A task scheduling method and system applied to a skill training platform and a medium

Abstract

The application provides a task scheduling method and system applied to a skill training platform and a medium, comprising the following steps: obtaining first task completion data of a virtual online task of a target object from the skill training platform; generating user portrait data of the target object based on the first task completion data and second task completion data of an entity offline task of the target object; determining a task arrangement constraint condition, wherein the task arrangement constraint condition comprises a skill dependency relationship in a skill knowledge graph of the skill training platform, a task proportion between the virtual online task and the entity offline task, and a total task target time length; and under the constraint of the task arrangement constraint condition, determining a target task sequence matched by the target object from a skill training task library of the skill training platform by using the user portrait data, and then determining task arrangement information of the target object. The application can improve the adaptability of the arranged task to the current actual ability of the operator, so that the task scheduling is more reasonable.

Description

Technical Field

[0001] This application relates to the field of skills training technology, specifically to a task scheduling method, system, and medium applied to a skills training platform. Background Technology

[0002] Skill training platforms typically integrate virtual simulation terminals for theoretical verification and physical equipment for practical exercises. They aim to improve operators' overall skill level through diverse task interactions and distribute corresponding skill training task instructions to operators according to the training progress.

[0003] However, task scheduling typically employs a linear progression logic based on a single data source, such as assigning skill training tasks sequentially according to a pre-defined, unchanging training syllabus. This task scheduling method often results in a low match between the assigned tasks and the operators' actual capabilities. It is clear that current task scheduling methods are often inadequate. Summary of the Invention

[0004] The embodiments of this application provide a task scheduling method, system, and medium for use in a skills training platform, aiming to improve the compatibility between the scheduled tasks and the actual capabilities of the operators, thereby making task scheduling more reasonable.

[0005] In a first aspect, embodiments of this application provide a task scheduling method applied to a skills training platform, the task scheduling method applied to the skills training platform comprising:

[0006] Obtain the first task completion data of the target object's virtual online task from the skills training platform;

[0007] Based on the first task completion data and the second task completion data of the target object's offline tasks, user profile data of the target object is generated.

[0008] The task scheduling constraints are determined, including the skill dependencies in the skill knowledge graph of the skill training platform, the task ratio between virtual online tasks and physical offline tasks, and the total duration of task objectives.

[0009] Under the constraints of the task arrangement conditions, using the user profile data, the target task sequence matching the target object is determined from the skill training task library of the skill training platform. The target task sequence includes target virtual online tasks and target physical offline tasks.

[0010] Based on the target task sequence, the task arrangement information of the target object on the skills training platform is determined, and the task arrangement information is used for task scheduling processing.

[0011] In the above embodiments, by integrating the completion data of the first task and the completion data of the second task, user profile data is generated, which quantifies the multi-dimensional capabilities of the target object. Then, by combining task arrangement constraints such as skill dependencies, task ratios, and the total duration of task objectives, the scientific nature of the task sequence in terms of logical order and resource consumption is ensured. Based on this, using the user profile data, target virtual online tasks and target physical offline tasks are selected from the skill training task library, and task arrangement information is determined, realizing personalized customization of training paths. This improves the adaptability of the arranged tasks to the actual capabilities of the operators, making task scheduling more reasonable.

[0012] In one embodiment, generating user profile data for the target object based on the first task completion data and the second task completion data of the target object's offline tasks includes:

[0013] The browsing duration parameter and interaction round parameter in the first task completion data, and the completion quality score parameter and collaboration evaluation parameter in the second task completion data are vectorized to obtain a multi-dimensional feature vector.

[0014] The multidimensional feature vector is input into a pre-trained semantic analysis model to extract the skill semantic features of the target object;

[0015] Based on the aforementioned skill semantic features, the target object's current skill proficiency level, skill deficiencies, and skill preferences are determined.

[0016] Based on the current skill proficiency level, the skill deficiencies, and the skill preferences, user profile data for the target object is generated.

[0017] In the above embodiments, by vectorizing browsing duration parameters, interaction round parameters, completion quality score parameters, and collaboration evaluation parameters, heterogeneous behavioral data is transformed into multi-dimensional feature vectors that can be processed by computers. Using a pre-trained semantic analysis model to deeply extract skill semantic features, user profile data generated based on this, including current skill proficiency level, skill deficiencies, and skill preferences, can comprehensively and three-dimensionally depict an individual's ability status and learning habits. This provides a reliable data foundation for subsequent high-precision personalized task matching, avoiding the one-sidedness of evaluation based on a single-dimensional data.

[0018] In one embodiment, determining the target task sequence matching the target object from the skill training task library of the skill training platform using the user profile data, under the constraints of the task orchestration conditions, includes:

[0019] Under the constraints of the task arrangement conditions, multiple skill training tasks in the skill training task library are encoded to generate a chromosome sequence population including multiple initial task chromosome sequences.

[0020] For each of the initial task chromosome sequences, determine the expected task matching degree between the initial task chromosome sequence and the user profile data;

[0021] Based on the expected matching degree of the task, determine the function value of the preset fitness function;

[0022] Under the constraints of the task arrangement conditions, the chromosome sequence population is subjected to selection, crossover and mutation operations using the function value of the preset fitness function to obtain a converged chromosome sequence population.

[0023] Based on the converged chromosome sequence population, the target task sequence matching the target object is determined.

[0024] In the above embodiments, a chromosome sequence population is constructed using genetic algorithms. This not only considers the compatibility between user profile data and task sequences but also introduces task orchestration constraints as a selection boundary. By calculating the expected matching degree of tasks and the function value of a preset fitness function, the quality of each initial task chromosome sequence is quantitatively evaluated. Through iterative optimization using crossover and mutation operations, a globally optimal solution can be determined in a large-scale solution space. This results in a target task sequence that conforms to skill-dependent logic and highly matches individual characteristics, improving the rationality of task orchestration.

[0025] In one embodiment, determining the expected task match between the initial task chromosome sequence and the user profile data includes:

[0026] Based on the task difficulty coefficient and task type of each skill training task in the initial task chromosome sequence, generate a task difficulty coefficient sequence and a task type sequence;

[0027] Based on the task difficulty coefficient sequence and the current skill proficiency level, determine the ability matching deviation value;

[0028] Based on the task type sequence and the skill missing items, a task type matching index is determined;

[0029] The expected matching degree of the task is determined by using the capability matching deviation value and the task type matching index.

[0030] In the above embodiments, abstract task characteristics are visualized by generating task difficulty coefficient sequences and task type sequences respectively. The ability matching deviation value is calculated based on the current skill proficiency level, avoiding frustration due to excessive difficulty or boredom due to insufficient difficulty. The task type matching index is determined by combining skill deficiencies, ensuring that the task content accurately covers skill gaps. By comprehensively utilizing both to determine the expected task matching degree, the calibration of difficulty and content is achieved, thereby improving the relevance and effectiveness of each skill training task in improving the target audience's abilities.

[0031] In one embodiment, determining the function value of the preset fitness function based on the expected task matching degree includes:

[0032] Based on the initial task chromosome sequence and the user profile data, the expected skill improvement of the target object is determined;

[0033] The total expected duration of the task is determined by the initial task chromosome sequence;

[0034] The function value of the preset fitness function is determined based on the sum of the expected task matching degree, the expected skill improvement degree, and the expected task duration.

[0035] In the above embodiments, when determining the value of the preset fitness function, a multi-objective optimization evaluation system was constructed by comprehensively considering three dimensions: expected task matching degree, expected skill improvement degree, and total expected task duration. This scheme not only focuses on whether the task is suitable for the target object (expected task matching degree), but also on whether the task can bring the maximum expected skill improvement degree, while taking into account time cost (total expected task duration). This multi-dimensional evaluation logic ensures that the selected target task sequence can maximize the training effect and efficiency within limited time resources.

[0036] In one embodiment, determining the expected skill improvement of the target object based on the initial task chromosome sequence and the user profile data includes:

[0037] Based on the task type, the missing skill item, and the skill preference item of each skill training task in the initial task chromosome sequence, determine the expected skill improvement coefficient of each skill training task in the initial task chromosome sequence.

[0038] Obtain the skill gain contribution value for each skill training task in the initial task chromosome sequence;

[0039] The expected skill improvement is determined based on the skill gain contribution value and the expected skill improvement coefficient.

[0040] In the above embodiments, the expected skill improvement coefficient is determined by introducing task type, skill gaps, and skill preferences, thus assigning personalized weights to the skill gain contribution value calculated under standard working conditions. This calculation logic fully considers the complementarity of different task types to specific individuals' skill gaps and the stimulating effect of personal preferences on learning efficiency. Based on this, the determined expected skill improvement can more realistically predict the training value of task sequences for the target object, ensuring that high-value tasks that can both fill skill gaps and conform to user operating habits are prioritized.

[0041] In one embodiment, after determining the target object based on the target task sequence and the task arrangement information of the skills training platform, the method further includes:

[0042] Obtain the first interaction log generated by the target object executing the target virtual online task;

[0043] Identify the skill confusion keywords in the first interaction log;

[0044] In the skills training task library, identify the offline tasks that match the keywords causing confusion in the skills.

[0045] The offline tasks matching the keywords causing the skill confusion are inserted before any unexecuted skill training tasks in the task scheduling information.

[0046] In the above embodiments, by analyzing the target object's first interaction log in the virtual environment in real time and extracting keywords indicating skill confusion, specific obstacles encountered by users at the theoretical cognition or simulated operation levels can be sensitively identified. Automatic matching and timely insertion of offline physical tasks into the task orchestration process achieves intervention from virtual cognition to physical verification. This scheduling adjustment based on real-time feedback ensures that users receive intuitive feedback and correction through real physical operations as soon as confusion arises.

[0047] In one embodiment, after determining the target object based on the target task sequence and the task arrangement information of the skills training platform, the method further includes:

[0048] Obtain the second interaction log generated by the target object executing the offline task of the target entity;

[0049] Based on the second interaction log, the target object's weak skills are determined;

[0050] In the skills training task library, determine the virtual online task that matches the weak skill;

[0051] The virtual online task matching the weak skill is inserted before the skill training task that has not yet been executed in the task scheduling information.

