Artificial intelligence-based monitoring digital employee generation and use method and device

Abstract

The application provides a kind of based on artificial intelligence's monitoring digital employee generation and use method and device, it is related to artificial intelligence technical field.Therein, method includes, the monitoring work that true personnel worker executes is identified, when monitoring work meets preset machine generation condition, start digital employee robot generation system, train and generate the digital employee robot corresponding to monitoring work;Digital employee robot is evaluated to reach the standard, and digital employee robot is formally put into operation after evaluation reaches the standard, using the mode of true personnel worker and digital employee robot shift work, compare the work effect of true personnel worker and digital employee robot, according to the comparison result, dynamically adjust the scheduling plan.The application overcomes the illusion risk of traditional large language model when directly applied to alarm analysis, large training cost, unable to continuously evolve and the problem that general robot is difficult to adapt to special scene, significantly improves the automation degree and dynamic scheduling efficiency of fault monitoring work.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for generating and using digital employees based on artificial intelligence. Background Technology

[0002] With the continuous expansion of communication networks, operator network management systems generate massive amounts of alarms daily. Rapid and accurate alarm processing is crucial for ensuring network service quality. However, the sheer number of alarms, their diverse types, and the complexity of device topologies mean that traditional manual analysis methods are insufficient to meet timeliness requirements. Industry attempts to build full-featured robots using large language models face the following core challenges: large model outputs can be misleading, posing a high risk when directly applied to production environments; model training is costly, and once deployed, they cannot leverage the new experiences of human employees for continuous evolution; and general-purpose robots struggle to adapt to the specific needs of different scenarios.

[0003] Existing alarm handling methods are mainly based on manual rules or isolated machine learning models, which make it difficult to incorporate the dynamic experience of real employees in monitoring work such as fault correlation analysis, level prediction, and scheduling tracking. Furthermore, they lack a low-cost continuous evolution mechanism, resulting in limited processing efficiency and accuracy.

[0004] Therefore, there is an urgent need for a monitoring process that can leverage real employee monitoring experience to generate evolving digital employee robots at low cost. Summary of the Invention

[0005] This invention provides a method for generating and using digital employees for monitoring based on artificial intelligence. This method can overcome the problems of high illusion risk, high training cost, inability to continuously evolve, and difficulty in adapting general robots to special scenarios when traditional large language models are directly applied to alarm analysis. It significantly improves the automation level and dynamic scheduling efficiency of fault monitoring work.

[0006] Firstly, it provides a method for generating and using digital employees based on artificial intelligence, including: Identify monitoring tasks performed by real employees, including at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm miss analysis. When the monitoring work meets the preset machine generation conditions, the digital employee robot generation system is activated to train and generate a digital employee robot corresponding to the monitoring work. The generated digital employee robot undergoes a compliance assessment. Once the digital employee robot passes the assessment, it is officially launched and a work mode of rotating between real employees and digital employee robots is adopted. The work performance of the real employees and the digital employee robot is compared, and the shift schedule is dynamically adjusted based on the comparison results.

[0007] In this way, by identifying the specific monitoring tasks performed by real employees, corresponding digital employee robots are automatically trained and generated when preset conditions are met, achieving intelligent replacement and assistance for monitoring work. The generated robots undergo a compliance assessment before going online, ensuring their reliability. Adopting a rotating work model between real employees and digital employee robots allows for a direct comparison of their work performance, and the scheduling plan can be dynamically adjusted based on the comparison results, enabling the system to continuously optimize human resource allocation. This method overcomes the problems of high illusion risk, high training costs, inability to continuously evolve, and difficulty in adapting general-purpose robots to specific scenarios when traditional large language models are directly applied to production environments. It significantly improves the automation level, dynamic adaptability, and overall operational efficiency of fault monitoring.

[0008] In some possible implementations, the monitoring tasks performed by identifying real employees include: Obtain real employees' operation logs and work order processing records; The task type of the real employee is determined from the operation log and work order processing records; The task type is matched with the monitoring work type to determine the monitoring work currently being performed by the real employee.

[0009] In this way, by acquiring the operation logs and work order processing records of real employees, the task type is automatically determined and matched with the monitored work type, without the need for manual annotation or additional configuration. On the one hand, it makes full use of existing operation and maintenance data, reducing the identification cost; on the other hand, based on the objective matching of logs and records, it avoids subjective judgment errors, improves the accuracy and real-time performance of monitoring work identification, and provides a reliable trigger basis for the subsequent on-demand generation of digital employee robots.

[0010] In some possible implementations, the step of activating the digital employee robot generation system and training to generate a digital employee robot corresponding to the monitoring task when the monitoring work meets preset machine generation conditions includes: Determine whether the monitoring work meets the preset machine generation conditions, the preset machine generation conditions including the difficulty of the work exceeding a first threshold and / or the workload exceeding a second threshold; When the conditions are met, the feature information of the monitoring work is extracted, wherein the feature information includes alarm logs, processing time, operation logs, manual feedback information and effect evaluation information; The feature information is vectorized to obtain the working feature vector; The K most similar historical workflows are retrieved from the vector database using the aforementioned work feature vectors; Based on the K historical workflows, a directed acyclic graph is generated that defines the execution logic of the digital employee robot. The directed acyclic graph includes the digital employee name, triggering conditions, and execution chain, wherein the triggering conditions are used to determine the robot's start time, and the execution chain is used to determine the sequence of robot execution steps. Based on the triggering conditions and the execution chain, an artificial intelligence model is trained to generate the digital employee robot.

[0011] In this way, suitable monitoring tasks for mechanization are automatically selected based on work difficulty or workload thresholds, avoiding invalid generation. Multi-dimensional feature information is extracted and quantified, and similar historical workflows are retrieved using a vector database. Existing experience is reused to generate a Directed Acyclic Graph (DAG), and then an artificial intelligence model is trained based on the triggering conditions and execution chains in the DAG. This method significantly reduces the generation cost and training difficulty of digital employee robots, improves generation efficiency and the interpretability of execution logic, while ensuring the robot's adaptability to specific monitoring tasks.

