simultaneously, a historical traffic event database is established to predict road network change patterns, enabling dynamic adaptation of spatial correlation strategies.

[0068] In this embodiment, continuing the cement transportation scenario described earlier, a cement plant in the suburbs of City A is taken as the core. Within a radius of L kilometers, there are n key stopping points (the east loading and unloading yard, the northwest temporary storage point, and the southwest logistics connection point). The transportation route covers rural roads (K0-K15), urban expressways (K15-K40), and ramps (K40-K45). Traffic congestion occurs during morning rush hour (7:00-9:00) and evening rush hour (17:00-19:00), and occasional unforeseen events such as expressway construction closures occur. A dynamic spatial association strategy is needed to ensure accurate association between key stopping points and the production site enterprises, providing a basis for subsequent transportation route planning.

[0069] 1. Traffic data collection and construction of traffic impact weighting factors: Data collection equipment was deployed on key sections of the transportation route: ① Loop sensors were installed on rural roads at K5 and K10 to collect traffic flow (vehicles / hour); ② High-definition cameras were installed on expressways at K20, K30, and K35 to obtain congestion levels through video analysis; ③ Road traffic status data was synchronously accessed from the regional traffic platform.

[0070] A weighted average algorithm was used to calculate the spatial dimension weights of each road segment: expressways, due to their high traffic volume and wide congestion impact, were assigned a weight of n1; rural roads, with lower traffic volume, were assigned a weight of n2; and ramps connecting expressways and consumer areas were assigned a weight of n3. A grey relational analysis algorithm was used to couple the temporal data of morning peak, off-peak, and evening peak hours with the spatial weights to construct traffic impact weight factors—m1 (high impact) for morning peak, m2 (low impact) for off-peak, and m3 (high impact) for evening peak—which were then used for subsequent correlation threshold adjustments.

[0071] 2. Three-level congestion early warning mechanism and dynamic adjustment of associated distance threshold: Set congestion index classification standards: ① Light congestion: such as the K25 section of the expressway during off-peak hours; ② Moderate congestion (index 0.4-0.6): such as the K8 section of the rural road during evening peak hours; ③ Severe congestion (index > 0.6): such as the K30 section of the expressway during morning peak hours.

[0072] Corresponding threshold benchmark values ​​are set: ① During mild congestion, the traffic impact weight factor is low, and the threshold remains at the benchmark value to ensure that the stopping points of normally driving vehicles are not mistakenly associated; ② During moderate congestion, the factor is increased, and the threshold is expanded—because vehicles may take detours; ③ During severe congestion, the factor is further increased, and the threshold is further expanded to avoid vehicles taking significant detours due to prolonged congestion.

[0073] 3. Road network reconstruction and optimal associated path calculation under emergency conditions: During the morning rush hour on a certain day, the K25-K30 section of the expressway was temporarily closed due to bridge maintenance (emergency situation). The real-time map update service (connected to the real-time traffic API of the Transportation Bureau of City A) synchronized the information of the closed road section to the source flow model within T4 seconds.

[0074] The model employs a hybrid strategy combining Dijkstra's algorithm and genetic algorithm: ① First, Dijkstra's algorithm is used to calculate the shortest detour route—exiting the expressway ahead of time at K25, detouring via suburban county road (X010) to K30 and re-entering the expressway, with a total detour route length of L4 kilometers; ② Then, genetic algorithm is used to optimize the traffic efficiency of the detour route, combining real-time traffic flow to select the optimal detour route of "county road X010 - township road Y005".

[0075] Meanwhile, considering the current severe traffic congestion, the associated distance threshold is maintained at L5 kilometers. The associated path between the key stop point (East Loading and Unloading Yard) and the production enterprise is recalculated. The original path was "Factory Area - Rural Road K0-K5 - Expressway K15-K25", which is reconstructed to "Factory Area - Rural Road K0-K3 - County Road X010 - Expressway K30". Although the straight-line distance between the stop point and the production enterprise increases, it is still within the L5-kilometer threshold, ensuring that the association is not interrupted.

