Material bill change management method and device, equipment and storage medium

Abstract

The invention relates to a change management method and device of a bill of materials, equipment and a storage medium. The method comprises the steps of obtaining material data streams of a business change event and a database change event, constructing a BOM topological structure according to the material data streams, performing risk analysis on the BOM topological structure based on a risk analysis model to obtain a risk analysis result, and triggering execution of a corresponding decision according to the risk analysis result to obtain an execution result, the risk analysis result comprises at least one of a first risk analysis result, a second risk analysis result, a third risk analysis result, a prediction analysis result and an anomaly identification result. According to the method, the automatic BOM management method of the full-link service closed loop from data flow change to construction of the BOM topological structure, then risk analysis of the BOM topological structure and finally decision execution is achieved, manual intervention is not needed in the management process of the full-link service closed loop, and the BOM management efficiency can be greatly improved.

Description

Technical Field

[0001] This application relates to the field of intelligent vehicle manufacturing technology, and in particular to a method, apparatus, equipment, and storage medium for managing changes to a bill of materials. Background Technology

[0002] In the automotive manufacturing industry, the Bill of Materials (BOM) serves as the core data of the automotive product structure, and its change management directly impacts vehicle production efficiency, supply chain collaboration, and product quality. Related technologies primarily employ methods for automotive BOM change management that include: daily monitoring of BOM data in the business system, constructing a BOM topology diagram based on this data, then using a depth-first search algorithm to traverse each node in the BOM topology diagram, calculating the direct and indirect impacts of changes, and finally analyzing the calculation results to update the changed BOM or obtain analysis reports.

[0003] However, current BOM change management methods suffer from poor management effectiveness. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, equipment, and storage medium for changing a bill of materials (BOM) that can improve the effectiveness of BOM change management, in order to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for managing changes to a bill of materials, including:

[0006] Obtain material data streams for business change events and database change events, and construct a BOM topology based on the material data streams;

[0007] Risk analysis is performed on the BOM topology based on a risk analysis model to obtain risk analysis results; the risk analysis results include at least one of the following: first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results;

[0008] The execution of the corresponding decision is triggered based on the risk analysis results, and the execution result is obtained.

[0009] Secondly, this application also provides a bill of materials change management device, comprising:

[0010] The acquisition module is used to acquire material data streams of business change events and database change events, and construct a BOM topology structure based on the material data streams;

[0011] The analysis module is used to perform risk analysis on the BOM topology based on the risk analysis model and obtain risk analysis results; the risk analysis results include at least one of the following: first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results;

[0012] The execution module is used to trigger the execution of corresponding decisions based on the risk analysis results and obtain the execution results.

[0013] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0014] Obtain material data streams for business change events and database change events, and construct a BOM topology based on the material data streams;

[0015] Risk analysis is performed on the BOM topology based on a risk analysis model to obtain risk analysis results; the risk analysis results include at least one of the following: first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results;

[0016] The execution of the corresponding decision is triggered based on the risk analysis results, and the execution result is obtained.

[0017] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0018] Obtain material data streams for business change events and database change events, and construct a BOM topology based on the material data streams;

[0019] Risk analysis is performed on the BOM topology based on a risk analysis model to obtain risk analysis results; the risk analysis results include at least one of the following: first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results;

[0020] The execution of the corresponding decision is triggered based on the risk analysis results, and the execution result is obtained.

[0021] The aforementioned method, apparatus, equipment, and storage medium for managing changes to the bill of materials (BOM) acquires material data streams from business change events and database change events, constructs a BOM topology based on the material data streams, performs risk analysis on the BOM topology based on a risk analysis model, obtains risk analysis results, and then triggers the execution of corresponding decisions based on the risk analysis results to obtain execution results. The aforementioned method achieves automated BOM management across the entire business loop, from data flow changes and BOM topology construction to risk analysis and decision execution. This closed-loop management process requires no manual intervention, significantly improving BOM management efficiency. Furthermore, whenever a business change event is triggered, the entire closed-loop management process can perform real-time risk analysis and decision execution based on the material data flow of both business and database change events, eliminating the need to wait for batch events to trigger changes and thus reducing analysis latency and overcoming the real-time limitations of traditional BOM management methods. Moreover, the automatic decision execution based on risk analysis results solves the problem of disconnect between traditional risk analysis and decision execution, greatly improving BOM management effectiveness. Additionally, the method implements multi-dimensional risk analysis, which can improve the accuracy of risk analysis and consequently increase the efficiency and accuracy of subsequent decision execution based on risk analysis results, thereby reducing BOM management costs. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is an application environment diagram of a bill of materials change management method in one embodiment;

[0024] Figure 2 This is one of the flowcharts illustrating a method for managing changes to a bill of materials in one embodiment;

[0025] Figure 3 This is a second flowchart illustrating a method for managing changes to a bill of materials in one embodiment;

[0026] Figure 4 This is the third flowchart illustrating a method for managing changes to a bill of materials in one embodiment;

[0027] Figure 5This is the fourth flowchart illustrating a method for managing changes to a bill of materials in one embodiment;

[0028] Figure 6 This is the fifth flowchart illustrating a method for managing changes to a bill of materials in one embodiment;

[0029] Figure 7 This is a flowchart illustrating the bill of materials change management method in one embodiment;

[0030] Figure 8 This is the seventh flowchart illustrating a method for managing changes to a bill of materials in one embodiment;

[0031] Figure 9 This is a structural block diagram of a bill of materials change management device in one embodiment;

[0032] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0033] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of this application.

[0034] It should be understood that although the terms “first,” “second,” etc., may be used herein to describe various elements, this does not indicate any order, quantity, or importance, but is merely used to distinguish different components. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. Words such as “comprising” or “including” mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, without excluding other elements or objects.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0036] In the automotive manufacturing industry, the Bill of Materials (BOM), as the core data of product structure, directly impacts production efficiency, supply chain collaboration, and product quality through change management. Current automotive BOM change management primarily employs the following technologies: First, batch-driven change analysis. For example, daily scheduled tasks scan the Product Lifecycle Management (PLM) system to obtain BOM change data. The analysis engine performs batch analysis based on preset SQL scripts, ultimately outputting a change impact report. Second, using business rule engines (e.g., the Drools rule engine) for BOM change risk assessment. The system hardcodes change rules into rule files; when a change request is received, the business rule engine matches the corresponding rules and outputs the assessment results. Third, calculating the scope of impact by constructing a fixed BOM topology graph. Specifically, the BOM data can be converted into a directed graph structure, and then a depth-first search algorithm can be used to traverse the graph nodes to calculate the direct and indirect impacts of the change. Fourth, the system relies on engineers to manually interpret the BOM analysis results. After the system generates a change impact report, engineers need to manually evaluate the report content, coordinate with departments such as production, procurement, and quality, formulate a change implementation plan, and synchronize the decision-making results via email or meeting.

[0037] The following shortcomings exist in automotive BOM change management technologies: First, insufficient real-time performance. Traditional BOM processing, being a batch processing model, results in change analysis delays of several hours, failing to meet the rapid iteration needs of new energy vehicles. Second, weak topology awareness. Traditional static topology analysis cannot capture dynamic changes brought about by vehicle platform design, leading to a high false positive rate, potentially exceeding 30%. Third, rigid traditional risk assessment models cannot adapt to complex business scenarios, resulting in a high false alarm rate, potentially exceeding 25%. Fourth, the efficiency of manual decision-making based on analysis results is extremely low. All of these factors contribute to the poor effectiveness of BOM change management in existing vehicle systems.

[0038] In view of this, embodiments of this application propose a method, apparatus, equipment, and storage medium for managing changes to a bill of materials (BOM), which can achieve efficient management of the BOM on vehicle production lines and in the supply chain.

[0039] It should be noted that the beneficial effects or technical problems solved by the embodiments of this application are not limited to this one, but may also be other implicit or related problems. For details, please refer to the description of the embodiments below.

[0040] The bill of materials change management method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can obtain material information and the relationships between all components in the vehicle system (i.e., the Bill of Materials, BOM) from the database storage system. When a business change event occurs, it analyzes the impact or risk of the components involved in the business change event on the components in the vehicle system, obtains the analysis results, and triggers corresponding decision execution based on the analysis results, such as updating the BOM, thus realizing change management of the BOM in the vehicle system. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Headset devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0041] In one exemplary embodiment, such as Figure 2 As shown, a method for managing changes to a bill of materials is provided, which can be applied to... Figure 1 Taking the terminal in the example, the explanation includes:

[0042] S201, obtain the material data streams for business change events and database change events, and construct the BOM topology based on the material data streams.

[0043] Among them, business change events are data flows published by the business system; database change events are basic data flows captured from the database. In the automotive manufacturing field, the topology of the Bill of Materials (BOM) can be represented by abstract "nodes" and "edges (connections)," which characterize the material composition network of automotive products.

[0044] In this embodiment, on the one hand, when data changes (e.g., order adjustments, design changes), business systems (e.g., Product Lifecycle Management PLM, Enterprise Resource Planning ERP) proactively push events to a Kafka message queue through domain services (e.g., BOM change approval service), and publish them to a topic (e.g., bom_change_events) via the Kafka message queue. This ensures that only approved changes enter the stream processing pipeline or stream processing layer, thereby constructing a material data stream. On the other hand, for non-business-triggered basic data changes (e.g., inventory quantities automatically changing due to inbound and outbound operations), FlinkCDC can directly capture change logs in the database to supplement the sampling of the database change data stream, thereby synchronizing the underlying data of the business system (e.g., material master data, process routes, etc.). When a published business change event is obtained and a database change event is captured, the business change event and the database change event can be associated to form a complete semantic data stream, i.e., a material data stream, so that change risk analysis, decision execution, decision strategy selection, and decision optimization can be performed based on this material data stream. When the terminal obtains the material data stream based on the aforementioned steps, it can use Gelly to construct the BOM topology structure based on the material data stream, so that risk analysis can be performed based on the BOM topology structure later.

[0045] S202, based on the risk analysis model, perform risk analysis on the Bill of Materials (BOM) topology structure corresponding to the material data flow, and obtain the risk analysis results.

