Intelligent learning dynamic optimization system introducing time sequence

Abstract

The invention relates to the technical field of dynamic optimization, and discloses an intelligent learning dynamic optimization system introducing a time sequence. The system comprises a multi-source data preprocessing module, an intelligent learning knowledge graph establishing module, a learning path optimizing module and a learning resource matching module. Firstly, a data detection model is constructed based on a neural network to perform invalid data filtering and abnormal data detection; secondly, knowledge node association strength is calculated based on a time sequence, and an intelligent learning knowledge graph is established; marking a learning path in the intelligent learning knowledge graph, and performing dynamic path planning by using an improved Harris eagle optimization algorithm to find an optimal learning path; and finally, generating learning resource configuration according to the optimal learning path, and establishing a learning resource feature library for resource matching to obtain an intelligent learning optimization strategy. According to the method, the learning behavior data are analyzed and processed by introducing the time sequence, and the purpose of intelligent learning dynamic optimization is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic optimization, and in particular to an intelligent learning dynamic optimization system that introduces time sequence. Background Art

[0002] In recent years, dynamic optimization algorithms, owing to their adaptive capabilities in complex dynamic environments, have gradually become a core research direction in fields such as industrial control, energy management, and financial decision-making. With the integration of technologies such as reinforcement learning and evolutionary algorithms, intelligent systems are now capable of real-time parameter adjustment and multi-objective collaborative optimization in some dynamic scenarios. However, existing methods generally lack explicit modeling of the temporal correlations within dynamic processes. This makes it difficult to effectively capture the temporal evolution of states in dynamic systems with strong temporal dependencies, such as production processes and supply chain scheduling. This limits long-term optimization performance and robustness in dynamic scenarios.

[0003] The main methods of traditional intelligent learning dynamic optimization are: first, an optimization framework based on static models, through offline training of fixed-parameter prediction models (such as neural networks and support vector machines); second, a hybrid strategy based on rolling optimization, such as combining model predictive control (MPC) with reinforcement learning to achieve dynamic optimization through short-term prediction and feedback mechanisms.

[0004] In traditional intelligent learning dynamic optimization methods, static models cannot accurately represent the time-varying characteristics of dynamic systems, resulting in a significant decrease in optimization effects in long-term tasks. At the same time, offline training models cannot adaptively capture the time-varying characteristics of dynamic systems due to the lack of time dimension modeling, and cannot achieve good learning dynamic optimization effects. Summary of the Invention

[0005] In response to the problems in the related art, the present invention provides an intelligent learning dynamic optimization system that introduces time sequence to overcome the technical problems existing in the existing related art.

[0006] To solve the above technical problems, the present invention is achieved through the following technical solutions: The present invention is an intelligent learning dynamic optimization system that introduces time sequence, which specifically includes: a multi-source data preprocessing module, an intelligent learning knowledge graph establishment module, a learning path optimization module and a learning resource matching module; The multi-source data preprocessing module is used to collect data and perform multi-source data fusion, then build a data detection model based on a neural network to filter invalid data and detect abnormal data, and output the final learning behavior data set; The intelligent learning knowledge graph establishment module is used to divide static attributes and dynamic attributes according to the final learning behavior data set, calculate the association strength of knowledge nodes based on the time sequence, and establish an intelligent learning knowledge graph; The learning path optimization module is used to mark the learning paths in the intelligent learning knowledge graph and use the improved Harris Eagle optimization algorithm to perform dynamic path planning to find the best learning path; The learning resource matching module is used to generate a learning resource configuration according to the optimal learning path, and then establish a learning resource feature library for resource matching.

[0007] Preferably, the data collection and multi-source data fusion includes: Time series data and interactive behavior data are collected, and several knowledge points are selected to collect learning behavior data in sequence to obtain a learning behavior data set.

[0008] Preferably, the step of constructing a data detection model based on a neural network to filter invalid data and detect abnormal data includes: The data detection model is set to include an invalid data filtering model and an anomaly detection model. The loss function of the data detection model is cross entropy loss. The invalid data filtering model includes an input layer, a hidden layer, and an output layer, wherein the hidden layer adopts a bidirectional LSTM network and an Attention mechanism; the anomaly detection model includes an encoder, a decoder, and anomaly determination; Acquire learning behavior data again to obtain a new learning behavior data set, normalize and encode the new learning behavior data set, and divide it into a sample training set and a sample test set, input them into the data detection model, and the output layer outputs the invalid data probability, and the abnormality judgment outputs the reconstruction error value to obtain the final data detection model; After the learning behavior data set is normalized and encoded, it is input into the final data detection model to identify invalid data and abnormal data to obtain the final learning behavior data set.

