DRL-based power employee occupational path generation system and method
The DRL-based career path generation system for power industry employees solves the problems of low matching degree and rigidity in traditional path planning, realizes personalized and dynamic career path planning, and improves the matching degree of employee skills and the adaptability of enterprise talent reserves.
Patent Information
- Application Number
- CN202511230120.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-29
- Publication Date
- 2025-11-11
AI Technical Summary
Traditional power companies' career path planning for employees suffers from low matching, rigidity, high costs, and an inability to respond in real time to changes in employee status and corporate strategic adjustments. Furthermore, it fails to effectively incorporate new skills requirements, resulting in lagging employee skills updates and talent shortages.
A career path generation system for power industry employees based on deep reinforcement learning (DRL) is adopted. Through data collection, processing and model training, a state space, action space and reward function are constructed. The model is optimized by combining PPO algorithm to generate personalized and dynamic career path plans, and is optimized in real time to adapt to industry changes.
It achieves a precise match between employee skills and job requirements, shortens the promotion cycle, reduces the waste of training resources, ensures that the company has a talent pool to adapt to new technologies, and improves the accuracy and dynamism of career path planning.
Smart Images
Figure CN120930902A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of career path generation technology, specifically a career path generation system and method for power industry employees based on DRL. Background Technology
[0002] As a core area of national energy security, the power industry's talent cultivation is characterized by "high technical specialization, clear job levels, and close connection with technological iteration." Currently, career path planning for power company employees mainly relies on traditional methods, which has the following significant defects and shortcomings: 1. Traditional career paths are mostly based on a uniform template and do not take into account employees' skill gaps and career preferences, resulting in a low degree of matching between the path and employees' abilities, insufficient employee motivation for growth, and about 30% of employees experience job burnout due to unsuitable paths; 2. The power industry is accelerating its transformation towards new energy and intelligentization. The traditional path has not incorporated new skills requirements, resulting in employees' skills updates lagging behind job requirements. Enterprises face the risk of a gap between an oversupply of employees with old skills and a shortage of talents with new skills. A survey by a power grid company shows that 75% of operation and maintenance employees need additional training to adapt to the needs of smart substations. 3. Once a traditional path is determined, it is only statically adjusted every year or two, which cannot respond in real time to changes in employee status or corporate strategy adjustments, resulting in a rigid path; 4. Traditional path planning lacks a quantitative "cost-benefit" model. The arrangement of training programs and project experiences relies on subjective judgment, which may lead to problems such as high-cost training with limited skill improvement and a disconnect between project resource investment and employee growth. Statistics from a power company show that the resource waste rate of traditional training is 25%, and the average employee promotion cycle is as long as 8 years, which is far higher than the industry's best level. Summary of the Invention
[0003] The purpose of this invention is to provide a DRL-based career path generation system and method for power industry employees, in order to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: A career path generation system for power industry employees based on DRL includes: a data acquisition module, a data processing module, a DRL model training module, a career path planning module, and a visualization and interaction module. Data acquisition module: Collects basic employee information, skill data, and career preference data from internal or external data sources. Data processing module: processes the collected data; DRL Model Training Module: Constructs DRL models for the power industry, including: State space definition: S =[S emp , S pos , S env ],in S emp This represents the employee's current state vector (skill vector, performance score, career preference). S pos This is a job requirement state vector (job skill requirements, promotion conditions). S env External environment state vector (technology trend weights, corporate strategy priorities); Action space definition: A ={ A 1, A 2, A 3, A 4}, A 1 refers to a job adjustment action. A 2 refers to training and certification actions. A 3 represents the training activities for the project. A 4 represents actions that need optimization; Reward function design: R = αR short + β Rlong - γC ,in R short For short-term rewards, R long For long-term rewards, C To constrain enterprise costs, α , β , c These are the weighting coefficients; Model training: The PPO algorithm is used, with historical employee growth data of power companies as the training set. The policy network and value network are optimized offline iteratively to make the model converge to the "reward maximization" objective. Career path planning module: Based on predictive models, it analyzes industry development trends and future skill requirements to provide employees with career development paths and training suggestions; Visual interactive module: Displays career paths in chart form.
[0005] As a further embodiment of the present invention: the data acquisition module includes a data acquisition unit; the data processing module includes a data cleaning unit, a data normalization processing unit, and a data integration and transformation unit; the data normalization processing unit is used to normalize numerical features and also to encode categorical features; the data integration and transformation unit is used to merge data sources, integrate information from different data sources to form a complete dataset, and also to perform feature transformation, transforming or combining the original features.
