Athlete body function evaluation system and method based on bidirectional data closed loop

Abstract

The invention provides an athlete body function evaluation system and method based on a bidirectional data closed loop, and relates to the technical field of data processing. The system comprises a data management module used for collecting multi-source data and performing time sequence alignment on the multi-source data to generate target to-be-evaluated data; wherein the multi-source data is used for representing multiple types of data of body functions of athletes; the large model service module is used for generating a corresponding preliminary evaluation conclusion based on the target to-be-evaluated data; wherein the preliminary evaluation conclusion comprises an interpretable evidence chain; and the evaluation optimization module is used for performing online revision and knowledge labeling on the preliminary evaluation conclusion so as to determine optimization information and output a corresponding target evaluation conclusion. According to the system, data pushing, intelligent evaluation, expert auditing and model optimization are integrated, the defects of unidirectional transmission and static evaluation of a traditional system are overcome, and then the accuracy of physical function evaluation of athletes and the self-evolution ability of the model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, for example to a player physical function evaluation system and method based on a bidirectional data closed loop. BACKGROUND

[0002] In the field of competitive sports and sports science, player physical function evaluation is a core link of scientific training. In the field of player physical function monitoring and evaluation, the technology development has experienced a stage from single data collection to preliminary analysis. In the early stage, mainly relying on simple physiological index measurement equipment such as heart rate belt, sphygmomanometer, etc., the basic physiological data of the players are collected, but these data can only provide basic health information, lacking the ability of in-depth analysis and comprehensive evaluation.

[0003] With the progress of computer technology and data analysis methods, in the related technology, some systems that can integrate multiple physiological data and perform preliminary analysis begin to appear. However, these systems mostly adopt a one-way data transmission mode, that is, data flows from the collection device to the evaluation model, and the evaluation results are output in a fixed report form, lacking effective interaction and feedback mechanism with professional field knowledge. SUMMARY

[0004] The present application aims to provide a player physical function evaluation system and method based on a bidirectional data closed loop, which can improve the evaluation accuracy in the form of effective interaction and feedback mechanism with professional field knowledge.

[0005] According to an aspect of the present application, a player physical function evaluation system based on a bidirectional data closed loop is proposed, comprising: a data management module for collecting multi-source data and performing time sequence alignment on the multi-source data to generate target evaluation data; wherein the multi-source data is used to represent multiple types of data of player physical function; a large model service module for generating a corresponding preliminary evaluation conclusion based on the target evaluation data; wherein the preliminary evaluation conclusion contains an explainable evidence chain; an evaluation optimization module for online revision and knowledge annotation of the preliminary evaluation conclusion to determine optimization information and output a corresponding target evaluation conclusion.

[0006] According to an aspect of the present application, a player physical function evaluation method based on a bidirectional data closed loop is proposed, comprising: collecting multi-source data and performing time sequence alignment on the multi-source data to generate target evaluation data; wherein the multi-source data is used to represent multiple types of data of player physical function; generating a corresponding preliminary evaluation conclusion based on the target evaluation data; wherein the preliminary evaluation conclusion contains an explainable evidence chain; The preliminary evaluation conclusion is revised and knowledge-labeled online to determine the optimized information and output the corresponding target evaluation conclusion.

[0007] According to an aspect of the present application, an electronic device is provided, comprising: a processor; a memory storing a computer program, when the computer program is executed by the processor, the processor executes the method as described above.

[0008] According to an aspect of the present application, a non-transitory computer readable medium is provided, which stores readable instructions, when the instructions are executed by a processor, the processor executes the method as described above.

[0009] It should be understood that the above general description and the following detailed description are only exemplary and do not limit the present application.

[0010] Through the above embodiments provided by the present application, the system of the present application constructs a bidirectional data closed loop, and the output result of the evaluation optimization module can be fed back to the data management module and the large model service module, forming a cycle of data feedback and model optimization. On the one hand, the data problems and model deficiencies found in the evaluation optimization process can guide the data management module to improve the data acquisition and processing method and improve the data quality; on the other hand, according to the evaluation optimization result, the large model is adjusted and knowledge is updated online, and the performance and accuracy of the model are continuously improved. This bidirectional data closed loop enables the system to continuously self-improve and optimize, adapt to the dynamic changes of the athlete's physical function and the needs of different training stages. Through the cooperative work of the three modules, the system realizes the comprehensive dynamic evaluation of the athlete's physical function. From data acquisition to preliminary evaluation, to evaluation optimization, the whole process covers various aspects of information of the athlete's physical function, and can track its changes in real time. This comprehensive dynamic evaluation helps coaches, athletes and medical personnel to understand the physical condition of athletes in a timely manner, to develop scientific and reasonable training and rehabilitation plans, and to improve the competitive level and health level of athletes. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art according to these drawings without departing from the scope of the present application.

[0012] Figure 1 A block diagram of an athlete physical function evaluation system based on a bidirectional data closed loop is provided for the embodiments of the present application; Figure 2 Another overall block diagram of an athlete physical function evaluation system based on a bidirectional data closed loop is provided for the embodiments of the present application; Figure 3 A flow chart of the method for evaluating the physical function of an athlete based on a bidirectional data closed loop provided by an embodiment of the present application; Figure 4 A flow chart of the model optimization provided by an embodiment of the present application; Figure 5 An architecture diagram of the system for evaluating the physical function of an athlete based on a bidirectional data closed loop provided by an embodiment of the present application; Figure 6 A structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0013] Example embodiments now will be described more fully hereinafter with reference to the accompanying drawings. Example embodiments, however, can be implemented in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example embodiments to those skilled in the art. Like reference numerals refer to like elements throughout the description.

[0014] Moreover, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the application can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, devices, implementations, and operations have not been shown or described in detail to avoid obscuring aspects of the application.

[0015] The block diagrams in the drawings show functions and functionality as they can exist in one or more embodiments. They can be implemented as software, hardware, firmware, or a combination thereof. In the context of software, the blocks represent computer program steps implemented on one or more network devices. The functions can be implemented using software, hardware, firmware, or a combination thereof.

[0016] The flow diagrams depicted herein are examples of sequences of operations that can be performed by some embodiments of the application. The depicted examples are not meant to be limiting in terms of operations or order. For example, one or more of the depicted operations could be performed in other orders or could be performed concurrently. Furthermore, one or more of the depicted operations could be omitted. In some examples, one or more additional operations could be added.

[0017] It should be understood that although the terms first, second, third, etc. can be used herein to describe various components, these components should not be limited by these terms. These terms are used only to distinguish one component from another. Thus, a first component discussed below could be termed a second component without departing from the teachings of the present application. As used herein, the term "and / or" includes any and all combinations of associated items, and / or one or more of the associated items.

[0018] In recent years, athlete physical function evaluation technology, early single machine analysis, mainly rely on laboratory equipment to collect static physiological data (such as blood lactic acid peak value), using statistical methods to generate evaluation report, there are data timeliness is poor, evaluation dimension single problem. With the discovery of science and technology, gradually began to use functional cloud monitoring stage, that is, through wearable devices to realize continuous data acquisition, but the evaluation model only supports one-way batch data processing, lack of real-time interaction ability. In the stage of artificial intelligence development, combined with for example SportsGPT (Sports General Performance Transformer, Sports General Performance Transformer) introduces natural language interaction, carries on the function evaluation, but has not solved the key problem of expert knowledge feedback model.