[0009] Preferably, dividing the static attributes and dynamic attributes according to the final learning behavior data set includes: The logical relationships between the knowledge points in the final learning behavior data set are marked to obtain static attributes of the knowledge nodes, and the logical relationships between the knowledge points are replaced according to the logical relationships between the knowledge points and time series data to obtain dynamic attributes of the knowledge nodes.

[0010] Preferably, the step of calculating the association strength of knowledge nodes based on the time sequence and establishing the intelligent learning knowledge graph includes: Calculate the association strength of knowledge nodes and set a decay threshold. When the association strength of knowledge nodes is greater than the decay threshold, add the same knowledge points to the binary group and form triples based on the logical relationship between the knowledge points. Then traverse the final learning behavior data set in sequence to extract several triples. The knowledge points in several triples are regarded as knowledge graph nodes, and the logical relationships between the knowledge points in several triples are regarded as knowledge graph edges. The knowledge graph nodes are connected in sequence using knowledge graph edges, and the knowledge node association strength corresponding to the knowledge graph nodes is calculated in sequence. The knowledge node association strength is marked on the knowledge graph edge to establish an intelligent learning knowledge graph.

[0011] Preferably, the learning path in the labeled intelligent learning knowledge graph includes: The knowledge graph nodes are marked in the intelligent learning knowledge graph to obtain a knowledge graph node sequence, and a learning path set is generated according to the knowledge graph node sequence; and then an objective function is established based on the maximum knowledge coverage.

[0012] Preferably, the use of the improved Harris Hawk optimization algorithm to perform dynamic path planning and find the best learning path includes: The Harris Hawk optimization algorithm is improved by fusing chaos fusion parameters and nonlinear control factors. The improved Harris Hawk optimization algorithm is obtained. The objective function is regarded as the fitness function, and the process of finding the best fitness function value is regarded as the process of finding the best bread function value. Randomly select a learning path from the learning path set and record it as an initial learning path, where the initial learning path includes a number of knowledge graph nodes; Assume that there is a Harris Hawk population in the search space, and the Harris Hawk individuals in the Harris Hawk population represent the initial learning path; Chaotic fusion parameters are introduced to initialize the Harris Hawk population, and then nonlinear control factors are used to balance the global exploration phase and the local exploitation phase. The Harris Hawk population enters the global exploration phase and the local development phase, updates the position of the Harris Hawk population, replaces the knowledge graph nodes in the initial learning path, and generates a new learning path; The maximum number of iterations is set, and the iteration is stopped until the current number of iterations reaches the maximum number of iterations, and the final Harris Hawk population is obtained. The Harris Hawk individual corresponding to the best fitness function value is selected from the final Harris Hawk population to obtain the best learning path.

[0013] Preferably, generating a learning resource configuration according to the optimal learning path and then establishing a learning resource feature library for resource matching includes: Obtaining optimal knowledge points and optimal knowledge node association strengths according to the optimal learning path to form a learning resource configuration; Establish a learning resource feature library, calculate the matching coefficient between the optimal knowledge point in the learning resource configuration and the learning resource feature library, find the learning resource data with the highest matching coefficient in the learning resource feature library, and generate an intelligent learning optimization strategy.

[0014] The present invention has the following beneficial effects: 1. This invention fuses multi-source data and builds a data detection model based on a neural network to filter invalid data and detect abnormal data. Through the bidirectional LSTM and Attention mechanism, it captures long-term dependencies in time series and focuses on key abnormal periods, thereby achieving invalid data filtering and anomaly detection, effectively eliminating illogical or obviously erroneous data, and ensuring that subsequent analysis is based on high-quality data.

[0015] 2. The invention divides the static and dynamic attributes of knowledge nodes, and then calculates the association strength of knowledge nodes based on time sequence to establish an intelligent learning knowledge graph. By using formulas to quantify the association strength, weak associations are automatically cut off to prevent invalid knowledge transfer. Triples are used to automatically screen strong associations, simplify the graph structure, and reduce redundant information interference. The full-process automated generation of the graph is suitable for large-scale learning groups.

[0016] 3. This invention uses the improved Harris Eagle optimization algorithm for dynamic path planning to find the optimal learning path. It improves the Harris Eagle optimization algorithm by fusing chaos fusion parameters and nonlinear control factors to solve the problem that traditional algorithms are prone to falling into local optimality, helps to escape from local optimality, and recommends personalized learning paths for different learners. It has good optimization effects in long-term sequence tasks, solves the problems of resource waste and poor adaptability in traditional methods, and is suitable for a variety of learning scenarios.

