Intelligent work order scheduling method and system driven by mechanism and data fusion

Abstract

The invention relates to the technical field of equipment asset management, in particular to a mechanism and data fusion driven intelligent work order scheduling method and system, and the method comprises the steps: constructing a mechanism model; training an AI prediction model by using historical data; acquiring a current work order pool, real-time states of all personnel and environment information; according to the constructed constraint condition set, preliminarily screening available personnel maintained in the personnel dynamic ability map to obtain a candidate personnel set; calling a trained work order complexity prediction model and a personnel performance prediction model, and predicting the predicted processing time of each work order and the expected result of processing each work order by each candidate personnel in combination with the analyzed work order attributes and the candidate personnel set; and inputting the current work order pool, the candidate person set, the predicted processing time, the expected result and the multi-objective optimization function into an AI scheduling engine, and outputting an optimal work order-person allocation scheme. The work order response time is shortened, and the resource utilization rate is improved.

Description

Technical Field

[0001] This application relates to the field of equipment asset management technology, specifically to an intelligent work order scheduling method and system driven by the fusion of mechanism and data. Background Technology

[0002] Traditional work order dispatching models heavily rely on dispatchers' manual experience, making it difficult to systematically coordinate multiple dynamic factors such as work order urgency, estimated time, required skills, spare parts availability, personnel location, real-time load, and performance levels. This results in low dispatching efficiency, response delays, and uneven resource allocation. Existing automated work order dispatching systems are mostly based on static rules or simple priority algorithms, failing to effectively handle the complex matching relationship between work order characteristics and personnel capabilities. They lack the ability to integrate and adaptively optimize contextual information such as equipment failure prediction, personnel performance fluctuations, and real-time traffic conditions. Consequently, they perform poorly in scenarios involving sudden failures, parallel task scheduling, and multi-objective collaborative optimization, becoming a bottleneck restricting the improvement of equipment asset management efficiency.

[0003] The massive historical work order data, personnel operation records, and equipment operation status data accumulated in the current system provide a solid foundation for building an intelligent scheduling engine that integrates business mechanisms and data-driven approaches. There is an urgent need for a solution that can deeply integrate domain knowledge and artificial intelligence algorithms to achieve precise, dynamic, and optimized work order scheduling. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an intelligent work order scheduling method and system driven by the fusion of mechanism and data. This method deeply integrates the characteristics of equipment maintenance services, enabling efficient and accurate work order dispatch, improving the efficiency and service quality of equipment asset management, and reducing operation and maintenance costs.

[0005] In a first aspect, the present invention provides an intelligent work order scheduling method driven by the fusion of mechanism and data, comprising the following steps: S1. Collect historical data from the equipment asset management system, including work order data, personnel data, equipment data, spare parts inventory data, and geographic information data; S2. Construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; S3. Train an AI prediction model using historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. S4. In response to the generation of a new work order or a system status change event, obtain the current work order pool, the real-time status of all personnel, and environmental information. S5. Based on the constructed set of constraints, perform a preliminary screening of the available personnel maintained in the personnel dynamic capability map to obtain a set of candidate personnel. S6. Call the trained work order complexity prediction model and personnel performance prediction model, and combine the work order feature attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. S7. Input the current work order pool, candidate set, predicted processing time and expected results, and the constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on the reinforcement learning algorithm. S8. Execute the work order-personnel allocation scheme to dispatch the work order to the terminal of the corresponding personnel.

[0006] As a preferred embodiment of the technical solution of the present invention, the step of constructing the mechanism model in S2 includes: S21. Define a set of characteristic attributes for each type of work order. The characteristic attributes include at least: work order type, urgency level, estimated time consumption, required skill chain, required tool or spare parts list, geographical location, planned time window, associated equipment identifier, security level requirements, and customer importance level. S22. Establish a dynamically updated capability map for each operations and maintenance personnel. The capability map shall include at least: a skill tag library, the proficiency level corresponding to each skill tag, real-time GPS location, current work status, historical performance indicators, current workload quantification value, and personal work preferences. S23. Define the hard constraints that must be met and the soft constraints that are expected to be met in the scheduling decision based on business rules; The hard constraints include: skill matching constraints, security compliance constraints, spare parts availability constraints, geographical accessibility constraints, and time conflict constraints. The soft constraints include: skill proficiency matching constraints, commuting time optimization constraints, workload balancing constraints, and personnel preference matching constraints. S24. Construct a comprehensive optimization objective function F to quantitatively evaluate the quality of the scheduling scheme. The objective function F is a weighted sum of multiple sub-objectives, expressed as:

[0007] In the formula, Total completion time For total commuting costs, To improve staff utilization, First-time repair rate refers to the percentage of work orders in which an employee successfully resolves a fault on their first execution out of all their work orders. These are the weighting coefficients for each sub-objective.

[0008] By introducing multi-objective optimization functions and a system of soft and hard constraints, the system can intelligently balance multiple objectives such as efficiency, cost, quality, and safety. This ensures the rigid execution of key business rules while optimizing flexible objectives, thus meeting the diversified management needs of complex industrial scenarios.

[0009] The introduction of the mechanistic model makes the AI ​​decision-making process highly interpretable, allowing dispatchers to clearly understand the logic behind order dispatching and thus increasing their trust in the system's recommendations. Simultaneously, the system supports human intervention and feedback, integrating expert experience into the automated dispatching process and achieving a new intelligent operation and maintenance model of human-machine collaboration.

[0010] As a preferred embodiment of the technical solution of the present invention, training the AI ​​prediction model in S3 includes: S31. Train the work order complexity prediction model, including the following steps: The text descriptions, equipment types, and fault codes in historical work order data are collected as training features, and the actual processing time of historical work orders is used as training labels. Vectorized feature extraction is performed on the text description, and then fused with the features of device type and fault code to generate a standardized feature vector; A regression model is trained based on the fused standardized feature vectors to obtain a work order complexity prediction model, which is used to predict the expected processing time of new work orders. S32. Training personnel performance prediction model, including the following steps: Construct a training sample set, where each sample's features include: personnel skill features, features of the work orders processed, and environmental context features; The labels for the training sample set are the actual processing results of the corresponding historical work orders, including success / failure labels, actual processing time, and customer satisfaction rating. The model is trained using machine learning classification and regression algorithms to obtain a personnel performance prediction model, which is used to predict the expected performance of a specified person in handling a specified work order, including the probability of first-time repair, the estimated time, and the expected satisfaction.

