Teacher team structure optimization and dynamic configuration method based on big data analysis

Abstract

The invention discloses a teacher team structure optimization and dynamic configuration method based on big data analysis, and the method comprises the following steps: S1, collecting the personal information of a teacher and the post demand data of a school, and constructing an original data set based on the personal information of the teacher; s2, preprocessing the original data set; s3, clustering is carried out on the preprocessed original data set through a VaDE clustering model; s4, matching a teacher structure clustering result with post demand data, and performing global search by adopting an improved wormhole behavior particle swarm optimization algorithm to generate a preliminary teacher resource allocation scheme; s5, updating the standardized data set and post demand data according to a scheme execution result; s6, dynamically updating the configuration scheme; and S7, outputting the teacher resource configuration scheme to an education management system. According to the method, VaDE clustering and the improved particle swarm optimization are fused, intelligent matching optimization of teacher resources is achieved, and the method has the advantages of being accurate in structure, dynamic in response and efficient in configuration.

Description

Technical Field

[0001] The present invention relates to the technical field of education management, and particularly to a method for optimizing the structure and dynamically allocating a teaching staff based on big data analysis. Background Art

[0002] In the prior art, the analysis and allocation of the teaching staff structure mostly rely on manual investigations, static statistics of human resource information systems, and expert rule formulation. Although some regions have constructed a teacher data platform based on an information system, its processing methods are mostly for the display and query of tabular information, lacking the ability of in-depth modeling and intelligent analysis of multi-dimensional teacher information. For example, during the teacher deployment process, manual judgments are usually made through single indicators such as teaching age, professional title, and subject matching degree, ignoring the comprehensive performance of teachers in dimensions such as teaching behavior, classroom feedback, and teaching and research activities, resulting in one-sided structure modeling and subjective deployment results.

[0003] In recent years, with the development of artificial intelligence and data mining technologies, some studies have attempted to introduce algorithms such as clustering analysis, decision trees, and support vector machines into education management systems for teacher classification, job recommendation, or teaching performance evaluation. However, most of these methods use shallow models or static clustering methods, unable to capture the complex temporal characteristics of the teaching staff under multi-source heterogeneous data, and lacking the adaptive update ability of the model under feedback data.

[0004] At the same time, most current resource allocation schemes are single-round decision-making mechanisms, lacking the closed-loop feedback ability of execution results. After the initial allocation of teacher resources, the education system rarely incorporates dynamic data such as teaching feedback, job changes, and teacher mobility back into the model for optimization iteration, making the resource allocation lack the ability of continuous optimization.

[0005] In summary, the existing teacher resource management methods generally have the following problems: First, there is a lack of means for in-depth modeling of multi-source, multi-dimensional, and dynamic teacher data, and the structure of the teaching staff cannot be truly depicted; second, job matching depends on manual judgment or simple rules, with low intelligence level and weak optimization ability; third, the model lacks a feedback mechanism and cannot adapt to the dynamic changes of the system structure and requirements; fourth, the optimization algorithm is limited in performance when facing complex constraints and a large-scale variable space, and cannot effectively support large-scale resource allocation scenarios. Summary of the Invention

[0006] An object of the present invention is to propose a method for optimizing the structure and dynamically allocating a teaching staff based on big data analysis. The present invention integrates VaDE clustering and an improved wormhole behavior particle swarm optimization algorithm to achieve intelligent matching and dynamic optimization adjustment with job requirements, and has the advantages of high structural modeling accuracy, strong resource allocation efficiency, strong system response ability, and strong adaptability, and can effectively improve the rationality and utilization rate of teacher resources in the education system.

[0007] A method for optimizing the structure and dynamically allocating a teaching staff based on big data analysis according to an embodiment of the present invention includes the following steps: S1. Collect personal information of teachers and post demand data of schools, and construct an original data set based on the personal information of teachers; S2. Preprocess the original data set to form a standardized data set; S3. Extract teacher feature vectors based on the standardized data set, construct a VaDE clustering model, and cluster the teacher feature vectors through the variational autoencoder and Gaussian mixture model of the VaDE clustering model, and output the teacher structure clustering result and the corresponding class probability distribution; S4. Construct an optimization model for teacher resource allocation, match the teacher structure clustering result with the post demand data, and perform global search using an improved wormhole behavior particle swarm optimization algorithm to generate a preliminary teacher resource allocation plan; S5. Execute the preliminary teacher resource allocation plan, and update the standardized data set and post demand data according to the execution result of the plan; S6. Feed the updated data back to the VaDE clustering model and the teacher resource allocation optimization model, adjust the teacher structure clustering result and the teacher resource allocation plan, and generate a dynamically updated allocation plan; S7. Output the final teacher resource allocation plan to the education management system.

[0008] Optionally, the personal information includes basic information, teaching behavior, professional title level, teaching subject, post resume, teaching and research records, and classroom feedback information.

[0009] Optionally, the post demand data includes post setting data, subject curriculum structure, student scale, teaching load standard, and post ability requirements.

