A logistics management system based on industrial big data

Abstract

The application discloses a logistics management system based on industrial big data, and relates to the technical field of industrial big data.The system comprises a source-side bypass agent layer, a semantic mapping and global view layer, a federal query and virtual index layer, a fault-tolerant compensation layer and a unified search and management application layer.The bypass agent reads only monitors local databases such as enterprise resource planning, manufacturing execution, warehouse management, transportation management and an Internet of Things platform, and does not need to physically migrate data.Equivalent mapping from local fields to a global view is established through an ontology graph and a semantic automatic mapping engine.Global queries are decoupled into local sub-queries and pushed to local execution through a cross-modal virtual index tree, and only the result set is returned to complete memory-level assembly.The application realizes cross-system low-delay unified view search, reduces mapping maintenance costs and source system load, and guarantees the continuity of full-link search in a weak network environment, and can be used for intelligent logistics full-link scheduling decision-making.

Description

Technical Field

[0001] This invention relates to the field of industrial control system technology, specifically to a logistics management system based on industrial big data. Background Technology

[0002] With the deepening development of the Industrial Internet, the value discovery of industrial big data has become an industry consensus. In actual industrial logistics management scenarios, covering the entire chain of raw material procurement, production line collaboration, finished product warehousing, trunk transportation and last-mile delivery, multiple information systems of different ages, architectures and developers usually coexist within enterprises and in the upstream and downstream of the industrial chain. These include enterprise resource planning systems, manufacturing execution systems, warehouse management systems, transportation management systems and various IoT device platforms. These systems maintain their own local databases and use different data model standards. For example, material codes are not unified and order status definitions are inconsistent. They are physically isolated and logically fragmented, forming data silos, making it difficult for enterprise managers to obtain a unified logistics view covering the entire chain.

[0003] Currently, the closest implementation to this invention in the industry is a logistics big data fusion solution based on a centralized data warehouse or data lake. This solution typically involves a data engineering team developing customized data extraction scripts for each business system. Incremental and full data are periodically extracted from the databases of each business system through scheduled batch processing tasks. The extracted raw data is then preprocessed through cleaning, format conversion, field mapping, and other operations to be uniformly converted into a target pattern predefined by the central data warehouse. The converted data is then written into the centralized physical data warehouse or data lake storage medium through loading jobs to form a physically aggregated full data copy. Managers or downstream analysis applications can directly initiate a unified query on the central data warehouse to obtain a cross-system logistics global view data. The system structure of this solution mainly consists of a four-layer architecture: a multi-source business system layer, a data extraction, conversion, loading, and scheduling layer, a central data warehouse layer, and a unified query and display layer.

[0004] The aforementioned existing technical solutions have the following shortcomings: First, because this solution relies on periodic batch processing jobs to physically migrate data to the central warehouse, it inevitably introduces a time window between data generation and queryability. Data updates are delayed, and when abnormal logistics events such as transportation disruptions or inventory shortages occur, managers find it difficult to make scheduling decisions based on real-time data. Second, because this solution requires data from each business system to be pre-converted to a target model uniformly defined by the central warehouse before physical aggregation, and the data model standards between different systems are not uniform, every time a new business system is added or an existing system undergoes a version upgrade, data engineers need to re-organize and hard-code the mapping relationships, resulting in poor system scalability and increased maintenance manpower costs as the number of systems increases. Thirdly, because this solution requires physically copying detailed data from each business system to the central warehouse through extraction operations, it causes redundant occupation of storage resources. On the other hand, the high-frequency extraction operation itself generates additional read load on the source system database, which poses a risk of performance fluctuations on the database that carries core production business. Fourthly, this solution usually assumes that the network connection of each business system is stable and the service is continuously available. When the network is unstable or the system is abnormal at edge nodes such as mobile logistics vehicles and remote warehouses, this solution lacks a corresponding fault tolerance mechanism, which causes the global query to be interrupted or return incomplete results when encountering local node abnormalities, making it difficult to guarantee the continuous availability of the entire logistics retrieval. Therefore, we propose a logistics management system based on industrial big data. Summary of the Invention

[0005] The purpose of this invention is to provide a logistics management system based on industrial big data.

