Control methods and systems for fitness equipment

Abstract

This invention discloses a control method and system for fitness equipment. The method includes: a data acquisition device synchronously acquiring multimodal data of a user during training and sending the multimodal data to a control device, wherein the multimodal data includes physiological characteristic data, movement data, and environmental data; the control device constructing a multidimensional ability profile of the user based on the multimodal data using a profile construction model, and generating a training plan based on the multidimensional ability profile, acquired training input data, and historical training data using an intelligent decision model, wherein both the profile construction model and the intelligent decision model are models sent from a cloud server; the fitness equipment receiving the training plan sent by the control device and controlling its own output parameters according to the training plan. This invention improves the effectiveness of exercise training while ensuring safety.

Description

Technical Field

[0001] This invention relates to the field of fitness equipment technology, and in particular to a control method and system for fitness equipment. Background Technology

[0002] With the improvement of living standards, maintaining good physical condition has become a basic pursuit for people, and going to the gym has become a daily routine for many. Existing fitness equipment (such as smart dumbbells and ab wheels) typically relies on a single type of sensor for data collection and outputs control commands based on a preset algorithm model. This "one-way open-loop" control mode cannot adapt to the user's real-time dynamic physiological state and training needs, resulting in poor training effects and certain safety risks. For example, when a user's movements become distorted due to fatigue, the equipment may still apply the load as planned, easily causing muscle or joint injuries; conversely, when a user is in good physical condition, the equipment may fail to increase the load in time to achieve efficient training input data. Summary of the Invention

[0003] This invention provides a control method and system for fitness equipment, aiming to solve the problem of poor exercise training effect of existing fitness equipment.

[0004] In a first aspect, embodiments of the present invention provide a method for controlling a fitness device, comprising:

[0005] The acquisition device synchronously collects multimodal data of the user during the training process and sends the multimodal data to the control device. The multimodal data includes physiological feature data, motion data and environmental data.

[0006] The control device constructs a multidimensional capability profile of the user through a profile construction model based on the multimodal data, and generates a training scheme through an intelligent decision model based on the multidimensional capability profile, the acquired training input data, and historical training data. Both the profile construction model and the intelligent decision model are models distributed by a cloud server.

[0007] The fitness equipment receives the training plan sent by the control device and controls its own output parameters according to the training plan.

[0008] Secondly, embodiments of the present invention also provide a control system for a fitness device, including the fitness device, a cloud server, a data acquisition device, and a control device. The data acquisition device is configured with a data acquisition and transmission unit, the control device is configured with a data generation unit, and the fitness device is configured with a receiving and control unit.

[0009] The acquisition and transmission unit is used to acquire multimodal data of the user during the training process synchronously and send the multimodal data to the control device. The multimodal data includes physiological feature data, motion data and environmental data.

[0010] The construction and generation unit is used by the control device to construct a multi-dimensional capability profile of the user through a profile construction model based on the multi-modal data, and to generate a training scheme through an intelligent decision model based on the multi-dimensional capability profile, the acquired training input data and historical training data. The profile construction model and the intelligent decision model are both models distributed by a cloud server.

[0011] The receiving and control unit is used by the fitness equipment to receive the training plan sent by the control device and control its own output parameters according to the training plan.

[0012] This invention provides a control method and system for fitness equipment. The method includes: a data acquisition device synchronously acquiring multimodal data of a user during training and sending the multimodal data to a control device. The multimodal data includes physiological characteristic data, movement data, and environmental data. The control device constructs a multidimensional ability profile of the user using a profile building model based on the multimodal data, and generates a training plan using an intelligent decision model based on the multidimensional ability profile, acquired training input data, and historical training data. Both the profile building model and the intelligent decision model are models sent from a cloud server. The fitness equipment receives the training plan sent by the control device and controls its own output parameters according to the training plan. This invention's technical solution, where the control device constructs a multidimensional ability profile of the user using a profile building model based on the multimodal data acquired by the data acquisition device, and generates a training plan based on the user's multidimensional ability profile using an intelligent decision model, and the fitness equipment controls its own output parameters based on the training plan, improves the training effect while ensuring safety. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram illustrating a scenario of a control method for a fitness device according to an embodiment of the present invention.

