Banking outlet consumable abnormal use root cause diagnosis method and system and computer readable storage medium

Abstract

The invention relates to the technical field of data processing, in particular to a bank outlet consumable abnormal use root cause diagnosis method and system and a computer readable storage medium. The method comprises the following steps: acquiring data by automatically capturing data of an acquisition system and manually inputting the data; processing the data in parallel based on four agents of service coupling degree analysis, time sequence mode detection, peer efficiency benchmarking and inventory strategy compliance verification, and respectively generating a service interpretation contribution rate, an abnormal mode label, an efficiency level and an inventory compliance label; performing standardized mapping and causal probabilistic reasoning on the structured evidence output by the intelligent agent through a Bayesian network, calculating the posterior probability of each root cause hypothesis, and screening competitive hypotheses; and generating a structured diagnosis report containing the most probable root cause, the competitive hypothesis, the key evidence chain and the action guidance according to the posterior probability. According to the method, multi-dimensional root cause diagnosis of abnormal use of the bank outlet consumables can be realized, and the management precision and interpretability are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, and computer-readable storage medium for diagnosing the root causes of abnormal use of consumables in bank branches. Background Technology

[0002] With the comprehensive advancement of digital transformation in the banking sector, although paperless office practices have been further implemented, bank branches, as frontline customer-facing entities, still have a significant demand for printing, photocopying, and form filling, resulting in a persistently high demand for related consumables. To ensure normal branch operations and improve customer satisfaction, a relatively extensive management model is adopted for consumables control. Branches request consumables as needed, and the purchasing department engages in over-purchasing. While this ensures the use of consumables, it also leads to waste. Therefore, a more refined management approach is needed. However, existing methods for consumables management suffer from insufficient refinement. Currently, the requisition, use, and procurement of consumables mainly rely on traditional manual statistics. Branches are accustomed to requesting consumables as needed, and there is a lack of digital and visual management of usage details, progress tracking, and over-limit reminders for various consumables. Traditional management relies on fixed thresholds or single time-series models, which can only reveal the surface phenomenon of "excessive usage" but cannot penetrate to the root cause of "why the usage exceeded the limit." Summary of the Invention

[0003] The present invention aims to at least partially solve one of the technical problems in the related art.

[0004] Therefore, the first objective of this invention is to provide a method for diagnosing the root causes of abnormal use of consumables in bank branches, comprising: S1, data is obtained through automatic capture of data from the centralized procurement system and manual input. The data includes consumable usage data, early warning rules, and inventory parameters. S2, based on business coupling analysis, time-series pattern detection, peer performance benchmarking and inventory strategy compliance verification, four intelligent agents process the data in parallel and generate business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label respectively. S3, the structured evidence output by the agent is standardized and mapped and subjected to causal probability inference through a Bayesian network, the posterior probability of each root cause hypothesis is calculated and competing hypotheses are screened. S4. Generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

[0005] In one embodiment of the present invention, the method for processing data based on business coupling degree analysis in step S2 is as follows: A21. Based on the acquired data, a linear model is constructed to establish a direct correlation between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the direct correlation is predicted based on the linear model. A22. Based on the acquired data, a nonlinear model is constructed to establish an indirect relationship between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the indirect relationship is predicted based on the nonlinear model. A23, the consumption of consumables under direct association and consumption of consumables under indirect association are merged according to the set weight to obtain the theoretical consumption driven by business volume. When the difference between actual consumption and theoretical consumption exceeds the preset threshold, residual abnormality is determined and the corresponding residual abnormality score is generated. A24, based on theoretical consumption, uses SHAP attribution analysis to quantify the total contribution of various business activities to consumption fluctuations and generates a business explanation contribution rate.