[0076] 4. Construction of historical traffic incident database and prediction of correlation strategy: Collect traffic incident data of this transportation route in the past year: ① construction incidents; ② accident incidents; ③ holiday congestion.

[0077] Time series analysis algorithms are used to predict event patterns, adjust the correlation distance threshold in advance, and preset detour paths to avoid the lag in correlation strategy adjustment when construction emergencies occur, thus achieving proactive adaptation of correlation strategy.

[0078] Specifically, the construction of the regional logistics topology map is based on the synergy of graph neural networks, node weight assignment technology, and supply and demand data integration technology. The implementation method is as follows:

[0079] The production enterprises, logistics hubs, and consumer enterprises within the region are abstracted as nodes in a network topology. Multi-source data acquisition technology is used to collect historical transportation frequency and capacity data for these nodes. The connection strength between nodes is used as a measure of the closeness of transportation links. A node connection strength calculation model is constructed using a transportation frequency weighting algorithm. The weights of production enterprise nodes are adjusted based on capacity fluctuation data from enterprise production plans, and the weights of consumer enterprise nodes are adjusted based on changes in consumer demand. Through the iterative learning mechanism of the graph neural network, the topology structure is optimized by combining node weights and connection strength data to construct a regional logistics topology map.

[0080] Specifically, the planning of the candidate transportation routes includes:

[0081] Based on the aforementioned demand forecasting-capacity matching dual-dimensional evaluation system, this system analyzes historical consumption data and market demand trends, and uses a time-series forecasting model to intelligently predict the short-, medium-, and long-term demand scale of consumers. It employs blockchain technology to construct a shared ledger of production capacity for enterprises in production areas, synchronizing current capacity, inventory levels, and production plans in real time. A federated learning framework is used to calculate the matching degree between demand and capacity based on the intelligent forecasting. Furthermore, it introduces BeiDou spatiotemporal big data, integrating transportation costs, transportation timeliness, and road safety risk indicators to construct a dynamic risk assessment system. Digital twin technology is used to visualize and comprehensively score potential transportation routes. Finally, based on a multi-objective optimization algorithm, the candidate transportation routes are dynamically planned and prioritized using the matching degree and the comprehensive score.

[0082] In this embodiment, the transportation scenario continues from a cement plant in the suburbs of City A (daily production capacity of 5,000 tons, current inventory of 3,000 tons). The consumer side covers three core demand groups: ① Urban mixing plant B (short-term demand, requiring an average of 800 tons of cement per day for emergency road repairs, requiring delivery within 24 hours); ② Precast component plant C (medium- to long-term demand, requiring an average of 12,000 tons of cement per month for residential project construction, allowing delivery within 3-5 days); ③ Building materials market D (fluctuating demand, requiring an average of 500 tons of cement per day, adjusted according to retail orders, requiring delivery within 48 hours). There are three possible transportation routes:

[0083] Route 1: Factory area → Rural road K0-K15 → Expressway K15-K40 → Ramp K40-K45 → Consumer end (Total length 45 kilometers, passing through 2 congestion points, travel time during morning rush hour is 1.5 hours).

[0084] Route 2: Factory area → Rural road K0-K8 → County road X010 → Expressway K30-K40 → Consumer end (50 kilometers in total, passing through only 1 congestion point, travel time during morning rush hour is 1.2 hours).

[0085] Route 3: Factory area → Ring Expressway G01 → Expressway K40-K45 → Consumer end (60 km in total, no congestion points, travel time 1 hour, but expressway tolls are required).

[0086] We need to conduct a two-dimensional evaluation and multi-objective optimization to plan suitable candidate routes for the three types of consumers.

[0087] 1. Intelligent forecasting of short, medium and long-term consumer demand: Collect historical data from three types of consumer sectors: ① B. Weekly demand data of mixing plants over the past year; ② C. Monthly demand data of precast component factories over the past three years; ③ D. Daily demand data of building materials markets over the past six months.