[0046] The risk analysis results include first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results. The first risk analysis results include critical nodes, critical paths, shortest critical paths, and highest-risk areas. The second risk analysis results include the edge weights of each edge in the BOM topology. The third risk analysis results include the number and scope of nodes actually affected by predicted business change events, as well as the risk level of the business change events. The predictive analysis results include the number and scope of nodes affected by predicted change events in the BOM topology, and may also include the dependency depth of the change, change risk trends, and high-risk nodes. The anomaly identification results include a list of anomalous nodes (e.g., a node corresponding to a component that simultaneously connects to two conflicting high-dependency clusters) and anomaly types (e.g., dependency loops, isolated nodes, redundant connections). The BOM topology is a tree structure and also a directed graph, representing the compositional relationships of the product. The BOM topology includes multiple nodes and multiple edges; each node represents a component, and edges represent the dependencies between components, for example, one component is a subcomponent of another component. Risk analysis models can be implemented through training based on various types of neural network models, machine learning, or intelligent models, and are used for risk analysis based on BOM topology.

[0047] In this embodiment, when the terminal obtains the BOM topology based on the aforementioned steps, it can combine the topological features of the BOM topology with the node features of each node in the BOM topology to analyze the importance value of each node in the BOM topology. This allows for subsequent risk or impact analysis based on the importance value of each node. Furthermore, the BOM topology can be divided into risk clusters to identify the risk cluster to which each node belongs, and the cluster density of each risk cluster can be calculated accordingly. It can also be used to determine the number and scope of components affected by business change events in the BOM topology. Additionally, it can detect abnormal nodes or edges in the BOM topology. The results obtained from these analyses or identifications can all be used as risk analysis results, enabling subsequent decision-making and execution based on these risk analysis results.

[0048] S203, based on the risk analysis results, triggers the execution of the corresponding decision and obtains the execution result.

[0049] The decision-making process includes full execution, partial execution, delayed execution, and may also include pre-verification. The execution result includes successful execution or execution failure.

[0050] In this embodiment, when the terminal obtains the risk analysis results based on the aforementioned steps, it can analyze these results, select the corresponding decision to execute, and obtain the execution result. For example, when the risk analysis results include the number and scope of nodes affected by the predicted change event in the BOM topology, and the execution priority of the nodes, the corresponding decision to execute can be selected based on these risk analysis results. Specifically, the various information types contained in the risk analysis results can be analyzed using the corresponding risk analysis models. For example, when the risk analysis results include the number and scope of nodes affected by the predicted change event in the BOM topology, the corresponding risk analysis model can be an impact prediction model; when the risk analysis results include key nodes, critical paths, shortest critical paths, and highest-risk areas, the corresponding risk analysis model can be a key information identification model; when the risk analysis results include the edge weights of each edge in the BOM topology, the corresponding risk analysis model can be an edge weight calculation model; when the risk analysis results include the risk level of the business change event, the corresponding risk analysis model can be a risk level estimation model; when the risk analysis results include a list of abnormal nodes and abnormal types, the corresponding risk analysis model can be an abnormal identification model. Any of the above models can be obtained through pre-training. When the risk analysis results are obtained through the above model, a pre-trained decision model can also be used to analyze the risk analysis results and output the corresponding decision.

[0051] In the above-mentioned bill of materials change management method, the material data flow of business change events and database change events is obtained, and the BOM topology is constructed based on the material data flow. Risk analysis is performed on the BOM topology based on the risk analysis model to obtain the risk analysis results. Then, the execution of corresponding decisions is triggered based on the risk analysis results to obtain the execution results. The risk analysis results include at least one of the following: key nodes, highest risk areas, vector features, cluster classification results, edge weights, complex event handling rules, risk scores, predictive analysis results, and anomaly identification results. The aforementioned method achieves automated BOM management across the entire business loop, from data flow changes and BOM topology construction to risk analysis and decision execution. This closed-loop management process requires no manual intervention, significantly improving BOM management efficiency. Furthermore, whenever a business change event is triggered, the entire closed-loop management process can perform real-time risk analysis and decision execution based on the material data flow of both business and database change events, eliminating the need to wait for batch events to trigger changes and thus reducing analysis latency and overcoming the real-time limitations of traditional BOM management methods. Moreover, the automatic decision execution based on risk analysis results solves the problem of disconnect between traditional risk analysis and decision execution, greatly improving BOM management effectiveness. Additionally, the method implements multi-dimensional risk analysis, which can improve the accuracy of risk analysis and consequently increase the efficiency and accuracy of subsequent decision execution based on risk analysis results, thereby reducing BOM management costs.

[0052] In one exemplary embodiment, a method for risk analysis based on BOM topology is provided, the method comprising:

[0053] S301 identifies the attribute characteristics of each node and edge in the BOM topology to obtain the first risk analysis result.

[0054] The first risk analysis results include at least one of the following: critical nodes, critical paths, shortest critical paths, and highest-risk areas. The first risk analysis results may also include Complex Event Processing (CEP) rules. Node attribute characteristics include the node's importance value (PageRank), and edge attribute characteristics include the regions connected by the edge. Critical nodes represent the parts that have the greatest impact on the BOM topology; the highest-risk areas represent high-risk areas for change propagation; the critical path is the path formed by connecting critical nodes; the shortest critical path is the path from the critical node on the final assembly line to each component.

[0055] In this embodiment, when the terminal constructs the BOM topology, it can use an importance algorithm (PageRank algorithm) to traverse each node in the BOM topology, calculate the importance value of each node, and then use the importance values ​​of each node to traverse each edge in the BOM topology, calculate the degree of tightness between nodes on each edge, determine the connected regions based on the degree of tightness between nodes on each edge, and use the connected regions as critical paths. Then, it determines the shortest critical path and the highest-risk region from multiple critical paths, and finally uses this information as the first risk analysis result. Optionally, when the terminal identifies the importance value, critical nodes, critical paths, shortest critical paths, and high-risk regions of each node in the BOM topology, it can fuse CEP rules based on this information to generate new CEP rules.

[0056] S302, the BOM topology is transformed into a vector representation to obtain the vector characteristics of each node in the BOM topology, and risk analysis is performed based on each vector characteristic to obtain the second risk analysis result.

[0057] The second risk analysis results include vector features, cluster classification results, edge weights, and risk scores.

[0058] In this embodiment, when the terminal constructs the BOM topology, it can further explore the deep semantic relationships of the BOM network and assign topological business semantics to the BOM topology result. Therefore, a graph embedding model or graph embedding algorithm (e.g., the random walk graph embedding algorithm Node2Vec) can be used to transform the BOM topology into a vector representation to obtain the vector features of each node in the BOM topology. The graph embedding model can be pre-learned through vector learning. Later, based on the vector features of each node in the BOM topology, a change risk analysis of the BOM topology can be performed, yielding risk scores for each node in the BOM topology, risk scores for business change events, cluster classification results for each node in the BOM topology, and edge weights for each edge in the BOM topology. The aforementioned risk analysis process based on vector features can employ corresponding risk analysis models or algorithms. The risk analysis model can also be determined based on the type of information contained in the second risk analysis result. For example, when the second risk analysis result includes cluster classification results, the corresponding risk analysis model can be a cluster classification model or algorithm; when the second risk analysis result includes edge weights, the corresponding risk analysis model can be an edge weight calculation model or algorithm; and when the second risk analysis result includes risk scores, the corresponding risk analysis model can be a risk scoring model. Any of these types of models can be pre-learned or trained intelligent models.

[0059] In an exemplary embodiment, when the first risk analysis result includes critical nodes and high-risk areas, an implementation method for identifying critical nodes and high-risk areas is also provided, which includes:

[0060] S401, determine the importance of each node in the BOM topology, and identify key nodes based on the importance of each node to obtain the key nodes in the BOM topology.

[0061] Importance can be represented by the PageRank value. Critical nodes are those components with high PageRank values ​​(such as crankshafts). Changes to these components will affect more downstream components, and therefore they are identified as critical nodes.

[0062] In this embodiment of the application, the terminal can use the following relationship (1) to calculate the importance (PageRank value) of each node in the BOM topology:

[0063] PR(A) = (1-d) + d * (PR(T1) / C(T1) + ... + PR(Tn) / C(Tn)) (1);

[0064] Where PR(A) represents the PageRank value of node A; d represents the damping coefficient, usually taken as 0.85; T1...Tn represent all nodes pointing to node A; C(Ti) represents the out-degree of node Ti; n represents the number of nodes; and i represents the index of any node between 1 and n.

[0065] Once the terminal calculates the importance of each node in the BOM topology (which can be identified by PR or PageRank value), it can sort these nodes by importance PR from largest to smallest. One approach is to select the top-ranked nodes as key nodes, where the number of nodes selected can be determined according to the identification requirements; for example, the top 5 nodes can be selected as key nodes. Another approach is to compare the importance PR of each node with a preset importance threshold, and select nodes with an importance greater than the preset importance threshold as key nodes, thus achieving the identification of key nodes. The preset importance threshold can be determined in advance according to the identification requirements.

[0066] Optionally, a method for identifying key nodes is also provided, the method comprising:

[0067] S501, Initialize the initial importance values ​​of each node in the BOM topology;

[0068] S502, substitute the initial importance value of each node into the importance calculation model for iterative calculation to obtain the importance value of each node;

[0069] S503, determine the key nodes in the BOM topology based on the importance value of each node.

[0070] This application embodiment relates to an iterative method for calculating the importance of nodes. First, a node association matrix (component association matrix) is constructed based on the BOM topology, and the PageRank value of each node is initialized. Specifically, the initial value of all nodes is set to 1 / N, where N represents the total number of nodes in the BOM. Then, each node in the BOM topology is traversed. First, the initial value is substituted into the importance calculation model (corresponding to the above relation (1)) to calculate the PageRank value of a certain node. Then, the PageRank value of the certain node in the BOM is updated based on the calculated PageRank value. Then, based on the new PageRank value and the PageRank values ​​of other nodes, the PageRank value of the certain node is calculated in the second iteration. This iterative calculation continues until the difference between two iterations is less than a preset threshold (e.g., 0.0001), at which point the iterative calculation stops. The PageRank value of a certain node calculated in the last iteration is the final calculated PageRank value of the aforementioned node. Here, only one node is used as an example. The calculation method for the PageRank values ​​of other nodes in the BOM topology can all be obtained using the above method, which will not be elaborated here.

[0071] Once the importance value of each node is determined using the method described above, nodes with high PageRank values ​​(such as the node corresponding to the crankshaft) are identified as critical nodes. For example, a node (part) that is directly or indirectly dependent on by multiple important components or has a high PageRank value will be identified as a critical node if its changes affect many other nodes. The pseudocode for the above method is as follows:

[0072] for i in range(max_iterations): # Iterates at most max_iterations times to prevent infinite loops when the loop fails to converge;

[0073] for node in graph.nodes: # Iterate through each node in the graph;

[0074] # Calculate the sum of the contributions of all incoming edge nodes;

[0075] sum_contribution = sum(PR[neighbor] / neighbor.out_degree forneighbor in node.in_neighbors);

[0076] # Update the PageRank value of the current node;

[0077] PR[node] = (1-d) + d * sum_contribution;

[0078] S402, determine the highest risk region based on each connected component in the BOM topology.