[0006] In one embodiment of the present invention, the method for processing data based on time-series pattern detection in step S2 is as follows: B21 uses STL decomposition to split the consumable usage data of the target network into trend items, seasonal items and residual items, and uses the SH-ESD algorithm to detect point anomalies and context anomalies in the residual items. The point anomaly is the case of a sudden increase in the amount of consumables used at a single point in time, and the context anomaly is the case of the residual items exceeding the threshold or a sudden peak for consecutive days while in a period of stable business. B22 calculates the anomaly intensity of point anomalies and context anomalies by combining the anomaly duration and deviation degree, and outputs the timestamp, anomaly type and anomaly intensity corresponding to the anomaly as anomaly pattern label.

[0007] In one embodiment of the present invention, the method for processing data based on peer performance benchmarking in step S2 is as follows: C21. Based on the acquired data, the K-Prototypes algorithm is used to perform peer group clustering on each network point. The clustering features include business structure proportion, scale level, equipment model and region. C22 uses the MAD (Modulation and Dispersion) index to measure the deviation of the target network from its peer group, and measures the performance level of the target network based on the deviation. The performance level includes leading, average, and lagging. C23. When the performance level of the target network point is lagging, analyze the reasons for the lagging performance of the network point, and output the clustering basis of the peer group list, the performance level of each network point and the reasons for lagging performance.

[0008] In one embodiment of the present invention, the method for processing data based on inventory strategy compliance verification in step S2 is as follows: D21 calculates the theoretical safety stock and maximum stock based on the traditional (s,S) inventory model, and introduces a correction formula for shelf-life constraints:

[0009]

[0010] in, For safety stock, Maximum inventory; D22 generates an inventory compliance label based on the current inventory, safety stock, and maximum inventory, provided the current inventory is greater than the adjusted maximum inventory. When the time since the last purchase is less than 50% of the purchase lead time and the current inventory is greater than the safety stock, it is considered "excessive stockpiling". At that time, it was judged as "purchasing too early".

[0011] In one embodiment of the present invention, the formula for calculating the safety stock is: = Daily consumption × Procurement lead time × Safety factor; The procurement lead time is determined by the supplier's delivery time; The formula for calculating the maximum inventory is: = Safety stock + 30 days of consumption.

[0012] In one embodiment of the present invention, S3 further includes: S31 converts the generated business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label into three-level labels, and maps the labels to Bayesian network evidence node probabilities. S32, input the evidence probability into the network, perform inference through variable elimination, and update the posterior probability of the root cause node using the preset causal relationship; S33, select root cause nodes from the root cause nodes whose posterior probability is greater than the set probability threshold, and sort the root cause nodes according to the posterior probability from large to small. Set the root cause node with the highest probability as the most likely root sound, and the top 30% of root cause nodes as competitive hypothesis root sounds.

[0013] In one embodiment of the present invention, if the posterior probability of all root cause nodes is less than the probability threshold, the staff is prompted to add a new root cause.

[0014] To achieve the above objectives, a second aspect of the present invention provides a root cause diagnosis system for abnormal use of consumables in bank branches, comprising: The data acquisition module is used to acquire consumable usage data, early warning rules, and inventory parameters through automatic data capture from the centralized procurement system and manual data entry. The intelligent agent processing module is used to process the data in parallel by four intelligent agents based on business coupling degree analysis, time sequence pattern detection, peer performance benchmarking and inventory strategy compliance verification, and generate business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label respectively. The Bayesian network processing module is used to standardize and map the structured evidence output by the agent through a Bayesian network and perform causal probability inference, calculate the posterior probability of each root cause hypothesis, and filter competing hypotheses. The diagnostic report generation module is used to generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

[0015] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0016] The methods, systems, and storage media of this invention can achieve multi-dimensional root cause diagnosis of abnormal use of consumables in bank branches, improve management accuracy and interpretability, and effectively reduce consumable waste and procurement costs.

[0017] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating a method for diagnosing the root causes of abnormal use of consumables in bank branches, provided in this application embodiment; Figure 2 This is a structural diagram of a root cause diagnosis system for abnormal use of consumables in bank branches, provided in an embodiment of this application. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] The following description, with reference to the accompanying drawings, describes a method and system for diagnosing the root causes of abnormal use of consumables in bank branches, according to an embodiment of the present invention.