[0012] In some possible implementations, the assessment of the generated digital employee robot for compliance includes: The digital employee robot described in the instructions performs the corresponding monitoring tasks on the backup database and obtains the task execution results; The accuracy, recall, and false alarm rates of the digital employee robot's work execution results were statistically analyzed. Determine whether the accuracy reaches a first preset threshold, whether the recall reaches a second preset threshold, and whether the false positive rate is lower than a third preset threshold; If all conditions are met, the digital employee robot is deemed to have met the standards and is approved for official launch. If the requirements are not met, the model is deemed substandard, triggering a retraining of the AI ​​model.

[0013] In this way, by evaluating the accuracy, recall, and false positive rate of the digital employee robot offline in a backup database, and setting three independent thresholds to determine whether it meets the standards, the production risks that may arise from direct deployment are avoided, while ensuring the comprehensive performance of the robot. If the standards are not met, the model is retrained, forming a closed-loop optimization mechanism to continuously improve the quality of the robot and ensure its reliability and stability after official deployment.

[0014] In some possible implementations, the method of employing a rotating work schedule of real employees and digital employees involves comparing the work performance of the real employees and the digital employee robots, and dynamically adjusting the shift schedule based on the comparison results, including: The digital employee robot works alongside human employees in a shift system, with tasks assigned according to a preset schedule. During the rotation process, the work result data of the real employees and the digital employee robots are continuously collected. The work result data includes one or more of the following: alarm processing timeliness, work quality score, false alarm rate and false alarm rate. The collected work results data are input into the scheduling model, which uses XGBoost, LightGBM or random forest algorithms; The scheduling model outputs an optimized scheduling vector, which indicates the task allocation ratio between the real employees and the digital employee robots in each time period. The scheduling plan is updated based on the optimized scheduling vector to achieve dynamic scheduling of the real employees and the digital employee robot.

[0015] In this way, by having human employees and digital employee robots work in parallel and on a rotating basis, multi-dimensional work result data (timeliness, quality, false alarm rate, and false negative rate) is continuously collected. Optimized scheduling vectors are then output using ensemble learning models such as XGBoost, enabling dynamic adjustment of task allocation ratios. Compared to fixed scheduling, this method can adaptively optimize human-machine collaboration strategies based on actual work performance, fully leveraging the strengths of both parties, continuously improving overall monitoring efficiency and accuracy, while reducing the cost of manual intervention.

[0016] Among the possible implementations are: When changes in employee turnover or alarm load fluctuations are detected, the scheduling model is automatically retrained and the scheduling plan is regenerated.

[0017] In this way, by monitoring changes in employee turnover rate and alarm load fluctuations in real time, the system automatically triggers the retraining of the scheduling model and the regeneration of the scheduling plan, enabling the system to dynamically adapt to changes in the external environment. When human resources decrease or alarm volume surges, the system can promptly adjust the task allocation ratio between human employees and digital employee robots, avoiding processing delays caused by staff shortages or sudden load increases, ensuring the continuity and stability of monitoring work, and reducing the operational costs of manual intervention in scheduling.

[0018] Among the possible implementations are: Collect creative work characteristics of real employees in monitoring work, including methods for handling new types of alarms, records of developing new processing procedures, and operations for optimizing alarm rules; The creative work characteristics are processed to form characteristic data; When the accumulated amount of the characteristic data reaches a preset threshold, the characteristic data is used to trigger the digital employee robot generation system to train and generate new digital employees or optimize existing digital employee robots.

[0019] In this way, by proactively collecting behavioral records of real employees in their creative work, such as handling new types of alarms, formulating new processes, and optimizing rules, and using these records as training data after characterization, the system automatically triggers the generation or optimization of a digital employee robot when the accumulated data reaches a certain threshold. This mechanism enables the system to continuously absorb the experience, innovation, and business evolution of real employees, breaking the limitations of traditional models that become fixed after deployment. It achieves self-evolution and scenario self-adaptation of the robot, effectively improving the responsiveness and long-term applicability of the digital employee to new fault modes and new processing logic.

[0020] Secondly, an artificial intelligence-based device for monitoring the generation and use of digital employees is provided, comprising: The first processing module is used to identify the monitoring work performed by real employees. The monitoring work includes at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm missed reporting analysis. The second processing module is used to start the digital employee robot generation system and train and generate a digital employee robot corresponding to the monitoring work when the monitoring work meets the preset machine generation conditions. The third processing module is used to conduct a compliance assessment on the generated digital employee robot. After the digital employee robot passes the assessment, it is officially launched and adopts a work rotation mode of real employees and digital employee robots. The work effects of the real employees and the digital employee robots are compared, and the shift schedule is dynamically adjusted based on the comparison results.

[0021] In some possible implementations, the first processing module is used to obtain the operation logs and work order processing records of real employees; The task type of the real employee is determined from the operation log and work order processing records; The task type is matched with the monitoring work type to determine the monitoring work currently being performed by the real employee.

[0022] In some possible implementations, the second processing module is used to determine whether the monitoring work meets preset machine generation conditions, the preset machine generation conditions including the difficulty of the work exceeding a first threshold and / or the workload exceeding a second threshold; When the conditions are met, the feature information of the monitoring work is extracted, wherein the feature information includes alarm logs, processing time, operation logs, manual feedback information and effect evaluation information; The feature information is vectorized to obtain the working feature vector; The K most similar historical workflows are retrieved from the vector database using the aforementioned work feature vectors; Based on the K historical workflows, a directed acyclic graph is generated that defines the execution logic of the digital employee robot. The directed acyclic graph includes the digital employee name, triggering conditions, and execution chain, wherein the triggering conditions are used to determine the robot's start time, and the execution chain is used to determine the sequence of robot execution steps. Based on the triggering conditions and the execution chain, an artificial intelligence model is trained to generate the digital employee robot.