[0017] 4. This invention generates a learning resource configuration, establishes a learning resource feature library, performs resource matching on the learning resource configuration and the learning resource feature library, transforms the abstract knowledge structure into quantifiable matching parameters, and achieves precise matching of the learning resource configuration and the feature library through real-time weight calculation, thus having a good dynamic learning optimization effect.

[0018] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying any creative work.

[0020] Figure 1 The present invention provides a flow chart of an intelligent learning dynamic optimization system that introduces a time sequence.

[0021] Figure 2 The present invention provides a flow chart of an intelligent learning dynamic optimization method that introduces a time sequence. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] In traditional intelligent learning dynamic optimization methods, static models cannot accurately represent the time-varying characteristics of dynamic systems, resulting in a significant decrease in optimization effects in long-term tasks. At the same time, offline training models cannot adaptively capture the time-varying characteristics of dynamic systems due to the lack of time dimension modeling, and cannot achieve good learning dynamic optimization effects.

[0024] In order to solve the above technical problems, Figure 1 As shown, an embodiment of the present invention provides an intelligent learning dynamic optimization system that introduces time sequence, which specifically includes: a multi-source data preprocessing module, an intelligent learning knowledge graph establishment module, a learning path optimization module and a learning resource matching module; the multi-source data preprocessing module is used to collect data and perform multi-source data fusion, and then build a data detection model based on a neural network to filter invalid data and detect abnormal data, and output a final learning behavior data set; the intelligent learning knowledge graph establishment module is used to divide static attributes and dynamic attributes according to the final learning behavior data set, calculate the association strength of knowledge nodes based on time sequence, and establish an intelligent learning knowledge graph; the learning path optimization module is used to mark the learning path in the intelligent learning knowledge graph, use the improved Harris Hawk optimization algorithm to perform dynamic path planning, and find the best learning path; the learning resource matching module is used to generate a learning resource configuration according to the best learning path, and then establish a learning resource feature library for resource matching.

[0025] In a specific embodiment, an interdisciplinary project-based learning platform aims to analyze learners' learning behaviors in mathematics courses. The platform collects learning data on 500 learners in 10 related courses, including algebra, geometry, probability, calculus, and statistics, providing good data support for the embodiments of the present invention. In the specific implementation process of the above embodiment, first, learning behavior data is collected and multi-source data is integrated, and then a data detection model is constructed based on a neural network to filter invalid data and detect abnormal data, and output the final learning behavior data set; this method uses a bidirectional LSTM and Attention mechanism to capture long-term dependencies in time series and focus on key abnormal periods, to achieve invalid data filtering and anomaly detection, and effectively eliminate illogical or obviously erroneous data, to ensure that subsequent analysis is based on high-quality data; secondly, the static attributes and dynamic attributes of knowledge nodes are divided according to the final learning behavior data set, and then the association strength of knowledge nodes is calculated based on the time sequence to establish an intelligent learning knowledge graph; this method uses a formula to quantify the association strength, automatically cuts off weak associations, prevents invalid knowledge transfer, and uses triples to automatically screen strong associations, simplify the graph structure, and reduce redundant information interference. The full-process automatic generation of the graph is suitable for large-scale learning groups; then the intelligent learning is marked The learning path in the learning knowledge graph is learned, and the objective function is established based on the maximum knowledge coverage. The improved Harris Eagle optimization algorithm is used for dynamic path planning to find the optimal learning path. This method improves the Harris Eagle optimization algorithm through algorithm improvement and multi-objective optimization, and integrates chaos fusion parameters and nonlinear control factors to improve the Harris Eagle optimization algorithm. It solves the problem that traditional algorithms are prone to fall into local optimality, helps to jump out of local optimality, recommends personalized learning paths for different learners, and has good optimization effect in long-term sequence tasks. It solves the problems of resource waste and poor adaptability in traditional methods and is suitable for a variety of learning scenarios. Finally, a learning resource configuration is generated according to the optimal learning path, and a learning resource feature library is established. The learning resource configuration and the learning resource feature library are matched to obtain an intelligent learning optimization strategy. This method converts the abstract knowledge structure into quantifiable matching parameters, and realizes the accurate matching of learning resource configuration and feature library through real-time weight calculation, which has a good learning dynamic optimization effect.