[0022] Example 1 Figure 1 This is a flowchart of a root cause diagnosis method for abnormal use of consumables in bank branches according to an embodiment of the present invention.

[0023] like Figure 1 As shown, the root cause diagnosis method for abnormal use of consumables in bank branches includes the following steps: S1. Data is obtained through automatic capture of data from the centralized procurement system and manual input. The data includes consumable usage data, early warning rules, and inventory parameters.

[0024] Specifically, the step of "obtaining consumable usage data, early warning rules, and inventory parameters by automatically capturing data from the centralized procurement system and manually entering data" in this invention is the primary step in achieving refined management of bank consumable expenses. Its technical implementation principle and operation method have a high degree of system integration and data standardization.

[0025] At the technical implementation level, this step employs a dual-channel data acquisition mechanism. On one hand, it automatically captures structured data such as historical purchase records, usage details, and supplier information from the centralized procurement system via API interfaces or ETL tools. On the other hand, it allows branch managers to manually input real-time usage data, custom warning rules, and inventory parameters through a front-end data entry interface. The automatic capture module must possess capabilities such as data cleaning, field mapping, and timestamp alignment to ensure that data extracted from heterogeneous systems is compatible with the local database structure. The manual entry module must support access control, data validation, and mandatory field settings, and provide a warning rule template library for users to select or customize, such as setting trigger conditions like "monthly usage exceeds 80% of the threshold" or "month-on-month growth of 150%".

[0026] In application scenarios, this step is widely used in the initialization and data update phases of bank branch consumables management systems. Data capture from centralized procurement systems is suitable for historical data modeling and trend analysis, while manual data entry is used for real-time data updates, rule configuration, and anomaly feedback. Through this step, the system can build a unified data source, providing high-quality, structured input data for subsequent modules such as visualization, early warning analysis, and intelligent attribution.

[0027] The technical effect of this step is that by combining automation and manual data entry, it effectively solves the problems of data lag, inconsistent rules, and missing inventory parameters in traditional bank consumables management. It provides the system with a data foundation that is highly comprehensive, timely, and accurate, thereby supporting subsequent intelligent analysis and decision optimization, and realizing the digitalization, intelligence, and visualization of bank consumables management.

[0028] S2, based on business coupling analysis, time-series pattern detection, peer performance benchmarking, and inventory strategy compliance verification, processes the data in parallel by four intelligent agents, generating business explanation contribution rate, abnormal pattern label, performance level, and inventory compliance label respectively.

[0029] Specifically, the method for processing data based on business coupling analysis is as follows: A21. Based on the acquired data, a linear model is constructed to establish a direct correlation between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the direct correlation is predicted based on the linear model. A22. Based on the acquired data, a nonlinear model is constructed to establish an indirect relationship between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the indirect relationship is predicted based on the nonlinear model. A23, the consumption of consumables under direct association and consumption of consumables under indirect association are merged according to the set weight to obtain the theoretical consumption driven by business volume. When the difference between actual consumption and theoretical consumption exceeds the preset threshold, residual abnormality is determined and the corresponding residual abnormality score is generated. A24, based on theoretical consumption, uses SHAP attribution analysis to quantify the total contribution of various business activities to consumption fluctuations and generates a business explanation contribution rate.

[0030] The method for processing data based on time-series pattern detection is as follows: B21 uses STL decomposition to split the consumable usage data of the target network into trend items, seasonal items and residual items, and uses the SH-ESD algorithm to detect point anomalies and context anomalies in the residual items. The point anomaly is the case of a sudden increase in the amount of consumables used at a single point in time, and the context anomaly is the case of the residual items exceeding the threshold or a sudden peak for consecutive days while in a period of stable business. B22 calculates the anomaly intensity of point anomalies and context anomalies by combining the anomaly duration and deviation degree, and outputs the timestamp, anomaly type and anomaly intensity corresponding to the anomaly as anomaly pattern label.