[0023] In some possible implementations, the third processing module is used to instruct the digital employee robot to perform the corresponding monitoring task on the backup database and obtain the task execution result; The accuracy, recall, and false alarm rates of the digital employee robot's work execution results were statistically analyzed. Determine whether the accuracy reaches a first preset threshold, whether the recall reaches a second preset threshold, and whether the false positive rate is lower than a third preset threshold; If all conditions are met, the digital employee robot is deemed to have met the standards and is approved for official launch. If the requirements are not met, the model is deemed substandard, triggering a retraining of the AI ​​model.

[0024] In some possible implementations, the third processing module is used to enable the digital employee robot to work in parallel with human employees in a shift manner, and to assign tasks according to a preset shift schedule. During the rotation process, the work result data of the real employees and the digital employee robots are continuously collected. The work result data includes one or more of the following: alarm processing timeliness, work quality score, false alarm rate and false alarm rate. The collected work results data are input into the scheduling model, which uses XGBoost, LightGBM or random forest algorithms; The scheduling model outputs an optimized scheduling vector, which indicates the task allocation ratio between the real employees and the digital employee robots in each time period. The scheduling plan is updated based on the optimized scheduling vector to achieve dynamic scheduling of the real employees and the digital employee robot.

[0025] In some possible implementations, a third processing module is used to automatically trigger the retraining of the scheduling model and the regeneration of the scheduling plan when changes in the turnover rate of real employees or fluctuations in alarm load are detected.

[0026] In some possible implementations, the second processing module is used to collect creative work characteristics of real employees in the monitoring work, including methods for handling new types of alarms, records of formulating new processing procedures, and operations for optimizing alarm rules. The creative work characteristics are processed to form characteristic data; When the accumulated amount of the characteristic data reaches a preset threshold, the characteristic data is used to trigger the digital employee robot generation system to train and generate new digital employee robots or optimize existing digital employee robots.

[0027] Thirdly, an electronic device is provided, comprising: one or more processors; one or more memories; and one or more programs, wherein the one or more programs are stored in the one or more memories, and the one or more programs include instructions that, when executed by the one or more processors, cause the configured device to perform the method as described in the first aspect.

[0028] Fourthly, a computer storage medium is provided, the computer storage medium storing instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect or the second aspect.

[0029] Fifthly, a computer program product is provided, the computer program product storing instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect or the second aspect.

[0030] The beneficial effects of the second to fifth aspects can be referred to the introduction of the beneficial effects of the first aspect above, and will not be repeated here. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for generating and using digital employees, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the alarm collection, filtering, and push process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the alarm filtering rule generation process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the instruction set provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the fault association robot process provided in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the startup process of a digital robot generation system provided in an embodiment of the present invention. Figure 7These are schematic diagrams illustrating various robot algorithms provided in embodiments of the present invention. Figure 8 A schematic diagram of the shift scheduling plan generation provided in this embodiment of the invention; Figure 9 This is a schematic diagram illustrating the collaborative evolution process between human and digital employee robots, provided in an embodiment of the present invention. Figure 10 A flowchart illustrating a method for generating and using digital employees based on artificial intelligence, provided in an embodiment of the present invention; Figure 11 This is a schematic diagram of an artificial intelligence-based device for monitoring the generation and use of digital employees, which can be used to implement the methods of the present invention. Figure 12 This is a schematic diagram of an electronic device provided by the present invention. Detailed Implementation

[0032] The solutions provided by the embodiments of the present invention will now be described with reference to the accompanying drawings. In the embodiments of the present invention, "multiple" refers to two or more objects, and "various kinds" refers to two or more types. Terms such as "first," "second," etc., are only used to distinguish similar objects and are not necessarily used to describe a specific order or number of objects.

[0033] With the continuous expansion of communication networks, operator network management systems generate massive amounts of alarms daily. Rapid and accurate alarm processing is crucial for ensuring network service quality. However, due to the large number of alarms, diverse types, and complex device topologies, traditional methods relying on manual analysis are no longer sufficient to meet timeliness requirements. While the industry has attempted to build full-featured robots using large language models, it faces the following core problems: first, the output of large models can be misleading, making direct application to production environments risky; second, model training is costly, and once deployed, it cannot continuously evolve using the new experiences of human employees; and third, general-purpose robots struggle to adapt to the specific needs of different scenarios.

[0034] To address the aforementioned technical issues, this invention proposes an artificial intelligence-based method for generating and using digital employees for monitoring. By breaking down complex monitoring tasks into multiple specialized stages, training dedicated robots for each stage independently, and employing a rotating and continuously evolving model of real and digital employees in the work, the level of automation is continuously improved while controlling the risk of hallucinations.

[0035] The following is in conjunction with the appendix Figure 1 The specific method of the present invention will be described in detail below.

[0036] Figure 1 This is a flowchart illustrating a method for generating and using digital employees based on artificial intelligence, as shown in an embodiment of the present invention. Figure 1As shown, the specific steps include: In step 110, alarms are collected, filtered, and pushed.

[0037] First, collect alarms from each network management system. Alarm collection involves collecting all device alarms, including but not limited to fault alarms, performance alarms, and status alarms. After collection, alarms need to be filtered using alarm filtering rules, and only critical alarms are pushed to subsequent processing stages.

[0038] Figure 2 This is a diagram illustrating the alarm collection, filtering, and push process, such as... Figure 2 As shown, eight network types are listed vertically by professional field, in the following order: wireless network, core network, broadband, data network, transmission network, dynamic environment network, cloud resources, and platform.

[0039] Each specialized network corresponds to three standard components, from left to right: alarm management, alarm push rules, and alarm robot. For example, the wireless network corresponds to "wireless network alarm management," "wireless network alarm push rules," and "wireless network alarm robot"; the core network corresponds to "core network alarm management," "core network alarm push rules," and "core network alarm robot," and so on for other specialized networks. The overall structure demonstrates that multi-specialized alarm processing adopts a standardized pipeline model of "network management data collection—rule filtering—robot push."