[0010] Optionally, the preprocessing includes format conversion, missing value filling, outlier removal, normalization, and standardization processing.

[0011] Optionally, the S3 specifically includes: S31. Based on the standardized data set, extract the feature vector of each teacher, and form an input matrix with all the feature vectors , where represents the total number of teachers, represents the th teacher's feature vector; S32. Construct a VaDE clustering model, input the input matrix into the encoder network of the variational autoencoder, and obtain the mean vector and logarithmic variance vector of the latent representation vector. The VaDE clustering model includes a variational autoencoder and a Gaussian mixture model in the latent space. The variational autoencoder includes an encoder network and a decoder network. The encoder network extracts the distribution parameters of the latent representation vector of the input matrix, the decoder network reconstructs the input features, and the Gaussian mixture model models the prior distribution of the latent representation vector in the latent space; S33. Adopt the reparameterization trick to map the mean vector and logarithmic variance vector of the latent representation vector output by the encoder network into a latent representation vector, and obtain the posterior distribution of the latent representation vector: ; where, represents the latent representation vector of the th teacher, represents the mean vector of the latent representation vector of the th teacher, represents the logarithmic variance vector of the latent variable of the th teacher, represents element-wise multiplication of vectors, represents a random noise vector sampled from the standard Gaussian distribution, represents a diagonal covariance matrix, represents the posterior distribution of the latent representation vector, generated by the encoder network, represents the Gaussian distribution, represents the training parameters of the encoder network; S34. Establish a Gaussian mixture model in the latent space as the prior distribution: ; where, represents the prior distribution of the latent representation vector, represents the total number of clusters in the Gaussian mixture model, represents the prior weight of the th cluster, represents the probability density function of the Gaussian distribution with the mean vector and covariance matrix as parameters at the point , represents the covariance matrix of the th cluster, represents the mean vector of the th cluster, represents the latent representation variable; S35. Input the latent representation vector into the decoder network to generate a reconstructed vector, and at the same time calculate the reconstructed probability distribution of the latent representation vector belonging to each cluster: ; Among them, represents the reconstruction probability distribution of the latent representation vector, describing the probability that the latent representation vector belongs to the th cluster, represents the cluster label of the th teacher, represents the latent representation vector of the th teacher, represents the prior weight of the th cluster, represents the covariance matrix of the th cluster, represents the mean vector of the th cluster, represents the probability density function of the Gaussian distribution with the mean vector and covariance matrix as parameters at the point ; S36. Jointly minimize the reconstruction error and the difference in the latent space distribution to construct a comprehensive loss function: ; Among them, represents the comprehensive loss function, represents the Kullback-Leibler divergence between the posterior distribution of the latent representation vector and the Gaussian mixture prior distribution, represents the mathematical expectation, represents the training parameters of the encoder network, represents the decoder parameters, represents the posterior distribution of the latent representation vector, generated by the encoder network, represents the prior distribution of the Gaussian mixture defined in the latent space, represents the mean vector of the th cluster, represents the probability density function of the Gaussian distribution with the mean vector and covariance matrix as parameters at the point ; represents the total number of teachers, represents the reconstruction probability distribution generated by the decoder network, represents the th teacher's feature vector; S37. After training the VaDE clustering model until convergence, output each latent representation vector , the teacher structure clustering category probability distribution and the teacher structure clustering result .

[0012] Optionally, the S4 specifically includes: S41. Represent the teacher structure clustering result as a set , where represents the clustering category to which the -th teacher belongs. Represent the job requirement data as a set , where represents the requirement data for the -th job, represents the total number of jobs. Based on the sets and , construct the matching optimization problem between teachers and jobs; S42. Initialize the improved wormhole behavior particle swarm optimization algorithm. Represent each particle as a position vector , where represents whether the teacher in particle is assigned to the job . Initialize the particle swarm population . Each particle corresponds to a teacher resource allocation scheme. The improvements of the improved wormhole behavior particle swarm optimization algorithm include a wormhole perturbation mechanism, a global optimal guidance mechanism, and an adaptive jump probability adjustment mechanism; S43. In each round of iteration, update the particle position and velocity according to the global optimal guidance mechanism; ; ; where represents the updated velocity of particle in the -th round of iteration at the -th job, represents the inertia weight, and represent the learning factors, and represent random numbers in the range [0, 1], represents the current velocity of particle in the -th round of iteration at the -th job, represents the historical optimal position of the particle, represents the current position of particle in the -th round of iteration at the -th job, represents the current global optimal position, represents the updated position of particle in the -th round of iteration at the -th job; S44, introduce the wormhole perturbation mechanism in each round of iteration, with the wormhole perturbation probability Some particles are selected to introduce disturbances. The wormhole disturbance mechanism simulates the long-distance jump of particles "traversing" the search space by introducing nonlinear and globally guided jump operations, thereby enhancing global optimization: ; in, Indicates Round Iteration Particle In the The updated position of the job, Indicates Round Iteration Particle In the Current location of the job, represents the disturbance intensity coefficient, represents a standard Gaussian distributed random variable with mean 0 and variance 1; S45. Set an adaptive jump probability adjustment mechanism to dynamically adjust the wormhole disturbance probability with the number of iterations: ; in, Indicates The probability of wormhole perturbation in round iterations, represents the initial jump probability value, represents a natural constant, represents the exponential decay coefficient; S46. After each round of iteration, compare the fitness function values ​​of all particles and select the particle with the largest fitness value. ,in represents the particle population set, represents the fitness function value, The corresponding particle position vector is the optimal teacher resource allocation scheme, and the particle allocation result is output. ,in Indicates that the Teachers are assigned to Positions, Indicates no allocation, and finally generates a preliminary teacher resource allocation plan.

