Cross-cycle intelligent assessment method for student exercise data
By constructing a motion trajectory dataset with non-uniform time sampling and performing manifold embedding and kernel regression, the problem of continuous modeling and personalized suggestions for students' physical fitness development trends was solved, enabling refined prediction and personalized feedback of students' physical fitness, and improving the stability and practicality of the system.
Patent Information
- Application Number
- CN202511020079.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-23
- Publication Date
- 2025-10-28
AI Technical Summary
Existing technologies cannot continuously model the development trend of students' physical fitness. Their predictive ability relies on complete sequences, their output suggestions lack personalization, and their system deployment is complex. Teachers face high learning costs when using them, making it difficult to achieve stable implementation in daily teaching.
By collecting students' physical fitness data over multiple time periods, a non-uniform time-sampled motion trajectory dataset is constructed. A manifold embedding algorithm is used for low-dimensional spatial mapping, and a kernel regression method is combined to estimate the trend of physical fitness changes and generate personalized assessment suggestions.
It enables more refined estimation of students' physical fitness development, enhances the continuity of prediction and the relevance of personalized suggestions, improves the system's compatibility and deployment capabilities, and solves the problems of poor robustness and low feedback accuracy of traditional methods in the case of missing data.
Smart Images
Figure CN120853808A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of educational informatization and intelligent assessment technology, and in particular to an intelligent assessment method for student movement data across cycles. Background Technology
[0002] In current physical education practices in primary, secondary, and higher education institutions, physical fitness assessment has gradually become an important component of teaching feedback and personalized intervention. With the increasing level of educational informatization, single assessment results are no longer sufficient for teachers and parents to fully understand students' growth trends; there is an urgent need for a technology that can continuously track individual physical fitness changes, identify potential risk trends, and provide intervention suggestions. This technology should not only possess predictive capabilities but also be able to handle complex situations such as missing or structurally incomplete data.
[0003] Current physical fitness management technologies mostly employ standardized template-based assessment systems. These systems compare individual student scores with national or regional standards to derive corresponding evaluation levels. This approach is simple to implement, provides fast feedback, and is suitable for large-scale assessments. In some advanced applications, statistical analysis methods are incorporated, such as mean comparison and historical distribution fluctuation analysis, which can identify patterns in performance changes to some extent. Some commercial products are exploring the use of neural networks to model complete assessment data sequences for detecting abnormal behavior or training responses.
[0004] However, current methods generally suffer from weak trend modeling capabilities, especially when data points are discontinuous or sample time periods are misaligned, rendering the models almost ineffective. Static, template-based suggestions cannot be integrated with students' individual developmental trajectories, offering only general feedback with limited practicality. Furthermore, the prediction module and suggestion system are typically independent, resulting in numerous data gaps and hindering the formation of an effective feedback loop; system deployment is complex, and the learning curve for teachers is high, making stable implementation in daily teaching difficult. These problems frequently arise in practice, impacting the true auxiliary value of educational data. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent evaluation method for student exercise data across cycles, which solves the problems in the existing technology that it is impossible to continuously model the development trend of students' physical fitness, the predictive ability depends on the complete sequence, and the suggested output lacks personalization.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent evaluation method for student movement data across cycles, comprising the following steps: Collect students' physical fitness data over multiple time periods to construct a non-uniform time-sampled motion trajectory dataset containing timestamps and corresponding physical fitness indicators; Based on the constructed trajectory dataset, the similarity of the movement trajectories of any two students is measured, and a trajectory similarity distance matrix reflecting the differences in trajectory evolution is generated. Based on the distance matrix, a manifold embedding algorithm is used to perform low-dimensional spatial mapping on the students' physical fitness trajectories, thereby obtaining the trajectory embedding representation of each student while preserving the trajectory structure features. Based on the trajectory embedding representation of the target students and combined with the embedding representation of known samples, the kernel regression method is used to estimate the trend of physical fitness changes of the target students at future preset time points; Based on the predicted results of physical fitness change trends, the system generates growth trend curves for target students over future time intervals and further outputs corresponding personalized assessment suggestions.
