Artificial intelligence-based professionalism course optimization method, system, device and medium

Abstract

This application relates to an artificial intelligence-based method, system, device, and medium for optimizing vocational skills courses. The method constructs a skill map by decomposing vocational skills, configuring the initial state of skill nodes based on individual diagnosis, and using cognitive dynamic parameters describing forgetting, consolidation, and other characteristics. This constructs a learner dynamic cognitive state model simulating skill decay and the impact of learning. Then, using this model as a state transition engine, a Markov decision process is constructed, and a forward-looking optimal learning strategy is obtained through reinforcement learning. This strategy, combined with the current cognitive state, is used for multi-step simulation to generate a personalized learning path that meets resource constraints and is converted into a schedule. Finally, learning feedback is collected to update the state and calibrate parameters, achieving closed-loop optimization. This method achieves forward-looking globally optimal planning for the learning process of composite skills, thereby effectively improving the long-term retention level of composite abilities and overall learning efficiency at a limited cost.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence-driven education technology, and in particular relates to a method, system, device and medium for optimizing vocational literacy courses based on artificial intelligence. Background Technology

[0002] In the current field of vocational education and lifelong learning, adaptive learning systems have been widely explored and applied. Their mainstream technology is typically based on memory models such as the Ebbinghaus forgetting curve, aiming to schedule reviews by predicting the forgetting threshold of knowledge points. These systems are mostly optimized for isolated, homogeneous knowledge points. Their core logic is to trigger review prompts when memory strength decays to a specific threshold, representing a passive, "retrospective" intervention model. However, when learning objectives shift to complex, multifaceted professional skills such as "data analysis skills" and "project management skills," existing technologies reveal significant limitations.

[0003] First, professional competence comprises a variety of heterogeneous skills, including statistical knowledge, software operation skills, and business thinking frameworks. Its acquisition and degradation are not simply a binary "memory-forgetting" process, but rather involve multiple stages such as "knowledge understanding," "skill solidification," and "proficient application." A single forgetting curve model cannot depict the crucial "solidification" process of transforming short-term memory into long-term stable skills through practice. Second, existing systems often optimize each individual skill independently, ignoring the correlation and synergistic effects between skills. They fail to plan from a holistic perspective of the overall robustness of "composite abilities," potentially leading to uneven skill development and gaps in competence. Finally, "retrospective" review strategies are essentially reactive resource allocation, lacking prediction of learners' long-term competence and forward-looking planning based on cost constraints. This makes it difficult to make optimal review and practice combinations within limited time and energy budgets, ensuring the long-term maintenance of overall composite abilities.

[0004] Therefore, there is an urgent need for a new method that can model the dynamic evolution of multiple skills and carry out forward-looking intelligent planning with the goal of maintaining composite capabilities in the long term, in order to address the shortcomings of existing technologies in terms of learning efficiency, long-term effects and optimal resource allocation. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, equipment, and medium for optimizing professional competence courses based on artificial intelligence to address the aforementioned technical problems.

[0006] Firstly, this application provides a method for optimizing vocational skills courses based on artificial intelligence, including:

[0007] S1. Decompose the domain knowledge of the target professional competence into skills to obtain the skill decomposition results; construct a skill map including multiple skill nodes and their relationships based on the skill decomposition results; configure the initial proficiency and initial solidification of each skill node in the skill map based on the learner's performance data on the ability diagnosis task, and configure a set of cognitive dynamic parameters for each skill node; among which, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills;

[0008] S2. Based on skill maps, initial proficiency, initial solidification, and cognitive dynamics parameters, a learner dynamic cognitive state model is constructed to simulate the evolution of skill states and the impact of learning activities.

[0009] S3. Based on the learner's dynamic cognitive state model, a Markov decision process is constructed with skill state as input and learning action as output. The Markov decision process is solved by deep reinforcement learning algorithm to obtain the optimal learning planning strategy.

[0010] S4. Based on the learner's current cognitive state, invoke the optimal learning planning strategy to perform multi-step sequence decision simulation, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule.

[0011] S5. Collect real-time feedback data on learners' execution of the learning schedule, update learners' cognitive state based on real-time feedback data and calculate the model prediction error of learners' dynamic cognitive state model; calibrate cognitive dynamic parameters based on model prediction error, and use the calibrated cognitive dynamic parameters to dynamically optimize learners' dynamic cognitive state model and optimal learning planning strategy.

[0012] Secondly, this application also provides an artificial intelligence-based vocational skills curriculum optimization system for implementing the method described in the first aspect, the system comprising:

[0013] The skill graph construction and parameter configuration module is used to decompose the domain knowledge of the target professional competence into skills, and obtain the skill decomposition results. Based on the skill decomposition results, a skill graph including multiple skill nodes and their relationships is constructed. Based on the learner's performance data on the ability diagnosis task, the initial proficiency and initial solidification of each skill node in the skill graph are configured, and a set of cognitive dynamic parameters are configured for each skill node. Among them, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills.

[0014] The dynamic cognitive state modeling module is used to construct a learner's dynamic cognitive state model based on skill graphs, initial proficiency, initial solidification, and cognitive dynamics parameters to simulate the evolution of skill states and the impact of learning activities.

[0015] The optimal learning strategy generation module is used to construct a Markov decision process with skill state as input and learning action as output based on the learner's dynamic cognitive state model, and solve the Markov decision process through a deep reinforcement learning algorithm to obtain the optimal learning planning strategy.

[0016] The personalized path planning module is used to simulate multi-step sequence decisions by calling the optimal learning planning strategy based on the learner's current cognitive state, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule.

[0017] The adaptive feedback optimization module is used to collect real-time feedback data of learners executing their learning schedules, update learners' cognitive states based on the real-time feedback data, and calculate the model prediction error of the learner's dynamic cognitive state model. Based on the model prediction error, the cognitive dynamics parameters are calibrated, and the calibrated cognitive dynamics parameters are used to dynamically optimize the learner's dynamic cognitive state model and the optimal learning planning strategy.

[0018] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement an artificial intelligence-based vocational literacy curriculum optimization method as described in the first aspect.

[0019] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an artificial intelligence-based vocational skills curriculum optimization method as described in the first aspect.

[0020] The aforementioned AI-based method, system, equipment, and medium for optimizing vocational skills courses first construct a skill map by structurally decomposing the target vocational skills. Based on individual diagnostic data, each skill node is configured with cognitive dynamic parameters and an initial state, including forgetting, consolidation, learning efficiency, and synergistic relationships. Then, a learner dynamic cognitive state model is constructed to simulate the natural decay of skills and the impact of learning activities. This model is then used as a state transition engine to construct and solve a Markov decision process with skill state as input and learning action as output, thereby obtaining an optimal strategy for proactively planning the learning sequence. This strategy, combined with the current cognitive state, is then used for multi-step simulation to generate a personalized learning path that meets actual resource constraints and is converted into a schedule. Finally, learning feedback is collected to update the cognitive state in real time and calibrate the model parameters, forming a closed loop of "modeling-planning-execution-optimization." This method collectively achieves proactive, personalized, and globally optimal path planning for the learning process of composite vocational skills, thereby significantly improving the long-term retention level of composite abilities and overall learning efficiency under limited time and energy constraints. Attached Figure Description

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

[0022] Figure 1 A flowchart illustrating a method for optimizing vocational skills courses based on artificial intelligence, provided by this invention.

[0023] Figure 2 This is a schematic diagram of the process of constructing an action influence model in one optional embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the structure of a professional competence curriculum optimization system based on artificial intelligence provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] refer to Figure 1 The document presents a flowchart illustrating an artificial intelligence-based method for optimizing vocational skills courses, as provided in this application. This method includes the following steps:

[0027] S1. Decompose the domain knowledge of the target professional competence into skills to obtain the skill decomposition results; construct a skill map including multiple skill nodes and their relationships based on the skill decomposition results; configure the initial proficiency and initial solidification of each skill node in the skill map based on the learner's performance data on the ability diagnosis task, and configure a set of cognitive dynamic parameters for each skill node; among which, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills.

[0028] Specifically, this step involves deconstructing the skills of the target professional competencies, constructing the skill map, configuring initial state parameters, and calibrating cognitive dynamics parameters, laying the foundation for subsequent model construction and strategy optimization. First, a domain knowledge framework for the target professional competencies is constructed using domain ontology tools, clarifying the core competency dimensions and their subordinate skill units. For typical target professional competencies (such as data analysis skills, project management skills, etc.), they are first broken down into multiple core competency dimensions, and then each dimension is further subdivided into sub-skills.

[0029] For example, the core connotations of the "Longjiang Spirit" (hard work, pioneering spirit, prioritizing the overall situation, and selfless dedication) can be combined with additional skill units related to the practice of the spirit during the skill deconstruction process. That is, specific skill units that reflect the requirements of practicing the Longjiang Spirit can be deconstructed separately (such as the ability to adapt to harsh environments, the ability to work together to overcome difficulties, the ability to innovate and break through, and the ability to practice the overall awareness). This ensures that the skill deconstruction not only covers the core professional needs but also takes into account the cultivation and practice goals of the Longjiang Spirit.

[0030] During the decomposition process, multiple domain experts were invited to conduct multiple rounds of review and revision of the skill decomposition results. The first round of review focused on verifying the completeness of the skill units to ensure no core skills were omitted; the second round focused on the independence of the skill units to avoid functional overlap between different skill units; the third round confirmed the rationality of the skill unit hierarchy to ensure that the decomposition granularity is adapted to the needs of subsequent learning status assessment and path planning. The final result is a skill decomposition result containing information such as skill ID, skill name, skill description, related competency dimension, prerequisite skill requirements, and skill difficulty coefficient. This result also references skill decomposition cases from similar professional competency courses and industry job competency standards.