[0052] In the above embodiments, the second interaction log is used to perform fine-grained analysis of the target object's performance in offline tasks, which can accurately pinpoint the weak skills exposed in actual operation. Virtual online tasks are then matched in reverse and inserted into the unexecuted sequence, providing targeted reinforcement training for weaknesses in physical operations through a low-cost, repeatable virtual environment. This closed-loop feedback adjustment of practical problem discovery and virtual reinforcement fully leverages the advantages of combining virtual and real learning. It not only solves the problem of limited practical resources preventing repeated trial and error but also ensures that each operational skill is supported by solid theoretical foundations and logical codification.

[0053] Secondly, embodiments of this application provide a task scheduling system applied to a skills training platform, wherein the task scheduling system applied to the skills training platform is used to execute the task scheduling method applied to the skills training platform as described in any of the preceding claims.

[0054] Thirdly, embodiments of this application provide an electronic device, the electronic device including a memory, a processor and a computer program stored in the memory, the processor executing the computer program to implement the task scheduling method applied to a skills training platform as described in any of the preceding claims.

[0055] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program configured to be executed by a processor to implement the task scheduling method applied to a skills training platform as described in any of the preceding claims. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments 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.

[0057] Figure 1 This is a schematic flowchart of an embodiment of the task scheduling method applied to a skills training platform provided in this application;

[0058] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. In addition, in the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0060] In a first aspect, embodiments of this application provide a task scheduling method applied to a skills training platform, wherein the executing entity is a task scheduling system (hereinafter referred to as "the system") applied to the skills training platform.

[0061] Specifically, refer to Figure 1 Task scheduling methods applied to skills training platforms may include:

[0062] S101. Obtain the first task completion data of the virtual online task of the target object from the skills training platform.

[0063] In the embodiments of this application, a skills training platform refers to a computer software system that provides skills training functions such as skills drills, process operation training, or engineering practice. The target audience refers to trainees using the skills training functions on the skills training platform. A virtual online task refers to a numerical simulation task, an interactive 3D model operation task, or a web-based theoretical verification task constructed based on computer simulation technology, augmented reality (AR), or virtual reality (VR) environments. The first task completion data refers to the structured or semi-structured logs generated by the target audience during the execution of the virtual online task, encompassing task access duration, human-computer interaction frequency, instruction execution accuracy, and feedback data for complex logic nodes.

[0064] In some embodiments of this application, the first task completion data also includes a record of the target object's operation path in the virtual environment, and identifies the pause time of the target object at specific technical nodes through trajectory analysis, thereby accurately locating its operational suspicious points.

[0065] S102. Based on the first task completion data and the second task completion data of the target object's offline tasks, generate user profile data of the target object.

[0066] In the embodiments of this application, the physical offline task refers to real-world tasks performed in a physical environment, such as operating production equipment, assembling physical objects, offline group collaboration projects, or calibrating experimental instruments. The second task completion data refers to the feature data resulting from the digitization of the output quality, material loss rate, collaboration score, and evaluation text provided by the technical supervisor for that physical stage. User profile data is a digitized representation vector of the target object's skill status, used to transform abstract human capabilities into a high-dimensional data structure that can be recognized by computers.

[0067] The generation of user profile data can be achieved by establishing a heterogeneous data fusion model. A heterogeneous data fusion model refers to a data processing module that integrates data with different structures (such as discrete numerical data in the first task completion data and textual evaluations in the second task completion data) into the same feature space. It uses a preset mapping dictionary and feature alignment algorithm to align and merge the first and second task completion data in dimensions to form a fused feature vector of the initial dimension, which is then used as user profile data.

[0068] The pre-defined mapping dictionary is a predefined lookup table that maps various field names, data types, and value ranges in heterogeneous data to standardized feature dimension names and numerical ranges. The feature alignment algorithm is an algorithmic logic that maps data fields from different sources to their corresponding capability dimensions based on feature name matching and semantic similarity calculation. Specifically, the system first maps the numerical fields (such as browsing time and interaction rounds) in the first task completion data to standardized values ​​of the corresponding dimensions according to the mapping dictionary. The textual evaluation fields in the second task completion data are converted into dense vectors of fixed length through a pre-trained word embedding model. Then, all mapped features are concatenated according to the dimensional order defined in the mapping dictionary to form a fused feature vector of uniform length.

[0069] In some embodiments of this application, incremental learning algorithms, such as the Follow-The-Regularized-Leader (FTRL) algorithm, are employed to update the feature weights of the user profile data in real time based on the latest first and second task completion data. Specifically, the FTRL algorithm updates the weights of each skill dimension in the profile vector in a sample-by-sample iterative manner by comprehensively minimizing the sum of historical gradients, an L1 regularization term (to generate sparse solutions to filter irrelevant features), and an L2 regularization term (to prevent weight overfitting). This approach ensures that the profile reflects the real-time fluctuations in the target object's abilities.

[0070] S103. Determine the task scheduling constraints, which include the skill dependencies in the skill knowledge graph of the skills training platform, the task ratio between virtual online tasks and physical offline tasks, and the total duration of task objectives.

[0071] In the embodiments of this application, task orchestration constraints are a set of boundary parameters that limit the rationality and compliance of task sequences. The skill knowledge graph is a technical graph storing the logical sequence relationships between skill units. In this skill knowledge graph, nodes represent specific discrete skill items, and directed edges represent pre- and post-dependencies between skill items. This skill knowledge graph is a dataset pre-extracted from business documents such as standard operating manuals and training syllabi of training equipment using knowledge extraction technology, and then structured and managed in the form of a graph database (e.g., Neo4j). Skill dependencies ensure that task orchestration conforms to the objective logic from basic to advanced levels. The task ratio between virtual online tasks and physical offline tasks is used to control the weighting of theoretical simulation and practical exercises, for example, setting a ratio of 6:4 to ensure the balance of virtual and real integration during training. The total duration of task objectives is a preset total training cycle constraint.

[0072] In some embodiments of this application, the task orchestration constraints also integrate hardware resource occupancy constraints, dynamically adjusting the admission time of physical offline tasks based on the idle status of offline training equipment, to ensure the executability of the scheduling scheme at the physical resource level.

[0073] S104. Under the constraints of task arrangement, using user profile data, determine the target task sequence matching the target object from the skill training task library of the skill training platform. The target task sequence includes target virtual online tasks and target physical offline tasks.

[0074] In the embodiments of this application, the skills training task library is a collection of task metadata containing different technical dimensions, different difficulty levels, and different presentation formats. The process of determining the target task sequence using user profile data is essentially solving the constraint satisfaction problem (CSP) under multi-objective optimization. The target task sequence refers to a heterogeneous task chain tailored for the target object, containing a series of sequential temporal relationships.

[0075] In some embodiments of this application, an improved heuristic search algorithm or genetic algorithm is employed, using skill gaps in user profile data as search guidance. Under strong constraints of task proportion and dependency relationships, the task path with the highest fitness is iteratively calculated by maximizing the skill gain function. This approach ensures that the issued task sequence accurately covers the weak points of the target object.

[0076] S105. Based on the target task sequence, determine the task arrangement information of the target object on the skills training platform, and the task arrangement information is used for task scheduling processing.

[0077] In the embodiments of this application, task orchestration information is a set of execution instructions instantiated from the target task sequence. Its scope includes the task's start timestamp, access token, rendering parameters of the task card in the user interface, and associated training resource index. Determining the task orchestration information signifies that the system has formally completed this scheduling operation and entered the execution and distribution phase.

[0078] In some embodiments of this application, task orchestration information is pushed to the target object's mobile terminal or workstation in real time via a full-duplex communication protocol (such as WebSocket), and the corresponding task indicator is dynamically loaded in the front-end component, ensuring the efficiency and continuity of the training process.

[0079] As can be seen, this application constructs a dynamic profile by integrating execution data from heterogeneous virtual and real tasks, uses skill dependencies and virtual-real ratios for constraint solving, coordinates the logical gradients of physical entity tasks and digital virtual tasks, reduces redundant configuration of training resources, and improves the adaptation accuracy of skill training under complex working conditions.

[0080] In some embodiments of this application, user profile data of the target object is generated based on the first task completion data and the second task completion data of the target object's offline tasks, including:

[0081] S201. Vectorize the browsing duration and interaction round parameters in the first task completion data, and the completion quality score and collaboration evaluation parameters in the second task completion data to obtain a multi-dimensional feature vector.

[0082] In the embodiments of this application, the browsing duration parameter refers to the time the target object resides in a specific virtual training module. The interaction round parameter refers to the number of cycles of instruction input and feedback response between the target object and the simulation system. The completion quality score parameter refers to the numerical result obtained by scoring the output or operation process of the entity task based on preset standards. The collaboration evaluation parameter refers to the quantitative rating of the cooperation and contribution of the system or other collaborative nodes to the target object in a multi-person collaborative task. Vectorization processing refers to the process of converting the above discrete, scalar numerical values ​​or text labels into a numerical matrix of a unified dimension. The multidimensional feature vector is a mathematical expression used to characterize the comprehensive behavioral state of the target object in that time slice.

[0083] In some embodiments of this application, one-hot encoding or normalization techniques (such as min-max standardization or Z-score standardization) are used to map data of different dimensions (e.g., time in seconds, score in points) into a unified feature space, constructing a high-dimensional sparse vector containing temporal behavioral features and result quality features. This provides standardized input data for subsequent deep learning models, solving the technical problem of the difficulty in directly fusing and calculating heterogeneous data. Specifically, for continuous numerical features such as browsing duration and interaction rounds, min-max standardization is used, that is, subtracting the minimum value of the feature from all samples and then dividing by the difference between the maximum and minimum values ​​to map it to the [0,1] interval; for completion quality scoring parameters, if they are continuous scores, min-max standardization is also used; for collaborative evaluation parameters, if they are discrete level labels (e.g., excellent, good, qualified, unqualified), one-hot encoding is used to convert each level into a sparse vector with only one element as 1 and the rest as 0. After processing, all standardized or encoded subvectors are concatenated according to a predefined dimensional order to form the final multidimensional feature vector.

