Intelligent decision-making method and device based on semantic knowledge graph, equipment and medium

Abstract

The invention relates to the technical field of knowledge maps, can be applied to an intelligent traffic operation management scene, and discloses an intelligent decision-making method, device, equipment and medium based on a semantic knowledge map, and the method comprises the steps: carrying out the analysis, cleaning and complementation of multi-source heterogeneous data, and generating a standardized event flow under a unified space-time reference; constructing a semantic knowledge graph containing entities, attributes, relationships and business logic, injecting the standardized event streams into the graph to complete pattern matching, generating a structured knowledge packet and realizing business logic updating; and scene analysis is carried out based on the structured knowledge packet and the semantic knowledge graph, an analysis result is output, a decision suggestion is generated, and closed-loop processing from data to decision is realized. According to the method, multi-source data is converted into structured knowledge and semantic reasoning is introduced, so that the analysis result can support decision generation, and the semantic understanding ability and decision support efficiency of data fusion are improved.

Description

Intelligent decision-making methods, devices, equipment, and media based on semantic knowledge graphs Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to an intelligent decision-making method, apparatus, device, and medium based on semantic knowledge graph. Background Technology

[0002] With the large-scale deployment of traffic sensing devices, traffic management systems are now able to continuously collect heterogeneous data from multiple sources, including video surveillance, vehicle checkpoints, floating car positioning, and signal control, providing a data foundation for traffic operation monitoring. However, existing technologies for processing and utilizing this data are still mainly limited to data collection, aggregation, and display, making it difficult to effectively support comprehensive analysis and decision support in complex traffic scenarios.

[0003] On the one hand, multi-source heterogeneous data differ significantly in format structure, temporal precision, spatial representation, and business meaning. Existing systems typically process data only through simple parsing or field alignment, lacking a unified expression standard and consistent organization. This makes it difficult to correlate and use data from different sources within the same analytical framework, hindering the full realization of data value. Even after basic cleaning and alignment, the data often exists in a loosely parallel form, making it difficult to form stable and reusable fusion results.

[0004] On the other hand, existing technologies lack sufficient semantic understanding of traffic data. Most systems still use raw fields or statistical indicators as the objects of analysis, lacking a structured expression of the business meaning behind the data. There is no clear mapping relationship between the data and specific management scenarios, making it difficult for the analysis results to directly correspond to actual handling needs. In this situation, the analysis conclusions often need to rely on human experience for interpretation and transformation, and the system itself is unable to output decision recommendations with clear business orientation.

[0005] Furthermore, existing platforms typically rely on fixed rules or static models during analysis. When faced with new data combinations or changes in business logic, they lack effective knowledge updates and evolution mechanisms, making it difficult for the system to adapt to complex and ever-changing traffic conditions. There is a significant disconnect between analytical capabilities and decision support capabilities, limiting the comprehensive application of multi-source data in traffic management. Summary of the Invention

[0006] The main objective of this invention is to provide an intelligent decision-making method, device, equipment, and storage medium based on semantic knowledge graphs. This invention aims to solve the technical problem that existing traffic management technologies are unable to perform semantic association and business logic integration on multi-source heterogeneous data under a unified spatiotemporal benchmark, resulting in analysis results remaining at the data level and failing to form structured knowledge and decision suggestions that can evolve with the scenario and directly support decision-making.

[0007] To achieve the above objectives, this invention provides an intelligent decision-making method based on a semantic knowledge graph, comprising: acquiring multi-source heterogeneous data; performing format parsing, cleaning, and completion on the multi-source heterogeneous data to obtain primary format data; establishing a unified spatiotemporal reference grid; aligning the primary format data to the spatiotemporal reference grid to generate a standardized event stream; constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic; injecting the standardized event stream into the semantic knowledge graph for pattern matching to obtain a matching result; adding semantic tags to the standardized event stream based on the matching result to generate a structured knowledge package; and generating new business logic to update the semantic knowledge graph when the matching result indicates the existence of business logic not defined in the semantic knowledge graph; performing scenario-based analysis based on the structured knowledge package and the semantic knowledge graph to generate analysis results; and generating decision suggestions based on the analysis results.

[0008] Furthermore, to achieve the above objectives, this invention provides an intelligent decision-making device based on a semantic knowledge graph, comprising: a data preprocessing module for acquiring multi-source heterogeneous data, performing format parsing and cleaning / completening on the multi-source heterogeneous data to obtain primary format data; a spatiotemporal alignment module for establishing a unified spatiotemporal reference grid, aligning the primary format data to the spatiotemporal reference grid, and generating a standardized event stream; a knowledge graph module for constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic; a pattern matching module for injecting the standardized event stream into the semantic knowledge graph for pattern matching, and obtaining a matching result; a knowledge evolution module for adding semantic tags to the standardized event stream based on the matching result to generate a structured knowledge package, and generating new business logic to update the semantic knowledge graph when the matching result indicates the existence of business logic not defined in the semantic knowledge graph; and a decision analysis module for performing scenario-based analysis based on the structured knowledge package and the semantic knowledge graph, generating analysis results, and generating decision suggestions based on the analysis results.

[0009] Furthermore, to achieve the above objectives, the present invention also provides a computer device, the computer device including a memory, a processor, and an intelligent decision-making program based on semantic knowledge graph stored in the memory and executable on the processor, wherein when the intelligent decision-making program based on semantic knowledge graph is executed by the processor, it implements the steps of the intelligent decision-making method based on semantic knowledge graph as described above.

[0010] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an intelligent decision-making program based on a semantic knowledge graph, wherein the intelligent decision-making program based on a semantic knowledge graph, when executed by a processor, implements the steps of the intelligent decision-making method based on a semantic knowledge graph as described above.

[0011] Beneficial Effects: This invention relates to the field of knowledge graph technology and can be applied to intelligent transportation operation and management scenarios. It discloses an intelligent decision-making method, device, equipment, and medium based on semantic knowledge graphs, including: parsing, cleaning, and completing multi-source heterogeneous data, and aligning it under a unified spatiotemporal benchmark to form a standardized event flow; constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic based on this, and injecting the standardized event flow into the graph to complete pattern matching; generating a structured knowledge package based on the matching results, and updating the business logic when undefined business logic is found in the graph; further utilizing the structured knowledge package and the semantic knowledge graph to conduct scenario-based analysis, outputting analysis results, and generating decision suggestions. This invention unifies the spatiotemporal representation of multi-source data and introduces a semantic knowledge graph for semantic association and business logic constraints, enabling data to be transformed into understandable and reasonable structured knowledge, and further generating decision suggestions through scenario-based analysis, thereby improving the depth of multi-source data fusion and decision support capabilities. Attached Figure Description

[0012] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the drawings: Figure 1 is a schematic diagram of an application environment of an intelligent decision-making method based on semantic knowledge graphs according to an embodiment of the present invention; Figure 2 is a schematic flowchart of an embodiment of the intelligent decision-making method based on semantic knowledge graphs according to the present invention; Figure 3 is a schematic diagram of the functional modules of a preferred embodiment of the intelligent decision-making device based on semantic knowledge graphs according to the present invention; Figure 4 is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 5 is another schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0013] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0014] The intelligent decision-making method based on semantic knowledge graphs provided in this invention can be applied in the application environment shown in Figure 1, where the client communicates with the server via a network. The server can use the client to parse, clean, and complete multi-source heterogeneous data, aligning it under a unified spatiotemporal benchmark to form a standardized event stream. Based on this, a semantic knowledge graph containing entities, attributes, relationships, and business logic is constructed, and the standardized event stream is injected into the graph to complete pattern matching. A structured knowledge package is generated based on the matching results, and business logic updates are implemented when undefined business logic is found in the graph. Furthermore, the structured knowledge package and the semantic knowledge graph are used to conduct scenario-based analysis, outputting analysis results and generating decision suggestions. This invention unifies the spatiotemporal representation of multi-source data and introduces a semantic knowledge graph for semantic association and business logic constraints, enabling data to be transformed into understandable and reasonable structured knowledge. Decision suggestions are further generated through scenario-based analysis, thereby improving the depth of multi-source data fusion and decision support capabilities. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The present invention will now be described in detail through specific embodiments.

[0015] Please refer to Figure 2, which is a flowchart illustrating an embodiment of the intelligent decision-making method based on semantic knowledge graphs provided by the present invention. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0016] As shown in Figure 2, the intelligent decision-making method based on semantic knowledge graph proposed in this invention includes the following steps: S10, acquiring multi-source heterogeneous data, performing format parsing and cleaning / completening on the multi-source heterogeneous data to obtain primary format data; in this embodiment, the source of multi-source heterogeneous data can be business interface returned data, message channel pushed data, or file delivery data. Heterogeneity is reflected in inconsistent field sets, inconsistent nesting levels, inconsistent record granularity, and inconsistent encoding and unit expression. The acquisition action corresponds to connecting the data to the processing link and forming a processable data record set. The data record set is obtained by boundary recognition. Boundary recognition divides continuous input into record units. Record units can be single records or batch records. The division criteria can be delimiters, length headers, batch markers, or file block indexes. After division, each record unit is associated with a source identifier and arrival time for conflict arbitration and priority determination during subsequent cleaning and completion.

[0017] Format parsing maps record units to a unified field structure and standardizes field types. The unified field structure originates from the defined set of fields required for analysis and the minimum set of fields required for cross-source fusion. The unified field structure must include at least one or more of the following: entity identifier-related fields, time-related fields, space-related fields, and event attribute-related fields. Format parsing first identifies the data format or access protocol, based on metadata fields, protocol version identifiers, or structure verification results. After identification, a conversion strategy is selected, including field extraction rules and field conversion rules. Field extraction rules map the original field names or hierarchical paths to the target fields of the unified field structure. Field conversion rules convert strings, numbers, booleans, time, coordinates, etc., into a unified internal type representation and perform normalization processing on encoding, units, and enumerated values, resulting in data after format parsing.