[0015] Figure 2A flowchart illustrating a control method for a fitness device according to an embodiment of the present invention;

[0016] Figure 3 This is a schematic diagram of a sub-process of a control method for a fitness device provided in an embodiment of the present invention;

[0017] Figure 4 This is a schematic block diagram of a control system for a fitness device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0023] This invention proposes a control method and system for fitness equipment, addressing the problem of poor exercise training effects in existing fitness equipment. In this embodiment, the control device constructs a multi-dimensional ability profile of the user through a profile building model based on multimodal data collected by the acquisition device. Based on the user's multi-dimensional ability profile, a training plan is generated through an intelligent decision-making model. The fitness equipment controls its own output parameters based on the training plan, improving the exercise training effect while ensuring safety.

[0024] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0025] This invention provides a control method for fitness equipment, which can be used in the control system of fitness equipment, such as... Figure 1 As shown, the control system 200 of the fitness equipment includes the fitness equipment 30, a cloud server 40, a data acquisition device 10, and a control device 20. The specific functions implemented within the units configured in the fitness equipment 30, cloud server 40, data acquisition device 10, and control device 20 will be described in detail in later embodiments; for simplicity, they will not be repeated here. Please refer to... Figure 2 , Figure 2 A flowchart illustrating the control method of the fitness equipment according to an embodiment of the present invention is shown, such as... Figure 2 As shown, the control method of the fitness equipment includes steps S110-S130.

[0026] S110. The acquisition device synchronously acquires multimodal data of the user during the training process and sends the multimodal data to the control device. The multimodal data includes physiological characteristic data, motion data and environmental data.

[0027] In this embodiment, a data acquisition device connected to the control device synchronously collects multimodal data during the user's training process and sends it to the control device in real time. Physiological characteristic data is acquired through physiological sensors in the acquisition device, including the user's heart rate, blood oxygen saturation, and surface electromyography (EMG) signals. Heart rate is monitored in real time using photoelectric sensors or electrode patches. Blood oxygen saturation is calculated based on the ratio of red light to infrared light absorbance. Surface EMG signals are captured by sensors attached to the target muscle groups to capture muscle activity electrical signals. Movement data is acquired by an infrared camera in the acquisition device, including the user's joint angles, movement trajectories, and force frequency during training. Environmental data is collected by environmental sensors in the acquisition device, including temperature, humidity, and air quality parameters of the training environment. Temperature and humidity are detected in real time by temperature and humidity sensors, and air quality parameters are monitored by gas sensors to detect indicators such as PM2.5 concentration. All data is processed for time synchronization and format standardization before being sent to the control device via wireless or wired communication. It should be noted that the acquisition device can be worn, attached, or affixed to the user's body.

[0028] S120. The control device constructs a multi-dimensional capability profile of the user through a profile construction model based on the multi-modal data, and generates a training scheme through an intelligent decision model based on the multi-dimensional capability profile, the acquired training input data, and historical training data. The profile construction model and the intelligent decision model are both models distributed by a cloud server.