[0019] The existing system cannot realize the two-way closed loop of "evaluation-expert correction-model iteration", which leads to the stagnation of model evolution. For example, when the blood lactic acid data and subjective fatigue degree appear contradiction, the system cannot actively request expert intervention, nor can it convert expert correction opinions into model parameters.

[0020] Based on this, the application provides an athlete physical function evaluation system and method based on a two-way data closed loop.

[0021] The specific implementation mode can refer to the following embodiments.

[0022] Figure 1 The block diagram of the athlete physical function evaluation system based on the two-way data closed loop provided by the embodiments of the application is shown in the figure. Figure 1 As shown in the figure, the system includes a data management module 10, a large model service module 11 and an evaluation optimization module 12.

[0023] The data management module 10 is used to collect multi-source data and time-align the multi-source data to generate target data to be evaluated; wherein the multi-source data is used to represent multi-class data of athlete physical function.

[0024] The present application can collect physiological data of athletes in real time through wearable devices such as smart bands, smart watches, heart rate bands, etc. These devices are usually equipped with sensors that can collect data at a certain frequency (such as once per second) and transmit the data to the data management module 10 through wireless communication technologies such as Bluetooth, Wi-Fi, etc. Motion sensors such as accelerometers, gyroscopes, etc. are installed on the athlete's sports equipment (such as shoes, rackets, bicycles, etc.) to collect motion data such as athlete's speed, acceleration, motion trajectory, step count, jumping height, etc. At the same time, the athlete's movement process is recorded through video capture devices (such as high-definition cameras), and the video is analyzed through computer vision technology to extract data such as athlete's action posture, motion amplitude, etc. During the training process of the athlete, the training management system records training data such as training items, training intensity, training duration, training load, etc. The training management system can interact with the terminal device of the coach, and the coach can input the training plan and related data on the terminal device and synchronize the data to the data management module 10. Regularly collect biological samples such as blood and urine of athletes, and detect biochemical indicators such as creatine kinase, lactic acid, and hemoglobin of athletes through laboratory analysis equipment (such as blood analyzers, urine analyzers, etc.). These data can be transmitted to the data management module 10 through manual input or by connecting with the laboratory information system (LIS). Based on this, the collection of multi-source data is realized.

[0025] When collecting multi-source data, an accurate timestamp can be added to each data point. The timestamp can use Global Positioning System (GPS) time or Network Time Protocol (NTP) time to ensure the time consistency of different data sources. Time synchronization algorithms such as timestamp-based synchronization algorithm and event-triggered synchronization algorithm are used to align the time sequence of multi-source data.

[0026] The timestamp-based synchronization algorithm can arrange the data in chronological order by comparing the timestamps of different data sources. The event-triggered synchronization algorithm associates the data of different data sources with a specific event (such as the athlete's starting, jumping, etc.) to achieve time sequence alignment of data.

[0027] In some implementations, during data collection, due to the differences in collection frequency and data transmission delay of different data sources, the data management module 10 needs to set up a data buffer area to temporarily store the collected data. At the same time, data processing techniques such as data cleaning, data filtering, data interpolation, etc. are used to process the buffered data in real time to ensure the accuracy and integrity of the data.

[0028] The multi-source data aligned in time sequence can be fused by preset methods such as weighted average method, Kalman filtering method, neural network fusion method, etc. to generate target evaluation data. The generated target evaluation data is stored in a database for subsequent analysis and processing. The database can be a relational database or a non-relational database, which is selected according to the type and size of the data.

[0029] The large model service module 11 is configured to generate a corresponding preliminary evaluation conclusion based on the target evaluation data; wherein the preliminary evaluation conclusion includes an explainability evidence chain.

[0030] The application can pre-select a large model suitable for athlete physical function evaluation, such as a pre-trained language model based on the Transformer architecture or a deep learning model developed specifically for the sports health field. These models have strong feature extraction and classification capabilities and can analyze and process complex multi-source data. The selected large model is deployed on a server or cloud computing platform for real-time processing of target evaluation data. During deployment, the model needs to be optimized and adjusted to improve the performance and efficiency of the model. For example, model quantization, model pruning and other techniques are used to compress the model, reduce the computational load and storage space of the model; distributed computing technology is used for parallel processing of the model to improve the processing speed of the model.

[0031] Feature extraction can be performed on the target evaluation data using principal component analysis, linear discriminant analysis, or deep learning methods to convert the original data into feature vectors suitable for processing by the large model. For example, for physiological data, features such as heart rate variability and blood pressure fluctuations can be extracted; for sports data, features such as movement speed and acceleration trends can be extracted. The extracted feature vectors are standardized to ensure that different features have the same scale and distribution.

[0032] The standardized feature vectors are input into the large model for model inference. The large model generates a preliminary evaluation conclusion based on the input feature vectors and the knowledge and rules within the model. The preliminary evaluation conclusion can include the athlete's physical function status, such as normal, fatigue, injury risk, and sports ability level, such as endurance, strength, speed, etc.

[0033] To make the preliminary evaluation conclusion interpretable, the large model service module 11 needs to generate an interpretable evidence chain. The interpretable evidence chain can be realized in various ways, such as attention mechanism visualization, feature importance analysis, decision rule extraction, etc. Attention mechanism visualization can show the degree of attention of the large model to different features when generating the preliminary evaluation conclusion; feature importance analysis can evaluate the contribution of each feature to the preliminary evaluation conclusion; decision rule extraction can convert the decision-making process of the large model into a rule form that is easy to understand.

[0034] The evaluation optimization module 12 is used to revise and knowledge label the preliminary evaluation conclusion online to determine optimization information and output the corresponding target evaluation conclusion.

[0035] In this application, the evaluation optimization module 12 can collect feedback from users, such as relevant experts, on the preliminary evaluation conclusion through a user interface, such as a webpage, a mobile application, etc. Users can modify, supplement or ask questions about the preliminary evaluation conclusion on the interface and submit feedback to the evaluation optimization module 12. According to the user feedback, the evaluation optimization module 12 adjusts the large model online. The adjustment method can include parameter fine-tuning, model structure adjustment, etc. Parameter fine-tuning refers to local adjustment of the parameters of the large model to improve the performance of the model on specific tasks; model structure adjustment refers to modifying the network structure of the large model, such as adding or reducing network layers, adjusting the connection method of network layers, etc., to improve the generalization ability and adaptability of the model.

[0036] The evaluation optimization module 12 can integrate professional knowledge in the field of athlete physical function evaluation and establish a knowledge base. The knowledge base can include knowledge in the fields of sports physiology, sports training, sports medicine, etc., as well as historical data, training plans, competition results, etc. of athletes. Correlate the field knowledge with the preliminary evaluation conclusion and knowledge label the preliminary evaluation conclusion. Knowledge labeling can use natural language processing technology to associate field knowledge in the form of text, image, video, etc. with the preliminary evaluation conclusion; it can also use knowledge graph technology to build a knowledge graph from the field knowledge and link it to the preliminary evaluation conclusion.