[0026] Furthermore, in order to better introduce the technical solutions of the embodiments of the present invention, Figure 2 As shown, combined with an intelligent learning dynamic optimization method that introduces time sequence, a detailed description of the intelligent learning dynamic optimization system that introduces time sequence is given, specifically including the following contents: S1. Collect learning behavior data and perform multi-source data fusion to obtain a learning behavior data set. Then, build a data detection model based on a neural network to filter invalid data and detect abnormal data, and output the final learning behavior data set. Said S1 comprises the following steps: S11. Collecting learning behavior data, including time series data and interactive behavior data; in the process of collecting time series data, recording the learning time of knowledge points, the switching frequency of knowledge points, and the learning interval of knowledge points; in the process of collecting interactive behavior data, recording the average learning speed of knowledge points, the percentage of knowledge points mastered, and the percentage of knowledge points skipped; selecting several knowledge points, and sequentially collecting learning behavior data to obtain a learning behavior data set; S12. Set the data detection model to include an invalid data filtering model and an anomaly detection model. The loss function of the data detection model is cross entropy loss. The invalid data filtering model includes an input layer, a hidden layer, and an output layer. The hidden layer adopts a bidirectional LSTM network and an Attention mechanism, and the output layer activation function is a Sigmoid function. The anomaly detection model includes an encoder, a decoder, and an anomaly determination. S13. Acquire learning behavior data again to obtain a new learning behavior data set, normalize and encode the new learning behavior data set, and divide it into a sample training set and a sample test set, input them into the data detection model, and output the final detection results. The specific steps are as follows: S131, inputting the sample training set into the invalid data filtering model of the data detection model for training, extracting time step features, and outputting invalid data probability at the output layer; then inputting the sample training set into the anomaly detection model of the data detection model for training, and outputting a reconstruction error value for anomaly determination; setting a maximum number of training rounds, and stopping training when the data detection model reaches the maximum number of training rounds, thereby obtaining a trained data detection model; S132, then input the sample test set into the invalid data filtering model and the anomaly detection model of the trained data detection model respectively, output the model results, set the accuracy threshold, and when the model result accuracy is greater than the accuracy threshold, obtain the final data detection model; otherwise, adjust the weight until the model result accuracy is greater than the accuracy threshold; S133, after normalizing and encoding the learning behavior data set, input it into the final data detection model and output the final detection result; set an invalid threshold and an error threshold, and when the invalid data probability in the final detection result is greater than the invalid threshold, the corresponding learning behavior data is recorded as invalid data and deleted; when the reconstruction error value in the final detection result is greater than the error threshold, the corresponding learning behavior data is recorded as abnormal data and deleted to obtain the final learning behavior data set; In this embodiment, learning behavior data is collected and multi-source data is fused. Then, a data detection model is constructed based on a neural network to filter invalid data and detect abnormal data, and the final learning behavior data set is output. This method uses a bidirectional LSTM and Attention mechanism to capture long-term dependencies in time series and focus on key abnormal periods, thereby filtering invalid data and detecting abnormalities, effectively eliminating illogical or obviously erroneous data, and ensuring that subsequent analysis is based on high-quality data. Specifically, for example, time series data is collected: learner A's learning time (25 minutes), switching frequency (switching to other knowledge points 3 times), and learning interval (learning the same knowledge point again every other day) in algebra are recorded; learner B's learning time in probability is recorded (5 minutes, which may be due to an error); interactive behavior data is collected: learner A's average learning speed of algebra basics (2 pages of content per minute), mastery rate (80% correct rate of exercises), and skip rate (skipping 10% of examples). Learner B's probability skip rate reaches 95% (possibly not actually learning). The above two types of data are collected for 10 knowledge points in sequence to form an initial database containing records of 500 learners. In the data detection model, the invalid data filtering model receives time series and interaction behavior data (such as learning duration and skip rate). The bidirectional LSTM extracts time step features (for example, identifying "too short learning intervals" as possible invalid operations). The Attention mechanism focuses on abnormal periods (such as learner B's 5-minute learning record). The Sigmoid activation function outputs the invalid probability. In the data detection model, the encoder compresses the data into a low-dimensional representation, and the decoder attempts to reconstruct the original data and then calculates the reconstruction error. For example, 70% of the data is set as the training set and 30% as the test set. These are normalized and input into the model. The maximum number of training rounds is set to 100, and early stopping is used to prevent overfitting. The model parameters are locked after the test set accuracy reaches 95%. Finally, the invalid probability and reconstruction error are calculated in the training set and used as the invalid probability threshold and reconstruction error threshold, respectively. The output results are: Learner B's probability data invalid probability 0.92 > the invalid probability threshold 0.85, so it is marked as invalid and deleted. Learner C's geometric reconstruction error 4.2 > the reconstruction error threshold 3, so it is judged as abnormal data and deleted, retaining high-quality data for subsequent analysis. S2. Dividing the final learning behavior data set to obtain static and dynamic attributes of knowledge nodes, and then calculating the association strength of the knowledge nodes based on the time sequence to establish an intelligent learning knowledge graph; The S2 comprises the following steps: S21, marking the static labels in the final learning behavior data set, wherein the static labels are logical relationships