Root cause analysis based internal and external collaborative OA system and process optimization method

Abstract

The application discloses an inside-outside collaborative OA system and a process optimization method based on root cause analysis, belongs to the technical field of enterprise office automation and intelligent optimization, and is characterized in that the system is divided into three-layer architecture: the bottom cause and effect discovery layer adopts an improved PC algorithm which fuses process time sequence constraint and hierarchical constraint, automatically learns the cause and effect structure from multi-source process data, and constructs a business process cause and effect graph model; the middle counterfactual reasoning layer estimates the average cause and effect of each node through a backdoor adjustment formula, accurately identifies bottleneck nodes, and adopts three-step counterfactual reasoning of traceability, intervention and prediction to pre-evaluate the input-output ratio of an optimization scheme; and the upper dynamic optimization layer configures three core modules of a reinforcement learning scheduling algorithm based on cause and effect, inside-outside business unified cause and effect model collaborative optimization and customer work order intelligent routing. The application realizes intelligent diagnosis and optimization of the process through the cause and effect inference technology, and improves the system operation efficiency and the intelligent level.

Description

Technical Field

[0001] This invention belongs to the field of enterprise office automation and intelligent optimization technology, specifically involving an internal and external collaborative OA system and process optimization method based on root cause analysis. Background Technology

[0002] Traditional enterprise OA office systems are mainly geared towards internal approval process management, driving document flow through a pre-set workflow engine.

[0003] However, existing technologies have the following technical shortcomings: First, they only count superficial indicators such as node time consumption, which cannot distinguish between causal and spurious correlations. They rely entirely on manual experience to diagnose bottlenecks, which is prone to misjudging hidden root causes and resulting in inefficient resource investment. Second, the division of labor among personnel is static and fixed. Existing scheduling only performs basic load balancing and cannot locate the real bottlenecks. When business fluctuates, resource allocation becomes unbalanced and utilization is low. Third, the data between the OA and customer work order systems is fragmented, and internal and external local optimizations are prone to conflicts, resulting in internal efficiency improvements but declining customer satisfaction. Fourth, the use of static RBAC permission configuration means that tightening permissions hinders business flow, while relaxing permissions raises security risks, making it difficult to balance security and efficiency. Fifth, security auditing relies on fixed rule matching, which can only identify known violations and has blind spots in protection against new and unknown operational anomalies.

[0004] Therefore, a new method is urgently needed to address the technical shortcomings of existing OA systems, such as reliance on experience for process optimization, static and rigid resource allocation, fragmented optimization between internal and external systems, static imbalance in access control, and limited audit detection scope. Summary of the Invention

[0005] The purpose of this invention is to provide an internal and external collaborative OA system and process optimization method based on root cause analysis. By constructing a three-layer technical system of causal discovery, counterfactual reasoning, and dynamic optimization, it can realize intelligent diagnosis of process bottlenecks, dynamic scheduling of resources, and collaborative optimization of internal and external business, thereby improving the operating efficiency and intelligence level of the OA system.

[0006] To achieve the above objectives, this invention provides an internal and external collaborative OA system based on root cause analysis, comprising a three-layer technical architecture: a bottom-layer causal discovery layer, a middle-layer counterfactual reasoning layer, and an upper-layer dynamic optimization layer; The underlying causal discovery layer includes a multi-source data acquisition and preprocessing module and an improved PC algorithm causal structure learning module. The multi-source data acquisition and preprocessing module collects internal approval process data, external customer work order data, and personnel performance data, and performs data standardization, missing value imputation, and feature engineering. The improved PC algorithm causal structure learning module adopts an improved PC algorithm that integrates process time sequence constraints and hierarchical constraints to learn the causal structure between variables from the preprocessed data and construct a business process causal graph model. The middle-layer counterfactual reasoning layer includes a causal effect estimation module and a counterfactual pre-assessment module. The causal effect estimation module uses a backdoor adjustment formula, with the node resource input level as the processing variable and the total process time as the result variable, to estimate the average causal effect of each node. It calculates the bottleneck comprehensive score and ranks the nodes based on the product of the absolute value of the causal effect and the business importance weight. The counterfactual pre-assessment module simulates the expected effect and filters the expected effect through the three-step counterfactual reasoning of the structural causal model: tracing, intervention, and prediction. The upper-layer dynamic optimization layer includes a reinforcement learning scheduling module based on causal effects, an internal and external business collaborative optimization module, and a customer work order intelligent routing module. The reinforcement learning scheduling module uses the load of each node, the causal effect value, and the length of the pending queue as the state space, the resource allocation adjustment amount of each node as the action space and satisfies the total resource conservation, and uses the causal effect weighted waiting cost minus the adjustment cost as the reward function. The internal and external business collaborative optimization module incorporates internal approval process variables and external customer experience variables into the same causal graph to construct a unified causal model for internal and external businesses. The customer work order intelligent routing module predicts the expected effects of different processing paths based on the causal model.