[0088] Based on market trends (City A will launch 3 municipal emergency repair projects and 2 new residential construction projects in the second half of the year), a time-series forecasting model (based on trend fitting and cycle analysis of historical data) is used to predict: In the short term, the demand of the B mixing plant will increase to an average of m1 tons per day due to the emergency repair projects, and the demand of the D building materials market will reach m2 tons on weekends; in the medium term, the demand of the C precast component plant will rise to an average of 30,000 tons per month due to the new construction projects; in the long term, with the arrival of the rainy season, the demand of the B mixing plant will fall back to an average of m4 tons per day, and the demand of the C precast component plant will remain stable.

[0089] 2. Capacity and Demand Matching Calculation: Blockchain nodes are deployed at the factory, logistics companies, and consumer ends to build a shared capacity ledger and synchronize factory data in real time; the participating parties (factory, Bilibili, C factory, D market) do not share raw data, but only calculate the matching degree through model parameter interaction.

[0090] B. Mixing Plant (Short-term Demand): Current plant capacity can meet demand, and inventory can be readily allocated; matching degree n1%;

[0091] C. Precast Component Plant (Medium-Term Demand): Current capacity can cover the demand; considering the production increase plan for next month, the matching degree is n2%.

[0092] D. Building materials market (fluctuating demand): Production capacity and inventory are both well-matched, with a matching degree of n3%.

[0093] 3. Dynamic Risk Assessment and Route Visualization Simulation: Introducing BeiDou spatiotemporal big data (real-time vehicle location and road speed), and integrating three types of indicators to construct an assessment system:

[0094] Transportation costs: Route 1 (fuel cost 80 yuan + no toll fees), Route 2 (fuel cost 90 yuan + no toll fees), Route 3 (fuel cost 100 yuan + highway toll fee 30 yuan);

[0095] Transportation time: Morning peak: Route 1 (90 minutes), Route 2 (72 minutes), Route 3 (60 minutes); Off-peak: Route 1 (60 minutes), Route 2 (50 minutes), Route 3 (45 minutes);

[0096] Road safety risks: Route 1 (K25 section accident rate s1% in the past 3 months), Route 2 (county road X010 has no accident record), Route 3 (expressway section accident rate s2%).

[0097] A 3D simulation platform was built using digital twin technology to simulate the status of various routes at different times: During the morning rush hour, Route 1 experienced a 2-kilometer congestion at kilometer marker K25, with the simulation showing a 20-minute delay in delivery to Station B; Route 3, although more expensive, showed no congestion throughout the route, ensuring delivery to Station B within 24 hours. A comprehensive score was calculated based on the weights of "cost (30%) + timeliness (40%) + safety (30%)".

[0098] Morning rush hour: Route 1 (p1 minutes), Route 2 (p2 minutes), Route 3 (p3 minutes);

[0099] Off-peak: Route 1 (p4 points), Route 2 (p5 points), Route 3 (p6 points).

[0100] 4. Multi-objective optimization ranking of candidate routes: A multi-objective optimization algorithm (considering both matching degree and comprehensive score) is used to prioritize routes for different consumer ends.

[0101] B Mixing Plant (Short-term, high-efficiency demand): Morning peak: Priority route 3, secondary route 2; Off-peak: Priority route 2, secondary route 1;

[0102] C. Precast component plant (medium to long term, low cost demand): preferred route 2, secondary route;

[0103] D Building Materials Market (Fluctuating, Balanced Demand): Weekends (High Demand): Priority Route 2; Weekdays (Low Demand): Priority Route 1.

[0104] Specifically, the method for classifying candidate transportation routes through multi-dimensional feature fusion is as follows:

[0105] The system integrates multi-dimensional feature information from the transportation process and normalizes these features to map them to a unified numerical range. The processed multi-dimensional features are then input into the intelligent prediction and analysis model. A decision tree algorithm is used to mine classification rules from the data, and the nonlinear fitting ability of neural networks is used to learn feature relationships. Through a multi-model fusion strategy, the prediction results are integrated and optimized. The model outputs the candidate transportation route classification results by comprehensively considering transportation time, cost, and congestion probability, and outputs differentiated solutions based on the classification results.