[0031] The method for processing data based on peer performance benchmarking is as follows: C21. Based on the acquired data, the K-Prototypes algorithm is used to perform peer group clustering on each network point. The clustering features include business structure proportion, scale level, equipment model and region. C22 uses the MAD (Modulation and Dispersion) index to measure the deviation of the target network from its peer group, and measures the performance level of the target network based on the deviation. The performance level includes leading, average, and lagging. C23. When the performance level of the target network point is lagging, analyze the reasons for the lagging performance of the network point, and output the clustering basis of the peer group list, the performance level of each network point and the reasons for lagging performance.

[0032] The method for processing data based on inventory strategy compliance verification is as follows: D21 calculates the theoretical safety stock and maximum stock based on the traditional (s,S) inventory model, and introduces a correction formula for shelf-life constraints:

[0033]

[0034] in, For safety stock, Maximum inventory; D22 generates an inventory compliance label based on the current inventory, safety stock, and maximum inventory, provided the current inventory is greater than the adjusted maximum inventory. When the time since the last purchase is less than 50% of the purchase lead time and the current inventory is greater than the safety stock, it is considered "excessive stockpiling". At that time, it was judged as "purchasing too early".

[0035] This step involves a parallel processing mechanism of four specialized analytical agents, which intelligently analyze the consumable usage data of bank branches from four dimensions: business coupling degree, time series pattern, peer performance, and inventory strategy. The results output the business explanation contribution rate, anomaly pattern label, performance level, and inventory compliance label, providing structured evidence input for the subsequent collaborative attribution engine.

[0036] At the technical implementation level, this step adopts a distributed computing architecture, synchronously distributing the original consumable usage data to four independent intelligent agent modules. Each module executes analysis tasks in parallel based on a pre-set algorithm model. The business coupling analysis agent models the relationship between business volume and consumable consumption by constructing both linear and nonlinear models. It also uses the SHAP (SHapley Additive exPlanations) attribution method to quantify the contribution rate of each business type to consumption fluctuations, outputting the business explanation contribution rate (e.g., 70%) and residual anomaly scores (0-5 points, ≥3 points indicating high anomaly). The time series pattern detection agent decomposes the time series into trend, seasonality, and residual terms based on the STL (Seasonal and Trend decomposition using Loess) algorithm, and then combines it with the SH-ESD (Seasonal Hybrid-ESD) algorithm to identify outliers in the residual terms, outputting anomaly pattern labels, including sudden spikes and persistently high levels. The peer performance benchmarking agent uses the K-Prototypes clustering algorithm to divide peer groups based on the business structure, scale, equipment model, and geographical characteristics of the network outlets. It then determines the performance level based on the unit weighted business consumable cost and the Median Absolute Deviation (MAD) indicator, outputting the Top 5 difference factors. The inventory strategy compliance agent, based on an improved (s, S) inventory model and combined with expiration date constraints, calculates the theoretical safety stock s and the maximum inventory S, and performs compliance verification on actual inventory and procurement behavior, outputting an inventory compliance label and expiration date risk value.

[0037] Regarding parameter settings, the fusion weight of linear and nonlinear models in the business coupling analysis is 4:6; the anomaly threshold for SH-ESD in time-series anomaly detection is set to three times the standard deviation of the residual mean; the K value in peer group clustering is dynamically adjusted according to the number of branches; and the safety factor in the inventory model is set to 1.65, corresponding to a 95% service level, with an expiration date correction factor of 30%. This process is widely used in bank branch consumables management scenarios, especially suitable for complex management environments with multiple business types, multiple equipment configurations, and multiple geographical distributions. Through multi-agent parallel analysis, multi-dimensional identification and interpretation of abnormal usage behavior can be achieved, significantly improving the scientific nature and response efficiency of management decisions.

[0038] S3, the structured evidence output by the agent is standardized and mapped and subjected to causal probability inference through a Bayesian network, the posterior probability of each root cause hypothesis is calculated and competing hypotheses are screened.