[0029] In this embodiment, the portrait construction model includes a cross-modal attention layer, a feature fusion layer, and a portrait output layer; as follows: Figure 3As shown, step S120 specifically includes steps S121-S127: S121, the control device performs data fusion and feature extraction on the multimodal data to obtain a multimodal feature vector; S122, the correlation between different modal features in the multimodal feature vector is calculated through the cross-modal attention layer, and attention weights are assigned to obtain a weighted fused feature vector; S123, based on the weighted fused feature vector, deep feature interaction and compression are performed through a fully connected network and nonlinear activation function in the feature fusion layer to obtain a comprehensive feature representation; S124, the comprehensive feature representation is linearly transformed and normalized through the portrait output layer to generate the user's multidimensional ability portrait, wherein the multidimensional ability portrait includes the user's strength level, endurance characteristics, movement pattern preferences, and fatigue recovery ability. S125. Obtain the historical training data. If the number of data entries in the historical training data is less than the preset number of data entries, obtain the user's questionnaire answers and physical test results. Generate pseudo-label data based on the questionnaire answers and physical test results, and supplement the historical training data with the pseudo-label data. S126. Obtain the training input data set by the user. Perform preprocessing and feature fusion on the training input data, the historical training data, and the multi-dimensional ability profile to obtain decision input data. S127. Input the decision input data into the intelligent decision-making model to make intelligent decisions and obtain the training scheme. In the intelligent decision-making process, the intelligent decision-making model aims to maximize training benefits. The training benefits include effect benefits, safety benefits, and persistence benefits. Training benefits = w1 * effect benefits + w2 * safety benefits + w3 * persistence benefits. The weights w1, w2, and w3 are set according to actual needs. It should be noted that step S121 specifically includes: the control device performing spatiotemporal alignment and confidence-weighted fusion on the multimodal data to obtain a multimodal data matrix; extracting multi-dimensional features from the multimodal data matrix through a deep feature extraction network based on an attention mechanism; and performing dimensionality reduction and standardization on the multi-dimensional features to obtain the multimodal feature vector. It should also be noted that when training the profile building model, the cloud server first preprocesses the sample multimodal data, removing outliers and normalizing it, and proportionally dividing the training and test sets. The basic model of the profile building model adopts a multi-task learning architecture, with Transformer as the core network. It extracts general features through a shared Vision Transformer encoder, and then sets four task branches: user strength level, endurance features, action pattern preferences, and fatigue recovery ability. Each branch uses a fully connected layer combined with an LSTM network to achieve specific feature mapping. During model training, the mean squared error loss function is used to optimize the regression tasks for strength and endurance dimensions, and the cross-entropy loss function is used to optimize the classification task for action pattern preferences. A dynamic attention mechanism is introduced to allocate the weights of each branch.After each training round, the prediction error of the profile dimension is evaluated using a test set. If the error exceeds a threshold, the network parameters and feature fusion strategy are adjusted, iterating until the model converges to obtain the profile construction model. When training the intelligent decision-making model on the cloud server, training samples are first constructed. The training samples include the user's multi-dimensional ability profile, historical training data (e.g., past resistance settings, action completion rate, training effect), and training input data (e.g., training objectives, duration preferences). The training samples are preprocessed and feature fused to form decision input data of a unified dimension. Its basic model adopts a Transformer-based multi-task decision architecture. The input layer receives the decision input data, the encoder captures the correlation between the profile, historical data, and training objectives through a multi-head self-attention mechanism, and the decoder has four sub-task output layers: resistance parameter decision, duration interval decision, action proportion decision, and frequency decision. Each output layer uses a fully connected layer combined with linear regression to achieve parameter prediction. During model training, the model parameters are optimized using a multi-task loss function, and iterated until the model converges to obtain the intelligent decision-making model. Understandably, in other embodiments, the multi-dimensional ability profile also includes ability profiles of other dimensions, such as training adaptation speed. It's important to note that after establishing a communication connection with the control device, the cloud server will send the entire profile building model and the entire intelligent decision-making model to the control device all at once. If the profile building model and the intelligent decision-making model are subsequently updated, the updated model data will be sent to the control device, which will then update the profile building model and the intelligent decision-making model accordingly. Pseudo-label data is generated based on the questionnaire answers and the physical test results; specifically, the questionnaire answers and the physical test results are converted into numerical values ​​to generate pseudo-label data. Understandably, if the number of data entries in the historical training data is not less than the preset number of data entries, then it is not necessary to generate pseudo-label data.

