Industrial chain operation detection method based on multi-dimensional fusion data

Abstract

The invention relates to the technical field of industrial chain management, in particular to an industrial chain operation detection method based on multi-dimensional fusion data. The method comprises the following steps: establishing a multi-dimensional standard hierarchical structure; performing quantitative weighting on the standard in the hierarchical structure by adopting an analytic hierarchy process to generate a weighted evaluation model; scoring candidate suppliers based on the weighted evaluation model; selecting the candidate supplier with the highest score as a new supplier; after the new supplier is integrated into the industrial chain, continuously collecting a plurality of key performance indicator data streams from the industrial chain; based on the key performance indicator data stream, identifying an operation deviation generated from a learned baseline by using an unsupervised anomaly detection algorithm; and when one operation deviation is identified, generating a risk early warning based on the operation deviation. According to the invention, active supplier evaluation and intelligent real-time operation monitoring are integrated, so that the visibility, decision scientificity and overall toughness of an industrial chain are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, specifically a supply chain operation monitoring method based on multi-dimensional fusion data. Background Technology

[0002] Driven by global "dual carbon" goals, the electric vehicle industry has entered a period of rapid development. As a comprehensive industry integrating mechanical manufacturing, electronic information, energy, and other fields, the electric vehicle industry chain is long, complex, and involves many links. Its stable operation is crucial to the development of enterprises and the entire industry. However, existing supply chain operation monitoring methods have many shortcomings when applied to the electric vehicle industry, failing to meet practical needs. On the one hand, supplier evaluation metrics are often too simplistic. Traditional evaluations tend to focus only on a few aspects such as product quality or delivery time, failing to comprehensively reflect a supplier's overall capabilities. Taking core component suppliers for electric vehicles as an example, the technological innovation capabilities, production scale, and supply stability of power battery suppliers directly impact the product performance and market competitiveness of vehicle manufacturers. However, traditional evaluation methods struggle to effectively assess battery suppliers' R&D capabilities and production capacity scalability; similarly, they fail to comprehensively evaluate suppliers' performance in key dimensions such as intelligent control technology development and product customization capabilities for drive motors and electronic control systems. This can lead to potential risks for vehicle manufacturers when selecting suppliers due to incomplete evaluations, such as suppliers lagging behind in technological upgrades causing the vehicle to gradually lose its technological advantage in the market.

[0003] On the other hand, the lack of effective data collection and analysis methods during the operation of the electric vehicle industry chain makes it difficult to detect potential risks and problems in real time. The electric vehicle industry chain involves the coordinated production and supply of numerous components, from upstream mineral resource supply to midstream manufacturing of core components such as batteries, motors, and electronic controls, and downstream vehicle assembly and sales. Problems in any link can trigger a chain reaction. Traditional methods cannot identify these risk factors that have a significant impact on the operation of the industry chain in a timely and accurate manner.

[0004] Furthermore, existing risk warning mechanisms are inadequate. In the electric vehicle industry, different risk factors have vastly different impacts on the supply chain. For example, battery safety issues can trigger recalls, severely damaging a company's reputation and market share; while delays in the supply of some minor components have a relatively smaller impact on vehicle delivery. However, existing warning mechanisms cannot provide targeted countermeasures based on the different levels of risk, leaving companies relatively passive in handling risks and unable to take timely and effective measures to mitigate losses.

[0005] Therefore, there is an urgent need for a more comprehensive and efficient supply chain operation testing method applicable to the electric vehicle industry, in order to improve the stability and competitiveness of the supply chain and ensure the healthy and sustainable development of the electric vehicle industry. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a supply chain operation monitoring method based on multi-dimensional fusion data. This method combines proactive supplier assessment with continuous and intelligent operational monitoring, aiming to overcome the limitations of the prior art.