[0018] Cleaning and completion consist of two parts. Cleaning targets redundant and duplicate records and conflicting data. Redundant and duplicate records originate from repeated reporting or retransmission. Identifying redundancy requires constructing a deduplication key, which is formed by combining at least two of the following: entity identifier-related fields, time-related fields, and spatially related fields. Records matching the same deduplication key are either removed or retained. Conflicting data arises from inconsistent field values ​​under the same deduplication key. Conflict determination depends on the set of conflicting fields and a conflict threshold. Conflict handling relies on arbitration based on source identifier priority or arrival time priority, resulting in cleaned data. Completion targets missing critical fields in the cleaned data. Critical fields originate from a set of fields marked as necessary in a unified field structure. Completion employs either a preset strategy or an interpolation strategy. The preset strategy corresponds to deterministic mapping and default assignment rules, while the interpolation strategy utilizes numerical estimation rules based on adjacent records of the same entity in the time dimension. After completion, a basic formatted data is output.

[0019] A source-identified, multi-channel access approach can be adopted, where data from different sources are placed into independent buffer queues and boundary identification is performed within the queues to reduce format interference between different sources. Alternatively, a unified access channel can be used, and data is routed to the corresponding conversion strategy execution unit after identifying the data format or access protocol. The unified field structure can be configured as an extensible field set, adapting to new sources by adding field mapping rules without changing existing field type specifications. Deduplication keys can be configured as multi-level composite keys, first filtering candidate duplicates with coarse-grained combinations and then confirming duplicates with fine-grained combinations to reduce computational overhead. Conflict arbitration can be configured as a source identifier priority table or an arrival time priority rule. The completion strategy can be configured with the order of rule priority and interpolation completion, and a maximum interval threshold can be set for the interpolation strategy; when the threshold is exceeded, missing values ​​are retained to avoid unreasonable estimations.

[0020] This embodiment forms record units through boundary recognition and associates them with source identifiers and arrival times, enabling stable processing of input data and supporting the determination of subsequent arbitration criteria; it maps multi-source heterogeneous data into a unified field structure and completes field type standardization through format parsing, reducing the impact of field expression differences on subsequent processing; it removes and arbitrates redundant and duplicate records and conflicting data through cleaning, reducing duplicate and inconsistent input; and it improves the field integrity and consistency of primary format data by applying preset strategies or interpolation strategies to missing key fields through completion.

[0021] S20, establish a unified spatiotemporal reference grid, align the primary format data to the spatiotemporal reference grid, and generate a standardized event stream. In this embodiment, the unified spatiotemporal reference grid provides reusable temporal and spatial references, enabling primary format data from different sources to be expressed within the same reference system. The temporal dimension of the spatiotemporal reference grid constrains the temporal location of events, and the spatial dimension constrains the spatial location of events; together, they constitute the unique positioning basis of the grid cell. The spatiotemporal reference grid can divide time segments at fixed time intervals, or it can adaptively adjust the length of time segments based on event density. The spatial part can be divided into spatial segments using latitude and longitude grids, road node areas, road segment intervals, or lane-level geometric units, so that spatial positioning matches the actual distribution of business objects.

[0022] Aligning primary format data to a spatiotemporal reference grid is a combination of mapping and correction processes. The mapping process maps timestamp information from the primary format data to corresponding time segments and spatial coordinate information to corresponding spatial cells, thus obtaining the target location of each record within the spatiotemporal reference grid. To reduce the impact of device sampling latency, clock drift, and positioning jitter, a time synchronization tolerance threshold can be set for time mapping, and a spatial positioning error range can be set for spatial mapping, ensuring that records near the boundaries can still be grouped into the same grid cell within acceptable limits. The correction process is used to converge alignment errors after mapping, including correcting timestamp deviations, correcting spatial location offsets, and constraining abnormally abrupt records, ensuring that the aligned data has consistent temporal and spatial semantics within the same grid cell.

[0023] Standardized event streams are a form of aligned, continuous data representation. The key is the structured encoding of the event's position within a spatiotemporal reference grid. Standardized event streams can be organized sequentially by time segments or aggregated by spatial units and then sorted temporally. This gives the event stream a stable input structure and a traceable spatiotemporal index, resulting in a unified event sequence that can be directly used for subsequent processing.

[0024] This embodiment establishes a unified spatiotemporal reference grid and maps and corrects primary format data to this grid, enabling the primary format data to obtain a unified reference and consistent semantic expression in time and space. This generates a standardized event stream that is structurally stable, indexable, and traceable, reducing the impact of inconsistencies in the timing and location of multi-source data on subsequent processing.

[0025] S30, construct a semantic knowledge graph containing entities, attributes, relationships, and business logic; in this embodiment, the semantic knowledge graph is used to express objects and semantic relationships between objects in a structured manner, enabling data to rise from the field level to a searchable and reasonable semantic level. Constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic includes three consecutive steps: semantic element definition, semantic element generation, and semantic element storage, so that entities, attributes, relationships, and business logic have a unified data organization method and consistent reference identifiers.

[0026] Entities represent sets of described objects, originating from business object directories, event subject identifier sets, object references in rule entries, or object noun phrases appearing in text. Entity generation includes entity type determination and entity identifier generation. Entity types distinguish different object categories, while entity identifiers uniquely locate entities in the graph. Attributes characterize the features of entities, originating from structured fields, semi-structured key-value pairs, feature phrases in text descriptions, or rule parameter tables. Attribute generation includes attribute name normalization, attribute value type constraints, and attribute value loading, ensuring that attributes with the same meaning maintain synonymy in the graph and are usable for retrieval.

[0027] Relationships are used to represent the way entities are associated. They can originate from business process definitions, event chains, verb phrases and prepositional structures in text, or association path expressions in existing rules. Relationship generation includes determining the relationship type, determining the relationship direction, and defining relationship constraints. The relationship type is used to distinguish the semantics of association, the relationship direction is used to express the direction of action or dependency, and the relationship constraints are used to limit the range of conditions under which the relationship is established, so that the relationship can not only connect entities but also carry computable constraint information.

[0028] Business logic represents semantically executable conditions and derivation rules. Its sources can include policy clauses, handling procedures, threshold condition sets, state transition constraints, or conditional sentences appearing in text. The generation of business logic includes defining trigger conditions, defining condition combination methods, and defining output conclusions. Trigger conditions describe when the condition is triggered, condition combination methods describe the logical connection between multiple conditions, and output conclusions describe the semantic conclusion or state result after triggering, enabling business logic to form actionable rule expressions with entities, attributes, and relationships.

[0029] To ensure consistent representation, semantic knowledge graphs can adopt a subject-predicate-object triple structure as their basic data organization. The subject carries entity identifiers, the predicate carries attribute names, relation types, or rule association names, and the object carries attribute values, associated entity identifiers, or rule conclusion identifiers. After mapping entities, attributes, relations, and business logic to the triple structure, the data is written to a graph database, enabling the semantic knowledge graph to have persistent storage, index retrieval, and association traversal capabilities, thus completing the construction of the semantic knowledge graph.

[0030] This embodiment defines and generates entities, attributes, relationships, and business logic using a unified semantic element system, and maps them into a subject, predicate, and object triple structure, which is then written into a graph database. This enables the semantic knowledge graph to carry object, feature, association, and rule information with consistent identifiers, thereby improving the structure and retrieval of semantic expression and reducing the problem of difficulty in unifying the organization of multi-source information at the semantic level.

[0031] S40, the standardized event stream is injected into the semantic knowledge graph for pattern matching to obtain a matching result. In this embodiment, the standardized event stream is used to carry event data that has been uniformly expressed in time and space, and its content has clear entity references and state descriptions. Injecting the standardized event stream into the semantic knowledge graph is the process of transforming events from a time-series form into something that can be understood and compared within a semantic structure. The meaning of injection is that events no longer exist as isolated data records, but rather establish computable associations with entities, attributes, relationships, and business logic in the semantic knowledge graph.

[0032] Pattern matching is used to determine whether an event conforms to the existing semantic structures and rule expressions in the semantic knowledge graph. This process first parses entity identifiers that uniquely point to objects, along with temporal, spatial, and state information related to the event state, from the standardized event stream. Entity identifiers are used to locate the corresponding nodes in the semantic knowledge graph, giving the event semantic context. State values ​​are used as matching input and compared with the condition expressions defined in the semantic knowledge graph.

[0033] In semantic knowledge graphs, patterns are constructed from the relationship structure between entities and the triggering conditions defined in business logic. Pattern matching compares and calculates event states with business logic triggering conditions, and combines this with existing association paths between entities to determine whether an event satisfies a defined semantic pattern. This comparison not only focuses on single attribute values, but also on whether attribute combinations and relationship paths are simultaneously valid, thus avoiding reliance on simple threshold judgments.

[0034] The matching results describe the correspondence between events and existing patterns in the semantic knowledge graph. The matching results must at least include a matching status identifier to indicate whether a matching pattern exists, and may also carry a matched business logic identifier or associated path identifier, enabling subsequent processing to clearly define the semantic context in which the event is identified. By generating matching results, events are transformed from raw state data into judgment results with clear semantic implications.

[0035] This embodiment injects standardized event streams into a semantic knowledge graph and performs pattern matching, enabling events to be calculated and compared with existing entity relationships and business logic at the semantic level. This transforms data that originally only had spatiotemporal meaning into event expressions with clear semantic judgment results, reducing the reliance on human experience for event understanding and improving the accuracy and consistency of semantic recognition.

[0036] S50, based on the matching results, semantic tags are added to the standardized event stream to generate a structured knowledge package, and when the matching results indicate the existence of business logic not defined in the semantic knowledge graph, new business logic is generated to update the semantic knowledge graph. In this embodiment, the matching results are used to characterize the correspondence between the standardized event stream and existing patterns in the semantic knowledge graph, and their content at least reflects whether the event satisfies the given semantic conditions and the corresponding logical direction. Adding semantic tags to the standardized event stream based on the matching results is the process of explicitly binding the matching conclusion to the event, transforming the event from a state description into a directly identifiable semantic unit. Semantic tags are used to express the meaning of the event at the business level, and their sources include matched business logic identifiers, relationship path features, or condition combination features. Semantic tags correspond one-to-one with events and maintain stable references.

[0037] When generating a structured knowledge package, semantic tags are combined with the original time information, spatial information, entity identifiers, and state fields of the event to form a data carrier with a fixed structure. The structured knowledge package emphasizes the semantic consistency between internal fields. Semantic tags are used to encompass the meaning of the event, while the remaining fields are used to support the factual basis of that meaning, so that the knowledge package retains the original data source while having a clear semantic orientation.