[0106] Specifically, the method for updating the modeling evaluation rules is as follows:

[0107] Multi-source heterogeneous data acquisition technology is used to obtain information on transportation safety standards, logistics optimization cases, and abnormal situation handling solutions in the logistics industry. A knowledge graph construction engine is used to transform the collected multi-source heterogeneous data into nodes in the graph. Based on the semantic links of temporal and causal relationships, the associations between nodes are established to form a structured expert experience knowledge base.

[0108] When new transportation scenarios, policy adjustments, or technological updates occur, an associative reasoning process is triggered. A graph neural network algorithm is used to analyze the logical relationship between the new situation and existing rules. Combined with real-time feedback data from actual application scenarios, a reinforcement learning algorithm is used to iteratively correct the parameters of existing rules, thereby achieving adaptive dynamic updates of the modeling and evaluation rules.

[0109] In this embodiment, the transportation scenario from a cement plant in the suburbs of City A to the consumer end is continued. The initial modeling and evaluation rules have already covered the transportation costs, timeliness, and safety risk assessments of traditional fuel-powered heavy trucks (such as the route scoring rules mentioned above). With industry development, three types of situations require rule updates: ① New transportation scenarios (the plant introduces 10 new energy heavy trucks, requiring consideration of charging time and charging station distribution); ② Policy adjustments (City A issues new regulations, reducing the weight limit for highway freight vehicles from 55 tons to 50 tons, affecting transport load and costs); ③ Technological updates (the implementation of BeiDou-3 positioning technology improves positioning accuracy, enabling more precise identification of road risk points). Adaptive updates of the evaluation rules need to be achieved through the construction of a multi-source knowledge base and dynamic reasoning.

[0110] 1. Constructing an expert experience knowledge base (knowledge graph and semantic association): Three types of core data are collected using web crawling, industry report parsing, and expert interviews. A knowledge graph construction engine is then used to transform the collected data into structured nodes and related links.

[0111] Key nodes: "New energy heavy trucks", "50-ton weight limit", "BeiDou-3 positioning", "charging cost", "load capacity", "risk point identification accuracy", and "charging time";

[0112] The temporal sequence is as follows: "Policy release (October 2024) → Weight limit implementation (November 2024) → Load capacity adjustment (November 2024) → Changes in transportation costs (from November 2024 onwards)".

[0113] Causal relationship: "Use of new energy heavy trucks → charging time required → extended transportation time → need to adjust timeliness assessment weight"; "Improved positioning accuracy of Beidou-3 → more accurate risk point identification → lower road safety risk scoring threshold";

[0114] Ultimately, an expert experience knowledge base is formed to provide knowledge support for rule updates.

[0115] 2. Triggering association reasoning and rule iteration correction: (1) New transportation scenario trigger (new energy heavy trucks are put into use), the factory puts 3 new energy heavy trucks into B mixing station for transportation (route 2), the first "new energy heavy truck + cement transportation" scenario appears, the system detects that there is no corresponding evaluation rule in the knowledge base, triggering association reasoning; through algorithm mining, the logical association between the "new energy heavy truck" node and the existing rules is explored - the existing cost rule only includes fuel cost, so the "charging cost" node needs to be associated; the existing timeliness rule does not include charging time, so the "charging time" node needs to be associated;

[0116] Collect transportation data of new energy heavy trucks in the first month as real-time feedback data; use reinforcement learning algorithms to adjust rule parameters: add a weight for "charging cost" (accounting for 10% of the total cost assessment).

[0117] (2) Policy adjustment trigger (weight limit reduced from 55 tons to 50 tons). City A implemented a new regulation on the 50-ton weight limit starting November 1, 2024. The system detected the change in the "weight limit policy" node attribute, triggering correlation reasoning. Analyze the causal relationship between "50-ton weight limit" and "loading capacity", "transportation cost" and "route selection" - the loading capacity is reduced from 55 tons / vehicle to 50 tons / vehicle, the single trip transportation volume is reduced, the transportation frequency needs to be increased, which may lead to an increase in cost;

[0118] After collecting transportation data for one month following the weight limit, the rules were revised, and a "loading capacity coefficient" was added. For precast component plant C (medium- to long-term demand), route 2 with suitable load capacity was prioritized (the original loading capacity of 50 tons just met the new regulations, and no increase in frequency was required). The priority of route 2 was consolidated from the original No. 1 to a mandatory route.

