A Smart Chemical Industry Planning and Consulting Method and System Based on AI Big Data Analysis

Abstract

This invention discloses an intelligent academic planning consultation method and system based on AI big data analysis, comprising the following steps: collecting student academic planning data and external educational constraint data; grouping and normalizing students by continuous learning cycles to generate student state feature vectors; writing the feature vectors into the dandelion optimization continuous search space to generate a set of potential planning variable vectors; performing constraint decoding to generate a set of candidate academic planning paths; constructing counterfactual candidate paths and calculating counterfactual regret values; correcting the dandelion seed positions based on the counterfactual regret values; re-decoding and filtering target academic planning paths to generate personalized academic planning consultation solutions. This invention improves the matching degree and executability of academic planning paths and enhances the dynamic optimization capability of consultation solutions.

Description

Technical Field

[0001] This invention relates to the fields of educational informatization and artificial intelligence data analysis technology, and in particular to an intelligent academic planning and consulting method and system based on AI big data analysis. Background Technology

[0002] Academic planning consultation primarily addresses students' choices of subjects, majors, university levels, learning commitment, and college application goals. Current academic planning methods often rely on analyzing grades, rankings, interest assessment results, university admission rankings, and major selection requirements, with consultants or recommendation programs providing suggestions on subject selection, majors, and universities. Some platforms can collect student learning data, create student profiles, and generate academic planning results based on rule-based matching.

[0003] Existing methods, when generating planning results, often directly map student status data to majors, institutions, or learning suggestions, lacking continuous search processing between student status characteristics and educational constraints. Changes in student grades, rankings, learning behavior stability, ability assessment results, interests, and subject selection intentions change over continuous learning cycles; single profiles or static rules are insufficient to reflect these phased changes in student status. Major selection, institution level, and learning resource allocation interact in actual planning, and direct matching methods easily lead to problems such as subject selection not meeting major requirements, mismatches between institution level and ranking, and learning input exceeding affordability.

[0004] Existing recommendation methods often rank candidate options based on single evaluation results, lacking the processing of counterfactual substitution and difference comparison of candidate academic planning paths. Changes in constraint satisfaction, goal attainability, and learning input matching between the current path and alternative paths are difficult to quantify and utilize. Updates to planning schemes still tend to be more about re-recommendation, lacking a correction process targeting specific sources of deviation in the original planning path. When students experience fluctuations in grades, changes in interests, adjustments to target institutions, or changes in learning input during subsequent learning cycles, existing methods struggle to make targeted modifications to the original planning path based on the advantages of alternative paths.

[0005] Existing optimization algorithms for academic planning often directly search for specific candidate solutions without converting student state feature vectors into potential planning variable vectors. They also fail to use constraint decoding to map these potential planning variables into candidate academic planning paths that meet the requirements of major selection, university admission ranking, and learning input constraints. Ordinary optimization searches are prone to producing inconsistent results between discrete major directions, university levels, and continuous learning resource allocation, lacking effective correlation between search direction and educational constraints, student ability gaps, and the advantages of alternative paths. Therefore, providing an intelligent academic planning consultation method and system based on AI big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent academic planning and consulting method and system based on AI big data analysis. This invention utilizes the dandelion optimization algorithm, constraint decoding, and counterfactual regret correction to search, correct, and filter planning paths for student academic data and educational constraint data. It has the advantages of high planning matching degree, strong constraint adaptability, and good dynamic adjustment capability.

[0007] According to an embodiment of the present invention, an intelligent chemical industry planning and consulting method and system based on AI big data analysis includes the following steps: Collect student academic planning data and external educational constraints data; The student academic planning data is grouped according to continuous learning cycles, and then coded and normalized to generate student state feature vectors. The student state feature vector is written into the continuous search space of the dandelion optimization algorithm, the variable boundary is determined according to the normalized value, the dandelion seed position is initialized, and a set of potential planning variable vectors is generated. Perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; Select the current candidate academic planning path from the set of candidate academic planning paths, perform single field preservation and adjacent field replacement to generate counterfactual candidate paths, and generate counterfactual regret values ​​based on the difference between the current candidate academic planning path and the counterfactual candidate path in terms of constraint satisfaction, goal attainability and learning input matching degree. When the evaluation value of the counterfactual candidate path exceeds the evaluation value of the current candidate academic planning path, the direction of the potential planning variable vector corresponding to the counterfactual candidate path relative to the current potential planning variable vector is used as the correction direction, and the counterfactual regret value is used as the correction step size weight to update the dandelion seed position and generate an updated set of potential planning variable vectors. The updated set of potential planning variable vectors is re-executed with constraint decoding to generate an updated set of candidate academic planning paths. The target academic planning path is determined through feasibility assessment and robustness screening, and a personalized academic planning consultation plan is generated based on the target academic planning path.

[0008] Optionally, the student academic planning data and external educational constraint data include: Student academic planning data includes academic performance data for each subject, exam ranking data, learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance examination goal data; External educational constraint data includes data on majors, subject selection requirements, university admission rankings, university level, and maximum learning input.

[0009] Optionally, the operation of generating the student state feature vector specifically includes: The student academic planning data is aligned and grouped according to the student identifier and the consecutive learning cycle number to generate learning cycle data groups; Within each learning cycle data group, extract the changes in grades and rankings, and encode the fields for learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance goal data. Normalize the changes in grades, changes in ranking, and field encoding results, and concatenate them according to a fixed field order to generate a student status feature vector.