[0084] S202. Input the multidimensional feature vector into the pre-trained semantic analysis model to extract the skill semantic features of the target object.

[0085] In the embodiments of this application, the pre-trained semantic analysis model refers to a deep neural network model pre-trained based on large-scale training log data, such as the Bidirectional Encoder Representations from Transformers (BERT) model based on a transformer architecture. Skill semantic features refer to the hidden state vectors output by the model, which not only contain explicit behavioral statistics but also implicitly contain higher-order logical connections behind the behavior, such as the potential causal relationship between excessive operation time and specific skill weaknesses.

[0086] In some embodiments of this application, to eliminate the input barrier between numerical features and the natural language model, a multidimensional feature vector containing continuous numerical parameters such as browsing duration and interaction rounds is first transformed into a structured behavioral description statement using a discretization template (e.g., a text sequence stating "The object has a high browsing duration but a very low interaction round in the simulation task"). Then, tokenization is performed and the result is input into the semantic analysis model. The discretization template refers to a set of predefined conditional-text mapping rules. Each rule contains a feature dimension name, a set of numerical interval thresholds, and a corresponding natural language description fragment. For example, for the browsing duration dimension, the discretization template defines: a normalized browsing duration value greater than 0.7 maps to a high browsing duration; a value between 0.3 and 0.7 maps to a moderate browsing duration; and a value less than 0.3 maps to a low browsing duration. The system iterates through each dimension of the multidimensional feature vector, queries the corresponding threshold interval in the discretization template based on its value, selects the matching description fragment, and finally concatenates the description fragments of all dimensions into a complete behavioral description statement according to a preset sentence template.

[0087] This semantic analysis model is fine-tuned to adapt to specific industrial training scenarios. The fine-tuning process involves collecting historical behavioral descriptions and their corresponding manually labeled skill tags accumulated on the skills training platform as a fine-tuning training set. Based on the pre-trained BERT model, the model is updated several times (e.g., 3 to 5 epochs) using the Adam optimizer with a low learning rate (e.g., 2e-5) and based on the cross-entropy loss function. This allows the model's output feature space to adapt to the skill semantic distribution in the industrial training scenario, enabling the model to identify implicit, non-linear ability patterns in behavioral description sequences. This abstracts low-level behavioral data into high-level semantic representations with business interpretability, improving the depth and accuracy of feature extraction.

[0088] Furthermore, before use, this pre-trained semantic analysis model undergoes comparative learning training using positive and negative behavior description samples constructed from historical training log data. The network weights are optimized by minimizing the contrastive loss function based on the backpropagation algorithm, enabling the model to accurately distinguish complex feature representations at different ability levels. The InfoNCE loss function can be used for this purpose. By minimizing this loss function, the model learns to cluster behavior descriptions at the same ability level in the feature space and push behavior descriptions at different ability levels apart, thereby enhancing the model's ability to identify ability differences.

[0089] S203. Based on the semantic features of skills, determine the current skill proficiency level, skill deficiencies, and skill preferences of the target object.

[0090] In the embodiments of this application, the current skill proficiency level refers to the ability level achieved by the target object at a specific skill tree node, such as being divided into different levels like memory, comprehension, application, and analysis using Bloom's Taxonomy. Skill deficiencies refer to specific operational points or theoretical knowledge points that the target object has failed to meet preset benchmark requirements, such as missing temperature control logic in precision welding or inadequate adherence to safety regulations. Skill preferences refer to specific learning habits or operational tendencies exhibited by the target object, such as a preference for visual guidance or a preference for hands-on trial and error. Determining the above information relies on the decoding of skill semantic features by a classifier or regression model.

[0091] In some embodiments of this application, abstract semantic vectors are mapped to a predefined skill evaluation index system through a classification mapping layer. The predefined skill evaluation index system refers to a hierarchical classification label system pre-constructed based on curriculum standards and job competency requirements in the practical training field. This system includes: a set of skill proficiency level labels (e.g., six levels according to Bloom's Taxonomy: memory, comprehension, application, analysis, evaluation, and creation); a set of skill deficiency label labels (e.g., a list of skill labels extracted from all leaf skill nodes of the skill knowledge graph, such as precision welding temperature control, safety standard execution, etc.); and a set of skill preference label labels (e.g., visual guidance preference, hands-on trial-and-error preference, document reading preference, etc.).

[0092] Specifically, the classification mapping layer comprises three parallel fully connected network branches, corresponding to the current skill proficiency level prediction, multi-label classification of missing skill items, and skill preference item classification tasks, respectively. The current skill proficiency level prediction branch is a single-label multi-classification task; its output layer node count equals the number of skill proficiency level labels. It uses the Softmax activation function to output the probability distribution of each level, selecting the level with the highest probability as the prediction result. The multi-label classification branch for missing skill items has an output layer node count equal to the total number of missing skill item labels. Each output node uses the Sigmoid activation function to independently output the probability value of whether the skill item is missing. Skill items with a probability value exceeding a preset threshold (e.g., 0.5) are classified as missing items. The skill preference item classification branch is also a single-label multi-classification task; its output layer node count equals the number of skill preference item labels. It uses the Softmax activation function to output the probability distribution of each preference type, selecting the preference type with the highest probability as the prediction result.

[0093] After calculation by the corresponding fully connected branches, each output node is processed by the Softmax activation function (or Sigmoid activation function) to obtain the probability distribution of each evaluation index. Finally, the category with the highest probability value is selected as the corresponding result parameter, realizing fine-grained diagnosis of personnel capabilities and providing specific decision-making basis for accurate task recommendations in the future. In the model training stage, historical behavioral feature vectors with manually labeled skill tags can be used as the training set input model. The cross-entropy loss function or multi-label classification loss between the probability distribution of the model output and the real historical skill tags is calculated (where the current skill proficiency level prediction branch and the skill preference item classification branch use the multi-class cross-entropy loss function, and the skill missing item multi-label classification branch uses the binary cross-entropy loss function and averages it over all label dimensions). The backpropagation algorithm and model optimizer (such as the Adam optimizer) are used to continuously update the connection weights of each fully connected network branch in the classification mapping layer to ensure the accuracy of the classification prediction output.

[0094] S204. Generate user profile data for the target object based on the current skill proficiency level, skill deficiencies, and skill preferences.

[0095] In the embodiments of this application, generating user profile data refers to aggregating the aforementioned isolated metrics into a structured JSON (JavaScript Object Notation) object or database record to form a comprehensive capability view of the target object. This user profile data is not only a static collection of tags, but also a dynamic data model that includes capability shortcomings and preferred improvement paths.

[0096] In some embodiments of this application, the user profile data also integrates the ability evolution trend over time, quantifying the skill acquisition rate of the target object by comparing historical profile snapshots. This profile data is directly used as the input state variable for subsequent path planning algorithms, ensuring that the task scheduling system can perform calculations based on comprehensive, objective, and business-in-depth personnel characteristics.

[0097] As can be seen, the embodiments of this application transform heterogeneous training process data into a unified multi-dimensional feature vector, extract high-order skill semantics using a semantic model, and construct a multi-dimensional, high-precision user profile by quantifying proficiency, missing items, and preferences. This solves the technical defect in related technologies where evaluation relies on a single score and cannot identify specific skill deficiencies.

[0098] In some embodiments of this application, under the constraints of task orchestration, user profile data is used to determine the target task sequence matching the target object from the skill training task library of the skill training platform, including:

[0099] S301. Under the constraints of task arrangement, encode multiple skill training tasks in the skill training task library to generate a chromosome sequence population including multiple initial task chromosome sequences.

[0100] In the embodiments of this application, encoding refers to the process of mapping discrete skill training tasks into gene sequences that can be processed by computer algorithms. The initial task chromosome sequence refers to a candidate task permutation scheme randomly generated or generated based on heuristic rules during the algorithm initialization phase, with each gene position corresponding to a specific training task identifier. The chromosome sequence population refers to a set composed of several initial task chromosome sequences, representing the initial solution set in the search space.

[0101] In some embodiments of this application, an integer encoding method is used, where each integer represents a task index in the skill training task library. Specifically, if the skill training task library contains P skill training tasks, each task is uniquely encoded as an integer index value in the range of 1 to P. An initial task chromosome sequence consists of Q gene positions, where Q is the number of tasks to be selected in this arrangement. Each gene position stores a task index integer value, and the order of the gene positions represents the execution order of the tasks. The system randomly selects Q non-repeating integer values ​​from the integer range of 1 to P using a random number generator and randomly arranges them into a chromosome sequence. This process is repeated to generate a preset number (e.g., 100) of initial task chromosome sequences, forming an initial chromosome sequence population. During the generation process, the task arrangement constraints act as a filter.

[0102] Specifically, the system uses the CSP solver to perform hard constraint verification on the generated sequence. The verification rules include: (1) Skill dependency verification, that is, traversing each task index in the chromosome sequence, querying all the predecessor dependent skill nodes of the skill node corresponding to the task in the skill knowledge graph, verifying whether the task indexes corresponding to these predecessor dependent skills have all appeared in the gene position before the current task index. If there is a situation where the predecessor task is ranked after the successor task, it is determined to be a violation of the constraint; (2) Virtual-to-real ratio verification, that is, counting the ratio of the number of virtual online tasks to the number of physical offline tasks in the chromosome sequence, verifying whether the ratio is within the deviation range allowed by the preset task ratio in the task arrangement constraint; (3) Total duration verification, that is, accumulating the standard execution duration of all tasks in the chromosome sequence, verifying whether the total does not exceed the total duration of the task target. Illegal sequences that violate skill dependency (such as predecessor skills not being ranked before successor skills) or violate the virtual-to-real ratio limit are eliminated to ensure that each individual in the initial population is logically executable.