[0031] During the review process, experts conducted additional checks on the rationality and suitability of the special skills units related to the Heilongjiang Spirit, confirmed the logical connection between the special skills units and the vocational skills units, ensured that the spirit-related skills units were not divorced from the professional scenario and could be cultivated in parallel through vocational skills learning, and verified whether the key points of practicing the Heilongjiang Spirit were incorporated into the skill descriptions, ensuring that the skill decomposition results were highly consistent with the goals of cultivating the Heilongjiang Spirit.

[0032] After the skills are decomposed, a skill graph is constructed based on the results. The skill graph is mathematically modeled using a directed weighted graph structure, and its expression is: .in For each vertex set, there is a set of all skill nodes. Each vertex corresponds to a skill unit. Vertex attributes include skill difficulty coefficient and skill importance weight. The skill difficulty coefficient ranges from 0 to 1 and is determined by domain experts based on the difficulty of mastering the skill and the learning time. The skill importance weight ranges from 0 to 1 and is determined based on the skill's contribution to the target professional competence.

[0033] To highlight the core objective of cultivating the Heilongjiang spirit, an additional "Heilongjiang spirit association weight" can be added to the vertex attributes of the skill map. This weight ranges from 0 to 1 and is used to quantify the degree of connection between each skill unit and the practice of the Heilongjiang spirit (e.g., the association weight of innovation and breakthrough skills, collaborative skills, and overall awareness skills is higher than that of basic operational skills). This weight is determined by domain experts in combination with the professional characteristics, spiritual connotations, and professional skill requirements of the Heilongjiang region, providing data support for balancing spirit cultivation and skill enhancement in subsequent learning path planning.

[0034] The set of edges represents the relationships between skill nodes. The types of relationships include three categories: prior relationships, collaborative relationships, and complementary relationships. Each edge corresponds to one type of relationship. Let be the weight set, representing the association strength, with values ​​ranging from 0 to 1. This set is determined by combining domain expert scores with statistical data on the association between previous learners' skill mastery and existing skills. For preceding association edges, the association strength is assessed based on the impact of prior skill mastery on subsequent skill learning. Simultaneously, by incorporating sufficient learners' skill learning sequence data, the correlation of skill mastery is calculated using the Pearson correlation coefficient to adjust the expert-assessed weights. The adjustment formula can be... . In the formula, The corrected correlation strength, The initial association strength as assessed by experts. The Pearson correlation coefficient is used to quantify the degree of linear correlation between two skill mastery levels, ensuring that the correlation strength can truly reflect the mutual influence between skills; the coefficients (0.6 and 0.4) in the formula can be adjusted according to actual needs.

[0035] Once the skill graph is constructed, the association structure is presented using visualization tools, the rationality of the association logic is verified, and the association structure is optimized and adjusted.

[0036] Next, we will configure the initial proficiency and initial solidification levels. First, we will design a targeted ability diagnostic task system, which includes three types of diagnostic content: theoretical tests, practical tasks, and scenario simulations. The theoretical tests will combine objective and subjective questions to cover the key knowledge points of each skill unit. The practical tasks will be designed based on real business scenarios and will require learners to complete them in designated data processing software. The scenario simulation tasks will require learners to comprehensively apply multiple skills to solve practical problems in complex business scenarios. The scenario designs will be based on typical professional scenarios.

[0037] All diagnostic content can be integrated into the evaluation dimensions of the practice of the Heilongjiang Spirit. The theoretical test includes questions related to the core connotations of the Heilongjiang Spirit (hard work, pioneering spirit, etc.); the practical tasks prioritize typical occupational scenarios in the Heilongjiang region (such as agricultural production optimization, industrial technology breakthroughs, grassroots service support, etc.), simultaneously assessing practical skills and the ability to practice the spirit, such as adapting to harsh environments and efficient collaboration; the scenario simulation task design focuses on the practice of the spirit and the application of skills in complex scenarios, such as task advancement under limited resources and tackling urgent projects, with a focus on assessing the core qualities of pioneering innovation and prioritizing the overall situation.

[0038] The learning management system collects learners' performance data on diagnostic tasks, including scores for each skill unit, task completion time, number of errors, and number of repetitions. Initial proficiency and initial retention are calculated based on these indicators. Initial proficiency is calculated using a weighted scoring method, with the formula: ,in Initial proficiency level, The total number of diagnostic indicators. For the first The standardized scores of the diagnostic indicators, ranging from 0 to 1, are obtained through min-max standardization. The standardization formula is as follows: , For the first The weight of each indicator is determined based on the ability of the diagnostic content to represent the level of skill mastery.

[0039] When collecting data, additional indicators related to the practice of the Heilongjiang Spirit can be added (such as tolerance to harsh environments, contribution to teamwork, effectiveness of innovative solutions, and manifestation of overall awareness). These indicators can be incorporated into the initial proficiency calculation system. The weight of each spirit indicator is determined according to its importance to the cultivation of professional qualities, ensuring that the initial proficiency reflects both the level of skill mastery and the foundation for practicing the Heilongjiang Spirit.

[0040] Initial consolidation degree is used to characterize the extent to which a skill transitions from short-term memory to long-term stable skill. It is calculated based on the stability and timeliness of learners' diagnostic task performance, using the following formula: .in Initial degree of curing These are weighting coefficients, determined by domain experts. The standard deviation of learners' scores when they complete the same diagnostic task multiple times. The average score, This indicates the degree of score fluctuation; the smaller the fluctuation, the more stable the skill mastery and the higher the degree of solidification. The average time it takes learners to complete a diagnostic task. The maximum allowed time for diagnosing this skill's task. The time taken to complete a task indicates the level of proficiency and the degree of skill solidification.

[0041] Finally, cognitive dynamics parameters are configured for each skill node; this parameter set includes the forgetting coefficient. Curing rate Learning efficiency factor Synergistic influence coefficient Four core parameters. Among them, the forgetting coefficient... This value describes the natural decay of skill proficiency over time, ranging from 0 to 1. It is determined based on the skill type and difficulty coefficient. Theoretical skills have a higher forgetting coefficient than practical skills, and skills with higher difficulty coefficients have a greater forgetting coefficient. The calculation formula is as follows: ,in The basal forgetting coefficient is set separately for theoretical skills and practical skills. This represents the skill difficulty level.

[0042] Curing rate This term describes the rate at which skill solidification increases with learning and practice activities. The value ranges from 0 to 0.1 per learning activity. It is positively correlated with skill proficiency; the higher the proficiency, the faster the solidification rate. The calculation formula is as follows: ,in Based on the curing rate, Initial proficiency. Learning efficiency factor. This value describes the efficiency with which learners improve skill proficiency through learning activities. It ranges from 0 to 1 and is determined by considering the learner's historical learning efficiency, skill difficulty, and type of learning activity. The calculation formula is as follows: ,in The learner's historical average learning efficiency is calculated based on the ratio of the increase in proficiency to the learning time in past learning activities. The skill difficulty level is... The adaptation coefficient is used for learning activities, and the adaptation coefficient varies for different types of learning activities.

[0043] Synergistic Influence Coefficient Used to describe skills skills The degree of synergistic impact, ranging from -1 to 1, with positive numbers indicating positive synergy, i.e., skills. Enhancement and promotion of skills Learning and maintenance; negative numbers represent negative interference, i.e., skills. Over-emphasis may inhibit skills The mastery level is zero, indicating no significant impact. This coefficient is determined through a combination of domain expert scores and regression analysis of correlation data with learners' skill mastery. Expert scores use a specific scoring system, which is standardized and converted into values ​​from -1 to 1. This is then corrected using linear regression with historical data to obtain the final result. The specific value.

[0044] During the expert scoring process, the synergistic effect between skills related to the Heilongjiang spirit and professional skills can be given priority consideration. For example, the positive synergy between skills related to hard work and perseverance and practical skills, the positive synergy between innovative skills and project management skills, and the positive synergy between skills related to overall awareness and teamwork skills can be ensured. This will ensure that the synergistic impact coefficient reflects both the professional relevance between skills and the role of skills related to the Heilongjiang spirit in promoting vocational skills learning, so that the coefficient calibration results are more in line with the needs of vocational quality cultivation in the Heilongjiang region.

[0045] S2. Based on skill maps, initial proficiency, initial solidification, and cognitive dynamics parameters, a learner dynamic cognitive state model is constructed to simulate the evolution of skill state and the impact of learning activities.

[0046] Specifically, the learner dynamic cognitive state model simulates the evolution of learners' skill states at different time points and the mechanism by which learning activities affect skill states. It overcomes the limitations of existing single forgetting curve models in that they cannot characterize the skill solidification process and the synergistic effect of multiple skills. The model is built on the core principle of cognitive state evolution.

[0047] The model is constructed using a state-space modeling approach, defining the learner's cognitive state as a vector composed of the proficiency and solidification of all skill nodes, i.e., a state vector. .in for The cognitive state vector at time step, The total number of skill nodes. For the first Each skill node is Proficiency at all times For the first Each skill node is The degree of solidification at any given moment For time variables, This represents the matrix transpose operation.