[0010] Optionally, the operation of generating the set of potential planning variable vectors specifically includes: Read the student status feature vector and build a variable dimension sequence table according to the field order; For each variable dimension in the variable dimension sequence list, extract the normalized field values ​​of all learning cycle data groups under the same variable dimension, sort them in ascending order, and take the first value as the initial lower bound of the variable and the last value as the initial upper bound of the variable. The constraint values ​​corresponding to the variable dimensions in the external education constraint data are converted into normalized constraint intervals, and the intersection of these intervals with the initial lower and upper bounds of the variables is processed to obtain the lower and upper bounds of the variables. Combine the lower and upper bounds of variables according to the variable dimension sequence list to form the variable boundaries of the continuous search space; Within the variable boundaries, dandelion seed position values ​​are generated, and all variable dimension values ​​under the same dandelion seed position are concatenated to form a latent planning variable vector, resulting in a set of latent planning variable vectors.

[0011] Optionally, the operation of generating a set of candidate academic planning paths specifically includes: Read the vector of potential planning variables and divide it into professional direction decoding dimension, institutional level decoding dimension and learning resource decoding dimension according to the variable dimension sequence list; The distance between the variable values ​​of the professional direction decoding dimension and the normalized code of the professional direction in the professional data is calculated one by one. The professional subject selection requirements are read according to the distance and the professional subject selection requirements are checked against the subject selection intention field one by one. Candidate subject selection combinations and candidate professional directions are retained. The distance between the variable values ​​of the institution level decoding dimension and the institution level normalized code in the institution level data is calculated one by one. The institution admission ranking data is read according to the distance. The institution admission ranking data and the ranking ranking change field are used to calculate the ranking difference. Institution levels whose ranking difference does not fall into the admission ranking range are deleted, and candidate institution levels are retained. The variable values ​​of the learning resource decoding dimension are converted into the proportion of learning time for each subject, and the learning time allocation results are formed based on the performance change field and the upper limit data of learning input. The candidate academic planning paths are generated by combining candidate majors, candidate subject combinations, candidate university levels, and study time allocation results.

[0012] Optionally, the operation of generating counterfactual candidate paths specifically includes: Retrieve the candidate majors, candidate university levels, and study time allocation results from the current candidate academic planning paths; Keeping the candidate major direction unchanged, the candidate institution level is replaced with the institution level whose level sorting number differs by one number, generating the institution level counterfactual candidate path; Keeping the candidate institution level unchanged, select adjacent professional directions under the same discipline category identifier, check the subject selection requirements of the adjacent professional directions against the subject selection intention field item by item, and generate counterfactual candidate paths for professional directions. Keeping the candidate major and candidate institution level unchanged, the proportion of study time for each subject is adjusted from small to large according to the field of grade change, and the total proportion of study time is limited to the upper limit of study input data to generate counterfactual candidate paths for learning strategies. The counterfactual candidate paths are composed of counterfactual candidate paths at the institutional level, counterfactual candidate paths of majors, and counterfactual candidate paths of learning strategies.

[0013] Optionally, the operation of generating counterfactual regret values ​​specifically includes: Read the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path for the major direction, and the counterfactual candidate path for the learning strategy; The following steps are performed on the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level: verify the subject selection requirements, calculate the difference in the admission ranking of the institution, and compare the upper limit of learning input. The results are used to generate the evaluation values ​​for the current path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level. The counterfactual path evaluation values ​​for institutional level, major direction, and learning strategy are calculated by comparing them with the current path evaluation value to generate the counterfactual difference values ​​for institutional level, major direction, and learning strategy. The counterfactual discrepancies of the institution level, major direction, and learning strategy are sorted in descending order. The counterfactual discrepancy at the top of the list is taken as the counterfactual regret value, and the counterfactual candidate path corresponding to the counterfactual discrepancy at the top of the list is taken as the target counterfactual candidate path. The constraint violation degree is generated based on the number of failed subjects in the professional subject selection requirements, the number of excesses in the calculation of the difference between the university admission ranking and the value of excess input in the comparison of the upper limit of learning input. The ability gap is generated based on the difference between the phased achievement goals and the achievement change field, the difference between the admission rank and the ranking rank change field corresponding to the candidate institution level, and the difference between the study time allocation results and the upper limit data of study input.

[0014] Optionally, the operation of generating the updated set of potential planning variable vectors specifically includes: Read the target counterfactual candidate path corresponding to the counterfactual regret value from the counterfactual candidate path, and map the candidate major direction, candidate institution level and learning time allocation results in the target counterfactual candidate path back to the major direction decoding dimension, institution level decoding dimension and learning resource decoding dimension, respectively, to obtain the target potential planning variable vector; Read the current potential planning variable vector corresponding to the current candidate academic planning path, and subtract the target potential planning variable vector from the current potential planning variable vector digit by digit according to the variable dimension sequence list to obtain the counterfactual correction direction; The counterfactual regret value is normalized to obtain the counterfactual correction step size; In the seed position update stage of the dandelion optimization algorithm, the update direction of the dandelion seed position is set as the counterfactual correction direction, and the dandelion seed position is moved along the counterfactual correction direction by the counterfactual correction step size to obtain the counterfactual corrected dandelion seed position. The constraint violation degree is normalized to obtain the constraint scaling factor, the capability gap degree is normalized to obtain the capability scaling factor, the constraint scaling factor is applied to the professional direction decoding dimension and the institutional level decoding dimension, the capability scaling factor is applied to the learning resource decoding dimension, and the dandelion seed position after counterfactual correction is scaled by dimension step. The positions of the dandelion seeds after the step-by-step scaling of each dimension are compared with the lower and upper bounds of the variables. Variable values ​​below the lower bound are replaced with the lower bound, and variable values ​​above the upper bound are replaced with the upper bound, thus obtaining the dandelion seed positions after boundary clipping. The positions of the dandelion seeds after boundary clipping are used as the updated latent planning variable vectors, and all the updated latent planning variable vectors are combined to form the updated latent planning variable vector set.