[0030] Furthermore, the training program includes a resistance baseline, resistance increment, single training duration, movement type, and training frequency; the decision input data includes quantified values ​​of user strength level, endurance characteristics, movement pattern preference, and fatigue recovery ability. It should be noted that user strength level typically includes beginner, intermediate, and advanced strength levels, with corresponding quantified values ​​of 0-45, 46-75, and 76-100, respectively. Endurance characteristics also include three levels: beginner, intermediate, and advanced, with corresponding quantified values ​​of 0-45, 46-75, and 76-100, respectively. Fatigue recovery ability also includes three levels: beginner, intermediate, and advanced, with corresponding quantified values ​​of 0-45, 46-75, and 76-100, respectively. Movement pattern preference includes stable, balanced, and flexible types, with corresponding quantified values ​​of 0-45, 46-75, and 76-100, respectively. The intelligent decision-making model makes decisions on the resistance benchmark, resistance increment, single training session duration, movement type, and training frequency based on the user's strength level quantification, endurance characteristic quantification, movement pattern preference quantification, and fatigue recovery ability quantification. Specifically, the intelligent decision-making model sets the resistance benchmark and resistance increment based on the user's strength level quantification; sets the single training session duration based on the endurance characteristic quantification; sets the proportion of compound movements in the movement type based on the movement pattern preference quantification; and sets the training frequency based on the fatigue recovery ability quantification. Specifically, when generating training plan parameters, the intelligent decision-making model first calls the quantitative values ​​of each dimension in the user's multi-dimensional ability profile: for the resistance baseline and resistance increment, if the user's strength level quantitative value is low, the resistance baseline is set to 60%-70% of the current strength level, and the resistance increment is reduced by 3%-5% per week; if the user's strength level quantitative value is high, the baseline is set to 80%-85%, and the increment is increased by 5%-8%; for the duration of a single training session, if the endurance characteristic quantitative value is low, it is shortened to 20-30 seconds; if the endurance characteristic quantitative value is high, it is extended to 40-60 seconds; regarding the proportion of compound movements in the movement type, if the movement pattern preference quantitative value is high, the proportion of compound movements is increased to 60%-70%; if the movement pattern preference quantitative value is low, the proportion is reduced and isolation movements are increased; regarding the training frequency, if the fatigue recovery ability quantitative value is low, it is set to 3-4 times per week; if the fatigue recovery ability quantitative value is high, it is set to 5-6 times per week, ensuring that each parameter matches the user's actual ability and adapts to the training needs of different fitness equipment. It should be noted that in other embodiments, the training scheme may also include other training requirements, such as action specification requirements.

[0031] S130. The fitness equipment receives the training plan sent by the control device and controls its own output parameters according to the training plan.

[0032] In this embodiment, the fitness equipment receives a training plan sent by the control device, adjusts its resistance according to the resistance baseline value and the resistance increment, controls the duration of each training session according to the single training duration, matches its own movement mode according to the movement type, and plans a cyclical training schedule according to the training frequency. Specifically, after receiving the training plan sent by the control device, the fitness equipment adjusts its resistance according to the resistance baseline value and resistance increment in the training plan. For example, smart dumbbells initialize their weight with reference to the resistance baseline value and gradually adjust the load by increment, while ab wheels change the resistance torque through an elastic resistance adjustment device. Then, it controls the duration of each training session according to the single training duration, automatically reminding or pausing the current training session when the time is up. Subsequently, it matches the corresponding movement mode of the equipment according to the movement type in the training plan. For example, strength training equipment switches to the resistance output mode of the corresponding movement to ensure that the movement is performed appropriately. Finally, it plans a cyclical training schedule according to the training frequency, records the completion status of each training session, and reminds the user of subsequent training times according to the frequency.

