Treadmill control method, treadmill, electronic device, and storage medium

Abstract

The application relates to the technical field of sports equipment, in particular to a treadmill control method, a treadmill, an electronic device and a storage medium. The treadmill control method is applied to controlling a treadmill, and target user training purpose information, user health data and static physical sign data need to be acquired first; dynamic physical sign data of the target user is collected through an intelligent wearable device; natural language instruction sentences are obtained by generating instruction language according to the training purpose information, the user health data, the static physical sign data and the dynamic physical sign data; the target large language model which is adjusted in advance is used to analyze the natural language instruction sentences, and user feature analysis information is obtained; a pre-trained decision-making agent generates a treadmill control strategy suitable for the target user on the basis of the user feature analysis information; and the running state of the treadmill is controlled based on the treadmill control strategy, so that personalized running modes can be provided for different users.

Description

Technical Field

[0001] This application relates to the field of sports equipment technology, and in particular to a treadmill control method, a treadmill, electronic equipment, and a storage medium. Background Technology

[0002] Treadmills, as an important piece of fitness equipment, are widely used in gyms and home workout settings. In the fitness industry, how to use intelligent technologies to help users better utilize treadmills to achieve their ideal fitness results has become an important research direction.

[0003] While existing treadmills possess basic functions such as speed and incline adjustment, users often face numerous challenges during use. First, users need to rationally set the treadmill parameters according to their own physical condition and fitness goals, which is difficult for those lacking fitness knowledge. Second, even users with some fitness knowledge find it difficult to consistently and stably control key indicators such as pace and heart rate during a run to ensure optimal fitness results. Clearly, current treadmill technology falls short in meeting users' personalized training needs. These issues limit the effective application of treadmills in the fitness field. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a treadmill control method, a treadmill, electronic equipment, and a storage medium, capable of providing personalized operating modes for different users.

[0005] A treadmill control method according to a first aspect of this application is applied to control a treadmill equipped with a smart wearable device, the method comprising:

[0006] Acquire training objective information, user health data, and static vital sign data of the target users;

[0007] The target user's dynamic vital signs data are collected through the smart wearable device;

[0008] Instructions are generated based on the training objective information, the user health data, the static vital signs data, and the dynamic vital signs data to obtain natural language instruction statements;

[0009] The natural language instruction statement is reasoned and parsed based on the pre-adapted target large language model to obtain user feature parsing information;

[0010] Based on the user feature parsing information, a pre-trained decision-making agent generates a treadmill control strategy adapted to the target user.

[0011] The treadmill's operating status is controlled based on the aforementioned treadmill control strategy.

[0012] According to some embodiments of this application, before performing reasoning and parsing on the natural language instruction statement based on the pre-adapted target large language model to obtain user feature parsing information, the method further includes pre-adapting and adjusting the target large language model, specifically including:

[0013] Acquire knowledge data in the sports domain, sample data for training purposes, user sample data, and the original large-scale natural language model;

[0014] Based on the sports domain knowledge data, the training objective sample data, and the user sample data, instruction texts are generated to obtain multiple instruction text sample data.

[0015] The indicator sample data is labeled to obtain indicator label information corresponding to each indicator sample data;

[0016] Based on multiple sample data of the indicated terms and the indicated term annotation information corresponding to each sample data of the indicated terms, an adaptive adjustment dataset is constructed.

[0017] The large natural language model is trained using the adaptive dataset.

[0018] In response to the natural language large model meeting the preset adjustment expectation conditions during adaptive training, the natural language large model is identified as the target large language model.

[0019] According to some embodiments of this application, the adaptive training of the large natural language model based on the adaptive dataset includes:

[0020] Select the indicator sample data from the adaptive adjustment dataset and input it into the natural language large model to obtain the output result of this round of adaptive adjustment;

[0021] The indicator annotation information corresponding to the indicator sample data is queried from the adaptive adjustment dataset, and the difference analysis is performed based on the indicator annotation information and the current round of adaptive adjustment output to obtain the current round of deviation data;

[0022] Based on the current round of deviation data, the natural language large model is subjected to model adaptation adjustment operations to update the natural language large model;

[0023] Based on the updated natural language big model, the process returns to selecting the indicator sample data from the adaptive training dataset and inputting it into the natural language big model until the current round of deviation data is less than the preset expected value, thus determining that the natural language big model meets the expected adjustment conditions in the adaptive training.

[0024] According to some embodiments of this application, before the pre-trained decision agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, the method further includes pre-training the decision agent, specifically including:

[0025] A motion simulation process is constructed to match multiple candidate users; wherein, the motion simulation process is used to simulate the environment of treadmill operation and the exercise process of the candidate users;

[0026] Acquire multiple user feature sample data and the original agent network;

[0027] The user feature sample data is processed by the intelligent agent network to obtain the simulation strategy for this round of training;

[0028] The training simulation strategy described above is used to conduct interactive simulation in the motion simulation process in order to determine the treadmill state simulation data and the user state simulation data in the motion simulation process.

[0029] The simulation reward signal is calculated based on the treadmill status simulation data and the user status data;

[0030] The agent network is updated based on the simulated reward signal;

[0031] Based on the updated agent network, the process returns to processing the user feature sample data through the agent network until the current training simulation strategy meets the preset training expectation conditions, and the decision agent is generated according to the agent network.