[0103] S302. For each initial task chromosome sequence, determine the expected task matching degree between the initial task chromosome sequence and the user profile data.

[0104] In the embodiments of this application, the expected task matching degree is a quantitative indicator that measures whether the current task sequence theoretically matches the ability characteristics of the target object. This indicator reflects the degree of fit between the task difficulty and the target object's current skill proficiency level, as well as the extent to which the task coverage compensates for the target object's skill deficiencies.

[0105] In some embodiments of this application, the preset attribute features of each skill training task (including but not limited to multi-dimensional feature indicators such as task preset difficulty, skill point coverage, and preference type) are first extracted and aligned and spliced ​​to generate a task attribute vector corresponding to the dimension of the user profile feature vector; then, the expected matching degree of the task is determined by calculating the cosine similarity or Euclidean distance between the task attribute vector and the user profile feature vector.

[0106] Skill point coverage refers to the ratio of the number of skill nodes in the skill knowledge graph involved in the skill training task to the total number of skill nodes in the skill knowledge graph. This parameter is pre-stored in the task metadata of the skill training task library. Preference type refers to the learning method tag (e.g., visual guidance type, hands-on trial and error type) that is suitable for the skill training task. This tag uses the same tag system as the skill preference items in the user profile data to achieve dimensional alignment.

[0107] The specific process of feature extraction and alignment is as follows: the normalized values ​​of the task's preset difficulty, the skill point coverage values, and the vectors of preference types after one-hot encoding are concatenated in the same order as the corresponding dimensions in the user profile feature vector (i.e., the normalized values ​​of the current skill proficiency level, the multi-label binary vectors of skill missing items, and the one-hot encoded vectors of skill preference items), thereby forming a task attribute vector with the same dimensions as the user profile feature vector.

[0108] S303. Based on the expected matching degree of the task, determine the function value of the preset fitness function.

[0109] In the embodiments of this application, the preset fitness function is a core function in genetic algorithms used to evaluate the quality of individual chromosomes, and it determines the direction of population evolution. The higher the function value, the better the quality of the task sequence.

[0110] S304. Under the constraints of task arrangement, the selection, crossover and mutation operations are performed on the chromosome sequence population using the function value of the preset fitness function to obtain the converged chromosome sequence population.

[0111] In the embodiments of this application, selection refers to the process of weeding out inferior individuals based on their fitness function values, retaining superior individuals for the next generation. For example, a roulette wheel selection method can be used, where the probability of each individual being selected is equal to its fitness function value divided by the sum of the fitness function values ​​of all individuals in the current population. This ensures that chromosomes with high fitness have a greater chance of survival and selection, while maintaining a certain level of population diversity. Crossover refers to exchanging partial gene segments between two parent chromosomes to generate new individuals. Specifically, a single-point crossover method is used, where a crossover point aligned at the same position is randomly selected from two parent chromosome sequences, and the gene sequences after that point are exchanged. Since crossover may result in duplicate task indices in the same chromosome sequence, the system performs a deduplication and repair operation after crossover: detecting duplicate task indices in the offspring chromosomes and replacing duplicates with compliant task indices not inherited from the parents, ensuring the uniqueness of task indices in each chromosome sequence. The execution probability of crossover is controlled by a preset crossover probability parameter, for example, set to 0.8, meaning that each selected parent pair performs crossover with an 80% probability; otherwise, it directly copies into the next generation.

[0112] Mutation operation refers to randomly changing the value of a gene locus in a chromosome to increase population diversity. For example, multi-point mutation can be used. According to the preset mutation probability parameter (e.g., set to 0.05, that is, each gene locus has a 5% probability of mutation), one or more gene loci are randomly selected in the chromosome sequence, and the old task index on the gene locus is directly replaced by the index of other compliant tasks with similar difficulty coefficients in the skill training task library. The converged chromosome sequence population refers to the population state when the fitness tends to be stable or the preset termination condition is reached after multiple rounds of iterative evolution. Among them, the preset termination conditions include: (1) the number of iterations reaches the preset maximum number of iteration generations (e.g., 500 generations); or (2) the fitness function value of the best individual in the population increases less than the preset convergence threshold (e.g., 0.001) within a consecutive preset number of generations (e.g., 50 generations). When any of the above conditions are met, the genetic algorithm stops iterating and outputs the current population as the converged chromosome sequence population.

[0113] In some embodiments of this application, task orchestration constraints must be reintroduced to address crossover and mutation operations. If a mutation operation results in a sequence that violates skill dependencies (e.g., performing high-level practical tasks before basic theoretical learning), the system can adjust the gene order or discard the mutated individual to ensure that the evolutionary process always occurs within the feasible solution space.

[0114] S305. Based on the converged chromosome sequence population, determine the target task sequence that matches the target object.

[0115] In the embodiments of this application, determining the target task sequence refers to selecting the individual with the highest fitness function value from the converged chromosome sequence population and decoding it into a final list of tasks that can be issued. The decoding process specifically involves: reading the integer index value stored at each gene position in the optimal chromosome sequence; querying the complete metadata of the corresponding skill training task (including task name, task type, standard execution time, associated device information, etc.) in the skill training task database based on this index value; and organizing all the queried task metadata into an ordered task list according to the order of the gene positions. This target task sequence includes a clear task execution order, a logic for interleaving virtual and real tasks, and expected time nodes.

[0116] As can be seen, the embodiments of this application solve the technical challenge of generating personalized training paths under complex skill dependencies and resource constraints by combining constraint satisfaction with evolutionary computation. This scheme utilizes hard constraints to ensure the compliance of task logic and uses a fitness function to guide the sequence to evolve towards the direction with the highest ability matching degree, achieving path optimization in a large-scale solution space and ensuring the accuracy of the training plan.

[0117] In some embodiments of this application, determining the expected task match between the initial task chromosome sequence and user profile data includes:

[0118] S401. Based on the task difficulty coefficient and task type of each skill training task in the initial task chromosome sequence, generate a task difficulty coefficient sequence and a task type sequence.

[0119] In the embodiments of this application, the task difficulty coefficient is a predefined quantitative value in the skills training task library, typically ranging from 0 to 1, used to characterize the cognitive load, operational precision, or logical complexity required to complete the task. This value is entered into the task library after being comprehensively evaluated by experts in the training field based on factors such as the depth of knowledge points involved in the task, the number of operational steps, and the error tolerance rate.

[0120] Task type is a label used to distinguish the business scope to which a skills training task belongs, such as theoretical knowledge, simulation operation, troubleshooting, or emergency response. The task difficulty coefficient sequence and task type sequence are ordered sets formed by replacing the task ID (Identifier) ​​corresponding to each gene position in the initial task chromosome sequence with the corresponding attribute value. Specifically, the system traverses each gene position in the initial task chromosome sequence, queries the skills training task database for the task difficulty coefficient and task type fields based on the task index value at that gene position, arranges the task difficulty coefficients corresponding to all gene positions in order to form the task difficulty coefficient sequence, and arranges the task types corresponding to all gene positions in order to form the task type sequence.

[0121] S402. Based on the task difficulty coefficient sequence and the current skill proficiency level, determine the ability matching deviation value.

[0122] In the embodiments of this application, the ability matching deviation value is a numerical indicator used to measure the degree of mismatch between task load and personnel ability. Ideally, the task difficulty should be slightly higher than the personnel's current skill proficiency level to form an effective skill improvement zone (i.e., zone of proximal development). If the difficulty is too low, it will lead to inefficient training; if it is too high, it will lead to cognitive overload.

[0123] In some embodiments of this application, the root mean square error (RMSE) algorithm is used to calculate the deviation value. Specifically, the system sets a dynamic learning gain step size and calculates the Euclidean distance between the sum of the current skill proficiency level and the gain step size, and the values ​​in the task difficulty coefficient sequence. The smaller this ability matching deviation value, the more the difficulty gradient design of the task sequence matches the target object's current learning ability, effectively avoiding the waste of training resources caused by difficulty mismatch.

[0124] S403. Determine the task type matching index based on the task type sequence and skill missing items.

[0125] In the embodiments of this application, the task type matching index is a numerical indicator used to measure the degree to which a task sequence specifically covers skill gaps. Skill gaps have been clearly identified in the user profile data.

[0126] In some embodiments of this application, step S403 can be implemented through set operations. Specifically, the system traverses the task type sequence, extracts all task types to form a task set, and maps the skill deficiency items to the required types to form a demand set. The system uses the Jaccard Similarity Coefficient to calculate the task type matching index, which is the ratio of the number of elements in the intersection of the two sets to the number of elements in the union. For core deficiency items, a weighted summation method can be introduced. For example, if the target object lacks the skill of precision assembly, but the sequence contains a high proportion of precision assembly-related entities or virtual tasks, the calculated task type matching index will be high.

[0127] S404. Use the capability matching deviation value and the task type matching index to determine the expected matching degree of the task.

[0128] In the embodiments of this application, the expected task matching degree is the final score that comprehensively evaluates the suitability of the task sequence, serving as a core component of the fitness function in the genetic algorithm. Since the ability matching deviation value is a negative indicator (the smaller the better), while the task type matching index is a positive indicator (the larger the better), normalization and weighted fusion are required.

[0129] In some embodiments of this application, a linear weighted formula is used to calculate: the expected task matching degree equals the task type matching index multiplied by a first preset weight, minus the ability matching deviation value multiplied by a second preset weight. This calculation logic ensures that the final selected task sequence not only accurately corresponds to the weak points of the target object in terms of content, but also strictly matches the capability boundaries of the target object in terms of difficulty.

[0130] As can be seen, this application's embodiments, by constructing a multi-dimensional assessment of difficulty and type, and utilizing numerical deviation calculation and index statistics, achieve an accurate measurement of the quality of task sequences. This solution can automatically eliminate task combinations with excessively large difficulty spans or weak targeting, ensuring that the final scheduled training content maintains an appropriate level of challenge to promote skill internalization while accurately covering capability gaps to achieve efficient improvement.