[0006] The technical problem this invention aims to solve is: In industrial logistics scenarios, the data models of multiple heterogeneous business systems existing within enterprises and in the upstream and downstream of the industrial chain are inconsistent and physically isolated. Traditional physical data aggregation solutions have problems such as poor real-time performance, high mapping and maintenance costs, and large impact on the load of source systems, which makes it impossible to achieve low-latency unified view retrieval of logistics information across the entire chain. In particular, it is difficult to support real-time scheduling decisions when logistics anomalies occur.

[0007] Meanwhile, in environments with weak networks or unstable nodes, such as mobile logistics vehicles and remote warehouses, existing cross-system retrieval solutions lack fault tolerance and result compensation capabilities, which can easily lead to interruptions or data loss in the global query chain.

[0008] To address the aforementioned technical problems, this invention provides a logistics management system based on industrial big data, comprising a source-side bypass agent layer, a semantic mapping and global view layer, a federated query and virtual index layer, a fault tolerance compensation layer, and a unified retrieval and management application layer.

[0009] The source-side bypass agent layer includes multiple bypass agents deployed in the enterprise resource planning system, manufacturing execution system, warehouse management system, transportation management system, and IoT device platform. Each bypass agent is connected to the local database of the corresponding business system through a database binary log listening interface, which is used to listen to the local database in read-only mode, obtain data change events, and not write data to the local database or perform physical data migration.

[0010] The bypass agent internally constructs an in-memory inverted index and a spatiotemporal R-tree index. The in-memory inverted index is used to store index information of frequently queried fields, and the spatiotemporal R-tree index is used to store index information of spatiotemporal data.

[0011] The semantic mapping and global view layer are connected to the source-end bypass proxy layer, including an industrial logistics basic ontology graph module and a semantic automatic mapping engine;

[0012] The semantic mapping and global view layer is used to extract metadata from each of the local databases, calculate the similarity between the metadata and the standard concept nodes in the industrial logistics basic ontology graph using the semantic automatic mapping engine, establish an equivalent mapping relationship from local fields to the global unified logical view based on the similarity determination result, and generate a global mapping relationship table.

[0013] The federated query and virtual index layer is connected to the semantic mapping and global view layer, and includes a distributed query engine, a cross-modal global virtual index tree, and an operator pushdown scheduling module;

[0014] The cross-modal global virtual index tree uses the logistics entity identifier as the primary key, and associates it with the time-series trajectory branch index node and the relation entity branch index node respectively. Each index node stores the position pointer of the memory index pointing to the corresponding bypass agent, but does not store the actual business data.

[0015] The distributed query engine is used to decouple global query requests into local subqueries corresponding to different business systems based on the global mapping relationship table. The operator pushdown scheduling module pushes down and distributes each local subquery to the corresponding bypass agent. The bypass agent performs filtering and aggregation operations on the local in-memory inverted index and spatiotemporal R-tree index to generate a local result set. The local result set is then sent back to the federated query and virtual index layer for memory-level association and assembly to obtain the global logistics view result.

[0016] The fault tolerance compensation layer is connected to the federated query and virtual index layer through a bidirectional channel, including a connection status monitoring module, a Markov state transition prediction submodule, and a Kalman filter trajectory completion submodule.

[0017] The connection status monitoring module is used to monitor the heartbeat signal and query response latency of each bypass agent. When a connection anomaly is detected in a bypass agent, a compensation process is triggered. The Markov state transition prediction submodule predicts the state distribution based on historical state transition probabilities for discrete business state fields. The Kalman filter trajectory completion submodule performs trajectory extrapolation and completion based on historical trajectories for continuous spatiotemporal fields. The confidence value is calculated based on the duration of missing data. The compensation result and the confidence value are encapsulated together and output.

[0018] The unified retrieval and management application layer is connected to the fault tolerance compensation layer and the federated query and virtual index layer, and is used to provide managers with a global logistics tracking, anomaly warning and scheduling decision interface.

[0019] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0020] 1. This invention utilizes a semantic mapping mechanism built upon semantic mapping and a global view layer. This mechanism eliminates the need for the system to migrate data from various heterogeneous business systems at the physical layer. Instead, it automatically establishes equivalent mapping relationships between local fields and the global unified logical view simply by extracting local database metadata. Combined with low-code manual confirmation feedback for closed-loop processing of long-tail fields, this shortens the access cycle for new business systems and reduces mapping maintenance costs. It overcomes the problems of poor scalability and high operation and maintenance costs caused by manual sorting and hard-coded mapping relationships in existing technologies.