[0007] Preferably, the physical fitness data includes: The assessment indicators in the National Student Physical Fitness and Health Standards include vital capacity, 50-meter sprint, standing long jump, sit-and-reach, pull-ups, 800-meter and 1000-meter running times. Exercise behavior data collected through wearable devices during classroom and extracurricular exercise, including cadence, heart rate, running distance and exercise duration; The timestamp information associated with the above indicators is used to construct a non-uniform time sampling trajectory sequence.
[0008] Preferably, the trajectory dataset includes: A set of time-series trajectories constructed for each student, each trajectory is represented as: ; in, Indicates the first The number of students The time point of the next physical fitness measurement; Indicates physical fitness index values; Indicates the first A collection of time-series trajectories of 100 students; Indicates the first A total of [number] students conducted [number] [activities]. Secondary physical fitness test; For student index number; This is the sequence number of the physical fitness test.
[0009] Preferably, the low-dimensional space mapping includes: Apply the manifold embedding algorithm to the trajectory similarity distance matrix; The manifold embedding algorithm is selected from diffusion mapping; The mapping result assigns a low-dimensional representation vector to each trajectory.
[0010] Preferably, the trajectory embedding representation includes: In the low-dimensional space, the local neighborhood relationships and overall evolutionary form of the original trajectory are preserved; The embedding is represented as This is used for subsequent similarity calculations and as input to the prediction model; The embedding is represented as It is scalable, meaning that new students can obtain their representations through incremental embedding without retraining the entire embedding space.
[0011] Preferably, the kernel regression method includes: Based on the target student embedding representation With history students Calculate the weight function based on the Euclidean distance between them: ; in, This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; a larger value indicates greater similarity. Indicates the target student and the first The square of the Euclidean distance between the students. It is a natural exponential function. The bandwidth parameter represents the kernel function; Using the sample's predicted physical fitness values at the time point A kernel-weighted estimate is then performed using the weights mentioned above: ; in, To predict physical fitness values at specific time points; This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; the larger the value, the more similar they are. For student index number; The total number of samples is the number of historical students used for training.
[0012] Preferably, the trend of physical fitness change includes: Estimated curves of continuous physical fitness indicators over a future period of time; The trends reflect the slope, acceleration, and phased changes in students' individual physical fitness over time; The trend data supports the detection of physical fitness inflection points, used to assess growth bottlenecks and excessive fatigue.
[0013] Preferably, the growth trend curve includes: A continuous curve generated based on multiple predicted time points; The curve, with time on the horizontal axis and predicted physical fitness value on the vertical axis, is smooth and differentiable. A confidence band is introduced during the curve plotting process to represent the range of prediction uncertainty.
[0014] Preferably, the personalized assessment recommendations include: Rating of the current rate of physical fitness growth; Recommended intensive training programs and suggested training frequency; Early warnings and rest recommendations are issued for detected downward trends; Provide training path reference templates based on the evolution of similar samples for different student types.
[0015] Preferably, the similarity measure includes: Based on the trajectory data sampled in a non-uniform time, a trajectory alignment method is used to perform multi-point correspondence matching, and then the dynamic evolution difference value between different student trajectories is calculated. The difference value is based on the maximum distance between corresponding points in each pair of trajectories and is used to measure the degree of deviation between the two trajectories in their overall shape. By calculating the difference values mentioned above, a trajectory similarity distance matrix is constructed, providing an input basis for subsequent low-dimensional space mapping.
[0016] This invention provides an intelligent evaluation method for student exercise data across different time periods. It has the following beneficial effects: 1. This invention employs a predictive modeling scheme combining trajectory embedding representation and kernel regression methods, achieving refined estimation of students' future physical fitness indicators. It achieves the technical effect of accurately depicting individual physical fitness evolution trends without relying on the original full-volume time series data. Compared to most existing predictive methods that directly use time series extrapolation, this invention solves the problems of poor robustness and inconsistent predictive structure in the event of missing data.
[0017] 2. This invention constructs a growth trend curve over continuous time and employs interpolation and multi-time-point prediction mechanisms to reflect the dynamic trajectory of physical fitness indicators over future periods. Compared to existing methods that only return discrete predicted values, this approach enhances information readability and interpretability, overcoming the shortcomings of traditional models in intuitively identifying trend directions during actual teaching interventions.