[0040] Figure 3 This is a schematic diagram of the alarm filtering rule generation process, such as... Figure 3 The alarm filtering rules shown are not statically configured, but dynamically generated using the following methods, such as... Figure 3 As shown.

[0041] First, input the full alarm fields, which come from the raw alarm data fields of various professional network administrators. Next, perform alarm field validity cleaning, including deduplication, completion, format standardization, and invalid field removal, to extract valid fields that can be used for subsequent rules.

[0042] Subsequently, a rule training algorithm is executed on the processed alarm fields, combined with a rule generation model, such as algorithms based on LLM + LoRA and FP-Growth, and referencing existing rule samples in the alarm rule library, to generate new alarm filtering and push rules.

[0043] Finally, the generated filtering rules use alarm fields to specify the specific set of fields that the newly generated rules depend on. This includes generating alarm filtering push rules, including outputting executable rule expressions such as PromQL, SQL, and regular expressions. The performance of the generated rules on actual or simulated alarm flows is evaluated. If the evaluation is satisfactory, the rule can be added to the database or deployed online. If the evaluation is unsatisfactory, feedback is given to the rule training algorithm or rule generation model for iterative optimization.

[0044] Through the above dynamic rule generation method, alarm filtering rules can adapt to the evolution of the network environment and new alarm patterns, while maintaining the accuracy of filtering.

[0045] The model training steps include: First, the existing rule base is modified into a <instruction, input, output> format that the large model can recognize. The rule base is then parsed: extracting the "matching conditions" and "execution actions / filtering logic" from the rules, and constructing an instruction set.

[0046] Figure 4 This is a diagram of the instruction set, such as... Figure 4 The table illustrates how to convert rules from the alert rule base into the training data format required for fine-tuning a large model. The table contains three rows, corresponding to three different data roles: Instruction: Example content: "Generate a filtering rule based on the following alarm characteristics to exclude false alarms." Its function is to tell the model the task objective that needs to be completed.

[0047] Input: The content example includes specific alarm fields such as "Alarm Name: Link Outage", "Host Role: Switch", "Duration: 30s", and "Historical False Alarm Rate: 80%", providing the model with factual information on which the generation rules are based.

[0048] Output: The content example shows two rule expression forms. One is a conditional statement (IF role == 'switch' AND duration > 30s THEN suppress_alert), and the other is a PromQL query expression (avg(rate(linkdown[30s])) > 0.8 AND role = 'switch'). These correspond to the mature rule logic already existing in the rule base, which is the expected output that the model needs to learn and generate.

[0049] By defining the <instruction, input, output> triple structure, it can be used to train large code models to automatically generate alarm filtering and push rules based on alarm characteristics.

[0050] Next, feature enhancement is performed. FP-Growth can be used to mine frequent itemsets, and historical alarm logs can be scanned to calculate support and confidence. For example, it was found that when the protocol goes down and the duration does not exceed 30 seconds, 90% of the time it is a brief outage that automatically recovers. When inputting the LLM, an associated feature field is added: [protocol down, duration not exceeding 30 seconds]. If the alarm content contains unstructured text, BERT or TF-IDF is used to convert it into vectors, which serve as additional input to the model to help it understand semantic similarity.

[0051] Next, fine-tune the model by loading a pre-trained model (such as CodeLlama-7b), configuring the LoRA adapter, and setting the rank r = 8 or 16. Set the target modules (target_modules) = ['q_proj', 'v_proj']. Define the loss function: use the cross-entropy loss function to make the model predict the next token as closely as possible to the standard rules in your rule base. Input the instruction and alert fields, and the model outputs the rules.

[0052] The differences between the rules generated by the calculation model and the rule base are used to update the model parameters.

[0053] During inference, a new stream of unclassified alarm fields is input, and the model outputs suggested rules, such as: "Suggest adding rule: Ignore intermittent alarms from the switch within 30 seconds." The generated rules are compared and run against a copied alarm database. Alarms that have been blocked are manually marked as valid. If the accuracy is >95%, the rule is added to the rule base; otherwise, the model continues to be trained and the algorithm is fine-tuned.

[0054] In step 120, fault correlation analysis is performed.

[0055] Even after alarm filtering, the daily alarm volume remains substantial. Furthermore, the devices are interconnected in a topology, meaning a failure in one device can easily cause service interruptions in multiple devices. Therefore, it is necessary to perform correlation analysis on the alarms to determine the root cause. Figure 5 This is a flowchart illustrating the fault-related robot process, which includes the following steps: First, alarm features are vectorized. Alarms from multiple specialties are vectorized separately. Each alarm may include fields such as alarm type, device identifier, timestamp, and additional text. Each specialty alarm vector can be obtained as a fixed-length vector representation using a bag-of-words model or a pre-trained text encoder (such as BERT). Then, the alarm vectors from each specialty are concatenated to obtain a joint vector: V_alert = Concat(V1, V2, V3, ...).

[0056] Next, a graph is constructed. A time window (e.g., 8 minutes) is set. Within this window, if two alarms correspond to devices that have a direct calling relationship in the network topology (e.g., device A is upstream of device B), an edge is added to the graph. Similarly, if the semantic similarity of two alarms (e.g., both are power outage alarms) is greater than 0.8 (calculated using cosine similarity), an edge is also added. This results in an alarm association graph where nodes represent alarms and edges represent potential associations.

[0057] Subsequently, model training is performed. Supervised learning is conducted using historical fault data (annotated multi-disciplinary alarms and associated root causes). The input is a vector containing multi-disciplinary alarms, and a graph neural network (GNN) is used to aggregate neighbor information: each alarm node collects features from its associated nodes (upstream services, downstream services), calculated using the following formula: : in These are the attention weights learned by the model, representing the degree of influence of alarm A on alarm B; W is the learnable weight matrix. is the feature vector of the neighboring node.

