Intelligent education management system based on artificial intelligence

Abstract

The application discloses an intelligent education management system based on artificial intelligence, which comprises a multi-source data acquisition module, an education direction selection module, an education path optimization module and an intelligent training education management module. The application relates to the technical field of training education data analysis, in particular to an intelligent education management system based on artificial intelligence. The scheme innovatively combines a student situation adaptive learning strategy, selects an education direction suitable for students in real time through the education direction selection module, effectively improves the adaptability and individualization of education content, introduces an improved bidirectional long short-term memory network to extract education time sequence dynamic characteristics, effectively improves the accuracy of education direction recommendation, and enhances the adaptability of the model to dynamic changes in the student situation; a particle optimization algorithm based on a group adaptability strategy and a score order dynamic memory mechanism is improved, the accuracy of the education path is improved, the education path recommendation is highly matched with the student demand, and the intelligentization of education management is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of training and education data analysis, in particular to an intelligent education management system based on artificial intelligence. BACKGROUND

[0002] The intelligent education management system based on artificial intelligence is a training and education management system for project personnel of municipal engineering design institutes. Through the collection, fusion and analysis of multi-source data by using artificial intelligence and big data technology, intelligent training and education management of training personnel is realized, and scientific and data-driven decision support is provided for training and education managers.

[0003] However, the traditional education management system has the technical problems of arranging based on fixed education content, ignoring the changes of the instant situation and task orientation of the students, so that the students cannot obtain adaptive learning content when facing the dynamic changes in actual work, and further affecting the education effect; the existing education direction recommendation model has the technical problems of isolated use of student characteristics and adoption of static data, ignoring the change of the situation of the students over time, so that the education direction recommendation result is inaccurate; the traditional education path optimization depends on fixed education path planning, lacks dynamic evaluation of education effect and personalized education path adjustment, resulting in low accuracy of education path recommendation and poor education effect. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides an intelligent education management system based on artificial intelligence, which is aimed at the technical problems of the traditional education management system that arranges fixed education content, ignores the changes of the instant situation and task orientation of the students, and thus leads to the students being unable to obtain adaptive learning content when facing dynamic changes in actual work, thereby affecting the education effect. The present scheme innovatively combines the student situation adaptive learning strategy, selects the education direction suitable for the students in real time through the education direction selection module, ensures that the training content is highly consistent with the current situation of the students, effectively improves the adaptability and individualization of the education content, enhances the learning effect of the students in the actual scene, realizes the flexibility and pertinence of education management, and improves the intelligent level of the education management system. In view of the technical problems that the student features are used in isolation and static data is used in the existing education direction recommendation model, the situation of the students changes over time is ignored, and thus the education direction recommendation result is inaccurate, the present scheme innovatively introduces an improved bidirectional long short-term memory network to extract the education timing dynamic features, improves the traditional long short-term memory network structure by fusing the causal convolution, up-down projection strategy and multilayer perceptron structure, the causal convolution can capture short-term timing dependence and avoid information leakage, the up-down projection strategy optimizes the dimension of the timing data and reduces the computational complexity, the multilayer perceptron structure enhances the analysis ability of the nonlinear timing data, can effectively improve the accuracy of the education direction recommendation, enhances the adaptability of the model to the dynamic changes of the situation of the students, makes the education path recommendation be able to respond to the needs and task changes of the students in time and accurately, and realizes the individualization and intelligentization of the education direction recommendation. In view of the technical problems that the traditional education path optimization depends on fixed education path planning, lacks dynamic evaluation of education effect and individualized education path adjustment, and thus leads to low accuracy of education path recommendation and poor education effect, the present scheme innovatively establishes an education effect prediction model, takes the output result thereof as the fitness value of the search individual, uses it to optimize the parameters of the education path, and combines the particle optimization algorithm improved based on the group adaptability strategy and the fractional order dynamic memory mechanism to dynamically search and optimize the education path parameters, so as to find the optimal education path. The group adaptability strategy makes the optimization process keep balance between exploration and utilization, improves the search efficiency and accuracy, and the fractional order dynamic memory mechanism enhances the retention ability of the algorithm to historical information, so that the particle optimization algorithm can better adapt to the changing needs and learning situation of the students, improves the individualization and accuracy of the education path. Through the above method, the accuracy of the education path is significantly improved, the education path recommendation is highly consistent with the needs of the students, the flexibility and adaptability of the education path optimization are enhanced, and the learning effect of the students under different paths can be evaluated in real time, thereby realizing the intelligentization of the education management.

[0005] The technical scheme adopted by the present application is as follows: The intelligent education management system based on artificial intelligence provided by the present application comprises a multi-source data acquisition module, an education direction selection module, an education path optimization module and an intelligent training education management module.

[0006] The multi-source data acquisition module specifically obtains education management optimization data through data acquisition and data optimization operation.

[0007] The education direction selection module is used for dynamically adjusting and recommending the education direction most suitable for the current situation of the student according to the real-time situation of the student and the task-oriented needs, specifically, an input matrix is first constructed, and an education environment matching feature and an education time sequence dynamic feature are extracted by using a graph convolution network and an improved bidirectional long short-term memory network respectively, then the dynamic attention mechanism is used to perform feature fusion on the two features to generate education direction recommendation fusion features, and finally the education direction recommendation results are generated through full connection layer and Softmax function processing, so as to complete the education direction recommendation model, the model is trained, and finally the target data is input into the trained education direction recommendation model to obtain the real-time education direction recommendation result of the student.

[0008] The education path optimization module is used for optimizing the education path of the student in combination with the real-time education direction of the student; specifically, an education effect prediction model is first constructed and trained, and the output result of the model is used as the objective function value of the optimization algorithm, then an improved optimization algorithm is used to optimize and search the education path parameters to obtain the optimal education path parameter combination, and finally the optimal education path and the education effect prediction value of the optimal education path are obtained.