[0079] Connected components are also closely related subgraphs in the BOM topology, with the highest risk areas being the subgraphs of strongly connected components.

[0080] In this embodiment, the nodes and edges in the BOM topology can be traversed to find the largest, internally fully connected subgraph, i.e., the strongly connected component, representing a set of interdependent parts. Furthermore, critical paths and shortest critical paths can be determined based on each connected component. Moreover, when the highest-risk area, critical path, and shortest critical path are determined, the highest-risk area and shortest critical path can be marked as "high-dependency clusters," and the critical path can be marked as a "cluster ID."

[0081] The code corresponding to the above method is as follows:

[0082] DataSet <long>connectedComponents = Gelly.connectedComponents(bomGraph); / / Mark nodes of the same connected component as "high-dependency cluster";

[0083] Shortest path algorithm: Calculate the critical path length and dynamically update the dependency urgency coefficient;

[0084] / / Calculate the shortest path length for each component, starting from the critical node of the final assembly line; DataSet<(Long,Double)> shortestPaths = Gelly.shortestPaths( bomGraph, Collections.singletonList(finalAssemblyNodeId), DistanceMetric.EUCLIDEAN, );

[0085] Optionally, the method for determining the highest risk area may include: determining the largest connected component based on each connected component in the BOM topology, and using the largest connected component as the highest risk area.

[0086] The terminal can first determine the subgraphs of each connected component in the BOM topology, and then select the largest subgraph range as the largest connected component by comparing the size of the subgraph range of each subgraph. This represents the area formed by the most closely connected nodes, which is also the area most affected by changes, and is used as the final component.

[0087] In an exemplary embodiment, a method for fusing CEP rules based on the results of a first risk analysis is also provided, including: generating complex event handling rules based on key nodes and the highest risk area; wherein the complex event handling rules are used to trigger risk alarms or update the BOM status.

[0088] In this embodiment, when the terminal quantifies the key nodes and highest-risk areas in the BOM topology based on the aforementioned steps, these two data points can be used as dynamic inputs to the CEP rule. This automatically adjusts the CEP rule threshold instead of fixing it, allowing the event processing logic to perceive changes in the BOM structure in real time, thereby achieving more accurate risk identification and smarter event response. For example, the generated complex event processing rules include: when the importance value (PageRank value) of a node (e.g., a changed node or a node to be identified) is greater than the top 5% and belongs to a high-risk area, the event corresponding to that node is determined to be a key event, and an event warning can be generated or the risk level of that node can be determined; when a node does not meet the above conditions (CEP rule), the event corresponding to that node is determined to be a normal event, and the PageRank value of each node in the BOM can be updated. The aforementioned top 5% can refer to 5% of the PageRank value of the top-ranked key node in the BOM topology.

[0089] Optionally, the terminal can periodically execute the above method (graph algorithm), that is, periodically update the attributes (PageRank value), critical path, shortest critical path, highest risk area, etc. of each node in the BOM topology structure and store them in HBASE, which can be queried in real time for later CEP rule processing, critical path changes, policy updates, or decision execution.

[0090] See Figure 3 As shown, based on the methods described in the above embodiments, an exemplary explanation is provided: when a business change event data flow occurs, i.e., when the business change event is triggered, a BOM structure is constructed based on the material data flow of the business change event and the material data flow of the database change event. The importance value of each node in the BOM structure can be calculated using the aforementioned graph algorithm, identifying key nodes and high-risk areas, loading graph attributes, and dynamically generating CEP rules based on key nodes and high-risk areas to achieve CEP rule matching. If the business change event is determined to be a critical change event according to the CEP rules, a risk alarm is generated; if the business change event is determined to be a normal change event according to the CEP rules, the BOM structure is updated, and the topological features of the updated BOM are stored. Furthermore, the above... Figures 2-3 The method described in the embodiments can be implemented by the terminal's stream processing layer.

[0091] The methods described in the above embodiments can automatically discover critical paths through a series of graph algorithms, without the need for manual rule maintenance.

[0092] This method addresses the issue of traditional CEP rules being unable to adapt to BOM topology evolution. Furthermore, it incorporates multi-dimensional metrics such as node PageRank values, connectivity component dependencies, and shortest path lengths, significantly improving the accuracy of critical path and critical node identification, for example, by 30%. Moreover, this method can periodically recalculate node attributes in the BOM, and CEP rule thresholds can automatically adjust with changes in the BOM structure. For instance, if the criticality of a component decreases due to the addition of substitutes, the rules will automatically weaken the risk level of that change.

[0093] In one exemplary embodiment, another risk analysis method is provided, wherein the second risk analysis result obtained by the method includes cluster classification results and edge weights, and the method includes:

[0094] S601 converts the BOM topology into a low-dimensional vector representation to obtain the vector features of each node in the BOM topology.

[0095] In this embodiment, the terminal can employ a graph embedding algorithm (e.g., Node2Vec) to convert the BOM topology into a low-dimensional vector representation, automatically learning the potential relationships between nodes. This essentially "translates" the BOM topology into machine-understandable digital vectors, capturing potential relationships between nodes (such as functional similarity and risk transmission paths). Specifically, the terminal can generate low-dimensional vectors (e.g., 128-dimensional) for each component's nodes in the BOM topology. The closer the vectors are, the more similar the functions or risk transmission patterns of the components within the BOM topology. For example, the similar vector features of gear P001 and bearing P002 indicate that changes to them may affect similar downstream components.

[0096] S602, perform cluster classification on each node based on the vector features of each node, and obtain the cluster classification results of each node.

[0097] The cluster classification results for a node include the cluster identifier ID to which the node belongs and the dependency density of each cluster.

[0098] In this embodiment, the terminal can employ an unsupervised clustering algorithm (e.g., Density-Based Spatial Clustering of Applications with Noise, DBSCAN) to label and classify the clusters to which each node belongs by analyzing the similarity between the vector features of each node and the vector features of other nodes, thus obtaining the cluster classification results for each node. These cluster classification results include the identifier (ID) of the cluster to which each node belongs and the dependency density of each cluster. For example, cluster 101 represents the core components of the gearbox. A higher dependency density indicates a tighter dependency between parts within the cluster. For instance, the density of cluster 101 is 0.8, indicating a tight dependency between parts, and changes could easily trigger a chain reaction. By classifying the clusters in the BOM topology, high-risk part clusters can be autonomously grouped, replacing manually defined "critical component" labels.

[0099] S603, based on the vector characteristics of each node, the cluster classification results of each node, and the material information associated with each edge, determine the edge weight of each edge in the BOM topology.

[0100] In this structure, edge weights represent the change frequency of the nodes connected by the edge; the higher the change frequency of the nodes connected by the edge, the higher the edge weight. The material information associated with each edge includes the assembly quantity, dependency level, and change frequency of the components associated with each edge. The assembly quantity represents the number of assemblies represented by each edge; the dependency level is a predefined dependency strength level, such as critical dependency, important dependency, or general dependency; and the change frequency indicates how frequently the components connected to each edge undergo changes. In the BOM topology, edges represent the dependencies between nodes and their corresponding parts, and edge weights represent the strength or risk impact of this dependency.

[0101] In this embodiment of the application, after the terminal determines the cluster ID and corresponding cluster density of each node in the BOM based on the aforementioned steps, it can combine the vector features of each node in the BOM and the material information associated with each edge, such as the assembly quantity, dependency level, and change frequency of the parts corresponding to each node, to calculate the edge weight of each edge. For example, the edge weight can be calculated using the following relationship (2):

[0102] edge_weight = number of assemblies × dependency level × (1 + α × change frequency) (2);

[0103] Where edge_weight represents the edge weight; α represents the influence coefficient, used to characterize the degree of influence of the change frequency. Here, α can be a fixed influence coefficient.

[0104] Optionally, one implementation of the above-mentioned S603 "determine the edge weights of each edge in the BOM topology based on the vector features of each node and the cluster classification results" includes:

[0105] S701, based on the vector characteristics of each node and the cluster classification results, determine the weight influence coefficient of each edge in the BOM topology.

[0106] The weight influence coefficient represents the contribution of the change frequency of an edge to its edge weight, and can also be understood as the degree of influence of the change frequency.

[0107] In this embodiment, the terminal can use a random forest regression algorithm or a pre-trained prediction model to predict the contribution of the change frequency of each edge in the BOM topology to the edge weight based on the vector features of each node and the cluster classification results. This contribution is represented by a weight influence coefficient, allowing for dynamic adjustment of the weight influence coefficients corresponding to each edge in the BOM topology, replacing fixed coefficient values ​​(e.g., the traditional fixed coefficient value is 0.1). For example, when a gear is changed, the weight influence coefficient α = 0.15, indicating frequent and significant historical changes. Furthermore, based on historical experience, the "magnification factor of change frequency on risk" can be automatically adjusted; for instance, the change frequency of precision parts has a greater impact, and the weight influence coefficient α is automatically increased to 0.2. In traditional methods, the weight influence coefficient α is a fixed value. The impact of change frequency on risk may vary for different components or different dependencies. Introducing a random forest regression algorithm to dynamically calculate the weight influence coefficient α, and then combining the weight influence coefficient α with the material information associated with each edge to calculate the edge weight, allows the edge weight to accurately reflect the actual situation.

[0108] S702, based on the weight influence coefficient of each edge in the BOM topology and the material information associated with each edge, the edge weight of each edge is obtained.

[0109] The material information associated with each side includes the assembly quantity, dependency level, and change frequency of the components associated with each side.

[0110] In this embodiment of the application, the terminal can calculate the edge weight of each edge based on the material information associated with each edge and the above-mentioned relation (2).

[0111] For an illustrative explanation of the above risk analysis method, see [link to relevant documentation]. Figure 4 As shown, the above method can use graph embedding algorithms (such as the Node2Vec algorithm) to convert the BOM topology into a low-dimensional vector representation, obtaining the vector features of each node; unsupervised clustering algorithms (such as the DBSCAN algorithm) can be used to label and classify the clusters to which each node belongs, obtaining the cluster classification results of each node; the random forest regression algorithm is used to calculate the weight influence coefficient 'a' of each edge in the BOM topology based on the vector features of each node and the cluster classification results of each node; based on the weight influence coefficient 'a' of each edge and combined with the material information associated with each edge, the edge weight of each edge in the BOM topology is calculated, and then risk scoring or impact analysis is performed based on the edge weight. The above method realizes the transformation from topology to intelligence, that is, by using graph embedding algorithms (such as the Node2Vec algorithm) and unsupervised clustering algorithms (such as the DBSCAN algorithm) to transform "purely structured data" into "features containing business semantics", and then by using random forest to endow it with dynamic computing capabilities, enabling the BOM topology to have "risk perception" capabilities.