[0058] The loss function is selected based on the task: if it is root cause localization, cross-entropy loss is used to make the root cause nodes predicted by the model consistent with the labels; if it is alarm aggregation, contrastive loss is used to make the alarm vectors of the same faulty link closer together and unrelated ones farther apart.

[0059] Finally, inference and correlation output are performed. When multiple specialties simultaneously generate new valid alarms, they are used as vector inputs for topological and temporal matching. The model outputs the probability P(B|A) that alarm A causes alarm B. If P(B|A) > a threshold (e.g., 0.8), then A is determined to be the root cause of B. The final output is a directed acyclic graph (DAG) that clearly shows the propagation chain of root cause and symptom, for example: power outage (root cause) -> cloud resource server downtime (symptom) -> causing platform page error (symptom).

[0060] In step 130, the digital robot generation system is started.

[0061] When an alarm is transmitted to subsequent stages, if the workload of manual processing reaches a certain level, or the work is difficult or ineffective, such as manual feedback that it is "too difficult" or "too cumbersome", or the accuracy rate of effect evaluation is less than 70%, then the digital robot generation system will be activated. Figure 6 This is the startup flowchart for the digital robot generation system, which includes the following sub-steps: First, analyze and vectorize the work characteristics of real employees. Extract features such as alarm logs, processing time, operation logs, human feedback information, and effect evaluation information, and use TF-IDF or BERT to extract keywords, transforming them into a work feature vector V_work.

[0062] Next, the robot generation process is initiated. The K most similar historical workflows are retrieved from the vector database using the work feature vector V_work.

[0063] In this embodiment, the master control model can be used to receive features and retrieved cases, and output a directed acyclic graph (DAG) that defines the robot's execution logic. An example DAG is as follows: { "Digital Employee Name": "Fault Correlation Analysis Robot", Triggering conditions: "Simultaneous generation of multiple professional alarms AND correlation probability > 80% AND duration > 5 minutes", "Execution chain": [ {"Steps": 1,"Algorithm":"BERT classifier","Action":"Confirm alarm authenticity"}, {"Steps": 2,"Algorithm":"Knowledge Graph Query","Action":"Search Related Reasoning Service"}, {"Steps": 3,"Action":"Call the model API","Action":"Generate inference results"} ] } The DAG defines the name of the digital employee, the triggering conditions, and the specific sequence of execution steps. Each step can specify the algorithm or action to be used.

[0064] Subsequently, the robot is trained. Based on the algorithmic requirements specified in the DAG, if a new lightweight model (such as an "alarm correlation analysis model") is needed, the system automatically extracts historical alarm correlation data and uses transfer learning to quickly fine-tune a random forest or XGBoost classifier. If a large language model (LLM) is used, a specific System Prompt is written for the LLM, granting it the persona of an "alarm analysis expert" and tool access permissions.

[0065] Example 7 shows schematic diagrams illustrating various robot algorithms. Figure 7 As shown, this document lists seven typical tasks in the field of alarm operation and maintenance, along with the corresponding problem type, recommendation algorithm, and selection rationale for each task. Specifically, these include: Alarm correlation analysis falls under the category of association rule or sequence mining problems. It is recommended to use FP-Growth, PrefixSpan, or Graph Neural Networks (GNN) because these algorithms can effectively uncover co-occurrence relationships or temporal dependencies between alarms.

[0066] Alarm rating is a multi-classification or ranking problem, and XGBoost or LightGBM is recommended because they offer high interpretability, support for multi-feature fusion, and are mature for industrial applications.

[0067] Alarm scheduling falls under the category of combinatorial optimization or matching recommendation problems. Multi-objective optimization or reinforcement learning (such as DQN) can be employed to achieve the optimal match between alarms and handlers in terms of skills, load, and timeliness.

[0068] Work order tracking falls under the category of status prediction or anomaly detection problems. Random forest, LSTM, or isolated forest are recommended for identifying situations such as work order timeouts, stuck work orders, and abnormal workflows.

[0069] Alarm escalation is a problem involving time-series classification and rules. A time-series model (such as TCN) can be used in conjunction with rule triggers to first predict whether an escalation is needed, and then execute a fallback strategy.

[0070] Alarm miss detection falls under the category of unsupervised anomaly detection. Isolation forests or autoencoders are recommended because miss samples are scarce, and unsupervised methods can discover uncaptured alarms even with unlabeled data.

[0071] Performance evaluation falls under the category of evaluation and comparative learning problems. It typically employs offline metrics combined with online A / B testing. It does not directly generate a model but is used to support closed-loop optimization of the algorithm.

[0072] In addition, the Large Model Prompt Project can be used to write a specific System Prompt for the LLM, giving it the persona of an alert analysis expert and the permissions to call tools.

[0073] Finally, application validation and effectiveness evaluation.

[0074] In this embodiment of the application, the newly generated robot can be verified through a synchronized backup database, such as an alarm database.

[0075] The verification mode operates in the background: the robot only outputs suggestions and does not perform any actual operations. Evaluation metrics include accuracy, recall, and false positive rate. If a human approves the robot's suggestion, the data is marked as a positive sample and added to the training set; if the robot makes an incorrect judgment, the system automatically corrects the parameters, such as adjusting the threshold, changing the algorithm, or modifying the trigger conditions in the DAG, and retrains.

[0076] The robot can only be officially launched when its accuracy on the validation set reaches a preset threshold and its false positive rate is lower than the preset threshold.

[0077] In step 140, real and digital employees take turns collaborating.

[0078] Once the robot passes the verification, it will be officially launched, but instead of an absolute replacement model, a rotation model will be used to dynamically evaluate the combination of real and digital employees.