[0048] The core of the model lies in constructing state evolution equations to describe the dynamic changes in proficiency and fixedness over time and learning activities. For the evolution of proficiency, the model comprehensively considers three major factors: natural forgetting, improvement through learning activities, and the synergistic effect of multiple skills, and constructs the following evolution equation: . In the formula, For the first Each skill node is Proficiency at all times For time step, For the first The forgetting coefficient of each skill node is derived from the cognitive dynamics parameters of step S1. For the first The learning efficiency factor for each skill node also comes from the cognitive dynamics parameters of step S1. for At any time for the first The intensity of learning activities for each skill node ranges from 0 to 1 and is determined by the duration and depth of the learning activity. For skills skills The synergistic influence coefficient is derived from the cognitive dynamics parameters of step S1. for Moment Skills The change in proficiency is calculated using the following formula: ; This is a multi-skill synergy impact factor, among which The total number of skill nodes. Indicates the absence of skills All outside Skill Nodes Summation is performed to quantify the impact of changes in proficiency of other skills on the current skill. The overall impact of proficiency. The physical meaning of this equation is: skill exist Proficiency in timing begins with... The system calculates proficiency at any given moment by subtracting natural forgetting, then adds the proficiency gains from learning activities, and finally adjusts this through a multi-skill synergy factor, reflecting the interaction between skills. This evolutionary logic aligns with the core principle of dynamic changes in skill proficiency while overcoming the shortcomings of traditional models that only consider forgetting factors.

[0049] The evolution of skill solidification focuses on depicting the transformation process from short-term proficiency to long-term stability, and its evolution equation is as follows: .in, For the first Each skill node is The degree of solidification at any given moment For the first The solidification rate of each skill node is derived from the cognitive dynamics parameters of step S1. The proficiency threshold for skill solidification is determined by domain experts, and the value can range from 0.6 to 0.8. The higher the difficulty of the skill, the higher the threshold. This means that learning activities will only promote the improvement of solidification when the skill proficiency reaches the threshold; when the threshold is not reached, the solidification remains basically stable. For the first The attenuation coefficient of a skill node's solidification level can range from 0.005 to 0.01 per day. The higher the solidification level, the smaller the attenuation coefficient. The calculation formula is as follows: This term describes the slight decay of skills when there has been no targeted learning over a long period of time.

[0050] To ensure model accuracy, association constraints based on the skill graph are introduced during the construction process. For skills with prior associations... and ,in for Prerequisite skills, if skills proficiency Below the pre-threshold (This prerequisite threshold is determined by domain experts), then the skill Learning efficiency factor It will be reduced by a specific percentage, the percentage of which will depend on the skill level. The difference between the proficiency level and the prerequisite threshold is used to determine the constraint effect of the prerequisite skill on the learning of subsequent skills. The constraint rules are formulated based on the influence mechanism of the prerequisite skill on the learning of subsequent skills.

[0051] After the model is built, initial calibration is performed. The initial proficiency obtained in step S1 is then used. With initial curing degree Substitute the state vector By combining cognitive dynamics parameters and the intensity of the initial learning activity (which is set to 0 when no targeted learning occurs in the initial stage), the evolutionary trend of cognitive state in the initial state is calculated using an evolutionary equation. The prediction error is then calculated by comparing the prediction with actual skill change data from a small sample of learners; the root mean square error is used to determine the prediction error. Its calculation formula is ,in For the proficiency of model prediction, For the level of proficiency in actual observation, This represents the number of samples.

[0052] like If the value exceeds a set threshold, the baseline coefficients in the cognitive dynamics parameters, including the baseline forgetting coefficient, will be adjusted. Basic curing rate The process continues until the prediction error meets the requirements, ensuring that the model can accurately simulate the learner's cognitive state evolution. During calibration, the sample data is selected from real-world datasets showing changes in learners' skills, improving the model's practicality and reliability.

[0053] S3. Based on the learner's dynamic cognitive state model, a Markov decision process is constructed with skill state as input and learning action as output. The Markov decision process is then solved using a deep reinforcement learning algorithm to obtain the optimal learning planning strategy.

[0054] Specifically, this step involves constructing a Markov decision process and solving for the optimal learning planning strategy using a deep reinforcement learning algorithm. This enables forward-looking decision-making aimed at maintaining comprehensive professional competence in the long term, overcoming the limitations of existing technologies that focus on independently optimizing single skills and lack global planning. The decision framework design is based on a core planning process driven by reinforcement learning.

[0055] First, based on the learner's dynamic cognitive state model constructed in step S2, the four core elements of a Markov decision process are defined: state space, action space, reward function, and state transition probability. State space Defined as the combination of the learner's cognitive state vector and the associated states of the skill map, its expression is: ,in for The cognitive state vector at any given moment includes the proficiency and solidification of each skill node. It is the associated state vector of the skill graph, which includes the association relationship and association strength of each skill node.

[0056] The dimension of the state space is ,in The total number of skill nodes. Let be the total number of skill-related edges, and let each state satisfy the Markov property, i.e. The state at any given moment depends only on The state at any given moment and the action being performed are independent of historical states. This characteristic ensures the rationality and computability of the decision-making process, which aligns with the core assumptions of Markov decision processes.

[0057] Action space Defined as all possible combinations of learning activities, each action A set of learning activities for each skill node is given by the following expression: .in In order to target the The learning activity intensity of each skill node ranges from 0 to 1, consistent with the definition of learning activity intensity in step S2, and satisfies resource constraints. ,in In order to target the The time resources required for a unit strength learning activity for each skill node. This represents the maximum available learning time resource per day.

[0058] The size of the action space depends on the total number of skill nodes. To ensure the feasibility and relevance of the actions, a discretization approach is adopted, dividing each... The values ​​are discrete into multiple levels. At the same time, invalid actions are filtered out through the association constraints of the skill graph. If the prerequisite skills are not met, action combinations that require high-intensity learning of subsequent skills are filtered out. The action filtering rules are formulated based on the feasibility judgment criteria of learning activities.

[0059] reward function This is the core of Markov decision process construction. Its design goal is to guide the reinforcement learning agent to learn the optimal strategy that can maintain composite professional competence in the long term. Therefore, it adopts a design approach that combines immediate rewards and long-term rewards. The total reward formula is: .in For the total reward, , These are the weighting coefficients. It serves as an immediate reward, used to evaluate the short-term improvement effect of the current action on cognitive state. This is a long-term reward used to evaluate the contribution of current actions to the long-term maintenance of comprehensive professional competence.

[0060] Instant rewards The calculation is based on the change in cognitive state after the action is performed, and the formula is: .,in The total number of skill nodes. For the first The proficiency weight of each skill node is consistent with the skill importance weight in step S1. For the first The solidification weight of each skill node can take the following values: Emphasizing the importance of improving the degree of curing. , Skills after the action is executed The change in proficiency and solidification is recorded as 0 if the change is negative, ensuring that the instant reward is non-negative.

[0061] Long-term rewards The calculation is based on the future The formula for the comprehensive score of the step's composite ability is as follows: .in, To predict the step size (i.e., to predict the future) (Evolution of cognitive states of the day) The total number of skill nodes. The step size index is predicted, and its value ranges from 1 to... , This is a resource consumption penalty coefficient, used to penalize actions that excessively consume time and resources. For the first The comprehensive score of the day's composite abilities For the future The average composite ability score over the days indicates a better long-term maintenance effect and a greater long-term reward. The time and resources consumed by the current action are considered; the more time and resources consumed, the greater the penalty and the smaller the long-term reward.

[0062] State transition probability Used to describe the state Next action After transitioning to state The probability is derived from the learner's dynamic cognitive state model in step S2. Since the cognitive state evolution equation contains a certain degree of randomness, including slight fluctuations in learner learning efficiency and external interference factors, a random perturbation term is introduced. The disturbance term follows a pattern with a mean of 0 and a variance of 0. The normal distribution follows a given pattern, and the formula for calculating the state transition probability is: . In the formula, Let be the probability density function of the normal distribution. For state The corresponding cognitive state vector, The predicted value of the cognitive state evolution equation in step S2 is the state vector without random perturbation terms (for each action). (Corresponding to a set of learning activities for each skill level). Let be the covariance matrix of the random disturbance term, and its diagonal elements are: The off-diagonal elements are 0. This probability distribution can characterize the uncertainty of cognitive state transition after the action is performed. The probability modeling method is based on the statistical characteristic analysis of state transition.

[0063] Specifically, state transition probability The specific calculation formula is as follows:

[0064]

[0065] in, Let be the probability density function of the normal distribution, and its complete expression is: ,in A vector of random variables Dimensions Covariance matrix The determinant, Covariance matrix The inverse matrix, This is the natural exponential function. Its core parameters include a vector of random variables. Mean vector and covariance matrix It is used to characterize the uncertainty distribution of cognitive state transition after an action is performed.

[0066] In the modeling scenario involving state transition probabilities, the corresponding relationships of the parameters are as follows: Random variable vector Corresponding state The core cognitive state vector components, namely ,in For the first Each skill node is Proficiency observations at any given time. For the first Each skill node is The solidification observation at time, the dimension of this vector ( The total number of skill nodes constitutes the core of the random variable that forms the state transition probability; the associated states of the skill graph have no random perturbation, and their transition probability is 1, so they are not included in the normal distribution modeling.

[0067] Mean vector The state prediction vector corresponding to the normal distribution, i.e. The equation for the evolution of cognitive states in step S2 is calculated as follows: For the first The evolution prediction function for the proficiency of each skill node (corresponding to the proficiency evolution equation in step S2) is expressed as follows: ; For the first The evolution prediction function for the solidification degree of each skill node (corresponding to the solidification degree evolution equation in step S2) is expressed as follows: , For action China's target The intensity of learning activities for each skill node.

[0068] covariance matrix Covariance matrix corresponding to random disturbance term ,Right now Matrix dimension and random variable vector Consistency, for The diagonal elements are all (Variance of random disturbance term), all off-diagonal elements are 0, indicating that the random disturbances of proficiency and solidification of each skill node are independent of each other and have no cross interference.

[0069] Substituting the corresponding parameters into the complete expression of the normal distribution probability density function yields the state transition probability. This method quantifies the transition probabilities of each cognitive state after an action is performed, providing a reliable probabilistic basis for solving subsequent Markov decision processes. The probability distribution characterizes the uncertainty of cognitive state transitions after an action is performed by superimposing independent normal perturbations on the deterministic predictions of cognitive state evolution as the mean. Its modeling method is based on the statistical characteristics of state transitions, ensuring the rationality and accuracy of the probability calculations.