[0007] This solution presents a supply chain operation monitoring method based on multi-dimensional fusion data, comprising the following steps: Establish a multi-dimensional standard hierarchical structure for supplier evaluation, which includes a quality dimension, a technical capability dimension, and a supply chain resilience dimension; Among them, the quality dimension includes product quality standards and process control capability standards; the technical capability dimension includes performance standards and innovation capability standards; and the supply chain resilience dimension includes production capacity standards and logistics performance standards. The analytic hierarchy process (AHP) is used to quantify and weight the standards within the hierarchical structure to generate a weighted evaluation model. The candidate suppliers are scored based on the weighted evaluation model to obtain a comprehensive evaluation score; The candidate supplier with the highest comprehensive evaluation score will be selected as the new supplier. After the new supplier is integrated into the industry chain, multiple key performance indicator data streams are continuously collected from the industry chain; The key performance indicators include on-time and full delivery rate, perfect order rate, supplier defect rate, and inventory turnover rate; Based on the key performance indicator data stream, an unsupervised anomaly detection algorithm is used to identify operational deviations from the learned baseline. Once an operational deviation is identified, a risk warning is generated based on the operational deviation.

[0008] Preferably, the evaluation method for the process control capability standard specifically includes: Collect quality characteristic data during the production process, including dimensions, weight, and performance parameters; Based on the collected quality characteristic data, a mean-range control chart or a single-value-moving range control chart is drawn in real time. Based on the collected quality characteristic data, the process performance index is calculated; If the data points in the control chart exceed the control limits, exhibit a non-random pattern, or the process capability index fails to meet the preset standard, then the process control capability is deemed insufficient.

[0009] Preferably, the production capacity standard is based on capacity scalability and historical shipment volume assessment, specifically including: In terms of capacity scalability assessment, we examine the supplier's investment in equipment upgrades as a percentage of fixed assets, the percentage of newly added professional and technical personnel, and the completion status of production line automation transformation projects in the past two years. In terms of historical shipment volume assessment, we statistically analyzed the supplier's quarterly shipment volume data for the past three years and analyzed its annual shipment volume growth rate and quarterly shipment volume fluctuation coefficient. If the capacity scalability-related indicators are lower than 80% of the industry average, or if the annual growth rate of historical shipments is negative for two consecutive years and the quarterly volatility coefficient is greater than the industry standard value, then the production capacity assessment is deemed unqualified.

[0010] Preferably, the step of using the analytic hierarchy process (AHP) to quantify and weight the standards within the hierarchical structure to generate a weighted evaluation model specifically includes: Within each level of the hierarchical structure, a 1-9 scale method is used, and the domain expert group pairs all the criteria together and compares them according to their relative importance to the target of the next higher level to construct a judgment matrix; Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector, and normalize the eigenvector to obtain the initial weight vector of each standard. Calculate the consistency index based on the largest eigenvalue of the matrix and the order of the matrix: ; In the formula, As a consistency indicator, The largest eigenvalue of the matrix. Let be the order of the matrix; Find the corresponding order The average random consistency index is used to calculate the consistency ratio based on the consistency index and the average random consistency index. ; In the formula, The consistency ratio, The average random consistency index; like If the judgment matrix is ​​considered to have satisfactory consistency, the initial weight vector is accepted as the final weights for each criterion, and a weighted evaluation model is generated; if If so, the judgment matrix is ​​readjusted until the consistency requirement is met.

[0011] Preferably, the on-time and full delivery rate is the ratio of the number of orders delivered on time and in full by the supplier to the total number of orders within a statistical period, and the statistical period is a calendar month.

[0012] Preferably, the perfect order rate is the ratio of the number of orders that simultaneously meet the requirements of on-time delivery, accurate order quantity, no product defects, and complete delivery documents to the total number of orders within a statistical period, and the statistical period is a calendar month.

[0013] Preferably, the supplier defect rate is the ratio of the number of defective products delivered by the supplier to the total number of delivered products within a statistical period, and the statistical period is a calendar month.

[0014] Preferably, the inventory turnover rate is the ratio of product sales cost to average inventory balance within a statistical period, where the statistical period is a calendar year.

[0015] Preferably, the unsupervised anomaly detection algorithm is the Isolation Forest algorithm.