[0033] In one embodiment, such as this embodiment, the method further includes, after step S130 of the above embodiment: the control device monitors the multimodal data in real time, and if abnormal index values ​​are detected in the multimodal data, a hierarchical security protection mechanism is triggered.

[0034] In this embodiment, the control device monitors the multimodal data sent by the acquisition device in real time and continuously compares each data indicator with the preset safety threshold. If the physiological characteristic data is found to be lower than the lower limit of physiological characteristics, or the motion data exceeds the safety threshold, or the environmental data is found to be greater than the preset environmental threshold, a graded safety protection mechanism is triggered according to the type and severity of the abnormality. For example, a voice warning is first sent. If the abnormality is not relieved, the device resistance is automatically reduced. In case of extreme abnormality, an emergency shutdown is triggered to ensure training safety.

[0035] In one embodiment, such as this embodiment, the method further includes, after step S130 of the above embodiment: if the control device detects that the user switches from the currently used fitness device to a new fitness device, it acquires new training input data and new historical training data corresponding to the new fitness device; generates a new training plan adapted to the new fitness device based on the multi-dimensional ability profile, the new training input data and the new historical training data, and sends the new training plan to the new fitness device to control the new fitness device.

[0036] In this embodiment, if a user switches from the currently used fitness equipment to a new fitness equipment, a connection is first established with the communication module of the new fitness equipment to obtain new training input data (including the user's training goals and duration preferences for the new equipment) and new historical training data (including records of resistance settings, movement completion, and training effects when the user previously used the same type or functional new equipment). Next, a stored multi-dimensional capability profile is invoked, and the multi-dimensional capability profile is preprocessed (e.g., data normalization, outlier removal) and features are fused with the new training input data and new historical training data to form decision input data adapted to the new fitness equipment. Subsequently, the built-in intelligent decision model analyzes the decision input data to generate a new training plan adapted to the new fitness equipment (covering parameters such as the resistance baseline value, resistance increment, single training duration, movement type, and training frequency corresponding to the new fitness equipment). Finally, the new training plan is sent to the new fitness equipment via wireless or wired communication. After receiving the new training plan, the new fitness equipment adjusts its operating state to achieve seamless training transition and precise control after the user switches equipment.

[0037] In one embodiment, such as this embodiment, the method further includes the following steps after step S130 of the above embodiment: if the control device receives a training end instruction, it acquires the entire training data, generates a multi-dimensional training report based on the entire training data, and displays the multi-dimensional training report to the user; the control device de-identifies the entire training data to obtain de-identified entire data, and uploads the de-identified entire data to a cloud server, so that the cloud service iteratively optimizes the profile building model and the intelligent decision-making model to obtain model update data; the control device receives the model update data sent by the cloud server, and updates the profile building model and the intelligent decision-making model based on the model update data.

[0038] In this embodiment, after receiving a training end command triggered by the user (such as the user manually pressing the device's end button or the device detecting the completion of the training plan), the control device immediately acquires the entire training data. This data covers real-time physiological characteristic data, movement data, device operating parameters (resistance adjustment records, training duration statistics), and environmental data during the training process. Based on the entire training data, a multi-dimensional training report is generated from four dimensions: physical energy consumption (calorie consumption calculated by combining metabolic equivalents), movement quality (analyzing the rate of compliance with movement standardization and deviation points), training progress (comparing the intensity and duration differences between the current training and historical training), and device adaptability (assessing the matching degree between the device's resistance adjustment and the user's ability). The report is then displayed to the user through the device's display screen. The control device anonymizes all training data, removing user identity information (such as name and account), and performs differential privacy processing on physiological characteristic data and motion data to generate anonymized full-process data. This anonymized full-process data is then uploaded to a cloud server. The cloud server uses this data to iteratively update the profile building model and the intelligent decision-making model to obtain model update data. The control device receives the model update data from the cloud server and updates the profile building model and the intelligent decision-making model accordingly, completing the model version upgrade and ensuring the accuracy and adaptability of subsequent user multidimensional capability profile generation and training scheme formulation.