[0021] 2. This invention utilizes a cross-modal global virtual index tree constructed through federated queries and a virtual index layer, along with an operator pushdown mechanism, to decentralize query processing to local execution in each business system. Only partial result sets are returned to the central layer for assembly, avoiding high-frequency extraction and access to the source system database and full physical copying. While achieving cross-system global logistics status retrieval, it reduces performance interference with the normal business of the source system, overcoming the problems of high data update latency and large load impact on the source system in existing technologies.

[0022] 3. This invention utilizes the Markov state transition prediction mechanism and Kalman filter trajectory completion mechanism built into the fault-tolerant compensation layer. Even when a connection anomaly occurs in a local bypass agent, it can still generate compensation data with confidence labels based on historical spatiotemporal cache, and automatically switch back to real data after the connection is restored. Compared with existing technologies that directly interrupt queries or return null results when nodes are abnormal, this invention enables global queries to return approximate results with complete structure and confidence labels during node anomalies through state prediction and trajectory extrapolation. This ensures the continuity of global logistics link retrieval in weak network environments such as mobile logistics vehicles and remote warehouses, and overcomes the problem of global query interruption or result loss caused by local node anomalies in existing technologies. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the data flow of the five-layer overall architecture of the system in this embodiment of the invention;

[0024] Figure 2 This is a schematic diagram of the semantic automatic mapping processing flow in an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the fault tolerance compensation process for weak network anomalies in an embodiment of the present invention. Detailed Implementation

[0026] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0027] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0028] I. System Overall Architecture:

[0029] The logistics management system based on industrial big data provided by this invention adopts a five-layer architecture: source-side bypass proxy layer, semantic mapping and global view layer, federated query and virtual index layer, fault tolerance compensation layer, and unified retrieval and management application layer.

[0030] In the multi-source heterogeneous business system layer, there are enterprise resource planning systems, manufacturing execution systems, warehouse management systems, transportation management systems, and several IoT device platforms. Each system has its own independent local database, and there is no need for any physical data migration between them. A lightweight bypass agent is deployed next to each business system. The bypass agent is connected to the database of the corresponding business system through the database binary log listening interface. It performs read-only listening to the table data change events of the local database, does not perform write operations on the local database, and does not physically copy the local data to other storage media. Thus, it achieves the technical feature of zero copying in the data acquisition stage. The bypass agent internally has an in-memory inverted index for high-frequency query fields and a spatiotemporal R-tree index for spatiotemporal data to support subsequent localized query execution.

[0031] Each bypass agent connects upwards to the semantic mapping and global view layer. This layer contains an industrial logistics basic ontology graph module and a semantic automatic mapping engine driven by a logistics domain language model. It is used to map the local data views of each local system to a unified global logical view. The semantic mapping layer connects upwards to the federated query and virtual index layer. This layer contains a distributed query engine, a cross-modal global virtual index tree, and an operator pushdown scheduling module. It is responsible for decoupling the global query initiated by the central layer into multiple local subqueries and pushing them down to each bypass agent for execution. The federated query layer and the fault tolerance compensation layer are connected through a bidirectional channel. The fault tolerance compensation layer has a built-in Markov state transition prediction submodule and a trajectory completion submodule based on Kalman filtering. When a bypass agent connection timeout or disconnection is detected, this layer generates a compensation result with confidence label based on historical spatiotemporal cache. The top layer of the system is the unified retrieval and management application layer, which provides managers with a global logistics tracking, anomaly warning, and scheduling decision interface. The layers are connected by data flow and control flow. The bypass agents and the business system database exhibit zero-copy characteristics of read-only listening, no writing, and no migration.

[0032] II. Automatic Semantic Mapping Method for Logistics Across Heterogeneous Systems:

[0033] This invention provides an automatic semantic mapping method for logistics across heterogeneous systems. The system does not need to migrate data from various heterogeneous business systems at the physical layer. Instead, it extracts local database metadata and uses a logistics domain language model to automatically establish a semantic mapping from a local view to a globally unified logical view. The method includes the following steps.