[0037] According to the results of online revision and knowledge labeling, the evaluation optimization module 12 determines the optimization information. The optimization information can include the adjustment parameters of the model, the results of knowledge labeling, the processing of user feedback, etc. The optimization information is fused with the preliminary evaluation conclusion to generate the target evaluation conclusion. The target evaluation conclusion is a more accurate and comprehensive evaluation result of the athlete's physical function, with higher credibility and interpretability. The evaluation optimization module 12 outputs the target evaluation conclusion to the user through the user interface for decision-making and guidance.

[0038] The application constructs a bidirectional data closed loop, and the output results of the evaluation and optimization module can be fed back to the data management module and the large model service module to form a cycle of data feedback and model optimization. On the one hand, the data problems and model deficiencies found in the evaluation and optimization process can guide the data management module to improve data collection and processing methods and improve data quality. On the other hand, according to the evaluation and optimization results, the large model is adjusted and updated online, and the performance and accuracy of the model are continuously improved. This bidirectional data closed loop enables the system to continuously self-improve and optimize, adapt to the dynamic changes of the athlete's physical functions and the needs of different training stages. Through the cooperative work of the three modules, the system realizes comprehensive dynamic evaluation of the athlete's physical functions. From data collection to preliminary evaluation, to evaluation and optimization, the entire process covers various aspects of information of the athlete's physical functions and can track their changes in real time. This comprehensive dynamic evaluation helps coaches, athletes and medical personnel to understand the physical condition of athletes in a timely manner, develop scientific and reasonable training and rehabilitation plans, and improve the competitive level and health level of athletes.

[0039] According to some embodiments, reference is made to Figure 2 The athlete physical function evaluation system based on bidirectional data closed loop further comprises a model optimization module 13 for determining a reinforcement learning reward signal according to the optimization information.

[0040] In this application, the reward signal is designed according to the optimization information (i.e. the specific revision degree and the explainability evidence chain) of the preliminary evaluation conclusion of the evaluation and optimization module 12. If the preliminary evaluation conclusion is highly consistent with the final target evaluation conclusion, it means that the evaluation accuracy is high, and a positive reward is given; on the contrary, if the revision range is large, it means that the preliminary evaluation has a large deviation, and a negative punishment is given. For example, set an evaluation error threshold, when the error between the preliminary evaluation conclusion and the target evaluation conclusion is within the threshold range, a certain value of positive reward is given; when the error exceeds the threshold, a negative punishment in proportion to the degree of excess is given.

[0041] In some implementations, the influence of the completeness of the explainability evidence chain in the preliminary evaluation conclusion on the reward signal is considered. A complete explainability evidence chain can enhance the credibility and practicality of the evaluation conclusion, so if the preliminary evaluation conclusion contains detailed and reasonable explainability evidence chain, a positive reward is given; if the evidence chain is incomplete or lacks reasonableness, a negative punishment is given. The value of the reward or punishment can be determined according to the scoring results by setting a scoring standard for the completeness of the evidence chain.

[0042] The optimization information includes online revision of the preliminary evaluation conclusion, and when the model optimization module determines the reward signal based on the optimization information, the bias in the evaluation can be quantified and fed back to the reinforcement learning model. The explainable evidence chain in the optimization information provides rationality support for the evaluation conclusion. The optimized target evaluation conclusion reflects the matching degree of the athlete's physical function and the training target. The model optimization module determines the reward signal based on this information, which can guide the reinforcement learning model to generate evaluation results that are more in line with the training target.

[0043] According to some embodiments, with reference to Figure 2 The model optimization module 13 includes a feedback filtering submodule 131 and an incremental learning scheduling submodule 132.

[0044] The feedback filtering submodule 131 is configured to determine an effective feedback amount based on the preset identity attribute of the user corresponding to the historical optimization information corresponding to the preliminary evaluation conclusion and the corresponding optimization score of the historical optimization information, and determine a training set from the preliminary evaluation conclusion based on the effective feedback amount. The incremental learning scheduling submodule 132 is configured to perform model fine-tuning when the effective feedback amount meets a preset adjustment threshold.

[0045] In the present application, a special data storage system can be constructed to collect and store historical optimization information corresponding to the preliminary evaluation conclusion. These information includes but is not limited to revision of the preliminary evaluation conclusion, knowledge annotation information, and finally determined target evaluation conclusion, etc. At the same time, record the preset identity attribute of the user corresponding to each piece of historical optimization information, such as the age, gender, sports project, training time limit of the athlete, etc. Associate an optimization score with each piece of historical optimization information, which can be based on the automatic evaluation result of the evaluation optimization module, or scored by professional coaches or experts according to the accuracy and practicality of the evaluation.

[0046] The feedback filtering submodule 131 includes a pre-established feedback weight calculation model, which comprehensively considers the preset identity attribute of the user and the corresponding optimization score of the historical optimization information. For example, the scoring of experts with different abilities and identities has different reliability weights; for athletes of different sports projects, the characteristics of their physical functions and evaluation needs may differ, so different weights are given to the historical optimization information of athletes of different sports projects. At the same time, historical optimization information with higher optimization score means higher quality and reliability, and should also be given greater weight.

[0047] According to the feedback weight calculation model, the feedback weight of each piece of historical optimization information is calculated, and then the feedback weights of all historical optimization information are added to obtain the total effective feedback amount. A threshold of effective feedback amount is set, and when the total effective feedback amount exceeds the threshold, it is considered that the currently collected historical optimization information is sufficient for model training. According to the effective feedback amount, representative data is selected from the preliminary evaluation conclusion as a training set. Random sampling, stratified sampling and other methods can be used to ensure that the data in the training set can cover different body function states and evaluation scenarios. For example, for the evaluation of heart rate indicators, the training set should include data of different situations such as normal heart rate, high heart rate and low heart rate.

[0048] The real-time monitoring feedback filtering submodule 131 calculates the effective feedback amount. When the effective feedback amount meets the preset adjustment threshold, the model fine-tuning operation of the incremental learning scheduling submodule 132 is triggered. The preset adjustment threshold can be dynamically adjusted according to the actual needs of the system and the performance of the model. For example, in the early stage of the system, due to the small amount of data, the adjustment threshold can be set lower to fine-tune the model as soon as possible; as the amount of data increases, the adjustment threshold can be appropriately increased to reduce unnecessary model fine-tuning times.

[0049] Before model fine-tuning, the training set is preprocessed, including data cleaning, feature extraction, data standardization and other operations, to improve the efficiency and accuracy of model training. A suitable incremental learning algorithm is selected to fine-tune the model. The incremental learning algorithm can update and optimize the model using new training data without retraining the entire model. For example, an incremental learning algorithm based on gradient descent can be used to update the model parameters by calculating the gradient of the new training data. During model fine-tuning, the performance indicators of the model, such as evaluation accuracy and training error, are monitored in real time. If the performance indicators of the model do not meet the expected target, the parameters of the incremental learning algorithm can be adjusted or the training set can be reselected for fine-tuning.