[0015] Optionally, the operation of determining the target academic planning path specifically includes: Read the updated set of candidate academic planning paths; For the updated candidate academic planning paths, the candidate subject combinations, candidate institution levels, and study time allocation results are checked against the professional subject selection requirements, compared with the institution admission ranking range, and compared with the upper limit of study input, respectively. Updated candidate academic planning paths that fail to pass the check or comparison are deleted. Add perturbations for admission ranking, grade change, interest tendency, and learning engagement to the retained updated candidate academic planning paths; After the disturbance, the updated candidate academic planning paths are re-verified for major selection requirements, compared with the university admission ranking range, and compared with the upper limit of learning input. The updated candidate academic planning paths that still pass the verification and comparison after the disturbance are ranked according to the path evaluation value, and the updated candidate academic planning path ranked first is determined as the target academic planning path.

[0016] An intelligent academic planning and consulting system based on AI big data analysis according to an embodiment of the present invention includes the following modules: The data acquisition module is used to collect student academic planning data and external educational constraints data; The feature vector generation module is used to group, encode, and normalize student academic planning data to generate student status feature vectors. The latent variable generation module is used to write the student state feature vector into the continuous search space of the Dandelion optimization algorithm to generate a set of latent planning variable vectors. The constraint decoding module is used to perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; The counterfactual path and regret value calculation module is used to generate counterfactual candidate paths and calculate counterfactual regret values. The seed location update module is used to update the dandelion seed locations based on the counterfactual regret value, generating an updated set of potential planning variable vectors. The path selection and solution generation module is used to re-execute constraint decoding, perform feasibility judgment and robustness selection, determine the target academic planning path and generate personalized academic planning consultation solutions.

[0017] The beneficial effects of this invention are: (1) This invention converts student academic planning data into student state feature vectors and writes them into the continuous search space of the Dandelion optimization algorithm to generate a set of potential planning variable vectors. This makes academic planning no longer directly dependent on single grades, rankings or static profiles for matching. It can search and process changes in student state in the continuous variable space, thereby improving the generation range and matching accuracy of candidate academic planning paths.

[0018] (2) This invention converts the potential planning variable vector into candidate academic planning paths through constraint decoding, so that the candidate major direction, candidate subject combination, candidate school level and learning time allocation results are restricted by the major subject requirements, school admission ranking data and learning input upper limit data during the generation stage, reducing the problems of subject selection not meeting major requirements, school level and ranking mismatch, and learning input exceeding the upper limit, and improving the feasibility of academic planning schemes.

[0019] (3) The present invention constructs counterfactual candidate paths and generates counterfactual regret values ​​based on the evaluation difference between the current candidate academic planning path and the counterfactual candidate paths. The counterfactual regret values ​​are used to correct the dandelion seed position, so that the search direction of the planning path can be adjusted according to the advantages of the alternative path, thereby improving the constraint satisfaction, target attainability and learning input matching of the target academic planning path. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows an intelligent chemical industry planning and consulting method and system based on AI big data analysis proposed in this invention. Figure 2 This is a flowchart illustrating the potential planning variable vector generation process of an intelligent chemical industry planning and consulting method and system based on AI big data analysis proposed in this invention. Figure 3 This is a flowchart illustrating the counterfactual regret correction process of an intelligent chemical industry planning and consulting method and system based on AI big data analysis proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A smart chemical industry planning and consulting method and system based on AI big data analysis includes the following steps: Collect student academic planning data and external educational constraints data; The student academic planning data is grouped according to continuous learning cycles, and then coded and normalized to generate student state feature vectors. The student state feature vector is written into the continuous search space of the dandelion optimization algorithm, the variable boundary is determined according to the normalized value, the dandelion seed position is initialized, and a set of potential planning variable vectors is generated. Perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; Select the current candidate academic planning path from the set of candidate academic planning paths, perform single field preservation and adjacent field replacement to generate counterfactual candidate paths, and generate counterfactual regret values ​​based on the difference between the current candidate academic planning path and the counterfactual candidate path in terms of constraint satisfaction, goal attainability and learning input matching degree. When the evaluation value of the counterfactual candidate path exceeds the evaluation value of the current candidate academic planning path, the direction of the potential planning variable vector corresponding to the counterfactual candidate path relative to the current potential planning variable vector is used as the correction direction, and the counterfactual regret value is used as the correction step size weight to update the dandelion seed position and generate an updated set of potential planning variable vectors. The updated set of potential planning variable vectors is re-executed with constraint decoding to generate an updated set of candidate academic planning paths. The target academic planning path is determined through feasibility assessment and robustness screening, and a personalized academic planning consultation plan is generated based on the target academic planning path.

[0023] In practice, student academic planning data is collected according to student identifiers, while external educational constraint data is collected according to major, institution level, and learning input limitations. After the student state feature vector is generated, it does not directly output planning suggestions but serves as the continuous search input for the Dandelion optimization algorithm. The dandelion seed positions in the continuous search space represent potential planning variables, not specific majors, institutions, or learning schemes. Each dandelion seed position is converted into a candidate academic planning path after constraint decoding. These candidate paths are then subjected to counterfactual substitution, counterfactual regret value calculation, and dandelion seed position correction to form an updated set of potential planning variable vectors. The updated set of potential planning variable vectors is then re-decoded to generate an updated set of candidate academic planning paths. After feasibility assessment and robustness screening, the target academic planning path is output.

[0024] In this implementation, student academic planning data and external educational constraint data include: Student academic planning data includes academic performance data for each subject, exam ranking data, learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance examination goal data; External educational constraint data includes data on majors, subject selection requirements, university admission rankings, university level, and maximum learning input.

[0025] The student academic planning data includes subject-specific performance data to represent changes in student grades across different subjects within a continuous learning cycle; exam ranking data to represent changes in a student's position within their class or grade; learning behavior data to represent study time, task completion rate, and number of incorrect answers; ability assessment data to represent assessment scores for different ability items; interest tendency data to represent areas of interest and intensity of interest; subject selection intention data to represent a student's current subject selection combination; and college application goal data to represent the target major and target university level. The external educational constraint data includes major selection requirements to limit candidate subject combinations, university admission ranking data to limit candidate university levels, and upper limit data on learning input to limit the allocation of study time.