[0079] Figure 8 This is a schematic diagram illustrating the generation of a shift scheduling plan. For example... Figure 8 As shown, different scheduling plans are used to create comparison vectors for digital and real employees. Specifically, the input is an effect evaluation vector, which can include indicators such as expected alarm handling timeliness (e.g., average handling time ≤ 5 minutes), employee turnover rate (e.g., current on-duty employee ratio), work quality score (e.g., customer complaint rate), and current alarm load. The scheduling model is trained using XGBoost, LightGBM, or Random Forest algorithms. The input to this model is the effect evaluation vector, and the output is a scheduling vector (e.g., the work order allocation ratio of real and digital employees in each time period). After generating a new model, the system's desired target effect is input, and the model can automatically generate the corresponding scheduling plan, achieving controllable scheduling of robots and real employees. For example, when employee turnover is high, robot scheduling increases while ensuring that certain performance indicators (e.g., false alarm rate) still meet standards.

[0080] During the shift rotation, the system continuously collects performance data from both real and digital employees, updates the performance evaluation vector, and triggers retraining of the scheduling model to achieve dynamic optimization.

[0081] In step 150, real and digital employees collaborate and evolve.

[0082] Figure 9 This is a schematic diagram of the collaborative evolution process between real and digital employees. In daily life, when real employees perform creative work or complete new types of tasks, such as handling unprecedentedly complex alarms, developing new processing procedures, or optimizing existing rules, the system collects their work characteristics, including processing methods, time consumption, operation logs, and final feedback. When the accumulated data reaches a preset condition, the digital robot generation system is triggered. These new characteristics are used to train and generate new digital employees or optimize existing models, such as through incremental training or fine-tuning. After the new employees or optimized models are deployed, they collaborate with real employees in a rotation model, dynamically evaluating their effectiveness and forming a continuous improvement loop.

[0083] In this way, the digital employee model can continuously evolve by absorbing the creative accumulation of real employees, while real employees can also gain inspiration from the results processed by the robot, achieving mutual promotion. The entire system automatically generates new specialized digital employees by continuously accumulating new knowledge from real people, gradually expanding the scope of automation and reducing the burden on human labor.

[0084] Figure 10 This is a flowchart illustrating a method for generating and using digital employees based on artificial intelligence, as shown in an embodiment of the present invention. Figure 10 As shown, the specific steps include: 1010: Identify monitoring tasks performed by real employees, wherein the monitoring tasks include at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm missed reporting analysis. 1020: When the monitoring work meets the preset machine generation conditions, the digital employee robot generation system is started to train and generate a digital employee robot corresponding to the monitoring work; 1030: Conduct a compliance assessment on the generated digital employee robot. Once the digital employee passes the assessment, it will be officially launched. A work rotation mode of real employees and digital employees will be adopted. The work performance of the real employees and the digital employees will be compared. Based on the comparison results, the shift schedule will be dynamically adjusted.

[0085] In some embodiments, 1010 includes: Obtain real employees' operation logs and work order processing records; The task type of the real employee is determined from the operation log and work order processing records; The task type is matched with the monitoring work type to determine the monitoring work currently being performed by the real employee.

[0086] In some embodiments, 1020 includes: Determine whether the monitoring work meets the preset machine generation conditions, the preset machine generation conditions including the difficulty of the work exceeding a first threshold and / or the workload exceeding a second threshold; When the conditions are met, the feature information of the monitoring work is extracted, wherein the feature information includes alarm logs, processing time, operation logs, manual feedback information and effect evaluation information; The feature information is vectorized to obtain the working feature vector; The K most similar historical workflows are retrieved from the vector database using the aforementioned work feature vectors; Based on the K historical workflows, a directed acyclic graph is generated that defines the execution logic of the digital employee robot. The directed acyclic graph includes the digital employee name, triggering conditions, and execution chain, wherein the triggering conditions are used to determine the robot's start time, and the execution chain is used to determine the sequence of robot execution steps. Based on the triggering conditions and the execution chain, an artificial intelligence model is trained to generate the digital employee robot.

[0087] This method automatically selects suitable monitoring tasks for mechanization by using a threshold of work difficulty or workload, avoiding the generation of ineffective tasks. It extracts and quantifies multi-dimensional feature information, retrieves similar historical workflows from a vector database, reuses existing experience to generate a Directed Acyclic Graph (DAG), and then trains an artificial intelligence model based on the triggering conditions and execution chains in the DAG. This approach significantly reduces the generation cost and training difficulty of digital employee robots, improves generation efficiency and the interpretability of execution logic, while ensuring the robot's adaptability to specific monitoring tasks.

[0088] In some embodiments, 1030 includes: The digital employee robot described in the instructions performs the corresponding monitoring tasks on the backup database and obtains the task execution results; The accuracy, recall, and false alarm rates of the digital employee robot's work execution results were statistically analyzed. Determine whether the accuracy reaches a first preset threshold, whether the recall reaches a second preset threshold, and whether the false positive rate is lower than a third preset threshold; If all conditions are met, the digital employee robot is deemed to have met the standards and is approved for official launch. If the requirements are not met, the model is deemed substandard, triggering a retraining of the AI ​​model.

[0089] By offline evaluation of the accuracy, recall, and false positive rate of the digital employee robot in a backup database, and setting three independent thresholds to determine whether they meet the standards, the production risks that may arise from direct deployment are avoided, while ensuring the comprehensive performance of the robot. If the standards are not met, the model is retrained, forming a closed-loop optimization mechanism to continuously improve robot quality and ensure reliability and stability after official deployment.

[0090] In some embodiments, 1030 includes: The digital employee robot works alongside human employees in a shift system, with tasks assigned according to a preset schedule. During the rotation process, the work result data of the real employees and the digital employee robots are continuously collected. The work result data includes one or more of the following: alarm processing timeliness, work quality score, false alarm rate and false alarm rate. The collected work results data are input into the scheduling model, which uses XGBoost, LightGBM or random forest algorithms; The scheduling model outputs an optimized scheduling vector, which indicates the task allocation ratio between the real employees and the digital employee robots in each time period. The scheduling plan is updated based on the optimized scheduling vector to achieve dynamic scheduling of the real employees and the digital employee robot.