[0050] The feedback filtering submodule of the present application determines the effective feedback amount and selects the training set, avoiding the use of a large amount of low-quality or redundant data for model training, thereby reducing the computational load and time cost of model training. The incremental learning scheduling submodule fine-tunes the model when the adjustment threshold is met, allowing new optimization information to be incorporated into the model in a timely manner, enabling the model to adapt more quickly to changes in data and accelerating the convergence process of the model. The training set selected by the feedback filtering submodule has high quality and representativeness, providing more accurate learning samples for the model, thereby improving the evaluation accuracy of the model for the body function of the athlete. The incremental learning scheduling submodule can continuously optimize the model according to new optimization information, enabling the model to continuously adapt to changes in the body function of the athlete and the development of evaluation needs, further improving the evaluation accuracy of the model.

[0051] According to some embodiments, the evaluation optimization module 12 is configured to determine the audit information of the target evaluation conclusion according to the preset identity attribute and the optimization score, and output the target evaluation conclusion and send the corresponding optimization information to the model optimization module if the actual audit process of the target evaluation conclusion meets the audit information.

[0052] The present application can construct a comprehensive database for storing the preset identity attribute information of athletes, such as age, gender, sports project, training years, body basic indicators (height, weight, etc.), and past physical function evaluation records. At the same time, the corresponding optimization score of each evaluation is collected and recorded, which can be scored by professional coaches, sports medicine experts, or specially trained evaluation systems according to the accuracy, comprehensiveness, practicality, and other dimensions of the evaluation. The collected preset identity attribute and optimization score data are preprocessed, including data cleaning (such as removing duplicate, incorrect, or excessive missing value data), data standardization, and other operations to ensure the quality and consistency of the data.

[0053] Machine learning algorithms such as decision trees, neural networks, or support vector machines are used to establish audit rule models based on historical data. The preset identity attribute and optimization score are used as input features, and the audit information of the target evaluation conclusion is used as output labels.

[0054] In some implementations, using the decision tree algorithm as an example, by analyzing and learning from historical data, the corresponding audit information rules under different combinations of preset identity attributes and optimization scores are determined. For example, for young athletes with shorter training years and lower optimization scores, the audit information of the target evaluation conclusion may require a more rigorous audit process, including more expert review links; while for experienced athletes with higher optimization scores, the audit information may be relatively lenient, requiring only regular audits. During model training, cross-validation and other methods are used to evaluate and optimize the model to improve its accuracy and generalization ability. For example, the data set is divided into training and test sets, the model is trained on the training set, and the model's performance is evaluated on the test set, and the model's parameters and structure are adjusted according to the evaluation results.

[0055] The preset identity attribute of the current athlete and the optimization score of this evaluation are input into the trained audit rule model, and the model outputs the audit information of the target evaluation conclusion according to the preset rules and algorithms. The audit information can include audit levels (such as primary audit, intermediate audit, and advanced audit), audit personnel requirements (such as requiring specific field experts to audit), audit time requirements (such as completing the audit within a specified time), etc.

[0056] An audit process management system can be established in advance to monitor the actual audit process of the target evaluation conclusion in real time. The system can record the start time, end time, and operation records of the auditors, etc. When the audit process starts, the system initializes the audit task according to the audit information and assigns the audit requirements to the corresponding auditors or audit team. For example, if the audit information requires a senior expert to audit, the system will push the evaluation conclusion to a sports medicine expert with a senior title for audit.

[0057] During the audit process, the audit process management system compares and matches the actual audit process with the audit information in real time. For example, check whether the current audit level meets the audit level specified in the audit information, whether the auditor meets the requirements in the audit information, etc.

[0058] When the actual audit process meets all the requirements in the audit information, the system determines that the audit is passed, and generates a pass signal. After receiving the pass signal, the evaluation optimization module 12 outputs the target evaluation conclusion to the corresponding interface of the system for the athletes, coaches, training managers, and other relevant personnel to view and use. At the same time, the evaluation optimization module 12 sends the optimization information corresponding to this evaluation, including the preliminary evaluation conclusion, historical optimization information, knowledge annotation information, target evaluation conclusion, and related records in the audit process, to the model optimization module 13. The model optimization module 13 can further optimize and adjust the evaluation model according to these optimization information to improve the evaluation accuracy and adaptability of the model.

[0059] The present application can provide personalized audit processes for athletes with different characteristics and needs by determining audit information based on preset identity attributes and optimization scores. By considering the preset identity attributes and optimization scores, the audit information can reflect the individual differences and evaluation quality of the athletes. The optimization score reflects the optimization degree and quality of the evaluation process, and in combination with the preset identity attributes, the rationality of the target evaluation conclusion can be more comprehensively evaluated, thereby improving the accuracy and reliability of the audit.

[0060] According to some embodiments, with reference to Figure 2 The data management module 10 includes a dynamic time warping sub-module 101 and a multi-scale aggregation sub-module 102.

[0061] The dynamic time warping sub-module 101 is configured to automatically detect and remove abnormal segments caused by device signal loss of the multi-source data during time alignment according to an improved dynamic time warping algorithm, to determine the data to be aggregated; and the multi-scale aggregation sub-module 102 is configured to calculate feature statistics of the data to be aggregated according to a preset scale period layering, to determine the target evaluation data based on the feature statistics.

[0062] The traditional dynamic time warping (DTW) algorithm is mainly used to measure the similarity between two time series. By constructing a distance matrix, an optimal path is found to make the two sequences nonlinearly aligned on the time axis, so as to calculate the minimum cumulative distance between them.

[0063] To deal with the abnormal segment detection and elimination caused by the loss of device signal, the traditional DTW algorithm can be improved. When constructing the distance matrix, signal strength and quality indicators are introduced. For example, for the data at each time point, in addition to calculating the numerical distance with the corresponding point of another sequence, the strength of the device signal at this point (which can be measured by parameters such as signal amplitude and signal-to-noise ratio) and the signal quality (such as whether there is interference, distortion, etc.) are also considered.

[0064] In the process of calculating the optimal path, the dynamic time warping submodule 101 sets thresholds for signal strength and quality. When a time point with signal strength below the threshold or signal quality not meeting the requirements is encountered during path calculation, the data segment within a certain time range around this point is marked as a potential abnormal segment. Further analysis of the influence of these potential abnormal segments on the overall sequence similarity calculation is performed, and if it is found that a certain segment causes a significant deviation in sequence similarity calculation (for example, significantly increases the minimum cumulative distance), the segment is determined to be an abnormal segment.

[0065] Once the dynamic time warping submodule 101 detects an abnormal segment, it uses interpolation or direct truncation to eliminate it. For the interpolation method, linear interpolation, spline interpolation, etc. can be selected to generate reasonable replacement data according to the characteristics of the normal data before and after the abnormal segment, to maintain the continuity of the time series. If the abnormal segment has a large impact on the overall data and is difficult to accurately interpolate, the segment can also be directly truncated, but it is necessary to ensure that the remaining data after truncation still meets the requirements of subsequent analysis. After the abnormal segment elimination process, the remaining normal data is the data to be aggregated. These data are sorted and stored in chronological order to facilitate subsequent multi-scale aggregation operations.