[0018] 3. This invention introduces a personalized suggestion generation mechanism for target students, combining trend slope, group comparison, and historical deviation multi-dimensional indicators to output intervention opinions. Compared with traditional systems that mainly output static scores or templates, this method significantly improves the relevance and practical value of the suggestions, thus solving the limitations of traditional systems, which have high generalization and low feedback accuracy.
[0019] 4. This invention constructs a unified overall process structure for trajectory representation, predictive modeling, trend visualization, and suggestion generation, enabling physical fitness development assessment to possess excellent system compatibility and platform deployment capabilities. Compared to the problems of separate functional modules and difficulty in integration in existing technologies, it effectively solves the technical bottlenecks of data fragmentation and low assessment efficiency between training and management systems. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] Please see the appendix Figure 1 This invention provides an intelligent evaluation method for student sports data across different cycles, comprising the following steps: S1. Collect students' physical fitness data over multiple time periods and construct a non-uniform time-sampled motion trajectory dataset containing timestamps and corresponding physical fitness indicators; The collection and trajectory construction of students' physical fitness data provide the data foundation and structural support for the entire intelligent assessment process. This step not only serves as the input of raw data, but its results also directly affect subsequent modules such as trajectory comparison, low-dimensional embedding, and physical fitness trend modeling. Therefore, the technical design of this part must balance data integrity and structural expressiveness to ensure that the assessment system can truly reflect the students' growth process across different life cycles.
[0023] In general, schools face problems such as inconsistent periodicity and incomplete indicator coverage in collecting students' physical fitness data, making it difficult to establish continuous growth models using traditional methods. To overcome this, this invention specifically designs a trajectory modeling method suitable for multi-time-period, asynchronous sampling, which can construct stable physical fitness trajectories with temporal expressive capabilities under non-uniform time sampling conditions.
[0024] The test data in the National Student Physical Fitness and Health Standards include standard indicators such as vital capacity, 50-meter sprint, standing long jump, sit-and-reach, pull-ups, 800-meter run (for female students), and 1000-meter run (for male students). Scoring records of athletic performance during classroom instruction; Wearable fitness devices automatically record behavioral data, such as daily running mileage, duration, average cadence, and heart rate change curves. Data uploaded through the exercise check-in platform during extracurricular training includes training items, timestamps, and goal completion status.
[0025] In one possible implementation, the system performs time-based clustering and preprocessing on each student's data, constructing all historical records into a physical fitness trajectory dataset, denoted as: ; in, Indicates the first The number of students The time point of the next physical fitness measurement; Indicates physical fitness index values; Indicates the first A collection of time-series trajectories of 100 students; Indicates the first A total of [number] students conducted [number] [activities]. Secondary physical fitness test; For student index number; This is the sequence number of the physical fitness test.
[0026] As an option, physical fitness index values This can be further refined into a joint vector of multiple sub-indicators, namely: ; in, Indicates physical fitness index values; For the first The number of students The first measurement Individual energy index values; The number of physical fitness sub-indicators included in each measurement represents the number of indicator dimensions.
[0027] In actual data collection, different students The data may vary significantly, and there may even be missing data for certain time periods. This invention does not require consistent data sampling time intervals; instead, it directly retains the actual sampling timestamps. This approach improves the integrity of the data structure and lays the foundation for subsequent distance calculations based on the trajectory structure.
[0028] In some embodiments, to enhance the expressive power of trajectory sequences, the system can perform trend smoothing or linear interpolation on adjacent time points based on local time windows, thereby enhancing the comparability of trajectories while ensuring the true characteristics of the original data.
[0029] Specifically, the system uniformly encodes and stores all student trajectories, and saves the time span, missing segment information, and indicator dimension coverage of each trajectory through a metadata structure, providing support for subsequent processing modules.
[0030] As a scalable mechanism, when a new student joins, their initial data can also be used to construct trajectories and append them to the dataset in the same way, without having to rebuild the entire sample space, thus possessing good incremental learning adaptability.