[0009] The intelligent training education management module specifically realizes the comprehensive intelligent management of the training education of the student in combination with the real-time education direction recommendation result of the student, the optimal education path and the education effect prediction value.

[0010] Further, the multi-source data acquisition module specifically acquires education management raw data by data acquisition from an education learning platform and a work task platform, and obtains education management optimization data by data optimization processing on the education management raw data; the education management raw data comprises reference education direction selection data, reference education effect prediction data and target education direction selection data; the data optimization processing specifically comprises data cleaning, data standardization and data coding processing to obtain the education management optimization data.

[0011] Further, the education direction selection module specifically comprises the following steps:

[0012] The education direction recommendation model is established, comprising the following steps:

[0013] An input matrix is constructed, specifically a static education feature matrix is constructed by education direction selection data in the education management optimization data and a time sequence dynamic education feature matrix ;

[0014] An education environment matching feature is extracted, specifically mutual information between each pair of features in the static education feature matrix is calculated first, and a sparse threshold is set , so as to construct a sparse adjacency matrix, and normalization is performed on the sparse adjacency matrix, and then a feature is extracted through an L-layer graph convolution network, each layer of the graph convolution network is embedded with a residual connection, and finally an education environment matching feature is obtained :

[0015] ;

[0016] In the formula, indicates the mutual information value between the feature and the feature , indicates the ith static education feature, indicates the jth static education feature, indicates an element in the sparse adjacency matrix;

[0017] An education time sequence dynamic feature is extracted, specifically the time sequence dynamic education feature matrix is input into an improved bidirectional long short-term memory network to obtain an education time sequence dynamic feature

[0018] The construction of the improved bidirectional long short-term memory network specifically includes the following steps:

[0019] An education time sequence causal perception convolution layer is constructed, specifically a one-dimensional causal convolution operation is performed to process, the time sequence dynamic education feature at each time step is combined with the time sequence dynamic education features at the previous time steps, and a short-term time sequence causal feature is obtained

[0020] A time sequence feature dimension optimization layer is constructed, specifically the short-term time sequence causal feature is raised to a high-dimensional space through up-projection, and then it is reduced back to the original dimension through down-projection, and a down-projection output time sequence causal feature is obtained

[0021] A nonlinear time sequence adjustment layer is constructed, specifically the nonlinear changes in the down-projection output time sequence causal feature are processed by introducing a multilayer perception structure, and a nonlinear time sequence dynamic education feature is obtained

[0022] An improved long short-term memory network internal architecture is constructed, specifically the Sigmoid function is replaced by an exp activation function in the input gate and the forgetting gate, and the candidate cell unit state calculation method is modified to be processed by introducing a multilayer perception structure, and the formula is as follows ​

[0023] ;

[0024] In the formula, represents the output of the forget gate, represents the output of the input gate, represents the current time cell state, represents the previous time cell state, represents the output of the output gate, represents the weight matrix corresponding to the output gate processing, represents the bias term parameter of the output gate, represents the hidden state of the t-th time step, represents the hyperbolic tangent function, represents the Sigmoid function, represents a nonlinear time series dynamic education feature.

[0025] A bidirectional output layer is constructed, specifically by concatenating the forward direction hidden state and the reverse direction hidden state according to the feature dimension to obtain an education time series dynamic feature ;

[0026] Adaptive feature fusion, specifically, first compress the education time series dynamic feature and the education environment matching feature into one-dimensional vectors through a global average pooling operation to generate a time series dynamic feature intensity vector and an environment matching feature intensity vector, then calculate the attention scores of the two intensity vectors through a fully connected layer, then obtain the attention weights of the two through Softmax normalization, and finally according to the calculated attention weights, the education time series dynamic feature and the education environment matching feature are weighted and fused to obtain an education direction recommendation fusion feature ;

[0027] Education direction recommendation output, specifically, the education direction recommendation fusion feature is mapped through a fully connected layer and the probability distribution of each education direction is calculated through a Softmax function, and the class corresponding to the maximum probability is selected as the final output result to obtain an education direction recommendation result.

[0028] Education direction recommendation model training, specifically, the reference education direction selection data is used as training data to train the recommendation model, and finally a trained education direction recommendation model is obtained.

[0029] Target student education direction real-time recommendation, specifically, the target education direction selection data is input into the trained education direction recommendation model to obtain a student real-time education direction recommendation result.

[0030] Further, the education path optimization module specifically includes the following steps:

[0031] The education effect prediction model is constructed and trained, specifically, an education effect prediction model is constructed through a multilayer perception neural network, and reference education effect prediction data is used as training data of the education effect prediction model to perform prediction model training, and a trained education effect prediction model is obtained;

[0032] The education path parameter boundary is set, specifically, the dynamic boundary interval of the education learning time and the education completion standard is generated according to the real-time education direction recommendation result of the student, combined with the basic information of the student;

[0033] The education path parameter is optimized, specifically, an improved optimization algorithm is used to dynamically optimize and search the education path parameter, and the optimal education path parameter combination of the student in the recommended education direction is obtained; including the following steps:

[0034] The particle population is initialized, specifically, the education path parameter is used as a particle individual position vector in the optimization algorithm, and the position vectors of M particle individuals are randomly generated in combination with the dynamic boundary interval of the education learning time and the education completion standard , so as to complete the initialization of the particle population;

[0035] The particle fitness value is calculated, specifically, the education path data represented by the particle, the real-time education direction recommendation result of the student and the basic information of the student are combined and input into the trained education effect prediction model to obtain an education effect prediction value as the particle fitness value;

[0036] The inertia weight is calculated, specifically, the inertia weight is adaptively adjusted through a group fitness strategy; the formula used is as follows:

[0037] ;

[0038] In the formula, represents the inertia weight of the i-th particle, represents the maximum value of the inertia weight, represents the minimum value of the inertia weight, represents the fitness value of the i-th particle, represents the average fitness value of the particle population, represents the optimal fitness value of the particle individual;

