Motion analysis method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a motion analysis method and device, electronic equipment and a storage medium, and belongs to the technical field of machine learning. The method comprises the following steps: acquiring current motion data and historical motion data; training a preset recurrent neural network according to the historical exercise data and the corresponding historical exercise scores, and constructing a target score prediction model; inputting the current motion data into the target score prediction model to obtain a predicted score result; constructing a problem instruction based on the predicted score result, the current exercise data, the historical exercise data and the historical exercise score; and inputting the question instruction into a preset large language model, and outputting an exercise training suggestion report. The exercise data can be comprehensively analyzed based on the target score prediction model and the preset large language model, the exercise training suggestion report is quickly and accurately generated, and the problems that an existing exercise analysis method depends on manual subjective judgment and is low in efficiency and score prediction accuracy are solved.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a motion analysis method and device, an electronic device, and a storage medium. Background Art

[0002] Athletes' competitive performance is affected by many factors, including body shape, physiological function, physical function, psychological index, special quality and exercise load, etc. Traditional analysis methods often rely on the personal experience and observation of coaches, mainly based on the experience and judgment of the naked eye, and are relatively inefficient. This method is time-consuming and labor-intensive, and is constrained by the coach's training experience. In addition, it is greatly affected by human subjective judgment, making it difficult to comprehensively and accurately evaluate the athlete's status, and it is also impossible to accurately predict the athlete's performance. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a motion analysis method and device, an electronic device and a storage medium, aiming to quickly and accurately generate a sports training recommendation report.

[0004] To achieve the above object, an embodiment of the present application provides a motion analysis method, which includes the following steps:

[0005] Get current motion data and historical motion data;

[0006] Training a preset recurrent neural network based on the historical sports data and the corresponding historical sports performance to construct a target performance prediction model;

[0007] Inputting the current sports data into the target performance prediction model to obtain a predicted performance result;

[0008] Constructing a question instruction based on the predicted performance result, the current sports data, the historical sports data, and the historical sports performance;

[0009] The question instruction is input into a preset large language model, and a sports training recommendation report is output.

[0010] In some embodiments, the obtaining of current motion data and historical motion data includes:

[0011] Collect the athlete's physiological data and characteristic data during exercise to generate current exercise data;

[0012] Obtain historical sports data and corresponding historical sports results;

[0013] Performing data preprocessing on the historical sports data and the historical sports performance, wherein the historical sports data and the historical sports performance are used for model training;

[0014] Data preprocessing is performed on the current motion data, and the current motion data is used for model input.

[0015] In some embodiments, the physiological data includes body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data, the characteristic data includes stroke rate data, strength data, force field data, stroke path data and hand speed data, and the physiological data of the athlete and the characteristic data during the exercise are collected to generate the current exercise data, including:

[0016] Collecting the athlete's body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data in real time through a wearable device;

[0017] Collecting the stroke rate data, the strength data, the force field data, the stroke path data, and the hand speed data of the athlete during exercise through the wearable device;

[0018] Current motion data is generated according to the physiological data and the characteristic data.

[0019] In some embodiments, the training of a preset recurrent neural network based on the historical sports data and the corresponding historical sports performance to construct a target performance prediction model includes:

[0020] Training a preset recurrent neural network according to the historical sports data and the corresponding historical sports performance to obtain a candidate performance prediction model and a corresponding weight file;

[0021] Determine whether the loss of the candidate performance prediction model is less than a preset loss threshold, if not, return to the step of training the preset recurrent neural network according to the historical sports data and the corresponding historical sports performance to obtain the candidate performance prediction model and the corresponding weight file, if yes, determine the weight file corresponding to the candidate performance prediction model as the target weight file;

[0022] A target performance prediction model is constructed according to the target weight file.

[0023] In some embodiments, the constructing a question instruction based on the predicted performance result, the current sports data, the historical sports data and the historical sports performance includes:

[0024] Vectorizing the predicted performance results to generate a set of question vectors;

[0025] Searching a preset vector database according to the question vector set to obtain a target segmented paragraph, wherein the target segmented paragraph matches the question vector set;

[0026] The target segmentation paragraph is used as a preceding context, and question instructions are constructed in combination with the current sports data, the historical sports data, and the historical sports performance.

[0027] In some embodiments, the method further comprises:

[0028] Access to certain athlete improvement report documents;

[0029] Loading the athlete improvement report document and segmenting the document into sections to obtain segmented document paragraphs;

[0030] Performing vectorization processing on the segmented document paragraphs through a preset embedding model to obtain a target vector set;

[0031] The target vector set is stored in a preset vector database.

[0032] In some embodiments, searching a preset vector database according to the question vector set to obtain a target segmented paragraph includes:

[0033] Searching for related documents in a preset vector database according to the question vector set to obtain related document fragments;

[0034] Determining the vector similarity between the associated document fragment and the question vector set;

[0035] Arrange the associated document fragments in descending order according to the vector similarity to obtain an associated document fragment sequence;

[0036] The first several associated document segments of the associated document segment sequence are selected as target segmented paragraphs, and the target segmented paragraphs are multiple segmented paragraphs that match the question vector set.