[0006] As a further aspect of the present invention, the data acquisition module also includes an interface unit that supports connection with power company HR systems, ERP systems, skills assessment platforms, and industry databases.
[0007] As a further aspect of the present invention: in the DRL model training module, the policy network adopts a 3-layer fully connected neural network, with the input layer dimension being the state space dimension and the output layer dimension being the action space dimension.
[0008] As a further aspect of the present invention, the career path planning module also includes a constraint verification unit to verify whether the generated career path complies with the policies and regulations of the power industry. If it does not comply, the action space adjustment is triggered.
[0009] As a further aspect of the present invention: the basic employee information includes: age, education, major, educational background, and work experience; skills data includes: high-voltage operation skills, relay protection skills, intelligent inspection skills, and new energy operation and maintenance skills, etc.; occupational preference data includes job type and work location.
[0010] As a further aspect of the present invention: the career path planning module also includes a collaborative filtering unit. The collaborative filtering unit is based on a collaborative filtering algorithm. By analyzing the similarity between employees and positions, it provides users with personalized job recommendation results. At the same time, it constructs a user profile based on the employee's personal characteristics and historical behavior data, and uses a similarity measurement method to calculate the similarity between the employee profile and the job requirements, thereby measuring the degree of matching between the employee and the job. The results are based on the similarity calculation.
[0011] A method for generating career paths for power industry employees based on DRL includes the following steps: S1. Data Acquisition: The data acquisition module obtains individual employee data, job requirement data, and external environment data. Employee skill data is collected through standardized assessment tools in the power industry. S2. Data Processing: The data processing module performs numerical conversion, outlier removal, and time-series alignment on the collected data to generate a standardized dataset. D ={ D emp , Dpos , D env}; S3 and DRL model construction and training; S31. Define the state space S =[ S emp , S pos , S env Action space A ={ A 1, A 2, A 3, A 4} Designing a reward function based on the talent development goals of power enterprises R = αR short + βR long - γC ; S32. Using the PPO algorithm, with historical employee growth data as the training set, the training parameters are set as follows: learning rate 0.001, number of iterations 500 rounds, number of samples per round 256, and training until the model reward value converges. S4. Initial career path generation; S41. Standardize the dataset D Input the trained DRL model, and the model outputs the optimal action sequence for each stage through the policy network; S41. The constraint verification unit verifies the compliance of the path. If a violation is found, it is fed back to the model adjustment action to generate the final initial career path. P 0; S5. Real-time optimization of career paths; S51. Collect the latest employee status data and external environment update data every quarter; S52. The path generation and optimization module updates the data input into the DRL model, the model recalculates the state value and action reward, and outputs the optimized career path. P t , t To optimize the number of iterations, adjustment nodes are marked; S6. Iterative updates of the model and path; S61. Collect new data monthly through the update and iteration module, perform incremental training on the DRL model, and update the parameters of the policy network and value network. S61, the visual interaction module will optimize the path. P t Feedback is provided to employees and HR to support manual fine-tuning and path execution tracking, forming a closed loop of "data collection - model training - path generation - optimization iteration".
[0012] As a further aspect of the present invention, it also includes a model verification step: using 5-fold cross-validation, the historical data is divided into a training set and a test set to verify the matching degree between the generated path and the actual excellent employee path, requiring a matching degree of ≥85%; at the same time, the employee promotion cycle shortening rate after applying the model is statistically analyzed, requiring a shortening rate of ≥20%.
[0013] As a further aspect of the present invention: step S5 also uses time series analysis or regression analysis models to predict industry development trends and future skill requirements, plans career development paths for employees based on the prediction results, and recommends training courses and development directions in combination with job requirements and employees' potential abilities.