[0131] In some embodiments of this application, the specific data processing logic for determining the capability matching deviation value is as follows:

[0132] First, the system establishes a capability-difficulty mapping coordinate system. The system reads the user's current skill proficiency level L. user (e.g., Level 3), and based on the user's historical learning rate V history (i.e., the rate at which skill level improves per unit time), calculate the dynamic learning gain step size Δ, the formula is: Δ=α×(1-L) user / L max )+β×V history Where α is the basic excitation factor, β is the inertia factor, and L max This represents the highest skill level. The logic behind this formula is: the lower the level, the larger the preset gain step size (encouraging challenge); the faster the historical learning speed, the larger the gain step size.

[0133] Then, the target difficulty value D is calculated. target =L user +Δ.

[0134] Next, obtain the task difficulty coefficient sequence S. diff ={d1, d2, ..., d n The system not only calculates the overall sequence RMSE, but also uses a weighted sliding window approach to calculate local bias. For each task difficulty d in the sequence... i Calculate its relationship with the target difficulty value D. target The square of the difference (d) i -D target ) 2 .

[0135] Finally, calculate the capability matching deviation value E. match :

[0136]

[0137] Where, d i The task difficulty coefficient sequence S diff The difficulty coefficient of the i-th task, w i The positional weighting coefficients are used to represent tasks at the beginning of the sequence with lower weights and tasks at the end with higher weights, simulating the process of a user's ability gradually approaching the target difficulty level as the task progresses. This logic ensures that the calculated ability matching deviation value reflects the user's expected ability growth during the training process.

[0138] In some embodiments of this application, the data processing flow for determining the expected task matching degree using capability matching deviation value and task type matching index is as follows:

[0139] The system first matches the acquired capability deviation value V. dev Task type matching index V type Perform extreme value normalization.

[0140] For the capability mismatch deviation value (negative indicator), the normalization formula is:

[0141] N dev = (V dev_max -V dev ) / (V dev_max -V dev_min )

[0142] Among them, V dev_max and V dev_min These represent the maximum and minimum ability matching deviation values ​​for all chromosome sequences in this generation of the population.

[0143] For the task type matching index (positive indicator), the normalization formula is:

[0144] N type = (V type -V type_min ) / (V type_max -V type_min )

[0145] Among them, V type_max and V type_min These represent the maximum and minimum values ​​of the task type matching index for all chromosome sequences in this generation of the population.

[0146] Furthermore, to prevent the overperformance of one dimension from masking the deficiencies of another, the embodiments of this application do not simply employ linear weighting, but instead introduce a nonlinear fusion model with a penalty factor. The expected task matching degree F fit The calculation formula is as follows:

[0147] F fit =W1×N type +W2×N dev ×P constraint

[0148] Where W1 and W2 are preset weights, and W1 + W2 = 1. P constraint To constrain the penalty factor, when N dev When the value is below a preset threshold (e.g., 0.2, indicating an extreme mismatch in difficulty), P... constraint The value is 0.5 if the difficulty is appropriate, and 1 otherwise. This logic ensures that a high type matching degree is only meaningful when the difficulty is suitable, thus guaranteeing the logical validity of the generated task sequence.

[0149] In some embodiments of this application, the function value of a preset fitness function is determined based on the expected task matching degree, including:

[0150] S501. Based on the initial task chromosome sequence and user profile data, determine the expected skill improvement of the target object.

[0151] In the embodiments of this application, the expected skill improvement refers to the total quantitative increase in the skill proficiency level of the target object after it has completed all the skill training tasks in the initial task chromosome sequence in sequence.

[0152] In some embodiments of this application, the system invokes a preset cognitive gain model, inputs the user's current skill state and the knowledge point coverage density of each task in the sequence, and outputs a scalar value. The knowledge point coverage density refers to the ratio of the total number of times skill nodes are covered in the skill knowledge graph involved in all tasks of the chromosome sequence to the total number of tasks in the sequence, used to characterize the degree of focus of the task sequence on knowledge points.

[0153] This cognitive gain model is based on Item Response Theory (IRT) and utilizes a Multi-Layer Perceptron (MLP) structure. It inputs the hidden layer network by concatenating the user's current ability feature vector, the task's ability requirement feature vector, and the inter-task association density matrix. The user's current ability feature vector is extracted from user profile data, containing normalized values ​​of the current skill proficiency level and scores for each skill dimension. The task's ability requirement feature vector is extracted from a skill training task library, containing attributes such as the task's difficulty coefficient, discrimination parameter, and skill point coverage. The inter-task association density matrix is ​​an n×n square matrix (where n is the number of tasks in the chromosome sequence). The (i, j)th element in the matrix represents the ratio of the number of skill nodes shared by task i and task j in the skill knowledge graph to the union of the skill nodes involved in both tasks, used to quantify the strength of knowledge associations between tasks. The splicing process is as follows: flatten the user capability feature vector and the capability requirement feature vector for each task, and then splice them sequentially with the upper triangular element vectors of the correlation density matrix to form the input vector of the multilayer perceptron.

[0154] The multilayer perceptron model comprises an input layer, multiple fully connected hidden layers, and a single-node output layer. The input concatenated vector, after being passed through the fully connected hidden layers and mapped using non-linear activation functions (such as ReLU or GeLU), is directly calculated and output by the linear neurons in the output layer, representing the expected skill improvement. Furthermore, before application, this multilayer perceptron undergoes supervised training using historical interaction data from past learners and actual skill improvement assessment values ​​as training data and labels. The network weights are continuously updated iteratively using a gradient descent algorithm based on the mean squared error (MSE) loss function. The model implicitly incorporates reinforcement effects between tasks; for example, if the sequence includes a reasonable combination of theory followed by practice, the model assigns additional gain weights, thereby calculating a higher expected improvement than simply adding up single tasks.

[0155] S502. Determine the total expected duration of the initial task chromosome sequence.

[0156] In the embodiments of this application, the total expected duration of tasks refers to the total time cost required to execute the chromosome sequence. This parameter includes not only the standard execution time of each virtual online task and physical offline task in the sequence, but also the buffer time or device preparation time between task switching.

[0157] In some embodiments of this application, the system retrieves historical average time data or standard working hour data for each task from the skills training task library and performs cumulative calculations. If a task in a physical location requires specific site turnover time, this will also be included in the sum, to prevent the generated task sequence from being effective but taking too long and exceeding the cycle limit of the training course.

[0158] S503. Determine the function value of the preset fitness function based on the sum of the expected task matching degree, expected skill improvement degree, and expected task duration.

[0159] In the embodiments of this application, the preset fitness function is a comprehensive evaluation formula used in genetic algorithms to assess the merits of individuals, which converges multi-dimensional optimization objectives into a single comparable value.

[0160] In some embodiments of this application, a pre-defined fitness function is constructed using a weighted sum. Specifically, the expected task matching degree and expected skill improvement degree are assigned positive weights as positive gain terms; the deviation between the total expected task duration and the pre-defined target duration is assigned negative weights as a negative penalty term. Before calculation, the above three parameters need to be normalized to eliminate dimensional differences. The larger the calculated function value, the more the chromosome sequence (i.e., the task arrangement scheme) best meets the time cost constraint while satisfying personalized matching and maximizing skill improvement. This function value will directly determine the survival probability of the sequence in the population evolution process.

[0161] As can be seen, this application's embodiments transform a simple matching problem into a multi-objective optimization problem by introducing skill enhancement degree and time cost as key dimensions of the fitness function. This scheme constructs a comprehensive evaluation system that includes both benefits and costs, ensuring that the ultimately selected task sequence not only fits user characteristics but also maximizes the benefits of skill growth within limited time resources.

[0162] In some embodiments of this application, the expected skill improvement of the target object is determined based on the initial task chromosome sequence and user profile data, including:

[0163] S601. Based on the task type, skill missing items, and skill preference items of each skill training task in the initial task chromosome sequence, determine the expected skill improvement coefficient of each skill training task in the initial task chromosome sequence.

[0164] In the embodiments of this application, the skill expectation enhancement coefficient is a moderating factor characterizing the marginal utility of a specific task for a specific target object. This coefficient reflects the coupling depth between task content and object requirements.

[0165] In some embodiments of this application, the process of determining the expected skill improvement coefficient is implemented through a conditional logic matrix: when the task type of a certain skill training task happens to match the skill missing item in the user profile data, that is, when the task type label of the task corresponds to at least one label in the set of skill missing item labels (the correspondence is determined by the "task type-skill node" mapping relationship in the skill knowledge graph), the system assigns a higher expected skill improvement coefficient to the skill training task (e.g., 1.1).

[0166] If the presentation method or operation logic of the skill training task matches the target object's skill preference items (e.g., the target object's preference is logically deduced through virtual simulation), that is, the preference type label marked in the skill training task library is consistent with the skill preference item label in the user profile data, since preference matching means that the target object already has a good operational habit foundation and its marginal improvement space is relatively small, the system assigns a low expected skill improvement coefficient (e.g., 0.9) to the skill training task.

[0167] If the task type matches a missing skill and the preference type also matches, then the value of the corresponding cell in the conditional logic matrix is ​​retrieved (e.g., 1.0). If the task type does not match a missing skill and the preference type does not match, then a base coefficient is assigned (e.g., 0.8).

[0168] In this way, the differentiated effects of different skills training tasks on different individuals can be accurately measured by adjusting the proportions in a personalized manner.

[0169] S602. Obtain the skill gain contribution value of each skill training task in the initial task chromosome sequence.