[0013] Optionally, the execution results of the scheme include teaching feedback data, job adjustment data and teacher mobility records.

[0014] Optionally, the S5 specifically includes: S51. Implement the preliminary teacher resource allocation plan, record the implementation of the allocation plan in each position, and collect the plan implementation result data; S52. Constructing a teaching feedback data matrix , where represents the teaching feedback score of the th teacher on the th position; S53. Construct a set of job adjustment data , where represents the adjustment status of the th position, represents the cancellation of the position, 0 represents no change, represents the addition of a new position; S54. Construct a teacher mobility record matrix , where represents the transfer of the th teacher to the th position, represents the transfer from the th position, represents no job change; S55. Update the feature vector of the teacher based on the teaching feedback data, and update the job demand data based on the job adjustment data and teacher mobility records; S56. Replace the original standardized data set and job data with the updated teacher feature vector and job demand data.

[0015] Optionally, the teaching feedback score consists of student satisfaction, teaching completion rate, and classroom assessment average.

[0016] Optionally, the specific steps of S6 are as follows: S61. Use the updated standardized data set and job demand data as input data to dynamically adjust the VaDE clustering model and teacher resource allocation optimization model respectively; S62. Input the updated teacher feature data into the VaDE clustering model and output the latest clustering result of the teacher structure; S63. Re-match the updated clustering result with the updated job demand data to construct a dynamic matching relationship between teachers and positions. Based on the new matching relationship, re-initialize the particle swarm in the teacher resource allocation optimization model; S64. Execute the resource allocation optimization process of the wormhole perturbation mechanism, global optimal guidance mechanism, and adaptive jump probability adjustment mechanism. Continuously update the particle state and evaluate the configuration fitness during the optimization process. After reaching the maximum number of iterations or meeting the convergence condition, select the particle position vector with the optimal fitness to generate a dynamically updated teacher resource allocation plan.

[0017] The beneficial effects of the present invention are: First, the present invention comprehensively collects multi-source data such as the basic information, teaching behaviors, professional titles, teaching subjects, teaching and research records, and classroom feedback of teachers, and constructs a multi-dimensional matching space between teachers and positions in combination with the school's job settings, curriculum structures, teaching loads, and job ability requirements, effectively enhancing the breadth and depth of data modeling and breaking through the limitations of traditional classification and allocation based on single factors such as teaching age or professional title.

[0018] Secondly, the present invention introduces the VaDE clustering model to deeply model and cluster-identify the standardized teacher feature vectors, fully utilizes the modeling ability of variational autoencoders in processing high-dimensional non-linear data, and combines Gaussian mixture models for soft clustering modeling in the latent space, which not only improves the expression ability and stability of clustering results but also avoids the uncertainty of traditional hard clustering methods in identifying boundary teachers, making the teacher structure portrait more real and refined.

[0019] In addition, based on the multi-objective optimization matching between the clustering results and job requirements, the present invention designs an improved wormhole behavior particle swarm optimization algorithm, introduces a wormhole perturbation mechanism to enhance the algorithm's ability to jump out of local optima, introduces a global optimal guidance mechanism to improve the search directionality, and at the same time improves the efficiency and stability of search in different stages through an adaptive jump probability adjustment mechanism, achieving efficient solution of multiple constraints and non-linear objectives in large-scale teacher and position matching tasks.

[0020] Finally, the present invention has a perfect feedback and update mechanism. After the initial configuration plan is executed, it can collect teaching feedback data, job adjustment information, and teacher flow records in real time, and re-enter the updated information data into the VaDE clustering model and resource allocation optimization model to achieve dynamic update and adaptive adjustment of clustering results and optimization strategies, thereby constructing a "perception-optimization-feedback-re-optimization" closed-loop mode for teacher resource allocation. This mode improves the model's response ability to actual teaching changes and the continuous optimization ability of the allocation plan, making teacher allocation more in line with actual job requirements and teaching effect performance, and effectively realizing dynamic scheduling and structural optimization of resources. Description of the Drawings

[0021] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for optimizing the structure and dynamically allocating a teacher team based on big data analysis proposed by the present invention; Figure 2 is a schematic structural diagram of the VaDE model of a method for optimizing the structure and dynamically allocating a teacher team based on big data analysis proposed by the present invention; Figure 3The matching optimization flowchart of the improved wormhole behavior particle swarm optimization algorithm for an optimization and dynamic configuration method of teacher team structure based on big data analysis proposed by the present invention. Specific embodiments