[0037] To achieve the above object, another aspect of the embodiment of the present application provides a motion analysis device, the device comprising:

[0038] A data collection module is used to obtain current motion data and historical motion data;

[0039] A model training module, used to train a preset recurrent neural network according to the historical sports data and the corresponding historical sports performance, and to construct a target performance prediction model;

[0040] A performance prediction module, used for inputting the current sports data into the target performance prediction model to obtain a predicted performance result;

[0041] A question construction module, configured to construct a question instruction based on the predicted performance result, the current sports data, the historical sports data and the historical sports performance;

[0042] The report generation module is used to input the problem instruction into a preset large language model and output a sports training recommendation report.

[0043] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the above-mentioned method when executing the computer program.

[0044] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0045] The embodiments of the present application include at least the following beneficial effects: the present application provides a motion analysis method and device, an electronic device and a storage medium, the scheme obtains current motion data and historical motion data; trains a preset recurrent neural network according to the historical motion data and the corresponding historical motion results, and constructs a target performance prediction model; inputs the current motion data into the target performance prediction model to obtain the predicted performance result; constructs a question instruction based on the predicted performance result, the current motion data, the historical motion data and the historical motion results; inputs the question instruction into the preset large language model, and outputs a sports training suggestion report. The embodiments of the present application can comprehensively analyze the various characteristics of athletes based on the two models of the target performance prediction model and the preset large language model, quickly and accurately generate a sports training suggestion report to guide improvement, thereby feeding back to the athlete to significantly improve the sports performance, solving the problem that the existing motion analysis method relies on artificial subjective judgment, is inefficient and has low performance prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of a motion analysis method provided in an embodiment of the present application;

[0047] Figure 2 yes Figure 1 Flow chart of step S101 in FIG.

[0048] Figure 3 yes Figure 1 Flow chart of step S102 in FIG.

[0049] Figure 4 yes Figure 1 Flow chart of step S104 in FIG.

[0050] Figure 5 It is a flowchart of the steps of constructing a preset vector database of the motion analysis method provided in an embodiment of the present application;

[0051] Figure 6 yes Figure 4 Flow chart of step S402 in FIG.

[0052] Figure 7 is a flow chart of the motion analysis method provided in an embodiment of the present application applied to swimming motion analysis;

[0053] Figure 8 is a flow chart of generating a sports training suggestion report provided by an embodiment of the present application;

[0054] Fig. 9 is a schematic diagram of the structure of a motion analysis device provided in an embodiment of the present application;

[0055] Fig.10 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0057] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0058] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0060] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0061] 1) Recurrent Neural Network (RNN) is a special neural network structure that is mainly used to process sequence data. RNN can capture the temporal dependencies in the sequence and use past information to improve the network's performance in processing current and future inputs. RNN is particularly suitable for processing sequential data of different lengths, so it has a wide range of applications in natural language processing, speech recognition, time series analysis and other fields.

[0062] 2) Long Short-Term Memory (LSTM) is a special RNN (recurrent neural network) structure designed specifically to solve the gradient vanishing and gradient exploding problems during long sequence training. It improves the modeling ability of sequence data by using a carefully designed gating structure to capture long-term dependencies. The core of the LSTM network is a memory cell that can save and update information at different time steps. Each LSTM cell contains three gates: input gate, forget gate, and output gate. These gating structures allow the LSTM cell to selectively retain or forget past information and update its internal state based on the current input and past state.

[0063] 3) Big models, in the field of computer science and machine learning, refer to deep learning models with a large number of parameters and complex computational structures. These models are usually built with deep neural networks and have billions or even hundreds of billions of parameters. Big models are designed to improve the expressiveness and predictive performance of the model, enabling it to handle more complex tasks and data. Big models are widely used in various fields, including natural language processing, computer vision, speech recognition, and recommendation systems.

[0064] 4) Vector databases are databases specifically used to store and query vectors. The vectors stored in them come from the vectorization of text, speech, images, videos, etc. This type of database can process more unstructured data, such as images and audio, so in machine learning and deep learning, when data is usually represented in vector form, vector databases are particularly important.

[0065] 5)Qwen1.5-72B: Qwen-72B is the Tongyi Qianwen 72 billion parameter model launched by Alibaba Cloud.

[0066] 6) Large Language Models (LLMs) are a class of AI models that are trained with large amounts of text data to learn the complex patterns and structures of language. These models are able to generate coherent text, answer questions, translate languages, summarize text content, perform sentiment analysis, etc. Large language models are trained with massive amounts of data and use billions of parameters to perform tasks such as answering questions, language translation, and generating sentences.

[0067] 7) Retrieval-augmented generation (RAG) is a method for optimizing the output of large language models so that they can reference authoritative knowledge bases outside of the training data source before generating a response. Building on the already powerful LLM, RAG expands its capabilities to access internal knowledge bases for specific domains or companies without retraining the model. This approach is cost-effective and can effectively improve LLM outputs to maintain relevance, accuracy, and usefulness in different contexts.

[0068] Athletes' competitive performance is affected by many factors, including body shape, physiological function, physical function, psychological index, special quality and exercise load, etc. Traditional analysis methods often rely on the personal experience and observation of coaches, mainly based on the experience and judgment of the naked eye, and are relatively inefficient. This method is time-consuming and labor-intensive, and is constrained by the coach's training experience. In addition, it is greatly affected by human subjective judgment, making it difficult to comprehensively and accurately evaluate the athlete's status, and it is also impossible to accurately predict the athlete's performance.