[0070] After constructing the four key elements of a Markov decision process, the proximal policy optimization algorithm from deep reinforcement learning is used to solve for the optimal learning plan policy. This algorithm features high sample efficiency and stable training, and is suitable for decision problems with complex state and action spaces. The core of the proximal policy optimization algorithm is the construction of a policy network and a value network, where the policy network... Used based on the current state Output Action The probability distribution, For the parameters of the policy network; value network Used based on the current state Predicting the expected future cumulative rewards, These are the parameters of the value network.

[0071] The policy network employs a three-layer fully connected neural network structure, with the input layer dimension being the state space dimension, i.e. The first network has two hidden layers, each with a specific number of neurons, and uses the ReLU activation function. The output layer has the dimension of the number of discrete actions in the action space, and uses the Softmax activation function to output the probability of selecting each action. The second network, the value network, also uses a three-layer fully connected neural network structure. Its input layer has the same dimension as the policy network, and it has two hidden layers, each with a specific number of neurons, using the ReLU activation function. The output layer has a dimension of 1 and outputs the predicted value of the state. The network structure is built upon the design framework of reinforcement learning networks.

[0072] The training process is divided into two stages: sample collection and parameter update. In the sample collection stage, the agent updates the parameters based on the current policy network. In a Markov decision process environment, sample state-action-reward-next state quadruples, i.e. An experience replay buffer is constructed, with its capacity set according to a specific scale. The buffer is updated using a first-in, first-out (FIFO) approach to ensure the timeliness of the samples. In the parameter update phase, the value network is first used as the basis... The objective value function is calculated using time-series difference error correction, and the formula is as follows: ,in For target value, for Instant rewards for each moment This is a discount factor used to balance the weights of current and future rewards. The predicted value for the next state.

[0073] Then, the policy gradient is calculated, and the value network parameters are updated by minimizing the mean squared error between the objective function and the value network output. For policy network parameters The update uses a trimming objective function, the formula of which is: .in, Let the loss function be the policy network. For the expectation operator, This represents the probability ratio between the current policy and the old policy. The action probabilities output by the old policy network. for The dominance function at time 1 is calculated using the generalized dominance estimation method, and the formula is as follows: ,in The step size for advantage estimation, For timing difference error, The pruning factor is used to limit the step size of policy updates and avoid training instability.

[0074] During training, training parameters such as learning rate, batch size, and number of training iterations are set. After each iteration, the average cumulative reward of the policy is calculated. When the average cumulative reward no longer increases after several consecutive iterations and the increase is less than a set threshold, training is stopped. The policy corresponding to the policy network parameters at this point is the optimal learning planning policy. This strategy can output the optimal combination of learning activities based on the current cognitive state, achieving the dual goals of maintaining comprehensive professional competence in the long term and optimizing resource allocation. The training effect is verified based on the evaluation indicators of reinforcement learning strategies.

[0075] S4. Based on the learner's current cognitive state, invoke the optimal learning planning strategy to perform multi-step sequence decision simulation, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule.

[0076] Specifically, based on the optimal learning planning strategy obtained in step S3, and combined with the learner's current cognitive state and resource constraints, this paper generates a personalized learning path through multi-step sequential decision simulation and transforms it into an executable learning schedule. This solves the problem of the lack of personalization and foresight in the learning path in the existing technology. The path generation process is based on the implementation framework of personalized learning planning.

[0077] First, the learner's current cognitive state is acquired and preprocessed. The latest cognitive state data of the learner is collected through the learning management system, including the current proficiency of each skill node. Current degree of curing ,in This includes the current time and historical learning activity records, such as the intensity, duration, and skill improvement effects of learning activities over a past period. These data are then substituted into the learner's dynamic cognitive state model from step S2 to correct the current cognitive state. The correction uses a Kalman filter algorithm, and the formula is as follows: . In the formula, This is the corrected current cognitive state vector. This is the vector of the current cognitive state predicted by the model. The Kalman gain is calculated using the following formula: , The prediction error covariance matrix; The observation matrix is ​​an identity matrix. To observe the noise covariance matrix, its diagonal elements are set with specific values. The cognitive state vector is calculated based on the latest feedback data and is obtained from actual observations. This correction process ensures the accuracy of the current cognitive state, laying the foundation for subsequent decision simulation. The correction method is implemented based on the core process of state data calibration.

[0078] Next, a multi-step sequential decision-making simulation is performed, with a simulation time span of... The time frame is determined based on the learning objectives, typically set as short-term, medium-term, or long-term, with a time step. The time step is consistent with the cognitive state evolution model, i.e., 1 day. During the simulation, each iteration is based on the current state. ,in from arrive Invoke the optimal learning planning strategy Output optimal action This refers to the combination of learning activity intensity for each skill node.

[0079] After invoking an action, it is necessary to verify whether the resource constraints are met. ,in The total number of skill nodes. for Moment Skills The intensity of learning activities For skills The time consumed by unit intensity learning activities This represents the maximum daily learning time resource. If the constraints are met, the action is executed, calculated using the cognitive state evolution equation. state of time If the constraints are not met, the action will be adjusted by reducing the intensity of the learning activity according to the skill importance weight from high to low until the resource constraints are met. The adjustment formula is as follows: ,in, To adjust the total time spent on all previous skill learning activities, To ensure the adjusted intensity of learning activities remains optimal and feasible.

[0080] The core of multi-step sequence decision simulation is to generate a priority sequence for skill learning. Based on the correlation of skill graphs and the current cognitive state, a greedy strategy combined with long-term prediction is used to determine the core skills to be learned each day. Specifically, skills whose current proficiency is below the target threshold, which contribute significantly to composite abilities, and which have strong synergy with other skills are prioritized for focused learning. The target proficiency is determined by occupational requirements and can range from 0.85 to 0.95. The target proficiency of core skills and auxiliary skills are set according to different standards: a high contribution to composite abilities corresponds to a high skill importance weight, and strong synergy with other skills corresponds to a large synergy impact coefficient. The priority determination criteria are based on an evaluation system for skill learning priorities.

[0081] During the simulation, a dynamic adjustment mechanism for learning outcomes can be introduced. If the improvement in proficiency of a certain skill falls below a set threshold for several consecutive days, the type of learning activity for that skill will be adjusted, including changing from theoretical learning to practical practice, or from single-skill practice to comprehensive application in various scenarios. At the same time, the learning efficiency factor will be adjusted. Appropriately increase the value of this factor to promote effective improvement in skill proficiency. If the solidification degree of a certain skill reaches the target solidification degree (the target solidification degree can range from 0.9 to 0.95), then reduce the intensity of learning activities for that skill and allocate the released resources to other skills that have not yet reached the target, thereby achieving dynamic optimization of resource allocation.

[0082] After the multi-step sequence decision simulation is completed, a personalized learning path is generated. This learning path is presented in the form of a time series, including information such as the core learning skills, learning activity type, learning activity intensity, and learning duration for each time step. To ensure the feasibility of the learning path, it needs to be transformed into a standardized learning schedule. The learning schedule is formulated according to the principles of time fragmentation and rationality, allocating daily learning activities by time slots. The learning duration of each time slot does not exceed a specific duration to avoid learning fatigue, and the number of consecutive times of the same type of learning activity does not exceed a specific number of time slots.

[0083] The learning schedule can also include information on the learning resources, such as the software tools for practical exercises, the textbook chapters for theoretical reviews, and the business case numbers for scenario applications, to ensure that learners can directly carry out learning activities based on the learning schedule, and that the matching of learning resources is based on relevant case libraries.

[0084] S5. Collect real-time feedback data on learners' execution of the learning schedule, update learners' cognitive state based on real-time feedback data and calculate the model prediction error of learners' dynamic cognitive state model; calibrate cognitive dynamic parameters based on model prediction error, and use the calibrated cognitive dynamic parameters to dynamically optimize learners' dynamic cognitive state model and optimal learning planning strategy.

[0085] Specifically, this step collects real-time feedback data from learners during the learning process to update cognitive states, calculate model prediction errors, calibrate cognitive dynamics parameters, and dynamically optimize models and strategies. This ensures that the entire course optimization method can adapt to learners' learning progress and changes, overcoming the limitations of fixed models and rigid strategies in existing technologies.

[0086] First, real-time feedback data is collected. Collection channels include the learning management system's operation logs, learners' learning behavior records, interim test results, skills practice completion status, and subjective feedback. The collection frequency is both real-time collection and daily summaries. Learning behavior data is collected in real-time, while test results and practice completion status are summarized daily. The core data items collected include learning activity execution data, skills performance data, and learner status data. Learning activity execution data covers actual learning time, learning activity type, completion progress, and error records. Skill performance data covers interim test scores, practical task completion quality, skills application accuracy, and completion time. Learner status data covers learning fatigue ratings, subjective satisfaction ratings, and feedback on learning difficulties.

[0087] Error operation records are captured through the learning platform's operation logs. The practical process for each skill is broken down step-by-step, recording the error type and frequency for each step. Error types include parameter setting errors, command invocation errors, and logical errors. Skill application accuracy is calculated based on task completion results in actual application scenarios, using the following formula: The number of correctly completed tasks refers to the number of tasks successfully completed by learners in actual application scenarios, and the total number of tasks refers to the total number of tasks in actual application scenarios. The learning fatigue score is submitted by learners after each day's learning through a specific scoring system, and after standardization, it is converted into a value of 0-1. The data collection dimensions are set based on the core indicators of feedback data collection.