[0031] S2. Based on the constructed trajectory dataset, the similarity of the movement trajectories of any two students is measured, and a trajectory similarity distance matrix reflecting the differences in trajectory evolution is generated. In the aforementioned trajectory construction step, a physical fitness evolution trajectory based on non-uniform time series sampling has been established for each student. This trajectory not only includes pairing information between timestamps and indicator values but also accurately preserves the original temporal structure of each student's physical fitness changes. Building upon this, to systematically quantify the differences among individual students in their athletic development, this step introduces a trajectory similarity measurement mechanism. This mechanism aims to establish a structure-sensitive, time-consistent distance measurement system, providing direct input for subsequent trajectory embedding learning and physical fitness trend prediction.
[0032] Generally, students' growth pace is not uniform, with sparse distribution of time points and varying fluctuations in indicators. Simple methods such as Euclidean distance, DTW, or weighted moving averages can easily obscure the actual form of path evolution. Therefore, to more accurately assess the overall differences in "growth patterns" among different students, this invention selects Fréchet distance as a tool for calculating the similarity between trajectory pairs.
[0033] In this embodiment, let the first... The students and the first The physical fitness trajectories of the students are represented as follows: ; ; in, , Students , The trajectory sequence; , The number of time points for each trajectory; For student index number; This refers to the sequence number of the physical fitness test; For the first The number of students The time point of the next physical fitness measurement; For the first Students at the time point The corresponding physical fitness measurement index values.
[0034] In one possible implementation, the Fréchet distance is defined. The calculation method is as follows: ; in, Students physical fitness trajectory With students physical fitness trajectory The Fréchet distance between them; and Let represent the matching path function between trajectories, which is monotonically non-decreasing; , Indicates the normalized path parameters The trajectory interpolation value; This represents the minimum upper bound among all valid matching paths; Indicates the path matching parameters Find the maximum distance within the range.
[0035] Specifically, the Fréchet distance takes into account both the overall alignment effect and local variation of the path, making it particularly suitable for comparing the physical development trajectories of students with similar structures but asynchronous sampling nodes.
[0036] In some embodiments, to accelerate the calculation of the aforementioned distance, the present invention employs a grid dynamic programming algorithm for the path function. , An optimal pairing search is performed, and a pairing matching table is constructed by minimizing the maximum value through double nesting, thereby achieving an approximately accurate Fréchet distance estimation.
[0037] As an alternative, for longer time series trajectories, a trajectory downsampling mechanism can be introduced before calculation. For example, key point extraction methods (such as curvature extrema points and turning points) can be used to simplify the trajectory, thereby reducing the computational complexity of the distance matrix while ensuring the preservation of similar structures.
[0038] Ultimately, through all The system pairs the student trajectories together and generates a trajectory similarity distance matrix. Its element definition: ; in, Students With students The Fréchet distance between them; Students physical fitness trajectory With students physical fitness trajectory The Fréchet distance between them.
[0039] This distance matrix serves as the direct input to subsequent manifold embedding algorithms, enabling the construction of a low-dimensional representation space for student trajectories and thus achieving data-driven modeling of "growth patterns".
[0040] It should be noted that this step does not normalize the physical fitness index values; instead, it directly uses the differences in the original indicators for path comparison, thereby enhancing the responsiveness to the actual increase in the indicators. However, if there are inconsistencies in units in the application scenario, these can be handled uniformly before trajectory construction, which will not be elaborated here.
[0041] S3. Based on the distance matrix, the manifold embedding algorithm is used to perform low-dimensional spatial mapping of the students' physical fitness trajectory, and obtain the trajectory embedding representation of each student while maintaining the trajectory structure features. A multidimensional time series structure has been constructed based on student physical fitness trajectories sampled at non-uniform time. The similarity of trajectories among all students has been quantified using the Fréchet distance metric, forming a complete trajectory distance matrix. To further explore the distribution structure and hidden patterns inherent in the student physical fitness evolution trajectories and to achieve effective representation of individual students in the evolution space, this invention introduces a low-dimensional manifold embedding algorithm. This algorithm maps the original high-dimensional structural relationships to the embedding space, generating student embedding representations that preserve the trajectory structure features.
[0042] In general, the similarity matrix between the original trajectories The structural information contained within these trajectories is difficult to directly use for model learning, especially in tasks such as group comparison, clustering, or evolutionary trend modeling, where the curse of dimensionality exists. To overcome this limitation, this invention constructs a low-dimensional mapping mechanism based on the distance-preserving principle, which can compress the trajectory of each student into a low-dimensional vector while preserving its relative structural position in the original trajectory space as much as possible.