[0011] As a preferred embodiment of the technical solution of the present invention, step S4 includes: S41. Continuously monitor various events from the equipment asset management system, including: new work order creation event, work order status change event, personnel status change event, equipment alarm event, and timer trigger event; S42. In response to any of the events mentioned above, obtain all currently unassigned and uncompleted work orders and form the current work order pool; S43. Obtain the latest real-time status of all operation and maintenance personnel from the personnel dynamic capability map. The latest real-time status includes at least the current skill proficiency, GPS location, work status and current workload. S44. Call external services through the application programming interface to obtain the latest environmental information, which includes at least real-time traffic data and spare parts warehouse inventory status.

[0012] As a preferred embodiment of the technical solution of the present invention, step S5 includes: S51. Iterate through all available maintenance personnel, and for each work order to be dispatched, apply the hard constraint rules defined in the constraint condition set to perform a feasibility check on the personnel-work order pairing. S52. Retain the personnel who pass all hard constraint rule verifications to form the candidate personnel set for this work order; The hard constraint rules include at least the following: Verify whether the skill tags of the personnel in the dynamic competency map completely cover the required skill chain defined in the work order feature model for that work order; Based on the real-time location of personnel and the geographical location of the work order, combined with real-time traffic conditions, it verifies whether personnel can arrive at the site within the time window required by the work order. Check whether the time period of the task currently assigned by the verification personnel overlaps with the estimated processing time period of the work order; For the critical spare parts specified in the work order, verify whether the spare parts inventory that the person can currently apply for or the inventory status of the warehouse he is responsible for meets the requirements; Verify whether the personnel's qualifications meet the security level requirements defined in the work order, and whether the specific operation meets the required number of personnel.

[0013] As a preferred embodiment of the technical solution of the present invention, step S6 includes: S61. For each work order in the current work order pool, extract the corresponding feature values ​​from the work order data according to the work order feature attributes to form the standardized feature vector of the work order. S62. Input the standardized feature vector of the work order into the trained work order complexity prediction model and output the predicted processing time of the work order. S63. For each person in the candidate set who passed the initial screening, perform the following operations: a. Obtain the current status and capability feature vector of the person from the personnel dynamic capability map; b. Concatenate and fuse the feature vectors of personnel with the standardized feature vectors of work orders to be dispatched to form a personnel-work order paired feature vector; c. Input the personnel-work order pairing feature vector into the trained personnel performance prediction model, and output the expected result of the personnel handling this work order. The expected result includes at least the predicted first repair probability and the predicted processing time.

[0014] As a preferred embodiment of the technical solution of the present invention, step S7 includes: S71. Represent the current system state as a state feature vector. This state feature vector integrates the features of all work orders in the current work order pool, the state and capability features of all personnel in the candidate personnel set, the predicted processing time and performance, and environmental information; S72, Based on the current state The included work order pool and candidate set will handle every possible work order dispatch decision. Defined as assigning a work order from the work order pool to a person in the candidate set, all possible assignment actions constitute the decision action space A; S73. Execute a dispatch action in a simulation environment. Then, an immediate reward value is calculated based on the multi-objective optimization function F. This reward value is used to quantify actions. In state The immediate results achieved in terms of comprehensive optimization goals; S74. The AI ​​scheduling engine, as a reinforcement learning agent, is in a state... The goal is to select an action in the decision action space A according to policy π. To maximize the expected value of the future cumulative discount reward Gt, the final output can be achieved in the current state. Find the optimal work order-personnel allocation scheme that maximizes Gt.

[0015] In the formula, γ This is a discount factor, with a value range of 0 ≤ γ ≤1, The reward obtained at time step t+k. E [•] represents the expected value, which is the average value over all possible states and actions.

[0016] As a preferred embodiment of the technical solution of the present invention, the method further includes: S9. Collect the actual processing data after the work order is completed, and send the actual processing data back to the database; use the sent actual processing data to update and optimize the AI ​​prediction model and the AI ​​scheduling engine.

[0017] Leveraging a closed-loop feedback mechanism, the system can collect real-time data on actual work order processing and use this data to incrementally optimize the prediction model and scheduling strategy. This continuous learning mechanism enables the system to dynamically adapt to business changes, maintain optimal scheduling performance, and possess strong long-term adaptability.

[0018] As a preferred embodiment of the technical solution of the present invention, step S9 includes: S91. Collect the actual processing data after the work order is completed and send it back to the central database; the actual processing data includes: the actual start time of the work order, the actual completion time, the actual working hours consumed, the actual spare parts list used, the final status of the work order, the root cause code of the fault, and the on-site difficulty score reported by the personnel. S92. Associate the actual processed data returned with the original dispatch records in the database. The dispatch records include the work order characteristics, personnel capability status and AI model prediction values ​​used when dispatching the work order, thereby forming training samples labeled with real results. S93. Periodically or triggered, newly formed training samples are added to the training set to perform incremental learning or fine-tuning on the work order complexity prediction model and the personnel performance prediction model, so as to reduce the deviation between the model prediction value and the actual value. S94. The associated data pairs are used as experience samples and stored in the experience replay buffer of the AI ​​scheduling engine. The AI ​​scheduling engine periodically samples batch experiences from the buffer, calculates the temporal difference error between the actual reward and the expected reward, and updates its internal policy network and value network parameters through the backpropagation algorithm.

[0019] Secondly, the technical solution of the present invention provides an intelligent work order scheduling system driven by mechanism and data fusion, comprising: The data acquisition module is used to collect historical data from the equipment asset management system. The historical data includes work order data, personnel data, equipment data, spare parts inventory data, and geographic information data. The mechanism model construction module is used to construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; The AI ​​prediction model training module is used to train an AI prediction model using historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. The system status acquisition module is used to obtain the current work order pool, the real-time status of all personnel, and environmental information in response to the generation of new work orders or system status change events. The candidate screening module is used to perform preliminary screening of available personnel maintained in the personnel dynamic capability map based on the constructed set of constraints, and obtain a set of candidate personnel. The prediction module is used to call the trained work order complexity prediction model and personnel performance prediction model, and combine the work order feature attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. The AI ​​scheduling engine module is used to input the current work order pool, candidate set, predicted processing time and expected results, and the constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on reinforcement learning algorithm. The work order execution module is used to execute the work order-personnel allocation plan and dispatch the work order to the corresponding personnel's terminal.