[0032] According to some embodiments of this application, the decision-making agent, which is pre-trained, generates a treadmill control strategy adapted to the target user based on the user feature parsing information, including:

[0033] The decision-making agent retrieves knowledge entity information related to the user feature parsing information from a pre-built knowledge graph database.

[0034] The user feature parsing information and the knowledge entity information are fused together to obtain motion context representation information corresponding to the target user;

[0035] A control strategy is generated based on the motion context representation information to obtain a treadmill control strategy adapted to the target user.

[0036] According to some embodiments of this application, the step of generating a natural language instruction statement based on the training objective information, the user health data, the static vital signs data, and the dynamic vital signs data includes:

[0037] Key information is extracted from the training objective information, the user health data, the static vital signs data, and the dynamic vital signs data to obtain key motion indication information corresponding to the target user;

[0038] Natural language is generated based on the key motion instruction information to obtain intermediate instruction statements, and the intermediate instruction statements are then checked.

[0039] In response to the intermediate instruction statement passing the statement check, the intermediate instruction statement is determined to be the natural language instruction statement.

[0040] According to some embodiments of this application, after controlling the operating state of the treadmill based on the treadmill control strategy, the method further includes:

[0041] The dynamic vital signs data of the target user are re-collected through the smart wearable device;

[0042] Based on the re-collected dynamic vital sign data, the system returns to the process of generating instructions based on the training objective information, the user's health data, the static vital sign data, and the dynamic vital sign data, in order to continuously update the treadmill control strategy.

[0043] According to some embodiments of this application, the step of returning to the execution of generating instructions based on the training objective information, the user health data, the static vital signs data, and the dynamic vital signs data, so as to continuously update the treadmill control strategy, includes:

[0044] Based on the re-collected dynamic vital sign data, the process returns to generate an instruction statement based on the training objective information, the user health data, the static vital sign data, and the dynamic vital sign data, resulting in an updated natural language instruction statement.

[0045] Based on the analysis of the updated natural language instruction statement through the target large language model, the updated user feature analysis information is obtained.

[0046] Based on the updated user feature parsing information and the previous treadmill control strategy, the decision-making agent updates the treadmill control strategy for the target user.

[0047] The treadmill's operating status is controlled based on the updated treadmill control strategy.

[0048] According to some embodiments of this application, updating the treadmill control strategy for the target user by the decision-making agent based on the updated user feature parsing information and the previous round of treadmill control strategy includes:

[0049] The decision-making agent evaluates the current motion state of the target user based on the updated user feature parsing information.

[0050] Based on the current exercise state, the updated user feature parsing information, and the treadmill control strategy of the previous round, calculate the speed change and the incline change for this round.

[0051] A smoothing control strategy is generated based on the changes in speed and slope in the current cycle to update the treadmill control strategy.

[0052] According to some embodiments of this application, the step of reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information includes:

[0053] The natural language instruction statement is uploaded to the target large language model deployed on the cloud server, and the natural language instruction statement is reasoned and parsed through the target large language model to obtain the user feature parsing information;

[0054] The pre-trained decision-making agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, including:

[0055] Based on the user feature parsing information, the decision-making agent deployed on the cloud server generates a treadmill control strategy adapted to the target user.

[0056] According to some embodiments of this application, the step of reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information includes:

[0057] The natural language instruction statement is uploaded to the target large language model deployed on the terminal device, and the target large language model is used to perform reasoning and parsing on the natural language instruction statement to obtain the user feature parsing information.

[0058] The pre-trained decision-making agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, including:

[0059] Based on the user feature parsing information, the decision-making agent deployed on the terminal device generates a treadmill control strategy adapted to the target user.

[0060] According to some embodiments of this application, the step of reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information includes:

[0061] The natural language instruction statement is uploaded to the target large language model deployed on the treadmill, and the natural language instruction statement is reasoned and parsed through the target large language model to obtain the user feature parsing information;

[0062] The pre-trained decision-making agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, including:

[0063] Based on the user feature parsing information, the decision-making agent deployed on the treadmill generates a treadmill control strategy adapted to the target user.

[0064] A treadmill according to a second aspect embodiment of the present application is configured to be controlled by a treadmill control method according to any one of the first aspect embodiments of the present application.

[0065] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the treadmill control method as described in any one of the embodiments of the first aspect of this application.

[0066] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that is executed by a processor to implement the treadmill control method as described in any one of the embodiments of the first aspect of this application.

[0067] The treadmill control method, treadmill, electronic device, and storage medium according to the embodiments of this application have at least the following beneficial effects:

[0068] The treadmill control method according to embodiments of this application is applied to control a treadmill equipped with a smart wearable device. The method requires first acquiring the target user's training objective information, user health data, and static vital sign data; then, collecting the target user's dynamic vital sign data through the smart wearable device; generating instructions based on the training objective information, user health data, static vital sign data, and dynamic vital sign data to obtain natural language instructions; performing reasoning and parsing on the natural language instructions using a pre-adapted target large language model to obtain user feature parsing information; and generating a treadmill control strategy adapted to the target user based on the user feature parsing information using a pre-trained decision-making agent; finally, controlling the treadmill's operating state based on the treadmill control strategy. In this way, personalized operating modes can be provided for different users.