[0026] In this embodiment, the operation of generating student state feature vectors specifically includes: The student academic planning data is aligned and grouped according to the student identifier and the consecutive learning cycle number to generate learning cycle data groups; Within each learning cycle data group, extract the changes in grades and rankings, and encode the fields for learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance goal data. Normalize the changes in grades, changes in ranking, and field encoding results, and concatenate them according to a fixed field order to generate a student status feature vector.

[0027] When generating student status feature vectors, the academic planning data of all students for the same student is first located according to their student identifiers. Then, the data is divided into multiple learning cycle data groups according to consecutive learning cycle numbers. Within each learning cycle data group, the grade data is arranged by exam time, and the grade change value between adjacent exams is calculated. The exam ranking data is arranged by exam time, and the ranking change value between adjacent exams is calculated. Learning behavior data is summarized according to learning cycles to form learning behavior field codes. Ability assessment data, interest data, subject selection intention data, and college entrance goal data are converted into numerical fields. After normalization, all fields are concatenated in a fixed field order to form the student status feature vector, which serves as the input for subsequent continuous search space.

[0028] In this embodiment, the operation of generating the set of potential planning variable vectors specifically includes: Read the student status feature vector and build a variable dimension sequence table according to the field order; For each variable dimension in the variable dimension sequence list, extract the normalized field values ​​of all learning cycle data groups under the same variable dimension, sort them in ascending order, and take the first value as the initial lower bound of the variable and the last value as the initial upper bound of the variable. The constraint values ​​corresponding to the variable dimensions in the external education constraint data are converted into normalized constraint intervals, and the intersection of these intervals with the initial lower and upper bounds of the variables is processed to obtain the lower and upper bounds of the variables. Combine the lower and upper bounds of variables according to the variable dimension sequence list to form the variable boundaries of the continuous search space; Within the variable boundaries, dandelion seed position values ​​are generated, and all variable dimension values ​​under the same dandelion seed position are concatenated to form a latent planning variable vector, resulting in a set of latent planning variable vectors.

[0029] When generating the set of potential programming variable vectors, each field in the student state feature vector corresponds to a variable dimension in the variable dimension sequence list. For the same variable dimension, the system reads the normalized field values ​​from all learning cycle data sets, sorts the normalized field values ​​in ascending order, and uses the first value as the initial lower bound and the last value as the initial upper bound. The values ​​corresponding to the variable dimensions in the external educational constraint data are normalized into constraint intervals. The intersection of the initial lower bound, initial upper bound, and normalized constraint intervals yields the final lower and upper bounds. The lower and upper bounds form the variable boundaries in the variable dimension sequence list, and the dandelion seed positions are generated within these boundaries. The variable dimension values ​​at each dandelion seed position are sequentially concatenated to form a potential programming variable vector. Multiple potential programming variable vectors constitute the set of potential programming variable vectors.

[0030] In this embodiment, the operation of generating a set of candidate academic planning paths specifically includes: Read the vector of potential planning variables and divide it into professional direction decoding dimension, institutional level decoding dimension and learning resource decoding dimension according to the variable dimension sequence list; The distance between the variable values ​​of the professional direction decoding dimension and the normalized code of the professional direction in the professional data is calculated one by one. The professional subject selection requirements are read according to the distance and the professional subject selection requirements are checked against the subject selection intention field one by one. Candidate subject selection combinations and candidate professional directions are retained. The distance between the variable values ​​of the institution level decoding dimension and the institution level normalized code in the institution level data is calculated one by one. The institution admission ranking data is read according to the distance. The institution admission ranking data and the ranking ranking change field are used to calculate the ranking difference. Institution levels whose ranking difference does not fall into the admission ranking range are deleted, and candidate institution levels are retained. The variable values ​​of the learning resource decoding dimension are converted into the proportion of learning time for each subject, and the learning time allocation results are formed based on the performance change field and the upper limit data of learning input. The candidate academic planning paths are generated by combining candidate majors, candidate subject combinations, candidate university levels, and study time allocation results.

[0031] When generating a set of candidate academic planning paths, the system separates the major direction decoding dimension, institution level decoding dimension, and learning resource decoding dimension from the potential planning variable vector. For the major direction decoding dimension, the distance between the variable values ​​and the normalized codes of major directions in the major data is calculated one by one. Major directions with higher distance rankings are then checked against the subject selection requirements. After checking each subject selection requirement against the subject selection intention field, candidate subject combinations and candidate major directions that meet the requirements are retained. For the institution level decoding dimension, the distance between the variable values ​​and the normalized codes of the institution level is calculated one by one. Combining the institution admission ranking data and the ranking change field, a ranking difference is calculated. Institution levels whose ranking differences fall within the admission ranking range are retained as candidate institution levels. For the learning resource decoding dimension, the variable values ​​are converted into the proportion of study time for each subject, and a study time allocation result is formed based on the grade change field and the upper limit data of study input. The candidate major directions, candidate subject combinations, candidate institution levels, and study time allocation results are combined to form the candidate academic planning path.

[0032] In this embodiment, the operation of generating counterfactual candidate paths specifically includes: Retrieve the candidate majors, candidate university levels, and study time allocation results from the current candidate academic planning paths; Keeping the candidate major direction unchanged, the candidate institution level is replaced with the institution level whose level sorting number differs by one number, generating the institution level counterfactual candidate path; Keeping the candidate institution level unchanged, select adjacent professional directions under the same discipline category identifier, check the subject selection requirements of the adjacent professional directions against the subject selection intention field item by item, and generate counterfactual candidate paths for professional directions. Keeping the candidate major and candidate institution level unchanged, the proportion of study time for each subject is adjusted from small to large according to the field of grade change, and the total proportion of study time is limited to the upper limit of study input data to generate counterfactual candidate paths for learning strategies. The counterfactual candidate paths are composed of counterfactual candidate paths at the institutional level, counterfactual candidate paths of majors, and counterfactual candidate paths of learning strategies.

