Hybrid expert mode intelligent inspection management system under cloud edge collaborative architecture

Abstract

The invention provides an intelligent inspection management system in a hybrid expert mode under a cloud edge collaborative architecture, and relates to the technical field of data processing, the system is used for cutting basic inspection data into a plurality of section units according to a business time window, and combining record elements and node sequences in each section unit to obtain a combined data segment; constructing a track series sequence by taking the occurrence sequence of each record element as a reference, and converting the track series sequence into track matrix data; constructing node load data and calculating a process aggregation factor; screening out candidate sections in section combination data and constructing a section traffic matrix; calculating a process shunting factor; and determining process node data, and constructing an inspection service link according to the process node data. According to the invention, the problem of multiple circulation of error events in the process can be solved, and the uniqueness of the inspection task is ensured.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent inspection management system with a hybrid expert mode under a cloud-edge collaborative architecture. Background Technology

[0002] In road inspection and management scenarios, existing AI-based inspection systems typically manage image and depth information acquisition, data processing, anomaly detection, and task reporting as a unified business process. This process involves edge devices acquiring road images, directly calling a single recognition model to generate preliminary inspection results, and then synchronously uploading the results along with the raw data to a backend management platform. The platform then registers events, processes tasks, and updates handling records according to fixed rules. This entire process is centralized; the analysis, reporting, and distribution modules in the inspection chain all rely on fixed algorithm paths and unified process templates, potentially making it difficult to dynamically adjust backend resources or schedule different recognition models based on input data conditions.

[0003] However, fixed processes may not be able to dynamically adjust analysis strategies for multi-target recognition problems in complex scenarios, leading to the entry of abnormal events, incorrect distribution, or repeated handling. For example, in the inspection of shadow areas on elevated roads, when edge devices continuously pass through areas of sudden changes in lighting, a single model may misjudge the shadow texture as road surface cracks. The erroneous event is directly written into the inspection process and triggers task distribution. Existing platforms may lack a dynamic scheduling mechanism based on data features. This false alarm event may flow multiple times in the backend process, including repeatedly creating handling tasks, generating multiple event records, and triggering invalid dispatches, causing process nodes to be occupied by erroneous events, reducing the accuracy and stability of road inspection services. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent inspection and management system with a hybrid expert mode under a cloud-edge collaborative architecture, which aims to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A cloud-edge collaborative architecture-hybrid expert mode intelligent inspection management system, the system comprising: The segment module is used to acquire basic inspection data and divide the basic inspection data into multiple segment units according to the business time window. The record elements and node order within each segment unit are combined to obtain segment combined data. The trajectory module is used to construct a trajectory concatenation sequence based on the occurrence order of each recorded element according to the segment combination data, and to convert the trajectory concatenation sequence into trajectory matrix data. The aggregation module is used to construct node load data based on trajectory matrix data and calculate process aggregation factors based on node load data. The flow splitting module is used to filter out candidate segments from the segment combination data based on the process aggregation factor, construct a segment flow matrix based on the candidate segments, and calculate the process flow splitting factor. The scheduling module is used to select the corresponding expert processing path based on the combination range of process aggregation factor and process diversion factor, and obtain scheduling path data. The link module is used to determine process node data based on scheduling path data and to construct inspection business links based on process node data.

[0006] Furthermore, the segment module includes: The data processing unit is used to arrange the basic inspection data in chronological order according to the time stamps in the basic inspection data to obtain time series data. The segment generation unit is used to segment time series data according to the range of the business time window, generate multiple segment units, and obtain segment unit data; The structural decomposition unit is used to break down the record content of each segment unit into record elements and node sequences based on the segment unit data, thereby obtaining segment element node data. The combined construction unit is used to arrange and combine the record elements and node sequences into a continuous structure according to the segment unit numbering order based on the segment element node data, so as to obtain the segment combined data.

[0007] Furthermore, the trajectory module includes: The sequential extraction unit is used to extract the sequential position corresponding to each record element according to the arrangement order of record elements in the segment combination data, so as to obtain the record sequence data; The index building unit is used to determine the numbering based on the sequential position of the record elements in the record sequence data, and to generate a feature index for each record element to obtain sequential index data; The trajectory concatenation unit is used to connect the feature indexes of the same record feature according to the arrangement order in the segment combination data based on the sequential index data, to construct a continuous index sequence and obtain the trajectory concatenation sequence. The matrix generation unit is used to generate trajectory matrix data by taking the index sequence number as the row, the connection order of each element in each index sequence as the column, and filling each feature index into the corresponding row and column positions according to the trajectory concatenation sequence.

[0008] Furthermore, the aggregation module includes: The row-level calculation unit is used to calculate row-level statistical values ​​based on the number of feature indexes in each row of the trajectory matrix data, and obtain row-level density data. The segment analysis unit is used to identify the changes in the occurrence of each row in different segment units based on the trajectory matrix data and segment combination data, and to obtain segment change data. The jump recognition unit is used to count the number of cross-segment jumps between segment units in each row based on trajectory matrix data and segment combination data, and to obtain cross-segment jump data. The load construction unit is used to combine row-level density data, segment change data, and cross-segment jump data into node load data.

[0009] Furthermore, the aggregation module also includes: Based on the node load data, the load concentration degree of each row in the trajectory matrix is ​​calculated to obtain the density score; the stability of the load between each segment is calculated to obtain the change suppression score; based on the density score and the change suppression score, nodes with concentrated and stable loads are identified to obtain the aggregation item. Based on the node load data, the jump frequency of the node between each segment unit is calculated to obtain the jump score; the density score is adjusted by power to obtain the density reinforcement score; based on the density reinforcement score and the jump score, the cross-segment continuity is calculated to obtain the balance term. Based on the node load data, the load fluctuation intensity of each row between each segment unit is calculated to obtain the change intensity score; the degree of discontinuity across segments of each row is calculated to obtain the jump intensity score; based on the change intensity score and the jump intensity score, nodes with high fluctuation and high jump are identified to obtain the suppression term. By integrating the aggregation term, balancing term, and suppression term, the aggregation stability of the corresponding nodes in each row in the entire business process is calculated, and the process aggregation factor is obtained.

[0010] Furthermore, the traffic splitting module includes: The segment contribution unit is used to calculate the contribution value of each segment unit to the process aggregation factor based on the process aggregation factor and segment combination data, and obtain the segment contribution data. The candidate filtering unit is used to identify segment units whose contribution values ​​reach a preset contribution threshold based on the segment contribution data, and to obtain candidate segment data. The element statistics unit is used to count the frequency and order changes of recorded elements in each candidate segment based on the candidate segment data, and to obtain candidate element data. The flow matrix unit is used to row and columnize the recorded elements in each candidate segment based on the candidate element data, and generate a matrix structure with the recorded elements as rows and the segment numbers as columns to obtain the segment flow matrix.

[0011] Furthermore, the traffic splitting module also includes: Based on the segment flow matrix and candidate element data, the degree of balance in the flow distribution of recorded elements within each candidate segment is calculated to obtain the balance allocation term: Based on the segment flow matrix and candidate segment data, the diffusion degree of recorded elements among multiple candidate segments is calculated to obtain the cross-segment diffusion term. Based on the segment flow matrix and segment contribution data, the degree of flow concentration in high contribution segments is calculated to obtain the flow concentration term; the degree of redundant loading in candidate segments is calculated to obtain the redundancy correction term. By integrating the traffic concentration term, balanced allocation term, cross-segment diffusion term, and redundancy correction term, the adaptability of each candidate segment to undertake the diversion task in the business process is calculated, and the process diversion factor is obtained.

[0012] Furthermore, the scheduling module includes: The interval division unit is used to divide the process aggregation factor and process diversion factor into intervals based on their numerical ranges, thereby obtaining combined interval data. The interval mapping unit is used to map different combined intervals to expert processing paths based on the combined interval data. It constructs a mapping structure with the combined interval as the key and the expert processing path as the value to obtain interval path mapping data. The path selection unit is used to match the expert processing path corresponding to the numerical combination of the process aggregation factor and the process diversion factor based on the interval path mapping data, so as to obtain the preliminary path data. The path construction unit is used to arrange multiple processing segments in the expert processing path according to the business sequence based on the preliminary path data, and generate scheduling path data.

[0013] Furthermore, the link module includes: The segment feature unit is used to identify the business characteristics of each processing segment in the scheduling path based on the scheduling path data, and to parse the execution object, processing depth and processing order of each processing segment to obtain segment feature data. The node weight unit is used to calculate the impact value of each processing segment on the inspection business process based on the segment feature data, and to map the impact value to the node weight value to obtain the node weight data. The node construction unit is used to define the processing segment as a process node based on the node weight data, and to combine the process nodes into a node sequence according to the execution order to obtain the process node data. The link generation unit is used to analyze the difference in node weight values ​​between process nodes based on process node data, map the node sequence into a chain connection structure, and obtain the inspection business link.