[0039] Further, step S3 includes: S31 converts the generated business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label into three-level labels, and maps the labels to Bayesian network evidence node probabilities. S32, input the evidence probability into the network, perform inference through variable elimination, and update the posterior probability of the root cause node using the preset causal relationship; S33, select root cause nodes from the root cause nodes whose posterior probability is greater than the set probability threshold, and sort the root cause nodes according to the posterior probability from large to small. Set the root cause node with the highest probability as the most likely root sound, and the top 30% of root cause nodes as competitive hypothesis root sounds.

[0040] If the posterior probability of all root cause nodes is less than the probability threshold, staff are prompted to add a new root cause.

[0041] Specifically, in some implementations, this step uses a Bayesian network to standardize and map the structured evidence output by multiple agents and perform causal probabilistic inference, thereby calculating the posterior probability of each root cause hypothesis and filtering out competing hypotheses. This step is the core component of the entire consumable abnormal usage cause analysis module, aiming to integrate analysis results from different dimensions and improve the accuracy and interpretability of root cause identification.

[0042] At the technical implementation level, this step first standardizes the structured evidence output by four specialized agents (business coupling analysis, time-series pattern anomaly detection, peer performance benchmarking, and inventory strategy compliance). Specifically, the output of each agent is mapped to a three-level label of "high / medium / low," and further converted into the confidence probability of evidence nodes in a Bayesian network. For example, high confidence corresponds to... medium confidence level corresponds to Low confidence level corresponds to Among these, core evidence (such as inventory irregularities) can be given higher weight, such as... This demonstrates its strong correlation with the determination of root causes.

[0043] In Bayesian networks, causal relationship structures are constructed based on domain knowledge. For example, there is a strong causal relationship between "equipment failure" and "time series anomaly," and a significant conditional dependency between "near-expiry consumables depletion" and "excess inventory." Probability propagation is performed using variable elimination to update the posterior probability of each root cause node. Furthermore, the system filters root cause hypotheses with a posterior probability >10%, sorts them from highest to lowest probability, selects the highest probability as the "most likely root cause," and retains hypotheses with a probability >30% as "competitive hypotheses" to support complex scenarios with multiple root causes.

[0044] At the application level, this step is widely used in intelligent diagnosis of abnormal consumable usage in bank branches. When the system detects abnormal usage of thermal paper or ink cartridges in a branch, the collaborative attribution engine will integrate the analysis results of multiple agents and output a structured diagnostic conclusion, including the most likely root cause, competing hypotheses, key evidence chains, and action guidelines. For example, if the system determines that "near-expiry consumables are being used up" is the most likely root cause with a probability of 85%, it can prompt managers to check the expiration date of consumables and the requisition approval process, thereby achieving precise intervention.

[0045] In terms of technical effectiveness, this step effectively solves the problems of misjudgment and ambiguous root cause localization in single-agent analysis by utilizing the causal reasoning mechanism of Bayesian networks. Through multi-source evidence fusion and probability propagation, the system can achieve high-confidence identification of the causes of abnormal usage while retaining multiple competing hypotheses, enhancing the robustness and interpretability of the diagnosis. This method significantly improves the level of intelligence in bank consumables management, providing managers with clear and actionable decision-making basis.

[0046] S4. Generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

[0047] Specifically, in some implementations, generating a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability is the core output step of the collaborative attribution engine agent in this invention. This step, based on the multi-agent analysis results, uses Bayesian networks for causal reasoning to achieve accurate attribution and interpretability analysis of abnormal consumable usage behavior.

[0048] At the technical implementation level, this step first standardizes the outputs from four agents: business coupling analysis, time-series pattern anomaly detection, peer performance benchmarking, and inventory strategy compliance. The output of each agent is mapped to a three-level label of "high / medium / low," and further converted into corresponding probability values ​​(e.g., high = 0.9, medium = 0.5, low = 0.1). The weight of core evidence nodes (such as inventory violations and business decoupling) can be increased to 0.95 to enhance their influence in the attribution process. Subsequently, this evidence is input into a pre-defined Bayesian network, where probability propagation is performed using variable elimination to update the posterior probabilities of each root cause node.