[0033] When generating counterfactual candidate paths, the system reads the candidate major direction, candidate institution level, and study time allocation results from the current candidate academic planning paths, and performs three types of replacement operations. The institution level counterfactual candidate path keeps the candidate major direction unchanged, only replacing the candidate institution level with an institution level whose level sorting number differs by one number. The major direction counterfactual candidate path keeps the candidate institution level unchanged, selects adjacent major directions under the same discipline category identifier, and re-verifies the subject selection requirements and subject selection intention fields for adjacent major directions. The study strategy counterfactual candidate path keeps the candidate major direction and candidate institution level unchanged, adjusting the study time proportion of each subject according to the grade change field from small to large, and limiting the total adjusted study time proportion to within the upper limit of study input data. These three types of counterfactual candidate paths are used for subsequent counterfactual regret value calculation.

[0034] In this embodiment, the operation of generating counterfactual regret value specifically includes: Read the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path for the major direction, and the counterfactual candidate path for the learning strategy; The following steps are performed on the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level: verify the subject selection requirements, calculate the difference in the admission ranking of the institution, and compare the upper limit of learning input. The results are used to generate the evaluation values ​​for the current path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level. The counterfactual path evaluation values ​​for institutional level, major direction, and learning strategy are calculated by comparing them with the current path evaluation value to generate the counterfactual difference values ​​for institutional level, major direction, and learning strategy. The counterfactual discrepancies of the institution level, major direction, and learning strategy are sorted in descending order. The counterfactual discrepancy at the top of the list is taken as the counterfactual regret value, and the counterfactual candidate path corresponding to the counterfactual discrepancy at the top of the list is taken as the target counterfactual candidate path. The constraint violation degree is generated based on the number of failed subjects in the professional subject selection requirements, the number of excesses in the calculation of the difference between the university admission ranking and the value of excess input in the comparison of the upper limit of learning input. The ability gap is generated based on the difference between the phased achievement goals and the achievement change field, the difference between the admission rank and the ranking rank change field corresponding to the candidate institution level, and the difference between the study time allocation results and the upper limit data of study input.

[0035] When generating counterfactual regret values, the system performs the same evaluation operation on the current candidate academic planning path and the three types of counterfactual candidate paths. The subject selection requirement verification is used to obtain the subject selection constraint evaluation result; the university admission ranking difference calculation is used to obtain the university level reachability evaluation result; and the learning input upper limit comparison is used to obtain the learning input matching evaluation result. After normalization and summarization of the three evaluation results, the current path evaluation value, university level counterfactual path evaluation value, major direction counterfactual path evaluation value, and learning strategy counterfactual path evaluation value are obtained respectively. Subtracting the current path evaluation value from each of the three types of counterfactual path evaluation values ​​yields three types of counterfactual differences. After sorting the three types of counterfactual differences in descending order, the counterfactual difference at the top of the list is used as the counterfactual regret value, and the counterfactual candidate path corresponding to the top-ranked difference is used as the target counterfactual candidate path. The constraint violation degree is generated by the number of failed subjects, the number of times the ranking exceeded the limit, and the amount of input exceeded. The ability gap degree is generated by the difference between the stage-based performance target, the performance change field, the ranking change field, the admission ranking corresponding to the candidate university level, the learning time allocation result, and the learning input upper limit data.

[0036] In this embodiment, the operation of generating the updated set of potential planning variable vectors specifically includes: Read the target counterfactual candidate path corresponding to the counterfactual regret value from the counterfactual candidate path, and map the candidate major direction, candidate institution level and learning time allocation results in the target counterfactual candidate path back to the major direction decoding dimension, institution level decoding dimension and learning resource decoding dimension, respectively, to obtain the target potential planning variable vector; Read the current potential planning variable vector corresponding to the current candidate academic planning path, and subtract the target potential planning variable vector from the current potential planning variable vector digit by digit according to the variable dimension sequence list to obtain the counterfactual correction direction; The counterfactual regret value is normalized to obtain the counterfactual correction step size; In the seed position update stage of the dandelion optimization algorithm, the update direction of the dandelion seed position is set as the counterfactual correction direction, and the dandelion seed position is moved along the counterfactual correction direction by the counterfactual correction step size to obtain the counterfactual corrected dandelion seed position. The constraint violation degree is normalized to obtain the constraint scaling factor, the capability gap degree is normalized to obtain the capability scaling factor, the constraint scaling factor is applied to the professional direction decoding dimension and the institutional level decoding dimension, the capability scaling factor is applied to the learning resource decoding dimension, and the dandelion seed position after counterfactual correction is scaled by dimension step. The positions of the dandelion seeds after the step-by-step scaling of each dimension are compared with the lower and upper bounds of the variables. Variable values ​​below the lower bound are replaced with the lower bound, and variable values ​​above the upper bound are replaced with the upper bound, thus obtaining the dandelion seed positions after boundary clipping. The positions of the dandelion seeds after boundary clipping are used as the updated latent planning variable vectors, and all the updated latent planning variable vectors are combined to form the updated latent planning variable vector set.