[0066] The multi-scale aggregation submodule 102 presets multiple different scale periods according to the characteristics and application scenarios of the target data to be evaluated. For example, in the evaluation of athletes' physical functions, if the focus is on short-term changes in physical state, minute-level, hour-level, etc. shorter scale periods can be selected; if the focus is on long-term physical training effects and trends, day-level, week-level, month-level, etc. longer scale periods can be selected. The preset scale periods are layered in order from small to large, forming a multi-scale hierarchical structure. For example, the first layer is the minute-level scale, the second layer is the hour-level scale, the third layer is the day-level scale, and so on. This hierarchical design helps to comprehensively analyze the characteristics of the data to be aggregated from different time scales.

[0067] For each scale period layer, a plurality of characteristic statistics of the data to be aggregated are calculated. Common characteristic statistics include mean, variance, maximum value, minimum value, median, skewness, kurtosis, etc. For example, on a minute scale, the mean and variance of the data per minute are calculated to reflect the concentration trend and dispersion degree of the data in a short time; on a day scale, the maximum value and minimum value of the data per day are calculated to understand the fluctuation range of the data in a day. After calculating the characteristic statistics of the lower scale period layer, it is used as input data to calculate the characteristic statistics of the higher scale period layer. For example, the mean data calculated on the hour scale is used as the original data for day scale calculation, and the characteristic statistics on the day scale are further calculated. This hierarchical progressive calculation method can fully utilize the information between different scales and extract more representative and comprehensive features.

[0068] The characteristic statistics of each scale period layer calculated by multi-scale aggregation are comprehensively analyzed. The correlation and complementarity between different scale characteristic statistics are considered, for example, the large variance on a short-term scale may indicate the fluctuation of the body state, and the change of the mean value on a long-term scale may reflect the accumulation of training effect. According to the results of comprehensive analysis, data that can accurately reflect the state and characteristics of the target to be evaluated object are determined. A plurality of key characteristic statistics can be combined or further processed to generate the final target to be evaluated data. For example, the mean on the day scale and the variance on the month scale are combined by weighting to obtain a comprehensive evaluation index as the target to be evaluated data.

[0069] The dynamic time warping submodule of the present application automatically detects and removes abnormal segments caused by equipment signal loss through an improved algorithm, avoiding the interference of abnormal data on subsequent analysis and evaluation, and improving the accuracy and reliability of the data. The multi-scale aggregation submodule calculates and analyzes the data from multiple different scale periods, which can fully mine the characteristics of the data on different time scales. The short-term scale can capture the instantaneous changes and detailed information of the data, and the long-term scale can reflect the trend and periodicity of the data.

[0070] According to some embodiments, with reference to Figure 2 The large model service module 11 includes a hybrid architecture submodule 111 and an explainability output submodule 112.

[0071] The hybrid architecture submodule 111 is configured to extract high-frequency physiological indicator fluctuation characteristics in the target to be evaluated data and generate a key conclusion based on the high-frequency physiological indicator fluctuation characteristics. The explainability output submodule 112 is configured to output material data corresponding to a correlation degree greater than a preset correlation threshold.

[0072] In this application, the hybrid architecture sub-module 111 receives target data to be evaluated from the data management module 10, which may include time series data of various physiological indicators of athletes such as heart rate, blood pressure, blood oxygen saturation, etc. First, the data is cleaned to remove noise and outliers, such as using moving average filtering, median filtering and other methods to smooth the data, reducing the influence of measurement errors and transient interference. Then the data is standardized to unify the data range of different physiological indicators to a similar interval, facilitating subsequent analysis.

[0073] A variety of feature extraction algorithms are combined to extract features. For physiological indicators with strong periodicity, such as heart rate, Fourier transform is used to convert time domain signals to frequency domain signals to extract main frequency components and power spectrum features to capture the high frequency fluctuation rules of heart rate. For non-periodic or complex dynamic characteristics of physiological indicators, such as instantaneous changes in blood pressure, wavelet transform method is used. Wavelet transform can analyze the local features of signals at different scales to extract the fluctuation characteristics of blood pressure signals in different time-frequency ranges. At the same time, combined with time domain statistical feature extraction method, the mean, variance, maximum, minimum, etc. of physiological indicators in a certain time window are calculated to fully describe the fluctuation of high frequency physiological indicators.

[0074] Feature selection algorithms such as correlation-based feature selection (Correlation-based Feature Selection, CFS) or machine learning-based feature selection methods (such as random forest feature importance evaluation) are used to select the most relevant and representative high frequency physiological indicator fluctuation features from the extracted large number of features. Remove redundant and irrelevant features, reduce data dimension, and improve model calculation efficiency and generalization ability.

[0075] A hybrid model is constructed to combine the advantages of deep learning models and traditional machine learning models. For example, long short-term memory (Long Short-Term Memory, LSTM) is used as the deep learning part, which can handle long-term dependencies in time series data and effectively capture the trend of high frequency physiological indicator fluctuation features over time. At the same time, support vector machine (Support Vector Machine, SVM) is used as the traditional machine learning part, which has good performance in handling small sample data and classification problems. The extracted high frequency physiological indicator fluctuation features are input into the LSTM and SVM models to obtain the preliminary prediction results of the two models.

[0076] A result fusion mechanism is designed to weight the preliminary prediction results of LSTM and SVM. According to the performance of the two models in different evaluation scenarios, the weight coefficient is dynamically adjusted. For example, when evaluating the physical fatigue level of athletes, if historical data shows that LSTM has a higher accuracy in this task, then LSTM is given a larger weight; on the contrary, if SVM performs better in certain specific situations, then the weight of SVM is appropriately increased. Through result fusion, more accurate and comprehensive key conclusions are generated, such as the current physical state level of athletes (healthy, sub-healthy, fatigue, etc.), potential sports injury risks, etc.

[0077] The explainable output submodule 112 can collect various material data related to the target data to be evaluated, including the training plan of the athlete, the competition record, the historical physical examination report, the environmental data (such as the temperature and humidity of the training site), etc. These material data are stored and sorted in a structured manner, a unified data format and index are established, and the correlation degree calculation is facilitated. Various correlation degree calculation methods are adopted, such as Pearson correlation coefficient, mutual information, decision tree feature importance, etc. Pearson correlation coefficient is used to measure the linear correlation degree between two variables, which is suitable for correlation analysis between numerical material data and target data to be evaluated. Mutual information can capture the nonlinear relationship between variables, and has better adaptability to some complex correlation situations. Decision tree feature importance can evaluate the influence degree of each material data feature on the generation of key conclusions from the perspective of the model.

[0078] The results obtained by various correlation degree calculation methods are comprehensively evaluated, and weighted average or voting mechanism is used to determine the comprehensive correlation degree between each material data and the key conclusion. For example, according to the reliability and accuracy of different algorithms in specific scenarios, each algorithm is assigned a corresponding weight, and the weighted average correlation degree is calculated.

[0079] According to the actual application requirements and experience, a correlation threshold is preset. The setting of this threshold needs to consider the balance between the accuracy and explainability of the evaluation. If the threshold is set too high, some important correlation material data may be missed; if the threshold is set too low, a large amount of irrelevant or weakly correlated material data will be output, increasing the difficulty of user understanding. Through experiments and data analysis, a suitable threshold range can be determined, and dynamic adjustment can be made according to different evaluation scenarios.