[0069] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0070] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0071] Figure 1 A flowchart illustrating a treadmill control method provided in an embodiment of this application;

[0072] Figure 2 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0073] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0074] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0075] In the description of this specification, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. In the description of this application, the identification of specific steps does not imply a limitation on the order of steps or execution logic; the execution order and execution logic between the various steps should be understood and inferred from the content described in the embodiments.

[0076] Treadmills, as an essential piece of fitness equipment, are widely used in gyms and home workout settings. With the rapid development of artificial intelligence (AI) technology, various industries are actively exploring how to apply AI to specific product functions to improve user experience and product performance. In the fitness field, how to use intelligent methods to help users better utilize treadmills to achieve their ideal fitness results has become an important research direction.

[0077] While existing treadmills possess basic functions such as speed and incline adjustment, users often face several challenges during use. First, users need to set the treadmill parameters appropriately based on their physical condition and fitness goals, which is difficult for those lacking fitness knowledge. Second, even users with some fitness knowledge often struggle to consistently and stably control key indicators such as pace and heart rate during a run to ensure optimal fitness results. Furthermore, the user experience of treadmills is relatively monotonous, lacking personalized and intelligent guidance, and failing to meet the diverse needs of different users.

[0078] The technical problems existing in the background technology include the following aspects:

[0079] The main drawback of existing treadmill technology lies in the difficulty for users to achieve their ideal results. Although treadmills have basic functions such as adjusting speed and incline, users often lack relevant fitness knowledge and cannot accurately set and adjust these parameters to achieve the best fitness effect. Furthermore, users find it difficult to monitor and adjust their physical condition in real time during a run, such as key indicators like heart rate and blood pressure, which may lead to poor training results or even health risks.

[0080] Existing treadmills cannot provide personalized training programs based on a user's individual physical condition, health characteristics, and training goals. Users need to figure out suitable running styles and parameter settings on their own, which is unrealistic for most users without professional fitness guidance. Therefore, existing treadmills have significant shortcomings in meeting users' personalized needs.

[0081] Current treadmills lack intelligent guidance features. Users cannot receive real-time feedback and adjustment suggestions during their runs, nor can they dynamically adjust the treadmill's operation based on their own physical condition. Such treadmills, lacking intelligent guidance, struggle to help users achieve scientific and efficient fitness training.

[0082] It is evident that current treadmill technology has significant shortcomings in areas such as personalized training program development, real-time body condition monitoring and feedback, intelligent guidance, and user experience. These issues limit the effective application of treadmills in the fitness field and also affect user experience and satisfaction. Therefore, developing an intelligent treadmill that can solve these problems has become an important development direction in the fitness equipment industry.

[0083] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a treadmill control method, a treadmill, electronic equipment, and a storage medium, capable of providing personalized operating modes for different users.

[0084] The following explanation is based on the accompanying drawings.

[0085] Reference Figure 1 The treadmill control method according to the embodiments of this application is applied to control a treadmill equipped with a smart wearable device. The treadmill control method of this application may include:

[0086] Step S101: Obtain the target user's training objective information, user health data, and static vital sign data;

[0087] Step S102: Collect dynamic vital sign data of the target user through a smart wearable device;

[0088] Step S103: Generate instructions based on training objective information, user health data, static vital signs data, and dynamic vital signs data to obtain natural language instruction statements.

[0089] Step S104: Based on the pre-adapted target large language model, perform reasoning and parsing on the natural language instruction statement to obtain user feature parsing information;

[0090] Step S105: Based on the user feature parsing information, the pre-trained decision-making agent generates a treadmill control strategy adapted to the target user.

[0091] Step S106: Control the operating status of the treadmill based on the treadmill control strategy.

[0092] The goal of the treadmill control method in this application is to provide users with a personalized treadmill exercise experience, which is based on clever adaptation to individual user needs. By applying a target large language model and decision-making intelligence to the field of motion analysis, it is possible to generate exclusive treadmill control strategies according to the unique needs and physical conditions of different users. This personalized service is not only reflected in the precise setting of exercise parameters, but also in real-time companionship and intelligent guidance during the user's exercise process. For example, this application embodiment can provide encouraging voice feedback or suggestions for adjusting the exercise plan in a timely manner based on the user's exercise performance and physical reactions, allowing the user to feel the thoughtful care brought by the intelligent device during exercise, thereby improving the user's exercise enthusiasm and long-term adherence to exercise. In this way, runners do not need to intervene too much or have professional fitness knowledge; it is like having a full-time personal trainer by their side, adjusting the training plan according to changes in physical condition at any time.

[0093] In some embodiments, step S101 involves acquiring the target user's training objective information, user health data, and static vital sign data.

[0094] It's important to note that obtaining information about the target user's training objectives is the first step in developing a personalized treadmill plan. In some implementations, this can be achieved by guiding users to complete a comprehensive fitness goal questionnaire via a mobile app or other terminal device that comes with the treadmill. This questionnaire includes their primary purpose for using the treadmill, such as weight loss, muscle gain, improving cardiovascular endurance, preparing for a marathon, or improving overall health. Furthermore, users can specify concrete goals they wish to achieve, such as losing a certain number of kilograms in three months or improving their running endurance to a certain number of minutes they can run continuously. This detailed goal information will serve as a crucial basis for generating subsequent treadmill control strategies.