[0020] As can be seen from the above technical solutions, this application has the following advantages: Through precise modeling of work order feature models and personnel dynamic capability maps, combined with a rapid candidate set screening mechanism under multiple constraints, the time cost and subjective errors of manual scheduling are significantly reduced. The AI ​​scheduling engine, based on reinforcement learning, performs global optimization decisions and can output the optimal dispatch plan within milliseconds, effectively shortening work order response time and improving the overall utilization rate of personnel and equipment resources. Through work order complexity prediction models and personnel performance prediction models, the system can deeply mine the implicit patterns in historical data, predict the time, success rate, and satisfaction of different personnel in handling specific work orders, thereby achieving refined and personalized dispatching that surpasses human experience, improving the first-time repair rate and customer satisfaction. Attached Figure Description

[0021] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the method provided in an embodiment of the present invention.

[0023] Figure 2 This is a system architecture diagram provided for an embodiment of the present invention.

[0024] Figure 3 This is a flowchart of the intelligent scheduling engine workflow. Detailed Implementation

[0025] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0027] like Figure 1 As shown, this embodiment of the invention provides an intelligent work order scheduling method driven by mechanism and data fusion, including the following steps: S1. Collect historical data from the equipment asset management system, including work order data, personnel data, equipment data, spare parts inventory data, and geographic information data; S2. Construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; The steps involved in constructing the mechanistic model in this process include: S21. Define a set of characteristic attributes for each type of work order. The characteristic attributes include at least: work order type, urgency level, estimated time consumption, required skill chain, required tool or spare parts list, geographical location, planned time window, associated equipment identifier, security level requirements, and customer importance level. S22. Establish a dynamically updated capability map for each operations and maintenance personnel. The capability map shall include at least: a skill tag library, the proficiency level corresponding to each skill tag, real-time GPS location, current work status, historical performance indicators, current workload quantification value, and personal work preferences. S23. Define the hard constraints that must be met and the soft constraints that are expected to be met in the scheduling decision based on business rules; The hard constraints include: skill matching constraints, security compliance constraints, spare parts availability constraints, geographical accessibility constraints, and time conflict constraints. The soft constraints include: skill proficiency matching constraints, commuting time optimization constraints, workload balancing constraints, and personnel preference matching constraints. S24. Construct a comprehensive optimization objective function F to quantitatively evaluate the quality of the scheduling scheme. The objective function F is a weighted sum of multiple sub-objectives, expressed as:

[0028] In the formula, Total completion time For total commuting costs, To improve staff utilization, First-time repair rate refers to the percentage of work orders in which an employee successfully resolves a fault on their first execution out of all their work orders. These are the weighting coefficients for each sub-objective.

[0029] For each dispatched work order, its completion time = actual completion time - creation time. These time points are obtained directly from the system logs. Total completion time refers to the sum of the completion times of all work orders, or the sum of the completion times of all completed work orders within a certain period. The goal of the scheduling engine is to generate a dispatch plan that minimizes the predicted total completion time. Predicted work order processing times, commuting times, etc., are used to estimate the total completion time under different dispatch plans.

[0030] Total commuting cost is calculated by calling a map service API to obtain route planning results based on the real-time location of personnel and the geographical location of work orders, thus obtaining the estimated commuting time or distance, and converting the commuting time or distance into cost. The simplest way is to directly use time or distance as the cost unit. A more complex approach is to multiply by a cost coefficient. In a scheduling scheme, the total commuting cost is obtained by adding up the commuting costs incurred by all personnel due to work assignments.

[0031] Personnel utilization rate is the percentage of total effective working time of personnel to total working hours (e.g., 8-hour shifts) within a statistical period.

[0032] Personnel utilization rate = (Total effective working time / Total working hours) × 100% Effective working time can be calculated by summing the actual processing time of all work orders for a given person from historical data. When making decisions, the scheduling engine uses the estimated processing time output by the work order complexity prediction model to estimate the future effective working time of a person, thereby predicting the overall personnel utilization rate under different dispatch schemes.

[0033] After a work order is completed, the reporting personnel or the system marks whether the initial repair was successful by indicating the work order status (e.g., completed, requires secondary dispatch). This data is recorded and used to update historical performance indicators in the personnel dynamic competency graph. The personnel performance prediction model learns the relationship between an employee's historical FTFR and factors such as their skills and work order type, thereby predicting their probability of successfully repairing a new work order on the first attempt. The scheduling engine tends to assign work orders to personnel with a higher predicted probability of successful first attempt, thus maximizing the overall first attempt rate.

[0034] In this embodiment of the invention, the mechanism model module mainly constructs a work order dispatch mechanism model based on the rules, priority definitions, process requirements, and constraints of equipment maintenance operations. The mechanism model module includes work order characterization, personnel dynamic profiling, constraint definitions, and objective function definitions. Work order characterization is mainly used to analyze work order type (repair / maintenance / inspection), urgency, predicted time consumption, required skill chain, spare parts dependency, etc.; the personnel dynamic profiling integrates the maintenance personnel skill library, real-time location, personnel status (idle / working / resting), and historical performance (average processing time, first-time repair rate, historical work order evaluation). The constraint definitions include hard constraints and soft constraints. Hard constraints include skill matching (minimum skill requirements must be met), safety compliance (e.g., specific operations require two people), spare parts availability (work orders can only be dispatched when critical spare parts are available), geographical accessibility (distance / time), and time conflict (personnel can only handle one work order at a time). Soft constraints include skill proficiency matching (prioritizing dispatch to more skilled personnel), workload balancing, and personnel preference (e.g., assigning personnel familiar with the equipment whenever possible). The objective function definition mainly includes the following objectives: ① Minimize the total completion time of work orders / maximize the on-time completion rate; ② Minimize the total commuting time / distance of personnel; ③ Maximize personnel utilization / load balancing; ④ Maximize the first-time repair rate / work order satisfaction; ⑤ Minimize the response time of emergency work orders; ⑥ Minimize the overall operation and maintenance cost (Min(total cost) = .