[0014] Furthermore, the node weight unit includes: The object parsing unit is used to structure the scope of the execution object in the inspection business process based on the segment feature data to obtain the object's scope of action data. The depth evaluation unit is used to map the processing depth of each processing segment to a depth score based on the segment feature data, thereby obtaining processing depth data. The sequence influence unit is used to identify the order of each processing segment in the scheduling path based on the segment feature data, and convert the order into a sequence influence score to obtain sequence influence data. The impact calculation unit is used to calculate the impact value of each processing segment on the inspection business process based on the object action data, processing depth data, and sequential impact data, and obtain the impact value data. The weight mapping unit is used to divide the influence value into several weight levels based on the influence value data, and to assign corresponding node weight values ​​to each weight level to obtain node weight data.

[0015] The above-described solution of the present invention has at least the following beneficial effects: This invention converts combined segment data into a trajectory sequence and generates a trajectory matrix. The occurrence of each type of record element in different segments is mapped to a unified row index, and the connection position is mapped to a unified column index, forming a directly comparable two-dimensional coordinate system. This coordinate system simultaneously reveals the continuity relationship of the same element in adjacent road segments and the positional relationship of the same element within the same road segment, facilitating unified statistics on sequential integrity, cross-segment coherence, duplicate entry tendency, and short-term spikes. For false detections caused by road shadows, shadow-related elements exhibit a pattern of concentrated short column segments and discontinuous cross-row and cross-segment patterns on the matrix, while actual road surface defects form longer continuous column segments in adjacent segments. Therefore, the trajectory matrix provides calculable structured differences before entering the aggregation metric, enabling subsequent statistics to perform consistency and coherence judgments based on row and column positions, limiting the formation of process instances from single-frame results without contextual support.

[0016] This invention uses node load data constructed through a trajectory matrix to compress the occurrence density of elements within a window, the stability across adjacent segments, and the frequency of transitions between segments into a comparable aggregation factor. This factor serves as an admission signal for entering the main process, directly limiting the statistical support for forming the main business line by combining segment data. When a record appears in a single segment but lacks continuity in preceding and following segments, the factor outputs a low value, and the data is constrained to lightweight or bypass paths in subsequent routing. When a record maintains sequential continuity across multiple adjacent segments, the factor outputs a high value, and the data obtains continuous support from the main processing segment. It translates the differences in carrying stability between scene disturbances in road inspection, such as sudden changes in lighting, occlusion, and attitude jitter, and actual defects into scalar thresholds, ensuring that the inputs for event registration and task construction meet the cross-segment consistency conditions, and providing a controlled dataset for subsequent candidate segment selection and diversion measurement.

[0017] This invention constructs a candidate set based on segment contribution within the dataset screened by aggregation factors and generates a segment flow matrix. This makes the frequency and position distribution of elements in each candidate segment visible in matrix form. The resulting flow diversion factor quantifies the flow balance, cross-segment diffusion, and high contribution concentration of the candidate set, which is used to determine the subsequent processing arrangements at the segment level. For road inspection, candidate segments with concentrated flow and low diffusion are more suitable for deep identification or verification, while candidate segments with high diffusion are more suitable for merging or delaying processing. The flow diversion factor incorporates this objective structure into the routing criteria, completing the horizontal allocation restriction at the segment level before event generation and task dispatch. This reduces the repetition of entries for the same event in multiple segments due to local peaks, and also reduces the competition of discrete segments for downstream models and channels. This ensures that subsequent path mapping is based on aggregated and diverted input construction, maintaining consistent boundaries between road objects and time windows.

[0018] This invention divides the process aggregation factor and process diversion factor into intervals by combining their numerical values, and maps a fixed expert processing path to each interval. It then generates scheduling path data, which transforms the carrying stability signal and the segment carrying structure signal into a defined sequence of processing segments. This ensures that homogeneous inputs achieve a consistent sequence of detection, verification, confirmation, and disposal, as well as a consistent resource selection scheme. For road inspection management systems, routing selection is transformed from manual rule overlay to factor-driven keyed retrieval, avoiding path differences caused by threshold drift over time. The backhaul from the device side, the verification from the platform side, and the reception from the disposal side are connected on the same sequence, ensuring stable channel selection and order for data from the same source, reducing repeated triggering caused by mid-process route changes, and providing a directly executable node sequence sketch for link generation.

[0019] This invention determines process node data and generates inspection business links by scheduling path data. Each processing segment obtains clear input and output parameter definitions and execution order. Node weights are used to reflect the contribution value of the segment to the overall inspection process. The link connects these nodes into a single path according to their weights and order. This structure ensures that the same event occupies only one main path within the system. Low-weight branches are restricted to edge positions or merged into the main chain node, preventing data from the same source from forming disposal entries on multiple paths simultaneously. Input and output parameter constraints between nodes ensure that the output of any node can obtain a matching position in the downstream node, preventing entries without a source from entering the disposal segment. The link-based representation also provides a basis for deduplication detection: when an upstream node has already output data for a certain section combination, the link refuses to construct the same instance again based on the node sequence, maintaining single-channel carrying of data for the same window of the same road segment. Attached Figure Description

[0020] Figure 1This is a flowchart of an intelligent inspection management system with a hybrid expert mode under a cloud-edge collaborative architecture, provided by an embodiment of the present invention. Detailed Implementation

[0021] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0022] like Figure 1 As shown, an embodiment of the present invention proposes an intelligent inspection management system with a hybrid expert mode under a cloud-edge collaborative architecture. The system includes: The segment module is used to acquire basic inspection data and divide the basic inspection data into multiple segment units according to the business time window. The record elements and node order within each segment unit are combined to obtain segment combined data. The trajectory module is used to construct a trajectory concatenation sequence based on the occurrence order of each recorded element according to the segment combination data, and to convert the trajectory concatenation sequence into trajectory matrix data. The aggregation module is used to construct node load data based on trajectory matrix data and calculate process aggregation factors based on node load data. The flow splitting module is used to filter out candidate segments from the segment combination data based on the process aggregation factor, construct a segment flow matrix based on the candidate segments, and calculate the process flow splitting factor. The scheduling module is used to select the corresponding expert processing path based on the combination range of process aggregation factor and process diversion factor, and obtain scheduling path data. The link module is used to determine process node data based on scheduling path data and to construct inspection business links based on process node data.

[0023] In this embodiment of the invention, the segment module is used to acquire basic inspection data and divide the basic inspection data into multiple segment units according to the business time window. The recorded elements and node order within each segment unit are combined to obtain segment combination data, which provides a unified time window scale for subsequent trajectory construction, statistical calculation, segment matching and process construction. At the same time, it limits the boundaries of data from the same source in the process, and prevents misalignment or duplicate instantiation across time windows. The trajectory module is used to construct a trajectory concatenation sequence based on the appearance order of each recorded element according to the segment combination data, and convert the trajectory concatenation sequence into trajectory matrix data. This enables the system to perform cross-segment association, sequence integrity analysis, recurrence statistics and jump identification in subsequent calculations, providing a directly operable, countable and comparable original basis for node load calculation and process factor calculation. The aggregation module is used to construct node load data based on the trajectory matrix data, and calculate process aggregation factors based on the node load data. It uniformly converts information such as the distribution density, cross-segment continuity and change trend of elements into measurable node loads and process aggregation factors, enabling the system to identify elements with stable repetitive patterns in continuous inspection segments.

[0024] The flow splitting module is used to filter candidate segments from the combined segment data based on the process aggregation factor, construct a segment flow matrix based on the candidate segments, calculate the process splitting factor, and quantify the segment carrying structure as the process splitting factor. This enables the system to identify which segments are more suitable for subsequent processing and which segments need to be merged or bypassed. The scheduling module is used to select the corresponding expert processing path based on the combined range of the process aggregation factor and the process splitting factor, and obtain scheduling path data. This ensures that data with the same structural characteristics follow a consistent processing segment order in subsequent processes, making the process predictable and with a clear execution chain. The link module is used to determine the process node data based on the scheduling path data and construct the inspection business link based on the process node data. This ensures the uniqueness of the path for the same inspection instance in the process, so that all data processing nodes are executed in a fixed order, without duplicate triggering, cross-triggering, or disordered flow.

[0025] The basic inspection data specifically includes: Basic inspection data is continuously collected and uploaded by edge devices during road inspection operations, and constitutes the most original input information in the entire cloud-edge collaborative processing. This basic inspection data includes various types of recorded content used to describe the road environment, collection status, identification trigger conditions, and device behavior. Specifically, basic inspection data includes at least: road scene image data collected in real time by vehicle-mounted inspection terminals or fixed road monitoring equipment, which can be color image sequences, infrared image sequences, or multispectral image sequences; depth point cloud or depth map data generated by depth sensors or structured light devices to reflect the spatial height changes of the road surface; preliminary identification markers generated by edge recognition models on the collection side, such as crack indication markers, foreign object type markers, obstacle boundary box markers, or detection confidence information of suspicious targets; timestamp data to characterize the time of record generation and location information to characterize the location of the inspection equipment, including latitude and longitude coordinates, road mileage points, vehicle attitude information, or road segment number of the collection location; and collection status markers reflecting the operating status of the inspection equipment, including frame validity markers, brightness anomaly markers, image jitter markers, occlusion markers, device cache status, trigger type, and other information. The basic inspection data also includes trigger types to indicate the reason for record generation, such as triggering due to changes in lighting, road deformation, suspected target in the model, or human intervention, which facilitates subsequent modules in determining the semantic location of the recorded elements within the segment.