[0038] When the matching result indicates that the logical combination satisfied by the current event is not already defined in the semantic knowledge graph, new business logic needs to be generated. This new business logic originates from the conditional combinations and relational features formed during the matching process. Based on entity relationships and state constraints, this logic is abstracted into reusable semantic rules. The generated business logic is written into the semantic knowledge graph to expand the existing semantic structure, enabling subsequent similar events to directly obtain semantic recognition through matching, thus avoiding the repeated generation of the same logic.

[0039] By synchronously executing semantic tag generation and business logic updates, the event processing results evolve in line with the semantic knowledge graph, enabling the knowledge system to gradually improve as events accumulate.

[0040] This embodiment generates semantic tags based on matching results and constructs structured knowledge packages, enabling events to have a stable and clear semantic expression. At the same time, by dynamically generating and updating business logic, the semantic knowledge graph can continuously absorb newly emerging logical relationships, thereby improving the completeness of event semantic expression and the self-expansion capability of the knowledge system.

[0041] S60, perform scenario-based analysis based on the structured knowledge package and the semantic knowledge graph, generate analysis results, and generate decision suggestions based on the analysis results.

[0042] In this embodiment, the structured knowledge package carries the semantic identifiers, temporal and spatial elements, and related entity information of events, providing semantically complete and clearly defined input units for analysis. The semantic knowledge graph provides cross-event and cross-entity contextual association capabilities, containing relationship constraints between entities, state evolution rules, and historical semantic accumulation. Contextual analysis is the process of filtering, aggregating, and reasoning about the structured knowledge package within a pre-defined business scenario context. This process relies on the relationship paths and logical constraints in the knowledge graph, extending the analysis beyond single-point events to the association patterns between events.

[0043] When performing scenario-based analysis, the first step is to select a data set from the structured knowledge package that satisfies the semantic tag combination and spatiotemporal constraints based on the scenario conditions. Then, by combining the relationship links in the semantic knowledge graph, the selected data is expanded and associated to form a context set for analysis. The analysis results are obtained by calculating the state changes, association strength, or trend characteristics in the context set, and their manifestation can be risk judgment, trend description, or state assessment. The key conclusions from the analysis results are further extracted and mapped into actionable actions to form decision recommendations. These recommendations emphasize action orientation and are used to express the recommended response methods or adjustment directions in the current scenario.

[0044] Contextualized analysis can be achieved through rule-driven or model-based reasoning, both approaches relying on a combination of structured knowledge packages and semantic knowledge graphs as input. Analysis results can be output as structured fields, scores, or ratings. Decision recommendations can be generated through template filling or rule combinations to adapt to varying business complexities and response time requirements.

[0045] This embodiment combines structured knowledge packages with semantic knowledge graphs for scenario-based analysis, enabling the analysis process to have clear semantic constraints and contextual association capabilities. The analysis results can be directly transformed into targeted decision suggestions, thereby improving the executability of the analysis conclusions and the accuracy of decision generation.

[0046] In one embodiment, step S10 includes: S101, configuring multi-type data access adapters to receive multi-source heterogeneous data containing structured records, semi-structured logs, and streaming messages respectively; S102, identifying the data format or access protocol of the received multi-source heterogeneous data, and using a conversion strategy corresponding to the identified data format or access protocol to map the multi-source heterogeneous data to a preset unified intermediate model to complete format parsing, thereby obtaining format-parsed data; S103, detecting whether there are redundant duplicate records and / or conflicting data in the format-parsed data, and performing a removal process on the redundant duplicate records and / or conflicting data to clean them, thereby obtaining cleaned data; S104, filling in the missing key fields in the cleaned data based on a preset strategy or interpolation strategy to perform completion, thereby obtaining primary format data.

[0047] In this embodiment, when multi-source heterogeneous data includes structured records, semi-structured logs, and streaming messages, the data needs to achieve unified access boundaries, unified field semantics, and convergence of quality risks before entering the same processing link. A multi-type data access adapter is used to establish access channels for different sources and transmission formats. The adapter can provide batch retrieval interfaces, incremental subscription interfaces, message consumption interfaces, or file disk persistence monitoring interfaces, enabling structured records to enter the processing link as table records or row events, semi-structured logs to enter the processing link as text fragments and key-value pair fragments, and streaming messages to enter the processing link as continuous event fragments. During access, the adapter synchronously collects source identifiers, access time, and partition identifiers for subsequent deduplication and conflict determination to establish data boundaries.

[0048] Identifying data format or access protocol is used to determine the data carrier structure and transmission constraints, thereby selecting an executable conversion strategy. Data format can be reflected in field delimiter types, serialization methods, hierarchical nesting methods, or character encoding methods. Access protocols can be reflected in message topic agreements, interface parameter agreements, push confirmation mechanisms, or batch boundary agreements. The conversion strategy is used to map multi-source heterogeneous data to a unified intermediate model. The unified intermediate model fixes the set of fields, the set of field types, and the set of field semantics, allowing the same semantics to be carried by the same field from different sources, and ensuring that the same field has comparable type constraints across different sources. During the mapping process, field extraction, field renaming, type conversion, unit normalization, and hierarchical expansion can be performed, enabling nested keys in semi-structured logs to be expanded into flat fields of the unified intermediate model, and enabling event payloads in streaming messages to be decomposed into event field sets of the unified intermediate model. After mapping, the data is obtained after format parsing. The parsed data structurally satisfies the constraints of the unified intermediate model, type-wise satisfies verifiability constraints, and semantically satisfies alignment constraints.

[0049] The cleaning phase performs removal processing on redundant and duplicate records and conflicting data to reduce uncertainty in subsequent processing. Redundant and duplicate records can be identified based on unique key combinations, source identifiers and event timestamps, payload digest fingerprints, or field set consistency. Removal can employ strategies such as retaining the earliest arriving version, the latest arriving version, or versions with higher quality scores. Conflicting data is identified when contradictory field values ​​exist for the same object within the same time window. Conflicts can manifest as mutually exclusive status fields, numerical fields exceeding reasonable ranges, inconsistent enumeration fields, or multiple versions of the same object covering each other without priority. Removal can isolate conflicting records to an anomaly set and remove them from the main data stream, or select trusted versions according to priority rules and remove the remaining versions, ensuring consistency and traceability of the cleaned data within the same object and time window.

[0050] The completion phase performs imputation on missing key fields to achieve usability closure. Missing key fields refer to the set of fields essential for subsequent alignment, association, or analysis processes; their absence will result in objects being unlocatable, time-sequential sorting being impossible, or semantic classification being unrecognizable. Preset strategies are used for deterministic imputation when business constraints are clear, such as using source default values, dictionary mapping values, last valid values, or fixed rule-derived values. Interpolation strategies are used for estimative imputation when fields have continuity or smoothability. Interpolation can be based on temporally adjacent samples, spatially adjacent samples, or samples of similar objects to obtain calculable approximate values ​​for missing fields. During the imputation process, imputation markers, imputation sources, and imputation confidence information are recorded simultaneously, ensuring that the primary format data simultaneously possesses a set of usable fields and a set of quality markers, facilitating selective use of imputed values ​​or risk control in subsequent stages. After imputation, primary format data is obtained, which, under the constraints of a unified intermediate model, possesses a consistent structure, comparable types, and traceable quality status.

[0051] This embodiment incorporates structured records, semi-structured logs, and streaming messages into a unified access boundary through multi-type data access adapters. It maps multi-source heterogeneous data to a unified intermediate model to form format-parsed data through data format or access protocol identification and conversion strategies. It removes redundant and duplicate records and conflicting data to form cleaned data. It fills in missing key fields through preset strategies or interpolation strategies to form primary format data. This achieves uniformity and convergence of cross-source data in terms of structure, type, and quality, reduces the sensitivity of subsequent processing to source differences and data defects, and improves data availability and consistency.

[0052] In one embodiment, step S20 includes: S201, constructing a unified spatiotemporal reference grid that includes time dimension, geographic location dimension, and lane-level coordinate dimension; S202, setting a time synchronization tolerance threshold and spatial positioning error range for the unified spatiotemporal reference grid; S203, extracting timestamp information and spatial coordinate information from the primary format data, and using the timestamp information and spatial coordinate information to map the primary format data to the unified spatiotemporal reference grid to obtain projected data; S204, using the time synchronization tolerance threshold and the spatial positioning error range to perform spatiotemporal calibration on the projected data to generate a standardized event stream.

[0053] In this embodiment, a unified spatiotemporal reference grid is used to converge different time precisions, spatial representations, and sampling frequencies in the primary format data into the same alignable coordinate system, ensuring that subsequent data fusion does not depend on the original coordinate caliber of the data source. When constructing a unified spatiotemporal reference grid that includes time, geographic location, and lane-level coordinate dimensions, the time dimension defines discrete time units and time indexing rules. Time units can use time slices with fixed step sizes or adaptive time slices based on business rhythms. Time indexing rules map timestamp information to time raster numbers to form a sortable time series. The geographic location dimension defines the geographic space partitioning method and spatial indexing rules. Partitioning methods can use latitude and longitude grids, linear mileage grids after road topology segmentation, or a combination of administrative divisions and road levels. Spatial indexing rules map spatial coordinate information to spatial raster numbers to form a searchable spatial key. Lane-level coordinates are used to distinguish different lanes, directions, and functions within the same road segment. Lane-level coordinates can consist of lane number, lane centerline mileage, lane width boundary, and direction indicator, making fine-grained locations under the same geographic location dimension distinguishable, thereby preventing different lane data from being merged into the same spatial unit after alignment.

[0054] The time synchronization tolerance threshold and spatial positioning error range are used to constrain acceptable deviations in alignment and calibration, ensuring that projected data meets consistency requirements in both time and space. The time synchronization tolerance threshold handles offsets, jitter, and sampling interval differences in timestamp information from different sources. This tolerance threshold can be expressed as the maximum allowable time deviation or the allowable time window width, determining which time unit an event should be assigned to and whether time correction is required. The spatial positioning error range handles measurement errors, map matching errors, and projection errors in spatial coordinate information from different sources. This error range can be expressed as the maximum allowable planar distance deviation or the allowable road mileage deviation, determining which spatial unit an event should be assigned to and whether spatial correction is required. Both thresholds serve as quality constraints, ensuring that the subsequently generated standardized event stream has controllable alignment accuracy and traceable calibration basis.