[0035] S3. Train an AI prediction model using the historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. In this step, training the AI ​​prediction model includes: S31. Train the work order complexity prediction model, including the following steps: The text descriptions, equipment types, and fault codes in historical work order data are collected as training features, and the actual processing time of historical work orders is used as training labels. Vectorized feature extraction is performed on the text description, and then fused with the features of device type and fault code to generate a standardized feature vector; A regression model is trained based on the fused standardized feature vectors to obtain a work order complexity prediction model, which is used to predict the expected processing time of new work orders. S32. Training personnel performance prediction model, including the following steps: Construct a training sample set, where each sample's features include: personnel skill features, features of the work orders processed, and environmental context features; The labels for the training sample set are the actual processing results of the corresponding historical work orders, including success / failure labels, actual processing time, and customer satisfaction rating. The model is trained using machine learning classification and regression algorithms to obtain a personnel performance prediction model, which is used to predict the expected performance of a specified person in handling a specified work order, including the probability of first-time repair, the estimated time, and the expected satisfaction.

[0036] The workflow of the intelligent scheduling engine is as follows: Figure 3 As shown: Work order generation / triggering: This includes work orders automatically generated by the system according to the set plan, temporary work orders created manually, and work orders directly triggered by IoT device alarms or predictive maintenance.

[0037] Work order feature extraction and prediction: Predict work order complexity, required time, required spare parts, etc. based on work order information and equipment information.

[0038] Matching available operations and maintenance personnel: Quickly filter out a candidate set of available operations and maintenance personnel based on hard constraints (skills, location, status).

[0039] AI-optimized decision-making: The IA scheduling engine combines the current system status (all pending work orders, all candidate personnel status, and environmental information), business objectives, and soft constraints to calculate the optimal work order-personnel allocation scheme.

[0040] Dispatch execution and feedback: Send the dispatch results to the corresponding maintenance personnel (mobile APP) and record the dispatch decision.

[0041] Closed-loop learning and optimization: After a work order is completed, a historical record is generated, collecting the actual processing results (time spent, success rate, satisfaction level), on-site personnel feedback, and changes in the actual system status. This feedback is used to continuously train and optimize the AI ​​model, adjusting the parameters in the mechanistic model.

[0042] S4. In response to the generation of a new work order or a system status change event, obtain the current work order pool, the real-time status of all personnel, and environmental information; this step specifically includes: S41. Continuously monitor various events from the equipment asset management system, including: new work order creation event, work order status change event, personnel status change event, equipment alarm event, and timer trigger event; S42. In response to any of the events mentioned above, obtain all currently unassigned and uncompleted work orders and form the current work order pool; S43. Obtain the latest real-time status of all operation and maintenance personnel from the personnel dynamic capability map. The latest real-time status includes at least the current skill proficiency, GPS location, work status and current workload. S44. Call external services through the application programming interface to obtain the latest environmental information, which includes at least real-time traffic data and spare parts warehouse inventory status.

[0043] S5. Based on the constructed set of constraints, perform preliminary screening of the available personnel maintained in the personnel dynamic capability map to obtain a candidate personnel set; specifically including: S51. Iterate through all available maintenance personnel, and for each work order to be dispatched, apply the hard constraint rules defined in the constraint condition set to perform a feasibility check on the personnel-work order pairing. S52. Retain the personnel who pass all hard constraint rule verifications to form the candidate personnel set for this work order; The hard constraint rules include at least the following: Verify whether the skill tags of the personnel in the dynamic competency map completely cover the required skill chain defined in the work order feature model for that work order; Based on the real-time location of personnel and the geographical location of the work order, combined with real-time traffic conditions, it verifies whether personnel can arrive at the site within the time window required by the work order. Check whether the time period of the task currently assigned by the verification personnel overlaps with the estimated processing time period of the work order; For the critical spare parts specified in the work order, verify whether the spare parts inventory that the person can currently apply for or the inventory status of the warehouse he is responsible for meets the requirements; Verify whether the personnel's qualifications meet the security level requirements defined in the work order, and whether the specific operation meets the required number of personnel.

[0044] S6. Call the trained work order complexity prediction model and personnel performance prediction model, and combine the parsed work order attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. It should be noted here that this step specifically includes: S61. For each work order in the current work order pool, extract the corresponding feature values ​​from the work order data according to the work order feature attributes to form the standardized feature vector of the work order. S62. Input the standardized feature vector of the work order into the trained work order complexity prediction model and output the predicted processing time of the work order. S63. For each person in the candidate set who passed the initial screening, perform the following operations: a. Obtain the current status and capability feature vector of the person from the personnel dynamic capability map; b. Concatenate and fuse the feature vectors of personnel with the standardized feature vectors of work orders to be dispatched to form a personnel-work order paired feature vector; c. Input the personnel-work order pairing feature vector into the trained personnel performance prediction model, and output the expected result of the personnel handling this work order. The expected result includes at least the predicted first repair probability and the predicted processing time.

[0045] S7. Input the current work order pool, candidate set, predicted processing time, expected result, and constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on a reinforcement learning algorithm; further including: S71. Represent the current system state as a state feature vector. This state feature vector integrates the features of all work orders in the current work order pool, the state and capability features of all personnel in the candidate personnel set, the predicted processing time and performance, and environmental information; S72, Based on the current state The included work order pool and candidate set will handle every possible work order dispatch decision. Defined as assigning a work order from the work order pool to a person in the candidate set, all possible assignment actions constitute the decision action space A; S73. Execute a dispatch action in a simulation environment. Then, an immediate reward value is calculated based on the multi-objective optimization function F. This reward value is used to quantify actions. In state The immediate results achieved in terms of comprehensive optimization goals; S74. The AI ​​scheduling engine, as a reinforcement learning agent, is in a state... The goal is to select an action in the decision action space A according to policy π. To maximize the expected value of the future cumulative discount reward Gt, the final output can be achieved in the current state. Find the optimal work order-personnel allocation scheme that maximizes Gt.

[0046] In the formula, γ This is a discount factor, with a value range of 0 ≤ γ ≤1, The reward obtained at time step t+k. E [•] represents the expected value, which is the average value over all possible states and actions.

[0047] The specific process of policy search and decision-making by the AI ​​scheduling engine described in S74 based on reinforcement learning algorithm includes: based on the current policy network πθ, for the current state... Estimate the execution of any possible action The value of the subsequent action Qπ ( , Based on the policy evaluation results, a new policy πθ' that is better than the current policy is generated using a policy gradient ascent algorithm or a greedy policy based on the value function, so as to increase the expected cumulative reward. The policy evaluation and policy improvement cycle is repeated until the policy network converges or a preset computation time threshold is reached. The finally converged policy is then applied to the current state. It outputs its recommended optimal work order-personnel allocation scheme.