[0026] The expert processing path specifically includes: Expert processing paths represent the sequence of processing segments that a process should employ under different combinations of input factors. This path is not a human decision but is predefined by the system and automatically matched during operation. Expert processing paths include multiple processing segments for analyzing, verifying, confirming, and generating results from inspection data. Each processing segment represents a processing node with a clearly defined operation object and processing rules. In actual implementation, an expert processing path includes at least: a pre-analysis segment for generating the processing starting point based on data input from the scheduling module; this segment is responsible for extracting key fields from the segment combination data, reconstructing the context window, and providing standardized input for subsequent nodes; a sequence verification segment for secondary analysis based on the trajectory matrix or segment structure; this verification segment verifies whether the order of occurrence of record elements in multiple segments remains consistent, and is used to construct a continuous element chain; a segment evaluation segment for verifying candidate segments and segment flow matrices; this evaluation segment determines the business segment to which the current inspection item should belong and selects an appropriate processing template according to the segment carrying structure; and a processing step based on input characteristics. The multi-level processing segment of the alienation analysis includes a parallel processing segment for cross-comparison of multi-source elements, a temporal reconstruction segment for sequence completion and missing element judgment, and an element consistency segment for judging the validity of recorded elements; a result aggregation segment for generating the final inspection structure result, which organizes the outputs of all preceding processing segments into a unified structure according to the order agreed upon in the scheduling path, and uses it as the direct input for link construction; and an output evaluation segment for providing node weight reference for the link module, which numbers and organizes the outputs of each processing segment according to the execution object, input source, processing depth, and relative order of the processing segment, so that the link module can form process nodes with progressive relationships in the path structure.

[0027] In a preferred embodiment of the present invention, the segment module includes: The data processing unit is used to arrange the basic inspection data in chronological order according to the time stamps in the basic inspection data to obtain time series data. The segment generation unit is used to segment time series data according to the range of the business time window, generate multiple segment units, and obtain segment unit data; The structural decomposition unit is used to break down the record content of each segment unit into record elements and node sequences based on the segment unit data, thereby obtaining segment element node data. The combined construction unit is used to arrange and combine the record elements and node sequences into a continuous structure according to the segment unit numbering order based on the segment element node data, so as to obtain the segment combined data.

[0028] In this embodiment of the invention, the data processing unit is used to arrange the basic inspection data in chronological order according to the timestamps in the basic inspection data to obtain time series data, ensuring that subsequent segmentation can accurately locate data boundaries based on continuous time windows, avoiding segment cross-boundary or loss due to time differences between different devices; the segment generation unit is used to segment the time series data according to the range of the business time window to generate multiple segment units, obtaining segment unit data, which serves as the boundary condition for subsequent structural decomposition, giving the extraction of record elements and node sequences clear range constraints; the structural decomposition unit is used to split the record content of each segment unit into record elements and node sequences according to the segment unit data, obtaining segment element node data, ensuring that business information and process information can participate in different measurements respectively in the future; the combination construction unit is used to arrange and combine the record elements and node sequences according to the segment unit number order according to the segment element node data into a continuous structure, obtaining segment combination data, providing the lowest-level sequence foundation for all subsequent calculations.

[0029] The segment generation unit is used to segment time series data according to the range of the business time window, generating multiple segment units to obtain segment unit data, specifically including: First, the timestamp and associated road identification information (such as road number, lane number, and inspection equipment number) of each record are read from the time-series data. Records with the same road number and equipment number are grouped into the same set of segments to avoid records from different inspection tasks being grouped into the same segment. Then, the record with the smallest timestamp is selected from each set of segments, and this timestamp is used as the start time of the current business time window. The start time is added to a preset window duration to obtain the end time of the window, forming the time range of the business time window. Within the window, records in the time-series data are scanned sequentially. When a record's timestamp is greater than or equal to the window start time and less than the window end time, the record is assigned to the current segment unit, and a unique segment number is assigned to the current segment unit. When a record's timestamp is found to be greater than or equal to the window end time, the end time of the current segment unit is recorded as the window end time or the previous timestamp of the current record. The current segment unit is closed, and a new business time window start time is calculated using the current record's timestamp, initiating a new segment unit generation process. For situations with large time intervals during the inspection process, the time difference between adjacent records is determined during window sliding. When the time difference exceeds a preset interval threshold, even if the records are still within the same window range, this time difference is used as a new segmentation point, ending the current segment and starting a new one to ensure time continuity within the segment. After all records have been traversed, the end time of the last unclosed segment unit is recorded, and the start and end times of each segment unit are assembled into segment unit data.

[0030] The structural decomposition unit is used to break down the record content of each segment unit into record elements and node sequences based on the segment unit data, thereby obtaining segment element node data, specifically including: Using the section number as the outer index, each record within a section unit is parsed and processed one by one. First, according to the system's preset field mapping rules, fields directly related to the inspection business content are extracted from each record, such as target category, target location, detection confidence level, road ancillary facility status, environmental conditions, and equipment operating status. These fields are combined into record elements, and each record element is assigned an element identifier for reference and statistics within the same section and across sections. Simultaneously, fields related to the processing flow are extracted from the records, such as processing node identifiers, processing stage markers, processing result types, upstream node source numbers, and downstream node pointer numbers. These flow-related fields are converted into node markers in a node sequence according to a preset node type mapping table, so that each record corresponds to one or more ordered process nodes within the section. During the parsing process, the chronological order of records within the section unit is maintained. Record elements and node markers are paired and mapped one-to-one according to chronological order. For cases where one record corresponds to multiple process nodes, the same record element is stored in the node sequence as a sequential chain, indicating the node sequence relationship and the reference relationship between nodes. After parsing, all record elements and node sequences under the same segment number are organized into segment element node data according to the time order within the segment.

[0031] In a preferred embodiment of the present invention, the trajectory module includes: The sequential extraction unit is used to extract the sequential position corresponding to each record element according to the arrangement order of record elements in the segment combination data, so as to obtain the record sequence data; The index building unit is used to determine the numbering based on the sequential position of the record elements in the record sequence data, and to generate a feature index for each record element to obtain sequential index data; The trajectory concatenation unit is used to connect the feature indexes of the same record feature according to the arrangement order in the segment combination data based on the sequential index data, to construct a continuous index sequence and obtain the trajectory concatenation sequence. The matrix generation unit is used to generate trajectory matrix data by taking the index sequence number as the row, the connection order of each element in each index sequence as the column, and filling each feature index into the corresponding row and column positions according to the trajectory concatenation sequence.

[0032] In this embodiment of the invention, the sequence extraction unit is used to extract the sequential position corresponding to each record element according to the arrangement order of the record elements in the segment combination data, thereby obtaining record sequence data. This ensures that the relative order between different record elements within the same segment remains consistent in subsequent calculation stages, providing a data foundation for constructing element trajectories and determining the execution order of process nodes. The index construction unit is used to determine the number based on the sequential position of the record elements in the record sequence data, and generate an element index for each record element, thereby obtaining sequence index data. This achieves unified identification of similar elements, enabling subsequent trajectory concatenation and matrix generation stages to directly perform merging, grouping, and statistical operations based on the numerical index during processing, avoiding issues caused by different representations of record content. The system addresses inconsistencies in matching. A trajectory concatenation unit connects the element indices of the same record element according to the order of arrangement in the segment combination data, constructing a continuous index sequence to obtain a trajectory concatenation sequence. This facilitates subsequent identification of behavioral patterns such as continuous occurrence, intermittent occurrence, or cross-segment jumps at the sequence dimension, providing basic sequence data for subsequent steps in the system. A matrix generation unit, based on the trajectory concatenation sequence, uses the sequence number of each index sequence as rows and the connection order of each element in each index sequence as columns, filling each element index into the corresponding row and column positions to obtain trajectory matrix data. This ensures that the output of the trajectory module maintains a traceable relationship with the preceding segment combination data and sequential index data, providing a stable trajectory expression foundation for subsequent analysis stages of the business process.

[0033] The trajectory concatenation unit is used to connect the feature indices of the same record element according to the arrangement order in the segment combination data based on the sequential index data, to construct a continuous index sequence and obtain the trajectory concatenation sequence, specifically including: First, based on the feature index field contained in the sequential index data, all records are categorized. All feature indexes are extracted from the sequential index data, and a grouped cache structure with the feature index as the key is established. Within this grouped cache structure, independent sequence storage space is reserved for each feature index. Then, each record in the sequential index data is traversed, and for each record, the feature index, segment number, and sequence position fields are read. The record is added to the corresponding grouped storage space according to its feature index, and the segment number, sequence position, and marker information related to the arrangement order in the segment combination data are temporarily stored in the group. Further, within each group corresponding to a feature index, the records in the group are sorted according to the arrangement order in the segment combination data. The segment combination data output by the segment module is called, and the segment number and sequence position of each record in the group are compared with the segment order in the segment combination data. The order in which the segments appear in the segment combination data is used as the first sorting key, and the sequence position within the same segment unit is used as the second sorting key. All records in the group are sorted to obtain an ordered record sequence that conforms to the business time progression logic. After sorting, the feature indexes of each record are read sequentially according to the sorting result. These feature indexes are then connected in order to construct a continuous index sequence. For cases where the same feature index appears consecutively in adjacent segments, continuous connection segments are formed in the continuous index sequence. For cases with missing segments or jumps across segments, connection patterns with interval markers are formed in the continuous index sequence. Finally, the continuous index sequence corresponding to each feature index is output as a trajectory concatenation sequence.