[0007] Preferably, in the improved PC algorithm causal structure learning module, the calculation formula for integrating domain knowledge constraints is as follows: ; in, For the dataset, The significance level for the conditional independence test. It is a set of causal edges in domain knowledge that includes temporal and hierarchical constraints. This is the causal graph obtained from the final learning process.

[0008] Preferably, in the causal effect estimation module, the formula for calculating the average causal effect using the backdoor adjustment formula is as follows: ; In the formula, The variable is handled by the node resource investment level. The total process time is the result variable. To satisfy the backdoor criterion's set of promiscuous variables, To achieve a high level of resource input, For low resource input levels, This represents the average causal effect.

[0009] Preferably, the reward function of the reinforcement learning scheduling module is as follows: ; In the formula, For the first The causal effect value of each node, for Time of the first The length of the pending queue for each node. Adjust the vector for resource allocation. To adjust the cost weighting coefficient, for The reward value at any given moment.

[0010] Preferably, in the internal and external business collaboration optimization module, the calculation formula for the causal transmission effect of internal and external businesses is as follows: ; In the formula, It is a set of mediating variables between internal and external business operations. The average causal effect of internal business variables on mediating variables. The average causal effect of the mediating variable on external customer satisfaction. The overall transmission effect of internal business on external customer experience. As a set of mediating variables connecting internal processes and external customer experience, For the set of mediator variables A single mediator variable.

[0011] Preferably, in the intelligent routing module for customer work orders, the multi-objective decision formula for path selection is: ; In the formula, For the first The normalized score of the expected processing speed for each path. For the first The expected processing quality normalized score for each path. The first The normalized score of the expected processing cost of each path. for The weight, for The weight, for The weight, Number the optimal path.

[0012] Preferably, it also includes an adaptive dynamic permission adjustment module, which dynamically assesses trust levels based on user behavior data, and the trust level update formula is: ; In the formula, For users exist Trust level at all times This is the decay coefficient of historical trust. A security score based on the user's recent behavior.

[0013] Preferably, it also includes a full-link intelligent auditing module, which constructs normal behavior patterns based on a cause-effect graph, and the abnormal score is calculated using the following formula: ; In the formula, The operation sequence to be detected. For the set of variables in a cause-effect graph, For variables The set of parent nodes, For operation sequence Abnormal scores, For variables The set of parent nodes, In the operation sequence to be detected Under the condition of a given parent node Time variable The conditional probability distribution, In the normal behavior pattern, given the parent node Time variable The conditional probability distribution; An anomaly score is calculated by comparing the conditional distribution under the operation sequence with the conditional distribution under normal mode. When the anomaly score exceeds the alarm threshold, a security alarm is triggered.

[0014] This invention also provides a method for optimizing internal and external collaborative OA processes based on root cause analysis, including the following steps: S1. Collect internal approval process data, external customer work order data, and personnel performance data; perform data standardization, missing value filling, and feature engineering. S2. An improved PC algorithm that integrates process time sequence constraints and hierarchical constraints is adopted to learn the causal structure between variables from the preprocessed data and construct a business process causal graph model. S3. Using the backdoor adjustment formula, with the node resource input level as the processing variable and the total process time as the result variable, estimate the average causal effect of each node, calculate the bottleneck comprehensive score based on the product of the absolute value of the causal effect and the business importance weight, and identify bottleneck nodes by sorting them by score. S4. Through the three-step counterfactual reasoning of a structural causal model—namely, causation, intervention, and prediction—simulate the expected effects of different optimization schemes, calculate the input-output ratio, and select the optimal optimization strategy. S5. A reinforcement learning algorithm is adopted, with the load of each node, the causal effect value, and the length of the queue to be processed as the state space, the resource allocation adjustment amount of each node as the action space and satisfying the total resource conservation, and the causal effect weighted waiting cost minus the adjustment cost as the reward function to dynamically adjust the human resource allocation. S6. Incorporate internal approval process variables and external customer experience variables into the same causal graph to construct a unified causal model for internal and external business. Identify the causal transmission path of internal and external business through mediating variables, and maximize the weighted objective of internal efficiency and external satisfaction under total resource constraints. S7. Based on the causal model, predict the expected effects of different processing paths and automatically plan the optimal processing path for each customer work order.