[0095] Collecting user health data is a crucial step in ensuring safe and effective training. This data includes the user's medical history, such as whether they have heart disease, high blood pressure, asthma, knee problems, or other health issues that may affect exercise. Additionally, it allows for understanding the user's current physical condition, such as any chronic pain or recent injury recovery. Users can manually enter this information or obtain more accurate data by integrating electronic health records (if the user authorizes access). Furthermore, users can be asked about their general health habits, such as sleep quality, dietary habits, and daily activity levels. This information helps in comprehensively assessing the user's health condition and supports the development of a suitable treadmill training plan.

[0096] The collection of static physical characteristics data provides users with basic physiological parameters for their treadmill exercise. This includes basic information such as the user's height, weight, gender, age, and body mass index (BMI). This data can be entered by the user through a mobile app or other terminal device, or automatically obtained through connection to devices such as smart body fat scales. For example, a user's height and weight can be used to calculate their calorie expenditure rate while running, and to determine the appropriate treadmill settings for their body size, such as lane length and handrail height. Furthermore, the user's body composition information, such as body fat percentage and muscle mass, can also serve as a reference for optimizing treadmill training plans, helping to determine whether a training mode focusing on fat loss or muscle gain is needed.

[0097] In step S102 of some embodiments, dynamic vital sign data of the target user are collected through a smart wearable device;

[0098] It's important to clarify that, firstly, smart wearable devices can include a wide variety of items such as smartwatches / bands, heart rate monitors, and smart insoles. These devices can monitor various physiological indicators of a user in real time during exercise. For example, smartwatches / bands can accurately measure a user's heart rate, blood pressure changes, and cadence and stride length during running; heart rate monitors provide more detailed heart rate data, including heart rate variability, which is very useful for assessing a user's exercise intensity and level of fatigue; smart insoles can monitor the distribution of plantar pressure, gait characteristics, and ground contact time and airtime during running. This wealth of dynamic vital sign data provides comprehensive real-time information for subsequent personalized control strategies.

[0099] Secondly, data collected by smart wearable devices helps reflect trends in a user's exercise status. In addition to basic physiological indicators, these devices can record time-series data, such as the duration of continuous exercise and the time spent in different intensity ranges. By integrating this time-series data, the decision-making agent can analyze the user's exercise patterns and endurance trends. For example, if data shows that a user's speed significantly decreases and heart rate increases in the latter half of each run, the intelligent system may appropriately reduce the intensity of the later stages of training or suggest increasing rest intervals in subsequent training plans, thereby helping the user gradually improve endurance and optimize training results.

[0100] In the process of collecting dynamic vital sign data of target users through smart wearable devices, in order to ensure that the data collection and transmission do not consume too much network resources, and at the same time ensure that changes in key vital signs are effectively monitored, the time interval method and the threshold overload method can be used.

[0101] For the time interval method, according to some embodiments provided in this application, step S102, which involves collecting dynamic vital sign data of the target user through a smart wearable device, may include:

[0102] Using smart wearable devices, vital sign data is collected from the target user at regular intervals to update dynamic vital sign data.

[0103] It should be noted that the time interval method refers to setting a time interval and collecting and transmitting vital sign data at each time interval. This method can effectively control the frequency of data transmission, thereby reducing the consumption of network resources.

[0104] It should be noted that the data collection interval can be fixed. For example, heart rate data can be collected every 10 seconds. The advantage of a fixed collection interval is that it provides evenly distributed data points, facilitating periodic analysis and observation of long-term trends. This setting is suitable for monitoring under steady-state conditions, such as when a user is engaged in continuous low-to-moderate intensity running, ensuring stable data acquisition and maintaining data continuity and integrity.

[0105] However, in some situations, setting a flexible data collection interval is more appropriate. For example, when a user is engaged in high-intensity interval training, their activity level changes rapidly. In this case, the data collection interval can be shortened, such as collecting data every 5 seconds, to capture changes in vital signs more promptly. Conversely, when a user is engaged in low-intensity exercise or is in a resting phase, the data collection interval can be appropriately extended, such as collecting data every 15 or 20 seconds, to reduce the amount of data and conserve network resources.

[0106] Furthermore, the data collection interval can be flexibly adjusted based on the user's individual needs and health condition. For example, users with specific health needs, such as heart disease patients, may require closer monitoring of their heart rate changes, in which case a shorter data collection interval can be set. Conversely, healthy users can use a relatively longer data collection interval during regular exercise.

[0107] Because smart wearable devices are programmable, the data collection time interval in this embodiment needs to be determined based on the specific data type of vital signs and the exercise scenario. For example, for heart rate monitoring during treadmill exercise, if the user's exercise intensity is low, the data collection interval can be appropriately extended, such as collecting data once every 15 seconds; while for high-intensity interval training, data may need to be collected more frequently, such as once every 5 seconds, in order to more accurately capture changes in heart rate.

[0108] This ability to flexibly set data collection intervals allows smart wearable devices to optimize data collection and transmission efficiency, reduce network resource consumption, improve device battery life, and provide users with more considerate and personalized monitoring services, while ensuring that key vital signs changes are effectively monitored.