[0034] The matrix generation unit, based on the trajectory concatenation sequence, uses the index sequence number as the row, the connection order of elements in each index sequence as the column, and fills each feature index into the corresponding row and column positions to obtain trajectory matrix data. Specifically, it includes: First, the number and length of each trajectory concatenation sequence are statistically analyzed to determine the matrix row and column configuration strategy. The sequence number or corresponding feature index number of each trajectory concatenation sequence is used as the row identifier, and the connection order of elements in the sequence is used as the column identifier. Each trajectory concatenation sequence is assigned a row, and column space matching the sequence length is reserved for that row. The trajectory concatenation sequence dataset is traversed in a preset order. For the i-th trajectory concatenation sequence, the row index is set to i or the associated feature index number. Starting from the first element, column indices are assigned to each element sequentially, forming the connection order identifier from the first column to the nth column. For each trajectory element, the corresponding feature index value is read and filled into the matrix cell position corresponding to the current row and column. For column positions that do not exist in a trajectory, a null value or a preset zero value is filled in the corresponding position to distinguish between the missing connection position and the actual feature index. After filling all the rows and columns of the trajectory concatenation sequences, the trajectory matrix data is formed.

[0035] In a preferred embodiment of the present invention, the aggregation module includes: The row-level calculation unit is used to calculate row-level statistical values ​​based on the number of feature indexes in each row of the trajectory matrix data, and obtain row-level density data. The segment analysis unit is used to identify the changes in the occurrence of each row in different segment units based on the trajectory matrix data and segment combination data, and to obtain segment change data. The jump recognition unit is used to count the number of cross-segment jumps between segment units in each row based on trajectory matrix data and segment combination data, and to obtain cross-segment jump data. The load construction unit is used to combine row-level density data, segment change data, and cross-segment jump data into node load data.

[0036] In this embodiment of the invention, a row-level calculation unit is used to calculate row-level statistical values ​​based on the number of element indexes in each row according to the trajectory matrix data, to obtain row-level density data, reflecting the cumulative participation degree of elements in multiple segment units and multiple business nodes; a segment analysis unit is used to identify the changes in the occurrence of each row in different segment units according to the trajectory matrix data and segment combination data, to obtain segment change data, characterizing the spatial distribution and business distribution of elements on the inspection route; a jump identification unit is used to count the number of cross-segment jumps of each row between segment units according to the trajectory matrix data and segment combination data, to obtain cross-segment jump data, which facilitates the subsequent suppression of elements with high jumps in node load data; and a load construction unit is used to combine row-level density data, segment change data, and cross-segment jump data into node load data, which simultaneously reflects the participation frequency of elements in the entire inspection process, their concentration or dispersion on specific road segments, and their continuity or jump between multiple road segments.

[0037] The row-level calculation unit is used to calculate row-level statistical values ​​based on the number of feature indices in each row of the trajectory matrix data, thereby obtaining row-level density data. Specifically, it includes: First, each row of the trajectory matrix is ​​associated with its corresponding record element identifier, forming a mapping relationship between record element identifiers and matrix rows. Under this mapping relationship, the trajectory matrix data is traversed row by row. When traversing a row, the matrix elements at each column position in that row are read sequentially. According to the preset validity judgment rules, matrix elements that are non-empty, non-zero, and represent valid element indices are marked as valid index positions, while empty, zero, or placeholder values ​​are marked as invalid index positions. After completing a full column scan of a row, the number of all valid index positions in that row is counted, and this number is used as the basic occurrence count of the record element in the entire trajectory matrix. When necessary, preset weights are assigned to the valid indexes at different column positions according to the business node category, time series order, or node type to which the column index belongs, and the weighted occurrence counts are accumulated during the statistics to obtain row-level statistical values ​​that take into account the occurrence counts and differences in business positions. Subsequently, the basic occurrence counts and optional weighted statistical values ​​corresponding to each row are used to form row-level density data.

[0038] The segment analysis unit is used to identify the changes in the occurrence of each row in different segment units based on the trajectory matrix data and segment combination data, thereby obtaining segment change data, specifically including: First, based on the correspondence between segment units, node sequences, and connection positions recorded in the segment combination data, a mapping table is established for each column index in the trajectory matrix, ensuring that each column can determine its corresponding segment number during analysis. Then, taking each row of the trajectory matrix as a unit, the matrix row corresponding to a specific record element is traversed. During traversal, the mapping table from column index to segment unit is looked up, and each valid element index in that row is assigned to its corresponding segment unit for counting and statistics, obtaining the distribution of the record element's occurrence frequency in each segment unit. After scanning all columns of a single row, the occurrence frequency of each segment is sorted according to the time or business order of the segment units, and the difference in occurrence frequency, direction of increase / decrease, and relative change ratio between adjacent segments are calculated to reflect the concentration of the record element in the segment dimension and its changing trend between preceding and subsequent segments. After analyzing all rows corresponding to all record elements, the occurrence frequency distribution of each row in each segment unit and the changes between segments are summarized and organized into segment change data.

[0039] The jump recognition unit is used to count the number of cross-segment jumps between segment units in each row based on the trajectory matrix data and segment combination data, thereby obtaining cross-segment jump data, which specifically includes: First, the mapping table between the column index and the segment unit, which is the same as that of the segment analysis unit, is invoked to ensure that the segment number corresponding to any column in the trajectory matrix remains consistent throughout the analysis. Then, taking each row of the trajectory matrix as a unit, the matrix rows corresponding to a certain record element are linearly traversed according to the column index order. During the traversal, only the column positions of valid element indexes are processed; columns corresponding to null values, zero values, or placeholders are skipped. When processing the first valid element index, its segment number is recorded as the current segment number. As subsequent valid indexes are traversed, each new valid index is read, and its segment number is retrieved based on the column index. This segment number is then compared with the segment number recorded in the previous valid index. When two adjacent valid indexes belong to the same segment, the segment number is considered the segment number. When the segment numbers are different and the corresponding segment units are not continuous, the change in segment number is recorded as a cross-segment jump event, and the number of cross-segment jumps is accumulated. When necessary, different types of cross-segment jumps are classified and recorded according to the distance between segment numbers, the number of segments crossed, or the time interval between crossings, and short-distance cross-segments and long-distance cross-segments are counted separately to form more detailed jump statistics. After completing the traversal of all valid indexes in the row, the number of cross-segment jumps, the distribution of cross-segment jump types, and the optional jump interval information corresponding to the record element are uniformly organized into a cross-segment jump record. After performing the above operations on all rows in the trajectory matrix, the cross-segment jump records corresponding to all record elements are indexed according to the record element identifier to form cross-segment jump data.

[0040] In a preferred embodiment of the present invention, the aggregation module further includes: Based on the node load data, the load concentration degree of each row in the trajectory matrix is ​​calculated to obtain the density score; the stability of the load between each segment is calculated to obtain the change suppression score; based on the density score and the change suppression score, nodes with concentrated and stable loads are identified to obtain the aggregation item. Based on the node load data, the jump frequency of the node between each segment unit is calculated to obtain the jump score; the density score is adjusted by power to obtain the density reinforcement score; based on the density reinforcement score and the jump score, the cross-segment continuity is calculated to obtain the balance term. Based on the node load data, the load fluctuation intensity of each row between each segment unit is calculated to obtain the change intensity score; the degree of discontinuity across segments of each row is calculated to obtain the jump intensity score; based on the change intensity score and the jump intensity score, nodes with high fluctuation and high jump are identified to obtain the suppression term. By integrating the aggregation term, balancing term, and suppression term, the aggregation stability of the corresponding nodes in each row in the entire business process is calculated, and the process aggregation factor is obtained.