[0022] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0023] Reference Figures 1-3 , an optimization and dynamic configuration method of teacher team structure based on big data analysis, includes the following steps: S1. Collect personal information of teachers and post demand data of schools, and construct an original data set based on the personal information of teachers; S2. Preprocess the original data set to form a standardized data set; S3. Extract teacher feature vectors based on the standardized data set, construct a VaDE clustering model, and cluster the teacher feature vectors through the variational autoencoder and Gaussian mixture model of the VaDE clustering model, and output the teacher structure clustering result and the corresponding category probability distribution; S4. Construct an optimization model for teacher resource allocation, match the teacher structure clustering result and the post demand data, and perform a global search using an improved wormhole behavior particle swarm optimization algorithm to generate a preliminary teacher resource allocation plan; S5. Execute the preliminary teacher resource allocation plan, and update the standardized data set and the post demand data according to the execution result of the plan; S6. Feed the updated data back to the VaDE clustering model and the teacher resource allocation optimization model, adjust the teacher structure clustering result and the teacher resource allocation plan, and generate a dynamically updated allocation plan; S7. Output the final teacher resource allocation plan to the education management system.

[0024] The present invention realizes the deep modeling of the teacher structure and the intelligent matching of the post demand by constructing a complete teacher resource allocation process including data collection, preprocessing, clustering modeling, intelligent optimization, dynamic feedback and result output, improves the scientificity, intelligence and dynamic adaptability of the teacher team allocation, and can effectively solve problems such as unreasonable teacher resource allocation structure, untimely response, and low optimization efficiency.

[0025] In this embodiment, the personal information includes basic information, teaching behavior, professional title level, teaching subject, post resume, teaching and research records, and classroom feedback information.

[0026] The present invention uses teachers' basic information, teaching behaviors, professional titles, teaching subjects, job resumes, teaching and research records, and classroom feedback information as the basic data sources for teacher modeling, making the dimensions of the teacher portrait more comprehensive and the feature expression more accurate, which helps to construct a teacher structure model that meets the actual teaching ability and job requirements, thereby improving the accuracy of subsequent clustering and matching.

[0027] In this embodiment, the job demand data includes job setting data, subject curriculum structure, student scale, teaching load standard, and job ability requirements.

[0028] The present invention collects multiple job demand data such as job setting, curriculum structure, student scale, teaching load standard, and ability requirements, expanding the expression of teacher job demands from a single job type to a multi-dimensional matching standard, enhancing the expression ability of the job portrait, and improving the accuracy and practicality of job matching.

[0029] In this embodiment, the preprocessing includes format conversion, missing value filling, outlier removal, normalization, and standardization processing.

[0030] The present invention ensures the consistency and usability of teachers' original data before inputting into the model by introducing preprocessing operations such as format conversion, missing value filling, outlier removal, normalization, and standardization, improving the robustness and data quality of subsequent feature extraction and clustering modeling, and reducing the model error risk.

[0031] In this embodiment, S3 specifically includes: S31. Based on the standardized data set, extract the feature vectors of each teacher, and form an input matrix with all the feature vectors , where represents the total number of teachers, represents the -th teacher's feature vector; S32. Construct a VaDE clustering model, input the input matrix into the encoder network of the variational autoencoder to obtain the mean vector and logarithmic variance vector of the latent representation vector. The VaDE clustering model includes a variational autoencoder and a Gaussian mixture model in the latent space. The variational autoencoder includes an encoder network and a decoder network. The encoder network extracts the distribution parameters of the latent representation vector of the input matrix, the decoder network reconstructs the input features, and the Gaussian mixture model models the prior distribution of the latent representation vector in the latent space; S33. Use the reparameterization trick to map the mean vector and logarithmic variance vector of the latent representation vector output by the encoder network to the latent representation vector, and obtain the posterior distribution of the latent representation vector: ; where, Represents the latent representation vector of the th teacher, Represents the mean vector of the latent representation vectors of the th teacher, Represents the logarithmic variance vector of the latent variables of the th teacher, Represents element-wise multiplication of vectors, Represents a random noise vector sampled from a standard Gaussian distribution, Represents a diagonal covariance matrix, Represents the posterior distribution of the latent representation vector, generated by the encoder network, Represents a Gaussian distribution, Represents the training parameters of the encoder network; S34. Establish a Gaussian mixture model as the prior distribution in the latent space: ; Where, Represents the prior distribution of the latent representation vector, Represents the total number of clusters in the Gaussian mixture model, Represents the prior weight of the th cluster, Represents the probability density function of the Gaussian distribution with mean vector and covariance matrix as parameters at the point , Represents the th covariance matrix of the cluster, Represents the th mean vector of the cluster, Represents the latent representation variable; S35. Input the latent representation vector into the decoder network to generate a reconstructed vector, and at the same time calculate the reconstructed probability distribution of the latent representation vector belonging to each cluster: ; Where, Represents the reconstructed probability distribution of the latent representation vector, describing the probability that the latent representation vector belongs to the th cluster, Represents the cluster label of the th teacher, Represents the latent representation vector of the th teacher, Represents the th prior weight of the cluster, Represents the th covariance matrix of the cluster, Represents the th mean vector of the cluster, Represents the Gaussian distribution with mean vector and covariance matrix as parameters at the point The probability density function at S36. Jointly minimize the reconstruction error and the difference in the latent space distribution to construct a comprehensive loss function: ; where represents the comprehensive loss function, represents the Kullback-Leibler divergence between the posterior distribution of the latent representation vector and the Gaussian mixture prior distribution, represents the mathematical expectation, represents the training parameters of the encoder network, represents the decoder parameters, represents the posterior distribution of the latent representation vector, generated by the encoder network, represents the prior distribution of the Gaussian mixture defined in the latent space, represents the th mean vector of the cluster, represents the probability density function of the Gaussian distribution with the mean vector and covariance matrix as parameters at the point ; represents the total number of teachers, represents the reconstruction probability distribution generated by the decoder network, represents the th feature vector of the teacher; S37. After training the VaDE clustering model until convergence, output each latent representation vector , the probability distribution of the teacher structure clustering category, and the teacher structure clustering result .