[0109] For the threshold overload method, according to some embodiments provided in this application, step S102, which involves collecting dynamic vital sign data of the target user through a smart wearable device, may include:

[0110] By using smart wearable devices, vital sign data monitoring operations are performed on target users to determine real-time vital sign monitoring data;

[0111] Real-time vital sign monitoring data is collected in response to the real-time vital sign monitoring data meeting the preset threshold conditions.

[0112] The dynamic vital signs data of the target user are updated based on real-time vital signs monitoring data.

[0113] It should be noted that the threshold overload method refers to setting thresholds for one or more key vital signs, and data collection and transmission only occur when the vital signs data exceed or fall below these thresholds. This method can further reduce unnecessary data transmission because transmission only occurs when there are significant changes in vital signs data, thereby saving network resources.

[0114] For example, for heart rate monitoring, an upper threshold and a lower threshold can be set. If the user's heart rate fluctuates within the normal range, the system will not transmit data frequently; only when the heart rate exceeds the upper threshold (e.g., 180 beats / minute) or falls below the lower threshold (e.g., 50 beats / minute) will the threshold setting conditions be triggered, thereby transmitting data and reminding the user or the system to take appropriate measures.

[0115] The threshold setting needs to be personalized based on the user's health condition and exercise goals. For example, for users with a history of heart disease, the upper limit of the heart rate threshold may need to be set lower to ensure the user's safety. For professional athletes, a higher threshold may be needed to accommodate their high-intensity training needs.

[0116] In some more specific implementations, the time interval method and the threshold overload method can be combined in practical applications to achieve better results. For example, a basic time interval (such as collecting data every 10 seconds) can be set, while a threshold for key vital signs can be set. Under normal circumstances, the system collects and transmits data according to the time interval; however, when the vital signs exceed the threshold, the system will collect and transmit data immediately without waiting for the next time interval.

[0117] This combined approach ensures real-time data transmission and monitoring of key vital signs while minimizing network resource consumption. For example, during a user's regular running training, heart rate data is transmitted every 10 seconds; however, if the user's heart rate suddenly rises to near a preset upper limit threshold, the system will immediately transmit the current heart rate data to adjust the treadmill's speed or incline in a timely manner, ensuring the user's exercise safety.

[0118] By employing time interval and threshold overload methods, smart wearable devices can optimize data collection and transmission strategies, reduce network resource consumption, and improve overall system efficiency while ensuring effective monitoring of key vital signs changes.

[0119] In step S103 of some embodiments, instructions are generated based on training objective information, user health data, static vital signs data, and dynamic vital signs data to obtain natural language instructions.

[0120] It's important to note that the core task of this step is to integrate multi-dimensional user data into a structured natural language text for subsequent large language model processing. This step parses data from different sources, including user-inputted training objectives, health status, and basic vital signs, as well as dynamic vital signs data monitored in real time by smart wearable devices. Then, using preset sentence templates or natural language generation algorithms, these scattered data points are woven into a fluent, coherent, and complete natural language instruction statement.

[0121] In some more specific embodiments, an automated instruction statement generation module can be constructed first. This module can receive and parse structured user data from various channels, including basic user information, health status, training goals, and real-time vital sign monitoring data. The module has multiple preset statement templates, optimized for different user characteristics and training scenarios. For example, for users aiming to lose weight, the template might mention low-intensity aerobic exercise and prolonged exercise duration; while for users aiming to improve cardiopulmonary function, the template might emphasize high-intensity interval training and heart rate monitoring.

[0122] Then, the instruction statement generation module uses natural language processing (NLP) algorithms to analyze and match the input data to determine the most suitable statement template. These NLP algorithms can fill in and adjust variables in the statement template based on the user's health data and training objectives. For example, by substituting specific values ​​such as the user's age, weight, and heart rate into the corresponding positions in the template, personalized natural language instructions are generated.

[0123] In some more specific embodiments, to improve the accuracy and usability of the instruction statements, the instruction statement generation module can also integrate data verification and feedback mechanisms. After generating the instruction statement, the module checks whether the data in the statement conforms to logic and medical common sense. For example, it checks whether the heart rate value is within a reasonable range and whether the exercise intensity suggestion matches the user's health condition. If data anomalies or logical errors are found in the statement, the instruction statement generation module automatically triggers a correction process to regenerate the instruction statement.

[0124] It is worth noting that multi-dimensional information such as the target user's training objectives, health data, and static and dynamic vital signs can be integrated into a semantically coherent and information-rich instruction statement through natural language description. This representation not only covers various user characteristics but also preserves the semantic details of the original data. For example, the statement "The user wants to train for a marathon, is 25 years old, weighs 60 kg, has mild knee discomfort, and currently has a heart rate of 120 beats per minute" organically integrates the user's training goal (marathon training), basic physical condition (age, weight), health problem (knee discomfort), and real-time physiological state (heart rate). Compared to structured data tables, this natural language form can more naturally express complex and multi-dimensional related information, providing input that is closer to human cognitive habits for subsequent model processing and helping to generate outputs that better meet the user's actual needs.

[0125] In some embodiments, step S104 involves reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information.