[0080] All material data are traversed, and material data with a comprehensive correlation degree greater than the preset correlation threshold are selected. These material data are output to the user in an intuitive and easy-to-understand manner, such as in the form of tables, charts or reports. When outputting, the correlation degree and correlation direction (positive or negative) between each material data and the key conclusion are marked, helping users understand why these material data have an important impact on the generation of key conclusions.

[0081] The mixed architecture sub-module of the present application can comprehensively and accurately extract high-frequency physiological indicator fluctuation characteristics in the target to-be-evaluated data by combining multiple feature extraction algorithms. These features cover the change information of physiological indicators in different time-frequency ranges, providing rich data support for generating accurate key conclusions. The interpretable output sub-module calculates the correlation between various material data and key conclusions, and outputs material data with a correlation greater than a preset threshold, providing the basis and explanation for the generation of key conclusions. Users can understand which factors have a significant impact on the evaluation results, thereby better understanding the rationality and reliability of the evaluation conclusions.

[0082] According to some embodiments, with reference to Figure 2 The evaluation optimization module 12 includes an information import sub-module 121, a difference annotation sub-module 122, and an associated knowledge annotation sub-module 123.

[0083] The information import sub-module 121 is configured to receive user input evaluation revision information; the difference annotation sub-module 122 is configured to generate the basic evaluation conclusion according to the evaluation revision information and the preliminary evaluation conclusion, and to perform a preset special display of the corresponding optimization information; and the associated knowledge annotation sub-module 123 is configured to annotate the basic evaluation conclusion according to the user-selected preset clause number, to determine and output the target evaluation conclusion after annotation.

[0084] In the present application, an intuitive and easy-to-use user interaction interface can be pre-set, which can be integrated into the operation platform of the evaluation system or can be a separate module window. A dedicated input area is provided on the interface, such as a text box, a table, or a file upload button, to facilitate user input of evaluation revision information.

[0085] For text input, a rich text editing function is provided to allow users to format the input content, such as bold, italic, underline, font size adjustment, and color adjustment, to clearly express the focus and intent of the revision. If file uploading is supported, the acceptable file formats are clearly specified, such as common document formats and table formats, and file format prompts and examples are provided on the upload interface to facilitate users to prepare and upload the correct files.

[0086] When the user inputs the evaluation revision information through the interface and submits, the information import submodule 121 is responsible for receiving these data. A secure network communication protocol is used to ensure the security and integrity of the data during transmission. The format of the data is checked to see if it meets the requirements, the content is complete, there are no illegal characters or malicious code, etc., to strictly verify the received data. In some implementations, if the user uploads a document file, the file is scanned for viruses and the format is parsed and verified; if it is a text input, the text length is checked to see if it is within a reasonable range, whether it contains system prohibited keywords, etc. If the data verification fails, return an explicit error prompt to the user to guide the user to re-enter the correct evaluation revision information.

[0087] The evaluation revision information received by the information import submodule 121 and the existing preliminary evaluation conclusion of the system. Align the evaluation revision information and the preliminary evaluation conclusion in text or data structure to ensure the accuracy of the comparison. Advanced text difference comparison algorithms are used, such as algorithms based on the longest common subsequence or more efficient difference detection algorithms. These algorithms can quickly and accurately find the differences between the evaluation revision information and the preliminary evaluation conclusion, including new content, deleted content, and modified content. For data structure difference comparison, for example, in the case of numerical evaluation indicators, calculate the difference between the revised value and the preliminary evaluation value, and determine whether the difference exceeds the pre-set threshold to determine whether it is a significant difference.

[0088] According to the difference comparison result, generate the basic evaluation conclusion. Integrate the reasonable and effective part of the evaluation revision information into the preliminary evaluation conclusion to form the updated basic evaluation conclusion. For optimization information, use a pre-set special display method for annotation. For example, use different colors (such as green for new content, red for deleted content, and blue for modified content) to highlight the difference part; or use special symbols (such as arrows, boxes, etc.) to mark the location and range of the modification. At the same time, add annotations next to it to explain the type and reason of the difference, making it easy for users to quickly understand the revised content and impact.

[0089] A database containing various evaluation-related clauses can be established in advance, which can come from industry standards, regulations and policies, enterprise internal specifications, etc. Each clause is numbered and described in detail, and an index is established for quick retrieval. Update and maintain the pre-set clause database regularly to ensure the timeliness and accuracy of the clauses. The latest relevant information can be collected and organized manually, or the network crawler technology can be used to automatically capture relevant regulatory policy files and perform parsing and input.

[0090] The query and selection function of the preset clause is provided on the user interface, and the user can find the required preset clause number through keyword search, classified browsing and the like. When the user selects the preset clause number, the associated knowledge annotation submodule 123 obtains the corresponding clause content from the preset clause database. The obtained clause content is associated with the basic evaluation conclusion for correlation analysis to determine the relevance and applicability of the clause content to the basic evaluation conclusion. If the clause content is relevant to the basic evaluation conclusion, it is added as annotation information to the basic evaluation conclusion, and the clause number and source are explicitly marked in the annotation information. Finally, the annotated target evaluation conclusion is output, which is presented in a clear and standardized form, including basic evaluation content, optimization information annotation and associated knowledge annotation, facilitating user review and use.

[0091] The application receives the evaluation revision information input by the user through the information import submodule, and the difference marking submodule compares and fuses the revision information with the preliminary evaluation conclusion, which can timely correct errors and biases in the preliminary evaluation and supplement missing information, thereby improving the accuracy of the evaluation conclusion. The associated knowledge annotation submodule annotates the basic evaluation conclusion according to the user-selected preset clause, providing a clear basis and reference for the evaluation conclusion and enhancing the reliability and authority of the evaluation conclusion. The user can better understand the rationality and compliance of the evaluation conclusion based on the annotated associated knowledge.

[0092] The device embodiment of the application is described below, which can be used to execute the method embodiment of the application. For details not disclosed in the device embodiment of the application, reference can be made to the method embodiment of the application.

[0093] Figure 3 A flowchart of the athlete physical function evaluation method based on a bidirectional data closed loop is provided for the embodiment of the application. As shown in Figure 3 The athlete physical function evaluation method based on a bidirectional data closed loop includes steps S31, S32 and S33.

[0094] In step S31, multi-source data is collected, and the multi-source data is time-aligned to generate target evaluation data; wherein the multi-source data is used to represent multiple types of data of athlete physical function.

[0095] In step S32, a corresponding preliminary evaluation conclusion is generated based on the target evaluation data; wherein the preliminary evaluation conclusion contains an explainable evidence chain.

[0096] In step S33, the preliminary evaluation conclusion is revised and annotated online to determine optimization information and output a corresponding target evaluation conclusion.

[0097] Optionally, the method further includes determining a reinforcement learning reward signal based on the optimization information.

[0098] Optionally, the method further comprises: determining an effective feedback amount according to a preset identity attribute of a user corresponding to the historical optimization information corresponding to the preliminary evaluation conclusion and a corresponding optimization score of the historical optimization information, and determining a training set from the preliminary evaluation conclusion according to the effective feedback amount; and when the effective feedback amount meets a preset adjustment threshold, performing model fine-tuning.