[0041] In this embodiment of the invention, based on node load data, the load concentration degree of each row in the trajectory matrix is ​​calculated to obtain a density score, quantifying the cumulative participation degree of nodes in different sections of the inspection operation; the stability degree of load between sections is calculated to obtain a change suppression score, and the change trajectory of load in different sections is converted into a change suppression score; based on the density score and the change suppression score, nodes with concentrated and stable loads are identified to obtain aggregation items, reflecting both density and stability dimensions, avoiding the misclassification of nodes that only appear frequently but fluctuate highly as backbone nodes, and also avoiding the misclassification of nodes that only appear stably in a few sections as mainline nodes; based on node load data, the jump frequency of nodes between section units is calculated to obtain a jump score, and by statistically analyzing the number of times load states are switched between present and absent, the state dispersion of nodes between different sections is converted into a jump score; the density score is adjusted exponentially to obtain a density enhancement score, quantifying the load concentration of nodes in the section dimension; based on the density enhancement score and the jump score, the cross-section continuity is calculated. The system obtains a balancing term, enabling it to clearly mark cross-segment connection nodes at the data level, preserving structural information about cross-segment continuity when calculating the aggregation factor in the subsequent process. Based on node load data, the system calculates the load fluctuation intensity of each row across segment units, obtaining a change intensity score. This provides an independent quantitative dimension regarding load amplitude changes, allowing for the subsequent inclusion of nodes with high fluctuations and cross-segment discontinuities into the suppression term. The system calculates the degree of cross-segment discontinuity for each row, obtaining a jump intensity score, distinguishing nodes that appear consecutively in multiple adjacent segments from nodes that only occasionally appear in mutually isolated segments. Based on the change intensity score and jump intensity score, the system identifies nodes with high fluctuations and high jumps, obtaining a suppression term that clearly distinguishes nodes that exhibit both large load fluctuations and strong cross-segment discontinuities across different segments. The system integrates the aggregation term, balancing term, and suppression term to calculate the aggregation stability of the corresponding nodes in each row throughout the entire business process, obtaining the process aggregation factor. This factor simultaneously reflects the load concentration, cross-segment stability, cross-segment continuity, and high fluctuations and high discontinuities of the nodes.

[0042] In a preferred embodiment of the present invention, the traffic splitting module includes: The segment contribution unit is used to calculate the contribution value of each segment unit to the process aggregation factor based on the process aggregation factor and segment combination data, and obtain the segment contribution data. The candidate filtering unit is used to identify segment units whose contribution values ​​reach a preset contribution threshold based on the segment contribution data, and to obtain candidate segment data. The element statistics unit is used to count the frequency and order changes of recorded elements in each candidate segment based on the candidate segment data, and to obtain candidate element data. The flow matrix unit is used to row and columnize the recorded elements in each candidate segment based on the candidate element data, and generate a matrix structure with the recorded elements as rows and the segment numbers as columns to obtain the segment flow matrix.

[0043] In this embodiment of the invention, the segment contribution unit is used to calculate the contribution value of each segment unit to the process aggregation factor based on the process aggregation factor and segment combination data, thereby obtaining segment contribution data. This allows the participation degree of each segment unit in the global inspection process to be explicitly expressed in the form of segment contribution values. The candidate screening unit is used to identify segment units whose contribution values ​​reach a preset contribution threshold based on the segment contribution data, thereby obtaining candidate segment data. Non-candidate segments are excluded from the diversion path construction process at this stage to avoid computational interference to subsequent construction stages. The element statistics unit is used to count the occurrence frequency and order changes of recorded elements in each candidate segment based on the candidate segment data, thereby obtaining candidate element data and separating the recorded element category from the segment number to provide basic data for subsequent processing. The flow matrix unit is used to row and columnize the recorded elements in each candidate segment based on the candidate element data, generating a matrix structure with recorded elements as rows and segment numbers as columns to obtain a segment flow matrix. This constructs the flow distribution pattern of the candidate segment set at the data level, providing a data foundation for calculating the adaptability of candidate segments in the diversion task.

[0044] The segment contribution unit is used to calculate the contribution value of each segment unit to the process aggregation factor based on the process aggregation factor and segment combination data, thus obtaining the segment contribution data, which specifically includes: First, the identifier information, aggregation value, and row index in the trajectory matrix corresponding to each record element in the process aggregation factor are stored in a preset index table. Simultaneously, the segment number, business time window, and sequence of record elements arranged in node order within each segment are read from the segment combination data, and a segment access table is built in memory using the segment number as the key. Based on this, a traversal process is initiated for each segment unit: according to the sequence of record elements recorded in the segment access table, the record element identifier is read one by one starting from the beginning of the sequence, and the corresponding aggregation value is queried in the index table of the process aggregation factor using the record element identifier. After finding the aggregation value, according to the preset contribution calculation rules, the aggregation value is combined with the number of times the record element appears in the current segment, its relative order position, or node type to obtain the local contribution value of the record element in the current segment, and this local contribution value is temporarily stored in an intermediate cache indexed by the segment number and the record element identifier. When the same record element appears multiple times in the same segment, the local contribution value of the record element is accumulated, so that the total contribution of the same record element in the segment is reflected through a single entry. After traversing all record elements within a certain segment and accumulating local contributions, the set of local contribution values ​​of all record elements within that segment is read from the intermediate cache. An aggregation operation is performed on this set to obtain the segment contribution value as a whole. The aggregation operation can perform weighted summation, normalization, or other linear combinations according to preset rules, so that the segment contribution value numerically reflects the comprehensive influence of the record elements within that segment on the overall process aggregation factor, thus forming segment contribution data.

[0045] The element statistics unit is used to count the frequency and order changes of recorded elements within each candidate segment based on the candidate segment data, thereby obtaining candidate element data, specifically including: First, a candidate segment list is read from the candidate segment data. The candidate segment list includes at least the candidate segment number, the corresponding business time window, and the road geographical location information. Using the candidate segment number as the key, a candidate segment index table is built in memory. Based on this index table, the record element sequence and node order information corresponding to each candidate segment are extracted from the segment combination data in sequence, so that the record elements in the candidate segment are presented in the form of a continuous sequence. For a candidate segment, a segment-local statistics table is first initialized to record the occurrence of various record elements within the candidate segment. Then, the record element sequence is read from the segment combination data in node order. For each record element in the sequence, the segment-local statistics table is queried to see if a statistical entry for the same record element already exists. If not, a statistical record with the record element identifier as the key is created in the segment-local statistics table, the occurrence count is initialized to 1, and the current sequential position is recorded as the first occurrence position. At the same time, the sequential position is written to the sequential position list. If a statistical entry for the same record element already exists, the occurrence count of the record element is incremented by 1, and the current sequential position is appended to the sequential position list to reflect the change in the order when the record element occurs multiple times within the segment.

[0046] After the sequence of record elements within a single candidate segment is traversed, the statistical entries of all record elements are read from the segment's local statistics table. The occurrence count, first appearance position, last appearance position, and complete sequence position list of each record element within the segment are associated with the candidate segment number and written to the global candidate element statistics cache. When processing multiple candidate segments, the records are merged in the global candidate element statistics cache based on the record element identifier. The statistical information of the same record element in different candidate segments is summarized into a set structure with the record element identifier as the primary key. The occurrence count and sequence change information under each candidate segment number are recorded in the set structure to generate candidate element data.

[0047] In a preferred embodiment of the present invention, the splitting module further includes: Based on the segment flow matrix and candidate element data, the degree of balance in the flow distribution of recorded elements within each candidate segment is calculated to obtain the balance allocation term: Based on the segment flow matrix and candidate segment data, the diffusion degree of recorded elements among multiple candidate segments is calculated to obtain the cross-segment diffusion term. Based on the segment flow matrix and segment contribution data, the degree of flow concentration in high contribution segments is calculated to obtain the flow concentration term; the degree of redundant loading in candidate segments is calculated to obtain the redundancy correction term. By integrating the traffic concentration term, balanced allocation term, cross-segment diffusion term, and redundancy correction term, the adaptability of each candidate segment to undertake the diversion task in the business process is calculated, and the process diversion factor is obtained.

[0048] In this embodiment of the invention, based on the segment flow matrix and candidate element data, the degree of balance in the flow distribution of recorded elements within each candidate segment is calculated to obtain a balanced allocation term. This expresses, through a unified numerical form, whether a single element holds an absolute flow advantage in the segment and whether the distribution of multiple elements in the segment is nearly uniform. Based on the segment flow matrix and candidate segment data, the degree of diffusion of recorded elements among multiple candidate segments is calculated to obtain a cross-segment diffusion term. This distinguishes candidate segments that are easily used as local feature bearer nodes from those suitable as shared general bearer nodes, reducing the likelihood of the same element triggering the same subsequent processing link simultaneously in multiple candidate segments. Based on the segment flow matrix and segment contribution data, the degree of flow concentration in high-contribution segments is calculated. This process yields a traffic concentration term, a quantitative indicator reflecting the position of the traffic carried by a segment within the overall business contribution framework, forming the basis for differentiated traffic diversion for segments of different business importance. It also calculates the degree of redundancy loading in candidate segments, obtaining a redundancy correction term, and identifying redundant traffic portions in each candidate segment caused by duplicate reporting, duplicate identification, or multi-channel aggregation. By integrating the traffic concentration term, balanced allocation term, cross-segment diffusion term, and redundancy correction term, the adaptability of each candidate segment to undertake diversion tasks in the business process is calculated, resulting in a process diversion factor. This forms a segment adaptability description system centered on process diversion, enabling the road inspection management system to drive the selection and execution of diversion strategies with unified scalar indicators while maintaining data integrity.