[0055] When extracting timestamp and spatial coordinate information from primary format data, the timestamp information is used to characterize the event occurrence time or collection time, and the spatial coordinate information is used to characterize the event occurrence location or collection location. The extraction process requires unifying the various time and coordinate fields that may exist in the primary format data into a computable set of fields. For example, parsing a string of timestamps into timestamps with a unified time zone and uniform precision, and converting latitude and longitude coordinates, road mileage coordinates, or grid numbers into a unified spatial coordinate expression. When mapping the primary format data to a unified spatiotemporal reference grid using timestamp and spatial coordinate information, the mapping process first determines the time unit according to the time dimension, then the spatial unit according to the geographical location dimension, and finally the lane unit according to the lane-level coordinate dimension. This ensures that each record obtains a unique spatiotemporal grid index triplet, forming projected data. The projected data retains the original timestamp and spatial coordinate information before mapping, while simultaneously generating the mapped grid index field. This gives the same record dual-track fields for both original and aligned expressions, facilitating subsequent calibration and quality marking.

[0056] When performing spatiotemporal calibration of projected data using time synchronization tolerance thresholds and spatial positioning error ranges, time calibration corrects deviations between timestamp information and time unit boundaries. Correction methods include merging data into adjacent time units, reallocating it to more reasonable time units according to time windows, or performing offset compensation for sources with stable offsets, ensuring events meet tolerance threshold constraints in the time dimension. Spatial calibration corrects deviations between spatial coordinate information and spatial unit boundaries. Correction methods include merging data into adjacent spatial units, performing map matching back projection onto the road centerline along the road topology, and performing lane assignment correction according to lane-level coordinate dimensions, ensuring events meet error range constraints in the spatial dimension. During calibration, calibration and anomaly markers are generated for data exceeding tolerance thresholds or error ranges to distinguish between three states: direct alignment, requiring calibration, and unable to be calibrated. After completing time and spatial calibration, a standardized event stream is output. The standardized event stream organizes event sequences using a unified spatiotemporal grid index as the main thread, enabling events from different sources to be aggregated, compared, and correlated within the same time unit, spatial unit, and lane unit.

[0057] This embodiment constructs a unified spatiotemporal reference grid that includes time, geographic location, and lane-level coordinate dimensions, enabling primary format data to obtain a consistent spatiotemporal index representation. By setting a time synchronization tolerance threshold and a spatial positioning error range, and performing spatiotemporal calibration on the projected data accordingly, the alignment process has controllable deviation boundaries and traceable calibration rules. This generates a standardized event flow organized according to the unified spatiotemporal grid, reducing the impact of inconsistencies in the time and spatial dimensions of multi-source data on the fusion results, and improving the comparability and correlation of cross-source events within the same spatiotemporal unit.

[0058] In one embodiment, step S30 includes: S301, defining a subject, predicate, and object triple structure under a resource description framework as the basic data organization form of the semantic knowledge graph; S302, using a natural language processing module to extract entities and entity attribute features from unstructured business text to construct entities and attributes of the semantic knowledge graph; S303, defining dynamic interaction paths and pre- and post-dependency strategies between the entities to construct relationships in the semantic knowledge graph describing the business flow; S304, configuring business logic strategies and defining state change triggering conditions and evolution models for the entities to construct the business logic of the semantic knowledge graph; S305, mapping the entities, attributes, relationships, and business logic to the subject, predicate, and object triple structure, and storing the mapped triple data in a graph database to complete the construction of the semantic knowledge graph.

[0059] In this embodiment, the subject, predicate, and object triple structure under the resource description framework is used to unify the knowledge representation in the semantic knowledge graph into computable, searchable, and reasonable data units. The subject carries the identifier of the described object and can correspond to business objects, event objects, or rule objects; the predicate carries the semantic relationship type between the subject and object and can correspond to semantic connections such as attribute attribution, behavior pointing, dependency constraints, and trigger associations; the object carries the target value or target object pointed to by the subject, which can be a numerical value, enumeration, text fragment, or another entity identifier. When the triple structure is used as the basic data organization form, naming conventions and type constraints need to be established simultaneously to ensure that subjects with the same name maintain the same identifier across different data sources, that predicates with the same name maintain the same semantic boundary across different business contexts, and that the value range and data type of the object can be verified, thereby providing a definite structural boundary for subsequent mapping and database entry.

[0060] When the Natural Language Processing (NLP) module extracts entities and their attribute features from unstructured business text, this text includes content written in natural language such as policy clauses, handling records, alarm descriptions, work order notes, and inspection descriptions. The extraction process requires segmenting the text into processable semantic fragments, identifying entity candidates within these fragments, and performing entity normalization. This ensures that the same entity, under different representations, is grouped under the same identifier. Simultaneously, the attribute features of the entities are extracted in a structured manner, converting them into key-value representations that can directly fall into a triplet structure. There is a sequential dependency between entity extraction and attribute feature extraction; the entity identifier must be determined first before attribute features can be bound to the corresponding entity, preventing attribute features from becoming isolated, unassigned information. The extraction results are used to construct entities and attributes for a semantic knowledge graph, where entities bear the semantic burden of nodes, and attributes provide queryable descriptions and constraints for those nodes.

[0061] Dynamic interaction paths and pre- and post-dependency strategies between entities are used to construct relationships and describe the business flow. Dynamic interaction paths express the sequence of interactions and connections that occur between entities as time progresses or their states evolve, such as transmission, influence, linkage, and co-occurrence. Pre- and post-dependency strategies express the constraints and order conditions for the establishment of relationships, ensuring that relationships describe not only connections but also the prerequisites and triggering order for those connections. Defining dynamic interaction paths requires specifying the types of endpoint entities, the types of relationships allowed in the path, and the path length or hierarchical boundaries to ensure the path can be retrieved and reused. Defining pre- and post-dependency strategies requires specifying the conditions for determining dependencies, the order in which dependencies are satisfied, and the conflict resolution principles to ensure the business flow has verifiable constraints. After the relationships are constructed, they serve both as edges in the graph structure to connect entities and as inputs for subsequent business logic construction, defining the context of triggering conditions.

[0062] Business logic strategies, state change triggering conditions, and evolutionary models are used to construct business logic and extend static knowledge into an evolvable set of rules. Business logic strategies describe the organization and scope of rules, such as categorizing by scenario, object type, or priority, allowing rules to be selected and combined. State change triggering conditions describe the criteria for determining when an entity's state changes from one value to another. Triggering conditions can originate from thresholds of entity attributes, satisfaction of relational paths, occurrence of interaction sequences, or aggregation results within a time window. Evolutionary models describe how business logic updates and converges with new information during continuous operation, including rule version evolution, parameter adjustment boundaries, failure conditions, and rollback conditions, ensuring the maintainability and traceability of the business logic. The construction of business logic is coupled with the construction of relationships; state change triggering conditions often require referencing the association paths defined by relationships, and evolutionary models require referencing type constraints of entities and attributes to control the scope of rule expansion.

[0063] When mapping entities, attributes, relationships, and business logic to a subject, predicate, and object triple structure, node information, edge information, and rule information need to be uniformly transcribed into triple data. For entity mapping, the subject typically uses an entity identifier, the predicate uses an entity type or identifier relationship, and the object uses a type value or name value, giving the entity identifiable type semantics. For attribute mapping, the subject uses an entity identifier, the predicate uses an attribute name, and the object uses an attribute value, making attributes searchable and comparable. For relationship mapping, the subject uses a starting entity identifier, the predicate uses a relationship type, and the object uses an ending entity identifier, allowing dynamic interaction paths and pre- and post-dependency strategies to be presented with edge semantics. If necessary, dependency constraints can be split into additional triples to express sequence conditions and constraints. For business logic mapping, the subject can use a rule identifier or scenario identifier, the predicate is used to distinguish fields such as trigger conditions, action semantics, and version information, and the object is used to carry condition expressions, action expressions, or version numbers, making state change trigger conditions and evolution models form storable, queryable, and updatable rule units. When storing triplet data into a graph database, an index needs to be created to support rapid location based on subjects and predicates, and a consistency check needs to be established to prevent multiple modeling of entities with the same name, broken relationship chains, or missing rule references, thereby completing the construction of a semantic knowledge graph and forming a data foundation for sustainable expansion.

[0064] This embodiment uses a subject, predicate, and object triple structure to uniformly organize the data representation of the semantic knowledge graph. It utilizes a natural language processing module to transcribe the entity and attribute features in unstructured business text into structured content. Then, it characterizes the relationship between entities through dynamic interaction paths and pre- and post-dependency strategies. Simultaneously, it solidifies the evolvable rule representation using business logic strategies, state change triggering conditions, and evolutionary models. Finally, it maps entities, attributes, relationships, and business logic into indexable triple data and stores it in a graph database. This ensures that the semantic knowledge graph has consistent representation boundaries in structure, queryable semantic associations in content, and maintainable evolutionary capabilities in rules. This reduces the difficulty of association caused by inconsistent multi-source knowledge representation and improves the operability of structured reuse of business semantics.

[0065] In one embodiment, step S40 includes: S401, parsing entity identifiers and spatiotemporal state data from the standardized event stream; S402, using the entity identifiers to perform index retrieval in the semantic knowledge graph to locate the graph node corresponding to the entity identifiers; S403, retrieving the associated business logic and relationships associated with the graph node from the semantic knowledge graph; S404, performing feature comparison between the spatiotemporal state data and the trigger conditions set in the associated business logic and the associated paths defined in the associated relationships to perform pattern matching; S405, generating a matching result containing a matching status identifier based on the feature comparison result.

[0066] In this embodiment, the standardized event stream is injected into a semantic knowledge graph to load data with unified temporal and spatial dimensions into a queryable knowledge structure, enabling subsequent pattern matching to be completed under the semantic constraints of the graph structure. The injection action includes input access, field loading, and session identifier binding. Input access is used to receive continuous data units from the standardized event stream. Field loading is used to split each data unit into a set of indexable key fields. Session identifier binding is used to merge data units injected in the same batch into the same processing context, avoiding the mixing of cross-batch data in the same matching process.