[0039] Figure 4 This is a schematic block diagram of a control system 200 for a fitness device provided in an embodiment of the present invention. Figure 4 As shown, this corresponds to the control method for the fitness equipment applied to the acquisition device 10, the control device 20, and the fitness equipment 30 described above. The control system 200 of this fitness equipment includes a unit for executing the aforementioned control method. Specifically, please refer to... Figure 4 The control system 200 of the fitness equipment includes fitness equipment 30, cloud server 40, acquisition device 10 and control device 20. The acquisition device 10 is equipped with acquisition and transmission unit 101, the control device 20 is equipped with construction and generation unit 201, and the fitness equipment 30 is equipped with receiving and control unit 301.

[0040] The acquisition and transmission unit 101 is used to acquire multimodal data of the user during the training process synchronously and send the multimodal data to the control device. The multimodal data includes physiological feature data, movement data, and environmental data. The construction and generation unit 201 is used by the control device to construct a multidimensional ability profile of the user based on the multimodal data using a profile construction model, and to generate a training plan based on the multidimensional ability profile, the acquired training input data, and historical training data using an intelligent decision model. Both the profile construction model and the intelligent decision model are models sent from a cloud server. The receiving and control unit 301 is used by the fitness device to receive the training plan sent by the control device and to control its own output parameters according to the training plan.

[0041] In some embodiments, such as this embodiment, the construction and generation unit 201 is specifically used for: the control device performing data fusion and feature extraction on the multimodal data to obtain a multimodal feature vector; calculating the correlation between different modal features in the multimodal feature vector through the cross-modal attention layer and assigning attention weights to obtain a weighted fusion feature vector;

[0042] Based on the weighted fusion feature vector, a comprehensive feature representation is obtained through deep feature interaction and compression using a fully connected network and nonlinear activation function in the feature fusion layer. The comprehensive feature representation is then linearly transformed and normalized by the profile output layer to generate the user's multidimensional ability profile, which includes the user's strength level, endurance characteristics, movement pattern preferences, and fatigue recovery ability. Historical training data is acquired; if the number of data entries in the historical training data is less than a preset number, the user's completed questionnaire answers and physical test results are acquired, and pseudo-label data is generated based on the questionnaire answers and physical test results, and added to the historical training data. The user-set training input data is acquired, and the training input data, historical training data, and multidimensional ability profile are preprocessed and feature-fused to obtain decision input data. The decision input data is input into the intelligent decision-making model for intelligent decision-making to obtain the training scheme, wherein the intelligent decision-making model aims to maximize training benefits, including effectiveness benefits, safety benefits, and persistence benefits.

[0043] In some embodiments, such as this one, the receiving control unit 301 is specifically configured to: adjust its own resistance according to the resistance reference value and the resistance increment; control the duration of each training session according to the single training duration; match its own action mode according to the action type; and plan a cyclical training schedule according to the training frequency.

[0044] In some embodiments, such as this one, the control system 200 of the fitness equipment further includes a monitoring triggering unit, a detection acquisition unit, a generation and sending unit, a generation and display unit, a desensitized uploading unit, and a receiving and updating unit configured in the control device 20.

[0045] The monitoring triggering unit is used by the control device to monitor the multimodal data in real time. If abnormal indicator values ​​are detected in the multimodal data, a graded security protection mechanism is triggered. The detection acquisition unit is used by the control device to acquire the new training input data and new historical training data corresponding to the new fitness device if it detects that the user has switched from the currently used fitness device to a new fitness device. The generation and sending unit is used to generate a new training plan adapted to the new fitness device based on the multidimensional ability profile, the new training input data, and the new historical training data, and send the new training plan to the new fitness device to control the new fitness device. The display unit is used by the control device to acquire the entire training data when it receives a training end command, and to generate a multi-dimensional training report based on the training data, and then display the multi-dimensional training report to the user. The anonymized upload unit is used by the control device to anonymize the entire training data to obtain anonymized full-process data, and to upload the anonymized full-process data to the cloud server, so that the cloud service can iteratively optimize the profile building model and the intelligent decision-making model to obtain model update data. The receiving update unit is used by the control device to receive the model update data sent by the cloud server, and to update the profile building model and the intelligent decision-making model based on the model update data.