[0049] In a preferred embodiment of the present invention, the scheduling module includes: The interval division unit is used to divide the process aggregation factor and process diversion factor into intervals based on their numerical ranges, thereby obtaining combined interval data. The interval mapping unit is used to map different combined intervals to expert processing paths based on the combined interval data. It constructs a mapping structure with the combined interval as the key and the expert processing path as the value to obtain interval path mapping data. The path selection unit is used to match the expert processing path corresponding to the numerical combination of the process aggregation factor and the process diversion factor based on the interval path mapping data, so as to obtain the preliminary path data. The path construction unit is used to arrange multiple processing segments in the expert processing path according to the business sequence based on the preliminary path data, and generate scheduling path data.

[0050] In this embodiment of the invention, the interval partitioning unit is used to partition different combinations of process aggregation factors and process diversion factors into intervals based on their numerical ranges, obtaining combined interval data. This achieves structured segmentation of the numerical space of process aggregation factors and process diversion factors, with each interval unit having a clear boundary range and identifier on a two-dimensional numerical plane. The interval mapping unit is used to map different combined intervals to expert processing paths based on the combined interval data. It constructs a mapping structure using the combined intervals as keys and the expert processing paths as values, obtaining interval path mapping data. This eliminates the need to re-parse business rules or redesign paths for each instance during runtime, thus implementing routing strategies. The system features centralized definition and unified management. A path selection unit matches expert processing paths with numerical combinations of process aggregation and process diversion factors based on interval path mapping data, generating preliminary path data. It assigns a clear path template to each instance, preventing inconsistencies in path selection across different instances under the same factor model, ensuring determinism and consistency in path decisions. A path construction unit arranges multiple processing segments in the expert processing path according to business sequence based on the preliminary path data, generating scheduling path data. This eliminates the need to infer processing order or data dependencies again during the link generation stage, ensuring logical continuity and data flow consistency from factor-determined paths to path-driven link construction.

[0051] The interval division unit is used to divide different combinations of process aggregation factors and process diversion factors into intervals based on their numerical ranges, resulting in combined interval data. Specifically, it includes: First, the system configuration database is used to retrieve pre-defined numerical range parameters for road inspection operations. These parameters include at least the global numerical range of the process aggregation factor, the global numerical range of the process diversion factor, and the corresponding segmentation precision. Based on these global numerical ranges, the number axis of the process aggregation factor is divided into several non-overlapping one-dimensional aggregation sub-intervals at set intervals, and a unique interval number is assigned to each aggregation sub-interval. Similarly, the number axis of the process diversion factor is divided into several non-overlapping one-dimensional diversion sub-intervals at set intervals, and a unique interval number is assigned to each diversion sub-interval. Then, a Cartesian combination is performed on all aggregation sub-intervals and all diversion sub-intervals. Each pair of aggregation and diversion sub-interval combinations is considered a two-dimensional combination interval. A unique combination interval identifier is generated for each two-dimensional combination interval, and the boundary values ​​of the aggregation and diversion sub-intervals corresponding to this combination interval are recorded. When a specific inspection business instance generates a value for a process aggregation factor and a process diversion factor, the process aggregation factor is assigned to a specific aggregation sub-interval and the process diversion factor is assigned to a specific diversion sub-interval by comparing the factor values ​​of that instance with the upper and lower boundaries of each sub-interval. Based on this, the two-dimensional combined interval identifier to which the instance belongs is determined. All the identifiers of the two-dimensional combined intervals, their corresponding numerical boundaries, and the associated business instance tags are organized into structured data to form combined interval data.

[0052] The interval mapping unit is used to map different combined intervals to expert processing paths based on the combined interval data. It constructs a mapping structure using the combined interval as the key and the expert processing path as the value to obtain interval path mapping data, specifically including: During system initialization, multiple predefined expert processing paths are loaded from the expert path configuration library. Each expert processing path corresponds to a set of processing segment sequences designed for road inspection business scenarios. These processing segment sequences include processing segments for calling different recognition models, processing segments for result verification, and processing segments for generating task records. Combining the numerical boundary information of each combined interval in the combined interval data, and according to the processing requirements of road inspection business for different aggregation and diversion modes, at least one expert processing path is selected for each combined interval. An internal key-value mapping table is constructed, storing the combined interval identifier as the key and the corresponding expert processing path identifier or complete path description structure as the value. For combined intervals with multiple candidate paths, the priority order, applicable inspection task type, and road level-related constraints of each candidate path are recorded in the mapping table. After construction, the mapping table containing the correspondence between all combined intervals and expert processing paths is encapsulated as interval path mapping data.

[0053] In a preferred embodiment of the present invention, the link module includes: The segment feature unit is used to identify the business characteristics of each processing segment in the scheduling path based on the scheduling path data, and to parse the execution object, processing depth and processing order of each processing segment to obtain segment feature data. The node weight unit is used to calculate the impact value of each processing segment on the inspection business process based on the segment feature data, and to map the impact value to the node weight value to obtain the node weight data. The node construction unit is used to define the processing segment as a process node based on the node weight data, and to combine the process nodes into a node sequence according to the execution order to obtain the process node data. The link generation unit is used to analyze the difference in node weight values ​​between process nodes based on process node data, map the node sequence into a chain connection structure, and obtain the inspection business link.

[0054] In this embodiment of the invention, the segment feature unit is used to identify the business characteristics of each processing segment in the scheduling path based on the scheduling path data, analyze the execution object, processing depth, and processing order of each processing segment, and obtain segment feature data. This provides a clear input basis for calculating node impact values ​​and dividing node weights, enabling the scope and relative position of each processing segment in the inspection business process to be expressed and distinguished through data fields. The node weight unit is used to calculate the impact value of each processing segment on the inspection business process based on the segment feature data, and map the impact value to the node weight value to obtain node weight data. This establishes a quantifiable level system for the importance of processing segments, providing weight information for subsequent node construction and link generation. The node construction unit is used to... Based on node weight data, processing segments are defined as process nodes, and process nodes are combined into a node sequence according to the execution order to obtain process node data. The processing segment set has a node representation and adjacency description that can be directly used for link construction. The link generation unit is used to analyze the difference in node weight values ​​between process nodes based on the process node data, and map the node sequence into a chain connection structure to obtain the inspection business link. A chain connection structure that reflects both the process execution order and the importance level of nodes is constructed, providing a clear and parsable representation of the flow path, state transition position and dependency relationship between processing segments of abnormal events in road inspection business, supporting subsequent path-level monitoring, analysis and adjustment of the inspection process.

[0055] The segment feature unit is used to identify the business characteristics of each processing segment in the scheduling path based on the scheduling path data, and to parse the execution object, processing depth, and processing order of each processing segment to obtain segment feature data, specifically including: First, using the processing segment identifier recorded in the scheduling path data as an index, the corresponding processing type, applicable road segment scenario, and input / output data types are queried one by one in the preset processing segment configuration table. The road inspection object information associated with the processing segment, including original road image data, depth data, historical inspection record data, defect candidate result data, and alarm event data, is parsed into structured execution object fields, and these execution object fields are stored in a one-to-one correspondence with the processing segment identifier. Based on the algorithm category, computational complexity, and data dependency level recorded in the processing segment configuration table, the processing scope, processing granularity, and processing stage location of each processing segment are mapped to a processing depth field. This ensures that different types of processing segments, such as coarse-grained filtering, basic target detection, fine-grained defect localization, rule validation, task generation, and status updates, are processed on the same numerical dimension. A normalized representation is used to make the processing depth of each processing segment comparable. Further, according to the arrangement order of each processing segment in the scheduling path data, a sequence number is generated for each processing segment. This sequence number corresponds one-to-one with the relative position of the processing segment in the entire scheduling path. Simultaneously, a processing order field is generated by combining the sets of preceding and succeeding processing segments, encoding the sequential dependencies of processing segments in the business process into explicit sequence features. For scheduling path data containing scenario information, the segment feature unit also parses scenario features such as road segment type, elevated road marking, tunnel marking, and ramp marking from the scheduling path data and associates them with the processing segments, writing them into the segment feature data. This ensures that each processing segment simultaneously possesses an execution object field, a processing depth field, a processing order field, and a scenario feature field in the segment feature data. Finally, these fields are organized into segment feature data.

[0056] The node construction unit is used to define processing segments as process nodes based on node weight data, and to combine process nodes into a node sequence according to the execution order to obtain process node data, specifically including: First, using the processing segment identifier in the scheduling path data as the primary key, each processing segment is traversed sequentially. For each processing segment, the corresponding node weight value, influence value, and associated object action data, processing depth data, and order influence data are read from the node weight data. The processing segment identifier, the execution object field output by the segment feature unit, and the processing depth field are then encapsulated into a node description structure. This node description structure is defined as a process node, and its unique identifier, associated processing segment identifier, node weight value, node influence value, execution object information, and processing depth information are written into it. Based on the processing segment arrangement order recorded in the scheduling path data, a sequence number is appended to each process node, and this sequence number serves as the execution position information of the process node in the business process. The system collects information and sorts all process nodes according to their sequential numbers to form a sequential node sequence. During the sorting process, it identifies the preceding and succeeding relationships between adjacent nodes and writes the predecessor and successor node identifiers into the corresponding node description structure, ensuring that each process node has at least one predecessor node reference and at least one successor node reference. For process nodes whose node weight value is below a preset threshold and whose execution object information is the same as the upstream node, a merging flag field and a mergeable group number are recorded in the node description structure. This flag is used by the link generation unit to refer to when constructing the inspection business link for node merging or composite node construction. Finally, all process nodes arranged in sequential numbers, their preceding and succeeding relationships, node weight values, and merging flags are uniformly organized into process node data.