[0039] The particle is updated, specifically, the particle speed is updated based on a fractional order dynamic memory mechanism, and then the position is updated according to the updated particle speed; the formula used is as follows:

[0040] ;

[0041] ;

[0042] In the formula, denotes the velocity of the i-th particle in the j-th iteration, denotes the velocity of the i-th particle in the j-th iteration, denotes the position of the i-th particle in the j-th iteration, denotes the local optimal position of the particle, denotes the global optimal position of the particle, and denotes a random number in the range of [0, 1], denotes the position of the i-th particle in the j-th iteration, and denote the individual learning factor and the group learning factor, respectively, denotes the velocity of the i-th particle in the j-th iteration, denotes the velocity of the i-th particle in the j-th iteration, denotes the velocity of the i-th particle in the j-th iteration, denotes the velocity of the i-th particle in the j-th iteration;

[0043] particle optimal position updating, specifically re-evaluating the fitness value of all updated particles, and comparing the fitness value of the current particle with the global optimal position of the current particle, if the fitness value of the current particle is better, updating the global optimal position of the particle;

[0044] particle search termination, specifically when the particle fitness value is higher than the fitness threshold or the maximum number of iterations is reached, terminating the search and obtaining the global optimal position of the particle; the global optimal position of the particle specifically refers to the optimal education path parameter combination;

[0045] optimal education path output, specifically generating and outputting the optimal education path of the student in the education recommendation direction based on the optimal education path parameter combination, and obtaining the education effect prediction value of the optimal education path through the fitness value of the global optimal position of the particle in the optimization algorithm.

[0046] Further, the intelligent training and education management module specifically provides personalized education direction recommendation and optimization for the student by combining the real-time education direction recommendation result of the student, the optimal education path and the education effect prediction value, thereby realizing comprehensive intelligent management of the training and education of the student.

[0047] The above scheme has the following beneficial effects:

[0048] ​​​​​​​​(1) In view of the technical problem that the traditional education management system is based on fixed education content arrangement, ignores the instant situation and task-oriented changes of students, and thus leads to the fact that students cannot obtain adaptive learning content when facing dynamic changes in actual work, thereby affecting the education effect, the scheme innovatively combines the student situation adaptive learning strategy, selects the education direction suitable for the students in real time through the education direction selection module, ensures that the training content is highly consistent with the current situation of the students, effectively improves the adaptability and individualization of the education content, enhances the learning effect of the students in the actual scene, realizes the flexibility and pertinence of education management, and improves the intelligent level of the education management system.

[0049] (2) In view of the technical problem that the existing education direction recommendation model uses student features and static data in isolation, ignores the change of the student situation over time, and thus leads to inaccurate education direction recommendation results, the scheme innovatively introduces an improved bidirectional long short-term memory network to extract education time sequence dynamic features, improves the traditional long short-term memory network structure by fusing causal convolution, up-down projection strategy and multi-layer perceptron structure, the causal convolution can capture short-term time sequence dependence and avoid information leakage, the up-down projection strategy optimizes the dimension of time sequence data and reduces the computational complexity, and the multi-layer perceptron structure enhances the analysis ability of nonlinear time sequence data, which can effectively improve the accuracy of education direction recommendation, enhance the adaptability of the model to the dynamic change of the student situation, make the education path recommendation timely and accurately respond to the needs and task changes of the students, and realize the individualization and intelligentization of education direction recommendation.

[0050] (3) In view of the technical problem that the traditional education path optimization relies on fixed education path planning, lacks dynamic evaluation of education effect and individualized education path adjustment, and thus leads to low accuracy of education path recommendation and poor education effect, the scheme innovatively establishes an education effect prediction model, takes the output result of the model as the fitness value of the search individual to optimize the parameters of the education path, and combines a particle optimization algorithm improved based on group adaptability strategy and fractional order dynamic memory mechanism to dynamically search and optimize the education path parameters, so as to find the optimal education path, the group adaptability strategy makes the optimization process keep balance between exploration and utilization, improves the search efficiency and accuracy, and the fractional order dynamic memory mechanism enhances the retention ability of the algorithm to historical information, so that the particle optimization algorithm can better adapt to the changing needs and learning situation of the students, improves the individualization and accuracy of the education path, and through the above method, the accuracy of the education path is significantly improved, the education path recommendation is highly consistent with the needs of the students, the flexibility and adaptability of the education path optimization are enhanced, the learning effect of the students in different paths can be evaluated in real time, and thus the intelligentization of education management is realized. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A module schematic diagram of the intelligent education management system based on artificial intelligence provided by the present application is provided.

[0052] Figure 2 A flowchart schematic diagram of the education direction selection module is provided.

[0053] Figure 3 A flowchart schematic diagram of the establishment of the education direction recommendation model in the education direction selection module is provided.

[0054] Figure 4 A flowchart schematic diagram of the construction of the improved bidirectional long short-term memory network is provided.

[0055] Figure 5 A flowchart schematic diagram of the education path optimization module is provided.

[0056] Figure 6 A flowchart schematic diagram of the education path parameter optimization in the education path optimization module is provided.

[0057] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, which is used together with embodiments of the present application to explain the present application, and do not constitute a limitation of the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0059] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the systems or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0060] Embodiment one, refer to Figure 1 The intelligent education management system based on artificial intelligence provided by the present application comprises a multi-source data acquisition module, an education direction selection module, an education path optimization module and an intelligent training education management module.

[0061] The multi-source data acquisition module specifically obtains education management optimization data through data acquisition and data optimization operation, and sends the data to the education direction selection module and the education path optimization module.