[0099] Optionally, the online revision and knowledge labeling of the preliminary evaluation conclusion are used to determine optimization information and output a corresponding target evaluation conclusion, and are specifically used for: determining audit information of the target evaluation conclusion according to the preset identity attribute and the optimization score; when an actual audit process of the target evaluation conclusion meets the audit information, outputting the target evaluation conclusion and sending corresponding optimization information to the model optimization module.

[0100] Optionally, the collection of multi-source data and the time sequence alignment of multi-source data to generate target to-be-evaluated data comprise: automatically detecting and removing abnormal fragments caused by device signal loss of the multi-source data during time sequence alignment according to an improved dynamic time warping algorithm to determine to-be-aggregated data; calculating feature statistics of the to-be-aggregated data according to a preset scale period layering to determine the target to-be-evaluated data based on the feature statistics.

[0101] Optionally, the generation of a corresponding preliminary evaluation conclusion based on the target to-be-evaluated data comprises: extracting high-frequency physiological indicator fluctuation features in the target to-be-evaluated data, and generating a key conclusion based on the high-frequency physiological indicator fluctuation features; outputting material data corresponding to a correlation degree greater than a preset correlation threshold.

[0102] Optionally, the online revision and knowledge labeling of the preliminary evaluation conclusion to determine optimization information and output a corresponding target evaluation conclusion further comprise: receiving evaluation revision information input by a user; generating the basic evaluation conclusion according to the evaluation revision information and the preliminary evaluation conclusion, and performing preset special display of corresponding optimization information; annotating the basic evaluation conclusion according to a preset clause number selected by the user to determine and output the target evaluation conclusion after annotation.

[0103] In some specific implementation processes, reference is made to Figure 4The system can include an athlete comprehensive monitoring management platform (hereinafter referred to as "data management platform") and an athlete physical function state and sports ability evaluation large model (hereinafter referred to as "large model").

[0104] The data management platform generates an athlete personal report interface, supports querying physiological and biochemical monitoring data of target athletes according to time range and monitoring indicators (such as heart rate, blood oxygen, creatine kinase, etc.), and generates a standardized time series data file containing time stamp, indicator value and normal reference range through dynamic time warping algorithm for time series alignment of multi-source heterogeneous data (such as real-time data of wearable devices and laboratory test data).

[0105] The platform function button "get large model evaluation and suggestion" triggers data transmission, and adopts RESTful interface (Representational State Transfer (REST) API) to push the time series data file of athlete monitoring data to the large model for physical function state and sports ability evaluation. The vertical domain large model can realize accurate evaluation of athlete function state and sports ability according to physiological and biochemical, body composition and other physical function detection data of athletes, and give training suggestions. The input layer of the large model is based on a preset time window (such as 72-hour training period, 30-day recovery period or any input time period) to aggregate features (calculate mean, standard deviation, fluctuation coefficient, etc.) of time series data, and generate evaluation conclusions through deep neural network (fusion of LSTM and Transformer architecture), including core conclusions, indicator analysis and training suggestions.

[0106] The large model returns the evaluation conclusions in the form of streaming data (similar to the line-by-line rendering effect of a typewriter), supporting real-time viewing of the inference process; the data management platform provides a dialogue input box, allowing researchers / coaches to input follow-up instructions (such as "what is the main data basis for this conclusion" and "how to adjust the training plan to improve the abnormal indicators"), and the system analyzes the problem through an intent recognition model, associates historical data and domain knowledge base to generate a deep analysis report.

[0107] The evaluation conclusions are displayed in modules in the athlete personal report interface of the data management platform, and experts can revise the conclusions: directly edit the text content (such as supplementing injury risk prompts), mark data abnormal points (such as marking a blood lactic acid test value that needs to be excluded due to sample contamination), or add professional suggestions. The revised content is confirmed by electronic signature to generate a formal evaluation report. The core content of the report is expert opinions, and in addition to expert opinions, the report also contains personnel basic information, recent test records, indicator statistical data and line graphs.

[0108] Through the "push expert evaluation" button to trigger the reverse data transmission, the following information is packaged as a JSON format data packet: original time series data (desensitization processing, hiding the names of athletes and other sensitive information), expert revised evaluation conclusion, and the data packet is transmitted to the large model training platform through a special secure channel. It should be noted that an audit system should be set up on the platform, that is, the expert evaluation used as the basis for optimizing the model needs to be audited by personnel with audit authority before it can be pushed to the large model. Avoid not very rigorous expert evaluation feedback to the large model, which may affect the accuracy of the large model.

[0109] The large model uses a transfer learning strategy for optimization: (1) Basic layer: use a general evaluation model pre-trained using public sports medical data sets, sports monitoring data and evaluation reports accumulated by scientific research institutions and other departments over a long period of time; (2) Adaptation layer: based on expert feedback data, adjust parameters through reinforcement learning, and the reward function is designed as expert annotation consistency score Reward. The calculation formula is: Reward= * accuracy + * domain knowledge matching degree Wherein, the weight coefficient = 0.7, = 0.3.

[0110] (3) Output layer: the updated model supports output of personalized evaluation conclusions according to sports events (such as synchronized swimming, handball).

[0111] Referring to Figure 5 , the system can include in detail: 1. Data management platform: including data query module, time series processing module, interactive interface module, data interface module; 2. Large model server: including inference engine module, training optimization module, data parsing module; 3. Communication interface: define lightweight, easy to read, support key-value pair structure, JSON data format standard, used for data exchange between data management platform and large model server, support HTTPS encryption transmission.

[0112] The present application optimizes the model through expert audit results, realizes the explicit injection of domain knowledge, solves the limitations of traditional models relying on fixed labeled data, and improves the evaluation accuracy of the model. The streaming output and multi-round dialogue technology enables professional users to ask questions in real time, dynamically adjusts the recommendations based on the training scene, and improves the report generation efficiency. The time window aggregation and time series alignment algorithm is designed for the periodic characteristics of athlete data, effectively handles the problem of inconsistent sampling frequency, and improves the accuracy of abnormal data recognition.

[0113] In some specific implementations, the functional evaluation process of a swimmer is taken as an example.

[0114] First, data preparation is performed: a researcher queries the data management platform to obtain 12 test batches of functional monitoring data of a certain athlete from a year b month c day to a year d month e day, and the platform can query the time series data, change value, historical average value, historical maximum value, historical minimum value, and other statistical result data of the 12 batches.

[0115] Model inference: after clicking "get large model evaluation and suggestion", the above time series data and various statistical result data are transmitted to the large model through the interface, the model identifies that the athlete's (female) hemoglobin is less than 120.00 in the last test, and outputs the preliminary conclusion: "the athlete belongs to anemia, hemoglobin decreases, affects oxygen carrying capacity, and appears palpitation, shortness of breath, rapid heartbeat, dizziness, nausea, etc. during or after exercise, thereby affecting competitive ability and competition results, and seriously reducing the possible harm to the athlete's health", and the inference process is displayed in a stream line by line (such as "the athlete belongs to anemia, hemoglobin decreases →, affects oxygen carrying → ability, → during or during exercise ……").