[0119] Specifically, the method for implementing the visualized digital mapping is as follows:

[0120] Based on satellite remote sensing technology, geospatial data in the transportation network is collected. Combined with computer vision and 3D reconstruction technology, the terrain and building structure are reconstructed to build a 3D entity model. The transportation route data of bulk products is preprocessed through edge computing and federated learning technology. The dynamic data is mapped to the 3D entity model, and spatiotemporal heat maps and topological relationship maps are introduced. Paths with different transportation flow or transportation status are distinguished by dynamic color coding.

[0121] Specifically, the mining of the potential associations includes:

[0122] This process involves acquiring dynamic production data, cargo transportation information, and core operational data indicators from enterprises to construct a multi-source heterogeneous dataset. Data cleaning techniques are used to remove duplicate records and logically erroneous data. Unstructured text data is converted into structured data storage through format transformation. Outlier detection algorithms are used to identify and correct abnormal data. Association rule mining algorithms are employed to establish a quantitative relationship model between production capacity fluctuations of enterprises in production areas and container throughput and railway freight volume at logistics hubs. Route analysis algorithms are used to analyze the impact of seasonal changes in consumer demand on transportation route selection preferences. Finally, spatiotemporal clustering algorithms are used to identify core hub nodes in the logistics network within a specific time period.

[0123] Specifically, the dynamic deduction of the source-flow-direction relationship includes:

[0124] Based on data mining technology, multi-dimensional data on production capacity fluctuations, regional market demand changes, industry policy adjustments, and natural disaster frequency are extracted from historical source-flow-direction databases. Granger causality tests are used to quantitatively analyze the impact of each factor on changes in transportation routes. A visualized causal relationship network is constructed using graph neural network algorithms, and dynamic impact weights are assigned to factors such as capacity utilization rate, demand gap rate, and policy implementation intensity through node weight matrices.

[0125] An elastic response mechanism to external factors is constructed, taking sudden policy adjustments and natural disasters as exogenous variables; a probability distribution map of transportation routes is generated using the Monte Carlo simulation method, and combined with a time series prediction model, the evolution trend of the source-flow-direction relationship is dynamically predicted, identifying key nodes and transportation corridors of route changes.

[0126] In this embodiment, the transportation scenario continues from a cement plant in the suburbs of City A (with a daily production capacity of 5,000 tons) to the consumer end (mixing plant B, precast component plant C, and building materials market D), involving three core routes (Route 1: Plant → Rural Road → Expressway; Route 2: Plant → County Road → Expressway; Route 3: Plant → Ring Expressway → Expressway). Dynamic simulation is needed to predict changes in the source and flow of the following scenarios: ① Surge in demand during peak season (construction of residential projects at Plant C, leading to increased demand); ② Sudden environmental policies (City A issues a heavy pollution weather warning, resulting in production restrictions at the plant); ③ Natural disasters during the rainy season (continuous heavy rain causes flooding and interruption of the rural road section of Route 1). Through multi-dimensional data and simulation technology, key nodes in the route changes are identified to ensure stable transportation.

[0127] Multi-dimensional data mining and Granger causality test: Extracting four types of core data from the source-flow-direction database of the factory area over the past 3 years:

[0128] Production capacity: monthly capacity (6,000 tons / day in peak season, 4,000 tons / day in off-season, 4,000 tons / day due to environmental restrictions); Demand: monthly demand for Plant B (fluctuation ±m%), Plant C (increase n% in peak season), and Market D (increase p% on weekends); Policies: environmental restrictions (2 times per year, capacity decreases by t%), weight restrictions in November 2024 (load volume decreases by k%); Disasters: heavy rain from June to August (3 times per year, Route 1 interrupted for 2-3 days each time), and heavy snow in winter (Route 3 closed for 1-2 days each time).