[0062] The education direction selection module receives the data sent by the multi-source data acquisition module, and dynamically adjusts and recommends the education direction most suitable for the current situation of the student according to the real-time situation and task-oriented needs of the student, specifically, first, an input matrix is constructed, and an education environment matching feature and an education time sequence dynamic feature are extracted by using a graph convolution network and an improved bidirectional long short-term memory network respectively, then a dynamic attention mechanism is used to fuse the two features to generate an education direction recommendation fusion feature, and finally, the education direction recommendation fusion feature is processed by a full connection layer and a Softmax function to generate an education direction recommendation result, so as to complete the education direction recommendation model, train the model, and input the target data into the trained education direction recommendation model to obtain the real-time education direction recommendation result of the student, and send the data to the education path optimization module and the intelligent training education management module;

[0063] The education path optimization module receives the data sent by the multi-source data acquisition module and the education direction selection module, and optimizes the education path of the student in combination with the real-time education direction of the student, specifically, first, an education effect prediction model is constructed and trained, and the output result of the model is used as the objective function value of the optimization algorithm, then an improved optimization algorithm is used to optimize and search the education path parameters to obtain the optimal education path parameter combination, and finally, the optimal education path and the education effect prediction value of the optimal education path are obtained, and the data is sent to the intelligent training education management module;

[0064] The intelligent training education management module receives the data sent by the education direction selection module and the education path optimization module, and specifically, the real-time education direction recommendation result of the student, the optimal education path and the education effect prediction value are combined to realize comprehensive intelligent management of the training education of the student.

[0065] By performing the above operations, the technical problem of ignoring the changes of the real-time situation and task orientation of the student based on fixed education content arrangement in the traditional education management system is solved, so that the student cannot obtain adaptive learning content when facing dynamic changes in actual work, thereby affecting the education effect, the student situation adaptive learning strategy is innovatively combined, the education direction selection module is used to select the education direction suitable for the student in real time, the training content is highly matched with the current situation of the student, the adaptability and individualization of the education content are effectively improved, the learning effect of the student in the actual scene is enhanced, the flexibility and pertinence of the education management are realized, and the intelligent level of the education management system is improved.

[0066] Embodiment two, refer to Figure 1The embodiment is based on the above-mentioned embodiment, the multi-source data acquisition module, specifically through data collection from an education learning platform and a work task platform, obtains education management original data, and performs data optimization processing on the education management original data to obtain education management optimized data;

[0067] The education management original data includes reference education direction selection data, reference education effect prediction data, and target education direction selection data;

[0068] The data optimization processing is used to improve the accuracy, availability and consistency of the original data, thereby providing higher quality data support for education path recommendation, learning effect prediction and the like, and specifically includes data cleaning, data standardization and data coding processing to obtain education management optimized data;

[0069] The reference education direction selection data and the target education direction selection data both include student basic information, student current situation data and education environment data;

[0070] The student basic information includes age, gender, education, professional background, interest field, student stress level, project position, skill level, learning style, available education time and career goal;

[0071] The student current situation data provides detailed information about the student in terms of projects, tasks, work environment and the like, which helps to infer the challenges and needs currently faced by the student, thereby recommending appropriate training content and education direction for the student, including project type, project progress status, project task completion, project challenge, work task type, team communication method and team environment;

[0072] The reference education direction selection data further includes historical student education selection direction;

[0073] The student education selection direction includes technical skill direction, project management direction, business communication direction and enterprise management direction;

[0074] The reference education effect prediction data includes education learning direction, student basic information, education path data and historical education effect prediction value;

[0075] The education path data includes education difficulty, education learning time, education completion standard and education learning method;

[0076] The education effect prediction value is a quantitative index for measuring the potential learning effectiveness of the student in the education path, using a percentage system;

[0077] Data cleaning, used to ensure the accuracy and completeness of the original data, specifically by cleaning the original data of education management, removing invalid, duplicate, missing or abnormal data in the data, including removing duplicate data, handling missing data and removing outliers; the removal of duplicate data is specifically by comparing student records to delete redundant duplicate data, the handling of missing data is specifically by the mean filling method to complete the missing values in the original data; the removal of outliers is specifically by Z-Score algorithm to detect and remove extreme values and logical outliers in the original data;

[0078] Data standardization processing, used to eliminate the difference in dimension and range, and unify the dimension of numerical data, specifically by the min-max normalization method to map all continuous variables to the [0, 1] interval;

[0079] Data encoding processing, used to successfully convert category variables into numerical data, specifically using one-hot encoding method to encode the category fields in the original data, and converting discrete text and label variables into sparse numerical vectors.

[0080] Embodiment three, see Figure 1 、 Figure 2 、 Figure 3 and Figure 4 , this embodiment is based on the above embodiment, the education direction selection module, specifically including the following steps:

[0081] Establishing an education direction recommendation model, including the following steps:

[0082] Building an input matrix, used to provide structured input data to the education direction recommendation model, specifically by constructing a static education feature matrix and a time-series dynamic education feature matrix from the education direction selection data in the education management optimization data.

[0083] The static education feature matrix is specifically selected from the education direction selection data in the education management optimization data, and the feature fields that do not change with time are arranged according to the dimensions of student samples and static features, to obtain a static education feature matrix , wherein N is the number of students, and M is the total number of static features.

[0084] The time-series dynamic education feature matrix is specifically selected from the education direction selection data in the education management optimization data, and the feature fields that change with time are extracted according to the time sequence of each student at each time step, and arranged according to the dimensions of student samples, time steps and time-series features, to obtain a time-series dynamic education feature matrix , wherein is the number of time steps, and K is the total number of time-series features; the time step is every week.