[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. By synchronizing the technology trends of the power industry in real time through external environmental data and adjusting the technology trend weights in the DR state space, the lag time between matching employee skills with new job requirements is shortened, and the shortage of new skilled personnel in enterprises is alleviated. 2. Replacing the cumbersome process of "manually creating path templates" by traditional HR, the system can generate personalized paths for a large number of employees in a short time and output visual reports through automatic data collection and automatic calculation of DRL models. 3. Through "precise action sequence planning", the promotion cycle from "junior position" to "senior position" for employees is significantly shortened; 4. By dynamically adapting to technological trends and optimizing pathways in real time, we ensure that enterprises always have a talent pool that can adapt to new technologies. 5. The reward function of the DRL model incorporates a cost constraint term (C) to quantify the relationship between training costs, project resource consumption, skill improvement, and promotion cycle. Attached Figure Description
[0015] Figure 1 This is an architecture diagram of a career path generation system for power industry employees based on DRL. Detailed Implementation
[0016] Please see Figure 1 A career path generation system for power industry employees based on DRL includes: a data acquisition module, a data processing module, a DRL model training module, a career path planning module, and a visualization and interaction module. Data acquisition module: Collects basic employee information, skill data, and career preference data from internal or external data sources. Employee basic information includes: age, education, major, educational background, and work experience; skills data includes: high-voltage operation skills, relay protection skills, intelligent inspection skills, and new energy operation and maintenance skills; career preference data includes job type and work location. Individual employee data: Basic information is obtained by connecting with the company's HR system; mastery of 15 core skills, such as high-voltage operation, relay protection, new energy operation and maintenance, and intelligent inspection, is collected through the "Power Industry Skill Level Assessment Standard"; project contribution and performance are obtained through the ERP system; and career preferences, such as management / technical inclination and job location preference, are collected through questionnaires. Job Requirements Data: We have compiled a hierarchical system for 10 core positions in the power industry, namely operation and maintenance, dispatching, repair, marketing, new energy, energy storage, digital operation and maintenance, safety, training, and management, such as junior → intermediate → senior → expert / supervisor. We have clearly defined the "skills requirements list" and promotion constraints for each level of position. For example, the skills requirements list for senior new energy positions requires mastery of "energy storage system commissioning" and "photovoltaic grid connection control". Promotion constraints include "dispatch supervisors need 5 years of dispatching experience + senior dispatching professional certification". External environment data: Connect with the State Grid technical standards library and the new energy industry report library to obtain the weight of technology trends, such as the weight of "smart grid technology" of 0.3 and the weight of "new energy operation and maintenance" of 0.25 in 2025. Combine with the company's strategic plan, such as "add 5 wind power projects within 3 years and need to reserve 100 wind power operation and maintenance talents", and then determine the environmental state vector. Data processing module: processes the collected data; DRL Model Training Module: Constructs DRL models for the power industry, including: State space definition: S =[ S emp , S pos , S env ],in S emp This represents the employee's current state vector, such as skill vector, performance score, and career preference. S pos This represents the job requirement state vector, such as job skill requirements and promotion criteria. S env External environment state vectors, such as technology trend weights and corporate strategic priorities; Action space definition: A ={ A 1, A 2, A 3, A 4}, A 1 refers to a job adjustment, such as from junior operations and maintenance (SEM) to intermediate operations and maintenance (EM).A 2 refers to training and certification activities, such as participating in new energy operation and maintenance training or obtaining a high-voltage electrician certificate. A 3. Project-based learning experiences, such as participating in intelligent inspection projects or being responsible for substation maintenance. A 4 represents actions that need optimization; Reward function design: R = αR short + β Rlong - γC ,in R short For short-term rewards (skill improvement rate, performance improvement rate). R long For long-term rewards, C To constrain corporate costs, such as training expenses and project resource consumption, α , β , c These are the weighting coefficients. α + β + c =1, dynamically adjusted based on enterprise needs; State space: A 30-dimensional vector is used, including 15 dimensions of employee state, namely 5 basic skills, 5 professional skills, 3 performance indicators and 2 career preferences; 10 dimensions of job requirements, namely 5 core skill requirements, 3 experience requirements and 2 certification requirements; and 5 dimensions of environment state, namely 3 technology trend weights and 2 corporate strategic priorities, which fully cover the three-dimensional information of "people-job-environment". Motion space: Design 4 core motion categories, each containing 5-8 sub-motions, for example: Job adjustments: Junior Operations and Maintenance → Intermediate Operations and