[0091] By having human employees and digital employee robots work in parallel and on a rotating basis, multi-dimensional work result data (timeliness, quality, false alarm rate, and false negative rate) is continuously collected. Optimized scheduling vectors are then output using ensemble learning models such as XGBoost, enabling dynamic adjustment of task allocation ratios. Compared to fixed scheduling, this method can adaptively optimize human-machine collaboration strategies based on actual work performance, fully leveraging the strengths of both parties, continuously improving overall monitoring efficiency and accuracy, while reducing the cost of manual intervention.

[0092] In some embodiments, 1030 further includes: When changes in employee turnover or alarm load fluctuations are detected, the scheduling model is automatically retrained and the scheduling plan is regenerated.

[0093] In this way, by monitoring changes in staff turnover rate and alarm load fluctuations in real time, the system automatically triggers the retraining of the scheduling model and the regeneration of the scheduling plan, enabling the system to dynamically adapt to changes in the external environment. When human resources decrease or alarm volume surges, the system can promptly adjust the task allocation ratio between human and digital employees, avoiding processing delays caused by staff shortages or sudden load increases, ensuring the continuity and stability of monitoring work, and reducing the operational costs of manual intervention in scheduling.

[0094] In some embodiments, 1030 further includes: Collect creative work characteristics of real employees in monitoring work, including methods for handling new types of alarms, records of developing new processing procedures, and operations for optimizing alarm rules; The creative work characteristics are processed to form characteristic data; When the accumulated amount of the characteristic data reaches a preset threshold, the characteristic data is used to trigger the digital employee robot generation system to train and generate new digital employees or optimize existing digital employee robots.

[0095] By proactively collecting behavioral records of real employees during creative tasks such as handling new types of alarms, developing new processes, and optimizing rules, these records are characterized and used as training data. When the accumulated data reaches a certain threshold, the generation or optimization of a digital employee robot is automatically triggered. This mechanism enables the system to continuously absorb the experience, innovation, and business evolution of real employees, breaking through the limitations of traditional models that become fixed after deployment. It achieves self-evolution and scenario self-adaptation of the robot, effectively improving the responsiveness and long-term applicability of the digital employee to new fault modes and new processing logic.

[0096] Figure 11 This is a schematic diagram illustrating an artificial intelligence-based device for monitoring the generation and use of digital employees, as shown in an embodiment of the present invention. Figure 11As shown, it includes: The first processing module is used to identify the monitoring work performed by real employees. The monitoring work includes at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm missed reporting analysis. The second processing module is used to start the digital employee robot generation system and train and generate a digital employee robot corresponding to the monitoring work when the monitoring work meets the preset machine generation conditions. The third processing module is used to conduct a compliance assessment on the generated digital employee robot. After the digital employee passes the assessment, it is officially launched and adopts a work rotation mode of real employees and digital employees. The work performance of the real employees and the digital employees is compared, and the shift schedule is dynamically adjusted based on the comparison results.

[0097] In some possible implementations, the first processing module is used to obtain the operation logs and work order processing records of real employees; The task type of the real employee is determined from the operation log and work order processing records; The task type is matched with the monitoring work type to determine the monitoring work currently being performed by the real employee.

[0098] In some embodiments, the second processing module is used to determine whether the monitoring work meets preset machine generation conditions, the preset machine generation conditions including the difficulty of the work exceeding a first threshold and / or the workload exceeding a second threshold. When the conditions are met, the feature information of the monitoring work is extracted, wherein the feature information includes alarm logs, processing time, operation logs, manual feedback information and effect evaluation information; The feature information is vectorized to obtain the working feature vector; The K most similar historical workflows are retrieved from the vector database using the aforementioned work feature vectors; Based on the K historical workflows, a directed acyclic graph is generated that defines the execution logic of the digital employee robot. The directed acyclic graph includes the digital employee name, triggering conditions, and execution chain, wherein the triggering conditions are used to determine the robot's start time, and the execution chain is used to determine the sequence of robot execution steps. Based on the triggering conditions and the execution chain, an artificial intelligence model is trained to generate the digital employee robot.

[0099] In some embodiments, the third processing module is used to instruct the digital employee robot to perform a monitoring task of the corresponding type on the backup database, and obtain the task execution result; The accuracy, recall, and false alarm rates of the digital employee robot's work execution results were statistically analyzed. Determine whether the accuracy reaches a first preset threshold, whether the recall reaches a second preset threshold, and whether the false positive rate is lower than a third preset threshold; If all conditions are met, the digital employee robot is deemed to have met the standards and is approved for official launch. If the requirements are not met, the model is deemed substandard, triggering a retraining of the AI ​​model.

[0100] In some embodiments, the third processing module is used to assign tasks to the digital employee robot to work in parallel with real employees in a shift manner, according to a preset shift schedule. During the rotation process, the work result data of the real employees and the digital employee robots are continuously collected. The work result data includes one or more of the following: alarm processing timeliness, work quality score, false alarm rate and false alarm rate. The collected work results data are input into the scheduling model, which uses XGBoost, LightGBM or random forest algorithms; The scheduling model outputs an optimized scheduling vector, which indicates the task allocation ratio between the real employees and the digital employee robots in each time period. The scheduling plan is updated based on the optimized scheduling vector to achieve dynamic scheduling of the real employees and the digital employee robot.

[0101] In some embodiments, the third processing module is configured to automatically trigger the retraining of the scheduling model and the regeneration of the scheduling plan when changes in the turnover rate of real employees or fluctuations in alarm load are detected.

[0102] In some embodiments, the second processing module is used to collect creative work characteristics of real employees in the monitoring work, the creative work characteristics including methods for handling new types of alarms, records of formulating new processing procedures, and operations for optimizing alarm rules; The creative work characteristics are processed to form characteristic data; When the accumulated amount of the characteristic data reaches a preset threshold, the characteristic data is used to trigger the digital employee robot generation system to train and generate new digital employees or optimize existing digital employee robots.

[0103] Those skilled in the art will readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is implemented in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the present invention.