[0016] Preferably, the step of identifying operational deviations from the learned baseline using an unsupervised anomaly detection algorithm based on the key performance indicator data stream specifically includes: A certain number of data samples are randomly selected from the collected key performance indicator data stream to form a training dataset; For each training dataset, recursively and randomly select an attribute and a split value for that attribute, and divide the data samples into left and right child nodes until each child node contains only one data sample or reaches the preset tree depth, thus constructing multiple isolated trees to form an isolated forest. For the data point to be detected, start from the root node of the isolated forest and traverse downwards according to the selected attribute and split value until the leaf node is reached, and record the path length of the data point on each isolated tree. Calculate the average path length on all isolated trees, and based on this average, calculate the anomaly score for each data point. The calculated anomaly score is compared with a preset anomaly threshold. If the anomaly score is greater than the anomaly threshold, it is determined that the key performance indicator corresponding to that data point has an operational deviation from the learned baseline. The formula for calculating the anomaly score is: ; In the formula, For data points Abnormal scores, For data points Average path length in an isolated forest For a sample size of The expected path length.

[0017] Another aspect of this solution is to provide a supply chain operation monitoring system based on multi-dimensional fusion data, including a data acquisition module, an evaluation model construction module, a supplier scoring module, a data analysis module, and a risk warning module; The data acquisition module is used to collect key performance indicator data streams from the industry chain, including on-time and full delivery rate, perfect order rate, supplier defect rate, inventory turnover rate, as well as product quality data, production progress data, logistics and transportation data, etc. The evaluation model construction module is used to construct a multi-dimensional standard hierarchical structure for supplier evaluation covering quality, technical capability, and supply chain resilience dimensions, and uses the analytic hierarchy process to quantify and weight the standards within the hierarchical structure to generate a weighted evaluation model. The supplier scoring module is used to score candidate suppliers based on the weighted evaluation model to obtain a comprehensive evaluation score; The data analysis module is used to identify operational deviations from the learned baseline based on the collected key performance indicator data stream and using an unsupervised anomaly detection algorithm. The risk warning module is used to generate risk warnings at three levels (high, medium, and low) when operational deviations are detected, and push the warning information through the enterprise's internal management system.

[0018] Compared with the prior art, the advantages of this invention are: By introducing the Analytic Hierarchy Process (AHP) and constructing a multi-dimensional evaluation model that includes quality, technological capabilities, and supply chain resilience, supplier selection is no longer based on vague feelings or single-dimensional considerations, but on a comprehensive, objective, and traceable data model, thereby significantly improving the scientific nature and accuracy of decision-making.

[0019] By employing an unsupervised anomaly detection algorithm, the system autonomously learns the "normal operating mode baseline" in the multidimensional key performance indicator data stream and identifies small and complex deviations from this baseline in real time. This enables the system to capture early abnormal signals before operational problems escalate into major disruptions, giving managers time to respond and achieving proactive risk avoidance.

[0020] This solution analyzes multiple key performance indicators (such as on-time and full delivery rate, perfect order rate, and inventory turnover rate) as a whole data vector, enabling the identification of combinations of patterns that are insignificant in a single dimension but highly unusual in a multi-dimensional space. This allows the method to gain insight into systemic risks that span multiple business processes and are caused by a single root cause. Attached Figure Description

[0021] Figure 1 This is a flowchart of a supply chain operation detection method based on multi-dimensional fusion data proposed in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] refer to Figure 1 This paper describes the specific process of a supply chain operation monitoring method based on multi-dimensional fusion data.

[0024] S1. Establish a multi-dimensional standard hierarchical structure for supplier evaluation.

[0025] The evaluation framework of this invention decomposes the complex goal of "selecting the best supplier" into manageable and analyzable sub-problems, constructing a hierarchical decision-making structure. This hierarchical structure includes three primary dimension criteria: Quality dimension: This aims to assess a supplier's ability to consistently and stably produce compliant products.

[0026] Technical capabilities dimension: This aims to assess the supplier's engineering design, technological innovation, and process implementation capabilities.

[0027] Supply chain resilience dimension: This aims to assess a supplier's ability to withstand external disruptions and maintain stable operations and delivery.

[0028] In one specific embodiment, the quality dimension includes product quality standards and process control capability standards, wherein: The product quality standards are evaluated using quantitative indicators, such as a defect rate of parts per million, which provides a standardized measure of product conformity.