[0043] In this embodiment, let there be a total Let there be n students, and let their trajectory similarity distance matrix be: ; in, Students With students The distance of Fréchet's trajectory; It is a real symmetric matrix.
[0044] In one possible implementation, this invention employs an isometric mapping algorithm (Isomap) for low-dimensional manifold embedding. This method obtains the embedding coordinates of each student through classical multidimensional scaling (MDS) while preserving the original distance relationships.
[0045] Specifically, firstly, the distance matrix Convert to inner product matrix The calculation formula is: ; in, It is an element-wise squared matrix; For a centered matrix; It is an inner product matrix.
[0046] Subsequently, the inner product matrix Perform eigenvalue decomposition: ; in, It is a diagonal eigenvalue matrix; This is the corresponding eigenvector matrix.
[0047] In some embodiments, if outlier trajectories or abnormal data fluctuations exist, a local manifold embedding method can be introduced for robust embedding optimization to enhance the stability and structural expressiveness of the low-dimensional representation.
[0048] It should be noted that this embedding result can be used not only as model input for predicting subsequent physical fitness growth trends, but also for cluster analysis, visualization, or cross-semester intervention strategy design. Therefore, its construction should possess stability, consistency, and interpretability.
[0049] As an extension, a kernel-based nonlinear mapping mechanism can also be introduced into the embedding process to modify the inner product matrix. The kernel matrix is replaced with a Gaussian kernel or a polynomial kernel to preserve the nonlinear trajectory difference structure. This mechanism is suitable for scenarios with high-dimensional index combinations or complex coupling relationships between physical fitness indicators.
[0050] S4. Based on the trajectory embedding representation of the target student and combined with the embedding representation of known samples, the kernel regression method is used to estimate the trend of physical fitness changes of the target student at a future preset time point. After completing the structured representation and low-dimensional embedding of students' physical fitness trajectories, this invention designs a kernel regression method based on the embedded spatial structure to estimate the changes in physical fitness of target students at specific future time points, in order to further achieve trend prediction of individual physical fitness growth paths. This method fully utilizes the structural features preserved by the trajectory embedding, making the prediction process not only dependent on single-point information but also reflecting its relative evolutionary position within the peer group, thereby enhancing the stability and generalization ability of the prediction results.
[0051] Generally, students' future physical development is subject to certain uncertainties, limited by issues such as sparse historical data and strong individual differences. Traditional prediction methods based on unidirectional extrapolation of time series data struggle to capture the underlying growth trend structure, especially exhibiting unstable performance in non-uniform time sampling scenarios. Therefore, this invention, based on trajectory embedding results, introduces a kernel regression strategy to construct a physical fitness prediction model, effectively integrating common change patterns from the historical trajectories of multiple students.
[0052] In this embodiment, the target student is set as follows: The embedding is represented as: ; in, To embed spatial dimensions; This indicates that the student is embedded.
[0053] Suppose that the system is known to have For a given set of historical students with complete trajectories and future trend samples, their embedding representations and corresponding future physical fitness values are denoted as follows: ; in, Represents history students; This indicates that it is at the preset prediction time point. Physical fitness value; For sample index; The total number of samples is the number of historical students used for training.
[0054] Based on the target student embedding representation With history students Calculate the weight function based on the Euclidean distance between them: ; in, This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; the larger the value, the more similar they are. Indicates the target student and the first The square of the Euclidean distance between the students; It is a natural exponential function; This represents the bandwidth parameter of the kernel function.
[0055] Using the sample's predicted physical fitness values at the time point A kernel-weighted estimate is then performed using the weights mentioned above: ; in, To predict physical fitness values at specific time points; This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; the larger the value, the more similar they are. For sample index; The total number of samples is the number of historical students used for training.
[0056] It is suitable for scenarios involving the joint prediction of multiple physical fitness indicators. For example, it can simultaneously predict the future performance of two indicators: the 800-meter run and pull-ups. In this case, the prediction output is a mapping function from the embedding vector to the indicator space.