[0069] In view of this, a motion analysis method, device, equipment and medium are provided in the embodiment of the present application. The scheme obtains current motion data and historical motion data; trains a preset recurrent neural network according to the historical motion data and the corresponding historical motion results to construct a target performance prediction model; inputs the current motion data into the target performance prediction model to obtain a predicted performance result; constructs a question instruction based on the predicted performance result, the current motion data, the historical motion data and the historical motion results; inputs the question instruction into a preset large language model to output a sports training suggestion report. The embodiment of the present application can comprehensively analyze the various characteristics of athletes based on the two models of the target performance prediction model and the preset large language model, quickly and accurately generate a sports training suggestion report to guide improvement, thereby feeding back to the athlete to significantly improve the sports performance, solving the problem that the existing motion analysis method relies on artificial subjective judgment, is inefficient and has low performance prediction accuracy.

[0070] The motion analysis method provided in the embodiment of the present application relates to the field of machine learning technology. The motion analysis method provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or it can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services. The cloud server, the server can also be a node server in the blockchain network; the software can be an application that implements the motion analysis method, etc., but is not limited to the above form.

[0071] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0072] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0073] Figure 1 is an optional flow chart of the motion analysis method provided in the embodiment of the present application. Figure 1The method may include but is not limited to steps S101 to S105.

[0074] Step S101, obtaining current motion data and historical motion data.

[0075] Specifically, current motion data is used for model input, and historical motion data and historical motion results are used for model training.

[0076] The present application does not specifically limit the sports data collection device, and it can be flexibly selected according to actual needs. For example, a wearable device can be used for data collection, a sensor can be used for data collection, or a professional sports analysis device can be used for data collection.

[0077] The present application does not specifically limit the motion data collection method, and it can be flexibly selected according to actual needs. For example, data collection can be performed automatically or manually.

[0078] In some embodiments, the athlete's physiological data and characteristic data during exercise are collected to generate current exercise data; historical exercise data and corresponding historical exercise results are obtained; data preprocessing is performed on the historical exercise data and historical exercise results; and data preprocessing is performed on the current exercise data.

[0079] It should be noted that data preprocessing is performed on historical sports data and historical sports performance for model training; and data preprocessing is performed on current sports data for model input.

[0080] Optionally, data preprocessing is performed on the current motion data, and the current motion data is cleaned, standardized, feature reorganized, etc., and extracted into a feature vector as an input of the target performance prediction model.

[0081] Specifically, the data collection methods include automatic and manual methods. Among them, the automatic method can collect current motion data and historical motion data based on wearable devices (such as hardware bracelet devices), pass the current motion data and historical motion data to the message middleware through the Internet of Things MQTT protocol, and transmit the collected data to the bracelet management app on the mobile terminal based on Bluetooth. The app connects the collected data to the background of the sports training recommendation report generation system through an interface.

[0082] In another embodiment, the current motion data and historical motion data can be manually added to the message middleware through the web system. For example, after the bracelet transmits the collected data to the mobile terminal based on Bluetooth, it exports the Excel file and imports the Excel file into the system on the system page.

[0083] Step S102: training a preset recurrent neural network based on historical sports data and corresponding historical sports performance to construct a target performance prediction model.

[0084] In some embodiments, a preset recurrent neural network is regularly trained based on historical sports data and corresponding historical sports performance to construct a target performance prediction model.

[0085] Among them, if the trained preset recurrent neural network meets the preset model requirements, the trained preset recurrent neural network is determined to be used to construct the target score prediction model.

[0086] It can be understood that the preset model requirement can be a preset loss threshold. If the loss of the trained preset recurrent neural network is less than the preset loss threshold, the trained preset recurrent neural network is determined to be used to construct the target performance prediction model.

[0087] Specifically, the preset recurrent neural network is trained according to the historical sports data and the corresponding historical sports results to obtain a candidate performance prediction model and a corresponding weight file; it is determined whether the loss of the candidate performance prediction model is less than a preset loss threshold, if not, the step of returning to the step of training the preset recurrent neural network according to the historical sports data and the corresponding historical sports results to obtain a candidate performance prediction model and a corresponding weight file, if so, the weight file corresponding to the candidate performance prediction model is determined as the target weight file; and a target performance prediction model is constructed according to the target weight file.

[0088] Step S103, inputting the current sports data into the target performance prediction model to obtain the predicted performance result.

[0089] Specifically, the target performance prediction model adopts a deep learning model.

[0090] In some embodiments, the current motion data is input into a target performance prediction model, and the target performance prediction model predicts the performance of the swimmer based on the input current motion data to obtain a predicted performance result.

[0091] Optionally, the target performance prediction model may adopt a recurrent neural network model or a long short-term memory network model.

[0092] Step S104, constructing a question instruction based on the predicted performance result, current sports data, historical sports data and historical sports performance.

[0093] It can be understood that the predicted performance results, current sports data, historical sports data and historical sports performance are integrated, and question instructions are constructed based on the integrated data.

[0094] In some embodiments, a question instruction prompt is constructed based on predicted performance results, real-time sports data, historical sports data and corresponding sports performance; the question instruction prompt is input into a preset large language model to output a corresponding sports training recommendation report.

[0095] Furthermore, the predicted performance results are vectorized to generate a question vector set; a preset vector database is searched based on the question vector set to obtain target segmented paragraphs, and the target segmented paragraphs match the question vector set; the target segmented paragraphs are used as a pre-context, and question instructions are constructed in combination with current sports data, historical sports data and historical sports performance.

[0096] It can be understood that the present application also includes the step of constructing a preset vector database, and the preset vector database is used to store segmented paragraphs.