[0112] In one exemplary embodiment, another risk analysis method is also provided, through which third risk analysis results, predictive analysis results, and anomaly identification results can be obtained. This method includes:

[0113] S801. Based on the results of the second risk analysis, the topological characteristics of the BOM topology, the change characteristics of business change events, and the time characteristics of business change events, the results of the third risk analysis are obtained.

[0114] The first risk analysis results include critical nodes, critical paths, shortest critical paths, and highest-risk areas; the second risk analysis results include vector features, cluster classification results, and edge weights; and the third risk analysis results include risk scores and the number and scope of nodes affected by predicted business change events in the BOM structure.

[0115] In this embodiment, the terminal can use the cluster identifier and cluster density of the node in the second risk result as input data to input into the pre-trained prediction model for impact analysis, so as to predict the number and range of nodes affected by the business change event in the BOM topology. The terminal can also use the second risk analysis result, the topological characteristics of the BOM topology, the time characteristics of the business change event, and the change characteristics of the business change event as input data to input into the pre-trained risk scoring model for risk analysis to obtain a risk score.

[0116] Optionally, an implementation method for the above S801 is provided, the method comprising:

[0117] S901, based on the results of the second risk analysis and the change prediction model, predicts the number and scope of nodes affected by business change events in the BOM topology.

[0118] The scope of an affected node refers to the dependency chain, connectivity, or connectivity region of the node. Change prediction models can be implemented through training based on various types of neural network models, machine learning, or intelligent models to predict the number and scope of nodes affected by business change events in the BOM topology.

[0119] The second risk analysis results include the cluster identifier and cluster density to which the node belongs.

[0120] In this embodiment, the terminal can predict the number and range of nodes affected by a business change event in the BOM topology based on the cluster identifier and cluster density of each node in the BOM, using a gradient boosting tree (XGBoost) or a pre-trained change prediction model. Specifically, when a business change event exists, the cluster identifier and cluster density of the nodes associated with the business change event in the BOM structure can be determined, and the cluster associated with the change event can be identified. Based on the number of all nodes in the cluster and the area involved, the number and range of nodes affected by the business change event in the BOM topology can be obtained. For example, a gear change is expected to affect 47 nodes. In addition, the second risk analysis result can also include dependency depth. In this case, the terminal can determine the dependency depth based on the number and range of affected nodes, and then use this dependency depth as the depth of graph traversal update, which is convenient for use when updating the BOM topology later or when selecting an execution strategy later. For example, only updating the second-level downstream nodes instead of a full update; or performing PageRank recalculation, connected component marking, and other operations on the predicted 47 nodes and their dependency chains instead of a full graph update. The above method is similar to "a courier sorter predicting the impact range of a package." The traditional approach is to "search the entire warehouse for related items whenever a package arrives." XGBoost, like an experienced sorter, predicts in advance which shelves (BOM nodes) need to be checked based on package type (e.g., "fragile" corresponds to high-impact changes), weight (scale of change), and other information, accurately locating the affected area and reducing unnecessary searching. This method can reduce redundant calculations by 80%, improving change response speed to the second level. Traditional solutions update the BOM topology according to fixed rules (e.g., "full update if it affects 3 downstream levels"), which may lead to over-updates (wasting computing resources) or missed updates (lagging risk assessment). The method described in the above embodiments achieves accurate updates and resource optimization through dynamic prediction and intelligent decision-making via machine learning. Furthermore, the terminal can use reinforcement learning models or Q-learning algorithms to make decisions based on the number and range of affected nodes in the BOM topology, such as updating execution strategies, updating decisions, and updating the BOM topology state. Q-learning is a classic model-free temporal difference learning algorithm. Its core objective is to enable an agent to learn to select the optimal action in different states through interaction with the environment in order to maximize the cumulative reward. It can dynamically select update strategies based on system load and change priority.

[0121] For example, see Figure 5 The diagram illustrates that when a business change event is triggered, a gradient boosting tree (XGBoost) is used to predict the changes in the BOM topology of the business change event, obtaining the number and range of nodes affected in the BOM topology. Then, the reinforcement learning algorithm Q-learning is used to update the execution strategy based on the number and range of affected nodes, and the decision is made according to the execution strategy to update the state of the BOM topology, thereby realizing BOM change management.

[0122] S902. Based on the results of the second risk analysis, the topological characteristics of the BOM topology, the temporal characteristics of business change events, and the change characteristics of business change events, a risk score is obtained.

[0123] The topological features of the BOM topology include the vector characteristics of nodes; the change features include the type, magnitude, and frequency of business change events; the second risk analysis results include the cluster identifier ID, cluster density, and dependency depth of each node in the BOM; and the time features include the time when the business change event occurred, such as during a weekday, weekend, or peak production period.

[0124] In this embodiment, the terminal can input the second risk analysis result, the topological features of the BOM topology, the temporal features of the business change event, and the change features of the business change event into a risk scoring model (e.g., a random forest classification model) for risk analysis to obtain a risk score. Optionally, the terminal can first input the second risk analysis result, the topological features of the BOM topology, the temporal features of the business change event, and the change features of the business change event into a risk scoring model (e.g., a random forest classification model) for risk analysis to obtain a risk adjustment factor and a dynamic risk threshold. Then, risk analysis is performed based on the risk adjustment factor and / or the dynamic risk threshold to obtain a risk score. For example, regarding the risk adjustment factor, if a gear is in a critical cluster and changes over the weekend, the risk adjustment factor can be 1.3; a higher risk adjustment factor corresponds to a higher risk score. Regarding the dynamic risk threshold, for example, the threshold for critical nodes is 80, and the threshold for ordinary nodes is 120. Using the above method for risk scoring can improve the accuracy of risk scoring and greatly reduce the false positive rate.

[0125] S802, based on historical change data of BOM topology, predicts the change risk trend and high-risk nodes of BOM topology in the future time period, and obtains the prediction analysis results.

[0126] Among them, the change risk trend indicates the probability of a node changing within a future time period; high-risk nodes indicate nodes that will change within a future time period.

[0127] In this embodiment, the terminal can first obtain historical change data of the BOM topology, such as nodes that changed during historical time periods in the BOM, the clusters to which the changed nodes belong, and the risk level of the changed nodes; future time periods can be determined according to actual prediction needs. The terminal can input the historical change data of the BOM topology into time series analysis (LSTM) for risk prediction, predicting future risk areas and high-risk nodes to achieve preventive management. For example, it can output the probability of high-risk changes (change risk trend) and potential risk hotspot components (high-risk nodes) for the next 30 days. For example, the change risk trend can represent: during the high-temperature period in July, the change risk of motor parts increases by 20%; a high-risk node can represent: predicting that a certain bearing will undergo a specification change within two weeks. The above method is similar to urban traffic management departments predicting congested road sections. Traditional methods only address current events, while LSTM acts like a "traffic big data system." By analyzing historical traffic flow data (change frequency) and accident records (historical impact) from different time periods (such as rush hours and holidays), it predicts which road segments (BOM nodes) are likely to experience congestion in the future (high-risk changes) and proactively deploys police forces (adjusting approval processes) or adds warning signs (enhanced monitoring). These methods can optimize resource allocation in advance and reduce the risk of production disruptions.

[0128] S803 identifies defects or abnormal dependencies in the BOM topology and obtains the anomaly identification results.

[0129] The anomaly identification results include a list of abnormal nodes (such as a node corresponding to a part that is simultaneously connected to two conflicting highly dependent clusters) and anomaly types (such as dependency loops, isolated nodes, and redundant connections).

[0130] In this embodiment, the terminal can scan the BOM topology in real time and automatically identify design flaws or abnormal dependencies (such as loops and redundant connections) within the BOM topology. This is similar to a building quality inspector using an infrared detector to check for hidden problems in walls. Traditional solutions rely on manual inspections (fixed rule checks), while the anomaly detection algorithm acts like an "intelligent detector," quickly locating hidden problems such as wall cracks (dependency loops) and hollow bricks (isolated nodes) by scanning the building structure (BOM topology). These issues can be corrected in time during the construction phase (design phase), avoiding rework later (change costs in the production phase). Specifically, the terminal can input the BOM topology or its vector representation into the anomaly detection model (pre-trained based on a large amount of sample data) for anomaly detection, and output anomaly identification results for the BOM topology, including a list of abnormal nodes and anomaly types. This method can intercept more than 80% of structural defects in the BOM topology and also reduce production change costs.

[0131] Additionally, see Figure 6 As shown, combining S801, S802, and S803, multiple dimensions of information can be obtained through random forest, LSTM, and anomaly detection models, respectively, such as risk scores, predictive analysis results, and anomaly identification results. These results can then be used to update strategies and implement decisions. Specifically, the BOM topology is vector-transformed to obtain its vector features. The first branch takes the vector features of the BOM topology, the number and scope of nodes affected by business change events in the BOM topology, the topological features of the BOM graph topology, the time characteristics of business change events, and the change characteristics of business change events as input data, and inputs them into a pre-trained random forest model for risk analysis to obtain a risk score. The second branch takes the vector features of the BOM topology and historical change data of the BOM topology as input data, and inputs them into a pre-trained LSTM model to predict change risk trends and high-risk nodes, obtaining predictive analysis results. The third branch identifies defects or abnormal dependencies in the BOM topology based on its vector features, obtaining anomaly identification results. When risk scores, predictive analysis results, and anomaly identification results are obtained based on the above methods, strategies and / or decision execution can be updated according to these results. Among these, the Random Forest model can achieve accurate scoring, improving decision credibility; the LSTM model can predict risk trends, gaining an advantage in handling risks; and the anomaly detection model can intercept design flaws, reducing the risk at the source of change. Through the synergy of these three algorithms, risk analysis can be upgraded from "passive response" to "proactive defense"; together, they construct a full-link intelligent risk control system covering "current assessment - future prediction - source governance." The risk analysis results mentioned above (such as risk scores, dynamic risk thresholds, potential risk hotspots, anomaly node lists, and anomaly node types) can be directly used as core input conditions for the incremental graph update strategy in the decision-making execution layer. For example, when the random forest outputs "a gear change risk adjustment factor = 1.3 (high risk)," the incremental graph update module will trigger a full update strategy to ensure that the topological impact of high-risk changes is fully assessed. If anomaly detection identifies a "dependency loop" anomaly, the system will automatically mark the node as "requiring manual review" and delay the execution of related changes in the update strategy. Through this closed-loop linkage of "risk assessment → strategy decision → execution feedback," BOM change management is optimized and upgraded from a single module to full-link intelligent collaboration, achieving a dual improvement in risk prevention and resource efficiency.