[0088] After real-time feedback data collection, data preprocessing is performed, including three stages: data cleaning, data standardization, and data fusion. Data cleaning employs outlier detection and missing value imputation methods. Outlier detection uses the 3σ principle or box plot method, where the 3σ principle is suitable for normally distributed data, and the box plot method is suitable for non-normally distributed data. Data exceeding a reasonable range are identified as outliers and removed. Missing value imputation uses mean imputation or mode imputation methods, where mean imputation is suitable for numerical data, and mode imputation is suitable for categorical data, ensuring data integrity.

[0089] Data standardization employs the min-max standardization method, transforming all numerical data into a value range of 0-1 to facilitate subsequent calculations and analysis. Data fusion uses a weighted fusion method, combining data on the same indicator collected from different channels. The fusion formula is as follows: ,in The total number of data channels for collecting this indicator, The merged index value For the first The weight of each data channel is determined based on the reliability of the data. For the first The original indicator values ​​for each channel are used, and the data preprocessing process is implemented based on standard data processing steps.

[0090] After data preprocessing, the learners' cognitive states are updated based on the processed data. The cognitive state update uses a Bayesian estimation method, combining the predicted values ​​from the dynamic cognitive state model in step S2 with the observed values ​​from real-time feedback data to calculate the posterior cognitive state. For skills... The proficiency update formula is: . In the formula, For the updated skills Proficiency The confidence level of the model prediction is represented by a value ranging from 0 to 1. The initial value is set according to a specific value and is gradually adjusted as the number of model calibrations increases. Skills for model prediction Proficiency The value ranges from 0 to 1, representing the reliability of the observed data, and is determined based on the reliability of data acquisition. The observation proficiency, calculated based on real-time feedback data, is derived by weighting indicators such as test score, practical accuracy, and completion time. The calculation formula is as follows: ,in The total number of indicators for calculating observation proficiency (here) (corresponding to standardized test scores, practical accuracy, and time-related indicators). To test standardized scores, To improve the accuracy of practical operations, The test standardized score is the value of the original test score after min-max standardization, and the time consumption standardized score is the value of the original time consumption data after min-max standardization.

[0091] Regarding skills The curing degree is updated, and the formula is: .in, For the updated skills degree of solidification, Skills for model prediction degree of solidification, To update the coefficients, which are used to balance the weights of predicted and observed values, The observational solidification degree, calculated based on real-time feedback data, is derived from indicators such as skill performance stability and learning activity continuity. Its calculation formula can be: ,in The standard deviation of scores for recent skill performances. The average score is calculated by taking the number of consecutive learning days (the number of consecutive days in which the skill has been studied) and the total learning days (the total number of days in which the skill has been studied).

[0092] After the cognitive state is updated, the model prediction error of the learner's dynamic cognitive state model is calculated. The prediction error is evaluated using a combination of two indicators: root mean square error (RMSE) and mean absolute percentage error (MAS), ensuring the comprehensiveness of the error assessment. The formula for calculating RMSE is as follows: ,in The total number of skill nodes. The root mean square error, Skills for model prediction Proficiency skills for observation Proficiency, an indicator used to measure the overall deviation between predicted and observed values.

[0093] The formula for calculating the mean absolute percentage error is: ,in The total number of skill nodes. The mean absolute percentage error (MASE) is used to measure the relative deviation of predicted values, avoiding inaccurate error assessments due to differences in the magnitude of the indicators. Typically, thresholds are set for both the root mean square error (RMSE) and the MSE. If either the calculated RMSE or MSE exceeds the set threshold, the model's prediction accuracy is insufficient, requiring calibration of the cognitive dynamics parameters. If both indicators meet the threshold requirements, only the cognitive state needs updating, without parameter calibration. The error assessment criteria are based on relevant indicators for model accuracy evaluation.

[0094] Cognitive dynamics parameter calibration employs gradient descent to minimize model prediction error; here, root mean square error is used as the optimization objective. This adjusts the core cognitive dynamics parameters from step S1, including the forgetting coefficient. Curing rate Learning efficiency factor Synergistic influence coefficient During calibration, the root mean square error is used as the loss function. ,in Given the cognitive dynamics parameter vector to be calibrated, the parameters are updated by calculating the partial derivatives of the loss function with respect to each parameter and then updating them along the negative direction of the partial derivatives. The parameter update formula is as follows: . In the formula, For the updated parameter values, The parameter values ​​before the update. To calibrate the learning rate and prevent instability caused by excessively rapid parameter updates, Let be the partial derivative of the loss function with respect to the old parameters. For the forgetting coefficient... For calibration, the formula for calculating the partial derivative is: ,in The total number of skill nodes. To predict proficiency for the model The partial derivatives are derived from the proficiency evolution equation in step S2, and their expression is: (in Indicates the absence of skills All skill nodes outside (Sum), determined by calculating the partial derivative. The update direction and step size are adjusted until the root mean square error is reduced to below the threshold.

[0095] After parameter calibration, the learner's dynamic cognitive state model is updated using the calibrated cognitive dynamics parameters. The state evolution equation is reconstructed, the original parameter values ​​are replaced, and the deep reinforcement learning policy is retrained based on the updated model. The proximal policy optimization algorithm in step S3 is used, and the number of training iterations is a certain proportion of the original number of iterations to ensure that the policy can quickly adapt to the updated model and achieve dynamic optimization of the model and policy.

[0096] Furthermore, to ensure the continuity of dynamic optimization, a periodic calibration mechanism is established. The model prediction error is comprehensively evaluated at specific intervals. Even if the error meets the threshold requirements, the cognitive dynamics parameters are slightly adjusted by a certain percentage of the original parameter values ​​to adapt to the long-term changing trends of learners' cognitive states, including improvements in learning ability and changes in learning habits. After dynamic optimization, based on the updated cognitive state, the optimized model, and the optimal learning planning strategy, the multi-step sequence decision simulation in step S4 is re-executed to generate new personalized learning paths and schedules. This achieves closed-loop adaptive adjustment of the course optimization method, ensuring that the entire method continuously adapts to learners' learning progress and always guides learners to improve their comprehensive professional skills in the best way, while achieving optimal allocation of resources such as time and energy. The closed-loop adjustment mechanism is built upon the overall framework of adaptive optimization.

[0097] The aforementioned AI-based method for optimizing vocational skills courses first constructs a skill map by structurally decomposing the target vocational skills. Based on individual diagnostic data, it configures cognitive dynamics parameters and initial states for each skill node, including forgetting, consolidation, learning efficiency, and synergistic relationships. Then, it builds a learner dynamic cognitive state model that simulates the natural decay of skills and the impact of learning activities. This model is then used as a state transition engine to construct and solve a Markov decision process with skill state as input and learning action as output, thereby obtaining an optimal strategy for proactively planning the learning sequence. This strategy, combined with the current cognitive state, is then used for multi-step simulation to generate personalized learning paths that meet actual resource constraints and are translated into a schedule. Finally, learning feedback is collected to update the cognitive state in real time and calibrate the model parameters, forming a closed loop of "modeling-planning-execution-optimization." This method achieves proactive, personalized, and globally optimal path planning for the learning process of composite vocational skills, significantly improving the long-term retention level of composite abilities and overall learning efficiency under limited time and energy constraints.

[0098] In one optional embodiment, a learner dynamic cognitive state model is constructed based on skill maps, initial proficiency, initial solidification, and cognitive dynamics parameters to simulate the evolution of skill states and the impact of learning activities. This includes the following steps:

[0099] S11. Based on all skill nodes in the skill graph, and combining initial proficiency and initial solidification, construct the learner's initial cognitive state vector.

[0100] Specifically, each skill node in the skill graph corresponds to a pair of core state parameters: initial proficiency (representing the initial mastery of the skill) and initial solidification (representing the initial stability of the skill). These two parameters of all skill nodes are arranged sequentially according to the skill node number to form an initial cognitive state vector. ,in The total number of skill nodes in the skill graph, with vector dimension . , For the first The initial proficiency of each skill node, For the first The initial solidification level of each skill node.

[0101] S12. Based on the parameters characterizing forgetting in cognitive dynamics, construct the first hyperbolic equation for calculating the natural decay of skill proficiency during periods without learning activities.

[0102] Specifically, considering that the natural forgetting of skill proficiency does not decay linearly, but rather follows a hyperbolic pattern of "rapid decay in the early stages, followed by a gradual slowdown and stabilization of the decay rate in the later stages," which more accurately reflects the actual forgetting characteristics compared to a linear decay model, a hyperbolic form is used to construct the equation. The specific expression of the first hyperbolic equation can be: ,in For the first Each skill node in duration (Time step, set to 1 day) The natural decay of proficiency when there is no learning activity within a time step; a negative sign indicates a decrease in proficiency. For the first The forgetting coefficient of each skill node is determined by the skill type and difficulty coefficient; for theoretical skills... The higher the value, the greater the difficulty level. The larger the value; The hyperbolic correction coefficient for proficiency decay was calibrated by domain experts using a large amount of experimental data on learners' skill forgetting, and is used to adjust the curvature of the hyperbola. For the first Each skill node is Proficiency at all times; The time step is consistent with the time step of the overall cognitive state evolution model. The principle of this equation is based on the hyperbolic denominator term. Constrained decay rate, as As the denominator increases, the absolute value of the decay gradually increases, and the rate of increase of the decay gradually slows down, eventually tending towards... This aligns with the actual pattern of long-term forgetting of skill proficiency.

[0103] S13. Based on the parameters characterizing the decay characteristics of solidification in cognitive dynamics, a second hyperbolic equation is constructed to calculate the natural decay of skill solidification during periods without learning activities.