[0067] When parsing entity identifiers and spatiotemporal state data from a standardized event stream, entity identifiers are used to uniquely identify the business object corresponding to the event. These identifiers can originate from vehicle identifiers, equipment identifiers, road unit identifiers, personnel identifiers, or other object identifiers. The parsing process locates the field containing the entity identifier using field mapping rules and performs normalization processing on the entity identifier to eliminate differences in case, prefixes, or encoding. Spatiotemporal state data describes the state of an entity at a specific point in time or within a time window. It includes elements such as timestamps, location coordinates, a set of state values, and the direction of state change. The parsing process requires time granularity normalization of timestamps, coordinate system consistency of location coordinates, and type validation and missing value handling of the state value set, ensuring that the spatiotemporal state data can be directly accessed in subsequent comparison stages. Entity identifiers and spatiotemporal state data are bound within the same event unit, forming input pairs with entity identifiers as keys and spatiotemporal state data as values. These input pairs serve as the common source for subsequent index retrieval and feature comparison.

[0068] Entity identifiers are used for indexing and retrieval in semantic knowledge graphs to quickly locate graph nodes corresponding to entity identifiers in the graph database. Indexing and retrieval includes index key construction and retrieval path selection. Index key construction maps entity identifiers to node keys that can be directly accessed in the graph database. Retrieval path selection determines whether to perform the retrieval using node identifier index, attribute index, or combined index, adapting to the storage method of entity identifiers in the graph. The location action outputs the node identifier and node type of the graph node and maintains the mapping relationship between entity identifiers and graph nodes. This mapping relationship ensures that subsequent retrieval of related business logic and relationships uses the same graph node as the scope benchmark, thereby avoiding matching object drift.

[0069] The semantic knowledge graph retrieves associated business logic and relationships linked to graph nodes to obtain the rule expressions and path constraints required for this matching. Associated business logic is a set of rules bound to graph nodes, including triggering conditions, scope of application, priority, and version identifiers. The retrieval process filters associated business logic based on the node type and node identifier of the graph nodes, and performs validity filtering to exclude outdated or inapplicable rules. Relationships are a set of relationships connected to graph nodes, including relationship type, relationship direction, relationship endpoints, and path combination constraints. The retrieval process extracts relationships based on the incoming and outgoing edge structure of the graph nodes and groups relationships by relationship type to form a set of relationships usable for path verification. A binding relationship is established between associated business logic and relationships during the retrieval phase, ensuring that each associated business logic points to a range of available relationships, thus providing a common context boundary for subsequent joint comparison of triggering conditions and associated paths.

[0070] The spatiotemporal state data is compared with the associated paths defined in the related business logic using feature comparison to perform pattern matching. Feature comparison consists of trigger condition comparison and associated path comparison. Trigger condition comparison determines whether the spatiotemporal state data meets the state change trigger conditions defined in the related business logic. The determination process includes state field mapping, threshold judgment, interval judgment, and time window constraint judgment, enabling different types of trigger conditions to be calculated within a unified judgment framework. Associated path comparison determines whether the object associations indicated by the spatiotemporal state data conform to the associated paths defined in the relationship definition. The determination process includes path endpoint verification, path direction verification, path length constraint verification, and path type sequence verification, ensuring that associated paths not only reflect connectivity but also the sequence constraints of semantic relationships. The feature comparison output results form the satisfaction status for each related business logic and each associated path. The satisfaction status serves as the basis for pattern matching. Pattern matching performs a combination judgment based on the output of feature comparison, so that the trigger condition and the associated path are combined based on the binding relationship of the same associated business logic and the associated relationship, thereby avoiding misjudgments such as only meeting the trigger condition but the path is inconsistent, or the path is consistent but the trigger condition is not met.

[0071] Based on feature comparison, matching results containing matching status identifiers are generated to structure the output of pattern matching judgments. Matching status identifiers express whether a match was successful and the matching category, and can include success, failure, and pending identifiers. Success identifiers indicate that the triggering condition and associated path meet the combined judgment; failure identifiers indicate that the triggering condition or associated path does not meet certain conditions; and pending identifiers indicate that insufficient input elements prevented a judgment from being completed. In addition to matching status identifiers, matching results can also carry the identifiers of the matched associated business logic, the matched associated path, and a list of unmet conditions. The list of unmet conditions clarifies the source of differences in triggering conditions or paths, facilitating direct reuse in subsequent processing. Matching results are bound to entity identifiers, enabling multiple event units corresponding to the same entity identifier to form a traceable sequence of matching results.

[0072] This embodiment parses entity identifiers and spatiotemporal state data in a standardized event flow to establish input pairs. Then, it uses entity identifiers to perform index retrieval in a semantic knowledge graph to locate graph nodes. Furthermore, it retrieves the associated business logic and relationships related to the graph nodes as the matching context, and simultaneously compares the spatiotemporal state data with the triggering conditions and associated paths to complete pattern matching. Finally, it generates a matching result containing matching status identifiers, making the matching process have clear object location basis, rule source basis, and path constraint basis, reducing mismatches caused by inconsistent matching objects and unclear rule application scope, and improving the structure and reusability of pattern matching output.

[0073] In one embodiment, step S50 includes: S501, when the matching result indicates a successful match, calling the semantic tag corresponding to the matched business logic from the semantic knowledge graph, and aggregating the semantic tag with the standardized event stream and associated graph nodes to generate a structured knowledge package; S502, when the matching result indicates the existence of business logic not defined in the semantic knowledge graph, extracting the corresponding data fragment from the standardized event stream as an anomaly analysis sample; S503, analyzing the anomaly analysis sample using clustering mining technology to identify potential causal dependencies; S504, transforming the potential causal dependencies into new business logic conforming to a triple structure, and adding the new business logic in the form of a triple to the semantic knowledge graph to complete the update.

[0074] In this embodiment, the matching result is used to characterize the pattern matching conclusion between the standardized event stream and the semantic knowledge graph. It can carry information such as the matching status identifier, the matched business logic identifier, the matched association path identifier, and the list of unsatisfied items. When adding semantic tags to the standardized event stream based on the matching result to generate a structured knowledge package, the semantic tags play the role of mapping the meaning of event fields into searchable and aggregable semantic units. The source of semantic tags is limited to the set of tags in the semantic knowledge graph that correspond to the matched business logic. The calling action can be implemented through table lookup of business logic identifiers, traversal of graph database relationships, and cache key-value mapping. After obtaining the semantic tags, the semantic tags are aggregated with the standardized event stream and associated graph nodes. The aggregation action requires the establishment of an aggregation key to ensure that the same event unit and the same associated graph node are merged into the same output object. The aggregation key can be formed by combining entity identifiers, timestamp granularity identifiers, spatial unit identifiers, and business logic identifiers. The aggregation process writes the original field set, semantic tag set, node identifiers and node types of the association graph nodes from the standardized event stream into the same structure. It then applies overriding, appending, or versioning strategies to handle field conflicts, ultimately forming a structured knowledge package. The structured knowledge package can be a combination of key-value pairs and lists, an object-oriented field model, or a serialized document structure. The key is maintaining the reference relationships between semantic tags and standardized event stream fields within the same output object, as well as maintaining the traceable relationships between association graph nodes and business logic identifiers.

[0075] When a matching result indicates a successful match, semantic tag invocation and aggregation are triggered to generate a structured knowledge package. This branch requires determining the matching status identifier and configuring a trigger threshold. The trigger threshold can be a strict success identifier or a combination of success and a confidence threshold. To avoid duplicate tagging and aggregation of the same standardized event stream, deduplication tags can be written before generating the structured knowledge package. These tags can be generated based on the event fingerprint of the standardized event stream, which can be obtained from entity identifiers, timestamps, spatial unit identifiers, and hashes of key status fields. After generation, the structured knowledge package can be serialized and written to a message queue, object storage, or index. The writing location does not change the logical closed loop of the processed content, but it will affect the efficiency of subsequent retrieval and tracking. Therefore, it is possible to choose to write by partitioning according to business logic identifiers or by bucketing according to time windows.

[0076] When the matching result indicates the presence of business logic not defined in the semantic knowledge graph, the process proceeds to the anomaly analysis branch. The determination of undefined business logic can be expressed by the miss category in the matching status identifier, or by the absence of both the list of non-satisfied items and the candidate business logic set. To ensure reproducible experiments and traceable origins for anomaly analysis samples, the truncation boundaries and rules must be clearly defined when extracting corresponding data segments from the standardized event stream. The truncation boundary can be a single event unit or a group of event units bounded by a time window. The time window can be determined by a timestamp and a preset window length, or it can be adaptively expanded by a state change point. The truncation rules need to retain the set of fields related to the determination of undefined business logic. This set of fields can include entity identifiers, timestamps, spatial unit identifiers, key state fields from the spatiotemporal state data, and mapping fields from the list of non-satisfied items in the matching result. After extraction, the data fragments are written into the anomaly analysis sample set. The anomaly analysis sample set can be appended and include sample metadata. The sample metadata can include the matching status identifier that triggered the sample, the identifier of the association graph node, the identifier of the business logic candidate set, and the sampling time, so that the sample context can be traced back during subsequent analysis.

[0077] When using clustering mining techniques to analyze anomaly samples, the first step is to convert these samples into computable sample representations. These representations can be directly composed of numerical fields, encoded discrete fields, or obtained from time-series fields through windowed statistics. The construction of these representations should maintain a reversible mapping relationship with the set of fields in the anomaly analysis samples, allowing for the rewriting of the original field semantics after identifying potential causal dependencies. The execution of clustering mining techniques includes sample representation normalization, distance metric selection, cluster number or density parameter setting, and intra-cluster consistency testing. Distance metrics can be Euclidean distance, cosine similarity, or graph-based distance metrics. Parameter settings can be fixed or automatically adjusted based on sample size and variance. The analysis output includes cluster partitioning results, cluster center representations, high-frequency co-occurrence field combinations within clusters, and candidate conditions for inter-field dependencies. When identifying potential causal dependencies, it is necessary to filter out relationship pairs that satisfy directional and temporal constraints from co-occurrence and conditional dependency candidates. Directional constraints can be inferred from the business semantic type of the fields or verified by the time stamp sequence. Temporal constraints require that changes in the cause field occur before changes in the result field or within an interpretable lag interval. Potential causal dependencies are ultimately output as a combination of field pairs and constraints. Constraints can include threshold ranges, trigger frequency thresholds, spatial unit limitations, and time window limitations to ensure executability when subsequently transformed into new business logic.