between knowledge points, to obtain static attributes of knowledge nodes; According to the logical relationship between the knowledge points and the time series data, a time threshold is set. When the learning interval of the knowledge points is greater than the time threshold, the logical relationship between the knowledge points is changed to obtain the dynamic attributes of the knowledge nodes; S22, then obtain the last learning time of the knowledge point based on the time series data in the final learning behavior data set, and obtain the number of knowledge point tags based on the interactive behavior data in the final learning behavior data set; calculate the knowledge node association strength ,in b Indicates the number of knowledge point tags, Indicates the last time a knowledge point was learned. Indicates the current learning time of the knowledge point; A decay threshold is set, and the association strength of the knowledge nodes is used to measure the association of the knowledge points. When the association strength of the knowledge nodes is greater than the decay threshold, the same knowledge points are added to the binary group, and based on the logical relationship between the knowledge points, a triple is formed: <knowledge point, logical relationship between knowledge points, knowledge point>; the final learning behavior data set is traversed in sequence to extract a number of triples; S23. Consider the knowledge points in the triples as knowledge graph nodes, and the logical relationships between the knowledge points in the triples as knowledge graph edges. Use the knowledge graph edges to sequentially connect the knowledge graph nodes, and then calculate the knowledge node association strengths corresponding to the knowledge graph nodes in sequence. Mark the knowledge node association strengths on the knowledge graph edges to establish an intelligent learning knowledge graph. In this embodiment, the static attributes and dynamic attributes of knowledge nodes are divided according to the final learning behavior data set, and then the association strength of knowledge nodes is calculated based on the time sequence to establish an intelligent learning knowledge graph; this method uses a formula to quantify the association strength, automatically cuts off weak associations, prevents invalid knowledge transfer, and uses triples to automatically filter strong associations, simplify the graph structure, and reduce redundant information interference. The full-process automatic generation of the graph is suitable for large-scale learning groups; specifically, for example, static attributes: the logical relationship between knowledge points A, B, and C is a prerequisite relationship, that is, A→B→C (you must master A before you can learn B, and you must master B before you can learn C), dynamic attributes: set the time threshold to 7 days, where A The learning interval of A→B is 4 days (not exceeding the threshold), and the logical relationship remains the premise relationship. The learning interval of B→C is 15 days (exceeding the threshold), and the logical relationship is dynamically adjusted to a premise relationship that needs to be reviewed. When the memory strength of the knowledge point is lower than 0.5, the activation of the hippocampus will drop sharply, that is, the attenuation threshold is set to 0.5. The attenuation threshold (0.5) is used to automatically filter strong associations, simplify the graph structure, and reduce redundant information interference. The association strength of A→B is 0.82>0.5, and the triple is retained. However, the association strength of B→C is 0.47<0.5, and the association is not retained. The knowledge points A, B, and C are used as nodes, and only the edge of A→B is retained. The logical relationship is the premise relationship and is marked with the number 0.82. S3. Marking the learning paths in the intelligent learning knowledge graph, establishing an objective function based on maximum knowledge coverage, and using an improved Harris Hawk optimization algorithm for dynamic path planning to find the optimal learning path; The S3 includes the following steps: S31. Mark the knowledge graph nodes in the intelligent learning knowledge graph to obtain a knowledge graph node sequence ,in Indicates the m Knowledge graph nodes are generated, and a set of learning paths is generated according to the sequence of knowledge graph nodes; the proportion of knowledge points mastered in the learning path is calculated to obtain the total mastery of the current knowledge points, and then the weighted average of the association strength of the knowledge nodes in the learning path is calculated, and the total learning time of the learning path is recorded. The objective function is established based on the maximum knowledge coverage. ,in 、 and represents the weight coefficient, Indicates the total amount of knowledge currently mastered. represents the weighted average of the association strength of knowledge nodes in the learning path, Indicates the total learning time of the learning path; S32. The Harris Eagle optimization algorithm is improved by integrating the chaos fusion parameter and the nonlinear control factor to obtain the improved Harris Eagle optimization algorithm. In the intelligent learning knowledge graph, the improved Harris Eagle optimization algorithm is used for dynamic path planning to find the optimal learning path. The specific steps are as follows: S321: Taking the objective function as the fitness function and the process of finding the optimal fitness function value as the process of finding the optimal bread function value, randomly selecting a learning path from the learning path set and recording it as the initial learning path, wherein the initial learning path includes a plurality of knowledge graph nodes; Set the search space, there is a Harris Hawk population in the search space, and the number of Harris Hawk population is p , Harris Hawk population dimension is q , the Harris Hawk individuals in the Harris Hawk population represent the initial learning path; the chaotic fusion parameters are introduced to initialize the Harris Hawk population, and the first i Harris Hawk individual locations To update, When i +1 Harris Hawk individual position ,when When i +1 Harris Hawk individual position ; S322, Harris Hawk population iteration process includes global exploration phase and local development phase, using escape energy to transform the global exploration phase and local development phase, setting the initial escape energy is a random number between the interval [-1, 1], and a nonlinear control factor is introduced to replace the escape energy. ; S323, when the nonlinear control factor is greater than or equal to 1, the Harris Hawk population enters the global exploration phase, and the current number of iterations is set to t , No. t The first iteration i Harris's Hawk individual location is , No. t The average position of Harris's hawk population at the iteration is , No. t The position of the random Harris hawk