[0015] Therefore, the present invention employs the above-mentioned internal and external collaborative OA system and process optimization method based on root cause analysis. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) This invention introduces causal inference to build a business process causal graph, and relies on counterfactual reasoning to accurately locate the root cause of process bottlenecks. It solves the shortcomings of traditional OA which can only count correlations and cannot identify false associations. Compared with the method of simply relying on average time to determine bottlenecks, it can uncover hidden bottlenecks, avoid ineffective resource allocation, and improve the accuracy of bottleneck diagnosis.

[0016] (2) This invention builds a three-layer architecture of causal discovery, counterfactual reasoning, and dynamic optimization, integrates internal approval and external work orders to construct a unified causal model, and combines it with a causal reinforcement learning scheduling algorithm to dynamically allocate manpower and adaptively plan work order paths; unlike conventional load balancing, this solution is based on causal effects to target resources, which significantly improves resource utilization.

[0017] (3) This invention integrates a causal engine and an integrated OA architecture, retains traditional office functions, and has the ability to self-diagnose and iteratively optimize, adapting to various enterprise business scenarios. Compared with the solution of optimizing internal and external modules separately, it optimizes in a holistic causal perspective, balances internal office efficiency and external customer satisfaction, and reduces manual management costs.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the internal and external collaborative OA system and process optimization method based on root cause analysis of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0021] Example 1 like Figure 1 As shown, this embodiment addresses the bottleneck diagnosis scenario of an enterprise's internal approval process, and details the specific implementation of the underlying causal discovery layer and the middle counterfactual reasoning layer of the present invention.

[0022] S1, the multi-source data acquisition and preprocessing module, collects approval process data from the OA system database for the past 12 months, including approximately 50,000 process instances across three main categories: personnel approval, financial approval, and administrative approval. The collected feature variables include: process type, initiating department, handler's rank, number of nodes, time spent at each node, document amount, number of attachments, and number of returns, totaling 32 feature variables.

[0023] Perform the following preprocessing operations on the collected data: The Z-score standardization method is used to eliminate the dimensional differences between different features. The calculation formula is as follows: ; In the formula, For the first The first sample The original values ​​of each feature, For the first The mean of each feature, For the first The standard deviation of each feature These are the standardized feature values.

[0024] The K-nearest neighbor imputation algorithm is used to impute samples with missing values. The K value is set to 5, and the similarity measure is Euclidean distance, calculated as follows: ; in, In order to be with the first The most similar sample A set of neighbor samples, These are the feature values ​​after filling.

[0025] Derivative features are constructed, such as node time consumption ratio, average processing time, return rate, and material integrity, expanding the feature dimensions to 45.

[0026] S2. The improved PC algorithm causal structure learning module uses an improved PC algorithm that incorporates domain knowledge constraints to learn causal structures. The specific steps are as follows: S201. In the process, the variable of the preceding node cannot be the result of the variable of the following node. That is, the causal direction can only flow from the preceding node to the following node. This is determined by the temporal nature of the approval process. The variables of higher-level departments are the causes of the corresponding variables of lower-level departments. That is, the higher the department level, the greater the impact on the overall process. This is determined by the hierarchical nature of the organizational structure.

[0027] S202. Based on the traditional PC algorithm, domain knowledge constraints are introduced as hard constraints, and the calculation formula is as follows: ; in, For the preprocessed dataset, The significance level for the conditional independence test is set to 0.05, and the maximum size of the condition set is set to 5. It is a set of causal edges in domain knowledge that includes temporal and hierarchical constraints. This is the causal graph obtained from the final learning process.

[0028] S203. Finally, a business process causal graph model containing 32 core nodes and 47 directed edges is constructed, which accurately reflects the causal relationship between various process variables.