[0048] S8. Execute the work order-personnel allocation scheme to dispatch the work order to the terminal of the corresponding personnel.

[0049] S9. Collect the actual processing data after the work order is completed, and send the actual processing data back to the database; use the sent-back actual processing data to update and optimize the AI ​​prediction model and the AI ​​scheduling engine. Specifically, this includes: S91. Collect actual processing data after a work order is completed through various methods, including automatic reporting by maintenance personnel's mobile terminal APP, manual entry into the work order system, or automatic collection by IoT device sensors, and send it back to the central database. The actual processing data includes: the actual start time of the work order, the actual completion time, the actual working hours consumed, the actual spare parts list used, the final status of the work order, the root cause code of the fault, and the on-site difficulty score reported by the personnel. S92. Associate the actual processed data returned with the original dispatch records in the database. The dispatch records include the work order characteristics, personnel capability status and AI model prediction values ​​used when dispatching the work order, thereby forming training samples labeled with real results. S93. Periodically or triggered, newly formed training samples are added to the training set to perform incremental learning or fine-tuning on the work order complexity prediction model and the personnel performance prediction model, so as to reduce the deviation between the model prediction value and the actual value. S94. The associated data pairs are used as experience samples and stored in the experience replay buffer of the AI ​​scheduling engine. The AI ​​scheduling engine periodically samples batch experiences from the buffer, calculates the temporal difference error between the actual reward and the expected reward, and updates its internal policy network and value network parameters through the backpropagation algorithm.

[0050] The associated data is used to construct training tuples. ),in, The system state feature vector representing the moment of the dispatch decision; Indicates the dispatching action performed; This represents the actual reward value calculated based on the actual processing results of the work order, according to the multi-objective optimization function. This indicates the new state the system enters after the dispatch action is executed; The status is evaluated through a network of commentators within the AI ​​scheduling engine. value And based on the actual reward and subsequent states value Calculate the timing difference error δ The calculation formula is as follows: ,in, This is used as a discount factor and thus as a signal for strategy evaluation.

[0051] Using the aforementioned timing difference error δ The parameters of the actor network in the AI ​​scheduling engine are updated using a policy gradient algorithm. Its update volume is:

[0052] in, For learning rate, For actors network in status Select action The strategy probability. Simultaneously, the time-series difference error is used. δ As a loss signal, the parameters of the critic network are updated using gradient descent to improve its accuracy in predicting future state values.

[0053] In the framework of reinforcement learning, the role of the critic network is to evaluate the value of the current state. Specifically, the critic network evaluates the system state. value That is, in the state Below, the cumulative value of expected future rewards. This value function It is a scalar, representing the state. Begin by continuing with the current strategy and consider the expected total reward.

[0054] The value function output of the critic network Rewards for accumulated discounts in the future Gt The estimated expected value is used to assess the long-term value of a state or action, i.e.: .

[0055] Specifically, historical work order data, personnel information, equipment information, spare parts inventory, and geographical information are collected from the equipment asset management system, and then cleaned and organized to provide data support for subsequent model training and decision-making. For example, all work order data from a factory over the past year, including maintenance work orders and repair work orders, are collected, and missing or erroneous values ​​in the data are processed.

[0056] Based on the defined work order characteristics and personnel competency models, a mechanistic model is constructed. For example, for a certain type of equipment maintenance work order, its urgency level is clearly defined as high, medium, and low. The estimated time is estimated based on historical data. Required skills include electrical and mechanical maintenance skills. Required tools and spare parts are determined based on the equipment model and maintenance manual. Geographic location records the equipment's installation location. Time windows are set according to the equipment's operating requirements. Related equipment is considered, including other equipment related to the faulty equipment. Safety level requirements are determined based on the working environment. Customer importance is classified according to the importance of the department or production line to which the equipment belongs. For the personnel competency model, skill tags are established for each maintenance worker, such as the types of maintenance they are good at and the certifications they possess. Proficiency is calculated based on work experience and past repair success rates. Current location is obtained through a positioning system. Current status is categorized as idle, working, or resting. Historical performance includes indicators such as average processing time, first-time repair rate, and customer satisfaction. Workload is assessed based on the number and difficulty of assigned work orders. Preference records are kept of personnel's tendencies towards different regions and types of work orders.

[0057] AI models are trained using collected data, including work order complexity / time prediction models, personnel performance prediction models, fault prediction models, and dynamic scheduling optimization models. For example, using work order descriptions, equipment types, and fault codes from historical work order data as input and actual processing time as output, an NLP+regression model is trained to predict the complexity and time consumption of new work orders. For the personnel performance prediction model, personnel skills, experience, and historical work order processing records are used as input to predict their success rate, estimated time consumption, and customer satisfaction in handling specific types of work orders. The fault prediction model combines IoT device sensor data, such as temperature, pressure, and vibration data, and uses time-series prediction models such as LSTM or Transformer to predict potential equipment faults. The dynamic scheduling optimization model can employ reinforcement learning algorithms to model the scheduling environment as a Markov decision process. The state includes the current pool of pending work orders, personnel status and location, spare parts status, and time. The action is to assign a work order to a specific personnel, and the reward is calculated based on objective functions such as work order completion time, commuting time, and personnel utilization rate.

[0058] When a work order is generated or triggered, the system operates according to deep integration and engine workflow. For example, when equipment malfunctions, the system automatically generates a repair work order. First, it parses the work order description and, combined with equipment information, predicts complexity, time consumption, and required spare parts. Then, it filters out a candidate set of available personnel based on hard constraints (such as skill matching, location, and status). Next, the core scheduling engine (such as an RL Agent) combines the current system status (all pending work orders, all candidate personnel status, and environmental information), business objectives, and soft constraints to calculate the optimal work order-personnel allocation scheme. Finally, the dispatch result is sent to the corresponding personnel (mobile app), and the dispatch decision is recorded.