[0029] The process control capability standard is evaluated based on statistical process control theory: First, conduct process stability analysis: use control charts to monitor key quality characteristic data (such as dimensions, weight, etc.) in the supplier's production process. Commonly used control charts include mean-range control charts (for subgroup data) or single-value-moving-range control charts (for individual data points). If all data points fall within the control limits calculated based on historical data, and the distribution of points exhibits a random pattern (without continuous increases / decreases, periodicity, or other abnormal patterns), then the production process is considered to be in a statistically stable state, meaning the process is predictable.

[0030] Next, process capability analysis is performed: capability analysis can only be conducted after the process has been proven to be stable. Specifically, this is done by calculating the process potential capability index. and process capability index To achieve: The process potential capability index measures the degree of dispersion of the process distribution (usually expressed as...). This is the ratio of the process mean to the specification tolerance width. It only reflects the magnitude of process fluctuations and does not consider whether the process mean is aligned with the specification center.

[0031] The process capability index, based on the process potential capability index, takes into account the deviation between the process mean and the specification center, and is a key indicator for measuring the actual performance of the process.

[0032] Finally, the calculated process potential capability index and process capability index are compared with preset process control index thresholds. Typically, This is a widely accepted benchmark in the industry for determining whether a process has sufficient capability. This invention uses this benchmark to determine whether a supplier's process control capability meets the standard.

[0033] In one specific embodiment, the technical capability dimension includes performance standards and innovation capability standards, wherein: The performance criteria are evaluated through specific performance indicators targeting key components of electric vehicles: Power batteries: energy density, charge-discharge cycle life, fast charging performance and safety rating.

[0034] Drive motor: rated / peak power, maximum efficiency, torque response speed.

[0035] Electrical control system: energy conversion efficiency, control precision, electromagnetic compatibility compliance, and fault diagnosis coverage.

[0036] The innovation capability standard is assessed through customization capabilities and R&D investment: Customization capability: This is evaluated based on the number of customized projects completed by the supplier in the past 12 months, the proportion of customized projects to total projects, and the customer satisfaction ratings for these projects.

[0037] Research and development investment: By calculating the supplier's financial and human resources data for the past three years, we assess the ratio of its R&D expenditure to operating revenue, the ratio of R&D personnel to total employees, and its intellectual property output.

[0038] In one specific embodiment, the supply chain resilience dimension includes production capacity standards and logistics performance standards, wherein: The production capacity criteria are assessed based on capacity scalability and historical shipment volume: Capacity scalability: Examine the supplier's investment in equipment upgrades as a percentage of fixed assets, the proportion of new professional and technical personnel, and the completion status of production line automation transformation projects over the past two years.

[0039] Historical shipments: Statistics on the supplier's shipment data for each quarter over the past three years, analyzing its annual shipment growth rate and quarterly shipment volatility coefficient to assess the stability and growth potential of its production.

[0040] The logistics performance criteria are assessed through delivery reliability and geographic location risk: Delivery reliability: Statistics on the on-time delivery rate of orders from suppliers over the past 12 months.

[0041] Geographical location risk: Analyze the geographical distance between the supplier's production bases and warehouses and major customers and logistics hubs; assess whether its main transportation routes pass through areas prone to natural disasters (such as floods and landslides); and examine the local traffic congestion and the completeness of logistics infrastructure.

[0042] S2. The analytic hierarchy process (AHP) is used to quantify and weight the standards within the hierarchical structure to generate a weighted evaluation model, specifically including: S201. Pairwise Comparison and Judgment Matrix Construction: Within each level of the hierarchical structure, a group of domain experts (e.g., procurement, quality, and technical experts) pairs all criteria together, comparing them based on their relative importance to the objectives of the next higher level. The comparison results are quantified using a 1-9 scale, where "1" indicates that both criteria are equally important, and "9" indicates that one criterion is far more important than the other. These quantified results are then filled into a... In the positive and negative judgment matrix.

[0043] S202. Weight Vector Calculation: Calculate the eigenvector corresponding to the largest eigenvalue of the positive and negative judgment matrix as the relative weight vector of each standard. Then, normalize the calculated eigenvector to obtain the initial weight of each standard at this level.