[0046] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A control method of an exercise device, characterized by, The method comprises the following steps: The acquisition device synchronously acquires multi-modal data of a user during training and sends the multi-modal data to a control device, wherein the multi-modal data comprises physiological feature data, motion action data and environmental data; The control device constructs a multi-dimensional ability portrait of the user through a portrait construction model according to the multi-modal data, and generates a training scheme through an intelligent decision model according to the multi-dimensional ability portrait and acquired training input data and historical training data, wherein the portrait construction model and the intelligent decision model are both models issued by a cloud server; The fitness device receives the training scheme sent by the control device and controls its own output parameters according to the training scheme; The portrait construction model comprises a cross-modal attention layer, a feature fusion layer and a portrait output layer; the step of constructing a multi-dimensional ability portrait of the user through a portrait construction model according to the multi-modal data by the control device comprises: The control device performs spatio-temporal alignment and confidence weighted fusion on the multi-modal data to obtain a multi-modal data matrix; multi-dimensional features are extracted from the multi-modal data matrix through a deep feature extraction network based on an attention mechanism; the multi-dimensional features are processed through dimension reduction and standardization to obtain a multi-modal feature vector; The correlation between different modal features in the multi-modal feature vector is calculated through the cross-modal attention layer, and attention weights are allocated to obtain a weighted fusion feature vector; Deep feature interaction and compression are performed on the weighted fusion feature vector through a full connection network and a nonlinear activation function in the feature fusion layer to obtain a comprehensive feature representation; The comprehensive feature representation is linearly transformed and normalized through the portrait output layer to generate the multi-dimensional ability portrait of the user, wherein the multi-dimensional ability portrait comprises the user's strength level, endurance feature, action mode preference and fatigue recovery ability; The step of generating a training scheme through an intelligent decision model according to the multi-dimensional ability portrait and acquired training input data and historical training data comprises: If the number of data in the historical training data is less than a preset number of data, the questionnaire answers and physical measurement results filled in by the user are obtained, pseudo-label data is generated according to the questionnaire answers and the physical measurement results, and the pseudo-label data is supplemented into the historical training data; The training input data set by the user is obtained, and the training input data, the historical training data and the multi-dimensional ability portrait are preprocessed and feature fused to obtain decision input data; The decision input data is input into the intelligent decision model for intelligent decision to obtain the training scheme, wherein, during intelligent decision, the intelligent decision model takes maximizing training benefits as the target, and the training benefits = w1*effect benefits + w2*security benefits + w3*adherence benefits, w1, w2 and w3 are weights.

2. The method of claim 1, wherein, The training program includes a resistance baseline, resistance increment, single training duration, movement type, and training frequency; the decision input data includes quantitative values ​​of user strength level, endurance characteristics, movement pattern preference, and fatigue recovery ability. The intelligent decision-making model sets the resistance baseline and the resistance increment based on the user's strength level quantification value, sets the single training duration based on the endurance characteristic quantification value, sets the proportion of compound movements of the movement type based on the movement pattern preference quantification value, and sets the training frequency based on the fatigue recovery ability quantification value.

3. The method of claim 2, wherein, The step of controlling its own output parameters according to the training scheme includes: Adjust its resistance according to the resistance baseline value and the resistance increment; The duration of each training session is controlled according to the duration of a single training session; Match its own action pattern according to the action type; The training schedule is planned according to the training frequency.