[0097] In this embodiment, question instructions are constructed based on predicted performance results, current sports data, historical sports data, and historical sports performance, which lays the foundation for the subsequent generation of training suggestion reports, helps improve the accuracy of information transmission, and helps the large language model understand the context.

[0098] Step S105, input the question instruction into the preset large language model, and output the sports training suggestion report.

[0099] In some embodiments, the question instruction is input into a preset large language model, and the preset large language model performs reasoning based on the information of the complete question instruction and generates an athlete performance improvement report, namely a sports training suggestion report.

[0100] Among them, the preset large language models include but are not limited to Qwen1.5-72B large model, ChatGPT, LaMDA and Wenxinyiyan, etc., and this embodiment does not limit this.

[0101] Optionally, the sports training advice report may include training plans, improvement suggestions, dietary guidance, etc.

[0102] In this embodiment, the question instructions are input into the preset large language model, and the sports training recommendation report is output. By comprehensively considering the current and historical data, it is helpful to provide scientific and personalized training suggestions, improve the feasibility and effectiveness of the training suggestions, and help athletes optimize the training effects.

[0103] Steps S101 to S105 shown in the embodiment of the present application, by obtaining current motion data and historical motion data; training a preset recurrent neural network according to historical motion data and corresponding historical sports performance, constructing a target performance prediction model; inputting the current motion data into the target performance prediction model to obtain a predicted performance result; constructing a question instruction based on the predicted performance result, current motion data, historical motion data and historical sports performance; inputting the question instruction into a preset large language model, and outputting a sports training suggestion report. The embodiment of the present application can comprehensively analyze the various characteristics of athletes based on the target performance prediction model and the preset large language model, quickly and accurately generate a sports training suggestion report for guiding improvement, thereby feeding back to athletes to significantly improve sports performance, solving the problem that the existing sports analysis method relies on artificial subjective judgment, is inefficient and has low performance prediction accuracy.

[0104] See also Figure 2 In some embodiments, step S101 may include but is not limited to steps S201 to S203:

[0105] Step S201, collecting the athlete's physiological data and characteristic data during exercise to generate real-time exercise data.

[0106] Specifically, current motion data is used for model input, historical motion data and historical motion results are used for model training, physiological data include but are not limited to body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data, characteristic data is related to the type of sport.

[0107] It can be understood that if an athlete performs swimming, the characteristic data includes stroke rate data, strength data, force field data, stroke path data and hand speed data.

[0108] Exemplarily, physiological data include basic information data such as height, weight, age, gender, body mass index, blood oxygen saturation, urine pH, urine protein, urobilinogen, white blood cell count, hemoglobin, platelets, red blood cell count, urea nitrogen and creatine kinase, etc.; characteristic data include maximum oxygen uptake, swimming time, training program type, number of leg swings and number of hand swings, etc.

[0109] Optionally, wearable devices are used to collect the athlete's body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data in real time; wearable devices are used to collect the athlete's stroke rate data, strength data, force field data, stroke path data and hand speed data during exercise, real-time motion data is generated based on the physiological data and feature data, and the real-time motion data is sent to the message middleware.

[0110] In some embodiments, current motion data and historical motion data are loaded simultaneously based on the message middleware.

[0111] In step S201 of some embodiments, the current motion data and the historical motion data are sent to the current motion data preprocessing module and the historical motion data preprocessing module through the message middleware.

[0112] Step S202, obtaining historical sports data and corresponding historical sports performance.

[0113] In step S202 of some embodiments, historical sports data is obtained based on current sports data, and sports performance corresponding to the historical sports data is obtained.

[0114] Exemplarily, if the exercise mode of the current exercise data is long-distance running, then the pre-stored historical exercise data of the exercise mode being long-distance running is obtained.

[0115] This application does not specifically limit the scoring method of sports performance, and it can be flexibly selected according to actual needs. For example, the sports performance can be obtained by evaluating the trained convolutional neural network model, or by manual scoring.

[0116] Step S203, pre-processing the historical sports data and historical sports performance.

[0117] Specifically, historical sports data and historical sports performance are used for preset recurrent neural network model training.

[0118] In step S203 of some embodiments, data preprocessing is performed on the acquired historical motion data, wherein the data preprocessing includes but is not limited to cleaning, denoising and standardization.

[0119] Optionally, the historical motion data of the message middleware is loaded through a historical motion data preprocessing module, and data preprocessing is performed on the historical motion data.

[0120] In some embodiments, data preprocessing is performed to make the preprocessed historical motion data meet the input requirements of a preset recurrent neural network.

[0121] In some embodiments, data preprocessing is performed on the current motion data. Optionally, the current motion data of the message middleware is loaded by a current motion data preprocessing module, and data preprocessing is performed on the current motion data.

[0122] Exemplarily, the current motion data is extracted into a feature vector as the input of the deep learning model through cleaning, standardization, feature reorganization and other means. Among them, data cleaning includes data denoising, processing of missing values, elimination of duplicate values, etc. The purpose of data cleaning is to make the data more accurate, complete and reliable. Data standardization includes converting data of different units, formats or ranges into a unified form to facilitate subsequent data analysis and processing.

[0123] See also Figure 3 In some embodiments, step S102 may include but is not limited to steps S301 to S304:

[0124] Step S301, training a preset recurrent neural network according to historical sports data and corresponding historical sports performance to obtain a candidate performance prediction model and a corresponding weight file.