individual at the iteration is , the prey location of Harris's Hawk population is , d Represents a random number between the interval (0, 1), and the upper bound of the search space is , the search space lower bound is ;when When t+ Harris hawk population location at iteration 1 ,when When ,in 、 、 and represents a random number between the interval (0, 1); at this time, the knowledge graph nodes in the initial learning path are replaced, a new learning path is generated, and the fitness function value is calculated; S324, when the nonlinear control factor is less than 1, the Harris hawk population enters a local development stage, which includes a soft encirclement strategy and a hard encirclement strategy; Set the escape probability. When the nonlinear control factor is between the interval [0.5, 1) and the escape probability is greater than or equal to 0.5, use the soft encirclement strategy to update the position of the Harris's Hawk population. At this time, adjust the order of adjacent knowledge graph nodes in the new learning path. When the nonlinear control factor is less than 0.5 and the escape probability is greater than or equal to 0.5, use the hard encirclement strategy to update the position of the Harris's Hawk population again. At this time, make fine adjustments to the new learning path, eliminate knowledge graph nodes with low knowledge point mastery, and select knowledge graph nodes with high knowledge node association strength. When the nonlinear control factor is between the interval [0.5, 1) and the escape probability is less than 0.5, use Levy flight to guide the position update of the Harris's Hawk population. This calculation obtains the current best fitness function value, generates the next generation of Harris's Hawk population, sets the maximum number of iterations, and stops iterating until the current number of iterations reaches the maximum number of iterations to obtain the final Harris's Hawk population. Select the Harris's Hawk individual corresponding to the best fitness function value in the final Harris's Hawk population to obtain the best learning path. In this embodiment, the learning paths in the intelligent learning knowledge graph are marked, and the objective function is established based on the maximum knowledge coverage. The improved Harris Eagle optimization algorithm is used for dynamic path planning to find the best learning path. This method improves the Harris Eagle optimization algorithm through algorithm improvement and multi-objective optimization, and integrates chaos fusion parameters and nonlinear control factors to solve the problem that traditional algorithms are prone to falling into local optimality, helps to jump out of local optimality, and recommends personalized learning paths for different learners. It has good optimization effect in long-term sequence tasks, solves the problems of resource waste and poor adaptability in traditional methods, and is applicable to various learning scenarios. Specifically, for example, the SHAP value is used to analyze historical data to obtain the weight coefficient, the knowledge point complexity explains 50% of the variance, the user's cognitive level contributes 30%, and the device interaction frequency has an influence of 20%, that is, the weight coefficients are set to 0.5 and 0 respectively. .3 and 0.2; randomly generate 10 initial learning paths (such as path 1: a1→a3→a5; path 2: a2→a4→a1), use chaotic fusion parameters to initialize the Harris hawk population, in the global exploration phase, in the path a1→a3→a5, if a6 can be replaced by a3 of equal difficulty, then generate a new path a1→a6→a5; in the local development phase, adjust the order of adjacent nodes in the path, for example, the path a1→a6→a5 is adjusted to a6→a1→a5, calculate the new fitness value, and eliminate low-value nodes, for example, in the path a2→a4→a1, a4 is replaced by a5 with high correlation strength, and use Levy flight guidance to randomly insert jump steps, for example, the path a2→a5→a1 may mutate to a2→a5→a3; when the number of iterations reaches the maximum iteration number of 100, select the path with the highest fitness, and the optimal path may be a2→a4→a5; S4. Generate a learning resource configuration based on the optimal learning path, establish a learning resource feature library, perform resource matching on the learning resource configuration and the learning resource feature library, and obtain an intelligent learning optimization strategy; The S4 comprises the following steps: S41, the optimal learning path includes the optimal knowledge graph node and the knowledge graph edge connecting the optimal knowledge graph node, and the optimal knowledge point and the optimal knowledge node association strength are obtained respectively to form a learning resource configuration; S42: Collect learning resource data, including learning resource preferences, learning resource difficulty levels, etc., and annotate the learning resource data to establish a learning resource feature library; set the knowledge point overlap and urgency, assign weights to each, and calculate the matching coefficient between the optimal knowledge point in the learning resource configuration and the learning resource feature library; search for the learning resource data with the highest matching coefficient in the learning resource feature library, and generate an intelligent learning optimization strategy; In this embodiment, a learning resource configuration is generated according to the optimal learning path, and then a learning resource feature library is established, and resource matching is performed on the learning resource configuration and the learning resource feature library to obtain an intelligent learning optimization strategy; this method converts the abstract knowledge structure into quantifiable matching parameters, and realizes accurate matching of the learning resource configuration and the feature library through real-time weight calculation, which has a good learning dynamic optimization effect; specifically, for example, after generating a learning resource configuration and analyzing the user's learning trajectory, the optimal knowledge graph path is determined to be: algebra basics → linear equation → inequality application, and the knowledge node association strengths are 0.85 and 0.75 respectively; after constructing the learning resource feature library, a two-dimensional matching algorithm is used to calculate the knowledge point overlap = (resource coverage knowledge points ∩ target knowledge points) / total number of target knowledge points × knowledge weight, urgency = resource urgency mark × time decay coefficient, and obtain a comprehensive matching coefficient = (knowledge point overlap × 0.6) + (urgency × 0.4); matching is performed according to the learning resource feature library, priority is given to learning algebra basics quickly, and the prerequisite knowledge required for subsequent equation learning is associated to ensure the timeliness and adaptability of the strategy at all times.