Maintenance; Junior New Energy Position → Intermediate New Energy Position; Operations and Maintenance Position → Dispatch Position; Training and certification activities: High-voltage electrician certificate training, new energy operation and maintenance special training, intelligent inspection algorithm training, and advanced dispatching professional certification; Project experience included: participating in the maintenance of a 110kV substation, being responsible for a photovoltaic power station grid connection project, and leading the commissioning of an intelligent inspection system; Reward function: Balances "short-term improvement" and "long-term development" through weighted coefficients, while constraining enterprise costs, for example: Short-term reward R short Skill improvement rate and performance improvement rate: Skill improvement rate, such as relay protection skills improving from level 3 to level 4, earns 20 points; Performance improvement rate, such as quarterly performance improving from 75 points to 85 points, earns 15 points. Long-term reward R longImproved job matching and shortened promotion cycle: Improved job matching, such as increasing the current job matching from 60% to 80%, earns 30 points; shortened promotion cycle, such as reducing the expected promotion cycle from 6 years to 5 years, earns 25 points. Cost constraint C: Training expenses + project resource consumption. For example, if the training expense for new energy is 5,000 yuan, deduct 5 points. If the project resource consumption is the occupation of core project resources, deduct 3 points. The weighting coefficients α=0.3, β=0.5, and γ=0.2 can be adjusted according to the needs of the enterprise. For example, if short-term skills improvement is emphasized, α should be increased. Model training: The PPO proximal policy optimization algorithm is adopted, and the historical employee growth data of power companies is used as the training set. The policy network and value network are optimized offline iteratively to make the model converge to the "reward maximization" objective. Career path planning module: Based on predictive models, it analyzes industry development trends and future skill requirements to provide employees with career development paths and training suggestions; The career path planning module also includes a constraint verification unit, which verifies whether the generated career path complies with the policies and regulations of the power industry. If it does not comply, it triggers an adjustment of the action space. Initial path: generated in stages of 1-3 years, 3-5 years, and 5-10 years, clearly defining the "core objectives + action sequence" for each stage, such as the 1-3 year path for newly hired electrical engineering employees; Year 1: Junior maintenance position → Participate in high voltage electrician certificate training → Participate in daily inspection of 35kV substation; Year 2: Junior Operation and Maintenance Position → Participate in Relay Protection Specialized Training → Responsible for Substation Relay Protection Commissioning Projects; Year 3: Promotion to intermediate maintenance position → Participation in intelligent inspection technology training → Participation in the maintenance of 110kV intelligent substations; Real-time optimization: Collect the latest employee data every quarter. If the intelligent inspection skill reaches level 4 in the middle of the second year, the model will recalculate the action reward and adjust the path: If the employee has mastered the intelligent inspection skill in advance, the original intelligent inspection training in the third year will be adjusted to new energy operation and maintenance training to adapt to the needs of the company's new energy projects. Iterative updates: The model is trained monthly with incremental new data, and the state space weights are updated to ensure that the model adapts to industry changes; Visualization and Interaction Module: Displays career paths in chart form, such as time-series line graphs and skill-job matching heatmaps. It allows employees to query goals at each stage, such as the three core skills to be mastered within three years. It also supports HR departments in exporting career path planning reports and provides a manual intervention interface, such as allowing HR to fine-tune path nodes based on short-term needs of the company.
[0017] The data acquisition module includes a data acquisition unit, and the data processing module includes a data cleaning unit, a data normalization processing unit, and a data integration and transformation unit. The data normalization processing unit is used to normalize numerical features and also to encode categorical features. The data integration and transformation unit is used to merge data sources, integrate information from different data sources to form a complete dataset, and also to perform feature transformation, changing or combining the original features.
[0018] The data acquisition module also includes an interface unit that supports integration with power company HR systems, ERP systems, skills assessment platforms, and industry databases.
[0019] In the DRL model training module, the policy network adopts a 3-layer fully connected neural network, with the input layer dimension being the state space dimension and the output layer dimension being the action space dimension.
[0020] The career path planning module also includes a collaborative filtering unit. Based on the collaborative filtering algorithm, the collaborative filtering unit provides users with personalized job recommendations by analyzing the similarity between employees and positions. At the same time, it constructs user profiles based on employees' personal characteristics and historical behavior data, and uses similarity measurement methods to calculate the similarity between employee profiles and job requirements, measuring the degree of matching between employees and positions, and based on the similarity calculation results.