[0044] S203. Consistency Check: Since the input of AHP relies on human subjective judgment, and human judgment may be logically inconsistent (e.g., believing A is more important than B, B is more important than C, but C is more important than A). To ensure the logical rigor of the model, a consistency check must be performed, specifically: Calculate the consistency index based on the largest eigenvalue of the matrix and the order of the matrix: ; In the formula, As a consistency indicator, The largest eigenvalue of the matrix. Let be the order of the matrix.

[0045] For completely consistent judgments At this point, the consistency index is 0.

[0046] Find the corresponding order The average random consistency index is the average value of multiple randomly generated judgment matrices and their consistency indices calculated. It serves as a benchmark for measuring the degree of deviation in random judgments.

[0047] Based on the aforementioned consistency index and the average random consistency index, the consistency ratio is calculated as follows: ; In the formula, The consistency ratio, This is the average random consistency index.

[0048] S204, Judgment and Adjustment: When When the expert judgment is considered to have acceptable consistency, the judgment matrix is ​​valid, and its calculated weight vector can be adopted. If If the result is negative, it indicates a significant logical contradiction in the judgment, and we must return to the first step, whereby the expert group will re-examine and adjust the pairwise comparisons until the consistency test is passed.

[0049] After determining the final weights of all criteria (from the first-level dimension to the lowest-level indicator) in the hierarchical structure through hierarchical analysis, the weighted evaluation model is completed. Subsequently, objective data or scores of each candidate supplier on the lowest-level indicator are collected and normalized to a unified scale (e.g., 1-100 points).

[0050] Each supplier's overall evaluation score is calculated as follows: the normalized score of each underlying indicator is multiplied by its corresponding final weight, and then the results are aggregated upwards level by level to obtain a final total score. The supplier with the highest score is identified as the most qualified supplier.

[0051] S3. After the new supplier is integrated into the industry chain, multiple key performance indicator data streams are continuously collected from the industry chain.

[0052] The key performance indicators include on-time and full delivery rate, perfect order rate, supplier defect rate, and inventory turnover rate.

[0053] The on-time and full delivery rate is the ratio of the number of orders delivered on time and in full by the supplier to the total number of orders within the statistical period, which is a calendar month or quarter.

[0054] The perfect order rate is the ratio of the number of orders that simultaneously meet the requirements of on-time delivery, accurate order quantity, no product defects, and complete delivery documents to the total number of orders within the statistical period, which is a calendar month or quarter.

[0055] The supplier defect rate is the ratio of the number of defective products delivered by a supplier to the total number of delivered products within a statistical period, which is either a calendar month or a quarter.

[0056] The inventory turnover rate is the ratio of product sales cost to average inventory balance within a statistical period, which is a calendar year.

[0057] S4. Based on the key performance indicator data stream, use an unsupervised anomaly detection algorithm to identify operational deviations from the learned baseline.

[0058] In this scheme, the unsupervised anomaly detection algorithm is preferably the Isolation Forest algorithm. The specific implementation process for identifying deviations using this algorithm is as follows: S401, Building the Forest: The algorithm first constructs a forest consisting of multiple "isolated trees." For each tree, a subsample is randomly selected from the historical key performance indicator (KPI) data stream. Then, the sample is partitioned by recursively selecting a feature (i.e., a KPI dimension) and a random split value within that feature range, until each data point is isolated to a leaf node or the preset tree depth is reached.

[0059] S402. Calculate Path Length: For a new real-time key performance indicator (KPI) data point (containing the values ​​of all current monitored indicators), traverse each isolated tree in the forest, starting from the root node and proceeding downwards according to the partitioning rules of each node until a leaf node is reached. The number of edges traversed from the root node to the leaf node is its "path length" on that tree. Because outliers are different, they are usually separated in the shallower layers of the tree, so their average path length will be significantly shorter than that of normal points.

[0060] S403. Calculate the outlier score: The final outlier score is obtained by calculating the average path length of the data point across all trees in the forest and normalizing it. The calculation formula is as follows: ; In the formula, These are abnormal scores. The path length of the data points. This represents the average path length of a data point across all trees in the forest. For a sample size The normalization factor represents the average path length of an unsuccessful search in a binary search tree, and is used to correct the effect of tree size on path length.