[0049] At the parameter level, the posterior probability of root cause nodes is used to quantify their likelihood of occurrence. For example, if the posterior probability of "consuming near-expiry consumables" is 85%, it is determined to be the most likely root cause. Simultaneously, the system retains hypotheses with posterior probabilities higher than 30% as competing hypotheses to provide multi-faceted explanations. The key evidence chain consists of multiple high-confidence evidence nodes, such as "excessive inventory accumulation," "premature requisition," "sudden anomalies in timing," and "business decoupling," forming a logical closed loop and enhancing the credibility of the diagnosis. Furthermore, the system outputs preliminary action guidelines, such as "verify the expiration date and requisition approval process for consumables," to guide subsequent management actions.

[0050] At the application level, this step is widely used in the automatic diagnosis of abnormal consumption of consumables in bank branches. When the system detects abnormal usage of thermal paper or ink cartridges in a branch, the collaborative attribution engine will automatically integrate multi-dimensional analysis results to generate a structured report, allowing managers to quickly identify the root cause of the problem and take targeted measures.

[0051] The technical effect of this step is that, through multi-source evidence fusion and causal reasoning, it significantly improves the accuracy and interpretability of anomaly attribution, avoids the risk of misjudgment in single-agent analysis, and provides scientific and operable decision support for bank consumables management.

[0052] The root cause diagnosis method for abnormal use of consumables in bank branches according to this invention realizes digital tracking and intelligent analysis of the entire process of bank consumable use, effectively solving the waste problem caused by traditional manual management. Through multi-agent collaborative diagnosis and visual early warning, it improves the precision level and decision-making efficiency of consumable management.

[0053] Example 2 Figure 2 This is a structural diagram of a root cause diagnosis system for abnormal use of consumables in bank branches, according to an embodiment of the present invention.

[0054] like Figure 1 As shown, the root cause diagnosis system for abnormal use of consumables in bank branches includes: The data acquisition module is used to acquire consumable usage data, early warning rules, and inventory parameters through automatic data capture from the centralized procurement system and manual data entry. The intelligent agent processing module is used to process the data in parallel by four intelligent agents based on business coupling degree analysis, time sequence pattern detection, peer performance benchmarking and inventory strategy compliance verification, and generate business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label respectively. The Bayesian network processing module is used to standardize and map the structured evidence output by the agent through a Bayesian network and perform causal probability inference, calculate the posterior probability of each root cause hypothesis, and filter competing hypotheses. The diagnostic report generation module is used to generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

[0055] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned root cause diagnosis method for abnormal use of consumables in bank branches.

[0056] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the 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. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0057] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for diagnosing the root causes of abnormal use of consumables in bank branches, characterized in that, include: S1, data is obtained through automatic capture of data from the centralized procurement system and manual input. The data includes consumable usage data, early warning rules, and inventory parameters. S2, based on business coupling analysis, time-series pattern detection, peer performance benchmarking and inventory strategy compliance verification, four intelligent agents process the data in parallel and generate business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label respectively. S3, the structured evidence output by the agent is standardized and mapped and subjected to causal probability inference through a Bayesian network, the posterior probability of each root cause hypothesis is calculated and competing hypotheses are screened. S4. Generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

2. The method as described in claim 1, characterized in that, The method for processing data based on business coupling degree analysis described in step S2 is as follows: A21. Based on the acquired data, a linear model is constructed to establish a direct correlation between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the direct correlation is predicted based on the linear model. A22. Based on the acquired data, a nonlinear model is constructed to establish an indirect relationship between the business volume of the target outlet and the consumption of consumables, and the consumption of consumables under the indirect relationship is predicted based on the nonlinear model. A23, the consumption of consumables under direct association and consumption of consumables under indirect association are merged according to the set weight to obtain the theoretical consumption driven by business volume. When the difference between actual consumption and theoretical consumption exceeds the preset threshold, residual abnormality is determined and the corresponding residual abnormality score is generated. A24, based on theoretical consumption, uses SHAP attribution analysis to quantify the total contribution of various business activities to consumption fluctuations and generates a business explanation contribution rate.