[0032] The present invention adopts a deep clustering method based on the VaDE model, extracts latent representations through a variational autoencoder in a high-dimensional space, and introduces a Gaussian mixture model in the latent space for clustering, solving the problem of poor processing ability of traditional clustering algorithms for high-dimensional non-linear data and improving the accuracy and stability of teacher structure clustering.

[0033] In this embodiment, the S4 specifically includes: S41. Represent the teacher structure clustering result as a set , where represents the clustering category to which the th teacher belongs, represent the job requirement data as a set , where represents the requirement data of the th job, represents the total number of jobs, according to the sets and Construct the matching optimization problem between teachers and positions; S42, initialize the improved wormhole behavior particle swarm optimization algorithm, and represent each particle as a position vector ,in Represents particles Are teachers assigned to positions? , initialize the particle swarm population , each particle corresponds to a teacher resource configuration scheme, and the improved points of the improved wormhole behavior particle swarm optimization algorithm include wormhole perturbation mechanism, global optimal guidance mechanism and adaptive jump probability adjustment mechanism; S43. In each round of iteration, the particle position and velocity are updated according to the global optimal guidance mechanism; ; ; in, Indicates Round Iteration Particle In the The speed at which positions are updated, represents the inertia weight, and represents the learning factor, and Represents a random number in the range [0,1]. Indicates Round Iteration Particle In the The current speed of each position, represents the best historical position of the particle, Indicates Round Iteration Particle In the Current location of the job, represents the current global optimal position, Indicates Round Iteration Particle In the The updated position of each position; S44, introduce the wormhole perturbation mechanism in each round of iteration, with the wormhole perturbation probability Some particles are selected to introduce disturbances. The wormhole disturbance mechanism simulates the long-distance jump of particles "traversing" the search space by introducing nonlinear and globally guided jump operations, thereby enhancing global optimization: ; in, Indicates Round Iteration Particle In the The updated position of a post, indicating the round of iterative particles at the current position of the th post, indicating the perturbation intensity coefficient, representing a standard Gaussian distribution random variable with a mean of 0 and a variance of 1; S45. Set an adaptive jump probability adjustment mechanism to dynamically adjust the wormhole perturbation probability with the number of iterations: ; wherein, represents the wormhole perturbation probability of the th round of iteration, represents the initial jump probability value, represents the natural constant, represents the exponential decay coefficient; S46. After each round of iteration is completed, compare the fitness function values of all particles and select the particle with the largest fitness value , where represents the particle population set, represents the fitness function value, The corresponding particle position vector is the optimal teacher resource allocation scheme, and output the allocation result of the particle , where represents allocating the th teacher to the th post, represents non-allocation, and finally generate a preliminary teacher resource allocation scheme.

[0034] The present invention constructs a teacher post matching optimization model based on the wormhole behavior particle swarm optimization algorithm, introduces the wormhole perturbation, global guidance and adaptive jump mechanisms, breaks through the bottleneck of the traditional particle swarm optimization falling into local optimum, improves the ability to search for the global optimal configuration scheme, and improves the teacher resource allocation efficiency and the algorithm convergence speed.

[0035] In this embodiment, the execution result of the scheme includes teaching feedback data, post adjustment data and teacher flow records.

[0036] The present invention feeds back teaching feedback, post adjustment and teacher flow data as execution results to the model, establishes a dynamic update path based on the configuration result, realizes the adaptive learning and correction of the model for the actual execution effect, and improves the dynamic response ability of the teacher configuration system and the closed-loop ability of resource optimization.