[0029] S3. The causal effect estimation module uses a backdoor adjustment formula to estimate the average causal effect of each node. The specific steps are as follows: S301, Define processing variables The node resource investment level is divided into two levels: high and low. High resource investment means adding one processing personnel, while low resource investment means normal configuration. Define result variables Total process time, in hours; Define a set of mixed variables Confounding variables that satisfy the backdoor criteria include process type, initiating department, document amount, etc.

[0030] S302. Calculate the average causal effect of each node using the backdoor adjustment formula: ; In the formula, To achieve a high level of resource input, For low resource input levels, The average causal effect represents the magnitude of the causal impact of changes in resource input at this node on the overall process time.

[0031] S303. Calculate the bottleneck comprehensive score for each node by combining the business importance weights. : ; In the formula, For the first The average absolute value of the causal effect at each node This represents the business importance weight of the node. The business importance weight is set according to the company's strategy, with a weight of 1.8 for financial approval nodes, 1.5 for human resources approval nodes, and 1.0 for administrative approval nodes.

[0032] S304. Based on the comprehensive bottleneck score, sort from high to low and identify the top 3 bottleneck nodes: financial audit node, department head approval node, and material integrity verification node.

[0033] S4. The counterfactual pre-evaluation module simulates the expected effects of different optimization schemes through three steps of counterfactual reasoning. The specific steps are as follows: S401. Update the posterior distribution of exogenous variables based on the observed data, that is, infer the value distribution of each exogenous variable based on the actual observed process data. S402. Perform the do operation on the target node, that is, simulate the implementation of different optimization schemes on the node, such as increasing manpower, simplifying processes, upgrading tools, training personnel, etc. S403. Based on the modified causal model and the updated exogenous variables, calculate the counterfactual value of the outcome variable, that is, predict the change in process time after the implementation of the optimization scheme.

[0034] Calculate the input-output ratio of each optimization scheme: ; In the formula, For the first The expected improvement in process efficiency from the optimization plan is measured in hours per order. The implementation cost of this plan is expressed in person-days.

[0035] Based on the input-output ratio, the optimal optimization strategy is selected. In practical applications, the optimization scheme for the material integrity verification node has the highest ROI because the causal effect of this node is large while the optimization cost is low. By adding a material pre-inspection function, the subsequent return rate can be significantly reduced.

[0036] Example 2 This embodiment describes in detail the specific implementation of the reinforcement learning scheduling module in the upper dynamic optimization layer of the present invention, specifically for the scenario of dynamic human resource scheduling.

[0037] The reinforcement learning scheduling module based on causal effects constructs a reinforcement learning environment, specifically defined as follows: The state space is defined as the real-time state vector of each node: ; In the formula, for The system state vector of the learning agent is constantly reinforced. for The load vector of each node at any given time represents the number of tasks currently being processed by each node; This is the causal effect vector of each node, representing the degree of influence of each node on the overall efficiency; This is a vector representing the length of the queue to be processed for each node. There are a total of 8 approval nodes, with a state dimension of 24.

[0038] Action space is defined as the amount of resource allocation adjustment for each node: ; In the formula, for Time-based full node resource allocation adjustment vector For the first The manpower adjustment for each node, with an adjustment granularity of 0.5 people, satisfies the total resource conservation constraint. ; This means that the total human resources remain unchanged, but are dynamically allocated between different nodes.

[0039] The reward function is designed as a causal effect-weighted waiting cost minus the adjustment cost: ; In the formula, the first term is the weighted waiting cost. The greater the causal effect of a node, the higher the waiting cost. Therefore, the algorithm will prioritize reducing the queue length of nodes with large causal effects. The second term is the adjustment cost, which avoids the management overhead caused by frequent adjustments. For the first The causal effect value of each node, for Time of the first The length of the pending queue for each node. To adjust the cost weighting coefficient, we set it to 0.1. for The reward value at any given moment.

[0040] The deep Q-network algorithm was used for training. The neural network structure consisted of a 24-dimensional input layer, two hidden layers each with 64 neurons, and an output layer corresponding to the action space. The training data was generated from historical process data in a simulated environment. The training epochs were 5000, the experience replay pool size was 10000, and the target network was updated every 100 steps.