[0104] Specifically, the natural decay of skill solidification is slower than the forgetting of proficiency, and it also exhibits a hyperbolic decay characteristic. Therefore, a hyperbolic form is used to construct the equation to ensure simulation accuracy. The specific expression of the second hyperbolic equation can be: ,in For the first Each skill node in duration The natural decay of curing degree when there is no learning activity inside; a negative sign indicates a decrease in curing degree. For the first The attenuation coefficient of solidification for each skill node; the higher the solidification, the greater the attenuation. The smaller the value; The hyperbolic correction factor for cure degree decay is greater than... This is to reflect the characteristic of slower decay of curing degree; For the first Each skill node is The degree of solidification at any given moment; The time step is 1 day. The logic of this equation is... The term reflects the basic characteristics of hardening fading – the lower the hardening degree, the greater the fading potential. At the same time, by constraining the fading rate through the hyperbolic denominator, it ensures that the hardening degree can still maintain a certain basic level after long-term fading, which is consistent with the actual process of skill slowly fading from long-term memory.

[0105] S14. Based on the parameters in cognitive dynamics that characterize the synergistic relationship between learning efficiency and skills, construct an action impact model to calculate the state improvement of the target skill and other related skills after performing a specific learning activity.

[0106] Specifically, the action influence model covers not only the target skill being directly learned, but also all related skills in the skill graph that have a relationship with the target skill (pre-requisite association, synergistic association, complementary association), ensuring a comprehensive depiction of the chain effects of learning activities. The specific expression of the action influence model consists of two parts: the calculation of the state improvement of the target skill and the calculation of the state improvement of related skills. The overall expression can be: , .in, For the first The amount of proficiency increase for each skill node (including the target skill and related skills). For the first The increase in the solidification level of each skill node; For the first The learning efficiency factor for each skill node is determined by combining the learner's historical learning efficiency, skill difficulty, and type of learning activity. for At any time for the first The intensity of learning activities for each skill node (target skill) Values ​​greater than 0 indicate skills that were not directly learned. (Value is 0) For skills skills The synergistic influence coefficient ranges from -1 to 1, with positive numbers indicating positive synergy (skills). Learning promotes skills (Improvement), negative numbers indicate negative interference (skills) Overlearning inhibition skills (Increase), zero indicates no significant impact; For the first The rate at which a skill node is solidified is positively correlated with its initial proficiency. The proficiency threshold for skill solidification; The model ensures that learning activities only promote skill consolidation when skill proficiency reaches a threshold; otherwise, the increase in consolidation is zero, aligning with the objective laws of skill consolidation. The core principle of this model is to quantify the direct improvement effect through a learning efficiency factor, quantify the indirect chain effect through a synergistic influence coefficient, and constrain the conditions for consolidation improvement through consolidation rate and proficiency thresholds, ensuring the rationality and accuracy of the calculated improvement amount.

[0107] S15. Sum the natural decay amount described by the first hyperbolic equation and the second hyperbolic equation with the state elevation amount described by the action influence model to construct a state transition equation; based on the state transition equation and the initial cognitive state vector, construct a learner dynamic cognitive state model.

[0108] Specifically, the equations are first integrated by algebraically summing the natural decay of proficiency described by the first hyperbolic equation and the increase in proficiency described by the action influence model, resulting in the core formula for the state transition of proficiency. Similarly, the natural decay of fixedness described by the second hyperbolic equation and the increase in fixedness described by the action influence model are algebraically summed to obtain the core formula for the state transition of fixedness. The complete state transition equations are as follows: , .in, , The first Each skill node is The level of proficiency and consistency at any given moment. , They are respectively Proficiency and solidification at each moment. The above state transition equation is applied to the initial cognitive state vector constructed in S11. Arbitrary time steps can be obtained through iterative calculation. Cognitive state vector ,Right now ,in The state transition function is composed of the aforementioned proficiency and solidification transition equations. Specifically, it integrates the proficiency and solidification transition formulas for each skill node as core components, arranged according to the dimensional order of the cognitive state vector, and inputs the current cognitive state vector. and learning activity intensity vector Then, the proficiency and solidification of all skill nodes in the next time period can be output simultaneously, forming a complete [system / process]. The core value of this learner dynamic cognitive state model lies in its ability to accurately simulate the natural decline of the state when there is no learning activity and the state improvement process when there is learning activity by integrating the hyperbolic decay equation and the action influence model. At the same time, it covers the chain influence of the target skill and related skills. Compared with traditional single-skill state simulation models, it has higher fit and comprehensiveness, and can effectively support the optimization and construction of subsequent learning planning strategies.

[0109] In an optional embodiment, the first hyperbolic equation is constructed based on the instantaneous forgetting rate, consolidation inhibition coefficient, and proficiency adjustment coefficient for any skill node in the cognitive dynamics parameters, and the expression of the first hyperbolic equation is:

[0110]

[0111] in, Indicates the time of the i-th skill proficiency Indicates the time of the i-th skill The degree of curing, Indicates a time interval without learning activities. It is a very small positive number that prevents the denominator from being zero. This represents the solidification inhibition coefficient of the i-th skill. This represents the proficiency adjustment coefficient for the i-th skill. The first hyperbolic equation represents the instantaneous forgetting rate of the i-th skill; it is used to calculate future moments based on current proficiency and stagnation. Proficiency prediction value .

[0112] Specifically, This represents the proficiency of the i-th skill at time t, indicating the degree of mastery of the skill at that time. Δt represents the degree of solidification of the i-th skill at time t, characterizing the extent to which the skill transforms from short-term memory into a long-term stable mastery state; Δt represents the time interval without learning activities, that is, the duration interval between two learning activities, usually set in days (e.g., Δt=1 means 1 day without learning activities).

[0113] This represents the consolidation inhibition coefficient of the i-th skill, and its function is to quantify the inhibitory effect of consolidation on the forgetting process—the higher the consolidation degree (…). The larger ( The smaller, the better After exponentiation, the more significant the decay of the overall numerator, the slower the forgetting rate, reflecting the inhibitory effect of solidification on forgetting.

[0114] This represents the proficiency adjustment coefficient for the i-th skill, used to adjust the impact of the current proficiency on the forgetting rate—when... At that time, the lower the proficiency ( The smaller the denominator, the better. The smaller the value, the larger the overall forgetting rate coefficient, meaning that low-proficiency skills are forgotten relatively faster; when When proficiency modulates the rate of forgetting linearly; when... At that time, the influence of proficiency on the forgetting rate weakens, adapting to the characteristic that the forgetting rate of highly proficient skills tends to be stable.

[0115] This represents the instantaneous forgetting rate of the i-th skill, characterizing the base natural forgetting rate of skill proficiency in the absence of consolidation inhibition and proficiency modulation. It is determined statistically from measured learner skill forgetting data. The instantaneous forgetting rate differs between different skill types (theoretical and practical). The instantaneous forgetting rate for theoretical skills... The value can be higher than that of practical skills.

[0116] The first hyperbolic equation passes through the numerator term. The inhibitory effect of quantified solidification on forgetting is measured through the denominator term. The moderating effect of current proficiency on the forgetting rate is quantified, and the two are combined to form a forgetting rate correction coefficient, which is then compared with the instantaneous forgetting rate. Multiply the time interval Δt between periods of no learning activity to obtain the natural decay of proficiency during the period of no learning activity. Finally, use the proficiency at the current time. Subtracting this attenuation amount yields the predicted proficiency value at the future time t+Δt. Compared to traditional linear forgetting models, this equation more accurately reflects the actual patterns of skill forgetting—simultaneously considering the inhibitory effect of fixation and the moderating effect of proficiency. The prediction results are more consistent with the objective reality of learners' skill state evolution, and can effectively support the construction and optimization of subsequent cognitive state models.

[0117] refer to Figure 2 In one optional embodiment, based on parameters in cognitive dynamics that characterize the synergistic relationship between learning efficiency and skills, an action impact model is constructed to calculate the state improvement of the target skill and other related skills after performing a specific learning activity, including the following steps:

[0118] S21. Based on the learning efficiency coefficients of different learning actions for skills in cognitive dynamics parameters, construct calculation rules for the direct improvement of proficiency in target skills.

[0119] Specifically, let's first explain the definition of the learning efficiency coefficient. For the i-th target skill, each learning action corresponds to a unique learning efficiency coefficient. Typical learning activities include theoretical memorization, practical exercises, and scenario simulations, among which... Indicates the learning action type number, The value range is determined by domain experts based on the actual situation, taking into account the relevance of the learning action, the difficulty of skill mastery, and historical learning performance data. The core logic of the direct proficiency improvement calculation rule is that the amount of direct proficiency improvement in the target skill is positively correlated with the intensity of the learning action, the learning duration, and the learning efficiency coefficient. The specific calculation expression is as follows:

[0120]

[0121] in, This represents the direct increase in proficiency of target skill i. This represents the learning efficiency coefficient for the k-th type of learning action corresponding to the i-th target skill; This represents the intensity of performing the k-th type of learning action for skill i at time t, and its value is determined quantitatively by indicators such as the proportion of action execution time and focus. Indicates the duration of the learning action; This represents the current proficiency adjustment function, used to adapt to the differences in learning efficiency at different proficiency stages; Current proficiency adjustment function A piecewise function can be used, and the specific expression can be:

[0122]

[0123] in, , , These are all correction coefficients, determined by domain experts based on the principles of skill learning. , All of these are thresholds for proficiency stages, also calibrated by domain experts, used to distinguish between low, medium, and high proficiency stages; Let represent the proficiency of the i-th skill at time t, used to characterize the degree of mastery of the skill at that time, as... The input parameters determine the value of the correction coefficient. Its function is to dynamically adjust the proficiency level based on the current skill mastery to improve efficiency. A higher correction coefficient is used in the low proficiency stage to accelerate skill acquisition, a baseline correction coefficient is used in the medium proficiency stage to stabilize improvement, and a lower correction coefficient is used in the high proficiency stage to match the actual pattern of slower efficiency in the skill refinement stage.