[0085] Educational environment matching feature extraction is used to extract educational environment features that are strongly correlated with educational direction recommendations. It models the relationships between features using a graph convolutional network to generate feature vectors that reflect the suitability of educational resources to student needs. Specifically, it first calculates the mutual information between each pair of features in the static educational feature matrix and then sets a sparsity threshold. This constructs a sparse adjacency matrix, which is then normalized. Features are extracted using an L-layer graph convolutional network, where each layer embeds residual connections, ultimately yielding educational environment matching features. :

[0086] The educational environment matching feature is output by the last layer of the graph convolutional network; the formula used is as follows:

[0087] ;

[0088] ;

[0089] ;

[0090] In the formula, Representation of features and characteristics The mutual information value between them measures the strength of their association. This represents the i-th static educational feature. Let j represent the j-th static educational feature. The elements in the sparse adjacency matrix represent the features. and characteristics The correlation weight between them This represents the adjacency matrix with added self-loops. This represents a normalized sparse adjacency matrix. express The degree matrix, Indicates the first Features of layer output, Indicates the first Features of layer output, This represents the learning weight matrix of the l-th layer;

[0091] The extraction of educational time-series dynamic features is used to capture the time-series dynamic changes of learners during the educational process due to changes in tasks, project progress, and adjustment of interest areas. Specifically, the time-series dynamic educational feature matrix is ​​input into an improved bidirectional long short-term memory network to obtain educational time-series dynamic features.

[0092] The construction of the improved bidirectional long short-term memory network specifically includes the following steps:

[0093] The education time sequence causal perception convolution layer is constructed, which is used for focusing on analyzing the influence of the student project progress change in the last few months on the education direction recommendation, and ensuring that the model can process the short-term dynamic change of the student; specifically, the one-dimensional causal convolution operation is used to process the time sequence dynamic education features at each time step and the time sequence dynamic education features of the previous time steps, to obtain the short-term time sequence causal features; the formula is as follows:

[0094]

[0095] In the formula, represents the one-dimensional causal convolution operation, represents the short-term time sequence causal feature at the t time step, represents the convolution kernel weight, represents the time sequence dynamic education feature at the t time step, represents the causal perception convolution bias parameter;

[0096] The time sequence feature dimension optimization layer is constructed, which is used for reducing the calculation complexity while retaining the details of the time sequence features; specifically, the short-term time sequence causal feature dimension is raised to a high-dimensional space through up-projection, and then it is reduced back to the original dimension through down-projection, to obtain the down-projection output time sequence causal feature, and the formula is as follows:

[0097]

[0098]

[0099] In the formula, represents the up-projection output time sequence causal feature, which is used for raising the short-term time sequence causal feature to a high-dimensional space and retaining the fine-grained dynamic features, represents the down-projection output time sequence causal feature, which is used for reducing the high-dimensional feature back to a low-dimensional space and being consistent with the original time sequence feature dimension, and are the up-projection and down-projection weight matrices respectively, and are the bias parameters of the up-projection and down-projection respectively;

[0100] The nonlinear time sequence adjustment layer is constructed, which is used for processing the nonlinear change in the time sequence data, adjusting the time sequence weight, and ensuring that the learning direction of the student can be flexibly adapted when the task and interest change; specifically, the multilayer perception structure is introduced to process the nonlinear change in the down-projection output time sequence causal feature, to obtain the nonlinear time sequence dynamic education feature; the formula is as follows:

[0101]

[0102] In the formula, ​​​​​This represents the nonlinear time-series dynamic educational characteristics. and These represent the first-level high-dimensional projection weight matrix and the second-level dimension-restored weight matrix, respectively. This represents the first-level bias term. This indicates the second-level bias term. This indicates the processing of a multilayer perceptron structure;

[0103] An improved internal architecture for the Long Short-Term Memory (LSTM) network was constructed by replacing the Sigmoid function with the exp activation function in the input and forget gates, and by modifying the calculation method for candidate cell states to be processed through a multilayer perceptron structure. The formulas used are as follows:

[0104] ;

[0105] In the formula, This represents the output of the forget gate. Indicates the first Hidden state in time step and These represent the weight matrices for processing the input data and processing the hidden state, respectively, for the forget gate. The bias term parameter represents the forget gate. Indicates the output of the input gate. and These represent the weight matrices for processing the input data and processing the hidden state at the corresponding input gates, respectively. This represents the bias term parameter of the input gate. This indicates the current state of the cell unit. This indicates the state of the cell unit at the previous moment. This indicates the output of the output gate. This represents the weight matrix processed by the corresponding output gate. This represents the bias term parameter of the output gate. This represents the hidden state at time step t. Represents the hyperbolic tangent function. Represents the Sigmoid function;

[0106] Construct a bidirectional output layer, specifically by concatenating the forward and reverse hidden states along the feature dimension to obtain the dynamic features of the educational time series. The formula used is as follows:

[0107] ;

[0108] In the formula, This represents the hidden state in the positive direction at time step t. This represents the hidden state in the reverse direction at time step t. representing a concatenation operation;

[0109] Adaptive feature fusion is used to adaptively adjust the contribution weight of the education time sequence dynamic feature and the education environment matching feature according to the current education situation of the student, so as to ensure that the fused feature can accurately match the current recommendation demand. Specifically, first, the education time sequence dynamic feature and the education environment matching feature are compressed into one-dimensional vectors through a global average pooling operation to generate a time sequence dynamic feature intensity vector and an environment matching feature intensity vector. Then, the attention scores of the two intensity vectors are calculated through a fully connected layer, and then the attention weights of the two are obtained through Softmax normalization. Finally, the education time sequence dynamic feature and the education environment matching feature are weighted and fused according to the calculated attention weights to obtain an education direction recommendation fusion feature. The formula used is as follows:

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] In the formula, represents the environment matching feature intensity vector, represents the time sequence dynamic feature intensity vector, represents a global average pooling operation, represents the environment matching feature attention score, represents the time sequence dynamic feature attention score, represents the environment matching feature attention weight, represents the time sequence dynamic feature attention weight, represents the education direction recommendation fusion feature, and respectively represent the weight matrix for calculating the environment matching feature attention score and the time sequence dynamic feature attention score, and respectively represent the bias parameter for calculating the environment matching feature attention score and the time sequence dynamic feature attention score;

[0118] The education direction recommendation output is specifically that the education direction recommendation fusion features are mapped through a full connection layer and the probability distribution of each education direction is calculated through a Softmax function, the category corresponding to the maximum probability is selected as the final output result, and an education direction recommendation result is obtained; the formula used is as follows:

[0119]

[0120] In the formula, represents the weight matrix of the output full connection layer, represents the bias parameter of the output full connection layer, represents the probability distribution of each education direction;

[0121] The education direction recommendation model training is specifically that the reference education direction selection data is taken as training data, the recommendation model is trained, and finally a trained education direction recommendation model is obtained.