[0034] The first step is metadata collection: The bypass agent periodically or event-triggeredly reads the table structure definition of the local business system database through a standard database interface, extracting table names, column names, field comments, and sample value sets of fields to form a local metadata set, denoted as the set. ,in Indicates the first Metadata description of the fields to be mapped;

[0035] The second step is semantic vectorization: This involves vectorizing each field in the metadata collection. The table name, column name, and annotation text are input into a logistics domain language model to obtain the semantic embedding vector of that field. Simultaneously, for each standard concept node in the global ontology graph Concept vectors are obtained by encoding using the same model. ;

[0036] The third step is similarity calculation: calculate the field to be mapped according to the following formula. With ontology concept Overall similarity between them:

[0037] ;

[0038] In the formula, Cosine similarity, representing the semantic vector of a field and the semantic vector of a concept, is used to measure their semantic similarity. Representation field Sample value word set With concept Example lexical set The Jaccard similarity coefficient between vectors is used to supplement the similarity features of value distribution beyond vector semantics. and The weighting coefficients are preset and satisfy the following conditions: In this embodiment, we take 0.7 The value is 0.3. By assigning a higher weight to semantic vector similarity and a lower weight to value distribution similarity, both the semantic information of field naming and the distribution characteristics of actual stored data are taken into account.

[0039] Step 4, Threshold determination and equivalence link establishment: Set the mapping determination threshold. In this embodiment, the value is taken as 0.85, when the field With concept When the overall similarity between the two is not less than the mapping judgment threshold, it is determined that an equivalent link relationship is established between them, and the equivalent link relationship is written into the global mapping relationship table. When the overall similarity is less than the mapping judgment threshold, the field is marked as pending manual confirmation.

[0040] Step 5, manual confirmation of long-tail fields: For the fields marked as pending confirmation in step 4, a list of candidate ontology concepts is presented through a low-code visualization interface. The list of candidate ontology concepts is arranged in descending order of similarity score, allowing data engineers to drag and select the correct mapping target. The confirmation results are synchronously written to the global mapping relationship table.

[0041] Step 6, Ontology Graph Feedback Update: The newly confirmed mapping samples in Step 5 are fed back as incremental training data to the semantic vectorization model and ontology graph, which is used to improve the accuracy of automatic matching in subsequent field mapping, forming a continuous learning loop;

[0042] For example, after the second to fourth steps of calculation, the fields Waybill_No in the transportation management system and Outbound_ID in the warehouse management system can be automatically determined to be semantically equivalent, and a unified mapping can be established to the logistics waybill identifier concept node in the ontology graph, thus eliminating the need for manual pre-sorting of the field mapping relationship.

[0043] III. Federated Query Pushdown and Cross-Modal Virtual Indexing Methods:

[0044] This invention constructs a cross-modal global virtual index tree for time-series trajectory data and relational entity data. When the manager initiates a global logistics tracking query, the system's distributed query engine decouples the query into multiple local sub-queries. Using operator pushdown technology, filtering and aggregation operations are performed directly on the bypass proxy memory index corresponding to the original system database. Only the processed local result set is returned to the central layer for memory-level association and assembly. The method includes the following steps.

[0045] The first step is query reception and parsing: The system query access module receives retrieval requests for global logistics entities and restores the unified field names in the request to the local field names of each business system based on the global mapping relationship table.

[0046] The second step is execution plan decomposition: Based on the data sources involved in the request, the distributed query engine decomposes the global query plan into several sets of local subqueries, denoted as... Each subquery Local filtering and aggregation operations corresponding to a specific business system;

[0047] The third step is virtual index positioning: The query engine retrieves data in the cross-modal global virtual index tree using the logistics entity identifier as the primary key. This virtual index tree is associated with two types of sub-index nodes: the time-series trajectory branch index node and the relational entity branch index node. Each index node only stores a position pointer to the corresponding bypass agent memory index and does not store actual business data. Based on this, the position pointer of the bypass agent corresponding to each subquery is obtained.

[0048] Step 4, Operator pushdown execution: Based on the position pointers obtained in step 3, each subquery... The results are pushed down to the corresponding bypass agents, and each agent performs filtering and aggregation operations directly in its local memory index layer to generate a local result set. The entire process avoids accessing the main database of the business system, thereby avoiding impacting its online transaction processing performance;

[0049] Step 5, Result Set Feedback: Each bypass agent only sends back the processed partial result set. The data is transmitted back to the central layer via a cross-network channel, instead of the original detailed data.