[0116] Expert review: after the researcher checks the conclusion, the researcher supplements and revises: "combined with ferritin greater than 20 ng / ml, the preliminary evaluation of the athlete is "exercise-induced low hemoglobin", other indicators are normal, and the training factors are considered, such as the above amount and intensity adaptation stage, at this time, monitoring should be strengthened, and under normal circumstances, it can recover in 3-5 days, if it continues to decline, the load should be adjusted; if it appears in the long-term training process, intervention measures should be taken, such as adjusting the training load, combining other indicators to use appropriate nutritional intervention measures".

[0117] Model optimization: after clicking "push expert evaluation", the revised data and metadata are transmitted to the large model, and the training system identifies the expert knowledge of "hemoglobin and ferritin correlation analysis", and enhances the feature weight in the optimization process, so that the evaluation accuracy of subsequent similar cases is improved by 18%.

[0118] The method performs similar functions to the device provided above, and other functions can be referred to the previous description, which will not be described here.

[0119] Figure 6 The structural schematic diagram of the electronic device provided in the embodiments of the present application is shown in FIG. 6. Figure 6 As shown in FIG. 6, the electronic device 600 of the embodiments can include a memory 601 and a processor 602.

[0120] The computer program is stored on the memory 601, and when the computer program is executed by the processor 602, the aforementioned processor 602 executes the method in the above embodiments.

[0121] The processor 602 and the memory 601 are connected, for example, through a bus.

[0122] Optionally, the electronic device 600 can further include a transceiver. It should be noted that the transceiver in actual application is not limited to one, and the structure of the electronic device 600 does not constitute a limitation to the embodiments of the present application.

[0123] The processor 602 can be a CPU (Central Processing Unit, central processor), a general-purpose processor, a DSP (Digital Signal Processor, data signal processor), an ASIC (Application Specific Integrated Circuit, application specific integrated circuit), an FPGA (Field Programmable Gate Array, field programmable gate array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure of the present application. The processor 602 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0124] The bus can include a channel for transmitting information between the above-mentioned components. The bus can be a PCI (Peripheral Component Interconnect, peripheral component interconnect) bus or an EISA (Extended Industry Standard Architecture, extended industry standard architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is shown in the figure, but it does not mean that there is only one bus or only one type of bus.

[0125] The memory 601 can be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, a magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program codes in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0126] The memory 601 is used to store application program codes for implementing the solutions of the present application, and is controlled by the processor 602 to perform. The processor 602 is used to execute the application program codes stored in the memory 601 to realize the content shown in the foregoing method embodiments.

[0127] The electronic device includes, but is not limited to, a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a PMP (Portable Multimedia Player), a vehicle terminal (for example, a vehicle navigation terminal), and the like, and a fixed terminal such as a digital TV, a desktop computer, and the like. It can also be a server or the like. Figure 6 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0128] The electronic device of the present embodiment can be used to execute the method of any one of the foregoing embodiments, and has similar implementation principles and technical effects, which will not be described herein again.

[0129] The present application also provides a non-transitory computer-readable storage medium having computer-readable instructions stored thereon, and the processor executes the method in the above embodiments when the instructions are executed.

[0130] Those skilled in the art can understand that all or part of the steps of the foregoing method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a non-transitory computer-readable storage medium. The program, when executed, performs steps including the foregoing method embodiments; and the foregoing storage medium includes a ROM, a RAM, a magnetic disk or an optical disk, and various media that can store program codes.

[0131] The above has introduced the embodiments of the present application in detail, and the principles and implementation manners of the present application have been described by applying specific examples. The above embodiment descriptions are only used to help understand the method of the present application and its core idea. Meanwhile, the changes or deformations made by the person skilled in the art according to the idea of the present application, based on the specific implementation manners and application scope of the present application, all belong to the protection scope of the present application. In summary, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A two-way data loop based athlete physical ability assessment system, characterized in that, Comprising: a data management module for collecting multi-source data and time-aligning the multi-source data to generate target data to be evaluated; wherein the multi-source data is used to represent multiple types of data characterizing the physical functions of athletes; a large model service module for generating a corresponding preliminary evaluation conclusion based on the target data to be evaluated; wherein the preliminary evaluation conclusion includes an explainable evidence chain; an evaluation optimization module for online revision and knowledge annotation of the preliminary evaluation conclusion to determine optimization information and output a corresponding target evaluation conclusion.

2. The system of claim 1, wherein, Further comprising a model optimization module for determining a reinforcement learning reward signal according to the optimization information.

3. The system of claim 2, wherein, The model optimization module comprises: a feedback filtering submodule for determining an effective feedback amount from the preliminary evaluation conclusion according to a corresponding optimization score of historical optimization information of a user corresponding to a preset identity attribute of the historical optimization information corresponding to the preliminary evaluation conclusion, and determining a training set from the preliminary evaluation conclusion according to the effective feedback amount; an incremental learning scheduling submodule for model fine-tuning when the effective feedback amount meets a preset adjustment threshold.

4. The system of claim 3, wherein, The evaluation optimization module is configured to: determine audit information of the target evaluation conclusion according to the preset identity attribute and the optimization score; output the target evaluation conclusion and send the corresponding optimization information to the model optimization module when the actual audit process of the target evaluation conclusion meets the audit information.

5. The system of claim 1, wherein, The data management module comprises: a dynamic time warping submodule for automatically detecting and removing abnormal segments caused by device signal loss of the multi-source data during time alignment to determine data to be aggregated according to an improved dynamic time warping algorithm; a multi-scale aggregation submodule for calculating feature statistics of the data to be aggregated according to a preset scale period layering to determine the target data to be evaluated based on the feature statistics.

6. The system of claim 1, wherein, The large model service module comprises: a hybrid architecture submodule for extracting high-frequency physiological indicator fluctuation features in the target data to be evaluated and generating a key conclusion based on the high-frequency physiological indicator fluctuation features; an explainable output submodule for outputting material data corresponding to a correlation degree greater than a preset correlation threshold.

7. The system of claim 1, wherein, The evaluation optimization module comprises: an information import submodule for receiving evaluation revision information input by a user; a difference annotation submodule for generating a basic evaluation conclusion based on the evaluation revision information and the preliminary evaluation conclusion, and performing a preset special display of the corresponding optimization information; an associated knowledge annotation submodule for annotating the basic evaluation conclusion according to a preset clause number selected by the user to determine and output the target evaluation conclusion after annotation.

8. A method for evaluating the physical fitness of an athlete based on a bidirectional data closed loop, characterized in that, Comprising: collecting multi-source data and time-aligning the multi-source data to generate target data to be evaluated; wherein the multi-source data is used to represent multiple types of data characterizing the physical functions of athletes; generating a corresponding preliminary evaluation conclusion based on the target data to be evaluated; wherein the preliminary evaluation conclusion includes an explainable evidence chain; online revision and knowledge annotation of the preliminary evaluation conclusion to determine optimization information and output a corresponding target evaluation conclusion.

9. An electronic device, comprising: Comprising: a processor; a memory storing a computer program which, when executed by the processor, causes the processor to perform the method of claim 8.

10. A non-transitory computer-readable storage medium, comprising: a computer readable medium having stored thereon computer readable instructions which, when executed by a processor, cause the processor to perform the method of claim 8.

Images