[0104] It should be noted that, Figure 11 The division of modules / units is illustrative and represents only one logical functional division; in actual implementation, other division methods are possible. For example, two or more functions can be integrated into a single data acquisition module. The integrated modules described above can be implemented either in hardware or as software functional modules.

[0105] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned near-field channel device positioning methods for very large-scale antenna arrays. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute, by calling the computer program, a method for generating and using artificial intelligence-based digital employees according to any embodiment of the present invention.

[0106] In one alternative embodiment, an electronic device is provided, such as Figure 12 As shown, Figure 12 The illustrated electronic device 1200 includes a processor 1201 and a memory 1203. The processor 1201 and the memory 1203 are connected, for example, via a bus 1202. Optionally, the electronic device 1200 may further include a transceiver 1204, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 1204 is not limited to one type, and the structure of the electronic device 1200 does not constitute a limitation on the embodiments of the present invention.

[0107] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0108] It should be noted that, Figure 12 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0109] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any one of the above-mentioned methods for generating and using artificial intelligence-based digital employees.

[0110] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0111] It should be noted that the terms "first," "second," etc., used in the specification and claims of this invention are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of the invention described herein can be implemented in an order other than that shown or described.

[0112] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0113] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for generating and using digital employees based on artificial intelligence, characterized in that, include: Identify monitoring tasks performed by real employees, including at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm miss analysis. When the monitoring work meets the preset machine generation conditions, the digital employee robot generation system is activated to train and generate a digital employee robot corresponding to the monitoring work. The generated digital employee robot undergoes a compliance assessment. Once the digital employee robot passes the assessment, it is officially launched and a work mode of rotating between real employees and digital employee robots is adopted. The work performance of the real employees and the digital employee robot is compared, and the shift schedule is dynamically adjusted based on the comparison results.

2. The method according to claim 1, characterized in that, The monitoring tasks performed by real employees include: Obtain real employees' operation logs and work order processing records; The task type of the real employee is determined from the operation log and work order processing records; The task type is matched with the monitoring work type to determine the monitoring work currently being performed by the real employee.

3. The method according to claim 1, characterized in that, When the monitoring work meets preset machine generation conditions, the digital employee robot generation system is activated to train and generate a digital employee robot corresponding to the monitoring work, including: Determine whether the monitoring work meets the preset machine generation conditions, the preset machine generation conditions including the difficulty of the work exceeding a first threshold and / or the workload exceeding a second threshold; When the conditions are met, the feature information of the monitoring work is extracted, wherein the feature information includes alarm logs, processing time, operation logs, manual feedback information and effect evaluation information; The feature information is vectorized to obtain the working feature vector; The K most similar historical workflows are retrieved from the vector database using the aforementioned work feature vectors; Based on the K historical workflows, a directed acyclic graph is generated that defines the execution logic of the digital employee robot. The directed acyclic graph includes the digital employee name, triggering conditions, and execution chain, wherein the triggering conditions are used to determine the robot's start time, and the execution chain is used to determine the sequence of robot execution steps. Based on the triggering conditions and the execution chain, an artificial intelligence model is trained to generate the digital employee robot.

4. The method according to claim 1, characterized in that, The assessment of the generated digital employee robot to meet the required standards includes: The digital employee robot described in the instructions performs the corresponding monitoring tasks on the backup database and obtains the task execution results; The accuracy, recall, and false alarm rates of the digital employee robot's work execution results were statistically analyzed. Determine whether the accuracy reaches a first preset threshold, whether the recall reaches a second preset threshold, and whether the false positive rate is lower than a third preset threshold; If all conditions are met, the digital employee robot is deemed to have met the standards and is approved for official launch. If the requirements are not met, the model is deemed substandard, triggering a retraining of the AI ​​model.

5. The method according to claim 1, characterized in that, The method of employing a rotating shift system of real and digital employees, comparing the work performance of real and digital employees, and dynamically adjusting the shift schedule based on the comparison results includes: The digital employee robot works alongside human employees in a shift system, with tasks assigned according to a preset schedule. During the rotation process, the work result data of the real employees and the digital employee robots are continuously collected. The work result data includes one or more of the following: alarm processing timeliness, work quality score, false alarm rate and false alarm rate. The collected work results data are input into the scheduling model, which uses XGBoost, LightGBM or random forest algorithms; The scheduling model outputs an optimized scheduling vector, which indicates the task allocation ratio between the real employees and the digital employee robots in each time period. The scheduling plan is updated based on the optimized scheduling vector to achieve dynamic scheduling of the real employees and the digital employee robot.

6. The method according to claim 5, characterized in that, Also includes: When changes in employee turnover or alarm load fluctuations are detected, the scheduling model is automatically retrained and the scheduling plan is regenerated.

7. The method according to claim 1, characterized in that, Also includes: Collect creative work characteristics of real employees in monitoring work, including methods for handling new types of alarms, records of developing new processing procedures, and operations for optimizing alarm rules; The creative work characteristics are processed to form characteristic data; When the accumulated amount of the characteristic data reaches a preset threshold, the characteristic data is used to trigger the digital employee robot generation system to train and generate new digital employee robots or optimize existing digital employee robots.

8. A device for monitoring the generation and use of digital employees based on artificial intelligence, characterized in that, include: The first processing module is used to identify the monitoring work performed by real employees. The monitoring work includes at least one of the following: fault correlation analysis, fault level prediction, fault scheduling, work order tracking, fault escalation, and alarm missed reporting analysis. The second processing module is used to start the digital employee robot generation system and train and generate a digital employee robot corresponding to the monitoring work when the monitoring work meets the preset machine generation conditions. The third processing module is used to conduct a compliance assessment on the generated digital employee robot. After the digital employee robot passes the assessment, it is officially launched and adopts a work rotation mode of real employees and digital employee robots. The work effects of the real employees and the digital employee robots are compared, and the shift schedule is dynamically adjusted based on the comparison results.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the AI-based method for generating and using digital employees as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to implement the artificial intelligence-based method for generating and using digital employees as described in any one of claims 1 to 7.

Images