[0125] In step S301 of some embodiments, the candidate performance prediction model may include multiple models.

[0126] It is understandable that each candidate score prediction model has a corresponding weight file.

[0127] In some embodiments, the preset recurrent neural network is trained using historical sports data and corresponding historical sports performance.

[0128] Optionally, an optimization algorithm such as stochastic gradient descent (SGD) or Adam may be used in combination with an appropriate loss function (such as mean square error, cross entropy, etc.) for model learning.

[0129] Step S302, determining whether the loss of the candidate performance prediction model is less than a preset loss threshold, if so, executing step S303, if not, returning to step S301.

[0130] In step S302 of some embodiments, a preset loss threshold is obtained, wherein the preset loss threshold may be determined by a user or automatically determined by a system.

[0131] Furthermore, the loss threshold can be dynamically adjusted using Bayesian optimization or grid search techniques.

[0132] For example, if the preset loss threshold is 0.05, the training ends when the loss of the candidate performance prediction model converges to below 0.05 and no longer fluctuates.

[0133] It is understandable that if the loss of the candidate performance prediction model is greater than or equal to the preset loss threshold, the process returns to step S301 to periodically train the preset recurrent neural network.

[0134] In this embodiment, judging whether the loss of the candidate score prediction model is less than a preset loss threshold is helpful to filter out the candidate score prediction models with poor performance and improve the accuracy of the target score prediction model.

[0135] Step S303: determine the weight file corresponding to the candidate score prediction model as the target weight file.

[0136] In step S303 of some embodiments, if the loss of the candidate performance prediction model is less than a preset loss threshold, the weight file corresponding to the candidate performance prediction model is saved as a target weight file.

[0137] Step S304, constructing a target performance prediction model according to the target weight file.

[0138] In step S304 of some embodiments, the target weight file is loaded through a deep learning framework, and a target performance prediction model is constructed based on the target weight file.

[0139] Exemplarily, the deep learning framework may adopt TensorFlow or PyTorch.

[0140] See also Figure 4 In some embodiments, step S104 may include but is not limited to steps S401 to S403:

[0141] Step S401, vectorize the predicted performance results to generate a set of question vectors.

[0142] In step S401 of some embodiments, each predicted performance result is converted into a vector of fixed length through an algorithm to form a question vector set.

[0143] Optionally, the predicted score results are vectorized using NLP technology to generate a set of question vectors.

[0144] Step S402: searching a preset vector database according to the question vector set to obtain a target segmented paragraph.

[0145] Specifically, the target segmented paragraph is matched with a set of question vectors.

[0146] In step S402 of some embodiments, a target segmentation paragraph matching the question vector set is retrieved from a preset vector database, wherein the target segmentation paragraph includes a plurality of segmentation paragraphs.

[0147] Optionally, similarity calculation (such as cosine similarity) is performed between the question vector set and the vectors in the database to retrieve the most relevant target segmentation paragraphs.

[0148] In one embodiment, a plurality of athlete improvement report document sets are vectorized and stored in a preset vector database.

[0149] Furthermore, a plurality of athlete improvement report document sets are obtained; a plurality of athlete improvement report document sets are loaded and the documents are segmented to obtain segmented document paragraphs; the segmented document paragraphs are vectorized through a preset embedding model to obtain a target vector set; and the target vector set is stored in a preset vector database.

[0150] Step S403, using the target segmented paragraph as a pre-context, and combining the real-time sports data, historical sports data and historical sports performance to construct a question instruction.

[0151] In step S403 of some embodiments, the target segmented paragraph is used as the preceding context, which means that when constructing the question instruction, the target segmented paragraph will be used as background information to help understand and explain the question more accurately.

[0152] Furthermore, the paragraphs are segmented according to the target, and the real-time sports data, historical sports data and historical sports performance are integrated into the question instructions.

[0153] The prompt consisting of the predicted result Y, real-time sports data, historical sports data and corresponding sports performance is used as the initial information, and the initial information is enhanced to obtain a complete prompt: first, the associated document fragments are retrieved from the vector database (knowledge base) based on similarity, and then the initial information and the associated document fragments are merged to generate a new prompt, that is, the enhanced complete prompt.

[0154] See also Figure 5 In some embodiments, the motion analysis method provided in the embodiment of the present application further includes the step of constructing a preset vector database. The step of constructing a preset vector database may include but is not limited to steps S501 to S504:

[0155] Step S501, obtaining several athlete improvement report documents.

[0156] In step S501 of some embodiments, a number of athlete improvement report documents are obtained from an internal system, a file management system, a cloud storage, or a database to form a document collection.

[0157] Optionally, the Athlete Improvement Report document may be fetched using an API or crawler tool.

[0158] It is understandable that the athlete improvement report document supports multiple document formats, such as PDF, Word, TXT, etc.

[0159] Step S502, loading the athlete improvement report document and dividing the document into blocks to obtain segmented document paragraphs.

[0160] In step S502 of some embodiments, a document collection is loaded and the documents are divided into smaller paragraphs or sentences.

[0161] Optionally, a text parsing library is used to read the document content, wherein the text parsing library is determined according to the document format.

[0162] For example, Apache PDFBox may be used to process PDF documents, and python-docx may be used to process Word documents.

[0163] In some embodiments, the document can be segmented according to paragraphs, chapter titles, or specific characters (such as periods, line breaks), or segmented according to a fixed word count, topic segmentation, or a natural language processing (NLP)-based approach.