[0041] After training, the system is deployed to the production environment, updating its status and executing scheduling decisions every 5 minutes. The scheduling decisions are executed as follows: the processing personnel configuration of each node is adjusted according to the action vector, and idle personnel are redeployed from low-load, low-causal-effect nodes to high-load, high-causal-effect nodes.

[0042] Example 3 This embodiment describes in detail the specific implementation of the internal and external business collaboration optimization module of the present invention, specifically for the internal and external business collaboration optimization scenario.

[0043] The internal and external business collaboration optimization module incorporates internal approval process data and external customer work order data into a unified data framework, and constructs a unified causal model that includes internal and external variables.

[0044] The internal variables include 18 variables such as: time spent at each approval stage, approval rate, resource input level, and material return rate; External variables include 12 variables such as customer work order response time, problem resolution rate, customer satisfaction score, and customer complaint rate.

[0045] Using the same improved PC algorithm as in Example 1, a unified cause-effect graph of internal and external business with 30 variables is constructed.

[0046] By analyzing the mediation effect, the causal transmission path between internal and external business operations is identified, and the total transmission effect is calculated. ; In the formula, It is a set of mediating variables between internal and external business operations. The average causal effect of internal business variables on mediating variables. The average causal effect of the mediating variable on external customer satisfaction. The overall transmission effect of internal business on external customer experience. As a set of mediating variables connecting internal processes and external customer experience, For the set of mediator variables A single mediator variable.

[0047] Three main causal transmission pathways were identified: Path 1: Approval efficiency → Work order response time → Customer satisfaction, with a transmission effect value of 0.35; Path 2: Approval accuracy → Problem resolution rate → Customer satisfaction, with a transmission effect value of 0.42; Path 3: Approval transparency → Customer trust → Customer satisfaction, with a transmission effect value of 0.23.

[0048] Construct an objective function for optimizing internal and external business collaboration: ; ; in, For internal process efficiency indicators. As an indicator of external customer satisfaction. Weights are allocated specifically for internal efficiency. Weighting is assigned specifically to external satisfaction. Due to total resource constraints, As constraints, Assigned to the Resource quantity of each business node This represents the total number of nodes in the business process. In this embodiment... Take 0.6, The value is set to 0.4, which can be adjusted according to the company's strategy.

[0049] For external customer work order distribution scenarios, the customer work order intelligent routing module uses multi-objective decision-making to select the optimal processing path. The multi-objective decision-making formula for path selection is: ; In the formula, For the first The normalized score of the expected processing speed for each path. For the first The expected processing quality normalized score for each path. The first The normalized score of the expected processing cost of each path. for The weight, for The weight, for The weight, Number the optimal path.

[0050] The optimal resource allocation scheme is solved using a gradient descent algorithm based on causal effects. The resource allocation ratio among internal nodes is adjusted according to the magnitude of the transmission effect of each internal variable on external satisfaction, maximizing the weighted sum of internal efficiency and external satisfaction under total resource constraints.

[0051] Example 4 This embodiment details the specific implementation of the adaptive permission dynamic adjustment module and the full-link intelligent audit module of the present invention.

[0052] The adaptive permission dynamic adjustment module dynamically assesses trust levels and adjusts permission levels based on user behavior data, specifically as follows: Collect user login behavior, operation behavior, data access behavior and other data, and extract behavioral characteristics in five dimensions: login anomaly rate, operation compliance rate, data access pattern anomaly degree, and sensitive data access frequency.

[0053] Update user trust using the exponential moving average method: ; In the formula, For users exist Trust level at all times The historical trust decay coefficient is set to 0.9. A security score is given for the user's recent behavior, ranging from 0 to 100.

[0054] Trust levels are updated daily.

[0055] Permission levels are automatically adjusted: Set 5 permission levels, which will be automatically adjusted based on the comparison between trust level and threshold: When the trust level is higher than the upgrade threshold (85 points), the permission level is upgraded by one level; When the trust level falls below the downgrade threshold (60 points), the permission level is reduced by one level; When the level of trust is between the two, the permission level remains unchanged.

[0056] By using an adaptive dynamic permission adjustment mechanism, system security is ensured while avoiding the efficiency losses caused by static permissions. High-trust users can be granted higher privileges, improving business processing efficiency; low-trust users have restricted privileges, reducing security risks.