[0078] When transforming potential causal dependencies into new business logic conforming to a triple structure, the field pairs, directions, and constraints in the potential causal dependencies need to be mapped to a set of elements in a subject, predicate, and object triple structure. The subject can be mapped to an entity or entity state, the object can be mapped to another entity state or event conclusion, and the predicate can be mapped to causal triggers, conditional constraints, or evolutionary directions. Constraints can be decomposed into additional triples or stored as triple attributes. When storing attributes, a set of attribute keys must be defined to avoid name conflicts. New business logic should carry a version identifier and scope of effect during generation. The version identifier is used for coexistence management with existing business logic, and the scope of effect is used to limit the applicable spatial unit and time window to prevent the newly generated logic from mistakenly triggering unrelated data. When adding new business logic in the form of triples to the semantic knowledge graph to complete the update, the write operation needs to undergo conflict detection and duplication detection. Conflict detection is used to identify cases where the trigger conditions of the new business logic are mutually exclusive or the conclusions are contradictory. Duplication detection is used to identify cases where the new business logic is equivalent to or contains the existing triple structure. Conflict handling can be achieved through priority overwrite, version fork, or parallel retention with mutual exclusion conditions. After the write is completed, the index can be updated so that the new business logic can be invoked in subsequent retrieval and matching.

[0079] This embodiment retrieves semantic tags corresponding to the matched business logic from the semantic knowledge graph when the matching result indicates a successful match. These semantic tags are then aggregated with standardized event streams and associated graph nodes to generate a structured knowledge package. This ensures that the standardized event streams have stable semantic annotations and context binding structures on the output side, reducing cross-field interpretation costs and improving the reusability of the structured output. Furthermore, when the matching result indicates the presence of business logic not defined in the semantic knowledge graph, data fragments from the standardized event stream are extracted to form anomaly analysis samples. Clustering mining techniques are then used to identify potential causal dependencies and transform them into new business logic conforming to a triple structure, which is then added to the semantic knowledge graph. This enables continuous completion of uncovered business logic and the accumulation of executable rules, reducing long-term misses caused by knowledge gaps and improving the scalability and iteration efficiency of the semantic knowledge graph.

[0080] In one embodiment, step S60 includes: S601, calling an intelligent analysis model adapted to the current business scenario from a pluggable model library, and inputting the structured knowledge package into the intelligent analysis model; S602, using the intelligent analysis model to retrieve associated contextual knowledge in the semantic knowledge graph, and performing scenario-based analysis based on the associated contextual information and the structured knowledge package to generate analysis results containing causal diagnosis or trend prediction data; S603, extracting key indicators from the analysis results, and retrieving response plan templates matching the key indicators from a preset strategy library; S604, instantiating and filling the response plan templates using specific parameter values ​​from the analysis results to generate decision suggestions containing specific handling instructions.

[0081] In this embodiment, a pluggable model library is used to host multiple intelligent analysis models for different business scenarios. It provides metadata such as model enumeration, model version, input / output interface contracts, and runtime parameter sets, enabling the calling end to switch model instances without altering the external process. The determination of adaptation to the current business scenario can be based on the mapping relationship between scenario identifiers, scenario parameter sets, and model metadata. This mapping relationship can be maintained in the form of rule tables, configuration files, or registry entries. Matching conditions can include constraints such as scenario type, spatial granularity, time window granularity, output target category, and data field coverage. Calling an intelligent analysis model requires model loading and runtime context initialization. The runtime context can include resource quotas, inference batch size, timeout threshold, concurrency, and caching strategies to ensure consistent input processing paths for the same scenario across different runtime environments. When a structured knowledge package is input into the intelligent analysis model, input adaptation must be completed first. Input adaptation includes field selection, field normalization, semantic tag encoding, and context reference expansion. Field selection is used to extract a subset of fields from the structured knowledge package that are relevant to the current model contract. Field normalization is used to map numerical fields of different dimensions to a unified scale. Semantic label encoding is used to transform the set of semantic labels into a discrete identifier sequence or sparse feature set that the model can process. Context reference expansion is used to transform references to graph nodes within the structured knowledge package into node attribute fragments or relation path summaries that can be directly consumed by the model, thereby giving the model input a definite data type and a definite field order.

[0082] When intelligent analysis models retrieve associated contextual knowledge in a semantic knowledge graph, they need to utilize the graph node identifiers, entity identifiers, and semantic categories in the semantic tags carried in the structured knowledge package as retrieval entry points. The retrieval action can be implemented using a graph database query language or a graph traversal interface. The associated contextual information points to graph substructure information semantically related to the current event, including a set of node attributes, a set of relation types, a set of relation directions and relation path constraints, and, if necessary, trigger condition fragments and evolution constraint fragments related to business logic. The retrieval results need to be pruned and sorted. Pruning limits path length, the range of returned node types, and the range of returned attribute fields. Sorting rearranges candidate contextual fragments based on path weight, node importance, or time freshness, thus forming stable associated contextual information. Contextual analysis uses associated contextual information and structured knowledge packages as parallel input sources. The fusion location of these two needs to be clearly defined; fusion can be performed at the model input level or in the middle layer of the model. Input-side fusion typically encodes associated contextual information into a set of contextual features and concatenates it with features from a structured knowledge package. Intermediate-layer fusion typically uses contextual features as attention gating, conditional constraints, or graph convolutional neighborhood sets in computation. Regardless of the fusion location, the computational goal of scenario-based analysis is to output analytical results. The data structure of these results must at least include fields carrying causal diagnosis or trend prediction data. Causal diagnosis can be represented as a set of influencing factors, a set of influencing factor weights, triggering link segments, or conflict constraint segments. Trend prediction can be represented as a sequence of indicators for future time windows, the slope of change, peak and trough times, and confidence intervals. To ensure the analytical results can be used for subsequent retrieval and template filling, the results need to be structured. Structured output can include field names, field types, value ranges, and unit information, and provides null and missing value markers for missing fields to avoid misinterpretation in subsequent processing.

[0083] When extracting key indicators (KPIs) from the analysis results, it is necessary to first define the screening criteria for the KPIs. These criteria can be defined by threshold conditions, ranking conditions, or a combination of multiple conditions. Threshold conditions are used to select indicators that exceed a threshold or fall within a given range. Ranking conditions are used to select indicators with the highest weight or the largest fluctuation range. Combinations of multiple conditions are used to simultaneously satisfy both importance and manageability constraints. In addition to outputting the name and value of the KPIs, the extraction process should also output the unit, time window, spatial range, and source field reference. The source field reference is used to form a specific search key when searching the strategy library. The strategy library stores response plan templates and template indexes. Indexes can contain applicable condition expressions, sets of key indicator constraints, priorities, and applicable scenario identifiers. When retrieving response plan templates that match the KPIs, the KPIs need to be converted into template search conditions. Search conditions can include a set of indicator names, indicator threshold ranges, scenario identifiers, and a set of context tags. The matching process can employ rule-based matching, expression evaluation, or vector similarity retrieval. The matching results can return one or more response plan templates, and the final template is selected based on priority and coverage. Coverage is used to measure the degree of matching between the template conditions and the set of key indicators, avoiding the selection of templates with many missing conditions.

[0084] Instantiation population uses specific parameter values ​​from the analysis results to assign values ​​to the populateable fields of the response plan template. These specific parameter values ​​point to numerical or enumerated values ​​that directly drive the action, including threshold parameters, resource quantity parameters, time window parameters, spatial range parameters, and action switch parameters. Instantiation population requires maintaining field mapping relationships, establishing a one-to-one or one-to-many mapping between response plan template field names and analysis result field names, and defining conversion functions to handle unit conversion, rounding rules, upper and lower limit truncation, and default value rollback. The population process also requires consistency checks, which verify whether parameter values ​​meet template field constraints. Field constraints can include value ranges, mutual exclusion relationships, and dependencies. Mutual exclusion relationships prevent the simultaneous triggering of contradictory instructions, while dependencies ensure that subsequent instructions are only generated if the preceding parameters exist. After instantiation population is complete, decision recommendations are generated. The data structure of these recommendations can be an instruction list, a structured document field set, or a process orchestration description, at least including specific action instructions and instruction parameter sets, along with references to the analysis results for audit tracing.

[0085] Example Description: The system acquires multi-source heterogeneous data from various sources, including checkpoint systems, electronic police systems, video surveillance, signal controllers, floating car GPS, traffic broadcasts, weather stations, road construction management systems, and internet navigation platforms. This multi-source heterogeneous data simultaneously contains structured records, semi-structured logs, and streaming messages. The system performs format parsing and cleaning / completening on the multi-source heterogeneous data to obtain a basic formatted data. The system is configured with multiple types of data access adapters to receive multi-source heterogeneous data containing structured records, semi-structured logs, and streaming messages. For example, it receives checkpoint vehicle passage records through a database connection adapter, signal controller operation logs through a log acquisition adapter, and streaming positioning messages from floating car GPS through a message subscription adapter. The system identifies the data format or access protocol of the received multi-source heterogeneous data and uses a conversion strategy corresponding to the identified data format or access protocol to map the multi-source heterogeneous data to a preset unified intermediate model to complete format parsing, obtaining data after format parsing. For example, vehicle identification fields from different systems are uniformly mapped to vehicle identification fields in the unified intermediate model, location fields from different systems are uniformly mapped to spatial coordinate fields in the unified intermediate model, and time fields from different systems are uniformly mapped to timestamp fields in the unified intermediate model. The system detects whether there are redundant duplicate records and / or conflicting data in the format-parsed data, and performs a cleaning process to remove redundant duplicate records and / or conflicting data, obtaining cleaned data. For example, duplicate vehicle passes reported by different sources with the same vehicle identification field and the same timestamp field are deduplicated, and conflicting records with inconsistent spatial coordinate fields corresponding to the same location number field are removed or retained by a trusted source. The system fills in the missing key fields in the cleaned data based on a preset strategy or interpolation strategy to perform completion, and obtains the primary format data. For example, for records with missing spatial coordinate fields, the system queries the geographic dictionary based on the location number field to fill in the missing field. For floating car messages with missing speed fields, the system performs an interpolation strategy based on the trajectory points of adjacent timestamp fields to obtain the speed field.