[0164] Step S503, vectorizing the segmented document paragraphs using a preset embedding model to obtain a target vector set.

[0165] Specifically, the target vector set is able to capture the semantic information of the content of the athlete improvement report document.

[0166] In step S503 of some embodiments, the segmented document paragraphs (or sentences) are converted into a target vector set through a preset embedding model.

[0167] Among them, a suitable embedding model is selected to generate a vector representation of the document paragraph.

[0168] It is understandable that the preset embedding model includes but is not limited to bce-reranker-base_v1, bce-embedding-base_v1 and OpenAI-embed-3-large, etc. In this embodiment, bce-reranker-base_v1 can be used, which has a more excellent RAG (retrieval enhancement generation) effect.

[0169] In this embodiment, the segmented document paragraphs are vectorized through a preset embedding model to obtain a target vector set, retaining the semantic information of the text, and converting the text data into a vector form, making subsequent machine learning and data analysis more efficient.

[0170] Step S504: storing the target vector set into a preset vector database.

[0171] In step S504 of some embodiments, the generated target vector sets are written into a database in batches, and corresponding document metadata is stored at the same time to facilitate subsequent retrieval and matching.

[0172] Optionally, an efficient index structure (such as an inverted index, a KNN index, etc.) can be constructed for the preset vector database to improve query performance.

[0173] See also Figure 6 In some embodiments, step S402 includes but is not limited to steps S601 to S604:

[0174] Step S601 , searching for related documents in a preset vector database according to the question vector set to obtain related document fragments.

[0175] In step S601 of some embodiments, a search for associated documents is performed in a preset vector database according to the question vector set to obtain associated document fragments returned by the preset vector database.

[0176] Step S602: Determine the vector similarity between the associated document fragment and the question vector set.

[0177] In step S602 of some embodiments, the vector similarity between the question vector set and the acquired associated document fragments is calculated.

[0178] Optionally, the vector similarity may adopt cosine similarity or Euclidean distance.

[0179] Step S603: Arrange the associated document fragments in descending order according to vector similarity to obtain an associated document fragment sequence.

[0180] In step S603 of some embodiments, the similarity results are sorted in descending order using a quick sort or merge sort algorithm.

[0181] Step S604: select the first several associated document segments in the associated document segment sequence as target segmentation paragraphs.

[0182] In some embodiments, a similarity threshold is set, and a number of associated document segments with a similarity higher than the threshold are selected as target segmentation paragraphs.

[0183] In some other embodiments, the number of target segmentation paragraphs N is set, and the first N document segments are selected as the target segmentation paragraphs.

[0184] Figure 7 is an optional flow chart of the motion analysis method provided in the embodiment of the present application applied to swimming motion analysis. Figure 7 The method may include, but is not limited to, the following steps:

[0185] Step 1: Collect real-time swimming data, historical swimming data and results.

[0186] Among them, real-time motion data (i.e. current motion data) and historical motion data are collected through the swimming bracelet device, and the real-time motion data and historical motion data are transmitted to the message middleware through the Internet of Things MQTT protocol.

[0187] It should be noted that the current sports data is the raw data (RAW Data) of body shape, physiological function, physical function, psychological index, special quality and sports load, etc. The raw data RAW Data is cleaned and preprocessed, and stored as feature vectors to obtain historical sports data. Real-time sports data and historical sports data are loaded simultaneously based on the message middleware.

[0188] Step 2: Regularly train the LSTM model using historical swimming data and performance to generate a weight file.

[0189] Among them, the swimming history data and results are obtained from the middleware through the data subscription mode, and the LSTM model is trained regularly to generate a weight file.

[0190] Specifically, the LSTM model is trained regularly according to the swimming history data and results after data preprocessing, and the results prediction model and the corresponding weight file are obtained after the training is completed; the latest version of the weight file is loaded, and the inference model, i.e., the target results prediction model, is constructed according to the latest version of the weight file.

[0191] Exemplarily, the LSTM model is trained regularly, and the training is terminated when the loss Loss of the LSTM model converges to below 0.05 and no longer fluctuates, and the weight file is the saved weight of the LSTM model.

[0192] It is understandable that the preset recurrent neural network is a recurrent neural network (RNN) variant LSTM network as an example for explanation. It should be noted that the recurrent neural network is a special neural network structure, which is mainly used to process sequence data. RNN can capture the time dependency in the sequence and use past information to improve the performance of the network in processing current and future inputs. RNN is suitable for processing sequential data of different lengths, so it has a wide range of applications in natural language processing, speech recognition, time series analysis and other fields. Long short-term memory network (LSTM) is a special RNN structure, which is specially designed to solve the gradient disappearance and gradient explosion problems in the long sequence training process. It improves the modeling ability of sequence data by using a carefully designed gating structure to capture long-term dependencies. The core of the LSTM network is a memory cell (MemoryCell), which can save and update information at different time steps. Each LSTM cell contains three gates: input gate, forget gate and output gate. These gating structures allow the LSTM cell to selectively retain or forget past information and update its internal state based on the current input and past state.

[0193] Specifically, a recurrent neural network (RNN) variant LSTM network is used to train the preprocessed data (preprocessed historical data and corresponding sports performance) to build a swimmer performance prediction model, where the sports performance is used as the correctly labeled ground truth and can be scored by the coach.