[0170] In the embodiments of this application, the skill gain contribution value refers to the baseline contribution of the skill training task to the improvement of skill proficiency under standard working conditions. Its range is pre-set by training experts or historical big data statistical analysis and stored in the metadata field of the skill training task database. Specifically, this value is set by: training domain experts comprehensively scoring based on the number of core knowledge points involved in the task, the complexity of the operational steps, and the importance weight of the skill in the industry's job competency standards; or by using historical big data statistical analysis methods, i.e., statistically analyzing the average change in skill assessment scores of a large number of historical trainees before and after completing the task as the baseline contribution value. This value is a dimensionless positive real number. For example, the skill gain contribution value of completing a high-voltage cable connector fabrication task is usually higher than that of completing a basic electrical tool identification task.

[0171] In some embodiments of this application, the skill gain contribution value is multidimensional, corresponding to different ability dimensions in the skill knowledge graph of the skill training platform, such as theoretical knowledge dimension, hands-on operation dimension, and safety awareness dimension.

[0172] S603. Determine the expected skill improvement degree based on the skill gain contribution value and the expected skill improvement coefficient.

[0173] In the embodiments of this application, the expected skill improvement is the final estimated total increase in the target object's capabilities across the entire task sequence. This determination process reflects the logic of on-demand allocation of task value.

[0174] In some embodiments of this application, a multiplicative summation algorithm is used for calculation: the skill gain contribution value of each task in the initial task chromosome sequence is multiplied by its corresponding expected skill improvement coefficient to obtain the actual expected gain of the task in this scheduling, and then the actual expected gains of all tasks in the sequence are summed.

[0175] As can be seen, the embodiments of this application quantify the synergistic effect between the value of the task itself and the suitability for individual needs by multiplying the expected skill enhancement coefficient and the skill gain contribution value. This scheme uses skill deficiency items and skill preference items to weight the baseline contribution, which can not only prioritize the selection of key tasks to fill skill gaps, but also take into account the individual's learning characteristics.

[0176] In some embodiments of this application, the process of determining the expected skill improvement is not a linear summation of the values ​​of individual tasks, but rather introduces a cognitive fatigue decay factor to simulate the attentional characteristics of the human brain. The cognitive fatigue decay factor is a dynamic variable that decreases non-linearly with the increasing number of consecutive executions of a high-cognitive-load task, used to characterize the law of diminishing marginal utility. The specific calculation logic is as follows:

[0177] The system first scans the initial task chromosome sequence to identify consecutive clusters of high-difficulty tasks (e.g., consecutive task segments with a difficulty coefficient exceeding a preset fatigue threshold (e.g., 0.7)). For the Nth task (N≥1) within this high-difficulty task cluster, the formula for calculating its single-point effective gain E(n) is:

[0178] E(n) = V × K × D(n)

[0179] Where V represents the skill gain contribution value of the task; K represents the expected skill improvement coefficient of the task; and D(n) is the cognitive fatigue decay factor. D(n) is constructed using an exponential decay model, specifically:

[0180] D(n)=exp(-λ×(N-1))

[0181] Here, λ is the fatigue decay constant, which can be negatively correlated with, for example, the focus persistence feature in user profile data. When N=1, the decay factor is 1, indicating that the first high-difficulty task has the best effect; as N increases (i.e., high-difficulty tasks are executed continuously), D(n) decreases rapidly, causing the effective part included in the total amount to shrink significantly even if the theoretical value of subsequent tasks is high. Finally, the system accumulates the single-point effective gains of all skill training tasks in the initial task chromosome sequence after decay correction to obtain the expected skill improvement of the target object. This calculation mechanism allows the genetic algorithm to naturally eliminate chromosome sequences that pile up high-difficulty tasks during the evolution process, and tend to select sequences that reasonably intersperse low-load buffer tasks (such as simple review) to reset the N value and restore the decay factor to a high level.

[0182] As can be seen, the embodiments of this application introduce a cognitive fatigue decay model into the task value assessment and construct a nonlinear skill gain calculation logic. This solves the problem that scheduling algorithms ignore continuous high-load tasks, which leads to a marginal decrease in learning efficiency. It forces the task scheduling strategy to actively adapt to the learner's attention physiological cycle and ensures that the generated task sequence achieves a balance between work and rest.

[0183] In some embodiments of this application, an alternative method for calculating the fatigue decay constant λ is provided.

[0184] Specifically, the first step is to identify the low-difficulty task clusters that precede consecutive high-difficulty task clusters in the initial task chromosome sequence. These low-difficulty task clusters are buffer sequences consisting of consecutive low-cognitive-load tasks immediately preceding the high-difficulty task clusters. Low-cognitive-load tasks typically refer to knowledge review or video viewing tasks with a difficulty coefficient below a preset recovery threshold (e.g., 0.3).

[0185] Secondly, the number M of skill training tasks within the low-difficulty task cluster and the recovery attribute weight of each skill training task are obtained. Here, the number M refers to the number of consecutive tasks contained in the low-difficulty task cluster, which directly quantifies the length of the cognitive recovery window. The recovery attribute weight is a predefined cognitive recovery efficacy parameter based on the specific interaction form of the task. For example, a purely passively watched instructional video has a high recovery attribute weight (e.g., 1.0), while a test task requiring simple multiple-choice interaction, although low in difficulty, still consumes a small amount of attention, resulting in a lower recovery attribute weight (e.g., 0.8). Obtaining these parameters aims to provide data support for calculating the amount of cognitive energy recovered.

[0186] Then, based on the quantity M, the recovery attribute weights, and the basic focus factor in the user profile data, the fatigue decay constant λ is calculated. The fatigue decay constant λ is not a static value but a dynamic variable influenced by the quality of preceding rest. The basic focus factor is an inherent baseline value λ stored in the user profile data, representing the target object's fatigue resistance. base The smaller this value, the stronger the endurance. The fatigue decay constant λ applicable to this high-difficulty task cluster can be calculated using a cognitive recovery compensation model. The calculation formula is:

[0187]

[0188] Where w_i is the recovery attribute weight of the i-th task in the low-difficulty task cluster, and C is a regularization constant to prevent the denominator from being too small. This formula reflects the following technical logic: as the number of preceding low-difficulty tasks M increases and the task recovery attribute is enhanced, the corrected λ value will be significantly smaller than the base value λ. base This means that if the target subject receives sufficient low-load rest before entering high-intensity training, their attention will decay more slowly when performing consecutive high-difficulty tasks, thus maintaining a more sustained and efficient learning state; conversely, if the M value is too small, λ base The value will approach or exceed the baseline, leading to a rapid decline in the expected improvement in skills.

[0189] As can be seen, the embodiments of this application dynamically correct the fatigue decay constant λ by analyzing the low-difficulty buffer blocks in the task sequence and introducing the quantity M and recovery attribute weights. This scheme constructs a quantitative correlation model of rest-anti-fatigue, simulating the physiological mechanism of reducing the subsequent cognitive decay rate by reasonably interspersing low-load tasks, enabling the task scheduling algorithm to accurately predict the actual human factor efficiency under different arrangement structures.

[0190] In some embodiments of this application, after determining the task arrangement information of the target object on the skills training platform based on the target task sequence, the method further includes:

[0191] S701. Obtain the first interaction log generated by the target object when executing the target virtual online task.

[0192] In the embodiments of this application, the target virtual online task refers to the training subject in the target task sequence that needs to be completed in a computer simulation environment or a virtual reality environment. The first interaction log refers to the underlying behavioral data of the target object in the virtual environment, recorded in real time by the system backend.

[0193] In some embodiments of this application, the first interaction log includes a series of timestamps, atomic opcodes, resource request identifiers, and system alarm information triggered due to improper operation, arranged in chronological order. This log is not only a record of the task completion result but also a digital reconstruction of the operation process logic, providing original evidence for subsequent analysis of cognitive difficulties related to the target object.

[0194] S702. Identify the skill confusion keywords in the first interaction log.

[0195] In the embodiments of this application, the term "skill confusion" refers to a professional technical term that can characterize the cognitive obstacles, logical blind spots, or operational bottlenecks encountered by the target object during practical training.

[0196] In some embodiments of this application, the system utilizes Named Entity Recognition (NER) algorithms from Natural Language Processing (NLP) to semantically model the abnormal behavior trajectories in the first interaction log. For example, a sequence labeling network based on BiLSTM-CRF (Bidirectional Long Short-Term Memory Network-Conditional Random Field) can be used as the core model of this named entity recognition algorithm. During the model training phase, a large number of pre-collected historical log samples, manually labeled with entity tags and boundaries, are used to supervise the training of the BiLSTM-CRF model. The weight parameters of the model are continuously updated by optimizing the maximum log-likelihood function. In practical use, the pre-prepared text corpus of the behavior of the first interaction log is input into the corresponding BiLSTM layer to extract the bidirectional semantic dependency feature information of the preceding and following context. This information is then passed to the parameter-shared CRF layer to learn the transmission probability between adjacent entities based on the features learned from the annotations. At the model output level, the Viterbi decoding algorithm is used uniformly to output the globally optimal probability entity tag sequence, thereby extracting the word combinations labeled by the model as skill or fault entities.

[0197] Specifically, log transformation rules can be pre-defined to translate structured first-interaction logs (such as timestamps, error codes, and click coordinates) into text corpora describing the behavior, which can then be analyzed using the NER algorithm. For example, when the logs show that the target object repeatedly performs ineffective component replacements in a specific circuit simulation and frequently consults the reactive power compensation principle help document, the system extracts reactive power compensation as a skill confusion keyword through text transformation and NER analysis. This step transforms discrete interactive behaviors into clear knowledge gaps, enabling accurate identification of training difficulties.

[0198] In some embodiments of this application, the data processing logic for determining skill confusion keywords in the first interaction log includes two stages: feature mapping and semantic reasoning.

[0199] The first stage is behavioral feature mapping. The system parses the first interaction log and extracts the following three types of feature vectors:

[0200] High-frequency error characteristics: Statistically analyze the triggering frequency of undo / redo commands and the number of times error codes are reproduced in the same operation step.

[0201] Dwell characteristics: Calculate a heatmap of mouse hover duration and click density on a specific virtual user interface (UI) or simulated component.

[0202] Help Feature: Extract contextual tags associated with when a user clicks on help documents or intelligent Q&A.