[0037] In this embodiment, the S5 specifically includes: S51. Execute the preliminary teacher resource allocation plan, record the implementation of the allocation plan for each position, and collect the data of the implementation results of the plan; S52. Construct a teaching feedback data matrix , where represents the teaching feedback score of the th teacher on the th position; S53. Construct a set of position adjustment data , where represents the adjustment status of position , represents position cancellation, 0 represents no change, represents a newly added position; S54. Construct a teacher mobility record matrix , where represents the th teacher transferred to the th position, represents transferred out from the th position, represents no position change; S55. Update the feature vector of the teacher based on the teaching feedback data, and update the position demand data based on the position adjustment data and the teacher mobility record; S56. Replace the original standardized data set and position data with the updated teacher feature vector and position demand data.

[0038] By refining the process of collecting and updating the implementation result feedback data, the present invention constructs a structured feedback matrix and a position mobility information matrix, further enhancing the standardization and computability of data update, providing high-quality input for model adjustment, ensuring that the optimization plan can be adjusted in a timely manner with changes in positions and teacher status, and improving the practicality and extensibility of the model.

[0039] In this embodiment, the teaching feedback score consists of student satisfaction, teaching completion rate, and average classroom evaluation.

[0040] The present invention uses student satisfaction, teaching completion rate, and average classroom evaluation as teaching feedback evaluation indicators, which can more truly reflect the teaching performance of teachers in actual positions, improve the guidance of feedback data for teacher feature update and the judgment ability of the model for position adaptation, and strengthen the coordination mechanism between teacher portraits and position feedback.

[0041] In this embodiment, the specific steps of S6 include: S61. Use the updated standardized data set and position demand data as input data to dynamically adjust the VaDE clustering model and the teacher resource allocation optimization model respectively; S62. Input the updated teacher feature data into the VaDE clustering model to output the latest clustering results of the teacher structure; S63. Re-match the updated clustering results with the updated job demand data to construct a dynamic matching relationship between teachers and jobs. Based on the new matching relationship, re-initialize the particle swarm in the teacher resource allocation optimization model; S64. Execute the resource allocation optimization process of the wormhole perturbation mechanism, global optimal guidance mechanism, and adaptive jump probability adjustment mechanism. Continuously update the particle state and evaluate the configuration fitness during the optimization process. After reaching the maximum number of iterations or meeting the convergence condition, select the particle position vector with the optimal fitness to generate a dynamically updated teacher resource allocation plan.

[0042] Through the dynamic feedback mechanism, the present invention re-inputs the updated teacher features and job demand data into the clustering and optimization model, realizes the iterative optimization of teacher structure clustering and job matching, effectively improves the adaptability of the model to actual teaching changes, enables the teacher resource allocation to have the ability of self-learning and continuous evolution, and ensures the continuous optimization of educational resource allocation.

[0043] Example 1: To verify the feasibility of the present invention in implementation, the present invention is applied to the joint teaching staff management platform of 15 high schools under the Education Bureau of City Y in a certain province, and a pilot project of optimizing the teacher team structure and dynamic allocation is carried out for half a year (from March 2024 to September 2024), aiming to solve the long-existing management problems such as the imbalance of resource allocation structure among high school teachers in this region, the long-term overwork of some subject teachers, and the dependence on manual experience for teacher deployment.

[0044] The total number of teachers involved in the pilot area is 1286, covering 10 major subjects such as Chinese, Mathematics, English, Physics, Chemistry, Biology, History, Geography, Politics, and Information Technology. Due to different school-running histories, orientations, and regions, the teacher structures of the 15 high schools are significantly different. The proportion of high-title teachers in some municipal demonstration high schools exceeds 45%, while the proportion of senior teachers in suburban ordinary high schools is only 17%; and in the actual job settings, there are phenomena such as repeated part-time teaching of Chinese, Mathematics, and English teachers and uneven distribution of science comprehensive teachers in many schools, which seriously affect the balanced development of teaching quality.

[0045] In the pilot process, first relying on the teacher personnel file management system and teaching management platform of each school, data such as the basic information, teaching behavior records, professional title levels, teaching subjects, participation in teaching and research, job resumes, and classroom teaching feedback of each teacher are collected. Combining the job settings, curriculum structures, teaching load standards, student scales, and job ability requirements exported from the academic affairs systems of each school, a data pool covering multi-dimensional characteristics of "person-position" is constructed. After format conversion, missing value filling, outlier removal, normalization, and standardization processing, a standardized data set is formed for the model to use.

[0046] In the modeling stage, the VaDE clustering model proposed in the present invention is used to perform structural modeling on 1,286 teachers. The variational autoencoder is used to extract the latent representation, and a Gaussian mixture model is constructed in the latent space to achieve deep clustering of teachers at the structural level. Finally, 12 types of teacher clustering structures are output. Each type of teacher has significant differences in subject expertise, professional title composition, teaching behavior patterns, and job suitability, providing a structured reference for subsequent resource optimization and allocation.