[0126] It's important to note that to achieve efficient inference and parsing, the target large language model underwent a pre-adaptation adjustment (also known as fine-tuning). This pre-adaptation adjustment involved training with a large amount of text data relevant to the treadmill application domain. This data covered various fitness scenarios, user health issues, and exercise physiology knowledge. For example, text data containing descriptions of different user characteristics, training goals, and health conditions, along with corresponding exercise suggestions and treadmill parameter settings, could be collected. By adapting to this data, the target large language model learned how to map key information in natural language instructions to specific user characteristics, thus providing an accurate foundation for subsequent parsing.

[0127] Then, when reasoning and parsing natural language instructions, a multi-stage processing flow can be adopted to improve the accuracy and comprehensiveness of the parsing. First, the target large language model performs preliminary semantic understanding and key information extraction on the input statement, identifying entities (such as the user's age, weight, heart rate, etc.) and relationships (such as the association between training goals and health status) in the statement. Next, using the attention mechanism within the target large language model, weights are assigned to different parts of the statement, highlighting key information and reducing the weight of irrelevant or secondary content. For example, if the statement mentions the user's history of heart disease and current high-intensity training goals, the target large language model will give higher weight to the history of heart disease because it has a significant impact on the user's exercise safety. Through this multi-stage processing approach, the target large language model can more accurately parse the user's feature information, providing reliable support for generating personalized treadmill control strategies.

[0128] Furthermore, to further enhance the reasoning performance of the target large language model, knowledge graph technology can be integrated. Knowledge graphs store entities, concepts, and relationships related to fields such as fitness and exercise physiology, including suitable exercise types for different health problems and the impact of various exercise parameters on fitness outcomes. During the parsing of natural language instructions by the target large language model, the knowledge graph can be queried in real time, matching and associating information from the statement with entities and relationships within the knowledge graph. For example, when the target large language model parses that a user's goal is to improve cardiopulmonary function and that they have mild asthma, suitable aerobic exercise methods and precautions for asthma patients can be retrieved from the knowledge graph and integrated into the user feature parsing results. This not only enriches the content of the parsed information but also ensures the scientific validity and rationality of the parsing results, providing a more comprehensive and accurate description of user characteristics for subsequent decision-making agents.

[0129] It is worth noting that the large language model has undergone extensive training in natural language processing, possessing powerful semantic understanding, contextual awareness, and text generation capabilities. It can deeply analyze the key information in these natural language instructions, understand the user's true needs and physical condition, and transform this into structured user feature parsing. For example, regarding the user feature parsing information "The user wants to train for a marathon, is 25 years old, weighs 60 kg, has mild knee discomfort, and currently has a heart rate of 120 beats per minute," it can identify the long-term, high-intensity endurance training requirement corresponding to "marathon training," and the potential risk of needing to appropriately control exercise intensity and reduce knee impact implied by "mild knee discomfort." Furthermore, the large language model can generate detailed, semantically logical user feature descriptions based on the parsing results, providing clear, accurate, and comprehensive evidence for further decision-making—something traditional data processing methods cannot match.

[0130] On the other hand, a wealth of knowledge in the field of sports and fitness, including training methods, exercise physiology, and health advice, often exists in the form of natural language, such as professional literature, fitness guides, and coaching recommendations. Therefore, by adaptively adjusting the large language model, this rich domain knowledge can be integrated into the model's training. In this way, when processing user data, the large language model can combine domain knowledge for reasoning and parsing, generating more professional and scientific user characteristic information. For example, after learning knowledge such as "marathon training requires gradually increasing running distance to improve endurance" and "knee problems should avoid prolonged high-intensity impact exercise," the model can better understand the relationship between user goals and physical condition, thereby providing more accurate guidance information for the decision-making agent and assisting it in generating more reasonable treadmill control strategies.

[0131] According to some embodiments of this application, the instruction is generated based on the training objective information, user health data, static vital signs data, and dynamic vital signs data in step S103, resulting in a natural language instruction statement, which may include:

[0132] Step S201: Extract key information from training objective information, user health data, static vital signs data, and dynamic vital signs data to obtain key information on movement indications corresponding to the target user.

[0133] Step S202: Generate natural language based on key motion instruction information to obtain intermediate instruction statements, and perform statement checks on the intermediate instruction statements;

[0134] Step S203: In response to the intermediate instruction statement passing the statement check, the intermediate instruction statement is determined to be a natural language instruction statement.

[0135] In step S201 of some embodiments, key information is extracted from training objective information, user health data, static vital signs data and dynamic vital signs data to obtain key information of motion indication corresponding to the target user.

[0136] It should be noted that the embodiments of this application first require the extraction of key information from the collected user data. This involves analyzing and filtering various information input by the user to determine which information is most crucial for generating exercise instruction statements. For example, static physical characteristics such as the user's age, weight, height, and gender, as well as dynamic physical characteristics such as heart rate and blood pressure, are all important factors influencing exercise prescriptions. Regarding training objective information, the embodiments of this application need to be able to identify the user's specific goals, such as weight loss, muscle gain, or improving cardiopulmonary function. User health data includes past medical history, chronic diseases, and allergies, which are crucial for ensuring exercise safety. For example, for a 35-year-old user weighing 80 kg and 175 cm tall, whose training goal is weight loss and who has mild hypertension, the embodiments of this application will extract this key information to form key information for exercise instructions. This information will serve as the basis for generating exercise instruction statements.