[0203] The second stage involves semantic reasoning and keyword generation. The system invokes a pre-built fault-principle knowledge graph. If a high-frequency error feature is detected pointing to "circuit breaker closing failure," and the lingering features are concentrated in the "energy storage handle" area, the knowledge graph reasoning engine maps these two behavioral entities to their common parent node—"spring energy storage principle." The system then marks the metadata name of this parent node as a candidate keyword.

[0204] Finally, the system combines the search terms in the help-seeking features (e.g., a user searched for "cannot close the circuit breaker") and calculates the relevance weight of candidate keywords using the TF-IDF (term frequency–inverse document frequency) algorithm. Term frequency is the ratio of the number of times the candidate keyword appears in the currently extracted behavioral feature document to the total number of words in that feature document. Inverse document frequency is the logarithm of the inverse of the proportion of historical log documents containing the candidate keyword to the total number of log documents. Multiplying term frequency and inverse document frequency yields the relevance weight of the candidate keyword. The term with the highest weight (e.g., circuit breaker energy storage mechanism) is then identified as the final skill confusion keyword.

[0205] S703. In the skills training task library, identify the offline tasks that match the keywords for skills confusion.

[0206] In the embodiments of this application, physical offline tasks refer to practical testing, installation, or maintenance tasks that need to be carried out in a real physical setting or using physical training devices. Determining matching physical offline tasks means searching for task items in the skills training task library that can physically verify or physically demonstrate the principles involved in the keywords of skill confusion.

[0207] In some embodiments of this application, the system maintains a many-to-many mapping matrix between knowledge points and tasks. This mapping matrix is ​​a pre-constructed two-dimensional matrix, where the row indices correspond to all skill knowledge points (i.e., skill nodes) in the skill knowledge graph, and the column indices correspond to all offline tasks in the skill training task library. The value of each cell in the matrix is ​​0 or 1, where 1 indicates that there is a teaching coverage relationship between the knowledge point corresponding to that row and the offline task corresponding to that column. The system performs semantic similarity matching between skill confusion keywords and the row indices of the mapping matrix to locate the corresponding row. Then, it filters out the columns with a value of 1 in that row, which is the set of matching candidate offline tasks. Finally, it sorts them according to their relevance (if there are multiple matching tasks, the task with the highest relevance to the confusion keyword and available on the current training equipment is selected first) to determine the final offline task. If the skill confusion keyword is "high-voltage cable stripping," the system will automatically match the offline task of cable accessory installation practice to make up for the shortcomings of virtual simulation in tactile feedback or complex spatial perception by utilizing the intuitiveness and certainty of offline practice, thereby eliminating cognitive misconceptions of the target object.

[0208] S704. Insert the offline tasks that match the keywords for skill confusion into the task arrangement information before the skill training tasks that have not yet been executed.

[0209] In the embodiments of this application, task orchestration information refers to a pre-generated set of scheduling instructions that includes the task execution order and time window. The insertion operation refers to dynamically modifying the execution sequence of this instruction set, assigning the highest execution priority to newly determined tasks under the entity line.

[0210] In some embodiments of this application, the scheduling algorithm automatically detects the availability of equipment in the current training site and, provided that hardware resources do not conflict, forcibly places the entity's offline task at the top of the queue. This dynamic response logic enables timely adjustments to the training plan, ensuring that the target object receives correction through offline practice as soon as confusion arises, thus preventing the extension of cognitive biases into subsequent tasks.

[0211] As can be seen, this embodiment of the application accurately captures skill difficulties by monitoring the interaction details during virtual training in real time and utilizing natural language processing technology, and dynamically triggers associated offline practical tasks. This solution constructs a feedback loop from virtual to real, enabling the task scheduling information to adaptively adjust to sudden confusion during training, thus solving the problem that preset scheduling schemes cannot cope with learners' real-time cognitive fluctuations.

[0212] In some embodiments of this application, after determining the task arrangement information of the target object on the skills training platform based on the target task sequence, the method further includes:

[0213] S801. Obtain the second interaction log generated by the target object executing the target entity's offline task.

[0214] In the embodiments of this application, the offline task of the target entity refers to a practical task performed in a physical training site using real training equipment, tools, or instruments. The second interaction log refers to a record collected through Internet of Things (IoT) sensing terminals, wearable detection devices, or intelligent video surveillance systems, which records the operation trajectory and results of the target object in the physical space.

[0215] In some embodiments of this application, the second interaction log includes high-frequency pose coordinate data, pressure sensing data, torque monitoring values, and start and end timestamps of the operation procedures. Unlike logs in a virtual online environment, the second interaction log reflects the actual performance of the target object in a real gravity environment, physical feedback pressure, and complex practical space.

[0216] S802. Based on the second interaction log, determine the target object's weak skills.

[0217] In the embodiments of this application, weak skills refer to specific skill items that the target object exhibits during the execution of offline tasks, failing to meet preset process standards or evaluation criteria. The process of determining weak skills involves comparing extracted operational features with a standard digital twin model.

[0218] In some embodiments of this application, the system employs a Dynamic Time Warping (DTW) algorithm to perform morphological matching between the operation sequences in the second interaction log and the expert standard operation sequences. For example, the DTW calculation process may include: first, obtaining the time series dataset in the second interaction log as a test sequence, and calculating the Euclidean distance between it and the data points corresponding to the expert standard operation sequences one by one to construct a local cost matrix; then, using the idea of ​​dynamic programming, recursively searching for a warped alignment path from the starting time point matrix to the ending time point matrix in the local cost matrix domain, and ensuring that the sum of local costs in the cells where the alignment path is located meets the minimum requirement during the calculation process, the minimum distance integral value is the total DTW distance measuring the difference between the two sequences. If, in a specific process (such as precision bearing press fitting), the force curve of the target object fluctuates too much or the time taken is far greater than the average (i.e., the total DTW distance calculated above is greater than the preset entity difference tolerance threshold), the system determines that the skill point corresponding to the task is a weak skill.

[0219] The preset entity difference tolerance threshold is a value pre-set by training experts based on the process standards and safety requirements of each process, and is stored in the system configuration parameters. The skill point corresponding to this task refers to the specific operation process that generates abnormal DTW distance, which is back-linked to the skill knowledge point being examined through the mapping relationship between process nodes and skill nodes in the skill knowledge graph. This skill knowledge point is then marked as a weak skill.

[0220] S803. In the skills training task library, identify virtual online tasks that match the weak skills.

[0221] In the embodiments of this application, determining the virtual online task that matches the weak skill refers to retrieving digital teaching resources from the task library that can strengthen the principle, break down the logic, or conduct high-frequency simulation practice of the weak skill.

[0222] In some embodiments of this application, the system uses a pre-constructed skill knowledge mapping graph to find the simulation logic module with the highest correlation to the weak skill. Specifically, the skill knowledge mapping graph maintains many-to-many associations between skill nodes and virtual online tasks. The system uses a identified weak skill as a query node, retrieves all associated virtual online tasks for that skill node in the skill knowledge mapping graph, and sorts them from high to low according to the weight value of the associated edge (this weight value represents the theoretical coverage depth of the virtual online task to the skill node, preset by the training expert). One or more virtual online tasks with the highest ranking are selected as the matching results. For example, if the target object exhibits logical confusion regarding "complex circuit fault diagnosis" in a physical offline task, the system will match corresponding virtual simulation exercises of circuit principles and online fault logic judgment tests. This leverages the zero-cost, reproducible, and highly interactive characteristics of the virtual online environment to specifically reinforce the target object's theoretical foundation and logical connections, clearing cognitive obstacles for secondary practical exercises.

[0223] S804. Insert virtual online tasks that match the weak skills into the task scheduling information before the skill training tasks that have not yet been executed.

[0224] In the embodiments of this application, the insertion operation refers to real-time priority intervention on the task queue while maintaining the consistency of the existing task scheduling logic.

[0225] In some embodiments of this application, after detecting a weak skill, the task scheduling system automatically generates an immediate interruption command in the current task execution chain, setting the matching virtual online task as the current mandatory prerequisite. Only after the target object completes the virtual online task and passes the evaluation will the system unlock the subsequent skill training tasks in the original task scheduling information. This strategy achieves a closed-loop feedback of practical problem discovery - virtual reinforcement of cognition - subsequent continuous verification, avoiding the accumulation of skill deficiencies in subsequent high-level tasks.

[0226] As can be seen, this embodiment of the application achieves reverse feedback adjustment from offline practice to online reinforcement by collecting physical operation logs and identifying skill weaknesses. This solution utilizes the low-cost replayability of virtual tasks to promptly address specific shortcomings exposed in physical operations, ensuring that the task scheduling scheme can be dynamically adjusted based on the actual hands-on performance of the target, thus solving the problem of disconnect between virtual and real-world training arrangements.

[0227] In some embodiments of this application, after determining the task arrangement information of the target object on the skills training platform based on the target task sequence, the method further includes:

[0228] S901. Monitor the task status feedback signal of the target object in completing any skill training task in the task arrangement information.

[0229] In the embodiments of this application, the task status feedback signal refers to the communication message triggered by the front-end component of the skills training platform or the IoT terminal when it detects a task completion event. This signal includes the specific timestamp of task completion, metadata of the execution result (such as score and error code), and the user's real-time behavioral characteristics. The monitoring process is implemented by establishing a long connection or a message queue subscription mechanism.

[0230] S902. Using an online learning algorithm, incrementally update the feature weights in the user profile data based on the task status feedback signal to generate updated user profile data and the corresponding profile version identifier.

[0231] In the embodiments of this application, the online learning algorithm refers to a machine learning algorithm that can dynamically adjust model parameters based on new samples without retraining the entire dataset.