[0132] The above Figures 4-6 The method described in the embodiments can be implemented by the graph processing layer of the terminal. It should also be noted that the above... Figure 3 The method described in this embodiment can be implemented by the terminal's stream processing layer. Here, the stream processing layer and the graph processing layer can communicate or transmit data. When the stream processing layer monitors the data streams of business change events and database change events in real time, processes them to obtain the BOM topology, and obtains the key nodes, critical paths, shortest critical paths, and highest-risk areas in the BOM topology, it can transmit this data to the graph processing layer. The graph processing layer performs risk analysis and decision analysis based on this data, obtaining analysis results, and can also feed these analysis results back to the data stream layer. The data stream layer can adjust the BOM topology, CEP rules, etc. The graph processing layer can also communicate or transmit data with the decision layer. The graph processing layer generates execution instructions based on the risk analysis results and transmits the execution instructions to the decision layer, which then executes the corresponding decisions based on the execution instructions, forming a fully automated closed-loop link of data collection, analysis or processing, and decision-making. This achieves automatic BOM change management, greatly improving the efficiency and effectiveness of BOM change management.

[0133] In one exemplary embodiment, a method for decision execution, namely the aforementioned, is provided. Figure 2 In the embodiment, S203 "triggering the corresponding decision execution based on the risk analysis results and obtaining the execution result" includes:

[0134] S1001, Generates an execution strategy based on the risk analysis results.

[0135] The update strategy includes at least one of the following: execution priority, alternative execution path, execution timing, and resource conflict prediction results.

[0136] In this embodiment, the terminal obtains the risk analysis results output by the stream processing layer and the graph processing layer based on the aforementioned steps. It can determine the execution priority, alternative execution path, or execution timing based on the risk analysis results. It can also determine the resource conflict prediction results based on the material data stream monitored in real time by the data stream layer and the risk analysis results, and finally obtain the update strategy.

[0137] S1002, make decisions and execute them based on the execution strategy and risk analysis results, and obtain the execution results.

[0138] When the terminal generates an execution policy, it can generate corresponding execution instructions based on the execution policy and send the execution instructions to the terminal's decision execution layer. The terminal can then execute the corresponding decisions based on the execution instructions. For example, it can perform operations such as updating the BOM topology structure, optimizing the BOM topology structure, issuing change warnings, and updating approval processes.

[0139] In an exemplary embodiment, the method for generating execution priorities based on risk analysis results may include: the execution priority can determine the resource allocation order of change processing. When the terminal obtains the risk analysis results based on the aforementioned steps, particularly the risk scores of each node in the BOM and the cluster density of the cluster to which the node belongs, execution priorities can be further generated. For example, the dynamic risk score output by the random forest (e.g., a battery pack change risk score = 150) and the cluster dependency density of DBSCAN (e.g., battery core cluster density = 0.92) can be obtained from the aforementioned graph computation layer (i.e., graph processing layer) to generate execution priorities (e.g., level 7, the highest level), and the security level of the node (e.g., ASIL-D) can also be obtained. Since the priority determines the resource allocation order of change processing, high-priority changes (e.g., core component risk changes) occupy equipment, manpower, and other resources first, ensuring that critical changes are not delayed and avoiding production line stoppages caused by delays in core component changes. For example, the above method can correspond to the following pseudocode:

[0140] def calculate_execution_priority(risk_score, cluster_density):

[0141] # Basic Priority (Levels 1-5) / Risk Score

[0142] base_priority = min(5, int(risk_score / 30))

[0143] # Cluster density bonus (priority +1 when density > 0.8)

[0144] density_bonus = 1 if cluster_density > 0.8 else 0

[0145] # Security level bonus (ASIL-D level +1)

[0146] safety_bonus = 1 if is_safety_critical() else 0

[0147] return min(5, base_priority + density_bonus + safety_bonus)

[0148] The data mapping is implemented using the pseudocode above: Risk score = 150 → Basic priority = 5, Cluster density = 0.92 → +1, ASIL-D level → +1 → Final priority = 7 (triggering the highest level execution strategy).

[0149] In an exemplary embodiment, the method for recommending alternative execution paths based on risk analysis results includes: when the terminal obtains the risk analysis results based on the aforementioned steps, particularly the vector features of each node in the BOM (such as the 128-dimensional vector of gear P001), it can calculate similar nodes in the BOM topology based on the vector features of each node, and then generate alternative execution paths. Specifically, it can use the cosine similarity of vector features. For example, if the cosine similarity between a node on a certain path and a node on the target path is greater than a threshold (such as a threshold of 0.8), that path can be used as an alternative execution path to the target path. Exemplarily, the basic logic of the above method is: BOM topology → Node2Vec random walk to generate sequences → Word2Vec training of 128-dimensional vectors → calculation of vector cosine similarity → setting a threshold based on historical cases (such as a threshold of 0.8) → filtering out alternative parts with similar functions or topology. This method ensures that when the original path cannot be executed due to resource conflicts (such as tooling fixtures being occupied), the alternative path can guarantee uninterrupted execution, improving the execution success rate and reducing production delays caused by path blocking.

[0150] In an exemplary embodiment, the method for determining the execution timing based on the risk analysis results includes: when the terminal obtains the risk analysis results based on the aforementioned steps, particularly the critical path in the BOM and the timestamps of each node in the BOM, it can determine the critical path (such as the "powertrain-chassis" path) and the change time involved in the business change event. Then, through Q-learning, the execution timing is selected based on the critical path involved in the change and the change time of the business change event. This avoids global interruptions caused by forcibly executing critical changes during busy production lines, balances "timeliness of change" with "production stability," and reduces downtime losses during peak periods. For example, if the change involves a critical path (such as "powertrain-chassis") and is during peak production (8:00-12:00), the execution timing is determined to be delayed until the current work order is completed (e.g., delayed by 30 minutes).

[0151] The above method is similar to an intelligent traffic light system, where critical path changes are like "ambulance passage," delayed during peak hours until the current traffic flow interval, thus ensuring priority while avoiding global congestion. Therefore, this method can improve execution efficiency and success rate.

[0152] In an exemplary embodiment, the above-mentioned resource conflict prediction based on material data flow and risk analysis results, to obtain the prediction result, includes: when the terminal is based on the aforementioned... Figure 2 The steps in S201 described in the embodiment obtain the material data stream, especially the equipment status stream and production scheduling stream monitored in real time by the aforementioned stream processing layer. Then, combined with the equipment IDs corresponding to the nodes affected by the change events in the risk analysis results, the probability of resource conflict is calculated, and resource conflicts are predicted based on the probability of resource conflict. For example, the probability of resource conflict (e.g., tooling fixture occupancy rate > 80%) and the affected equipment IDs are combined to calculate the probability of resource conflict (e.g., 0.6). The above method can identify potential equipment contention risks in advance (e.g., two changes require the same machine tool at the same time), pre-schedule backup resources (e.g., switch to an idle machine tool), and reduce the execution failure rate caused by resource conflicts.

[0153] In an exemplary embodiment, when the decision is to update the BOM, the decision may include any one of full execution, partial execution, or delayed execution. The corresponding execution method based on the execution strategy and risk analysis results includes: obtaining the system load, and triggering the execution of the decision based on the system load, execution strategy, and risk analysis results to obtain the execution result. The decision may be any one of full execution, partial execution, or delayed execution.

[0154] Specifically, the terminal can generate optimal decisions based on system load, execution strategy (e.g., execution priority), and risk analysis results (e.g., critical path) using reinforcement learning (Q-learning). For example, the terminal can use Q-learning to select the optimal decision (full execution, partial execution, or delayed execution) based on system load (high, medium, or low) and change priority (e.g., level 5 is the highest). Another example is delaying low-priority changes by 30 minutes under high load. This method prioritizes critical changes (such as engine part changes) when resources are limited, efficiently handles full changes under low load, improves system resource utilization, and ensures that the response time for critical changes remains stable within 1 second.

[0155] The pseudocode for the above method is as follows:

[0156] class ExecutionScheduler:

[0157] def choose_action(self, system_load, priority_level, key_path_flag):

[0158] # Status coding: Load (3 levels) + Priority (5 levels) + Critical path (Yes / No)

[0159] state = f"{system_load}_{priority_level}_{1 if key_path_flag else 0}"

[0160] # Q-value table lookup (Example Q-table snippet)

[0161] # state="HIGH_LOAD_5_1" → action="FULL_EXECUTE" (Q value=9.2)

[0162] return self.q_table.get_best_action(state)

[0163] def update_q_table(self, state, action, reward, next_state):

[0164] # Reward function design: Success +10, Failure -5, -0.5 for every minute of delay.

[0165] self.q_table.update(

[0166] state, action

[0167] reward + 0.9 * max(self.q_table.get_q_values(next_state)) -

[0168] self.q_table.get_q_value(state, action) )

[0170] The above method is similar to an airport traffic control system. It dynamically adjusts aircraft takeoff and landing sequences based on runway load (system load), flight priority (change priority), and weather warnings (critical path indicators). For example, it prioritizes high-priority flights (high-risk changes) even during peak hours, while optimizing overall traffic efficiency by delaying low-priority flights (non-urgent changes). Therefore, this method can significantly improve the efficiency and accuracy of decision-making.

[0171] In an exemplary embodiment, the decision-making process may include pre-verification. Specifically, when the terminal executes the step of making a decision based on the execution strategy and risk analysis results to obtain the execution result, it performs the following: predicting the probability of change execution failure based on the execution strategy and risk analysis results; and triggering pre-verification if the probability of change execution failure exceeds a preset probability threshold, thereby obtaining the execution result. The preset probability threshold can be determined based on prediction requirements.

[0172] Pre-verification includes operations such as tooling inspection and material re-inspection. Execution results indicate whether the execution was successful, failed, or in progress, providing relevant information about the execution.