[0059] Continuously collect actual processing results and feedback information for closed-loop learning and optimization of parameters in AI and mechanistic models. For example, collect information such as actual work order processing results (time taken, success rate, satisfaction level), personnel feedback, and changes in the system's actual state. If it is found that a certain person frequently takes a long time to process a particular type of work order, the analysis suggests the reason might be a mismatch between the person's skills and the work order, or complex on-site conditions. Based on this feedback, adjust the skill tags or proficiency in the personnel competency model, and simultaneously optimize the prediction parameters in the AI ​​model to make subsequent work order dispatch more accurate.

[0060] This invention also provides an intelligent work order scheduling system driven by mechanism and data fusion, the system architecture of which is as follows: Figure 2 As shown, the system includes: The data acquisition module is used to collect historical data from the equipment asset management system. The historical data includes work order data, personnel data, equipment data, spare parts inventory data, and geographic information data. The mechanism model construction module is used to construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; The AI ​​prediction model training module is used to train an AI prediction model using historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. The system status acquisition module is used to obtain the current work order pool, the real-time status of all personnel, and environmental information in response to the generation of new work orders or system status change events. The candidate screening module is used to perform preliminary screening of available personnel maintained in the personnel dynamic capability map based on the constructed set of constraints, and obtain a set of candidate personnel. The prediction module is used to call the trained work order complexity prediction model and personnel performance prediction model, and combine the work order feature attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. The AI ​​scheduling engine module is used to input the current work order pool, candidate set, predicted processing time and expected results, and the constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on reinforcement learning algorithm. The work order execution module is used to execute the work order-personnel allocation plan and dispatch the work order to the corresponding personnel's terminal.

[0061] In some embodiments, the mechanism model building module includes: The work order characterization submodule is used to define a set of characteristic attributes for each type of work order. The characteristic attributes include at least: work order type, urgency level, estimated time consumption, required skill chain, required tool or spare parts list, geographical location, planned time window, associated equipment identifier, security level requirements, and customer importance level. The personnel dynamic capability graph submodule is used to establish a dynamically updated capability graph for each operations and maintenance personnel. The capability graph includes at least: a skill tag library, the proficiency of each skill tag, real-time GPS location, current work status, historical performance indicators, current workload quantification value, and personal work preferences. The constraint definition submodule is used to define the hard constraints that must be met and the soft constraints that are expected to be met in scheduling decisions based on business rules. The hard constraints include: skill matching constraints, safety compliance constraints, spare parts availability constraints, geographical accessibility constraints, and time conflict constraints. The soft constraints include: skill proficiency matching constraints, commuting time optimization constraints, workload balancing constraints, and personnel preference matching constraints. The multi-objective optimization function construction submodule is used to construct a comprehensive optimization objective function F, which is used to quantitatively evaluate the quality of the scheduling scheme. The objective function F is a weighted sum of multiple sub-objectives.

[0062] In some embodiments, the AI ​​prediction model training module includes: The work order complexity prediction model training submodule is used to collect text descriptions, equipment types, and fault codes from historical work order data as training features, and the actual processing time of historical work orders as training labels; the text descriptions are vectorized and features are extracted, and then fused with the features of equipment type and fault codes; a regression model is trained based on the fused features to obtain the work order complexity prediction model, which is used to predict the expected processing time of new work orders. The personnel performance prediction model training submodule is used to construct a training sample set, wherein the features of each sample include: personnel skill features, features of the work order being processed, and environmental context features; the labels of the training sample set are the actual processing results of the corresponding historical work orders, including success / failure labels, actual processing time, and customer satisfaction rating; the model is trained using machine learning classification and regression algorithms to obtain the personnel performance prediction model, which is used to predict the expected performance of a specified personnel in handling a specified work order, including the probability of first-time repair, estimated time consumption, and expected satisfaction.

[0063] The AI ​​prediction model training module is essentially an AI model enabling module, encompassing work order complexity / time prediction, personnel performance prediction, fault prediction and preventative maintenance triggering, and dynamic scheduling optimization. The work order complexity / time prediction utilizes historical work order data (description text, equipment type, fault code, historical processing time) to train an NLP+regression model, more accurately predicting the actual processing time of new work orders, replacing static estimation. The personnel performance prediction, based on personnel skills, experience, historical data, and work order characteristics, predicts a person's success rate (first-time repair rate), estimated processing time, and work order satisfaction for specific types of work orders, enabling more precise matching. The fault prediction and preventative maintenance triggering module combines IoT device sensor data, using a time-series prediction model (LSTM, Transformer) to predict potential equipment faults, automatically generating preventative maintenance work orders, and incorporating them into the scheduling pool to achieve "predictive dispatching." The dynamic scheduling problem is formalized as a variant of the job shop scheduling problem. An AI scheduling engine is built by combining reinforcement learning agents (RL agents) with graph neural networks (GNNs). The AI ​​engine continuously monitors the arrival of new work orders, changes in work order status (such as completed or suspended), changes in personnel status (task completed, on leave, location moved), traffic conditions, and changes in spare parts inventory. Once a change occurs, the scheduling engine is immediately triggered to recalculate or adjust.

[0064] In some embodiments, the system status acquisition module includes: The event listening submodule is used to continuously listen for various events from the equipment asset management system. These events include: new work order creation events, work order status change events, personnel status change events, equipment alarm events, and timer trigger events. The work order pool formation submodule is used to respond to any of the events mentioned above, obtain all currently unassigned and uncompleted work orders, and form the current work order pool. The personnel status acquisition submodule is used to obtain the latest real-time status of all operation and maintenance personnel from the personnel dynamic capability map. The latest real-time status includes at least the current skill proficiency, GPS location, work status and current workload. The environmental information acquisition submodule is used to call external services through the application programming interface to obtain the latest environmental information, which includes at least real-time traffic data and spare parts warehouse inventory status.

[0065] In some embodiments, the candidate screening module includes: The hard constraint verification submodule is used to iterate through all available maintenance personnel and, for each work order to be dispatched, sequentially apply the hard constraint rules defined in the constraint condition set to verify the feasibility of personnel-work order pairings. Personnel who pass all hard constraint rule verifications are retained to form a candidate personnel set for that work order. The hard constraint rules include at least: verifying whether the personnel's skill tags in the personnel's dynamic capability map completely cover the required skill chain defined for that work order in the work order's feature model; verifying whether the personnel can arrive at the site within the time window required by the work order based on the personnel's real-time location and the work order's geographical location, combined with real-time traffic conditions; verifying whether the time period of the personnel's currently assigned tasks overlaps with the estimated processing time period of the work order; verifying whether the personnel's current available spare parts inventory or the inventory status of their responsible warehouse meets the requirements for critical spare parts specified in the work order; verifying whether the personnel's qualification certification meets the security level requirements defined in the work order, and whether specific operations meet the required number of personnel.