[0050] Step 6, Memory-level Associative Assembly: The central layer, based on the global mapping table, assembles each local result set. Perform associative assembly operations in memory to generate a unified global logistics view result and return it to the query access module;

[0051] To measure the overall cost of query processing and guide execution plan optimization, this embodiment defines the total query cost function as follows:

[0052] ;

[0053] In the formula, Indicates the first The processing cost of performing filtering and aggregation operations on the corresponding bypass proxy's local in-memory index for each subquery. This represents the processing cost of the central layer performing memory-level joins and assemblies of each local result set. The query engine dynamically adjusts the pushdown order and parallelism of subqueries based on this cost function to reduce the overall overhead of query processing.

[0054] IV. Query Fault Tolerance and Result Compensation Methods for Industrial Weak Network Environments:

[0055] To address potential network instability or system anomalies at nodes such as mobile logistics vehicles and remote warehouses, this invention designs a data compensation and prediction method based on Markov chains and historical spatiotemporal caching. In cases of local system query timeouts or connection drops, it provides compensation data with confidence through global virtual view caching and spatiotemporal extrapolation. The method includes the following steps.

[0056] The first step is connection anomaly detection: The fault tolerance compensation layer continuously monitors the heartbeat response of each bypass agent. When an agent fails to return the subquery result within a preset timeout period, it is determined that the node is in a connection anomaly state and the compensation process is triggered.

[0057] The second step is historical state sequence extraction: extract the state observation sequence of the abnormal node within a certain period of time before the disconnection from the historical spatiotemporal cache. For discrete business states, extract the state number sequence; for continuous spatiotemporal data, extract the location and timestamp sequence.

[0058] The third step is discrete-state Markov prediction: For discrete business states, based on the state transition probability matrix obtained from historical state sequences, the state distribution of abnormal nodes in the missing time period is predicted according to the following formula:

[0059] ;

[0060] In the formula, Indicates to The predicted estimate of the state at time step. This represents the total number of possible states in the discrete state space. Indicates the first A possible discrete state value. Indicates that the system has gone through After a time step, the current state is... Transition to state The multi-step transition probability is obtained by exponentiation of the historical state transition matrix;

[0061] Step 4, Continuous Spatiotemporal Kalman Completion: For continuous spatiotemporal data such as vehicle position, a state transition model is established based on the preset navigation route and historical trajectory, and prediction is performed according to the following formula:

[0062] ;

[0063] ;

[0064] In the formula, Indicates to The prior prediction of the vehicle's state at any given time, where the vehicle state includes position and speed. The system state transition matrix is ​​determined by the vehicle kinematics model. express State estimate at time 10:00 Represents the control input matrix. express Time-based control input, This represents the prior estimation error covariance matrix. This represents the error covariance matrix at the previous time step. The process noise covariance matrix is ​​represented when there are observation data from adjacent vehicles or roadside equipment. Then, further state correction is performed according to the following formula:

[0065] ;

[0066] ;

[0067] In the formula, Represents the Kalman gain matrix. Represents the observation matrix. Represents the observation noise covariance matrix. express The actual observed value at time , This represents the corrected state estimate;

[0068] Step 5, Confidence Calculation and Labeling: Calculate the confidence level of the compensation result based on the duration of data loss, and determine the confidence level value according to the following formula:

[0069] ;

[0070] In the formula, This indicates the confidence level of the compensated data, with a value ranging from 0 to 1. This represents the preset confidence decay coefficient. This represents the time interval since the last real data reception. The longer the data is missing, the lower the confidence level. This function reflects the characteristic that the confidence level decreases monotonically as the time interval increases.

[0071] Step 6, result encapsulation and labeling return: Encapsulate the compensation estimate obtained in step 3 or 4 and the confidence value obtained in step 5 together, label them as estimated data tags and return them to the query access module for upper-layer applications to distinguish and prompt when displaying;

[0072] Step 7, Connection Recovery Exit: When the connection status monitoring module detects the recovery response of an abnormal node, it automatically terminates the compensation process. Subsequently, it queries and recovers the real data of the node and supplements the real data during the recovery period into the historical spatiotemporal cache library for subsequent compensation prediction model updates.