[0086] After obtaining the initial format data, the system establishes a unified spatiotemporal reference grid and aligns the initial format data into the grid to generate a standardized event stream. The system constructs a unified spatiotemporal reference grid that includes time, geographic location, and lane-level coordinate dimensions. For example, the time dimension is divided into continuous time windows with fixed time granularity, the geographic location dimension is divided into grid cells covering the roads within the jurisdiction, and the lane-level coordinate dimension is used to describe the relative position or absolute coordinate range of each lane within a road segment. The system sets a time synchronization tolerance threshold and a spatial positioning error range for the unified spatiotemporal reference grid. For example, the time synchronization tolerance threshold is configured to a range of several seconds, and the spatial positioning error range is configured to a range of several meters, to adapt to clock and positioning deviations from different data sources. The system extracts timestamp and spatial coordinate information from the primary format data and maps it to a unified spatiotemporal reference grid to obtain projected data. For example, the timestamp of a vehicle passage record is placed into the corresponding time window, and its spatial coordinates are placed into the corresponding geographic grid cell. Furthermore, its lane-level location is determined based on lane-level coordinate dimensions. The system uses time synchronization tolerance thresholds and spatial positioning error ranges to perform spatiotemporal calibration on the projected data, generating a standardized event stream. For instance, when the same traffic event is reported separately by video surveillance and checkpoint systems and there is a slight offset in the timestamp information, it is calibrated to the same event identifier within the same time window within the time synchronization tolerance threshold. When the floating car GPS jitters around the same grid cell within the spatial positioning error range, it is calibrated to a stable event location within the same geographic grid cell. This allows the standardized event stream to be organized with unified time and geographic dimensions and maintains a consistent expression of lane-level coordinate dimensions.

[0087] After generating standardized event streams, the system constructs a semantic knowledge graph containing entities, attributes, relationships, and business logic. The system defines a subject-predicate-object triple structure under a resource description framework as the basic data organization form for the semantic knowledge graph. For example, vehicles, intersections, road segments, traffic events, weather conditions, and construction areas are used as subjects or objects, and words like "passed through," "located in," "caused," "triggered," and "represented" are used as predicates to form queryable and reasonable triple data. The system utilizes a natural language processing module to extract entities and their attribute features from unstructured business text to construct the entities and attributes of the semantic knowledge graph. For example, traffic event entities and location attribute features are extracted from police report texts, and illegal behavior entities and license plate color attribute features are extracted from law enforcement record texts. The system defines dynamic interaction paths and pre- and post-dependent policies between entities to construct a semantic knowledge graph describing the flow of business operations. For example, it defines dynamic interaction paths where vehicles pass through road segments and then intersections, and pre- and post-dependent policies where severe weather occurs first, then affects visibility, and then affects vehicle speed, forming a set of relationships that express the evolution of traffic. The system also configures business logic strategies and defines state change triggering conditions and evolutionary models for entities to construct the business logic of the semantic knowledge graph. For example, it defines state change triggering conditions for multiple consecutive days of morning and evening rush hours occurring at the same congestion point to form business logic related to commuting behavior, and defines state change triggering conditions for multiple speeding violations in a single day to form business logic related to key monitoring. On the system side, entities, attributes, relationships, and business logic are mapped to a subject, predicate, and object triple structure. The mapped triple data is stored in a graph database to complete the construction of a semantic knowledge graph. For example, the vehicle entity and license plate color attribute feature are mapped to a triple of vehicle—has—license plate color; the rainstorm entity and vehicle speed decrease state are mapped to a triple of rainstorm—cause—vehicle speed decrease; and the commuting determination-related business logic is mapped to a triple of commuting trigger condition—trigger—commuting label and written into the graph database.

[0088] After the semantic knowledge graph is constructed, the system injects standardized event streams into the semantic knowledge graph for pattern matching to obtain matching results. The system parses entity identifiers and spatiotemporal state data from the standardized event streams, such as vehicle entity identifiers, road segment entity identifiers, and corresponding spatiotemporal state data such as time windows, geographic grid units, and lane-level coordinate dimensions. The system uses entity identifiers to perform index retrieval in the semantic knowledge graph, locating the graph nodes corresponding to the entity identifiers; for example, locating vehicle graph nodes by vehicle entity identifiers and intersection graph nodes by intersection entity identifiers. The system retrieves the associated business logic and relationships related to the graph nodes from the semantic knowledge graph; for example, retrieving commuting determination-related business logic and key supervision-related business logic associated with vehicle graph nodes, and retrieving the defined transit relationships and associated paths between vehicle graph nodes and road segment graph nodes. The system performs feature comparison between spatiotemporal state data and the associated paths defined in the relevant business logic, using trigger conditions and relationships to perform pattern matching. For example, it compares the frequency of a vehicle's appearance within multiple time windows with the frequency thresholds in the trigger conditions, and compares the vehicle's migration paths between multiple geographic grid cells with associated paths to identify whether commuting-related or abnormal driving-related patterns are met. Based on the feature comparison results, the system generates matching results including matching status identifiers. For example, it generates a matching status identifier for successful matches along with the matched business logic identifier and the matched associated path identifier, or it generates a matching status identifier for unmatched matches along with a description of the differences in the conditions not met.

[0089] After obtaining the matching results, the system adds semantic tags to the standardized event stream to generate a structured knowledge package. When the matching results indicate the presence of business logic not defined in the semantic knowledge graph, new business logic is generated to update the semantic knowledge graph. When the matching result indicates a successful match, the system retrieves the semantic tags corresponding to the matched business logic from the semantic knowledge graph and aggregates these tags with the standardized event stream and associated graph nodes to generate a structured knowledge package. For example, when matching business logic related to commuting determination, the system retrieves semantic tags related to commuting hotspots and aggregates them with standardized event fragments of the vehicle within the corresponding time window and geographic grid unit, as well as vehicle and road graph nodes, to generate a structured knowledge package. This allows the structured knowledge package to carry directly understandable semantic tags and corresponding event context. When the matching result indicates the presence of business logic not defined in the semantic knowledge graph, the system extracts corresponding data fragments from the standardized event stream as anomaly analysis samples. For example, it extracts continuous event fragments containing hazardous chemical vehicles entering a restricted area at night without escort information as anomaly analysis samples for subsequent analysis. The system utilizes clustering mining techniques to analyze anomaly analysis samples to identify potential causal dependencies. For example, it clusters time windows, geographic grid units, vehicle type attributes, illegal behavior attributes, and weather condition attributes in the anomaly analysis samples to identify potential causal dependencies where anomaly events frequently occur under specific combinations of conditions. The system then transforms these potential causal dependencies into new business logic conforming to a triple structure and adds this new business logic as triples to the semantic knowledge graph for updates. For instance, the dependency between nighttime hours and restricted area entry behavior and vehicle type conditions is transformed into trigger condition triples, and the corresponding regulatory warning semantic tag triples are written into the graph database. This allows subsequent similar events to be directly matched and generate stable semantic tags.

[0090] After generating the structured knowledge package, the system performs scenario-based analysis based on the structured knowledge package and semantic knowledge graph to generate analysis results, and then generates decision recommendations based on the analysis results. The system calls intelligent analysis models adapted to the current business scenario from a pluggable model library, and inputs the structured knowledge package into the intelligent analysis models. For example, in the congestion management scenario, it calls intelligent analysis models related to congestion cause tracing, and in the accident assessment scenario, it calls intelligent analysis models related to accident black spot cause diagnosis, and uses the structured knowledge package carrying semantic tags and graph node context as model input. The system utilizes intelligent analysis models to retrieve relevant contextual knowledge from a semantic knowledge graph. Based on this contextual information and structured knowledge packages, it performs scenario-based analysis to generate analytical results containing causal diagnosis or trend prediction data. For example, it retrieves contextual knowledge related to road segments, such as road grade attributes, construction area attributes, weather conditions, and signal timing. Combined with event fragments reflected in the structured knowledge packages, such as decreased vehicle speed and increased queue length, it generates causal diagnosis data to identify the combined factors causing reduced traffic capacity due to construction areas and heavy rain, or generates trend prediction data to predict a potential increase in the congestion index within several future time windows. The system extracts key indicators from the analysis results and retrieves response plan templates matching these indicators from a pre-set strategy library. For example, it extracts key indicators such as congestion index, average speed, and secondary accident risk level from the analysis results and retrieves corresponding service deployment suggestion templates, traffic organization optimization plan templates, or risk warning notice templates from the strategy library. The system uses specific parameter values ​​from the analysis results to instantiate and populate the response plan template to generate decision suggestions containing specific handling instructions. For example, specific parameter values ​​such as suggested speed limit, recommended number of police officers, signal timing adjustment range, and guidance information release time window are filled into the template fields to generate decision suggestions containing specific handling instructions such as dispatching additional police officers to designated intersections, adjusting speed limits on designated road sections, and releasing guidance information to navigation platforms. The decision suggestions can be pushed to the command center or mobile terminals in a structured form for rapid execution and feedback.

[0091] This embodiment utilizes a pluggable model library to call intelligent analysis models adapted to the current business scenario, and incorporates structured knowledge packages into the intelligent analysis model through a defined input adaptation process, ensuring interface stability during scenario changes and model upgrades. The intelligent analysis model retrieves relevant contextual knowledge from the semantic knowledge graph and, together with the structured knowledge packages, drives scenario-based analysis, enabling the analysis results to simultaneously carry causal diagnosis or trend prediction data while maintaining structured output. By extracting key indicators from the analysis results and retrieving matching response plan templates from the strategy library, template selection and analysis output are bound together in an interpretable manner. By instantiating and filling the response plan templates with specific parameter values ​​from the analysis results and performing consistency checks, decision recommendations are output as specific handling instructions with traceable parameter sources, thereby improving the efficiency of knowledge-to-decision transformation and output consistency.