[0057] The end-to-end intelligent auditing module constructs normal behavior patterns based on cause-effect graphs and detects abnormal operations through deviation detection, specifically: Based on nearly three months of normal operation logs, a causal graph model of user behavior is constructed to learn the normal conditional probability distribution among various operation variables.

[0058] For the operation sequence to be detected Calculate its anomaly score: ; In the formula, The operation sequence to be detected. For the set of variables in a cause-effect graph, For variables The set of parent nodes, For operation sequence Abnormal scores, For variables The set of parent nodes, In the operation sequence to be detected Under the condition of a given parent node Time variable The conditional probability distribution, In the normal behavior pattern, given the parent node Time variable The conditional probability distribution.

[0059] The anomaly score represents the magnitude of the difference between the conditional distribution of the operation sequence and the conditional distribution of the normal pattern; a larger difference indicates a higher degree of anomaly. Conditional probability estimation uses the kernel density estimation method.

[0060] When the abnormal score exceeds the alarm threshold When this occurs, a security alarm is triggered. The alarm threshold is dynamically adjusted based on the target false alarm rate, which is kept below 1%.

[0061] Compared to traditional rule-matching auditing, the causal pattern auditing of this invention can detect unknown and novel anomalous operations, significantly improving security protection capabilities. Traditional rules can only detect known violation patterns, while this invention, by learning the causal patterns of normal behavior, can identify any anomalous behavior that deviates from the normal pattern.

[0062] Therefore, this invention adopts the above-mentioned internal and external collaborative OA system and process optimization method based on root cause analysis. By constructing a three-layer technical system of causal discovery, counterfactual reasoning, and dynamic optimization, it realizes intelligent diagnosis of process bottlenecks, dynamic scheduling of resources, and collaborative optimization of internal and external business, thereby improving the operating efficiency and intelligence level of the OA system.

[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An internal and external collaborative OA system based on root cause analysis, characterized in that: It includes a three-layer technical architecture: a bottom-layer causal discovery layer, a middle-layer counterfactual reasoning layer, and an upper-layer dynamic optimization layer; The underlying causal discovery layer includes a multi-source data acquisition and preprocessing module and an improved PC algorithm causal structure learning module. The multi-source data acquisition and preprocessing module collects internal approval process data, external customer work order data, and personnel performance data, and performs data standardization, missing value imputation, and feature engineering. The improved PC algorithm causal structure learning module adopts an improved PC algorithm that integrates process time sequence constraints and hierarchical constraints to learn the causal structure between variables from the preprocessed data and construct a business process causal graph model. The middle-layer counterfactual reasoning layer includes a causal effect estimation module and a counterfactual pre-assessment module. The causal effect estimation module uses a backdoor adjustment formula, with the node resource input level as the processing variable and the total process time as the result variable, to estimate the average causal effect of each node. It calculates the bottleneck comprehensive score and ranks the nodes based on the product of the absolute value of the causal effect and the business importance weight. The counterfactual pre-assessment module simulates the expected effect and filters the expected effect through the three-step counterfactual reasoning of the structural causal model: tracing, intervention, and prediction. The upper-layer dynamic optimization layer includes a reinforcement learning scheduling module based on causal effects, an internal and external business collaborative optimization module, and a customer work order intelligent routing module. The reinforcement learning scheduling module uses the load of each node, the causal effect value, and the length of the pending queue as the state space, the resource allocation adjustment amount of each node as the action space and satisfies the total resource conservation, and uses the causal effect weighted waiting cost minus the adjustment cost as the reward function. The internal and external business collaborative optimization module incorporates internal approval process variables and external customer experience variables into the same causal graph to construct a unified causal model for internal and external businesses. The customer work order intelligent routing module predicts the expected effects of different processing paths based on the causal model.

2. The internal and external collaborative OA system based on root cause analysis according to claim 1, characterized in that, In the improved PC algorithm causal structure learning module, the calculation formula for integrating domain knowledge constraints is as follows: ； in, For the dataset, The significance level for the conditional independence test. It is a set of causal edges in domain knowledge that includes temporal and hierarchical constraints. This is the causal graph obtained from the final learning process.

3. The internal and external collaborative OA system based on root cause analysis according to claim 2, characterized in that, In the causal effect estimation module, the formula for calculating the average causal effect using the backdoor adjustment formula is as follows: ； In the formula, The variable is handled by the node resource investment level. The total process time is the result variable. To satisfy the backdoor criterion's set of promiscuous variables, To achieve a high level of resource input, For low resource input levels, This represents the average causal effect.