[0124] S22. Based on the efficiency coefficient of skill consolidation for different learning actions in cognitive dynamics parameters, construct calculation rules for the direct consolidation improvement of target skills.

[0125] Specifically, the consolidation efficiency coefficient corresponds one-to-one with the learning efficiency coefficient. For the k-th type of learning action of the i-th target skill, the consolidation efficiency coefficient is: . The value of is positively correlated with the depth of the learning action and negatively correlated with the current level of skill mastery. The specific value is determined by domain experts. The calculation rule for direct mastery improvement must satisfy a core constraint: the learning action will only produce a significant mastery improvement when the current proficiency of the target skill reaches the mastery threshold. The specific calculation expression is as follows:

[0126] .

[0127] in, This represents the direct increase in the solidification level of target skill i; This represents the consolidation efficiency coefficient of the i-th target skill corresponding to the k-th type of learning action; This represents the intensity of performing the k-th type of learning action for skill i at time t; Indicates the duration of the learning action; This represents the threshold for skill mastery, calibrated by domain experts. This represents the proficiency level of the i-th skill at time t; This is used to ensure that the degree of cure increase is 0 when the cure threshold is not reached; This represents the curing degree correction function, and its expression contains... This represents the degree of solidification of the i-th skill at time t, used to characterize the extent to which a skill transitions from short-term memory to a long-term stable mastery state. The solidification correction function... The specific expression can be: The purpose of the curing degree correction function is to quantify the impact of curing degree itself on curing efficiency; the lower the curing degree, the better. The larger the calculation result, the larger the correction coefficient, and the higher the curing improvement efficiency, reflecting the objective law that low curing degree technology has a stronger curing improvement potential; the higher the curing degree, the smaller the correction coefficient, and the smoother the curing improvement efficiency, which is consistent with the actual process of technology curing tending to stabilize.

[0128] S23. Based on the inter-skill collaborative gain matrix in the cognitive dynamics parameters, and combined with the direct proficiency improvement calculation rule, construct the indirect proficiency gain calculation rule for non-target skills.

[0129] Specifically, let's first define the synergistic gain matrix between skills. Suppose there are n skill nodes in the skill graph, and the synergistic gain matrix is ​​an n×n matrix. Matrix elements This represents the synergistic gain coefficient of skill j on skill i. Positive numbers indicate positive synergy, negative numbers indicate negative interference, and zero indicates no significant synergistic effect. The matrix element values ​​are determined by domain experts based on skill association types and measured data. Typical skill association types include pre-association, complementary association, and competitive association. The logic of the indirect proficiency gain calculation rule is that the indirect gain of non-target skills is obtained by summing the products of the direct proficiency increase of all target skills and their corresponding synergistic gain coefficients. The specific calculation expression can be:

[0130]

[0131] in, This represents the indirect proficiency gain for non-target skill i; This represents the set of target skills corresponding to the current learning activity; This represents the cooperative gain coefficient of skill j with respect to skill i, taken from the cooperative gain matrix between skills. ; This represents the direct increase in proficiency of the target skill j, derived from the calculation rules in S21.

[0132] S24. Based on the inter-skill collaborative gain matrix and combined with the direct solidification degree enhancement calculation rule, construct the indirect solidification degree gain calculation rule for non-target skills.

[0133] Specifically, this step, based on the inter-skill synergy gain matrix and combined with the direct solidification enhancement calculation rule of S22, constructs the indirect solidification gain calculation rule for non-target skills. The logic of this rule is similar to that of the indirect proficiency gain calculation, but it requires additional consideration of the solidification base state of the non-target skill itself to ensure the rationality of the indirect solidification gain. The specific calculation expression can be:

[0134]

[0135] in, This represents the indirect solidification gain of non-target skill i; This represents the cooperative gain coefficient of skill j with respect to skill i, taken from the cooperative gain matrix between skills. ; This represents the set of target skills corresponding to the current learning activity; This represents the direct increase in the solidification level of the target skill j, derived from the calculation rules of S22; The basic correction function for non-target skills can be expressed as follows:

[0136]

[0137] in, Basic correction factor, Both are calibration coefficients, determined by domain experts in conjunction with the co-consolidation rules of non-target skills. Together, they determine the range and trend of the correction function. This represents the degree of solidification of the i-th skill at time t, as... The input parameters are used to characterize the current solidified basic state of non-target skills. Its core function is to adjust the conversion efficiency of indirect solidification gains from non-target skills; the higher the solidification degree, the better. The larger the value of , the better. The larger the calculation result, the higher the indirect curing gain conversion efficiency, avoiding unreasonable curing improvement of low curing degree skills due to indirect gain, and ensuring that the indirect curing gain matches the curing foundation of non-target skills themselves.

[0138] S25. Integrate the calculation rules for direct proficiency improvement, direct solidification improvement, indirect proficiency gain, and indirect solidification gain to construct an action impact model.

[0139] Specifically, the integration method is as follows: for the target skill, the final increase in proficiency is the sum of the direct increase in proficiency and the indirect increase in proficiency, that is... If the target skill is simultaneously affected by other target skills, this stacking logic also applies; the final increase in solidification is the sum of the direct increase in solidification and the indirect increase in solidification, i.e. .

[0140] For non-target skills, the final increase in proficiency is only the indirect proficiency gain, i.e. The final increase in curing degree is only the indirect increase in curing degree, i.e. After integration, the overall output of the action-influence model consists of the proficiency increase vector and the fixedness increase vector for all skill nodes, where the proficiency increase vector is expressed as follows: The vector expression for the increase in curing degree is: .in, This represents the final increase in proficiency for the i-th skill node; This represents the direct increase in proficiency for the i-th skill node; This represents the indirect proficiency increase of the i-th skill node; This represents the final increase in the solidification level of the i-th skill node; This represents the direct increase in the solidification level of the i-th skill node; This represents the amount of indirect solidification improvement for the i-th skill node; This represents the vector of proficiency increase for all skill nodes. This represents the vector representing the increase in the solidification level of all skill nodes; This represents the total number of skill nodes in the skill graph.

[0141] In one alternative embodiment, the direct proficiency improvement calculation rule is defined using the following formula:

[0142]

[0143] in, This represents the direct increase in proficiency for the target skill j. Let be the learning efficiency coefficient of the learning action k on the target skill j in cognitive dynamics parameters. To learn the intensity of action k, This represents the proficiency level of target skill j at the current time t.

[0144] Specifically, the formula logic for calculating direct proficiency improvement is as follows: the direct proficiency improvement effect of the target skill is determined by the learning efficiency of the learning action for that skill, the intensity of the learning action itself, and the current proficiency level of the skill. The lower the current proficiency level of the skill, the greater the potential for improvement and the higher the direct improvement amount. This aligns with the objective laws of skill learning from beginner to advanced levels and accurately reflects the direct effect of the learning action on the proficiency of the target skill. In the formula, This represents the direct increase in proficiency of the target skill j, used to characterize the direct increment gained by the target skill j in the proficiency dimension after performing the learning action k; , is the learning efficiency coefficient of learning action k on target skill j in cognitive dynamics parameters. It is used to quantify the degree of adaptation of learning action k to the improvement of proficiency of target skill j. Its value is determined by domain experts in combination with the type of learning action and the characteristics of target skill. The learning intensity of learning action k is used to characterize the execution depth and level of engagement of learning action k. Its value is determined quantitatively through core indicators such as the execution duration and focus of the learning action. This represents the proficiency of target skill j at the current time t, and is used to characterize the degree of mastery of target skill j before performing the learning action k; The potential coefficient for improving proficiency is used to quantify the current level of proficiency improvement for the target skill j. The lower the proficiency, the larger the value of this coefficient, and the higher the direct proficiency improvement value.

[0145] The calculation rule for direct cure degree improvement is defined using the following formula:

[0146]

[0147] in, This represents the direct increase in the degree of fixation of the target skill j. Let be the efficiency coefficient of the learning action k on the solidification of the target skill j in cognitive dynamics parameters. This indicates the degree of fixation of the target skill j at the current time t.

[0148] Specifically, the formula logic for calculating the direct solidification improvement rule is as follows: the direct solidification improvement effect of the target skill must meet the core constraint condition: only when the current proficiency of the skill is higher than the current solidification level can the learning action produce an effective direct solidification improvement. The solidification improvement effect is jointly determined by the solidification efficiency of the learning action on the skill, the strength of the learning action itself, and the difference between proficiency and solidification level, ensuring that the solidification improvement matches the skill state and conforms to the objective law of skill development from proficiency to stable solidification. In the formula, This represents the direct improvement in the solidification of the target skill j, which characterizes the direct increment gained by the target skill j in the solidification dimension after performing the learning action k. , is the consolidation efficiency coefficient of learning action k on target skill j in cognitive dynamics parameters. It is used to quantify the degree of consolidation improvement of learning action k on target skill j. Its value is determined by domain experts in combination with the depth of learning action and the characteristics of target skill. This represents the degree of solidification of target skill j at the current time t, and is used to characterize the current stable mastery of target skill j; The solidification enhancement trigger coefficient is used to determine whether an effective solidification enhancement occurs. When the proficiency is higher than the solidification level, the coefficient is the difference between the two, triggering a solidification enhancement. The larger the difference, the greater the potential for solidification enhancement. When the proficiency is not higher than the solidification level, the coefficient is 0, and there is no effective solidification enhancement, ensuring the rationality of the solidification enhancement.