4. The method according to any one of claims 1 to 3, characterized in that, After the step of controlling its own output parameters according to the training scheme, the method further includes: The control device monitors the multimodal data in real time. If abnormal index values ​​are detected in the multimodal data, a hierarchical security protection mechanism is triggered.

5. The method according to any one of claims 1 to 3, characterized in that, After the step of controlling its own output parameters according to the training scheme, the method further includes: If the control device detects that the user has switched from the currently used fitness equipment to a new fitness equipment, it acquires the new training input data and new historical training data corresponding to the new fitness equipment. A new training plan adapted to the new fitness equipment is generated based on the multidimensional capability profile, the new training input data, and the new historical training data, and the new training plan is sent to the new fitness equipment to control the new fitness equipment.

6. The method of claim 5, wherein, The method further includes: If the control device receives a training end command, it acquires the entire training data and generates a multi-dimensional training report based on the training data, then displays the multi-dimensional training report to the user.

7. The method of claim 6, wherein, The method further includes: The control device de-identifies the entire training data to obtain de-identified full data, and uploads the de-identified full data to the cloud server so that the cloud service can iteratively optimize the profile building model and the intelligent decision-making model to obtain model update data; The control device receives model update data from the cloud server and updates the profile building model and the intelligent decision-making model based on the model update data.

8. A control system for an exercise device, characterized by, The system includes fitness equipment, a cloud server, data acquisition devices, and control devices. The data acquisition devices are equipped with a data acquisition and transmission unit, the control devices are equipped with a data generation and construction unit, and the fitness equipment is equipped with a receiving and control unit. The acquisition and transmission unit is used to acquire multimodal data of the user during the training process synchronously and send the multimodal data to the control device. The multimodal data includes physiological feature data, motion data and environmental data. The construction and generation unit is used by the control device to construct a multi-dimensional capability profile of the user through a profile construction model based on the multi-modal data, and to generate a training scheme through an intelligent decision model based on the multi-dimensional capability profile, the acquired training input data and historical training data. The profile construction model and the intelligent decision model are both models distributed by a cloud server. The receiving and control unit is used for the fitness equipment to receive the training plan sent by the control device, and to control its own output parameters according to the training plan; The portrait construction model includes a cross-modal attention layer, a feature fusion layer, and a portrait output layer; the construction generation unit is also used for: The control device performs spatiotemporal alignment and confidence-weighted fusion on the multimodal data to obtain a multimodal data matrix; it then extracts multidimensional features from the multimodal data matrix using a deep feature extraction network based on an attention mechanism; and finally performs dimensionality reduction and standardization on the multidimensional features to obtain a multimodal feature vector. The correlation between different modal features in the multimodal feature vector is calculated through the cross-modal attention layer, and attention weights are assigned to obtain a weighted fusion feature vector; Based on the weighted fusion feature vector, a comprehensive feature representation is obtained by deep feature interaction and compression through a fully connected network and a nonlinear activation function in the feature fusion layer; The comprehensive feature representation is linearly transformed and normalized through the image output layer to generate the user's multidimensional ability profile, which includes the user's strength level, endurance characteristics, movement pattern preferences, and fatigue recovery ability. The construction generation unit is also used for: If the number of data entries in the historical training data is less than the preset number of data entries, then the user's questionnaire answers and physical test results are obtained, pseudo-label data is generated based on the questionnaire answers and physical test results, and the pseudo-label data is added to the historical training data. Obtain the training input data set by the user, and perform preprocessing and feature fusion on the training input data, the historical training data, and the multidimensional capability profile to obtain decision input data; The decision input data is input into the intelligent decision model to make intelligent decisions and obtain the training scheme. In the intelligent decision-making process, the intelligent decision model aims to maximize the training benefit, where the training benefit = w1 * effect benefit + w2 * safety benefit + w3 * persistence benefit, and w1, w2, and w3 are weights.

Images