[0037] When generating the updated set of potential planning variable vectors, the system converts the candidate major direction, candidate institution level, and learning time allocation results in the target counterfactual candidate path back to the variable values ​​of the corresponding decoding dimensions, forming the target potential planning variable vector. The current potential planning variable vector corresponding to the current candidate academic planning path is subtracted bit by bit from the target potential planning variable vector to obtain the counterfactual correction direction. The counterfactual regret value is normalized to obtain the counterfactual correction step size. The dandelion seed position is moved along the counterfactual correction direction by the counterfactual correction step size to form the counterfactually corrected dandelion seed position. The constraint violation degree is normalized to a constraint scaling factor and applied to the major direction decoding dimension and the institution level decoding dimension. The ability gap degree is normalized to an ability scaling factor and applied to the learning resource decoding dimension, completing the dimension-specific step size scaling. The scaled dandelion seed position is compared bit by bit with the lower and upper bounds of the variables. Values ​​exceeding the variable boundaries are pruned to the corresponding boundaries, resulting in the updated set of potential planning variable vectors.

[0038] In this embodiment, the specific steps for determining the target academic planning path include: Read the updated set of candidate academic planning paths; For the updated candidate academic planning paths, the candidate subject combinations, candidate institution levels, and study time allocation results are checked against the professional subject selection requirements, compared with the institution admission ranking range, and compared with the upper limit of study input, respectively. Updated candidate academic planning paths that fail to pass the check or comparison are deleted. Add perturbations for admission ranking, grade change, interest tendency, and learning engagement to the retained updated candidate academic planning paths; After the disturbance, the updated candidate academic planning paths are re-verified for major selection requirements, compared with the university admission ranking range, and compared with the upper limit of learning input. The updated candidate academic planning paths that still pass the verification and comparison after the disturbance are ranked according to the path evaluation value, and the updated candidate academic planning path ranked first is determined as the target academic planning path.

[0039] When determining the target academic planning path, the system reads the updated set of candidate academic planning paths and performs checks on each updated candidate path, including verification of subject selection requirements, comparison of university admission ranking ranges, and comparison of maximum learning effort. Updated candidate paths that fail the checks or comparisons are deleted. The retained updated candidate paths are then subjected to perturbations based on admission ranking, academic performance, interest, and learning effort, and are checked again for subject selection requirements, university admission ranking ranges, and maximum learning effort. Updated candidate paths that still pass the checks and comparisons after perturbation are ranked according to their path evaluation values, and the updated candidate path ranked first is determined as the target academic planning path.

[0040] An intelligent academic planning and consulting system based on AI big data analysis according to an embodiment of the present invention includes the following modules: The data acquisition module is used to collect student academic planning data and external educational constraints data; The feature vector generation module is used to group, encode, and normalize student academic planning data to generate student status feature vectors. The latent variable generation module is used to write the student state feature vector into the continuous search space of the Dandelion optimization algorithm to generate a set of latent planning variable vectors. The constraint decoding module is used to perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; The counterfactual path and regret value calculation module is used to generate counterfactual candidate paths and calculate counterfactual regret values. The seed location update module is used to update the dandelion seed locations based on the counterfactual regret value, generating an updated set of potential planning variable vectors. The path selection and solution generation module is used to re-execute constraint decoding, perform feasibility judgment and robustness selection, determine the target academic planning path and generate personalized academic planning consultation solutions.

[0041] Example 1: To verify the feasibility of this invention in practice, it was applied to an academic planning consultation scenario. This scenario targets a group of students with continuous learning records. The consultation objectives include confirming subject combinations, selecting majors, determining university levels, allocating learning resources, and setting phased academic goals. The students participating in the verification possess complete academic planning data and external educational constraint data. The academic planning data covers grades, rankings, learning behaviors, aptitude assessments, interests, subject selection intentions, and university application goals. The external educational constraint data covers major selection requirements, university admission rankings, university levels, and maximum learning resources.

[0042] In this scenario, the system aligns and groups student academic planning data according to continuous learning cycles, generates student state feature vectors through encoding and normalization, and writes these vectors into the continuous search space of the dandelion optimization algorithm. The system determines variable boundaries based on normalized values ​​and external educational constraint data, generates dandelion seed positions within these boundaries, and forms a set of potential planning variable vectors.

[0043] During the constraint decoding phase, the system divides the potential planning variable vector into professional direction decoding dimension, college level decoding dimension, and learning resource decoding dimension, respectively completing the verification of professional subject selection requirements, calculation of college admission ranking difference, and allocation of learning time proportion, generating a set of candidate academic planning paths.

[0044] In the counterfactual regret correction phase, the system generates counterfactual candidate paths at the institution level, major direction, and learning strategy levels for the current candidate academic planning paths, and calculates the counterfactual regret value. The system converts the target counterfactual candidate path corresponding to the counterfactual regret value into a target latent planning variable vector. Using the direction of the target latent planning variable vector relative to the current latent planning variable vector as the counterfactual correction direction, it controls the movement of the "dandelion seed" position and performs multi-dimensional scaling and boundary pruning based on constraint violation degree and ability gap degree. The corrected set of latent planning variable vectors is then re-decoded to obtain an updated set of candidate academic planning paths.

[0045] Three comparison methods were set up in the verification: manual rule matching, direct use of the dandelion optimization algorithm to search for specific planning paths, and the method of this invention. Each method processed the same batch of student data, and manual and constraint verification were performed after the output scheme. The verification content included major selection conflicts, mismatch of university rankings, excessive learning input, target path acceptance rate, path retention rate after perturbation, and average generation time. The verification results show that this invention has significant improvements in reducing constraint conflicts, improving path stability, and increasing generation efficiency. The experimental comparison results are shown in the table below: Table 1: Comparison of the Optimization Effects of Academic Planning Paths

[0046] The data in the table shows that the major selection conflict rate of the manual rule matching method is 12.8%, the university ranking mismatch rate is 18.4%, and the learning input excess rate is 15.6%. This indicates that when relying solely on grades, rankings, interests, and rule matching to generate academic planning paths, problems such as unmet major admission requirements, mismatch between university level and student ranking, and learning time exceeding the student's capacity are likely to occur.