[0149] The aforementioned AI-based method for optimizing vocational skills courses first constructs a skill map by structurally decomposing the target vocational skills. Based on individual diagnostic data, it configures cognitive dynamics parameters and initial states for each skill node, including forgetting, consolidation, learning efficiency, and synergistic relationships. Then, it builds a learner dynamic cognitive state model that simulates the natural decay of skills and the impact of learning activities. This model is then used as a state transition engine to construct and solve a Markov decision process with skill state as input and learning action as output, thereby obtaining an optimal strategy for proactively planning the learning sequence. This strategy, combined with the current cognitive state, is then used for multi-step simulation to generate personalized learning paths that meet actual resource constraints and are translated into a schedule. Finally, learning feedback is collected to update the cognitive state in real time and calibrate the model parameters, forming a closed loop of "modeling-planning-execution-optimization." This method achieves proactive, personalized, and globally optimal path planning for the learning process of composite vocational skills, significantly improving the long-term retention level of composite abilities and overall learning efficiency under limited time and energy constraints.

[0150] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0151] Based on the same inventive concept, this application also provides a system for implementing the aforementioned artificial intelligence-based vocational skills curriculum optimization method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations in one or more embodiments of the artificial intelligence-based vocational skills curriculum optimization system provided below can be found in the limitations of the artificial intelligence-based vocational skills curriculum optimization method described above, and will not be repeated here.

[0152] In one exemplary embodiment, such as Figure 3 As shown, an artificial intelligence-based professional competence curriculum optimization system 30 is provided to implement the methods in the above-described method embodiments. The system includes:

[0153] The skill graph construction and parameter configuration module 31 is used to decompose the domain knowledge of the target professional competence into skills and obtain the skill decomposition results; construct a skill graph including multiple skill nodes and their relationships based on the skill decomposition results; configure the initial proficiency and initial solidification of each skill node in the skill graph based on the learner's performance data on the ability diagnosis task, and configure a set of cognitive dynamic parameters for each skill node; among which, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills.

[0154] The dynamic cognitive state modeling module 32 is used to construct a learner dynamic cognitive state model based on skill graphs, initial proficiency, initial solidification, and cognitive dynamics parameters to simulate the evolution of skill states and the impact of learning activities.

[0155] The optimal learning strategy generation module 33 is used to construct a Markov decision process with skill state as input and learning action as output based on the learner's dynamic cognitive state model, and solve the Markov decision process through a deep reinforcement learning algorithm to obtain the optimal learning planning strategy.

[0156] The personalized path planning module 34 is used to call the optimal learning planning strategy to perform multi-step sequence decision simulation based on the learner's current cognitive state, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule.

[0157] The adaptive feedback optimization module 35 is used to collect real-time feedback data of learners executing the learning schedule, update the learners' cognitive state based on the real-time feedback data and calculate the model prediction error of the learners' dynamic cognitive state model; calibrate the cognitive dynamic parameters based on the model prediction error, and use the calibrated cognitive dynamic parameters to dynamically optimize the learners' dynamic cognitive state model and the optimal learning planning strategy.

[0158] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the aforementioned method embodiments.

[0159] Embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0160] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0161] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for optimizing vocational skills courses based on artificial intelligence, characterized in that, The method includes: S1. Decompose the domain knowledge of the target professional competence into skills to obtain skill decomposition results; construct a skill map including multiple skill nodes and their relationships based on the skill decomposition results; configure the initial proficiency and initial solidification of each skill node in the skill map based on the learner's performance data on the ability diagnosis task, and configure a set of cognitive dynamic parameters for each skill node; wherein, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills; S2. Based on the skill map, the initial proficiency, the initial solidification, and the cognitive dynamics parameters, construct a learner dynamic cognitive state model to simulate the evolution of skill state and the impact of learning activities; S3. Based on the learner's dynamic cognitive state model, construct a Markov decision process with skill state as input and learning action as output, and solve the Markov decision process through a deep reinforcement learning algorithm to obtain the optimal learning planning strategy. S4. Based on the learner's current cognitive state, invoke the optimal learning planning strategy to perform multi-step sequence decision simulation, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule. S5. Collect real-time feedback data of the learner executing the learning schedule, update the learner's cognitive state based on the real-time feedback data and calculate the model prediction error of the learner's dynamic cognitive state model; calibrate the cognitive dynamics parameters based on the model prediction error, and dynamically optimize the learner's dynamic cognitive state model and the optimal learning planning strategy using the calibrated cognitive dynamics parameters.

2. The method according to claim 1, characterized in that, The learner dynamic cognitive state model, constructed based on the skill map, the initial proficiency, the initial solidification, and the cognitive dynamics parameters, to simulate the evolution of skill states and the impact of learning activities, includes: S11. Based on all skill nodes in the skill graph, and combining the initial proficiency and the initial solidification, construct the learner's initial cognitive state vector; S12. Based on the parameters characterizing forgetting characteristics in the cognitive dynamics parameters, construct a first hyperbolic equation for calculating the natural decay of skill proficiency during periods without learning activities. S13. Based on the parameters characterizing the decay characteristics in the cognitive dynamics parameters, construct a second hyperbolic equation for calculating the natural decay of skill solidification during periods without learning activities. S14. Based on the parameters in the cognitive dynamics parameters that characterize the synergistic relationship between learning efficiency and skills, construct an action influence model for calculating the state improvement amount of the target skill and other related skills after performing a specific learning activity. S15. Sum the natural decay amount described by the first hyperbolic equation and the second hyperbolic equation with the state improvement amount described by the action influence model to construct a state transition equation; based on the state transition equation and the initial cognitive state vector, construct the learner dynamic cognitive state model.

3. The method according to claim 2, characterized in that, The first hyperbolic equation is constructed based on the instantaneous forgetting rate, consolidation inhibition coefficient, and proficiency adjustment coefficient for any skill node in the cognitive dynamics parameters. The expression of the first hyperbolic equation is: in, Indicates the time of the i-th skill proficiency Indicates the time of the i-th skill The degree of curing, Indicates a time interval without learning activities. It is a very small positive number that prevents the denominator from being zero. This represents the solidification inhibition coefficient of the i-th skill. Let represent the proficiency adjustment coefficient for the i-th skill. The first hyperbolic equation represents the instantaneous forgetting rate of the i-th skill; it is used to calculate future moments based on current proficiency and entrenchment. Proficiency prediction value .

4. The method according to claim 2, characterized in that, The model, constructed based on the parameters representing the synergistic relationship between learning efficiency and skills in the cognitive dynamics parameters, is used to calculate the action impact model on the state improvement of the target skill and other related skills after performing a specific learning activity, including: S21. Based on the learning efficiency coefficients of different learning actions for skills in the cognitive dynamics parameters, construct a calculation rule for the direct proficiency improvement of the target skill. S22. Based on the consolidation efficiency coefficients of different learning actions for skills in the cognitive dynamics parameters, construct calculation rules for the direct consolidation improvement of the target skill. S23. Based on the inter-skill collaborative gain matrix in the cognitive dynamics parameters, and combined with the direct proficiency improvement calculation rule, construct the indirect proficiency gain calculation rule for non-target skills; S24. Based on the skill-to-skill collaborative gain matrix and the direct solidification degree enhancement calculation rule, construct the indirect solidification degree gain calculation rule for non-target skills. S25. Integrate the direct proficiency improvement calculation rule, the direct solidification improvement calculation rule, the indirect proficiency gain calculation rule, and the indirect solidification gain calculation rule to construct the action influence model.

5. The method according to claim 4, characterized in that, The direct proficiency improvement calculation rule is defined using the following formula: in, This represents the direct increase in proficiency for the target skill j. The learning efficiency coefficient of the learning action k on the target skill j in the cognitive dynamics parameters is denoted as . To learn the intensity of action k, This represents the proficiency level of target skill j at the current time t; The calculation rule for the direct cure degree improvement is defined using the following formula: in, This represents the direct increase in the degree of fixation of the target skill j. The solidification efficiency coefficient of the learned action k for the target skill j in the cognitive dynamics parameters is given. This indicates the degree of fixation of the target skill j at the current time t.

6. A professional competency curriculum optimization system based on artificial intelligence, used to implement the method according to any one of claims 1 to 5, characterized in that, The system includes: The skill graph construction and parameter configuration module is used to decompose the domain knowledge of the target professional competence into skills, and obtain the skill decomposition results; based on the skill decomposition results, a skill graph including multiple skill nodes and their relationships is constructed; based on the learner's performance data on the ability diagnosis task, the initial proficiency and initial solidification of each skill node in the skill graph are configured, and a set of cognitive dynamic parameters are configured for each skill node; wherein, the cognitive dynamic parameters include parameters used to describe the forgetting characteristics, solidification characteristics, learning efficiency, and synergistic relationships between skills; The dynamic cognitive state modeling module is used to construct a learner dynamic cognitive state model based on the skill map, the initial proficiency, the initial solidification, and the cognitive dynamics parameters to simulate the evolution of skill state and the impact of learning activities. The optimal learning strategy generation module is used to construct a Markov decision process with skill state as input and learning action as output based on the learner's dynamic cognitive state model, and solve the Markov decision process through a deep reinforcement learning algorithm to obtain the optimal learning planning strategy. The personalized path planning module is used to call the optimal learning planning strategy to perform multi-step sequence decision simulation based on the learner's current cognitive state, generate a personalized learning path that meets preset resource constraints, and convert the personalized learning path into a corresponding learning schedule. An adaptive feedback optimization module is used to collect real-time feedback data of learners executing the learning schedule, update the learner's cognitive state based on the real-time feedback data and calculate the model prediction error of the learner's dynamic cognitive state model; calibrate the cognitive dynamics parameters based on the model prediction error, and dynamically optimize the learner's dynamic cognitive state model and the optimal learning planning strategy using the calibrated cognitive dynamics parameters.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 5.

Images