[0073] V. Other implementation methods:

[0074] In another embodiment of the present invention, the semantic mapping and global view layer may not adopt the method of combining the language model of the logistics domain with the ontology graph, but instead use the knowledge graph embedding technology. The field metadata of each business system and the constructed logistics knowledge graph nodes are embedded together into the same low-dimensional vector space. The distance metric between the embedded vectors is used to replace the aforementioned semantic similarity calculation, which can also realize the automatic mapping from the local view to the global view and has the ability to reason about relationships in scenarios with more complex conceptual relationships.

[0075] In another embodiment of the present invention, the incremental data capture method of the bypass agent may not adopt the method of listening to the database binary log, but instead use a message queue as a data change notification channel between the business system and the bypass agent. When the data changes, the business system actively publishes the change event to the message queue, and the bypass agent subscribes to the corresponding topic and updates the local memory index synchronously. This method is suitable for business system scenarios that do not support binary log export or have limited access to database logs, as a supplementary implementation of the present invention.

[0076] 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 possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A logistics management system based on industrial big data, characterized in that, include: The source-side bypass agent layer includes multiple bypass agents deployed in the enterprise resource planning system, manufacturing execution system, warehouse management system, transportation management system, and IoT device platform. Each bypass agent is connected to the local database of the corresponding business system through a database binary log listening interface. It is used to listen to the local database in read-only mode, obtain data change events, and does not write data to the local database or perform physical data migration. The bypass agent internally constructs an in-memory inverted index and a spatiotemporal R-tree index. The in-memory inverted index is used to store index information of frequently queried fields, and the spatiotemporal R-tree index is used to store index information of spatiotemporal data. The semantic mapping and global view layer, connected to the source-end bypass proxy layer, includes an industrial logistics basic ontology graph module and a semantic automatic mapping engine. It is used to extract metadata from each of the local databases, use the semantic automatic mapping engine to calculate the similarity between the metadata and the standard concept nodes in the industrial logistics basic ontology graph, establish an equivalent mapping relationship from local fields to the global unified logical view based on the similarity judgment result, and generate a global mapping relationship table. The federated query and virtual index layer, connected to the semantic mapping and global view layer, includes a distributed query engine, a cross-modal global virtual index tree, and an operator pushdown scheduling module. The cross-modal global virtual index tree uses the logistics entity identifier as the primary key and associates it with time-series trajectory branch index nodes and relation entity branch index nodes respectively. Each index node stores a position pointer to the memory index of the corresponding bypass agent without storing actual business data. The distributed query engine is used to decouple the global query request into local subqueries corresponding to different business systems based on the global mapping relationship table. The operator pushdown scheduling module pushes down and distributes each local subquery to the corresponding bypass agent. The bypass agent performs filtering and aggregation operations on the local memory inverted index and spatiotemporal R-tree index to generate a local result set. The local result set is then sent back to the federated query and virtual index layer for memory-level association and assembly to obtain the global logistics view result. The fault tolerance compensation layer, connected to the federated query and virtual index layer via a bidirectional channel, includes a connection status monitoring module, a Markov state transition prediction submodule, and a Kalman filter trajectory completion submodule. The connection status monitoring module monitors the heartbeat signals and query response latency of each bypass agent. When a connection anomaly is detected in a bypass agent, a compensation process is triggered. The Markov state transition prediction submodule predicts the state distribution for discrete business state fields based on historical state transition probabilities. The Kalman filter trajectory completion submodule performs trajectory extrapolation and completion for continuous spatiotemporal fields based on historical trajectories, and calculates the confidence value based on the duration of data loss. The compensation result and the confidence value are then encapsulated and output together. The unified retrieval and management application layer, connected to the fault tolerance compensation layer and the federated query and virtual index layer, is used to provide managers with a global logistics tracking, anomaly warning, and scheduling decision interface.

2. The logistics management system based on industrial big data according to claim 1, characterized in that: The semantic automatic mapping engine is used to calculate the cosine similarity between the semantic vector corresponding to the field to be mapped and the semantic vector corresponding to the standard concept node, and the Jaccard similarity coefficient between the sample value lexicon set of the field to be mapped and the example lexicon set of the standard concept node. The cosine similarity and the Jaccard similarity coefficient are weighted and summed according to a preset weighting coefficient to obtain a comprehensive similarity score. When the comprehensive similarity score is not less than a preset threshold, the equivalent mapping relationship is established and written into the global mapping relationship table. When the comprehensive similarity score is less than the preset threshold, the corresponding field is marked as a field to be confirmed.