[0129] According to Granger causality tests, for every n1% increase in the demand gap of Plant C, the frequency of choosing Route 2 increases by m1%; when policies or environmental protection-related production restrictions lead to capacity limitations, the selection rate of Route 3 increases from m1% to m3%; when disasters cause Route 1 to be interrupted, the substitution rate of Route 2 increases by p1%; and the impact of non-production restrictions on capacity fluctuations is not significant.

[0130] 2. Visualization of Causal Relationship Network Construction: Based on the graph neural network algorithm, a causal network with core nodes and dynamic weights is constructed: the core nodes are capacity utilization rate, demand gap rate, policy intensity, and disaster scope; the dynamic weights are: peak season demand weight k1 is the highest, environmental protection policy weight k2 is the highest, and rainy season disaster weight k3 is the highest; the weights are represented by node size, and the thickness of the arrows indicates the intensity of the impact for visualization.

[0131] 3. Construction of an elastic response mechanism for external factors: response rules for exogenous variables, environmental protection early warning, production capacity decline, shortage at Plant C, priority route 3, supplementary route 2, suspension of route 1; rainstorm and water accumulation, order from route 1 is switched to route 2, and loading and unloading personnel are reassigned.

[0132] 4. Monte Carlo Simulation and Trend Forecasting: Simulate the probability distribution of transport routes, predict source flow trends, and identify key nodes. Combined with ARIMA model predictions, the trend is that the selection rate for Route 2 rises to t3% in June, peaks at t4% in July, and then rebounds to t5% in August.

[0133] The key nodes are the connection point of Route 2, the highway entrance of Route 3, and the K8 section of Route 1 (which is prone to interruption due to heavy rain).