3. The method as described in claim 1, characterized in that, The method for processing data based on time-series pattern detection in step S2 is as follows: B21 uses STL decomposition to split the consumable usage data of the target network into trend items, seasonal items and residual items, and uses the SH-ESD algorithm to detect point anomalies and context anomalies in the residual items. The point anomaly is the case of a sudden increase in the amount of consumables used at a single point in time, and the context anomaly is the case of the residual items exceeding the threshold or a sudden peak for consecutive days while in a period of stable business. B22 calculates the anomaly intensity of point anomalies and context anomalies by combining the anomaly duration and deviation degree, and outputs the timestamp, anomaly type and anomaly intensity corresponding to the anomaly as anomaly pattern label.

4. The method as described in claim 1, characterized in that, The method for processing data based on peer performance benchmarking in step S2 is as follows: C21. Based on the acquired data, the K-Prototypes algorithm is used to perform peer group clustering on each network point. The clustering features include business structure proportion, scale level, equipment model and region. C22 uses the MAD (Modulation and Dispersion) index to measure the deviation of the target network from its peer group, and measures the performance level of the target network based on the deviation. The performance level includes leading, average, and lagging. C23. When the performance level of the target network point is lagging, analyze the reasons for the lagging performance of the network point, and output the clustering basis of the peer group list, the performance level of each network point and the reasons for lagging performance.

5. The method as described in claim 1, characterized in that, The method for processing data based on inventory strategy compliance verification in step S2 is as follows: D21 calculates the theoretical safety stock and maximum stock based on the traditional (s,S) inventory model, and introduces a correction formula for shelf-life constraints: in, For safety stock, Maximum inventory; D22 generates an inventory compliance label based on the current inventory, safety stock, and maximum inventory, provided the current inventory is greater than the adjusted maximum inventory. When the time since the last purchase is less than 50% of the purchase lead time and the current inventory is greater than the safety stock, it is considered "excessive stockpiling". At that time, it was judged as "purchasing too early".

6. The method as described in claim 5, characterized in that, The formula for calculating the safety stock is: = Daily consumption × Procurement lead time × Safety factor; The procurement lead time is determined by the supplier's delivery time; The formula for calculating the maximum inventory is: = Safety stock + 30 days of consumption.

7. The method as described in claim 1, characterized in that, S3 further includes: S31 converts the generated business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label into three-level labels, and maps the labels to Bayesian network evidence node probabilities. S32, input the evidence probability into the network, perform inference through variable elimination, and update the posterior probability of the root cause node using the preset causal relationship; S33, select root cause nodes from the root cause nodes whose posterior probability is greater than the set probability threshold, and sort the root cause nodes according to the posterior probability from large to small. Set the root cause node with the highest probability as the most likely root sound, and the top 30% of root cause nodes as competitive hypothesis root sounds.

8. The method as described in claim 7, characterized in that, If the posterior probability of all root cause nodes is less than the probability threshold, staff are prompted to add a new root cause.

9. A root cause diagnosis system for abnormal use of consumables in bank branches, characterized in that, include: The data acquisition module is used to acquire consumable usage data, early warning rules, and inventory parameters through automatic data capture from the centralized procurement system and manual data entry. The intelligent agent processing module is used to process the data in parallel by four intelligent agents based on business coupling degree analysis, time sequence pattern detection, peer performance benchmarking and inventory strategy compliance verification, and generate business explanation contribution rate, abnormal pattern label, performance level and inventory compliance label respectively. The Bayesian network processing module is used to standardize and map the structured evidence output by the agent through a Bayesian network and perform causal probability inference, calculate the posterior probability of each root cause hypothesis, and filter competing hypotheses. The diagnostic report generation module is used to generate a structured diagnostic report containing the most likely root cause, competing hypotheses, key evidence chains, and action guidelines based on the posterior probability.

10. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as claimed in any one of claims 1-8.

Images