[0061] S404. Anomaly Identification: The anomaly score ranges from 0 to 1. A score closer to 1 indicates a shorter average path length and a higher probability that the data point is an anomaly; a score closer to 0 indicates a higher probability that it is a normal point; a score around 0.5 indicates high uncertainty. By setting a preset anomaly threshold (e.g., 0.6), when the anomaly score of a data point exceeds this threshold, it is determined to be an "operational deviation."

[0062] S5. Once the system identifies an operational deviation, it generates a risk warning based on the operational deviation.

[0063] Specifically, in this scheme, the relationship between the severity level of the early warning and the anomaly score calculated by the Isolation Forest algorithm is as follows: High Risk: Triggered when the outlier score exceeds a high-risk threshold (e.g., >0.85). This indicates a serious and rare deviation. The alert includes the outlier data point, the relevant key performance indicators, the magnitude of the deviation, and the potential business impact (e.g., "There is a risk of production line downtime").

[0064] Medium Risk: Triggered when anomaly scores fall within a medium risk range (e.g., 0.7 to 0.85). This indicates a significant but not extreme deviation. In addition to the high-risk warning, the alert will include a recent trend analysis of relevant key performance indicators.

[0065] Low Risk: Triggered when the anomaly score exceeds the low-risk threshold but falls below the medium-risk threshold (e.g., 0.6 to 0.7). This is primarily used as an early warning signal. The warning message includes a brief description of the deviation and initial observation recommendations.

[0066] Ultimately, the generated tiered alerts are automatically pushed to designated supply chain managers and relevant department heads via email, SMS, or dashboard pop-ups through the notification module of the enterprise's internal management system, ensuring rapid response and handling.

[0067] A supply chain operation monitoring system based on multi-dimensional fusion data includes a data acquisition module, an evaluation model construction module, a supplier scoring module, a data analysis module, and a risk warning module.

[0068] The data acquisition module is used to collect key performance indicator data streams from the industry chain, including on-time and full delivery rate, perfect order rate, supplier defect rate, inventory turnover rate, as well as product quality data, production progress data, logistics and transportation data, etc.

[0069] The evaluation model construction module is used to construct a multi-dimensional standard hierarchical structure for supplier evaluation that covers quality, technical capability, and supply chain resilience dimensions. It also uses the analytic hierarchy process (AHP) to quantify and weight the standards within the hierarchical structure, generating a weighted evaluation model.

[0070] The supplier scoring module is used to score candidate suppliers based on the weighted evaluation model to obtain a comprehensive evaluation score.

[0071] The data analysis module is used to identify operational deviations from the learned baseline based on the collected key performance indicator data stream and an unsupervised anomaly detection algorithm.

[0072] The risk warning module is used to generate risk warnings at three levels (high, medium, and low) when operational deviations are detected, and push the warning information through the enterprise's internal management system.