[0232] In some embodiments of this application, the online learning algorithm may employ the FTRL algorithm. The specific logic of step S902 is as follows: the system extracts only the deviation features (e.g., performance of a skill is far below expectations) from the task status feedback signal. The deviation features are extracted by comparing the actual completion score, actual time consumption, and other result data contained in the task status feedback signal with the expected performance value of the corresponding skill dimension in the user profile data. When the deviation between the actual value and the expected value exceeds a preset deviation detection threshold, the skill dimension is marked as a deviation feature dimension, and the direction (positive improvement or negative regression) and magnitude of the deviation are extracted as input features for the FTRL algorithm. Utilizing the sparsity characteristic of the FTRL algorithm, the gradient of the instantaneous loss function generated by the deviation feature is calculated (the instantaneous loss function uses a logarithmic loss function or a squared loss function, with the magnitude of the deviation feature as the prediction error input), and combined with L1 regularization to generate a sparse penalty term. Gradient descent is only applied to the weights of the skill dimensions related to the deviation feature, while keeping the weights of other irrelevant dimensions unchanged. Simultaneously, the system generates a unique profile version identifier (such as a hash value or an incrementing sequence number) for each updated profile data, used to identify the snapshot version of the user's capability status. This step avoids the high computational cost of retraining the entire model, keeping the profile update time within a very low range.

[0233] S903. Compare the updated user profile data with the pre-stored baseline profile version identifier to determine whether to trigger a replanning instruction.

[0234] In the embodiments of this application, the pre-stored baseline profile version identifier refers to the version identifier of the user profile data used during the last full path planning execution. The comparison process aims to establish a computational filtering barrier.

[0235] In some embodiments of this application, the system not only compares whether the version number has changed, but also calculates the rate of change of the Euclidean distance between the feature weights before and after the update. The system only generates a replanning instruction when the version number changes and the change in feature weights exceeds a preset threshold (i.e., the user's capabilities have undergone a significant change, rendering the original plan inapplicable); otherwise, the system ignores minor capability fluctuations and continues to maintain the original task orchestration information. This mechanism effectively prevents system jitter and wasted computing power due to oversensitivity.

[0236] S904. If a replanning instruction is triggered, the executed task nodes in the task orchestration information are locked, and the local paths of the skill training tasks that have not yet been executed in the task orchestration information are replanned.

[0237] In the embodiments of this application, local path replanning refers to recalculating only the task sequence within a future time window while keeping the historical execution record unchanged.

[0238] In some embodiments of this application, the system uses the updated user profile data as the new input state and restarts the genetic algorithm and constraint satisfaction problem solver. However, in this operation, the completed tasks are fixed as immutable hard constraints at the head of the chromosome sequence, and the algorithm only performs mutation and optimization on subsequent gene positions. The generated new task sequence seamlessly replaces the remaining parts of the original scheduling scheme and is pushed to the user terminal via real-time communication technologies such as WebSocket, realizing dynamic refresh of the user interface.

[0239] As can be seen, this application's embodiments construct a low-latency, high-efficiency path dynamic replanning system by introducing the FTRL incremental learning algorithm and the profile version control mechanism. This scheme solves the problems of slow response and high resource consumption in full computation, and achieves adaptive adjustment of the training path to the user's state by triggering local reordering only when the user's capabilities change significantly.

[0240] Secondly, embodiments of this application provide a task scheduling system applied to a skills training platform, which is used to execute the task scheduling method applied to a skills training platform as described in any of the above embodiments.

[0241] Thirdly, embodiments of this application provide an electronic device. This electronic device may include a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to run any of the task scheduling systems provided in the embodiments of this application for a skills training platform, implementing the task scheduling method for a skills training platform as described in any of the above embodiments.

[0242] Fourthly, embodiments of this application provide an electronic device for running any of the task scheduling systems provided in this application for a skills training platform. For example... Figure 2 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0243] The electronic device includes a Central Processing Unit (CPU) 201, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 202 or a program loaded from storage portion 208 into Random Access Memory (RAM) 203, such as performing the methods described in the above embodiments. The RAM 203 also stores various programs and data required for system operation. The CPU 201, ROM 202, and RAM 203 are interconnected via a bus 204. An Input / Output (I / O) interface 205 is also connected to the bus 204.

[0244] The following components are connected to I / O interface 205: input section 206 including audio input devices, push-button switches, etc.; output section 207 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 208 including a hard disk, etc.; and communication section 209 including a network interface card such as a local area network (LAN) card, modem, etc. Communication section 209 performs communication processing via a network such as the Internet. Drive 210 is also connected to I / O interface 205 as needed. Removable media 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 210 as needed so that computer programs read from them can be installed into storage section 208 as needed.

[0245] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by CPU 201, it performs the various functions defined in this application.

[0246] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0247] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0248] Specifically, the electronic device of this embodiment includes a processor and a memory. The memory is coupled to one or more processors and is used to store computer program code. The computer program code includes computer instructions. One or more processors call the computer instructions to cause the electronic device to perform the method provided in the above embodiment.

[0249] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program configured to be executed by a processor to implement the task scheduling method applied to a skills training platform as described in any of the above embodiments.

[0250] Sixthly, embodiments of this application provide a computer program product, including a computer program or instructions, which are executed by a processor to implement the task scheduling method applied to a skills training platform as described in any of the preceding claims.

[0251] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A task scheduling method applied to a skills training platform, characterized in that, The task scheduling method applied to the skills training platform includes: Obtain the first task completion data of the target object's virtual online task from the skills training platform; Based on the first task completion data and the second task completion data of the target object's offline tasks, user profile data of the target object is generated. The task scheduling constraints are determined, including the skill dependencies in the skill knowledge graph of the skill training platform, the task ratio between virtual online tasks and physical offline tasks, and the total duration of task objectives. Under the constraints of the task arrangement conditions, using the user profile data, the target task sequence matching the target object is determined from the skill training task library of the skill training platform. The target task sequence includes target virtual online tasks and target physical offline tasks. Based on the target task sequence, the task arrangement information of the target object on the skills training platform is determined, and the task arrangement information is used for task scheduling processing.

2. The task scheduling method applied to a skills training platform as described in claim 1, characterized in that, The step of generating user profile data for the target object based on the first task completion data and the second task completion data of the target object's offline tasks includes: The browsing duration parameter and interaction round parameter in the first task completion data, and the completion quality score parameter and collaboration evaluation parameter in the second task completion data are vectorized to obtain a multi-dimensional feature vector. The multidimensional feature vector is input into a pre-trained semantic analysis model to extract the skill semantic features of the target object; Based on the aforementioned skill semantic features, the target object's current skill proficiency level, skill deficiencies, and skill preferences are determined. Based on the current skill proficiency level, the skill deficiencies, and the skill preferences, user profile data for the target object is generated.

3. The task scheduling method applied to a skills training platform as described in claim 2, characterized in that, Under the constraints of the task arrangement conditions, the step of determining the target task sequence matching the target object from the skill training task library of the skill training platform using the user profile data includes: Under the constraints of the task arrangement conditions, multiple skill training tasks in the skill training task library are encoded to generate a chromosome sequence population including multiple initial task chromosome sequences. For each of the initial task chromosome sequences, determine the expected task matching degree between the initial task chromosome sequence and the user profile data; Based on the expected matching degree of the task, determine the function value of the preset fitness function; Under the constraints of the task arrangement conditions, the chromosome sequence population is subjected to selection, crossover and mutation operations using the function value of the preset fitness function to obtain a converged chromosome sequence population. Based on the converged chromosome sequence population, the target task sequence matching the target object is determined.

4. The task scheduling method applied to a skills training platform as described in claim 3, characterized in that, Determining the expected task match between the initial task chromosome sequence and the user profile data includes: Based on the task difficulty coefficient and task type of each skill training task in the initial task chromosome sequence, generate a task difficulty coefficient sequence and a task type sequence; Based on the task difficulty coefficient sequence and the current skill proficiency level, determine the ability matching deviation value; Based on the task type sequence and the skill missing items, a task type matching index is determined; The expected matching degree of the task is determined by using the capability matching deviation value and the task type matching index.

5. The task scheduling method applied to a skills training platform as described in claim 3, characterized in that, The step of determining the function value of the preset fitness function based on the expected matching degree of the task includes: Based on the initial task chromosome sequence and the user profile data, the expected skill improvement of the target object is determined; The total expected duration of the task is determined by the initial task chromosome sequence; The function value of the preset fitness function is determined based on the sum of the expected task matching degree, the expected skill improvement degree, and the expected task duration.

6. The task scheduling method applied to a skills training platform as described in claim 5, characterized in that, The step of determining the expected skill improvement of the target object based on the initial task chromosome sequence and the user profile data includes: Based on the task type, the missing skill item, and the skill preference item of each skill training task in the initial task chromosome sequence, determine the expected skill improvement coefficient of each skill training task in the initial task chromosome sequence. Obtain the skill gain contribution value for each skill training task in the initial task chromosome sequence; The expected skill improvement is determined based on the skill gain contribution value and the expected skill improvement coefficient.

7. The task scheduling method applied to a skills training platform as described in claim 1, characterized in that, The step of determining the target object based on the target task sequence, after the task arrangement information of the skills training platform, further includes: Obtain the first interaction log generated by the target object executing the target virtual online task; Identify the skill confusion keywords in the first interaction log; In the skills training task library, identify the offline tasks that match the keywords causing confusion in the skills. The offline tasks matching the keywords causing the skill confusion are inserted before any unexecuted skill training tasks in the task scheduling information.

8. The task scheduling method applied to a skills training platform as described in claim 1, characterized in that, The step of determining the target object based on the target task sequence, after the task arrangement information of the skills training platform, further includes: Obtain the second interaction log generated by the target object executing the offline task of the target entity; Based on the second interaction log, the target object's weak skills are determined; In the skills training task library, determine the virtual online task that matches the weak skill; The virtual online task matching the weak skill is inserted before the skill training task that has not yet been executed in the task scheduling information.

9. A task scheduling system applied to a skills training platform, characterized in that, The task scheduling system applied to the skills training platform is used to execute the task scheduling method applied to the skills training platform as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program configured to be executed by a processor to implement the task scheduling method for a skills training platform as described in any one of claims 1 to 8.

Images