[0027] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0028] The preferred embodiments of the invention disclosed above are intended only to help illustrate the invention. These preferred embodiments do not exhaust all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. An intelligent learning dynamic optimization system that introduces time sequence, characterized in that: include: The multi-source data preprocessing module is used to collect data and perform multi-source data fusion, then build a data detection model based on a neural network to filter invalid data and detect abnormal data, and output the final learning behavior data set; The intelligent learning knowledge graph establishment module is used to divide static attributes and dynamic attributes according to the final learning behavior data set, calculate the association strength of knowledge nodes based on the time sequence, and establish the intelligent learning knowledge graph; The learning path optimization module is used to mark the learning paths in the intelligent learning knowledge graph and use the improved Harris Eagle optimization algorithm to perform dynamic path planning to find the best learning path; The learning resource matching module is used to generate learning resource configuration according to the optimal learning path, and then establish a learning resource feature library for resource matching.

2. The intelligent learning dynamic optimization system with time sequence introduction according to claim 1 is characterized in that: The data collection and multi-source data fusion includes: Time series data and interactive behavior data are collected, and several knowledge points are selected to collect learning behavior data in sequence to obtain a learning behavior data set.

3. The intelligent learning dynamic optimization system with time sequence introduction according to claim 2 is characterized in that: The method of constructing a data detection model based on a neural network to filter invalid data and detect abnormal data includes: The data detection model is set to include an invalid data filtering model and an anomaly detection model. The loss function of the data detection model is cross entropy loss. The invalid data filtering model includes an input layer, a hidden layer, and an output layer, wherein the hidden layer adopts a bidirectional LSTM network and an Attention mechanism; the anomaly detection model includes an encoder, a decoder, and anomaly determination; Acquire learning behavior data again to obtain a new learning behavior data set, normalize and encode the new learning behavior data set, and divide it into a sample training set and a sample test set, input them into the data detection model, and the output layer outputs the invalid data probability, and the abnormality judgment outputs the reconstruction error value to obtain the final data detection model; After the learning behavior data set is normalized and encoded, it is input into the final data detection model to identify invalid data and abnormal data to obtain the final learning behavior data set.