[0047] After directly using the Dandelion optimization algorithm, the relevant indicators improved compared to the manual rule matching method, indicating that the optimized search can expand the search range of candidate paths and reduce some poorly fit paths. However, since this method directly searches for specific candidate academic planning paths without undergoing latent planning variable vector transformation and constraint decoding, there are still issues of incoordination between major direction, institution level, and learning resource allocation, and candidate paths still require further screening.

[0048] After adopting this invention, the major selection conflict rate decreased to 2.1%, the university ranking mismatch rate decreased to 4.3%, and the learning input exceeding limit rate decreased to 3.7%. The above data shows that constraint decoding can introduce major selection requirements, university admission ranking data, and learning input limit data into the candidate academic planning path generation stage, reducing candidate paths that do not meet educational constraints.

[0049] Regarding path stability, the robustness pass rate of this invention reaches 91.2%, indicating that the target academic planning path maintains high feasibility even under disturbances such as admission ranking, grade changes, interest tendencies, and learning engagement. The target path acceptance rate increased from 72.5% with manual rule matching to 88.9%, and the average generation time decreased from 18.6 minutes to 6.8 minutes, indicating that after correcting the dandelion seed position with counterfactual regret value, the target path is improved in terms of matching degree, executability, and generation efficiency.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart academic planning and consulting method based on AI big data analysis, characterized in that, Includes the following steps: Collect student academic planning data and external educational constraints data; The student academic planning data is grouped according to continuous learning cycles, and then coded and normalized to generate student state feature vectors. The student state feature vector is written into the continuous search space of the dandelion optimization algorithm, the variable boundary is determined according to the normalized value, the dandelion seed position is initialized, and a set of potential planning variable vectors is generated. Perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; Select the current candidate academic planning path from the set of candidate academic planning paths, perform single field preservation and adjacent field replacement to generate counterfactual candidate paths, and generate counterfactual regret values ​​based on the difference between the current candidate academic planning path and the counterfactual candidate path in terms of constraint satisfaction, goal attainability and learning input matching degree. When the evaluation value of the counterfactual candidate path exceeds the evaluation value of the current candidate academic planning path, the direction of the potential planning variable vector corresponding to the counterfactual candidate path relative to the current potential planning variable vector is used as the correction direction, and the counterfactual regret value is used as the correction step size weight to update the dandelion seed position and generate an updated set of potential planning variable vectors. The updated set of potential planning variable vectors is re-executed with constraint decoding to generate an updated set of candidate academic planning paths. The target academic planning path is determined through feasibility assessment and robustness screening, and a personalized academic planning consultation plan is generated based on the target academic planning path.

2. The intelligent academic planning and consulting method based on AI big data analysis according to claim 1, characterized in that, The student academic planning data and external educational constraint data include: Student academic planning data includes academic performance data for each subject, exam ranking data, learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance examination goal data; External educational constraint data includes data on majors, subject selection requirements, university admission rankings, university level, and maximum learning input.

3. The intelligent academic planning and consulting method based on AI big data analysis according to claim 2, characterized in that, The operation of generating student state feature vectors specifically includes: The student academic planning data is aligned and grouped according to the student identifier and the consecutive learning cycle number to generate learning cycle data groups; Within each learning cycle data group, extract the changes in grades and rankings, and encode the fields for learning behavior data, ability assessment data, interest data, subject selection intention data, and college entrance goal data. Normalize the changes in grades, changes in ranking, and field encoding results, and concatenate them according to a fixed field order to generate a student status feature vector.

4. The intelligent academic planning and consulting method based on AI big data analysis according to claim 3, characterized in that, The operation of generating the set of potential planning variable vectors specifically includes: Read the student status feature vector and build a variable dimension sequence table according to the field order; For each variable dimension in the variable dimension sequence list, extract the normalized field values ​​of all learning cycle data groups under the same variable dimension, sort them in ascending order, and take the first value as the initial lower bound of the variable and the last value as the initial upper bound of the variable. The constraint values ​​corresponding to the variable dimensions in the external education constraint data are converted into normalized constraint intervals, and the intersection of these intervals with the initial lower and upper bounds of the variables is processed to obtain the lower and upper bounds of the variables. Combine the lower and upper bounds of variables according to the variable dimension sequence list to form the variable boundaries of the continuous search space; Within the variable boundaries, dandelion seed position values ​​are generated, and all variable dimension values ​​under the same dandelion seed position are concatenated to form a latent planning variable vector, resulting in a set of latent planning variable vectors.

5. The intelligent academic planning and consulting method based on AI big data analysis according to claim 4, characterized in that, The operation of generating a set of candidate academic planning paths specifically includes: Read the vector of potential planning variables and divide it into professional direction decoding dimension, institutional level decoding dimension and learning resource decoding dimension according to the variable dimension sequence list; The distance between the variable values ​​of the professional direction decoding dimension and the normalized code of the professional direction in the professional data is calculated one by one. The professional subject selection requirements are read according to the distance and the professional subject selection requirements are checked against the subject selection intention field one by one. Candidate subject selection combinations and candidate professional directions are retained. The distance between the variable values ​​of the institution level decoding dimension and the institution level normalized code in the institution level data is calculated one by one. The institution admission ranking data is read according to the distance. The institution admission ranking data and the ranking ranking change field are used to calculate the ranking difference. Institution levels whose ranking difference does not fall into the admission ranking range are deleted, and candidate institution levels are retained. The variable values ​​of the learning resource decoding dimension are converted into the proportion of learning time for each subject, and the learning time allocation results are formed based on the performance change field and the upper limit data of learning input. The candidate academic planning paths are generated by combining candidate majors, candidate subject combinations, candidate university levels, and study time allocation results.