[0122] The recommendation model training is specifically that a cross-entropy loss function is taken as an optimization target in the training process, and a back propagation algorithm and a gradient descent optimization method are used to train and iteratively update the weight matrix and the bias parameter of the model. In the training process, the recommendation model parameters are continuously optimized through multiple rounds of iteration, and when a preset maximum training number or a loss function value converges to a set recommendation model iteration termination threshold, the iteration training is stopped.

[0123] The target student education direction real-time recommendation is specifically that the target education direction selection data is input into the trained education direction recommendation model, and a student real-time education direction recommendation result is obtained.

[0124] By performing the above operations, in view of the technical problem that the existing education direction recommendation model is used in isolation and static data is used, the student's situation is ignored, and the education direction recommendation result is inaccurate, the improved bidirectional long short-term memory network is innovatively introduced to extract the education time sequence dynamic feature, the traditional long short-term memory network structure is improved through the fusion of causal convolution, up-down projection strategy and multi-layer perception structure, the causal convolution can capture short-term time sequence dependence and avoid information leakage, the up-down projection strategy optimizes the dimension of the time sequence data and reduces the computational complexity, the multi-layer perception structure enhances the analysis ability of the nonlinear time sequence data, can effectively improve the accuracy of the education direction recommendation, enhances the adaptability of the model to the dynamic change of the student's situation, makes the education path recommendation respond to the needs and task changes of the student in time and accurately, and realizes the personalization and intelligentization of the education direction recommendation.

[0125] Embodiment four, refer to Figure 1 , Figure 5 and Figure 6 ​The embodiment is based on the above-mentioned embodiment, and the education path optimization module is used for predicting learning effects of the student under different education paths, and optimizing education path parameters by using a particle optimization algorithm, so as to provide the most suitable personalized learning path for the student; specifically comprising the following steps:

[0126] The education effect prediction model is constructed and trained, and is used for predicting learning effects of the student under a specific education direction and education path. Specifically, the education effect prediction model is constructed by using a multilayer perception neural network, and reference education effect prediction data is used as training data of the education effect prediction model to perform prediction model training, so as to obtain the trained education effect prediction model.

[0127] The prediction model training is specifically performed by using a mean square error as a loss function to calculate differences between prediction results and true values, and by using a back propagation algorithm and a gradient descent optimization method to iteratively update weight matrices and bias parameters of the model. In the training process, the model parameters are continuously optimized through multiple iterations. When a preset maximum training number or a loss function converges to a set prediction model iteration termination threshold, the iteration training is stopped.

[0128] The education path parameter boundary is set, and is used for setting dynamic boundary conditions for the to-be-optimized education path parameters, so as to ensure that the parameter search is performed within a reasonable allowable range. Specifically, according to the student real-time education direction recommendation result, the student basic information is combined, and a dynamic boundary interval of education learning time and education completion standard is generated accordingly.

[0129] The education path parameters include education difficulty, education learning time, education completion standard and education learning mode. The education difficulty and the education learning mode are category data, and do not need to be set boundaries. The education learning time and the education completion standard are continuous values, and need to be set boundary intervals. The dynamic boundary interval of the education learning time is specifically , wherein, represents a basic requirement time for completing education learning of the education direction, represents a maximum time amount that the student can invest; and the dynamic boundary interval of the education completion standard is specifically , wherein, represents a minimum standard of education completion, represents a highest standard of historical students completing the education direction;

[0130] The education path parameter optimization is specifically performed by using an improved optimization algorithm to dynamically optimize and search the education path parameters, so as to obtain an optimal education path parameter combination of the student in the recommended education direction; comprising the following steps:

[0131] ​​Particle swarm initialization specifically involves using the educational path parameters as the individual particle position vectors in the optimization algorithm, and randomly generating M individual particle position vectors based on the dynamic boundary interval of the educational learning time and educational completion criteria. This completes the initialization of the particle population;

[0132] Each individual particle represents a candidate combination of educational path parameters;

[0133] The particle fitness value calculation is used to evaluate the effectiveness of the educational path represented by each particle, thereby guiding the search process of the optimization algorithm. Specifically, it calculates the particle fitness value in the population, combines the educational path data represented by the particles, the real-time educational direction recommendation results of students, and the basic information of students, and inputs them into the post-training educational effect prediction model to obtain the educational effect prediction value, which is used as the particle fitness value.

[0134] The inertia weight is calculated by adaptively adjusting the inertia weight using a group adaptation strategy; the formula used is as follows:

[0135] ;

[0136] In the formula, This represents the inertial weight of the i-th particle. This represents the maximum value of the inertia weight. This represents the minimum inertial weight. This represents the fitness value of the i-th particle. This represents the average fitness value of the particle population. This represents the optimal fitness value of an individual particle.

[0137] Particle updates specifically involve updating particle velocity based on a fractional-order dynamic memory mechanism, followed by position updates based on the updated particle velocity; the formula used is as follows:

[0138] ;

[0139] ;

[0140] In the formula, It indicates that the i-th particle is in the... Dynamic memory speed in each iteration It indicates that the i-th particle is in the... Speed ​​in the next iteration It indicates that the i-th particle is in the... Position in the next iteration This represents the local optimal position of an individual particle. This represents the global optimal position of the particle. and Represents a random number in the range [0,1]. xi, i = 1, 2, 3,..., N xi, i = 1, 2, 3,..., N and respectively represent individual learning factor and group learning factor, vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N vi, i = 1, 2, 3,..., N

[0141] particle optimal position updating, specifically re-evaluating fitness value of all updated particles, and comparing fitness value of current particle with global optimal position of current particle, if fitness value of current particle is better, then updating global optimal position of particle;

[0142] particle search termination, specifically when fitness value of particle is higher than fitness threshold or reaches maximum iteration number, then terminating search and obtaining global optimal position of particle; the global optimal position of particle specifically refers to optimal education path parameter combination;

[0143] optimal education path output, specifically generating and outputting optimal education path of student in education recommendation direction based on optimal education path parameter combination, and obtaining education effect prediction value of the optimal education path through fitness value of the global optimal position of particle in optimization algorithm.