4. The intelligent learning dynamic optimization system with time sequence introduction according to claim 3 is characterized in that: The division of static attributes and dynamic attributes according to the final learning behavior data set includes: The logical relationships between the knowledge points in the final learning behavior data set are marked to obtain static attributes of the knowledge nodes, and the logical relationships between the knowledge points are replaced according to the logical relationships between the knowledge points and time series data to obtain dynamic attributes of the knowledge nodes.

5. The intelligent learning dynamic optimization system with time sequence introduction according to claim 4 is characterized in that: The calculation of the association strength of knowledge nodes based on the time sequence and the establishment of the intelligent learning knowledge graph include: Calculate the association strength of knowledge nodes and set a decay threshold. When the association strength of knowledge nodes is greater than the decay threshold, add the same knowledge points to the binary group and form triples based on the logical relationship between the knowledge points. Then traverse the final learning behavior data set in sequence to extract several triples. The knowledge points in several triples are regarded as knowledge graph nodes, and the logical relationships between the knowledge points in several triples are regarded as knowledge graph edges. The knowledge graph nodes are connected in sequence using knowledge graph edges, and the knowledge node association strength corresponding to the knowledge graph nodes is calculated in sequence. The knowledge node association strength is marked on the knowledge graph edge to establish an intelligent learning knowledge graph.

6. The intelligent learning dynamic optimization system with time sequence introduction according to claim 5 is characterized in that: The learning paths in the labeled intelligent learning knowledge graph include: The knowledge graph nodes are marked in the intelligent learning knowledge graph to obtain a knowledge graph node sequence, and a learning path set is generated according to the knowledge graph node sequence; and then an objective function is established based on the maximum knowledge coverage.

7. The intelligent learning dynamic optimization system with time sequence introduction according to claim 6 is characterized in that: The improved Harris Eagle optimization algorithm is used to perform dynamic path planning and find the best learning path, including: The Harris Hawk optimization algorithm is improved by fusing chaos fusion parameters and nonlinear control factors. The improved Harris Hawk optimization algorithm is obtained. The objective function is regarded as the fitness function, and the process of finding the best fitness function value is regarded as the process of finding the best bread function value. Randomly select a learning path from the learning path set and record it as an initial learning path, where the initial learning path includes a number of knowledge graph nodes; Assume that there is a Harris Hawk population in the search space, and the Harris Hawk individuals in the Harris Hawk population represent the initial learning path; Chaotic fusion parameters are introduced to initialize the Harris Hawk population, and then nonlinear control factors are used to balance the global exploration phase and the local exploitation phase. The Harris Hawk population enters the global exploration phase and the local development phase, updates the position of the Harris Hawk population, replaces the knowledge graph nodes in the initial learning path, and generates a new learning path; The maximum number of iterations is set, and the iteration is stopped until the current number of iterations reaches the maximum number of iterations, and the final Harris Hawk population is obtained. The Harris Hawk individual corresponding to the best fitness function value is selected from the final Harris Hawk population to obtain the best learning path.

8. The intelligent learning dynamic optimization system with time sequence introduction according to claim 7 is characterized in that: Generating a learning resource configuration according to the optimal learning path and then establishing a learning resource feature library for resource matching includes: Obtaining optimal knowledge points and optimal knowledge node association strengths according to the optimal learning path to form a learning resource configuration; Establish a learning resource feature library, calculate the matching coefficient between the optimal knowledge point in the learning resource configuration and the learning resource feature library, find the learning resource data with the highest matching coefficient in the learning resource feature library, and generate an intelligent learning optimization strategy.

9. An intelligent learning dynamic optimization method with time sequence as described in any one of claims 1 to 8, characterized in that: Specifically include: S1. Collect learning behavior data and perform multi-source data fusion to obtain a learning behavior data set. Then, build a data detection model based on a neural network to filter invalid data and detect abnormal data, and output the final learning behavior data set. S2. Dividing the final learning behavior data set to obtain static and dynamic attributes of knowledge nodes, and then calculating the association strength of the knowledge nodes based on the time sequence to establish an intelligent learning knowledge graph; S3. Marking the learning paths in the intelligent learning knowledge graph, establishing an objective function based on maximum knowledge coverage, and using an improved Harris Hawk optimization algorithm for dynamic path planning to find the optimal learning path; S4. Generate a learning resource configuration based on the optimal learning path, then establish a learning resource feature library, perform resource matching on the learning resource configuration and the learning resource feature library, and obtain an intelligent learning optimization strategy.

Images