[0073] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting operation of an industry chain based on multi-dimensional fused data, characterized in that, The method comprises the following steps: establishing a multi-dimensional standard hierarchy for supplier evaluation, the hierarchy comprising a quality dimension, a technical capability dimension, and an industry chain resilience dimension; wherein the quality dimension comprises product quality standards and process control capability standards; the technical capability dimension comprises performance standards and innovation capability standards; and the industry chain resilience dimension comprises production capability standards and logistics performance standards; quantitatively weighting the standards in the hierarchy using the analytic hierarchy process to generate a weighted evaluation model; scoring candidate suppliers based on the weighted evaluation model to obtain a comprehensive evaluation score; selecting the candidate supplier with the highest comprehensive evaluation score as a new supplier; after the new supplier is integrated into the industry chain, continuously collecting a plurality of key performance indicator data streams from the industry chain; the key performance indicators comprising on-time full-delivery rate, perfect order rate, supplier defect rate, and inventory turnover rate; based on the key performance indicator data streams, using an unsupervised anomaly detection algorithm to identify operational deviations from a learned baseline; when an operational deviation is identified, generating a risk warning based on the operational deviation. 2.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The evaluation method of the process control capability standards specifically comprises: collecting quality characteristic data in the production process, the quality characteristic data comprising size, weight, and performance parameters; based on the collected quality characteristic data, real-time plotting of mean-range control charts or single-value-moving-range control charts; based on the collected quality characteristic data, calculating a process performance index; if the data points in the control chart exceed the control limits, or exhibit a non-random pattern, and the process capability index does not meet the preset standards, it is determined that the process control capability is insufficient. 3.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The production capability standards are evaluated based on capacity scalability and historical shipment volume, specifically comprising: for capacity scalability evaluation, examining the proportion of equipment updates to fixed assets, the proportion of the number of newly added professional and technical personnel, and the completion of production line automation transformation projects in the past two years; for historical shipment volume evaluation, statistically analyzing the shipment volume data of the supplier in each quarter of the past three years, and analyzing the annual shipment volume growth rate and the quarterly shipment volume fluctuation coefficient; if the capacity scalability-related indicators are lower than 80% of the industry average, or the annual growth rate of historical shipment volume is negative for two consecutive years, or the quarterly fluctuation coefficient is greater than the industry standard value, it is determined that the production capability evaluation is not up to standard. 4.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The quantitative weighting of the standards in the hierarchy using the analytic hierarchy process to generate a weighted evaluation model specifically comprises: in each layer of the hierarchy, using the 1-9 scale method, pairing all standards two by two by a group of domain experts, comparing their relative importance to the target of the previous layer, and constructing a judgment matrix; calculating the largest eigenvalue of the judgment matrix and the corresponding eigenvector, and normalizing the eigenvector to obtain an initial weight vector of each standard; calculating the consistency index based on the largest eigenvalue of the matrix and the order of the matrix: ； wherein is a consistency index, is the largest eigenvalue of the matrix, is the order of the matrix; finding an average random uniformity index of the corresponding order based on the uniformity index and the average random uniformity index, calculating a uniformity ratio: ； wherein is the consistency ratio, is the average random consistency index; If , it is considered that the judgment matrix has satisfactory consistency, and the initial weight vector is accepted as the final weight of each criterion to generate a weighted evaluation model; if , the judgment matrix is adjusted again until the consistency requirement is met.

5. The method according to claim 1, characterized in that, the on-time full-delivery rate is the ratio of the number of orders delivered on time and in full to the total number of orders in a statistical period, and the statistical period is a natural month.

6. The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The perfect order rate is a ratio of the number of orders meeting the requirements of on-time delivery, accurate order quantity, no product defects and complete delivery documents to the total number of orders in a statistical period, and the statistical period is a natural month.

7. The method according to claim 1, characterized in that, The supplier defect rate is a ratio of the number of defective products in the products delivered by a supplier to the total number of delivered products in a statistical period, and the statistical period is a natural month. 8.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The inventory turnover rate is a ratio of product sales cost to average inventory balance in a statistical period, and the statistical period is a natural year. 9.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The unsupervised anomaly detection algorithm is an isolation forest algorithm. 10.The industry chain operation detection method based on multi-dimensional fusion data according to claim 1, characterized in that, The operation deviation from the learned baseline is identified based on the key performance indicator data stream and the unsupervised anomaly detection algorithm, and specifically includes the following steps: a certain number of data samples are randomly extracted from the collected key performance indicator data stream to form a training data set; for each training data set, an attribute and a split value of the attribute are randomly selected recursively to divide the data samples into left and right child nodes until each child node contains only one data sample or reaches a preset tree depth, a plurality of isolated trees are constructed to form an isolation forest; for a data point to be detected, the data point is started from the root node of the isolation forest, and is traversed downward according to the selected attribute and the split value until reaching a leaf node, and the path length of the data point on each isolated tree is recorded; an average value of the path lengths on all isolated trees is calculated, and based on the average value, an anomaly score of the data point is calculated; the calculated anomaly score is compared with a preset anomaly threshold value, and if the anomaly score is greater than the anomaly threshold value, it is determined that the key performance indicator corresponding to the data point has an operation deviation from the learned baseline, the formula for calculating the anomaly score is: ； wherein is the anomaly score for data point is the average path length in the Isolation Forest, is the anomaly score for data point is the average path length in the Isolation Forest, is the expected path length when the sample size is is the expected path length when the sample size is

Images