[0021] A method for generating career paths for power industry employees based on DRL includes the following steps: S1. Data Acquisition: The data acquisition module obtains individual employee data, job requirement data, and external environment data. Employee skill data is collected through standardized assessment tools in the power industry. S2. Data Processing: The data processing module performs numerical conversion, outlier removal, and time-series alignment on the collected data to generate a standardized dataset. D ={ D emp , D pos , D env}; S3 and DRL model construction and training; S31. Define the state space S =[ S emp , S pos , S env Action space A ={ A 1, A 2, A 3, A 4} Designing a reward function based on the talent development goals of power enterprisesR = αR short + βR long - γC ; S32. Using the PPO algorithm, with historical employee growth data as the training set, the training parameters are set as follows: learning rate 0.001, number of iterations 500 rounds, number of samples per round 256, and training until the model reward value converges. S4. Initial career path generation; S41. Standardize the dataset D Input the trained DRL model, and the model outputs the optimal action sequence for each stage through the policy network; S41. The constraint verification unit verifies the compliance of the path. If a violation is found, it is fed back to the model adjustment action to generate the final initial career path. P 0; S5. Real-time optimization of career paths; S51. Collect the latest employee status data and external environment update data every quarter; S52. The path generation and optimization module updates the data input into the DRL model, the model recalculates the state value and action reward, and outputs the optimized career path. P t , t To optimize the number of iterations, adjustment nodes are marked; S53. Furthermore, based on time series analysis or regression analysis models, it predicts industry development trends and future skill demands, and plans career development paths for employees based on the prediction results. Combining job requirements and employees' potential abilities, it recommends training courses and development directions. S6. Iterative updates of the model and path; S61. Collect new data monthly through the update and iteration module, perform incremental training on the DRL model, and update the parameters of the policy network and value network. S61, the visual interaction module will optimize the path. P t Feedback is provided to employees and HR to support manual fine-tuning and path execution tracking, forming a closed loop of "data collection - model training - path generation - optimization iteration".
[0022] It also includes a model validation step: using 5-fold cross-validation, historical data is divided into training and testing sets to verify the matching degree between the generated path and the actual path of outstanding employees, requiring a matching degree of ≥85%; at the same time, the rate of reduction in the employee promotion cycle after applying the model is statistically analyzed, requiring a reduction rate of ≥20%.
[0023] This invention system synchronizes with power industry technology trends in real time using external environmental data and updates the "technology trend weights" in the DRL state space; the lag time between matching employee skills and new job requirements is shortened, alleviating the shortage of new skilled personnel in enterprises; the DRL model automatically eliminates redundant skills training that is out of touch with industry trends through the "long-term value assessment" of the reward function; the system collects the latest employee data every quarter, triggering the DRL model to recalculate the optimal action sequence; the reward function of the DRL model incorporates a cost constraint (C), quantifying the correlation between training costs, project resource consumption, skills improvement, and promotion cycle; through "precise action sequence planning," the promotion cycle from "junior" to "senior" positions is significantly shortened; this design replaces traditional HR... The cumbersome process of "manually creating career path templates" is eliminated by the system. Through automatic data collection and DRL model calculation, the system can generate personalized career paths for a large number of employees in a short time and output visual reports. The system has a built-in constraint verification unit that strictly conforms to the safety regulations and job qualification requirements of the power industry. Through dynamic adaptation to technological trends and real-time path optimization, the system ensures that the company always has a talent reserve that can adapt to new technologies. This invention not only solves the problems of "accuracy and dynamism" in career path planning for individual employees of power companies, but also creates value for enterprises from the perspectives of "cost control, strategic support, and risk avoidance". At the same time, it provides core technical support for the entire power industry to adapt to energy transformation and achieve standardized talent training.
[0024] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A career path generation system for power industry employees based on DRL, characterized in that, include: Data acquisition module, data processing module, DRL model training module, career path planning module, and visualization interaction module; Data acquisition module: Collects basic employee information, skill data, and career preference data from internal or external data sources. Data processing module: processes the collected data; DRL Model Training Module: Constructs DRL models for the power industry, including: State space definition: S =[ S emp , S pos , S env ],in S emp Let the current state vector of the employee be... S pos Let the job requirement state vector be... S env This represents the external environment state vector. Action space definition: A ={ A 1, A 2, A 3, A 4}, A 1 refers to a job adjustment action. A 2 refers to training and certification actions. A 3 represents the training activities for the project. A 4 represents actions that need optimization; Reward function design: R = αR short + β Rlong - γC ,in R short For short-term rewards, R long For long-term rewards, C To constrain enterprise costs, α , β , γ These are the weighting coefficients. α + β + γ =1; Model training: The PPO (Proximal Policy Optimization) algorithm is adopted, using historical employee growth data (career path records of 5 years or more) from power companies as the training set. The policy network and value network are optimized offline iteratively to make the model converge to the "reward maximization" objective. Career path planning module: Based on predictive models, it analyzes industry development trends and future skill requirements to provide employees with career development paths and training suggestions; Visual interactive module: Displays career paths in chart form.