4. The internal and external collaborative OA system based on root cause analysis according to claim 3, characterized in that, The reward function of the reinforcement learning scheduling module is as follows: ； In the formula, For the first The causal effect value of each node, for Time of the first The length of the pending queue for each node. Adjust the vector for resource allocation. To adjust the cost weighting coefficient, for The reward value at any given moment.

5. The internal and external collaborative OA system based on root cause analysis according to claim 4, characterized in that, In the internal and external business collaboration optimization module, the calculation formula for the causal transmission effect of internal and external businesses is as follows: ； In the formula, It is a set of mediating variables between internal and external business operations. The average causal effect of internal business variables on mediating variables. The average causal effect of the mediating variable on external customer satisfaction. The overall transmission effect of internal business on external customer experience. A set of mediating variables connecting internal processes and external customer experience. For the set of mediator variables A single mediator variable.

6. The internal and external collaborative OA system based on root cause analysis according to claim 5, characterized in that, In the intelligent routing module for customer work orders, the multi-objective decision-making formula for path selection is: ； In the formula, For the first The normalized score of the expected processing speed for each path. For the first The expected processing quality normalized score for each path. The first The normalized score of the expected processing cost of each path. for The weight, for The weight, for The weight, Number the optimal path.

7. The internal and external collaborative OA system based on root cause analysis according to claim 1, characterized in that, It also includes an adaptive dynamic permission adjustment module, which dynamically assesses trust levels based on user behavior data. The trust level update formula is: ； In the formula, For users exist Trust level at all times This is the decay coefficient of historical trust level. A security score based on the user's recent behavior.

8. The internal and external collaborative OA system based on root cause analysis according to claim 1, characterized in that, It also includes a full-link intelligent auditing module, which constructs normal behavior patterns based on cause-effect graphs, and the anomaly score is calculated using the following formula: ； In the formula, The operation sequence to be detected. For the set of variables in a cause-effect graph, For variables The set of parent nodes, For operation sequence Abnormal scores, For variables The set of parent nodes, In the operation sequence to be detected Under the condition of a given parent node Time variable The conditional probability distribution, In normal behavior mode, given a parent node Time variable The conditional probability distribution; An anomaly score is calculated by comparing the conditional distribution under the operation sequence with the conditional distribution under normal mode. When the anomaly score exceeds the alarm threshold, a security alarm is triggered.

9. A method for optimizing internal and external collaborative OA processes based on root cause analysis, applied to any one of claims 1 to 5, characterized in that, Includes the following steps: S1. Collect internal approval process data, external customer work order data, and personnel performance data; perform data standardization, missing value filling, and feature engineering. S2. An improved PC algorithm that integrates process time sequence constraints and hierarchical constraints is adopted to learn the causal structure between variables from the preprocessed data and construct a business process causal graph model. S3. Using the backdoor adjustment formula, with the node resource input level as the processing variable and the total process time as the result variable, estimate the average causal effect of each node, calculate the bottleneck comprehensive score based on the product of the absolute value of the causal effect and the business importance weight, and identify bottleneck nodes by sorting them by score. S4. Through the three-step counterfactual reasoning of a structural causal model—namely, causation, intervention, and prediction—simulate the expected effects of different optimization schemes, calculate the input-output ratio, and select the optimal optimization strategy. S5. A reinforcement learning algorithm is adopted, with the load of each node, the causal effect value, and the length of the queue to be processed as the state space, the resource allocation adjustment amount of each node as the action space and satisfying the total resource conservation, and the causal effect weighted waiting cost minus the adjustment cost as the reward function to dynamically adjust the human resource allocation. S6. Incorporate internal approval process variables and external customer experience variables into the same causal graph to construct a unified causal model for internal and external business. Identify the causal transmission path of internal and external business through mediating variables, and maximize the weighted objective of internal efficiency and external satisfaction under total resource constraints. S7. Based on the causal model, predict the expected effects of different processing paths and automatically plan the optimal processing path for each customer work order.

10. A computer-readable medium, characterized in that, The computer-readable medium stores computer program code that, when executed on a computer, causes the computer to perform the method as described in claim 9.

Images