[0194] Exemplarily, in the present embodiment, the LSTM network can adopt a standard bidirectional LSTM network, 512 dimensions, 6 hidden layers (hidden layer), adjust the hyperparameters, and obtain the trained performance prediction model. In practical applications, the deep learning technology is used to extract the features that have a significant impact on the swimmer's performance from the pre-processed data, and the model that can predict the swimmer's performance is trained by the extracted feature data and the corresponding athlete performance data, that is, the trained performance prediction model is obtained. Among them, the feature data include but are not limited to the length of the swimmer's head, palm, arm, leg, foot, body trunk, maximum oxygen uptake, swimming time, training project type, leg swing number and hand swing number, etc.

[0195] Step 3: Input the real-time swimming data into the inference model loaded with the latest weight file to obtain the inference result and predict the performance.

[0196] Specifically, the inference model is the target performance prediction model, and the inference result predicted performance is the predicted performance result.

[0197] Step 4: Assemble the real-time swimming data, historical swimming data, performance, and the predicted performance of the inference result to obtain prompt.

[0198] Exemplarily, the flow chart of generating the exercise training recommendation report is as follows: Figure 8 Specifically, generating a report includes the following steps:

[0199] 1. The prediction model (inference model) outputs the prediction result Y (predicted score), and uses the prediction result Y, real-time sports data, historical sports data, and prompts consisting of corresponding sports scores as inputs for vectorization processing (embedding). For example, as in step 3 of establishing the knowledge base described above, a preset embedding model such as bce-reranker-base_v1 is used for vectorization processing, and converted into a vector form to form a problem vector set C;

[0200] 2. Retrieve the k segmented paragraphs with the highest similarity to the question vector set C from the vector database;

[0201] 3. Take the k segmented paragraphs as the pre-context, combine the athlete's historical sports data and sports performance, real-time sports data, and the predicted performance output by the inference model to form a complete prompt (i.e. Figure 8 complete question shown);

[0202] 4. Input the assembled complete prompt into a preset large language model, such as the Qwen1.5-72B large model. The large language model performs reasoning based on the information of the complete prompt and generates an athlete performance improvement report, i.e., a sports training recommendation report. The preset large language model includes but is not limited to the Qwen1.5-72B large model, ChatGPT, LaMDA, and Wenxinyiyan, etc., which are not limited in this embodiment.

[0203] Among them, retrieving the k segmented paragraphs with the highest similarity to the question vector set C in the vector database specifically includes: searching for related documents in the vector database according to the question vector set C, the vector database returns the related document fragments, and according to the vector similarity between the related document fragments and the question vector set, the related document fragments are arranged in descending order according to the vector similarity; the first k related document fragments are selected as multiple segmented paragraphs matching the question vector set.

[0204] In practical applications, a prompt consisting of the predicted result Y, real-time sports data, historical sports data and corresponding sports performance can be used as initial information, and the initial information can be enhanced to obtain a complete prompt: first, related document fragments are retrieved from the vector database (knowledge base) based on similarity, and then the initial information and the related document fragments are merged to generate a new prompt, that is, the enhanced complete prompt, and the enhanced complete prompt is input into the large language model (LLM), allowing the LLM to judge and answer the question by itself, and generate a response, that is, a sports training recommendation report.

[0205] Step 5: Input the prompt into the RAG system, and the RAG system generates a targeted training optimization recommendation report.

[0206] Specifically, the RAG system is a large language model.

[0207] In some embodiments, the inference results (predicted performance), real-time data, historical data and performance are combined into prompts and input into the RAG system. The big model (RAG system) will generate targeted training optimization suggestion reports for swimmers, including adjusting training load, improving technical movements, and improving psychological quality.

[0208] For example, Figure 8 As shown, building a knowledge base includes the following steps:

[0209] 1. Prepare a large number of existing manually written athlete improvement report documents (such as swimmer improvement reports) to form set A;

[0210] 2. Document loading and segmentation (text extraction and segmentation): Load a set A consisting of manually written athlete improvement report documents, such as PDF files, and segment the documents into smaller paragraphs or sentences;

[0211] 3. Vectorization (embedding): The segmented document paragraphs (or sentences) are converted into a vector set B through a preset embedding model, where the vector set B can capture the semantic information of the document content;

[0212] 4. Store vector set B into the vector database.

[0213] It should be noted that the preset embedding models include but are not limited to bce-reranker-base_v1, bce-embedding-base_v1, and OpenAI-embed-3-large, etc. In this embodiment, bce-reranker-base_v1 can be used, which has a more excellent RAG (retrieval enhancement generation) effect.

[0214] It is understandable that the general training mode of the large language model (LLM) includes a three-stage training mode, the first stage is pre-training (pre-train), the second stage is supervised fine-tuning (SFT), and the third stage is human feedback reinforcement learning (RLHF). Since the large language model (LLM) has been trained and supervised fine-tuned SFT with a large amount of text, the input of the large language model (LLM) in this embodiment basically uses RAW data (plain text) composed of text, which can help the large language model (LLM) generate better responses and make it easier to obtain matching results when performing retrieval.

[0215] See also Fig. 9 The present application also provides a motion analysis device that can implement the above-mentioned motion analysis method. The device includes:

[0216] Data collection module 901, used to obtain current motion data and historical motion data;

[0217] Model training module 902, used to train a preset recurrent neural network according to historical sports data and corresponding historical sports performance, and construct a target performance prediction model;

[0218] The performance prediction module 903 is used to input the current sports data into the target performance prediction model to obtain the predicted performance result;

[0219] A question construction module 904, for constructing a question instruction based on the predicted performance results, current sports data, historical sports data and historical sports performance;

[0220] The report generation module 905 is used to input the question instruction into the preset large language model and output the sports training recommendation report.