[0137] In some embodiments, step S202 involves generating natural language based on key motion instruction information to obtain intermediate instruction statements, and then performing statement checks on the intermediate instruction statements.

[0138] It should be noted that after extracting the key information, this embodiment of the application needs to utilize a natural language generation model to convert this information into intermediate instruction statements in natural language form. This process needs to consider the fluency, accuracy, and readability of the language. This embodiment of the application will embed the key information into a suitable sentence structure according to preset sentence templates and grammatical rules. For example, based on the key information extracted above, the intermediate instruction statement generated by this embodiment of the application might be: "A 35-year-old male user, 175 cm tall, weighing 80 kg, aims to lose weight and has a mild hypertension problem. Please generate a suitable treadmill training plan for him." After generating the statement, this embodiment of the application will check the statement to ensure that it is free of grammatical errors, the information is complete, and the logic is clear. If errors or ambiguities are found in the statement, this embodiment of the application will correct them until the statement meets the high-quality standard.

[0139] In some embodiments, step S203 involves determining the intermediate instruction statement as a natural language instruction statement in response to the intermediate instruction statement passing a statement check.

[0140] It should be noted that after the intermediate instruction statement passes the statement check, this embodiment of the application determines it as a natural language instruction statement. This step marks the completion of the instruction statement generation process, and the natural language instruction statement will be used for subsequent large language model inference and parsing. The quality of the natural language instruction statement is directly related to the accuracy and effectiveness of subsequent model inference and parsing; therefore, this method can be used to improve the quality of the natural language instruction statement before determining the statement.

[0141] According to some embodiments of this application, before step S104, which involves reasoning and parsing the natural language instruction statement based on the pre-adapted target large language model to obtain user feature parsing information, the method further includes pre-adapting and adjusting the target large language model, which may specifically include:

[0142] Step S301: Obtain sports domain knowledge data, training objective sample data, user sample data, and the original natural language large model;

[0143] Step S302: Generate instruction text based on sports domain knowledge data, training objective sample data and user sample data to obtain multiple instruction text sample data;

[0144] Step S303: Perform annotation processing on each instruction sample data to obtain instruction annotation information corresponding to each instruction sample data;

[0145] Step S304: Based on multiple indicator sample data and the indicator annotation information corresponding to each indicator sample data, an adaptive adjustment dataset is constructed;

[0146] Step S305: Adapt and train the large natural language model based on the adaptive dataset;

[0147] Step S306: In response to the natural language large model meeting the preset adjustment expectation conditions during adaptive training, the natural language large model is identified as the target large language model.

[0148] In some embodiments, step S301 involves acquiring motion domain knowledge data, training objective sample data, user sample data, and the original large natural language model.

[0149] It's important to note that collecting a large amount of knowledge data related to the sports and fitness field is crucial. This data can include professional sports training methods, exercise physiology knowledge, strategies for achieving common fitness goals, and measures for preventing sports injuries. For example, professional literature and guidelines on how to improve cardiovascular function through treadmill training and how to conduct effective weight loss training can be collected. Additionally, sample data on training goals is needed, covering common user fitness goals such as weight loss, muscle gain, and improved endurance. User sample data includes basic information and exercise history of users of different ages, genders, and physical conditions. This data can be obtained from fitness apps, health surveys, and other sources. Furthermore, a raw natural language processing model is required as a basis for adaptive adjustments.

[0150] In some embodiments, sports domain knowledge data may include treadmill training plans, heart rate control methods, and principles of exercise physiology. Training purpose sample data may include purposes such as weight loss, muscle gain, and improving cardiopulmonary function. User sample data may include basic information such as the user's age, weight, health status (e.g., heart disease, knee problems), and exercise history.

[0151] In step S302 of some embodiments, instructions are generated based on sports domain knowledge data, training objective sample data, and user sample data to obtain multiple instruction sample data.

[0152] It should be noted that the collected data is used to generate sample instruction data. This sample instruction data is designed to simulate user needs and physical conditions that may be encountered in real-world usage scenarios. It should be understood that the sample instruction data needs to cover different user characteristics, training objectives, and health conditions.

[0153] In some embodiments, the generated instruction sample data may include: "User A, 30 years old, weighing 80 kg, goal is to lose weight, has mild knee pain, and wants to do low-intensity treadmill training", "User B, 45 years old, weighing 70 kg, goal is to improve cardiopulmonary function, has a history of hypertension, and wants to do moderate-intensity treadmill training", "User C, 25 years old, weighing 60 kg, goal is to increase endurance, has no health problems, and wants to do high-intensity treadmill training".

[0154] In some embodiments, step S303 involves annotating each instruction sample data to obtain instruction annotation information corresponding to each instruction sample data.

[0155] It should be noted that the annotation process is to provide corresponding annotation information for each instruction sample data. This annotation information includes user feature analysis, training objective analysis, and health status analysis. The annotation information will serve as a supervisory signal for model training, helping the model learn how to extract key information from natural language instructions.

[0156] In step S304 of some embodiments, an adaptive adjustment dataset is constructed based on multiple indicator sample data and indicator annotation information corresponding to each indicator sample data;

[0157] It should be noted that the adaptation dataset consists of sample data of adaptation instructions and corresponding annotation information. This adaptation dataset will be used to train a large natural language model to adapt it to the specific needs of the sports and fitness domain. The construction of the adaptation dataset needs to ensure the diversity and representativeness of the samples to cover different combinations of user characteristics, training goals, and health conditions.