[0066] In some embodiments, the prediction module includes: The work order feature extraction submodule is used to extract corresponding feature values ​​from the work order data for each work order in the current work order pool, based on the attribute set defined in the work order feature model, to form a standardized feature vector for the work order; the standardized feature vector of the work order is input into the trained work order complexity prediction model to output the predicted processing time of the work order. The personnel performance prediction submodule is used to perform the following operations for each person in the candidate set that has passed the initial screening: obtain the current state and capability feature vector of the person from the personnel dynamic capability map; concatenate and fuse the feature vector of the person with the standardized feature vector of the work order to be dispatched to form a personnel-work order paired feature vector; input the personnel-work order paired feature vector into the trained personnel performance prediction model, and output the expected result of the person handling this work order, wherein the expected result includes at least the predicted first repair probability and the predicted processing time.

[0067] In some embodiments, the AI ​​scheduling engine module includes: The system state characterization submodule is used to characterize the current system state as a feature vector. This feature vector integrates the characteristics of all work orders in the current work order pool, the status and capability characteristics of all personnel in the candidate personnel set, the predicted processing time and performance, and environmental information. The decision action space definition submodule is used to define possible dispatch decision actions based on the current work order pool and the candidate set, such that one work order in the current work order pool is assigned to one person in the candidate set. All possible dispatch actions constitute the decision action space. The reward value calculation submodule is used to calculate a comprehensive reward value based on the multi-objective optimization function after simulating the execution of a dispatch action. The strategy search submodule is used to search for a strategy in the decision action space through reinforcement learning algorithms. This strategy can maximize the expected reward accumulated in the future and output a comprehensive optimal work order-personnel allocation scheme.

[0068] In some embodiments, the system further includes: The model update and optimization module is used to collect the actual processing data after the work order is completed and send the actual processing data back to the database; and to update and optimize the AI ​​prediction model and the AI ​​scheduling engine using the sent actual processing data.

[0069] In some embodiments, the model update optimization module includes: The data feedback submodule is used to collect the actual processing data after the work order is completed and send it back to the central database. The actual processing data includes: the actual start time of the work order, the actual completion time, the actual working hours consumed, the actual spare parts list used, the final status of the work order, the root cause code of the fault, and the on-site difficulty score reported by the personnel. The sample association submodule is used to associate the actual processed data returned with the original dispatch records in the database. The dispatch records include the work order characteristics used when dispatching the order, the personnel's ability status, and the prediction values ​​of the AI ​​model, thereby forming training samples labeled with real results. The incremental learning submodule is used to periodically or trigger newly formed training samples to be added to the training set to perform incremental learning or fine-tuning on the work order complexity prediction model and the personnel performance prediction model, so as to reduce the deviation between the model prediction value and the actual value. The policy network update submodule is used to store the associated data pairs as experience samples in the experience replay buffer of the AI ​​scheduling engine. The AI ​​scheduling engine periodically samples batch experiences from the buffer, calculates the temporal difference error between the actual reward and the expected reward, and updates its internal policy network and value network parameters through the backpropagation algorithm.

[0070] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A mechanism- and data-driven intelligent work order scheduling method, characterized in that, Includes the following steps: S1. Collect historical data from the equipment asset management system, including work order data, personnel data, equipment data, spare parts inventory data, and geographic information data; S2. Construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; S3. Train an AI prediction model using historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. S4. In response to the generation of a new work order or a system status change event, obtain the current work order pool, the real-time status of all personnel, and environmental information. S5. Based on the constructed set of constraints, perform a preliminary screening of the available personnel maintained in the personnel dynamic capability map to obtain a set of candidate personnel. S6. Call the trained work order complexity prediction model and personnel performance prediction model, and combine the work order feature attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. S7. Input the current work order pool, candidate set, predicted processing time and expected results, and the constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on the reinforcement learning algorithm. S8. Execute the work order-personnel allocation scheme to dispatch the work order to the terminal of the corresponding personnel.

2. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 1, characterized in that, The steps for constructing a mechanistic model in S2 include: S21. Define a set of characteristic attributes for each type of work order. The characteristic attributes include at least: work order type, urgency level, estimated time consumption, required skill chain, required tool or spare parts list, geographical location, planned time window, associated equipment identifier, security level requirements, and customer importance level. S22. Establish a dynamically updated capability map for each operations and maintenance personnel. The capability map shall include at least: a skill tag library, the proficiency level corresponding to each skill tag, real-time GPS location, current work status, historical performance indicators, current workload quantification value, and personal work preferences. S23. The hard constraints that must be met in scheduling decisions based on business rules include: skill matching constraints, security compliance constraints, spare parts availability constraints, geographical accessibility constraints, and time conflict constraints. S24. Construct a comprehensive optimization objective function F to quantitatively evaluate the quality of the scheduling scheme. The objective function F is a weighted sum of multiple sub-objectives, expressed as: In the formula, Total completion time For total commuting costs, To improve staff utilization, First-time repair rate refers to the percentage of work orders in which an employee successfully resolves a fault on their first execution out of all their work orders. These are the weighting coefficients for each sub-objective.

3. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 2, characterized in that, Training AI prediction models in S3 includes: S31. Train the work order complexity prediction model, including the following steps: The text descriptions, equipment types, and fault codes in historical work order data are collected as training features, and the actual processing time of historical work orders is used as training labels. Vectorized feature extraction is performed on the text description, and then fused with the features of device type and fault code to generate a standardized feature vector; A regression model is trained based on the fused standardized feature vectors to obtain a work order complexity prediction model, which is used to predict the expected processing time of new work orders. S32. Training personnel performance prediction model, including the following steps: Construct a training sample set, where each sample's features include: personnel skill features, features of the work orders processed, and environmental context features; The labels for the training sample set are the actual processing results of the corresponding historical work orders, including success / failure labels, actual processing time, and customer satisfaction rating. The model is trained using machine learning classification and regression algorithms to obtain a personnel performance prediction model, which is used to predict the expected performance of a specified person in handling a specified work order, including the probability of first-time repair, the estimated time, and the expected satisfaction.

4. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 3, characterized in that, The steps in S4 include: S41. Continuously monitor various events from the equipment asset management system, including: new work order creation event, work order status change event, personnel status change event, equipment alarm event, and timer trigger event; S42. In response to any of the events mentioned above, obtain all currently unassigned and uncompleted work orders and form the current work order pool; S43. Obtain the latest real-time status of all operation and maintenance personnel from the personnel dynamic capability map. The latest real-time status includes at least the current skill proficiency, GPS location, work status and current workload. S44. Call external services through the application programming interface to obtain the latest environmental information, which includes at least real-time traffic data and spare parts warehouse inventory status.

5. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 4, characterized in that, The steps in S5 include: S51. Iterate through all available maintenance personnel, and for each work order to be dispatched, apply the hard constraint rules defined in the constraint condition set to perform a feasibility check on the personnel-work order pairing. S52. Retain the personnel who pass all hard constraint rule verifications to form the candidate personnel set for this work order; The hard constraint rules include at least the following: Verify whether the skill tags of the personnel in the dynamic competency map completely cover the required skill chain defined in the work order feature model for that work order; Based on the real-time location of personnel and the geographical location of the work order, combined with real-time traffic conditions, it verifies whether personnel can arrive at the site within the time window required by the work order. Check whether the time period of the task currently assigned by the verification personnel overlaps with the estimated processing time period of the work order; For the critical spare parts specified in the work order, verify whether the spare parts inventory that the person can currently apply for or the inventory status of the warehouse he is responsible for meets the requirements; Verify whether the personnel's qualifications meet the security level requirements defined in the work order, and whether the specific operation meets the required number of personnel.

6. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 5, characterized in that, The steps in S6 include: S61. For each work order in the current work order pool, extract the corresponding feature values ​​from the work order data according to the work order feature attributes to form the standardized feature vector of the work order. S62. Input the standardized feature vector of the work order into the trained work order complexity prediction model and output the predicted processing time of the work order. S63. For each person in the candidate set who passed the initial screening, perform the following operations: a. Obtain the current status and capability feature vector of the person from the personnel dynamic capability map; b. Concatenate and fuse the feature vectors of personnel with the standardized feature vectors of work orders to be dispatched to form a personnel-work order paired feature vector; c. Input the personnel-work order pairing feature vector into the trained personnel performance prediction model, and output the expected result of the personnel handling this work order. The expected result includes at least the predicted first repair probability and the predicted processing time.

7. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 6, characterized in that, The steps in S7 include: S71. Represent the current system state as a state feature vector. This state feature vector integrates the features of all work orders in the current work order pool, the state and capability features of all personnel in the candidate personnel set, the predicted processing time and performance, and environmental information; S72, Based on the current state The included work order pool and candidate set will handle every possible work order dispatch decision. Defined as assigning a work order from the work order pool to a person in the candidate set, all possible assignment actions constitute the decision action space A; S73. Execute a dispatch action in a simulation environment. Then, an immediate reward value is calculated based on the multi-objective optimization function F. This reward value is used to quantify actions. In state The immediate results achieved in terms of comprehensive optimization goals; S74. The AI ​​scheduling engine, as a reinforcement learning agent, is in a state... The goal is to select an action in the decision action space A according to policy π. To maximize the expected value of the future cumulative discount reward Gt, the final output can be achieved in the current state. Find the optimal work order-personnel allocation scheme that maximizes Gt. In the formula, γ This is a discount factor, with a value range of 0 ≤ γ ≤1, The reward obtained at time step t+k. E [•] represents the expected value, which is the average value over all possible states and actions.

8. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 7, characterized in that, The method also includes: S9. Collect the actual processing data after the work order is completed, and send the actual processing data back to the database; use the sent actual processing data to update and optimize the AI ​​prediction model and the AI ​​scheduling engine.

9. The intelligent work order scheduling method driven by mechanism and data fusion according to claim 8, characterized in that, The steps in S9 include: S91. Collect the actual processing data after the work order is completed and send it back to the central database; the actual processing data includes: the actual start time of the work order, the actual completion time, the actual working hours consumed, the actual spare parts list used, the final status of the work order, the root cause code of the fault, and the on-site difficulty score reported by the personnel. S92. Associate the actual processed data returned with the original dispatch records in the database. The dispatch records contain the work order characteristics, personnel capability status and AI model prediction values ​​used when dispatching the work order, thereby forming training samples labeled with real results. S93. Periodically or triggerively add newly formed training samples to the training set to incrementally learn or fine-tune the work order complexity prediction model and the personnel performance prediction model in order to reduce the deviation between the model prediction value and the actual value. S94. The associated data pairs are used as experience samples and stored in the experience replay buffer of the AI ​​scheduling engine. The AI ​​scheduling engine periodically samples batch experiences from the buffer, calculates the temporal difference error between the actual reward and the expected reward, and updates its internal policy network and value network parameters through the backpropagation algorithm.

10. A mechanism- and data-driven intelligent work order scheduling system, characterized in that, include: The data acquisition module is used to collect historical data from the equipment asset management system. The historical data includes work order data, personnel data, equipment data, spare parts inventory data, and geographic information data. The mechanism model construction module is used to construct a mechanism model, which includes work order feature attributes, personnel dynamic capability map, constraint set and multi-objective optimization function; The AI ​​prediction model training module is used to train an AI prediction model using historical data. The AI ​​prediction model includes at least a work order complexity prediction model and a personnel performance prediction model. The system status acquisition module is used to obtain the current work order pool, the real-time status of all personnel, and environmental information in response to the generation of new work orders or system status change events. The candidate screening module is used to perform preliminary screening of available personnel maintained in the personnel dynamic capability map based on the constructed set of constraints, and obtain a set of candidate personnel. The prediction module is used to call the trained work order complexity prediction model and personnel performance prediction model, and combine the work order feature attributes with the candidate personnel set to predict the expected processing time of each work order and the expected results of each candidate personnel in processing each work order. The AI ​​scheduling engine module is used to input the current work order pool, candidate set, predicted processing time and expected results, and the constructed multi-objective optimization function into the AI ​​scheduling engine. The AI ​​scheduling engine outputs the optimal work order-personnel allocation scheme based on reinforcement learning algorithm. The work order execution module is used to execute the work order-personnel allocation plan and dispatch the work order to the corresponding personnel's terminal.

Images