[0092] In one embodiment, a semantic knowledge graph-based intelligent decision-making device is provided, which corresponds one-to-one with the semantic knowledge graph-based intelligent decision-making method in the above embodiments. Referring to Figure 3, Figure 3 is a functional module diagram of a preferred embodiment of the semantic knowledge graph-based intelligent decision-making device of the present invention. The module includes a data preprocessing module 10, a spatiotemporal alignment module 20, a knowledge graph module 30, a pattern matching module 40, a knowledge evolution module 50, and a decision analysis module 60. The functional modules are described in detail below: Data preprocessing module 10 is used to acquire multi-source heterogeneous data, perform format parsing and cleaning / completening on the multi-source heterogeneous data to obtain primary format data; Spatiotemporal alignment module 20 is used to establish a unified spatiotemporal reference grid, align the primary format data to the spatiotemporal reference grid, and generate a standardized event stream; Knowledge graph module 30 is used to construct a semantic knowledge graph containing entities, attributes, relationships, and business logic; Pattern matching module 40 is used to inject the standardized event stream into the semantic knowledge graph for pattern matching to obtain matching results; Knowledge evolution module 50 is used to add semantic tags to the standardized event stream based on the matching results to generate a structured knowledge package, and generate new business logic to update the semantic knowledge graph when the matching results indicate the existence of business logic not defined in the semantic knowledge graph; Decision analysis module 60 is used to perform scenario-based analysis based on the structured knowledge package and the semantic knowledge graph, generate analysis results, and generate decision suggestions based on the analysis results.

[0093] Specific limitations regarding the semantic knowledge graph-based intelligent decision-making device can be found in the aforementioned limitations on the semantic knowledge graph-based intelligent decision-making method, and will not be repeated here. Each module in the aforementioned semantic knowledge graph-based intelligent decision-making device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0094] In one embodiment, a computer device is provided, which can be a server, and its internal structure diagram is shown in Figure 4. The computer device includes a processor, memory, a network interface, and a database connected via a system bus. The processor of the computer device provides determination and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side intelligent decision-making method based on semantic knowledge graphs.

[0095] In one embodiment, a computer device is provided, which can be a client, and its internal structure diagram is shown in Figure 5. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device provides determination and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a client-side intelligent decision-making method based on semantic knowledge graphs.

[0096] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: acquiring multi-source heterogeneous data; parsing, cleaning, and completing the format of the multi-source heterogeneous data to obtain primary format data; establishing a unified spatiotemporal reference grid; aligning the primary format data to the spatiotemporal reference grid to generate a standardized event stream; constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic; injecting the standardized event stream into the semantic knowledge graph for pattern matching to obtain a matching result; adding semantic tags to the standardized event stream based on the matching result to generate a structured knowledge package; and generating new business logic to update the semantic knowledge graph when the matching result indicates the existence of business logic not defined in the semantic knowledge graph; performing scenario-based analysis based on the structured knowledge package and the semantic knowledge graph to generate analysis results; and generating decision suggestions based on the analysis results.

[0097] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, storing a computer program thereon. When the computer program is executed by a processor, it performs the following steps: acquiring multi-source heterogeneous data; parsing, cleaning, and completing the format of the multi-source heterogeneous data to obtain primary format data; establishing a unified spatiotemporal reference grid; aligning the primary format data to the spatiotemporal reference grid to generate a standardized event stream; constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic; injecting the standardized event stream into the semantic knowledge graph for pattern matching to obtain matching results; adding semantic tags to the standardized event stream based on the matching results to generate a structured knowledge package; and generating new business logic to update the semantic knowledge graph when the matching results indicate the existence of business logic not defined in the semantic knowledge graph; performing scenario-based analysis based on the structured knowledge package and the semantic knowledge graph to generate analysis results; and generating decision suggestions based on the analysis results.

[0098] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0100] It should be noted that if any AI models, software tools, or components not belonging to this company appear in the embodiments of this application, they are merely illustrative examples and do not represent actual use. The above-described embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An intelligent decision-making method based on semantic knowledge graphs, characterized in that, The process includes the following steps: acquiring multi-source heterogeneous data, performing format parsing, cleaning, and completion on the multi-source heterogeneous data to obtain primary format data; establishing a unified spatiotemporal reference grid, aligning the primary format data to the spatiotemporal reference grid, and generating a standardized event flow; and constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic. The standardized event stream is injected into the semantic knowledge graph for pattern matching to obtain a matching result; based on the matching result, semantic tags are added to the standardized event stream to generate a structured knowledge package; and when the matching result indicates the existence of business logic not defined in the semantic knowledge graph, new business logic is generated to update the semantic knowledge graph. Contextual analysis is performed based on the structured knowledge package and the semantic knowledge graph to generate analysis results, and decision recommendations are generated based on the analysis results.

2. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, The process involves acquiring multi-source heterogeneous data, performing format parsing and cleaning / completening on the multi-source heterogeneous data to obtain primary format data, including: configuring multiple types of data access adapters to receive multi-source heterogeneous data containing structured records, semi-structured logs, and streaming messages respectively; identifying the data format or access protocol of the received multi-source heterogeneous data, and using a conversion strategy corresponding to the identified data format or access protocol to map the multi-source heterogeneous data to a preset unified intermediate model to complete format parsing, obtaining format-parsed data; detecting whether there are redundant duplicate records and / or conflicting data in the format-parsed data, performing removal processing on the redundant duplicate records and / or conflicting data to clean the data, obtaining cleaned data; and filling in missing key fields in the cleaned data based on a preset strategy or interpolation strategy to complete the primary format data.

3. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, Establishing a unified spatiotemporal reference grid, aligning the primary format data to the spatiotemporal reference grid, and generating a standardized event stream include: constructing a unified spatiotemporal reference grid containing time, geographic location, and lane-level coordinate dimensions; setting a time synchronization tolerance threshold and spatial positioning error range for the unified spatiotemporal reference grid; extracting timestamp information and spatial coordinate information from the primary format data, and using the timestamp information and spatial coordinate information to map the primary format data to the unified spatiotemporal reference grid to obtain projected data; and performing spatiotemporal calibration on the projected data using the time synchronization tolerance threshold and the spatial positioning error range to generate a standardized event stream.

4. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, Constructing a semantic knowledge graph that includes entities, attributes, relationships, and business logic includes: defining a subject, predicate, and object triple structure under a resource description framework as the basic data organization form of the semantic knowledge graph; extracting entities and their attribute features from unstructured business text using a natural language processing module to construct the entities and attributes of the semantic knowledge graph; defining dynamic interaction paths and pre- and post-dependency strategies between the entities to construct relationships in the semantic knowledge graph that describe the business flow; configuring business logic strategies and defining state change triggering conditions and evolution models for the entities to construct the business logic of the semantic knowledge graph; mapping the entities, attributes, relationships, and business logic to the subject, predicate, and object triple structure, and storing the mapped triple data in a graph database to complete the construction of the semantic knowledge graph.

5. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, The process of injecting the standardized event stream into the semantic knowledge graph for pattern matching to obtain a matching result includes: parsing entity identifiers and spatiotemporal state data from the standardized event stream; using the entity identifiers to perform index retrieval in the semantic knowledge graph to locate the graph node corresponding to the entity identifiers; retrieving the associated business logic and relationships associated with the graph node from the semantic knowledge graph; performing feature comparison between the spatiotemporal state data and the trigger conditions set in the associated business logic and the associated paths defined in the associated relationships to perform pattern matching; and generating a matching result containing a matching status identifier based on the feature comparison result.

6. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, Based on the matching results, semantic tags are added to the standardized event stream to generate a structured knowledge package. When the matching results indicate the existence of business logic not defined in the semantic knowledge graph, new business logic is generated to update the semantic knowledge graph. This includes: when the matching results indicate a successful match, calling the semantic tags corresponding to the matched business logic from the semantic knowledge graph, and aggregating the semantic tags with the standardized event stream and associated graph nodes to generate a structured knowledge package; when the matching results indicate the existence of business logic not defined in the semantic knowledge graph, extracting corresponding data fragments from the standardized event stream as anomaly analysis samples; analyzing the anomaly analysis samples using clustering mining techniques to identify potential causal dependencies; transforming the potential causal dependencies into new business logic conforming to a triple structure, and adding the new business logic in triple form to the semantic knowledge graph to complete the update.

7. The intelligent decision-making method based on semantic knowledge graph as described in claim 1, characterized in that, Based on the structured knowledge package and the semantic knowledge graph, scenario-based analysis is performed to generate analysis results. Based on these results, decision recommendations are generated, including: calling an intelligent analysis model adapted to the current business scenario from a pluggable model library and inputting the structured knowledge package into the intelligent analysis model; using the intelligent analysis model to retrieve associated contextual knowledge from the semantic knowledge graph, and performing scenario-based analysis based on the associated contextual information and the structured knowledge package to generate analysis results containing causal diagnosis or trend prediction data; extracting key indicators from the analysis results, and retrieving response plan templates matching the key indicators from a preset strategy library; and instantiating and populating the response plan templates using specific parameter values ​​from the analysis results to generate decision recommendations containing specific handling instructions.

8. An intelligent decision-making device based on semantic knowledge graphs, characterized in that, The intelligent decision-making device based on semantic knowledge graph includes: a data preprocessing module for acquiring multi-source heterogeneous data, performing format parsing and cleaning / completening on the multi-source heterogeneous data to obtain primary format data; a spatiotemporal alignment module for establishing a unified spatiotemporal reference grid, aligning the primary format data to the spatiotemporal reference grid, and generating a standardized event stream; a knowledge graph module for constructing a semantic knowledge graph containing entities, attributes, relationships, and business logic; a pattern matching module for injecting the standardized event stream into the semantic knowledge graph for pattern matching to obtain a matching result; a knowledge evolution module for adding semantic tags to the standardized event stream based on the matching result to generate a structured knowledge package, and generating new business logic to update the semantic knowledge graph when the matching result indicates the existence of business logic not defined in the semantic knowledge graph; and a decision analysis module for performing scenario-based analysis based on the structured knowledge package and the semantic knowledge graph, generating analysis results, and generating decision suggestions based on the analysis results.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a semantic knowledge graph-based intelligent decision-making program stored in the memory and executable on the processor. When executed by the processor, the semantic knowledge graph-based intelligent decision-making program implements the steps of the semantic knowledge graph-based intelligent decision-making method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores an intelligent decision-making program based on a semantic knowledge graph, which, when executed by a processor, implements the steps of the intelligent decision-making method based on a semantic knowledge graph as described in any one of claims 1-7.

Images