[0047] In the stage of optimizing job matching, the clustering results of 12 types of teachers are mapped to the job requirements proposed by 15 schools to construct a teacher-job matching space. The improved wormhole behavior particle swarm optimization algorithm of the present invention is used to execute resource allocation. Based on the global optimal guidance and wormhole perturbation mechanism, the global nature and stability of resource allocation are improved. In the optimization process, comprehensive fitness indicators such as teaching performance and job fitness are introduced. After about 100 rounds of iteration, a preliminary teacher resource allocation plan is output, covering a total of 1,374 items including job assignment suggestions and personnel deployment suggestions.

[0048] After the preliminary allocation plan is implemented, the system tracks and collects the implementation effects of the plan, including teaching satisfaction, classroom evaluations, lists of transferred-in and transferred-out teachers, and job dynamic data. The system feedback data shows after two months.

[0049] Table 1 Comparison table of core indicators before and after the implementation of teacher resource optimization based on big data analysis ; In terms of structural rationality, the teacher job vacancy rate has dropped from the original 14.2% to 5.7%, and the job duplication and concurrent job rate has dropped from 36.2% to 19.4%. This shows that the present invention performs excellently in the precise allocation of job resources and the balance of teachers' burdens, significantly alleviating the resource imbalance problem of the coexistence of "vacant jobs without suitable candidates" and "repeated teaching" in some schools.

[0050] In terms of configuration efficiency, the improved wormhole behavior particle swarm optimization algorithm adopted by the present invention has significantly improved the convergence speed of the optimization process. The average number of convergence iterations has decreased from 180 to 98, a decrease of 45.6%, showing the efficient search ability and good solution convergence characteristics of the optimization model in a large-scale variable space, ensuring the output of a highly feasible teacher resource configuration plan within a controllable time.

[0051] In terms of post matching and execution response, the VaDE clustering model constructed by the present invention realizes the deep modeling and clustering classification of the teacher structure. The average classification confidence of the clustering model reaches 91.5%, significantly improving the accuracy of teacher feature recognition and providing a stable structural basis for subsequent matching. At the same time, the adoption rate of the post transfer suggestion reaches 88.7%, indicating that the method has good practicability and feasibility, and the generated configuration suggestions are highly executable.

[0052] In terms of teaching effect, the satisfaction score of teacher configuration has increased from 3.76 to 4.29, and the classroom feedback score after post transfer has increased from 3.84 to 4.37, with increases of 14.1% and 13.8% respectively, verifying that the present invention not only optimizes the structural configuration, but also significantly enhances the post adaptability and teaching site feedback, truly realizing an intelligent allocation closed-loop oriented by teaching results.

[0053] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered within the protection scope of the present invention.

Claims

1. A method for optimizing and dynamically configuring the teacher team structure based on big data analysis, characterized in that: The steps include: S1. Collect teachers’ personal information and school job demand data, and build an original data set based on teachers’ personal information; S2, preprocessing the original data set to form a standardized data set; S3, extracting teacher feature vectors based on the standardized data set, building a VaDE clustering model, clustering teacher feature vectors through the variational autoencoder and Gaussian mixture model of the VaDE clustering model, and outputting the teacher structure clustering results and the corresponding category probability distribution; S4. Construct a teacher resource allocation optimization model, match the teacher structure clustering results with the job demand data, use the improved wormhole behavior particle swarm optimization algorithm to perform global search, and generate a preliminary teacher resource allocation plan; S5. Execute the preliminary teacher resource allocation plan and update the standardized data set and job demand data according to the results of the plan execution; S6. Feed the updated data back to the VaDE clustering model and the teacher resource allocation optimization model, adjust the teacher structure clustering results and the teacher resource allocation plan, and generate a dynamically updated configuration plan; S7. Output the final teacher resource allocation plan to the education management system.

2. According to the method of optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1, it is characterized in that: The personal information includes basic information, teaching behavior, professional title level, teaching subjects, job experience, teaching and research records and classroom feedback information.

3. According to the method of optimizing and dynamically configuring the teacher team structure based on big data analysis in claim 1, it is characterized in that: The job demand data include job setting data, subject course structure, student size, teaching load standards and job ability requirements.

4. According to the method of optimizing and dynamically configuring the teaching staff structure based on big data analysis according to claim 1, it is characterized in that: The preprocessing includes format conversion, missing value filling, outlier removal, normalization and standardization.

5. According to the method of optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1, it is characterized in that: The S3 specifically includes: S31. Based on the standardized data set, extract the feature vector of each teacher and form an input matrix with all feature vectors ,in represents the total number of teachers, Indicates The characteristic vector of each teacher; S32, constructing a VaDE clustering model, inputting the input matrix into the encoder network of the variational autoencoder, and obtaining the mean vector and logarithmic variance vector of the potential representation vector; S33, using the reparameterization technique, the mean vector and logarithmic variance vector of the potential representation vector output by the encoder network are mapped to the potential representation vector, and the posterior distribution of the potential representation vector is obtained: ； in, Indicates The latent representation vector of each teacher is Indicates The mean vector of the latent representation vectors of the teachers, Indicates The logarithmic variance vector of the latent variables of the teachers, represents vector element-wise multiplication, represents a random noise vector sampled from a standard Gaussian distribution, represents the diagonal covariance matrix, represents the posterior distribution of the latent representation vector, generated by the encoder network, represents a Gaussian distribution, represents the training parameters of the encoder network; S34. Establish a Gaussian mixture model as a prior distribution in the latent space; S35, inputting the potential representation vector into the decoder network to generate a reconstruction vector, and calculating the reconstruction probability distribution of the potential representation vector belonging to each cluster; S36, jointly minimize the reconstruction error and the latent space distribution difference to construct a comprehensive loss function; S37. After training the VaDE clustering model to convergence, output each latent representation vector, the teacher structure clustering category probability distribution and the teacher structure clustering result.