[0057] The link generation unit is used to analyze the difference in node weight values ​​between process nodes based on process node data, map the node sequence into a chain connection structure, and obtain the inspection business link, specifically including: First, following the sequential numbering in the node sequence, traverse each pair of adjacent process nodes. For each pair, read its node weight value, calculate the weight difference between the two nodes, and simultaneously calculate the difference in their influence values. Use the node weight difference and influence value difference as the initial data for the edge attributes of the connection relationship between the two nodes. Then, judge the edge attributes based on a preset weight difference threshold. When the weight difference between adjacent process nodes is less than or equal to the first threshold and the execution object information of the two nodes is the same or there is an inclusion relationship, establish a directed edge of continuous processing type for the two process nodes in the connection structure. Mark this directed edge as a continuous edge at the same level, indicating that the two nodes belong to the same processing level of continuous execution relationship in the inspection business link. When the weight difference between adjacent process nodes is greater than the second threshold, establish a directed edge of hierarchical jump type for the two process nodes in the connection structure. Mark this directed edge as a cross-level jump edge, indicating that the business process at this position... The process transitions from the upstream identification level to the downstream task generation or processing record level. For process nodes marked as candidates for merging nodes and sequentially adjacent, multiple merging nodes are merged into a composite node in the view layer of the link structure based on the merging mark field and node weight difference when constructing the connection structure. The atomic node list and the internal connection relationship between atomic nodes are maintained inside the composite node. At the same time, only the composite node is retained as a single point in the link at the outer layer of the link to simplify the external structure of the inspection business link. After completing the generation of directed edges of all adjacent nodes and the construction of the composite node, starting from the starting node with the smallest sequential number, the process traverses along the continuous edges of the same layer and the cross-layer transition edges, and connects all reachable nodes into a chain connection structure with a clear beginning and end according to the connection relationship. The node access order, edge attributes, and internal structure of the composite node contained in the chain connection structure are written into the inspection business link description data, and the inspection business link is output.

[0058] In a preferred embodiment of the present invention, the node weight unit includes: The object parsing unit is used to structure the scope of the execution object in the inspection business process based on the segment feature data to obtain the object's scope of action data. The depth evaluation unit is used to map the processing depth of each processing segment to a depth score based on the segment feature data, thereby obtaining processing depth data. The sequence influence unit is used to identify the order of each processing segment in the scheduling path based on the segment feature data, and convert the order into a sequence influence score to obtain sequence influence data. The impact calculation unit is used to calculate the impact value of each processing segment on the inspection business process based on the object action data, processing depth data, and sequential impact data, and obtain the impact value data. The weight mapping unit is used to divide the influence value into several weight levels based on the influence value data, and to assign corresponding node weight values ​​to each weight level to obtain node weight data.

[0059] In this embodiment of the invention, the object parsing unit is used to structure the scope of action of the executing object in the inspection business process according to the segment feature data to obtain object action data, avoiding confusion between different action objects in the weight division process, and ensuring that the subsequently generated inspection business link maintains clear boundaries and clear road segment affiliation in the road object dimension; the depth evaluation unit is used to map the processing depth of each processing segment to a depth score according to the segment feature data to obtain processing depth data, and translate the processing type characteristics and scope characteristics of the processing segment into a single numerical depth score, so that the structure of the inspection business link in the data processing depth dimension has a visual and computable hierarchical characteristic; the sequence influence unit is used to identify the order of each processing segment in the scheduling path according to the segment feature data, and convert the order into a sequence influence score to obtain sequence influence data, avoiding some segments in the path that only undertake the task The processing segment of the transition function obtains the same weight level as the key control segment in the weight mapping, so that the final inspection business link has a clear control level in the time sequence dimension. The impact calculation unit is used to calculate the impact value of each processing segment on the inspection business process based on object action data, processing depth data, and sequence impact data, and obtain impact value data. It integrates the three essentially different feature dimensions of execution object coverage, processing depth, and path sequence into a single impact value measure, so that the differences of different processing segments in multi-dimensional features converge into a numerical sequence that can be directly sorted and segmented. The weight mapping unit is used to divide the impact value into several weight levels based on the impact value data, and assign corresponding node weight values ​​to each weight level, and obtain node weight data. This ensures that the inspection process presents a clear trunk and branch structure at the node level, while maintaining the adjustability and scalability of the impact calculation model.

[0060] The object parsing unit is used to structure the scope of the execution object in the inspection business process based on the segment feature data, to obtain object action data, specifically including: First, the execution object field, object type field, road segment identifier field, lane identifier field, business time window identifier field, and additional tags related to spatial location and time range are read from the segment feature data corresponding to each processing segment. Based on the pre-configured object classification rule table, the execution object field is matched with the object type field, mapping execution objects belonging to different categories such as road defect detection, obstacle recognition, sign and marking detection, work order generation, work assignment execution, and result confirmation to a unified object category code, and storing the same category code in the object category dimension of the object action data. Based on the road segment identifier field and lane identifier field, the road spatial range associated with each processing segment is... Quantitative processing involves calculating the road segment and lane intervals covered by each processing segment according to a predefined road segment granularity, and writing the resulting spatial coverage value into the spatial dimension of the object's impact data. Based on the business time window identifier field and timestamp information, the start and end times of each processing segment on the business time axis are parsed to calculate the duration of the processing segment within the corresponding time window and its coverage ratio in the overall inspection cycle, and the resulting temporal coverage value is written into the temporal dimension of the object's impact data. Using the processing segment number as an index, the object category code, spatial impact data, and temporal impact data are organized according to a unified data structure format to generate object impact data.

[0061] The depth evaluation unit is used to map the processing depth of each processing segment to a depth score based on the segment feature data, thereby obtaining processing depth data, specifically including: First, the processing type field, processing function description field, and parameter fields related to processing complexity of each processing segment are read from the segment feature data. Using the processing type field as the primary key, the processing segments are divided into preset types such as data acquisition, preprocessing and filtering, feature extraction and model inference, multi-stage inference and cross-validation, task generation and updating, and manual review and termination confirmation. According to the preset depth mapping rules, a basic depth level is assigned to different processing types. Processing segments that only complete the raw data access are mapped to a lower depth, those that perform feature calculation and single-model inference are mapped to a medium depth, those that perform multi-stage inference or multi-model interactive judgment are mapped to a higher depth, and those that directly generate inspection event records, disposal tasks, or final result confirmations are mapped to the highest depth. Based on the basic depth level, the processing complexity parameters in the segment feature data, such as the number of dimensions of input data required for processing, the number of levels of the algorithm called, and the number of business status fields involved, are numerically corrected to form a refined depth score. Using the processing segment number as an index, the depth score calculated for each processing segment is written into the processing depth data.

[0062] The sequence influence unit is used to identify the order of each processing segment in the scheduling path based on segment feature data, and convert this order into a sequence influence score to obtain sequence influence data, specifically including: First, the order information of processing segments in each scheduling path is read from the scheduling path data. The first processing segment executed in the path is marked as the entry segment, the last processing segment that completes output and ends the path is marked as the exit segment, and the processing segments located between the entry and exit segments are marked as intermediate segments. A sequence position index is assigned to each processing segment based on its position in the path. Combining the path branch information and the recurrence information of processing segments in the segment feature data, position aggregation is performed on processing segments that are in key positions in multiple paths to determine their effective sequence position in the overall inspection workflow. Then, according to the preset sequence influence scoring rules, the sequence position indices of the entry and exit segments are mapped to higher sequence influence benchmark values, while the sequence position indices of processing segments located in the middle of the path and only performing intermediate conversion functions are mapped to lower sequence influence benchmark values. The sequence influence benchmark values ​​are then weighted and adjusted based on the depth score to generate a sequence influence score that reflects the comprehensive sequence influence of the processing segment in the path structure. The sequence influence score is written into the sequence influence data using the processing segment number as an index.

[0063] The impact calculation unit is used to calculate the impact value of each processing segment on the inspection business process based on object action data, processing depth data, and sequential impact data, and obtain impact value data, specifically including: First, using the processing segment number as the association key, the object category, spatial scope, and temporal scope data corresponding to each processing segment are read from the object impact data. Depth scores and order impact scores are read from the processing depth data. These three types of data are then concatenated into a comprehensive feature record at the processing segment level according to a unified field layout. Based on a preset impact calculation strategy, weight coefficients are assigned to the object scope, depth score, and order impact score. Object impact factors are constructed according to the spatial and temporal coverage ratios of the object scope; depth factors are constructed according to the depth score; and order factors are constructed according to the order impact score. Then, the object impact factor, depth factor, and order factor are combined according to the weight coefficients to calculate a single impact value for each processing segment. The impact values ​​of all processing segments are normalized or scaled to ensure that the calculated impact values ​​under different scenarios remain within a uniform numerical range, facilitating subsequent threshold division and level mapping. Finally, using the processing segment number as an index, the impact value corresponding to each processing segment is written into the impact value data.