[0057] It should be noted that kernel regression models do not directly rely on the sequential information of points within the time series, but rather estimate location based on the structure of the embedding space, exhibiting robustness against time window misalignment. This method is also suitable for predicting student trajectories with significant data gaps, as it does not require complete historical sequence input.
[0058] S5. Based on the predicted results of physical fitness change trends, generate the growth trend curve of the target student in the future time interval, and further output the corresponding personalized assessment suggestions. By constructing an embedded representation of physical fitness trajectory and combining it with a kernel regression algorithm, the physical fitness indicators of target students at future time points were predicted. Building upon this, to comprehensively reflect students' growth trends over a future period and enhance the practicality of data interpretation, this invention further proposes a trend curve generation mechanism based on the prediction results, while simultaneously outputting personalized physical fitness development assessment suggestions. This aims to provide auxiliary decision support for physical education teaching, training plan design, and individual growth record management.
[0059] Generally, physical fitness indicators exhibit non-linear evolution over time and are influenced by a variety of external and internal factors. Therefore, relying solely on single-point predictions is insufficient to comprehensively depict a student's growth trajectory. This invention addresses this by continuously predicting at multiple time points and generating trend curves using interpolation or fitting methods, thereby outputting a dynamic representation of individual physical fitness growth over a continuous time interval.
[0060] In this embodiment, the target student is set as follows: The predicted time series is: ; in, Indicates the future number A preset time node; To predict the number of time points; For predicting time point series, it is represented as a set.
[0061] Correspondingly, the predicted value sequence of physical fitness indicators is as follows: ; in, Indicates the first Physical fitness prediction vectors at each time point; The predicted sequence of physical fitness index values represents the set of predicted outputs of the model at multiple future time points; This represents the total number of time points to be predicted, which is also the length of the output sequence.
[0062] In one possible implementation, a cubic spline interpolation method is used to construct the trend curve for each physical fitness indicator: ; in, For physical fitness; For the first Spline functions for segments of intervals; This represents the total number of time points to be predicted, which is also the length of the output sequence; This is the index of the interval segment, used to enumerate the interpolation segments between adjacent prediction time points; For indicator functions, when The value is 1 for the current time and 0 for other times, which controls which interpolation interval the current time belongs to.
[0063] Specifically, each trend curve is visualized and provided to teachers, parents, or students for reference, showing their physical development trend over the next few periods.
[0064] In some embodiments, to enhance the operability of the prediction results, the present invention further designs an assessment suggestion generation module based on a rule engine, which combines the student's current physical performance, the predicted trend slope and the results of comparative analysis with the same group to output personalized intervention suggestions.
[0065] For example, if a student's performance in a certain metric, such as the 800-meter run, shows a negative downward trend within the prediction period, and the historical value is lower than the average level for the same age, the system will output the following suggestions: Assessment Recommendations: Current endurance indicators show a downward trend. It is recommended to schedule at least two moderate-intensity long-distance running training sessions per week, and combine them with physical activities during breaks to improve cardiopulmonary function. Intervention suggestion: Combine heart rate monitors or smart bracelets to monitor training effects and moderately control the heart rate in the training zone to keep it within the target heart rate range.
[0066] As an extension mechanism, the personalized suggestion output module can set suggestion triggering conditions based on a multi-dimensional rule matrix, for example: A downward trend + indicators below the group average; The trend is upward, but the initial value is below the threshold. The trend is flat and the indicators are in a dangerous range.
[0067] The mechanism has the ability to update rules and can dynamically adjust parameters according to grade level, gender, and training seasonality to ensure that the assessment recommendations are real-time and adaptable.
[0068] It should be noted that the generated trend curves and evaluation suggestions should be stored in the individual's growth portfolio for subsequent periodic evaluations and longitudinal comparative analysis, and can also serve as an important basis for feedback on training effectiveness.