6. The method for optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1 is characterized in that: The S4 specifically includes: S41. Represent the teacher structure clustering results as a set ,in Indicates The cluster category to which the teacher belongs, and the job demand data is represented as a set ,in Indicates Demand data for each position, Indicates the total number of positions, according to the set and Construct the matching optimization problem between teachers and positions; S42, initialize the improved wormhole behavior particle swarm optimization algorithm, and represent each particle as a position vector ,in Represents particles Are teachers assigned to positions? , initialize the particle swarm population , each particle corresponds to a teacher resource configuration scheme, and the improved points of the improved wormhole behavior particle swarm optimization algorithm include wormhole perturbation mechanism, global optimal guidance mechanism and adaptive jump probability adjustment mechanism; S43. In each round of iteration, the particle position and velocity are updated according to the global optimal guidance mechanism; ； ； in, Indicates Round Iteration Particle In the The speed at which positions are updated, represents the inertia weight, and represents the learning factor, and Represents a random number in the range [0,1]. Indicates Round Iteration Particle In the The current speed of each position, represents the best historical position of the particle, Indicates Round Iteration Particle In the Current location of the job, represents the current global optimal position, Indicates Round Iteration Particle In the The updated position of each position; S44, introduce the wormhole perturbation mechanism in each round of iteration, with the wormhole perturbation probability Some particles are selected to introduce disturbances. The wormhole disturbance mechanism simulates the long-distance jump of particles "traversing" the search space by introducing nonlinear and globally guided jump operations, thereby enhancing global optimization: ； in, Indicates Round Iteration Particle In the The updated position of the job, Indicates Round Iteration Particle In the Current location of the job, represents the disturbance intensity coefficient, represents a standard Gaussian distributed random variable with mean 0 and variance 1; S45. Set an adaptive jump probability adjustment mechanism to dynamically adjust the wormhole disturbance probability with the number of iterations: ； in, Indicates The probability of wormhole perturbation in round iterations, represents the initial jump probability value, represents a natural constant, represents the exponential decay coefficient; S46. After each round of iteration, compare the fitness function values ​​of all particles and select the particle with the largest fitness value. ,in represents the particle population set, represents the fitness function value, The corresponding particle position vector is the optimal teacher resource allocation scheme, and the particle allocation result is output. ,in Indicates that the Teachers are assigned to Positions, Indicates no allocation, and finally generates a preliminary teacher resource allocation plan.

7. The method for optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1 is characterized in that: The results of the program implementation include teaching feedback data, job adjustment data and teacher mobility records.

8. The method for optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1 is characterized in that: The S5 specifically includes: S51. Implement the preliminary teacher resource allocation plan, record the implementation of the allocation plan in each position, and collect the plan implementation result data; S52. Constructing a teaching feedback data matrix ,in Indicates Teacher in Teaching feedback scores for each position; S53. Constructing a data set for job adjustment ,in Indicates position The adjustment status, Indicates that the position is cancelled, 0 indicates no change, Indicates newly added positions; S54. Constructing a teacher mobility record matrix ,in Indicates Teachers transferred to Positions, Indicates that from Positions transferred out, Indicates no job changes; S55. Update the teacher's feature vector based on the teaching feedback data, and update the job demand data based on the job adjustment data and teacher flow records; S56. Replace the original standardized data set and job data with the updated teacher feature vector and job requirement data.

9. The method for optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 8 is characterized in that: The teaching feedback score is composed of student satisfaction, teaching completion rate and classroom evaluation mean.

10. The method for optimizing and dynamically configuring the teacher team structure based on big data analysis according to claim 1 is characterized in that: The S6 specifically includes: S61, using the updated standardized data set and job demand data as input data to dynamically adjust the VaDE clustering model and the teacher resource allocation optimization model respectively; S62, input the updated teacher feature data into the VaDE clustering model, and output the latest clustering result of the teacher structure; S63, re-matching the updated clustering results with the updated job demand data, building a dynamic matching relationship between teachers and positions, and re-initializing the particle swarm in the teacher resource allocation optimization model based on the new matching relationship; S64, execute the resource allocation optimization process of wormhole perturbation mechanism, global optimal guidance mechanism and adaptive jump probability adjustment mechanism, continuously update particle status and evaluate configuration adaptation during the optimization process, and after reaching the maximum number of iterations or meeting the convergence conditions, select the particle position vector with the best fitness to generate a dynamically updated teacher resource allocation plan.

Images