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A bulk product source flow direction modeling method based on point source inventory and Beidou trajectory, characterized in that, include: S1: Based on point source emission inventories and satellite remote sensing data, artificial intelligence image recognition technology is used to extract enterprise geographical location, production scale and product output information to build a database of bulk cargo sources; 5G cellular positioning data and Beidou trajectory data of freight vehicles are obtained, and spatiotemporal fusion algorithms are used to complete data cleaning and trajectory reconstruction, and key stopping points are identified based on the trajectory pattern recognition engine; The bulk commodity source database is constructed through a combination of point source list data preprocessing technology, satellite remote sensing data parsing technology, and dynamic data verification technology. Specifically, this includes: standardizing the fields of enterprise data in the point source emission list, removing duplicate enterprise information and invalid production capacity data, and unifying industry classification standards and data formats; optimizing the texture features of satellite remote sensing data using image enhancement technology, and combining terrain correction technology to eliminate the impact of terrain undulations on enterprise location, extracting spatial range parameters of enterprise factory buildings and storage areas; using entity matching technology to correlate and compare enterprise names and industry types in the point source emission list with the spatial information of enterprises extracted by remote sensing, correcting geographical coordinate offsets and production scale statistical deviations in the list, and outputting a geographically calibrated-production fusion dataset; and establishing a periodic data update mechanism to collect monthly production reports and inventory data from enterprises, and synchronously updating product output and inventory balance information in the database. The method for data cleaning and trajectory reconstruction using a spatiotemporal fusion algorithm is as follows: Signal strength parameters of the 5G cellular positioning data and BeiDou trajectory data of freight vehicles are collected, and fusion weight coefficients are paired based on a dynamic weight allocation mechanism; a three-dimensional road network constraint model is constructed, and a road network topology feature vector is generated by combining the road orientation, lane number, and node distribution of the regional road network; the fused trajectory segments are smoothed based on semantic understanding, and trajectory inflection points that do not conform to the actual road network are identified and removed through adversarial generative networks to correct trajectory deviations; incremental trajectory reconstruction technology is adopted, combined with edge computing, to achieve real-time cleaning and trajectory reconstruction of positioning data. The identification of key stopping points is achieved by: collecting vehicle speed, acceleration, and direction change data through the trajectory pattern recognition engine, establishing a driving state characteristic threshold model, and capturing the state transition nodes of the vehicle from driving to stationary; using motion parameter analysis technology to perform secondary verification on the state transition nodes; constructing a quantitative index system for stopping characteristics by calculating the stopping time and statistically analyzing the vehicle density per unit area; and establishing a Bayesian classification model by combining the layout of loading and unloading sites and parking lot location characteristics around bulk cargo sources in the geographic information spatial database, and distinguishing between temporary stops and loading / unloading stops by analyzing the spatial distance between stopping points and facilities and the matching degree of historical stopping patterns, thereby identifying key stopping points related to cargo sources. S2: Construct a dynamic spatial matching model, combining real-time traffic conditions and road network changes to adjust the spatial association strategy between production enterprises and the key stopping points; construct a regional logistics topology map based on graph neural networks, integrate enterprise production plans and consumer demand forecast data to generate potential transportation routes; plan candidate transportation routes based on a demand forecasting-capacity matching dual-dimensional evaluation system, combined with the spatial association strategy; wherein, the adjustment of the spatial association strategy includes: Traffic flow, congestion levels, and road conditions data are collected within the region. A traffic impact weighting factor, incorporating both temporal and spatial dimensions, is constructed using a weighted average algorithm and a grey relational analysis algorithm. A three-tiered congestion early warning mechanism is established, dynamically adjusting the correlation distance threshold between production enterprises and key stopping points based on congestion levels. In the event of unforeseen circumstances, the bulk product source-flow model utilizes real-time map updates to capture changes in road network topology. A hybrid path planning strategy combining Dijkstra's algorithm and genetic algorithm is invoked, and the road network is reconstructed using the correlation distance threshold. The optimal correlation path between key stopping points and production enterprises is recalculated. Simultaneously, a historical traffic event database is established to predict road network change patterns, enabling dynamic adaptation of spatial correlation strategies. S3: Based on blockchain technology, collect information on product category, transportation distance, and vehicle type corresponding to the candidate transportation routes; construct an intelligent predictive analysis model, classify the candidate transportation routes through multi-dimensional feature fusion, and formulate differentiated solutions based on the classification results; use adversarial generative networks to generate virtual paths for comparison and verification, and dynamically update the preset modeling and evaluation rules based on expert knowledge graphs; S4: Based on digital twin technology, a three-dimensional visualization platform is built to visualize and digitally map the actual transportation path distribution of bulk products; using spatiotemporal big data analysis technology, multi-source heterogeneous data is integrated to mine the potential connections between production enterprises, regional logistics hubs and consumer enterprises, and the source-flow-direction relationship is dynamically extrapolated using a causal inference model to predict the direction of transportation path optimization.

2. The method of claim 1, wherein, The construction of the regional logistics topology map described in S2 is based on the synergy of graph neural networks, node weight assignment technology, and supply and demand data integration technology. The implementation method is as follows: The production enterprises, logistics hubs, and consumer enterprises in the region are abstracted as nodes in the network topology, and multi-source data acquisition technology is used to collect historical transportation frequency and transportation capacity data of the nodes. Using the connection strength between nodes as a measure of the closeness of transportation links, a node connection strength calculation model is constructed through a transportation frequency weighting algorithm. The node weights of enterprises in the production area are adjusted in combination with the capacity fluctuation data in the enterprise's production plan, and the node weights of enterprises in the consumption area are adjusted according to changes in consumer demand. By using the iterative learning mechanism of the graph neural network, and combining node weights and connection strength data, the topology graph structure is optimized to construct a regional logistics topology graph.

3. The method of claim 1, wherein, The planning of the candidate transportation routes described in S2 includes: Based on the aforementioned demand forecasting-capacity matching dual-dimensional evaluation system, this system analyzes historical consumption data and market demand trends, and uses a time-series forecasting model to intelligently predict the short-, medium-, and long-term demand scale of consumers. It employs blockchain technology to construct a shared ledger of production capacity for enterprises in production areas, synchronizing current capacity, inventory levels, and production plans in real time. A federated learning framework is used to calculate the matching degree between demand and capacity based on the intelligent forecasting. Furthermore, it introduces BeiDou spatiotemporal big data, integrating transportation costs, transportation timeliness, and road safety risk indicators to construct a dynamic risk assessment system. Digital twin technology is used to visualize and comprehensively score potential transportation routes. Finally, based on a multi-objective optimization algorithm, the candidate transportation routes are dynamically planned and prioritized using the matching degree and the comprehensive score.