[0064] The weight mapping unit is used to divide the influence value into several weight levels based on the influence value data, and assign corresponding node weight values ​​to each weight level to obtain node weight data, specifically including: First, the impact values ​​of all processing segments are read from the impact value data. The distribution range of the impact value set is statistically analyzed to determine the minimum, maximum, and segment intervals of the impact values. Based on the requirements for hierarchical management of nodes in the road inspection business process, the number of weight levels and the boundaries of the impact value intervals corresponding to each weight level are determined. The impact value intervals are divided into multiple continuous and non-overlapping sub-intervals, and a unique node weight code is assigned to each sub-interval. For example, a lower-level node weight is assigned to the sub-interval with lower impact values, and a higher-level node weight is assigned to the sub-interval with higher impact values. Then, the impact values ​​of each processing segment are traversed to determine the impact value sub-interval to which each impact value belongs, and the node weight code of the corresponding sub-interval is assigned to the processing segment, forming a correspondence between processing segments and node weights. Using the processing segment number as an index, the node weight code is written into the node weight data structure to generate node weight data.

[0065] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A smart inspection management system with a hybrid expert mode under a cloud-edge collaborative architecture, characterized in that, The system includes: The segment module is used to acquire basic inspection data and divide the basic inspection data into multiple segment units according to the business time window. The record elements and node order within each segment unit are combined to obtain segment combined data. The trajectory module is used to construct a trajectory concatenation sequence based on the occurrence order of each recorded element according to the segment combination data, and to convert the trajectory concatenation sequence into trajectory matrix data. The aggregation module is used to construct node load data based on trajectory matrix data and calculate process aggregation factors based on node load data. The flow splitting module is used to filter out candidate segments from the segment combination data based on the process aggregation factor, construct a segment flow matrix based on the candidate segments, and calculate the process flow splitting factor. The scheduling module is used to select the corresponding expert processing path based on the combination range of process aggregation factor and process diversion factor, and obtain scheduling path data. The link module is used to determine process node data based on scheduling path data and to construct inspection business links based on process node data.

2. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture as described in claim 1, characterized in that, The segment module includes: The data processing unit is used to arrange the basic inspection data in chronological order according to the time stamps in the basic inspection data to obtain time series data. The segment generation unit is used to segment time series data according to the range of the business time window, generate multiple segment units, and obtain segment unit data; The structural decomposition unit is used to break down the record content of each segment unit into record elements and node sequences based on the segment unit data, thereby obtaining segment element node data. The combined construction unit is used to arrange and combine the record elements and node sequences into a continuous structure according to the segment unit numbering order based on the segment element node data, so as to obtain the segment combined data.

3. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture as described in claim 2, characterized in that, The trajectory module includes: The sequential extraction unit is used to extract the sequential position corresponding to each record element according to the arrangement order of record elements in the segment combination data, so as to obtain the record sequence data; The index building unit is used to determine the numbering based on the sequential position of the record elements in the record sequence data, and to generate a feature index for each record element to obtain sequential index data; The trajectory concatenation unit is used to connect the feature indexes of the same record feature according to the arrangement order in the segment combination data based on the sequential index data, to construct a continuous index sequence and obtain the trajectory concatenation sequence. The matrix generation unit is used to generate trajectory matrix data by taking the index sequence number as the row, the connection order of each element in each index sequence as the column, and filling each feature index into the corresponding row and column positions according to the trajectory concatenation sequence.

4. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture as described in claim 3, characterized in that, The aggregation module includes: The row-level calculation unit is used to calculate row-level statistical values ​​based on the number of feature indexes in each row of the trajectory matrix data, and obtain row-level density data. The segment analysis unit is used to identify the changes in the occurrence of each row in different segment units based on the trajectory matrix data and segment combination data, and to obtain segment change data. The jump recognition unit is used to count the number of cross-segment jumps between segment units in each row based on trajectory matrix data and segment combination data, and to obtain cross-segment jump data. The load construction unit is used to combine row-level density data, segment change data, and cross-segment jump data into node load data.

5. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture according to claim 4, characterized in that, The aggregation module further includes: Based on the node load data, the load concentration degree of each row in the trajectory matrix is ​​calculated to obtain the density score; the stability of the load between each segment is calculated to obtain the change suppression score; based on the density score and the change suppression score, nodes with concentrated and stable loads are identified to obtain the aggregation item. Based on the node load data, the jump frequency of the node between each segment unit is calculated to obtain the jump score; the density score is adjusted by power to obtain the density reinforcement score; based on the density reinforcement score and the jump score, the cross-segment continuity is calculated to obtain the balance term. Based on the node load data, the load fluctuation intensity of each row between each segment unit is calculated to obtain the change intensity score; the degree of discontinuity across segments of each row is calculated to obtain the jump intensity score; based on the change intensity score and the jump intensity score, nodes with high fluctuation and high jump are identified to obtain the suppression term. By integrating the aggregation term, balancing term, and suppression term, the aggregation stability of the corresponding nodes in each row in the entire business process is calculated, and the process aggregation factor is obtained.

6. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture according to claim 5, characterized in that, The traffic splitting module includes: The segment contribution unit is used to calculate the contribution value of each segment unit to the process aggregation factor based on the process aggregation factor and segment combination data, and obtain the segment contribution data. The candidate filtering unit is used to identify segment units whose contribution values ​​reach a preset contribution threshold based on the segment contribution data, and to obtain candidate segment data. The element statistics unit is used to count the frequency and order changes of recorded elements in each candidate segment based on the candidate segment data, and to obtain candidate element data. The flow matrix unit is used to row and columnize the recorded elements in each candidate segment based on the candidate element data, and generate a matrix structure with the recorded elements as rows and the segment numbers as columns to obtain the segment flow matrix.

7. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture as described in claim 6, characterized in that, The splitting module also includes: Based on the segment flow matrix and candidate element data, the degree of balance in the flow distribution of recorded elements within each candidate segment is calculated to obtain the balance allocation term: Based on the segment flow matrix and candidate segment data, the diffusion degree of recorded elements among multiple candidate segments is calculated to obtain the cross-segment diffusion term. Based on the segment flow matrix and segment contribution data, the degree of flow concentration in high contribution segments is calculated to obtain the flow concentration term; the degree of redundant loading in candidate segments is calculated to obtain the redundancy correction term. By integrating the traffic concentration term, balanced allocation term, cross-segment diffusion term, and redundancy correction term, the adaptability of each candidate segment to undertake the diversion task in the business process is calculated, and the process diversion factor is obtained.

8. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture as described in claim 7, characterized in that, The scheduling module includes: The interval division unit is used to divide the process aggregation factor and process diversion factor into intervals based on their numerical ranges, thereby obtaining combined interval data. The interval mapping unit is used to map different combined intervals to expert processing paths based on the combined interval data. It constructs a mapping structure with the combined interval as the key and the expert processing path as the value to obtain interval path mapping data. The path selection unit is used to match the expert processing path corresponding to the numerical combination of the process aggregation factor and the process diversion factor based on the interval path mapping data, so as to obtain the preliminary path data. The path construction unit is used to arrange multiple processing segments in the expert processing path according to the business sequence based on the preliminary path data, and generate scheduling path data.

9. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture according to claim 8, characterized in that, The link module includes: The segment feature unit is used to identify the business characteristics of each processing segment in the scheduling path based on the scheduling path data, and to parse the execution object, processing depth and processing order of each processing segment to obtain segment feature data. The node weight unit is used to calculate the impact value of each processing segment on the inspection business process based on the segment feature data, and to map the impact value to the node weight value to obtain the node weight data. The node construction unit is used to define the processing segment as a process node based on the node weight data, and to combine the process nodes into a node sequence according to the execution order to obtain the process node data. The link generation unit is used to analyze the difference in node weight values ​​between process nodes based on process node data, map the node sequence into a chain connection structure, and obtain the inspection business link.

10. The intelligent inspection management system with hybrid expert mode under a cloud-edge collaborative architecture according to claim 9, characterized in that, The node weight unit includes: The object parsing unit is used to structure the scope of the execution object in the inspection business process based on the segment feature data to obtain the object's scope of action data. The depth evaluation unit is used to map the processing depth of each processing segment to a depth score based on the segment feature data, thereby obtaining processing depth data. The sequence influence unit is used to identify the order of each processing segment in the scheduling path based on the segment feature data, and convert the order into a sequence influence score to obtain sequence influence data. The impact calculation unit is used to calculate the impact value of each processing segment on the inspection business process based on the object action data, processing depth data, and sequential impact data, and obtain the impact value data. The weight mapping unit is used to divide the influence value into several weight levels based on the influence value data, and to assign corresponding node weight values ​​to each weight level to obtain node weight data.

Images