[0173] In this embodiment, the terminal can predict the probability of change execution failure using a failure risk prediction model (e.g., Gradient Boosting Tree XGBoost) based on execution strategies (e.g., resource conflict probability) and risk analysis results (risk score, cluster density, load characteristics). Specifically, the execution strategy and risk analysis results can be used as input data to the failure risk prediction model for prediction, resulting in a predicted probability of change execution failure. This predicted probability is then compared to a preset probability threshold. If the predicted probability of change execution failure is greater than the preset threshold, it indicates that the change is likely to fail, triggering a warning and corresponding pre-verification to proactively identify potential problems. For example, based on risk analysis results and features in the execution strategy such as risk score, cluster density, system load, execution priority, execution timing, and resource conflict probability, a change execution failure probability (e.g., 0.45) can be predicted. If this probability is greater than the preset probability threshold (e.g., 0.4), pre-verification (e.g., checking the tooling 2 hours in advance) is triggered. Optionally, the terminal can also predict the probability of change execution failure based on historical failure rates, risk analysis results, and execution strategies. The above method can eliminate potential problems in advance (such as assembly errors caused by tooling wear) through pre-verification, and greatly shorten the downtime caused by execution failure, for example, by nearly 60%. It can reduce production losses caused by rework due to failure, and thus greatly reduce production costs.

[0174] The pseudocode for the above method is as follows:

[0175] def predict_execution_risk(change_event):

[0176] # Obtain risk score and cluster density from graph computation layer

[0177] risk_score = graph_computing_layer.get_risk_score(change_event.part_id)

[0178] cluster_density = graph_computing_layer.get_cluster_density(change_event.part_id)

[0179] # Obtain current load and resource conflict probability from the stream processing layer

[0180] system_load = stream_processing_layer.get_system_load()

[0181] resource_conflict = predict_resource_conflict(change_event)

[0182] # Historical failure rate (number of failed similar changes in the past 30 days / total number of changes)

[0183] historical_failure_rate = get_historical_failure_rate(change_event.change_type)

[0184] # XGBoost Feature Vectors

[0185] features = [

[0186] risk_score, cluster_density, system_load,

[0187] resource_conflict, historical_failure_rate ]

[0189] # Probability of model prediction failure (Example: Input [150, 0.92, 85, 0.6, 0.2], corresponding output 0.45)

[0190] return xgboost_model.predict([features])[0]

[0191] The above method is similar to a hospital emergency triage system. Based on patient condition scores (risk scores), infectivity (cluster density), number of patients waiting (system load), equipment usage (resource conflicts), and historical misdiagnosis rates (historical failure rates), it predicts the probability of emergency treatment failure. High-risk patients (failure probability > 0.4) are then prioritized for CT pre-examination (pre-validation process), reducing the risk of resuscitation failure. Therefore, this method can significantly reduce the execution failure rate.

[0192] In an exemplary embodiment, a method for continuously improving decisions by using execution results to reverse-optimize decisions is also provided. The method includes: adjusting the association value of the state action corresponding to the decision when the execution result indicates execution failure, and optimizing the decision based on the adjusted association value; and / or learning new sample features based on the cause of execution failure, and updating the change prediction model based on the new sample features; the change prediction model is used to predict the number and range of nodes affected by change events in the BOM topology.

[0193] In one embodiment of this application, when the execution result indicates failure, the result can be fed back to the aforementioned gradient tree (Q-learning). The reward value (represented by the Q-value) of the state-action pair corresponding to the decision is adjusted using Q-learning, and the decision is optimized based on the adjusted reward value. For example, if the execution result indicates a full update failure under high load, the reward value (Q-value) of the state-action pair is reduced from 9.2 to 4.5, and the corresponding decision can be adjusted from "full update" to "partial update." Subsequent similar scenarios will then automatically select "partial update." Optionally, the terminal can also adjust the reward value of the state-action pair corresponding to the decision based on resource consumption rate and change timeliness when the execution result indicates failure. The above method allows decisions to update the BOM topology (e.g., full / partial / delayed update) to autonomously adapt to dynamic scenario characteristics. These scenario characteristics include continuously changing system load (e.g., early morning vs. peak hours), change priority (engine crankshaft vs. ordinary bolts), and impact range (number of nodes predicted by XGBoost), which can automatically adapt to the above decisions without manual rule modification, thus improving the adaptability of change processing. The above method, through feedback on execution results, automatically optimizes decision-making, replacing fixed rules maintained manually, and can greatly improve the efficiency of change management. For example, the input-output design of the above gradient tree (Q-learning) can be shown in Table 1 below: Table 1

[0194] Module Specific input Business significance State space S 1. System load: CPU utilization (0-100%, discrete as low / medium / high), memory usage; 2. Change attributes: priority (high / medium / low, business tag), number of nodes affected by XGBoost prediction (N, discrete as small / medium / large). Describe the "current system capabilities" and the "risk level of the change itself". Action Space A <![CDATA[Three update strategies: Full update (A1): Traverse the entire graph for update, with high resource consumption but comprehensive coverage; Partial update (A2): Only update the affected areas predicted by XGBoost; Delayed update (A3): Temporarily store the queue and execute during peak hours.]]> Provides "possibilities for decision-making," covering different resource / time-efficiency trade-off scenarios. Reward function R 1. Result dimension: Successful update (+8) / Failed update (-5); 2. Resource dimension: Resource consumption rate (e.g., CPU usage <30% +5, >80% -3); 3. Timeliness dimension: High-priority change delay (-2 / minute), low-priority delay (+1 / minute, off-peak benefit); Guide strategy evolution with "reward and punishment signals": encourage strategies that are "successful + resource-saving + timely".

[0195] The pseudocode for the above method is as follows:

[0196] # Scenario: Full update fails under high load, feedback is sent to Q-learning

[0197] def handle_execution_failure(state, action):

[0198] # Original state: HIGH_LOAD_5_1, Original action: FULL_EXECUTE

[0199] reward = -5 # Failure penalty

[0200] # Get the new state (load may have decreased)

[0201] new_load = get_system_load()

[0202] new_priority = get_change_priority()

[0203] new_key_path = is_key_path_change()

[0204] next_state = f"{new_load}_{new_priority}_{1 if new_key_path else 0}"

[0205] # Perform Q-value update (original Q-value = 9.2 → new Q-value = 9.2 + 0.1 * (-5 + 0.9 * max_next_q - 9.2))

[0206] execution_scheduler.update_q_table(state, action, reward, next_state)

[0207] # Result: The Q value of this state-action pair changed from 9.2 to 4.5. For similar scenarios in the future, PARTIAL_EXECUTE will be the preferred choice.

[0208] For example, the above method, through a closed loop of "state-action-reward," allows the system to autonomously learn "which strategy to choose in which scenario." Taking "full update failure under high load" as an example, the detailed logic is as follows:

[0209] 1) The State describes the current "situation" of the system and consists of three dimensions: System load includes: CPU utilization 85% (high load), memory usage 70%; Change priority includes: changing part to engine crankshaft (priority level 5, highest); Critical path indicator: Yes (this part belongs to the powertrain critical path). The state code is: S = "High Load_Level 5_Yes".

[0210] 2) Action: The system can choose from three execution strategies:

[0211] Full update (A1): Traverses all nodes in the entire graph to update the topology. It has high resource consumption but comprehensive coverage.

[0212] Partial Update (A2): Only updates the affected nodes predicted by XGBoost (e.g., 47 nodes), with moderate resource consumption;

[0213] Delayed Update (A3): The queue is temporarily stored for 30 minutes, and the execution is staggered. It has low resource consumption but poor real-time performance.

[0214] 3) Reward (Q-value): A scoring mechanism for evaluating the quality of an action, with the following rules:

[0215] Action successful (update completed and timeout not exceeded): +10 points;

[0216] Action failed (update interrupted or timed out): -5 points;

[0217] Resource consumption (CPU > 80%): -3 points (penalty for excessive resource consumption);

[0218] High-priority change delay: -2 minutes / minute (penalty for critical task delays).

[0219] When a case of execution failure occurs based on the aforementioned 1), 2), and 3), the iterative logic in the failed case includes:

[0220] 1) Scenario: During peak production hours (9:00-11:00), the system load is 85% (high load), the engine crankshaft (priority level 5, critical path) is changed, and the system initially selects "full update (A1)".

[0221] 2) Execution result: A full update requires traversing 1000+ nodes, causing CPU usage to spike to 95%, triggering the system protection mechanism and interrupting the update (failure).

[0222] 3) Feedback reward: Failure penalty (-5) + high resource consumption (-3) = total reward -8 points.

[0223] 4) Q-value update logic: In Q-learning, each "state-action pair (SA)" has a Q-value (representing the "potential reward of choosing action A in state S"). The update formula is shown in the following relation (3):

[0224] (3);

[0225] in: The Q-value represents the current state-action pair; alpha represents the learning rate, used to control the iteration range, for example, alpha=0.1; R represents the immediate reward, the reward obtained immediately after performing the action, for example, R=-8; gamma represents the discount factor, used to balance the importance of current reward and future gains, for example, gamma=0.9; The Q-value represents the optimal action in the new state, for example, =5.0; This indicates the new state entered after the action is performed (e.g., "High Load Level 5_Yes" → "Medium Load Level 5_Yes", as some nodes have already been updated).

[0226] Substituting into the above formula, we get: New Q value = 9.2 + 0.1 * (-8 + 0.9 * 5.0 - 9.2) = 9.2 + 0.1 * (-12.7) = 7.93.

[0227] Subsequent strategy adjustments: When the state S = "High Load_Level 5_Yes" reappears, the system will compare the Q values ​​of each action:

[0228] The Q value of A1 is 4.5, the Q value of A2 is 7.8 (hypothetically), and the Q value of A3 is 3.2; the system automatically selects A2 (partial update) with the highest Q value to avoid repeated failures.

[0229] Through repeated iterations of similar cases, the system develops a mapping between "scenario and optimal strategy." For example, high load + high priority: partial update (balancing resources and real-time performance); low load + high priority: full update (ensuring comprehensive coverage); high load + low priority: delayed update (saving resources by avoiding peak loads). This method enables "systems to autonomously adapt to day-night load fluctuations without manual rule changes," significantly improving the adaptive capability of change handling, for example, by approximately 70% (i.e., 70% of dynamic scenarios can automatically match the optimal execution strategy).

[0230] Another scenario is that when the execution result indicates failure, the result can be fed back to the aforementioned change prediction model (XGBoost). The change prediction model can learn new sample features based on the reason for the failure and update the model based on these new features. Specifically, the change prediction model is used to predict the number and range of nodes affected by change events in the BOM topology.