3. The logistics management system based on industrial big data according to claim 1, characterized in that: The semantic mapping and global view layer also includes a low-code manual confirmation module, which is used to display candidate standard concept nodes marked as fields to be confirmed in descending order of similarity scores, receive mapping confirmation operations for the fields to be confirmed, write the confirmation results into the global mapping relationship table, and feed the confirmation results back to the semantic automatic mapping engine and the industrial logistics basic ontology graph as incremental training data for updating.

4. The logistics management system based on industrial big data according to claim 1, characterized in that: The distributed query engine is also used to dynamically adjust the pushdown order and parallelism of each local subquery based on the processing cost of performing filtering and aggregation operations on the corresponding bypass proxy for each local subquery, as well as the processing cost of the memory-level association assembly, so as to reduce the overall overhead of query processing.

5. A logistics management system based on industrial big data according to claim 1, characterized in that: The Markov state transition prediction submodule is used to perform exponentiation based on the historical state transition probability matrix to obtain the multi-step transition probability from the current discrete business state to each possible discrete business state after a preset time step, and to obtain the state distribution prediction result for the missing time period by combining the current discrete business state.

6. A logistics management system based on industrial big data according to claim 1, characterized in that: The Kalman filter trajectory completion submodule is used to establish a state transition equation based on the vehicle kinematics model, obtain the prior predicted value of the vehicle state and the prior estimation error covariance matrix. The vehicle state includes position and speed. When there is observation data of adjacent vehicles or roadside equipment, the Kalman gain matrix is ​​calculated based on the prior predicted value, the prior estimation error covariance matrix and the observation data, and the prior predicted value is corrected to obtain the corrected state estimate.

7. A logistics management system based on industrial big data according to claim 1, characterized in that: The fault-tolerant compensation layer is also used to calculate the confidence value of the compensation result based on the time interval since the last receipt of real data, according to a function relationship in which the confidence value monotonically decreases as the time interval increases. The confidence value is used to characterize the credibility of the compensation result.

8. A logistics management system based on industrial big data according to claim 1, characterized in that: The connection status monitoring module is also used to terminate the corresponding compensation process after detecting the recovery response of the bypass agent in the case of the connection anomaly, control the federated query and virtual index layer to resume calling the real-time data of the bypass agent, and write the real-time data received during the recovery period into the historical spatiotemporal cache.

9. A logistics management method based on industrial big data, applied to multiple business systems with different data models, characterized in that, include: By deploying a bypass agent next to each business system, read-only monitoring is performed on the local database of each business system to obtain data change events, without physically migrating the data in the local database. Extract metadata from each local database, calculate the similarity between the metadata and standard concept nodes in the pre-built industrial logistics basic ontology graph, establish an equivalent mapping relationship from local fields to the global unified logical view based on the similarity determination result, and generate a global mapping relationship table. Receive a global query request for a logistics entity, and decouple the global query request into local sub-queries corresponding to different business systems based on the global mapping relationship table; In a cross-modal global virtual index tree with logistics entity identifier as primary key and associated with time-series trajectory branch index nodes and relation entity branch index nodes respectively, locate the position pointer of the bypass agent corresponding to each local subquery; Based on the location pointer, each of the local subqueries is pushed down and distributed to the corresponding bypass agent, which then performs filtering and aggregation operations on the local memory index to obtain a local result set; The local result sets are then assembled in memory to obtain the global logistics view result.

10. A logistics management method based on industrial big data according to claim 9, characterized in that: Also includes: When a connection anomaly is detected in a bypass agent, the historical state observation sequence corresponding to that bypass agent is extracted from the historical spatiotemporal cache. For discrete business status fields, the status distribution for missing time periods is predicted based on historical status transition probabilities; For continuous spatiotemporal fields, trajectory extrapolation and completion are performed based on historical trajectories; The confidence level is calculated based on the duration of data loss, and the compensation result is encapsulated with the confidence level and then output.

Images