2. The DRL-based career path generation system for power industry employees according to claim 1, characterized in that, The data acquisition module includes a data acquisition unit; the data processing module includes a data cleaning unit, a data normalization processing unit, and a data integration and transformation unit; the data normalization processing unit is used to normalize numerical features and also to encode categorical features; the data integration and transformation unit is used to merge data sources, integrate information from different data sources to form a complete dataset, and also to perform feature transformation, changing or combining the original features.
3. The DRL-based career path generation system for power industry employees according to claim 1, characterized in that, The data acquisition module also includes an interface unit that supports integration with power company HR systems, ERP systems, skills assessment platforms, and industry databases.
4. The DRL-based career path generation system for power industry employees according to claim 1, characterized in that, In the DRL model training module, the policy network adopts a 3-layer fully connected neural network, with the input layer dimension being the state space dimension and the output layer dimension being the action space dimension.
5. A career path generation system for power industry employees based on DRL according to claim 1, characterized in that, The career path planning module also includes a constraint verification unit, which verifies whether the generated career path complies with the power industry's policies and regulations. If it does not comply, it triggers an action space adjustment.
6. The DRL-based career path generation system for power industry employees according to claim 1, characterized in that, The employee's basic information includes: age, education, major, educational background, and work experience; skills data includes: high-voltage operation skills, relay protection skills, intelligent inspection skills, and new energy operation and maintenance skills; career preference data includes job type and work location.
7. A career path generation system for power industry employees based on DRL as described in claim 1, characterized in that, The career path planning module also includes a collaborative filtering unit. Based on the collaborative filtering algorithm, the collaborative filtering unit provides users with personalized job recommendations by analyzing the similarity between employees and positions. At the same time, it constructs user profiles based on employees' personal characteristics and historical behavior data, uses similarity measurement methods to calculate the similarity between employee profiles and job requirements, and measures the degree of matching between employees and positions based on the similarity calculation results.
8. A method for generating career paths for power industry employees based on DRL, characterized in that, Includes the following steps: S1. Data Acquisition: The data acquisition module obtains individual employee data, job requirement data, and external environment data. Employee skill data is collected through standardized assessment tools in the power industry. S2. Data Processing: The data processing module performs numerical conversion, outlier removal, and time-series alignment on the collected data to generate a standardized dataset. D ={ D emp , D pos , D env }; S3 and DRL model construction and training; S31. Define the state space S =[ S emp , S pos , S env Action space A ={ A 1, A 2, A 3, A 4} Designing a reward function based on the talent development goals of power enterprises R = αR short + βR long - γC ; S32. Using the PPO algorithm, with historical employee growth data as the training set, the training parameters are set as follows: learning rate 0.001, number of iterations 500 rounds, number of samples per round 256, and training until the model reward value converges. S4. Initial career path generation; S41. Standardize the dataset D Input the trained DRL model, and the model outputs the optimal action sequence for each stage through the policy network; S41. The constraint verification unit verifies the compliance of the path. If a violation is found, it is fed back to the model adjustment action to generate the final initial career path. P 0; S5. Real-time optimization of career paths; S51. Collect the latest employee status data and external environment update data every quarter; S52. The path generation and optimization module updates the data input into the DRL model, the model recalculates the state value and action reward, and outputs the optimized career path. P t , t To optimize the number of iterations, adjustment nodes are marked; S6. Iterative updates of the model and path; S61. Collect new data monthly through the update and iteration module, perform incremental training on the DRL model, and update the parameters of the policy network and value network. S61, the visual interaction module will optimize the path. P t Feedback is provided to employees and HR to support manual fine-tuning and path execution tracking, forming a closed loop of "data collection - model training - path generation - optimization iteration".
9. A method for generating career paths for power industry employees based on DRL according to claim 8, characterized in that, It also includes a model validation step: using 5-fold cross-validation, historical data is divided into training and testing sets to verify the matching degree between the generated path and the actual path of outstanding employees, requiring a matching degree of ≥85%; at the same time, the rate of reduction in the employee promotion cycle after applying the model is statistically analyzed, requiring a reduction rate of ≥20%.
10. A method for generating career paths for power industry employees based on DRL according to claim 8, characterized in that, Step S5 also uses time series analysis or regression analysis models to predict industry development trends and future skill requirements, plans career development paths for employees based on the prediction results, and recommends training courses and development directions based on job requirements and employees' potential abilities.