[0231] This application's embodiments relate to an incremental training method, specifically, the modified prediction model (XGBoost) uses "new features + incremental learning" to allow the model to remember "new risk patterns" and avoid repeating the same mistakes. Taking "tooling wear leading to execution failure" as an example, the detailed logic is as follows:

[0232] 1) The initial feature set corresponding to the change prediction model (XGBoost) includes: risk score, cluster density, system load, resource conflict probability, historical failure rate, etc. (these features can be obtained through risk analysis as mentioned above). This initial change prediction model can predict the probability of execution failure based on the above feature set. However, the risks not covered by the above feature set include: tooling and fixture wear (e.g., when the wear degree > 0.6mm, insufficient part assembly accuracy leads to change execution failure). Because this feature is not recorded in historical data, the model cannot predict it.

[0233] 2) Scenario: A gear change fails. Suppose that the tooling fixture wear is found to be 0.7mm (exceeding the threshold of 0.6mm), but the original model change prediction model did not include this feature, resulting in a very low probability of prediction failure, such as a prediction failure probability of only 0.2 (actual failure).

[0234] 3) New feature extraction: Extract "tool wear degree" from the equipment logs of the MES system and associate it with failure cases: when the wear degree is > 0.6mm, the failure case accounts for 60%.

[0235] 4) Incremental training process: retain 80% of historical samples (to avoid forgetting old patterns) + add 100 failure samples containing "tool wear"; freeze some parameters of the original model (such as the weights of historical features), and only update the weights and associated parameters of the new features; fine-tune the model using the new dataset so that the association between "tool wear" and failure probability is learned (e.g., for every 0.1mm increase in wear, the failure probability increases by 15%).

[0236] 5) Reasons for improved prediction accuracy: The original model completely misjudged "failure caused by tooling wear" due to the lack of key features (for example, the prediction accuracy was about 80%). After adding features, the model can identify the composite risk of "high wear + high risk score" (such as wear of 0.7mm + risk score of 150 → failure probability of 0.7), covering unforeseen scenarios and improving accuracy, for example, it can be improved to 92%.

[0237] The above method, by continuously incorporating features of new failure causes such as "equipment aging" and "material batch differences", enables the model to upgrade from "passively responding to known risks" to "actively predicting new risks". It can also greatly shorten the prediction lag time for sudden scenarios (such as a batch of materials not meeting the strength standard), for example, from 24 hours to 2 hours. Based on this, it can ultimately reduce execution failures caused by "unforeseen factors" by about 60%, reducing downtime losses.

[0238] The iterative essence of the aforementioned reinforcement learning Q-learning and XGBoost is to transform "execution feedback" into "nutrients for algorithmic evolution." For example, the reinforcement learning model Q-learning, through a "state-action-reward" closed loop, allows the system to autonomously optimize its strategy ("what to do") in dynamic scenarios, adapting to real-time changes in load, priority, etc.; XGBoost, through "adding new features + incremental training," allows the model to continuously learn new risk patterns ("what to pay attention to"), covering unforeseen factors such as equipment and materials. Together, they achieve the leap from "fixed rules" to "autonomous evolution," enabling the BOM change management system to continuously adapt to the complex and ever-changing production environment of the manufacturing industry, ultimately achieving a complete closed loop of "analysis-decision-execution-optimization."

[0239] In one exemplary embodiment, a method for optimizing a risk analysis model by combining human experience and machine intelligence is also provided. This method can address special business scenarios not covered by the risk analysis model and improve system adaptability. The method includes: obtaining an optimization strategy through human-computer interaction, and optimizing the execution strategy based on the optimization strategy; using the optimized execution strategy as sample data to train an initial risk analysis model, thereby obtaining a trained risk analysis model.

[0240] The trained risk analysis model is used to perform risk analysis on the BOM topology. Optimization strategies include manually adjusted execution strategies based on experience, and these strategies include optimized execution priorities and high-risk paths.

[0241] For example, the optimization strategy includes the optimized execution priority. The corresponding risk analysis model can be a priority algorithm. Engineers can manually adjust the system's suggested priority level 7 to level 6 (delaying execution until after production line changeover). The system records the execution strategy and optimizes the priority algorithm, which can solve special scenarios not covered by the algorithm (such as production line changeover and temporary process adjustments), greatly improving the accuracy of the system's execution strategy in unconventional scenarios.

[0242] The above scenario can be explained as follows: An engineer discovers that a high-risk change is due to an upcoming production line change and manually adjusts the priority from 7 to 6 (delaying execution until after the change). The system records this decision, and the priority algorithm is automatically adjusted for similar scenarios in the future.

[0243] For example, optimization strategies include optimized high-risk paths. The corresponding risk analysis model can be a high-risk path identification algorithm (the model or algorithm used in the aforementioned key node identification or high-risk area identification). The optimized high-risk paths marked by engineers (such as "P001→P002→P000") are transformed into GNN training samples to improve path recommendation accuracy. This method transforms human experience into model capabilities, shortening the strategy configuration cycle for new business scenarios from 2 weeks to 2 days, accelerating the system's implementation on new production lines and new products. Historical high-risk paths marked by engineers are automatically transformed into training samples for the GNN model (corresponding to the model or algorithm used in the aforementioned key node, high-risk area, and key path identification). The effect of applying the above method is that after engineers mark multiple paths (such as 30 paths), the accuracy of GNN path recommendation is improved.

[0244] For example, the optimization strategy includes the optimized reward value (which can correspond to the Q value mentioned above), and the corresponding risk analysis model can be the decision execution algorithm (which can correspond to the model or algorithm used when triggering the corresponding decision mentioned above). For example, manually adjusting the data as "expert samples" for reinforcement learning and updating the Q value table.

[0245] The above methods can automatically optimize and handle a large proportion (e.g., about 80%) of routine scenarios, and through human-machine collaboration, solve a certain proportion (e.g., about 20%) of special scenarios, improving adaptability. For example, the above methods complement each other through data sharing (such as Q-value tables and training samples), forming a complete closed loop. Through bidirectional strategy optimization in human-machine collaboration, the system has achieved a technological leap from "machine autonomous evolution" to "human-machine intelligent symbiosis." That is, the engineer's domain knowledge fills the decision gap of the algorithm in special scenarios (such as production line changeovers and unconventional process constraints), while the algorithm transforms human experience into reusable model features (such as GNN training samples and Q-value table expert rewards), forming a closed-loop ecosystem of "human intervention → experience accumulation → algorithm evolution → strategy optimization." This mechanism enables the system to maintain high automation efficiency in a large proportion (80%) of routine scenarios, while significantly reducing the decision-making error rate by a certain proportion (e.g., 20%) of special scenarios through human collaboration. For example, it can reduce the error rate by about 80%, and shorten the strategy configuration cycle for new business scenarios from 2 weeks to 2 days. Ultimately, it builds an intelligent decision-making system that combines efficiency and adaptability, providing a complete solution for automotive BOM change management that integrates technological innovation and industry experience.

[0246] The methods described in the above-described implementation decision-making process examples achieve end-to-end linkage from "analysis results → execution instructions → feedback optimization," solving the problems of "decision lag, resource waste, and poor adaptability" in traditional BOM change management. For example, by generating execution priorities, it is clear "who does it first," ensuring that high-risk, high-value changes receive priority resources and avoiding production interruptions caused by delays in key component changes; by recommending alternative paths, it solves the "what if it can't be done" problem, ensuring execution continuity and improving change success rates when resource conflicts occur; dynamic delays and resource prediction balance "efficiency and stability," avoiding forced execution during peak production periods or resource shortages, reducing downtime losses; and through optimization using Q-learning and XGBoost, it achieves "high-efficiency resource utilization," dynamically matching system load and execution strategies, improving resource utilization and reducing failure risks, greatly enhancing the management effectiveness of change management.

[0247] In an exemplary embodiment, an implementation method for obtaining material data streams is provided, wherein the above-mentioned S201 "obtaining material data streams of business change events and database change events" includes:

[0248] S1101: Obtain business change events pushed by the business system when data changes, and capture database change events from the database.

[0249] This application embodiment involves methods for obtaining business change events and methods for capturing database change events. Specifically, the method for obtaining business change events includes: business systems such as PLM / ERP (i.e., business event centers) can proactively push business change events to a Kafka message queue through domain services (such as BOM change approval services), and publish them to a topic (e.g., bom_change_events) via the Kafka message queue. During this process, business change events can be standardized according to a predefined "BOM Change Event Protocol" to generate events in a preset format, which are then sent to the topic bom_change_events via the Kafka Producer, ensuring that only approved changes enter the stream processing pipeline. The preset format can be determined through standardized processing according to a unified "BOM Change Event Protocol," and this preset format may include the following core fields:

[0250] {"event_type": "BOM_CHANGE"; / / Event type (required) "change_scope": "STRUCTURE / SPEC"; / / Scope of change (structure / specification, required) "part_id": "P001"; / / Part ID (required) "change_actor": "USER_001"; / / Change operator (required) "change_reason": "Design optimization"; / / Reason for change (required) "approval_status": "APPROVED"; / / Approval status (required to prevent incorrect processing of ineffective changes) "biz_context": { / / Business context (required, such as associated work order, design document ID) "workorder_id": "WO_20231001"; "design_doc_id": "DOC_007";},"data_payload": { / / Data change payload (structured field) "prev_spec": "φ80mm"; "new_spec": "φ82mm"; "parent_id": "P000";}

[0251] }

[0252] Specifically, methods for capturing database change events include: for non-business-triggered basic data changes (such as automatic changes in inventory quantity due to inbound and outbound operations), Flink CDC can be used to directly capture change logs in the database to supplement the data stream of database changes, so as to synchronize the underlying data of the business system (such as material master data, process routes, etc.). In this process, capturing change logs in the database can specifically capture the MySQL inventory table.

[0253] S1102, associate business change events and database change events to obtain material data stream.

[0254] In this embodiment, when the terminal obtains business change events and database change events based on the aforementioned steps, it can align the business change events and database change events using event timestamps (event_time) to ensure time consistency in subsequent change impact analysis based on material data streams. Optionally, the terminal can also encrypt sensitive fields (such as cost, supplier_id, etc.) in the business change events using AES-256 during transmission on the Kafka side, and dynamically decrypt them during Flink CDC processing to ensure the security of data stream processing and transmission. When the material data stream is obtained, the stream processing engine can be started to transmit the material data stream to the stream processing layer for processing.