4. The method of claim 1, wherein, The method for classifying candidate transportation routes through multi-dimensional feature fusion, as described in S3, is as follows: The system integrates multidimensional feature information from the transportation process, normalizes the multidimensional features, and maps them to a unified numerical range. The processed multidimensional features are then input into the intelligent prediction and analysis model, which uses a decision tree algorithm to mine classification rules from the data and learns feature relationships based on the nonlinear fitting ability of neural networks. By employing a multi-model fusion strategy, the prediction results are integrated and optimized, and the candidate transportation routes are classified based on multiple dimensions, including transportation time, cost, and congestion probability. Differentiated solutions are then output based on the classification results.

5. The method of claim 1, wherein, The method for updating the modeling evaluation rules described in S3 is as follows: We use multi-source heterogeneous data acquisition technology to obtain information on transportation safety standards, logistics optimization cases and abnormal situation handling solutions in the logistics industry, and use a knowledge graph construction engine to transform the collected multi-source heterogeneous data into nodes in the graph. Based on the semantic links of temporal and causal relationships, the associations between nodes are established to form a structured expert experience knowledge base; When new transportation scenarios, policy adjustments, or technological updates occur, an associative reasoning process is triggered. A graph neural network algorithm is used to analyze the logical relationship between the new situation and existing rules. Combined with real-time feedback data from actual application scenarios, a reinforcement learning algorithm is used to iteratively correct the parameters of existing rules, thereby achieving adaptive dynamic updates of the modeling and evaluation rules.

6. The method of claim 1, wherein, The method for implementing the visualized digital mapping described in S4 is as follows: Based on satellite remote sensing technology, geospatial data in the transportation network is collected. Combined with computer vision and 3D reconstruction technology, the terrain and building structure are reconstructed to build a 3D entity model. The transportation route data of bulk products is preprocessed through edge computing and federated learning technology. The dynamic data is mapped to the 3D entity model, and spatiotemporal heat maps and topological relationship maps are introduced. Paths with different transportation flow or transportation status are distinguished by dynamic color coding.

7. The method of claim 1, wherein, The mining of potential associations described in S4 includes: This process involves acquiring dynamic production data, cargo transportation information, and core operational data indicators from enterprises to construct a multi-source heterogeneous dataset. Data cleaning techniques are used to remove duplicate records and logically erroneous data. Unstructured text data is converted into structured data storage through format transformation. Outlier detection algorithms are used to identify and correct abnormal data. Association rule mining algorithms are employed to establish a quantitative relationship model between production capacity fluctuations of enterprises in production areas and container throughput and railway freight volume at logistics hubs. Route analysis algorithms are used to analyze the impact of seasonal changes in consumer demand on transportation route selection preferences. Finally, spatiotemporal clustering algorithms are used to identify core hub nodes in the logistics network within a specific time period.

8. The method of claim 1, wherein, The dynamic deduction of the source-flow-direction relationship described in S4 includes: Based on data mining technology, multi-dimensional data on production capacity fluctuations, regional market demand changes, industry policy adjustments, and natural disaster frequency are extracted from historical source-flow-direction databases. Granger causality tests are used to quantitatively analyze the impact of each factor on changes in transportation routes. A visualized causal relationship network is constructed using graph neural network algorithms, and dynamic impact weights are assigned to factors such as capacity utilization rate, demand gap rate, and policy implementation intensity through node weight matrices. An elastic response mechanism to external factors is constructed, taking sudden policy adjustments and natural disasters as exogenous variables; a probability distribution map of transportation routes is generated using the Monte Carlo simulation method, and combined with a time series prediction model, the evolution trend of the source-flow-direction relationship is dynamically predicted, identifying key nodes and transportation corridors of route changes.

Images