[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for intelligent evaluation of student exercise data across cycles, characterized in that, Includes the following steps: Collect students' physical fitness data over multiple time periods to construct a non-uniform time-sampled motion trajectory dataset containing timestamps and corresponding physical fitness indicators; Based on the constructed trajectory dataset, the similarity of the movement trajectories of any two students is measured, and a trajectory similarity distance matrix reflecting the differences in trajectory evolution is generated. Based on the distance matrix, a manifold embedding algorithm is used to perform low-dimensional spatial mapping on the students' physical fitness trajectories, thereby obtaining the trajectory embedding representation of each student while preserving the trajectory structure features. Based on the trajectory embedding representation of the target students and combined with the embedding representation of known samples, the kernel regression method is used to estimate the trend of physical fitness changes of the target students at future preset time points; Based on the predicted results of physical fitness change trends, the system generates growth trend curves for target students over future time intervals and further outputs corresponding personalized assessment suggestions.
2. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The physical fitness data includes: The assessment indicators in the National Student Physical Fitness and Health Standards include vital capacity, 50-meter sprint, standing long jump, sit-and-reach, pull-ups, 800-meter and 1000-meter running times. Exercise behavior data collected through wearable devices during classroom and extracurricular exercise, including cadence, heart rate, running distance and exercise duration; The timestamp information associated with the above indicators is used to construct a non-uniform time sampling trajectory sequence.
3. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The trajectory dataset includes: A set of time-series trajectories constructed for each student, each trajectory is represented as: ； in, Indicates the first The number of students The time point of the next physical fitness measurement; Indicates physical fitness index values; Indicates the first A collection of time-series trajectories of 100 students; Indicates the first A total of [number] students conducted [number] [activities]. Secondary physical fitness test; For student index number; This is the sequence number of the physical fitness test.
4. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The low-dimensional space mapping includes: Apply the manifold embedding algorithm to the trajectory similarity distance matrix; The manifold embedding algorithm is selected from diffusion mapping; The mapping result assigns a low-dimensional representation vector to each trajectory.
5. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The trajectory embedding representation includes: In the low-dimensional space, the local neighborhood relationships and overall evolutionary form of the original trajectory are preserved; The embedding is represented as This is used for subsequent similarity calculations and as input to the prediction model; The embedding is represented as It is scalable, meaning that new students can obtain their representations through incremental embedding without retraining the entire embedding space.
6. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The kernel regression method includes: Based on the target student embedding representation With history students Calculate the weight function based on the Euclidean distance between them: ； in, This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; a larger value indicates greater similarity. Indicates the target student and the first The square of the Euclidean distance between the students. It is a natural exponential function. The bandwidth parameter representing the kernel function; Using the sample's predicted physical fitness values at the time point A kernel-weighted estimate is then performed using the weights mentioned above: ； in, To predict physical fitness values at specific time points; This represents the Gaussian kernel function, used to measure the similarity between the target student and the historical students in the embedding space; the larger the value, the more similar they are. For student index number; The total number of samples is the number of historical students used for training.
7. The intelligent evaluation method for cross-cycle student motion data according to claim 1, characterized in that, The trends in physical fitness changes include: Estimated curves of continuous physical fitness indicators over a future period of time; The trends reflect the slope, acceleration, and phased changes in students' individual physical fitness over time; The trend data supports the detection of physical fitness inflection points, used to assess growth bottlenecks and excessive fatigue.
8. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The growth trend curve includes: A continuous curve generated based on multiple predicted time points; The curve, with time on the horizontal axis and predicted physical fitness value on the vertical axis, is smooth and differentiable. A confidence band is introduced during the curve plotting process to represent the range of prediction uncertainty.
9. The intelligent evaluation method for cross-cycle student motion data according to claim 1, characterized in that, The personalized assessment recommendations include: Rating of the current rate of physical fitness growth; Recommended intensive training programs and suggested training frequency; Early warnings and rest recommendations are issued for detected downward trends; Provide training path reference templates based on the evolution of similar samples for different student types.
10. The intelligent evaluation method for cross-cycle student sports data according to claim 1, characterized in that, The similarity measure includes: Based on the trajectory data sampled in a non-uniform time, a trajectory alignment method is used to perform multi-point correspondence matching, and then the dynamic evolution difference value between different student trajectories is calculated. The difference value is based on the maximum distance between corresponding points in each pair of trajectories and is used to measure the degree of deviation between the two trajectories in their overall shape. By calculating the difference values mentioned above, a trajectory similarity distance matrix is constructed, providing an input basis for subsequent low-dimensional space mapping.