6. The intelligent academic planning and consulting method based on AI big data analysis according to claim 5, characterized in that, The operation of generating counterfactual candidate paths specifically includes: Retrieve the candidate majors, candidate university levels, and study time allocation results from the current candidate academic planning paths; Keeping the candidate major direction unchanged, the candidate institution level is replaced with the institution level whose level sorting number differs by one number, generating the institution level counterfactual candidate path; Keeping the candidate institution level unchanged, select adjacent professional directions under the same discipline category identifier, check the subject selection requirements of the adjacent professional directions against the subject selection intention field item by item, and generate counterfactual candidate paths for professional directions. Keeping the candidate major and candidate institution level unchanged, the proportion of study time for each subject is adjusted from small to large according to the field of grade change, and the total proportion of study time is limited to the upper limit of study input data to generate counterfactual candidate paths for learning strategies. The counterfactual candidate paths are composed of counterfactual candidate paths at the institutional level, counterfactual candidate paths of majors, and counterfactual candidate paths of learning strategies.

7. The intelligent academic planning and consulting method based on AI big data analysis according to claim 6, characterized in that, The operation of generating counterfactual regret values ​​specifically includes: Read the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path for the major direction, and the counterfactual candidate path for the learning strategy; The following steps are performed on the current candidate academic planning path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level: verify the subject selection requirements, calculate the difference in the admission ranking of the institution, and compare the upper limit of learning input. The results are used to generate the evaluation values ​​for the current path, the counterfactual candidate path at the institution level, the counterfactual candidate path at the major direction, and the counterfactual candidate path at the learning strategy level. The counterfactual path evaluation values ​​for institutional level, major direction, and learning strategy are calculated by comparing them with the current path evaluation value to generate the counterfactual difference values ​​for institutional level, major direction, and learning strategy. The counterfactual discrepancies of the institution level, major direction, and learning strategy are sorted in descending order. The counterfactual discrepancy at the top of the list is taken as the counterfactual regret value, and the counterfactual candidate path corresponding to the counterfactual discrepancy at the top of the list is taken as the target counterfactual candidate path. The constraint violation degree is generated based on the number of failed subjects in the professional subject selection requirements, the number of excesses in the calculation of the difference in the university admission ranking, and the number of excesses in the comparison of the upper limit of learning input. The ability gap is generated based on the difference between the phased achievement goals and the achievement change field, the difference between the admission rank and the ranking rank change field corresponding to the candidate institution level, and the difference between the study time allocation results and the upper limit data of study input.

8. The intelligent academic planning and consulting method based on AI big data analysis according to claim 7, characterized in that, The operation of generating the updated set of potential planning variable vectors specifically includes: Read the target counterfactual candidate path corresponding to the counterfactual regret value from the counterfactual candidate path, and map the candidate major direction, candidate institution level and learning time allocation results in the target counterfactual candidate path back to the major direction decoding dimension, institution level decoding dimension and learning resource decoding dimension, respectively, to obtain the target potential planning variable vector; Read the current potential planning variable vector corresponding to the current candidate academic planning path, and subtract the target potential planning variable vector from the current potential planning variable vector digit by digit according to the variable dimension sequence list to obtain the counterfactual correction direction; The counterfactual regret value is normalized to obtain the counterfactual correction step size; In the seed position update stage of the dandelion optimization algorithm, the update direction of the dandelion seed position is set as the counterfactual correction direction, and the dandelion seed position is moved along the counterfactual correction direction by the counterfactual correction step size to obtain the counterfactual corrected dandelion seed position. The constraint violation degree is normalized to obtain the constraint scaling factor, the capability gap degree is normalized to obtain the capability scaling factor, the constraint scaling factor is applied to the professional direction decoding dimension and the institutional level decoding dimension, the capability scaling factor is applied to the learning resource decoding dimension, and the dandelion seed position after counterfactual correction is scaled by dimension step. The positions of the dandelion seeds after the step-by-step scaling of each dimension are compared with the lower and upper bounds of the variables. Variable values ​​below the lower bound are replaced with the lower bound, and variable values ​​above the upper bound are replaced with the upper bound, thus obtaining the dandelion seed positions after boundary clipping. The positions of the dandelion seeds after boundary clipping are used as the updated latent planning variable vectors, and all the updated latent planning variable vectors are combined to form the updated latent planning variable vector set.

9. The intelligent academic planning and consulting method based on AI big data analysis according to claim 8, characterized in that, The specific steps involved in determining the target academic planning path include: Read the updated set of candidate academic planning paths; For the updated candidate academic planning paths, the candidate subject combinations, candidate institution levels, and study time allocation results are checked against the professional subject selection requirements, compared with the institution admission ranking range, and compared with the upper limit of study input, respectively. Updated candidate academic planning paths that fail to pass the check or comparison are deleted. Add perturbations for admission ranking, grade change, interest tendency, and learning engagement to the retained updated candidate academic planning paths; After the disturbance, the updated candidate academic planning paths are re-verified for major selection requirements, compared with the university admission ranking range, and compared with the upper limit of learning input. The updated candidate academic planning paths that still pass the verification and comparison after the disturbance are ranked according to the path evaluation value, and the updated candidate academic planning path ranked first is determined as the target academic planning path.

10. An intelligent chemical industry planning and consulting system based on AI big data analysis, applied to the intelligent chemical industry planning and consulting method based on AI big data analysis as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition module is used to collect student academic planning data and external educational constraints data; The feature vector generation module is used to group, encode, and normalize student academic planning data to generate student status feature vectors. The latent variable generation module is used to write the student state feature vector into the continuous search space of the Dandelion optimization algorithm to generate a set of latent planning variable vectors. The constraint decoding module is used to perform constraint decoding on the set of potential planning variable vectors to generate a set of candidate academic planning paths; The counterfactual path and regret value calculation module is used to generate counterfactual candidate paths and calculate counterfactual regret values. The seed location update module is used to update the dandelion seed locations based on the counterfactual regret value, generating an updated set of potential planning variable vectors. The path selection and solution generation module is used to re-execute constraint decoding, perform feasibility judgment and robustness selection, determine the target academic planning path and generate personalized academic planning consultation solutions.

Images