[0255] The above method for obtaining material data streams of business change events and database change events is illustrated by example. (See [link to relevant documentation]). Figure 7 As shown in Figure A, a BOM change event protocol is predefined, including structured fields (e.g., event_type, part_id, approval_status, etc.). In Figure B, PLM / ERP generates business change events according to the BOM change event protocol through domain services and actively pushes these events to the Kafka Producer message queue. The Kafka message queue then publishes the business change events to the topic bom_change_events to filter out unapproved events. In Figure C, for non-business-triggered basic data changes such as inventory, Flink CDC captures database change events based on the change log (blinlog) in the database. The captured database change events are then mapped to obtain a data stream of database change events. The data streams of published business change events and database change events are aligned, and sensitive data is protected in the aligned data streams before the processed data stream is output to the stream processing layer.

[0256] In addition to the methods described in all the above embodiments, a BOM change management system and a corresponding BOM change management method are also provided, such as... Figure 8 As shown, the BOM change management system includes: a data acquisition layer, a stream processing layer, a graph computation layer, and a decision execution layer; wherein, the data acquisition layer includes a business event center, a Flink CDC acquisition module, and a stream processing engine; for example, the business event center, the Flink CDC acquisition module, and the stream processing engine can be used to execute the aforementioned... Figure 7 The methods described in the embodiments are detailed in the foregoing description and will not be repeated here. The stream processing layer includes a CEP rule engine and a graph computation engine; the CEP rule engine can be used to execute... Figure 3 The method described in the embodiments is detailed in the foregoing description and will not be repeated here. The graph computing layer includes a semantic transformation module for the BOM topology, an execution strategy update module, a decision engine, and a risk analysis module; the graph computing layer is used to execute the aforementioned... Figures 4-6 The method described in this embodiment is detailed in the foregoing description and will not be repeated here. The decision execution layer includes a human-machine collaboration module, an execution instruction generation module, an execution strategy optimization module, and a model self-evolution module; the decision execution layer can be used to execute the aforementioned... Figure 6 The method executed by the module corresponding to the update strategy / decision execution in the embodiment is detailed in the foregoing description and will not be repeated here.

[0257] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0258] Based on the same inventive concept, this application also provides a bill of materials change management device for implementing the bill of materials change management method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more bill of materials change management device embodiments provided below can be found in the limitations of the bill of materials change management method described above, and will not be repeated here.

[0259] In one exemplary embodiment, such as Figure 9 As shown, a bill of materials change management device is provided, comprising:

[0260] The acquisition module 181 is used to acquire material data streams for business change events and database change events, and to construct a BOM topology structure based on the material data streams.

[0261] The analysis module 182 is used to perform risk analysis on the Bill of Materials (BOM) topology corresponding to the material data flow and obtain risk analysis results; the risk analysis results include at least one of the following: first risk analysis results, second risk analysis results, third risk analysis results, predictive analysis results, and anomaly identification results.

[0262] The execution module 183 is used to trigger the execution of the corresponding decision based on the risk analysis results and obtain the execution results.

[0263] The modules in the aforementioned bill of materials change management device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0264] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a bill of materials change management method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0265] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0266] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0267] Obtain material data streams for business change events and database change events, and construct the BOM topology based on the material data streams;

[0268] A risk analysis is performed on the Bill of Materials (BOM) topology corresponding to the material data flow to obtain risk analysis results; the risk analysis results include at least one of the following: a first risk analysis result, a second risk analysis result, a third risk analysis result, a predictive analysis result, and an anomaly identification result;

[0269] The execution of the corresponding decision is triggered based on the risk analysis results, and the execution result is obtained.

[0270] The computer device provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0271] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0272] Obtain material data streams for business change events and database change events, and construct the BOM topology based on the material data streams;

[0273] A risk analysis is performed on the Bill of Materials (BOM) topology corresponding to the material data flow to obtain risk analysis results; the risk analysis results include at least one of the following: a first risk analysis result, a second risk analysis result, a third risk analysis result, a predictive analysis result, and an anomaly identification result;

[0274] The execution of the corresponding decision is triggered based on the risk analysis results, and the execution result is obtained.

[0275] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0276] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0277] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0278] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.< / long>

Claims

1. A change management method of a bill of materials, characterized by, The method comprises: acquiring material data streams of business change events and database change events, and constructing a BOM topology structure according to the material data streams; performing risk analysis on the BOM topology structure based on a risk analysis model to obtain a risk analysis result; the risk analysis result comprises at least one of a first risk analysis result, a second risk analysis result, a third risk analysis result, a prediction analysis result, and an anomaly identification result; triggering execution of a corresponding decision according to the risk analysis result to obtain an execution result.

2. The method of claim 1, wherein, The risk analysis result comprises the first risk analysis result and the second risk analysis result, the risk analysis on the bill of material (BOM) topology structure corresponding to the material data stream to obtain a risk analysis result comprises: identifying attribute features of each node and each edge in the BOM topology structure to obtain the first risk analysis result; the first risk analysis result comprises a key node and a highest risk area; performing vector representation conversion on the BOM topology structure to obtain vector features of each node in the BOM topology structure, and performing risk analysis according to each vector feature to obtain the second risk analysis result.

3. The method of claim 2, wherein, The identification of the attribute features of each node and each edge in the BOM topology structure to obtain the first risk analysis result comprises: determining the importance of each node in the BOM topology structure, and identifying the key node according to the importance of each node to obtain the key node in the BOM topology structure; determining the highest risk area according to each connected component in the BOM topology structure; The determination of the importance of each node in the BOM topology structure and the identification of the key node according to the importance of each node to obtain the key node in the BOM topology structure comprises: initializing an initial importance value of each node in the BOM topology structure; substituting the initial importance value of each node into an importance calculation model for iterative calculation to obtain the importance value of each node; determining the key node in the BOM topology structure according to the importance value of each node; The determination of the highest risk area according to each connected component in the BOM topology structure comprises: determining a maximum connected component according to each connected component in the BOM topology structure, and taking the maximum connected component as the highest risk area.

4. The method of claim 3, wherein, The first risk analysis result further comprises a complex event processing rule, and the method further comprises: generating a complex event processing rule according to the key node and the highest risk area; the complex event processing rule is used to trigger risk alarm or update the BOM state.

5. The method of claim 2, wherein, The second risk analysis result comprises vector features, cluster classification results, and edge weights, the vector representation conversion on the BOM topology structure to obtain the vector features of each node in the BOM topology structure, and the risk analysis according to each vector feature to obtain the second risk analysis result comprise: converting the BOM topology structure into a low-dimensional vector representation to obtain the vector features of each node in the BOM topology structure; According to the vector features of each node, the nodes are clustered and classified to obtain a cluster classification result of each node; According to the vector features and the cluster classification result of each node, a weight influence coefficient of each edge in the BOM topology structure is determined; According to the weight influence coefficient of each edge in the BOM topology structure and the material information associated with each edge, an edge weight of each edge is obtained.

6. The method of claim 2, wherein, The risk analysis result includes a third risk analysis result, a prediction analysis result and an anomaly identification result, and the method further comprises: According to the second risk analysis result, the topological features of the BOM topology structure, the change features of the business change event and the time features of the business change event, the third risk analysis result is obtained; According to the historical change data of the BOM topology structure, the change risk trend and the high-risk node of the BOM topology structure in a future time period are predicted, and the prediction analysis result is obtained; The defect or abnormal dependency relationship in the BOM topology structure is identified, and the anomaly identification result is obtained.

7. The method of claim 6, wherein, The third risk analysis result includes a risk score, and the number and range of influence nodes of the business change event in the BOM topology structure, and the third risk analysis result is obtained according to the second risk analysis result, the topological features of the BOM topology structure, the change features of the business change event and the time features of the business change event, comprising: According to the second risk analysis result and the change prediction model, the number and range of influence nodes of the business change event in the BOM topology structure are predicted; According to the second risk analysis result, the topological features of the BOM topology structure, the time features of the business change event and the change features of the business change event, risk analysis is performed to obtain the risk score.

8. The method according to any one of claims 1 to 7, characterized in that, According to the risk analysis result, the corresponding decision execution is triggered to obtain an execution result, comprising: According to the risk analysis result, an execution strategy is generated; the execution strategy includes at least one of execution priority, alternative execution path, execution opportunity, resource conflict prediction result; According to the execution strategy and the risk analysis result, the execution of the decision is performed to obtain an execution result.

9. The method of claim 8, wherein, When the decision includes one of full execution, partial execution, and delayed execution, the execution of the decision is performed according to the execution strategy and the risk analysis result to obtain an execution result, comprising: The system load is obtained, and the execution of the decision is triggered according to the system load, the execution strategy and the risk analysis result to obtain an execution result; When the decision includes pre-verification, the execution of the decision is performed according to the execution strategy and the risk analysis result to obtain an execution result, comprising: According to the execution strategy and the risk analysis result, the change execution failure probability is predicted, and in the case that the change execution failure probability is greater than a preset probability threshold, the execution of pre-verification is triggered to obtain an execution result.

10. The method of any of claims 7, wherein, The method further comprises: In the case that the execution result indicates execution failure, the reward value of the state-action pair corresponding to the decision is adjusted, and the decision is optimized according to the adjusted reward value; And / or, learn new sample features according to the reason for the execution failure, and update the change prediction model based on the new sample features; the change prediction model is used to predict the number and range of nodes affected by the business change event in the BOM topology structure.

11. The method of claim 8, wherein, The method further comprises: An optimization strategy is obtained through human-computer interaction, and the execution strategy is optimized according to the optimization strategy; The optimized execution strategy is used as sample data to train an initial risk analysis model, and a trained risk analysis model is obtained; the trained risk analysis model is used to perform risk analysis on the BOM topology structure.

12. The method of claim 1, wherein, The method further comprises: The business change event and the database change event are obtained through the material data flow, comprising: The business change event pushed by the business system when the data is changed is obtained, and the database change event of the database is captured; 13. A change management apparatus of a bill of materials, characterized by comprising: The business change event and the database change event are associated to obtain the material data flow. The device comprises: An acquisition module is configured to acquire the material data flow of the business change event and the database change event, and construct a BOM topology structure according to the material data flow; An analysis module is configured to perform risk analysis on the BOM topology structure based on a risk analysis model to obtain a risk analysis result; the risk analysis result comprises at least one of a first risk analysis result, a second risk analysis result, a third risk analysis result, a prediction analysis result, and an abnormality identification result; 14. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, An execution module is configured to trigger execution of a corresponding decision according to the risk analysis result to obtain an execution result.

15. A computer readable storage medium having stored thereon a computer program, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 12. The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 12.

Images