[0144] By performing the above operation, the optimization of the traditional education path depends on the fixed education path planning, and lacks dynamic evaluation of the education effect and individualized education path adjustment, which leads to the technical problems of low accuracy of education path recommendation and poor education effect. The present scheme innovatively establishes an education effect prediction model, uses the output results of the model as the fitness value of the search individual to optimize the parameters of the education path, and combines the particle optimization algorithm improved based on the group adaptability strategy and the fractional order dynamic memory mechanism to dynamically search and optimize the education path parameters, so as to find the optimal education path. The group adaptability strategy balances the exploration and utilization in the optimization process, improves the search efficiency and accuracy, and the fractional order dynamic memory mechanism enhances the retention ability of the algorithm to historical information, so that the particle optimization algorithm can better adapt to the changing needs and learning situation of the students, and improves the individualization and accuracy of the education path. Through the above method, the accuracy of the education path is significantly improved, the education path recommendation is highly matched with the needs of the students, the flexibility and adaptability of the education path optimization are enhanced, and the learning effect of the students in different paths can be evaluated in real time, so that the intelligentization of education management is realized.

[0145] In an embodiment, the intelligent training and education management module is specifically used to provide individualized education direction recommendation and optimization for the students by combining the real-time education direction recommendation result of the students, the optimal education path and the education effect prediction value, so as to realize the comprehensive intelligent management of the training and education of the students. Figure 1

[0146] The comprehensive intelligent management is specifically to dynamically adjust the education direction according to the real-time education direction recommendation result of the students, adapt to the needs and goals of the students, solve the current actual difficulties and problems of the students, generate individualized learning paths for the students according to the optimal education path, maximize the learning effect and training efficiency of the students, and finally evaluate the learning effect of the students in the path according to the education effect prediction value, so as to provide a basis for further optimization of the education path and ensure continuous improvement and optimization of the intelligent education scheme.

[0147] It should be noted that, in this document, the relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0148] ​While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary and that changes can be made in the embodiments without departing from the spirit and scope of the application.

[0149] The above description of the application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the application, without creative design, similar structure and embodiments of the technical solution should belong to the protection scope of the application.