[0221] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0222] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above motion analysis method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0223] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0224] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0225] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0226] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the motion analysis method of the embodiment of this application;

[0227] Input / output interface 1003, used to implement information input and output;

[0228] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0229] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0230] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0231] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program implements the above-mentioned motion analysis method when executed by a processor.

[0232] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0233] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0234] The motion analysis method, motion analysis device, electronic device and storage medium provided by the embodiment of the present application obtain current motion data and historical motion data; train a preset recurrent neural network according to the historical motion data and the corresponding historical motion results to construct a target performance prediction model; input the current motion data into the target performance prediction model to obtain the predicted performance result; construct a question instruction based on the predicted performance result, current motion data, historical motion data and historical motion results; input the question instruction into the preset large language model to output a sports training suggestion report. The embodiment of the present application can comprehensively analyze the various characteristics of athletes based on the two models of the target performance prediction model and the preset large language model, quickly and accurately generate a sports training suggestion report to guide improvement, thereby feeding back to the athlete to significantly improve the sports performance, solving the problem that the existing motion analysis method relies on artificial subjective judgment, is inefficient and has low performance prediction accuracy.

[0235] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0236] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0237] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0238] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0239] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0240] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0241] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0242] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0243] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0244] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0245] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A motion analysis method, characterized in that: The method comprises the following steps: Get current motion data and historical motion data; Training a preset recurrent neural network based on the historical sports data and the corresponding historical sports performance to construct a target performance prediction model; Inputting the current sports data into the target performance prediction model to obtain a predicted performance result; Constructing a question instruction based on the predicted performance result, the current sports data, the historical sports data, and the historical sports performance; The question instruction is input into a preset large language model, and a sports training recommendation report is output.

2. The method according to claim 1, characterized in that The obtaining of current motion data and historical motion data includes: Collect the athlete's physiological data and characteristic data during exercise to generate current exercise data; Obtain historical sports data and corresponding historical sports results; Performing data preprocessing on the historical sports data and the historical sports performance, wherein the historical sports data and the historical sports performance are used for model training; Data preprocessing is performed on the current motion data, and the current motion data is used for model input.

3. The method according to claim 2, characterized in that The physiological data includes body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data, the characteristic data includes stroke rate data, strength data, force field data, stroke path data and hand speed data, and the physiological data of the athlete and the characteristic data during the exercise are collected to generate the current exercise data, including: Collecting the athlete's body shape data, physiological function data, physical function data, psychological index data, special quality data and exercise load data in real time through a wearable device; Collecting the stroke rate data, the strength data, the force field data, the stroke path data, and the hand speed data of the athlete during exercise through the wearable device; Current motion data is generated according to the physiological data and the characteristic data.

4. The method according to claim 1, characterized in that: The step of training a preset recurrent neural network according to the historical sports data and the corresponding historical sports performance to construct a target performance prediction model includes: Training a preset recurrent neural network according to the historical sports data and the corresponding historical sports performance to obtain a candidate performance prediction model and a corresponding weight file; Determine whether the loss of the candidate performance prediction model is less than a preset loss threshold, if not, return to the step of training the preset recurrent neural network according to the historical sports data and the corresponding historical sports performance to obtain the candidate performance prediction model and the corresponding weight file, if yes, determine the weight file corresponding to the candidate performance prediction model as the target weight file; A target performance prediction model is constructed according to the target weight file.

5. The method according to claim 1, characterized in that The constructing of the question instruction based on the predicted performance result, the current sports data, the historical sports data and the historical sports performance includes: Vectorizing the predicted performance results to generate a set of question vectors; Searching a preset vector database according to the question vector set to obtain a target segmented paragraph, wherein the target segmented paragraph matches the question vector set; The target segmentation paragraph is used as a preceding context, and question instructions are constructed in combination with the current sports data, the historical sports data, and the historical sports performance.

6. The method according to claim 1, characterized in that The method further comprises: Access to certain athlete improvement report documents; Loading the athlete improvement report document and segmenting the document into sections to obtain segmented document paragraphs; Performing vectorization processing on the segmented document paragraphs through a preset embedding model to obtain a target vector set; The target vector set is stored in a preset vector database.

7. The method according to claim 5, characterized in that The step of searching a preset vector database according to the problem vector set to obtain a target segmented paragraph includes: Searching for related documents in a preset vector database according to the question vector set to obtain related document fragments; Determining the vector similarity between the associated document fragment and the question vector set; Arrange the associated document fragments in descending order according to the vector similarity to obtain an associated document fragment sequence; The first several associated document segments of the associated document segment sequence are selected as target segmented paragraphs, and the target segmented paragraphs are multiple segmented paragraphs that match the question vector set.

8. A motion analysis device, characterized in that: The device comprises: A data collection module is used to obtain current motion data and historical motion data; A model training module, used to train a preset recurrent neural network according to the historical sports data and the corresponding historical sports performance, and to construct a target performance prediction model; A performance prediction module, used for inputting the current sports data into the target performance prediction model to obtain a predicted performance result; A question construction module, configured to construct a question instruction based on the predicted performance result, the current sports data, the historical sports data and the historical sports performance; The report generation module is used to input the problem instruction into a preset large language model and output a sports training recommendation report.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images