Claims

1. An intelligent education management system based on artificial intelligence, characterized in that: The application relates to an intelligent training and education management system, which comprises a multi-source data acquisition module, an education direction selection module, an education path optimization module and an intelligent training and education management module. The multi-source data acquisition module is specifically used for obtaining education management optimization data through data acquisition and data optimization operation. The education direction selection module is used for dynamically adjusting and recommending an education direction most suitable for a current situation of a student according to real-time situation and task-oriented requirements of the student, specifically, an input matrix is first constructed, and an education environment matching feature and an education time sequence dynamic feature are extracted by adopting a graph convolution network and an improved bidirectional long short-term memory network respectively, then the two features are fused by using a dynamic attention mechanism to generate education direction recommendation fusion features, finally, the education direction recommendation fusion features are processed by a full connection layer and a Softmax function to generate an education direction recommendation result, so that an education direction recommendation model is established, the model is trained, and finally, target data is input into the trained education direction recommendation model to obtain a real-time education direction recommendation result of the student. The education direction recommendation model is established by the following steps: Constructing an input matrix, specifically a static education feature matrix by education direction selection data in the education management optimization data and a time-series dynamic education feature matrix ; The education environment matching feature extraction specifically comprises the following steps: firstly, mutual information between each pair of features in a static education feature matrix is calculated, a sparsity threshold is set , so as to construct a sparse adjacency matrix, and the sparse adjacency matrix is normalized, then a feature is extracted through an L-layer graph convolution network, each layer of the graph convolution network is embedded with a residual connection, and finally, an education environment matching feature is obtained : ； wherein representing a feature and a feature between the features, represents the ith static educational feature, represents the jth static educational feature, represents an element in the sparse adjacency matrix; education time sequence dynamic feature extraction; The adaptive feature fusion specifically comprises: firstly, compressing the education time sequence dynamic feature and the education environment matching feature into one-dimensional vectors through a global average pooling operation to generate a time sequence dynamic feature intensity vector and an environment matching feature intensity vector; then, calculating attention scores of the two intensity vectors through a fully connected layer; then, obtaining attention weights of the two through Softmax normalization; and finally, weighting and fusing the education time sequence dynamic feature and the education environment matching feature according to the calculated attention weights to obtain an education direction recommendation fusion feature ; education direction recommendation output, specifically, the education direction recommendation fusion features are mapped by a full connection layer and the probability distribution of each education direction is calculated by a Softmax function, a category corresponding to the maximum probability is selected as a final output result, and an education direction recommendation result is obtained. The improved bidirectional long short-term memory network is improved by introducing a causal convolution, applying an up-down projection strategy and combining a multilayer perception structure. The education path optimization module is used for optimizing an education path of a student in combination with a real-time education direction of the student, specifically, an education effect prediction model is first constructed and trained, the output result of the model is used as a target function value of an optimization algorithm, then a group fitness strategy is adopted to adaptively adjust an inertia weight and a fractional order dynamic memory mechanism is adopted to improve a particle swarm optimization algorithm, an education path parameter is optimized and searched, an optimal education path parameter combination is obtained, and finally, an optimal education path and an education effect prediction value of the optimal education path are obtained. The intelligent training and education management module is specifically used for realizing comprehensive intelligent management of student training and education in combination with a real-time education direction recommendation result of a student, an optimal education path and an education effect prediction value. 2.The intelligent education management system based on artificial intelligence according to claim 1, characterized in that: The education direction selection module specifically comprises the following steps: establishing an education direction recommendation model; education direction recommendation model training, specifically, reference education direction selection data is used as training data to train the recommendation model, and finally, a trained education direction recommendation model is obtained; target student education direction real-time recommendation, specifically, target education direction selection data is input into the trained education direction recommendation model to obtain a real-time education direction recommendation result of a student. 3.The intelligent education management system based on artificial intelligence according to claim 1, characterized in that: The education time sequence dynamic feature extraction is specifically used for inputting a time sequence dynamic education feature matrix into the improved bidirectional long short-term memory network to obtain an education time sequence dynamic feature. The improved bidirectional long short-term memory network is constructed by the following steps: The time sequence causal perception convolution layer is constructed, specifically, time sequence dynamic education features of each time step are combined with time sequence dynamic education features of previous time steps through one-dimensional causal convolution operation to obtain short-term time sequence causal features; The time sequence feature dimension optimization layer is constructed, specifically, the dimension of the short-term time sequence causal features is lifted to a high-dimensional space through up-projection, and then the dimension is reduced back to the original dimension through down-projection to obtain down-projection output time sequence causal features; The nonlinear time sequence adjustment layer is constructed, specifically, the nonlinear changes in the down-projection output time sequence causal features are processed through the introduction of a multilayer perception structure to obtain nonlinear time sequence dynamic education features; The improved long short-term memory network internal architecture is constructed, specifically, the Sigmoid function is replaced with an exp activation function in the input gate and the forgetting gate, and the candidate cell state calculation method is modified to be processed through the introduction of a multilayer perception structure, and the formula used is as follows: ； In the formula, This represents the output of the forget gate. Indicates the output of the input gate. This indicates the current state of the cell unit. This indicates the state of the cell unit at the previous moment. This indicates the output of the output gate. This represents the weight matrix processed by the corresponding output gate. This represents the bias term parameter of the output gate. This represents the hidden state at time step t. Represents the hyperbolic tangent function. This represents the Sigmoid function. This represents the nonlinear time-series dynamic educational characteristics. This indicates the temporal causal characteristics of the downprojection output. Indicates the first The temporal dynamics of education at the time step; A bidirectional output layer is constructed, specifically, the forward direction hidden state and the reverse direction hidden state are spliced according to a feature dimension to obtain an educational time sequence dynamic feature . 4.The intelligent education management system based on artificial intelligence according to claim 1, characterized in that: The education path optimization module specifically includes the following steps: The education effect prediction model is constructed and trained, specifically, the education effect prediction model is constructed through a multilayer perception neural network, and the reference education effect prediction data is used as the training data of the education effect prediction model for model training to obtain the trained education effect prediction model; The education path parameter boundary is set, specifically, the dynamic boundary interval of the education learning time and the education completion standard is generated according to the real-time education direction recommendation result of the student, combined with the basic information of the student; Education path parameter optimization; The optimal education path output, specifically, based on the optimal education path parameter combination, the optimal education path of the student in the education recommendation direction is generated and output, and the fitness value of the global optimal position of the particle in the optimization algorithm is obtained to obtain the education effect prediction value of the optimal education path. 5.The intelligent education management system based on artificial intelligence according to claim 4, characterized in that: The education path parameter optimization specifically includes the following steps: The particle population is initialized, specifically, the education path parameters are taken as the particle individual position vectors in the optimization algorithm, and the position vectors of M particle individuals are randomly generated in combination with the dynamic boundary interval of the education learning time and the education completion standard , thereby completing the initialization of the particle population; Particle fitness value calculation, specifically, the particle fitness value in the population is calculated, the education path data represented by the particle, the real-time education direction recommendation result of the student, and the basic information of the student are combined and input into the trained education effect prediction model to obtain the education effect prediction value as the particle fitness value; Inertia weight calculation, specifically, the inertia weight is adaptively adjusted through the group adaptability strategy; the formula used is as follows: ； wherein, represents the inertia weight of the i-th particle, represents the maximum inertia weight, represents the minimum inertia weight, represents the fitness value of the i-th particle, represents the average fitness value of the particle population, represents the optimal fitness value of the particle individual; Particle update, specifically, the particle velocity is updated based on the fractional order dynamic memory mechanism, and then the position is updated according to the updated particle velocity; the formula used is as follows: ； ； In the formula, It indicates that the i-th particle is in the... Dynamic memory speed in each iteration It indicates that the i-th particle is in the... Speed ​​in the next iteration It indicates that the i-th particle is in the... Position in the next iteration This represents the local optimal position of an individual particle. This represents the global optimal position of the particle. and Represents a random number in the range [0,1]. It indicates that the i-th particle is in the... Position in the next iteration and These represent individual learning factors and group learning factors, respectively. It indicates that the i-th particle is in the... Speed ​​in the next iteration It indicates that the i-th particle is in the... Speed ​​in the next iteration It indicates that the i-th particle is in the... Speed ​​in the next iteration It indicates that the i-th particle is in the... Speed ​​in the next iteration; Particle optimal position update, specifically, the fitness value of all updated particles is re-evaluated, and the fitness value of the current particle is compared with the global optimal position of the current particle, if the fitness value of the current particle is better, the global optimal position of the particle is updated; Particle search termination, specifically, when the particle fitness value is higher than the fitness threshold or the maximum iteration number is reached, the search is terminated and the global optimal position of the particle is obtained; the global optimal position of the particle specifically refers to the optimal education path parameter combination. 6.The intelligent education management system based on artificial intelligence according to claim 1, wherein: The multi-source data acquisition module specifically acquires data from an education learning platform and a work task platform, obtains education management original data, and performs data optimization processing on the education management original data to obtain education management optimized data; the education management original data includes reference education direction selection data, reference education effect prediction data, and target education direction selection data; the data optimization processing specifically includes data cleaning, data standardization, and data coding processing to obtain the education management optimized data.

Images