[0158] In some embodiments, step S305 involves adaptively training a large natural language model based on an adaptively adapted dataset.

[0159] It should be noted that in this stage, the original large-scale natural language processing model is trained using the constructed adaptive training dataset. The goal of adaptive training is to enable the large-scale natural language processing model to accurately extract user feature parsing information from natural language instructions. During training, the large-scale natural language processing model learns how to map key information in instructions to structured data in the labeled information.

[0160] According to some specific embodiments of this application, 305 performing adaptive training on a large natural language model based on an adaptive dataset may include:

[0161] Select indicator sample data from the adaptive dataset and input it into the large natural language model to obtain the output results of this round of adaptive adjustment;

[0162] The system queries the indicator annotation information corresponding to the indicator sample data from the adaptive adjustment dataset, and performs difference analysis based on the indicator annotation information and the output results of this round of adaptive adjustment to obtain the deviation data of this round.

[0163] Based on the deviation data from this round, the large-scale natural language model is adapted and adjusted to update the large-scale natural language model;

[0164] Based on the updated natural language big model, return to the process of selecting indicator sample data from the adaptive training dataset and inputting it into the natural language big model until the current round of bias data is less than the preset expected value, thus determining that the natural language big model meets the expected conditions for adaptation training.

[0165] In some embodiments of this application, the process of adaptively training a large natural language model based on an adaptive dataset is an iterative optimization process aimed at enabling the large natural language model to accurately extract user feature parsing information from natural language instruction statements.

[0166] It's important to note that at the start of the training process, a batch of instruction sample data is selected from the adaptation dataset as input to the natural language processing (NLP) model. This sample data is carefully designed and prepared to cover diverse user characteristics, training objectives, and health conditions. The NLP model processes these input statements according to the current parameter settings and generates corresponding output results, i.e., the adaptation output results for this round. This output represents the NLP model's understanding and parsing of the input instruction statements in the current training state.

[0167] After obtaining the output of the natural language processing model (NLP) large-scale model, these results need to be compared with the corresponding instruction annotations in the adaptive dataset. The instruction annotations can be pre-generated by experts in the field of sports and fitness or through other reliable methods, containing all the key information that should be extracted from the instructions. By comparing the NLP large-scale model output and the annotations, the difference between the two can be calculated, i.e., the current bias data. This bias data quantifies the gap between the current output of the NLP large-scale model and the expected output. For example, if the NLP large-scale model's output for the above instruction sample data omits the health condition information of "mild knee pain," this omission will be identified when comparing with the annotations and reflected in the bias data.

[0168] Based on the obtained bias data from this round, optimization algorithms (such as gradient descent) are used to adjust the parameters of the large-scale natural language processing model (NLP). The aim is to reduce the discrepancy between the NLP output and the labeled information, enabling the NLP to more accurately parse instructions in the next training round. This process involves a large number of parameter updates and calculations within the NLP model, aiming to gradually improve its performance.

[0169] It is important to emphasize that adaptive training is an iterative process. After each parameter update, the large-scale natural language processing (NLP) model will again select indicator sample data from the adaptive training dataset for processing and repeat the above steps. Each iteration calculates new bias data and further adjusts the NLP model parameters accordingly. This cycle continues until the bias data is less than the preset expected value, meaning the difference between the NLP model's output and the labeled information is small enough to meet the expected training conditions.

[0170] For example, after multiple rounds of iterative training, when the natural language processing model can accurately cover all key information in the parsing of most instruction sample data, and the deviation data reaches a preset low threshold, it can be considered that the natural language processing model has completed adaptive training and can accurately parse user feature information, providing reliable support for the subsequent generation of treadmill control strategies.

[0171] Through the above process, the embodiments of this application demonstrate how to systematically adapt and train a large natural language model to accurately understand and parse user feature information, laying the foundation for the development of personalized treadmill training programs.

[0172] In some embodiments, step S306, in response to the natural language large model meeting preset adjustment expectation conditions during adaptive training, identifies the natural language large model as the target large language model.

[0173] It's important to note that during model training, certain adaptive adjustment conditions need to be set, such as the natural language processing model achieving a certain accuracy threshold on the validation set and the loss function reaching a certain level. When the natural language processing model meets these conditions during training, it can be identified as the target large language model. This target large language model will be used for subsequent inference and parsing to generate user feature parsing information.

[0174] In some embodiments, it is assumed that the preset condition is that the model achieves an accuracy of over 90% on the validation set. When the model achieves an accuracy of 92% on the validation set after multiple rounds of training, the model is considered to meet the expected condition and can be identified as the target large language model.

[0175] In this way, the embodiments of this application ensure the professionalism and accuracy of the target large language model in processing information in the field of sports and fitness, providing a reliable foundation for the subsequent generation of personalized treadmill control strategies.

[0176] In step S105 of some embodiments, a treadmill control strategy adapted to the target user is generated by a pre-trained decision-making agent based on user feature parsing information.

[0177] It's important to note that during the pre-training process, the decision-making agent needs to utilize training data that encompasses diverse user characteristics and corresponding treadmill control strategies. This data can be obtained from historical treadmill usage records, including users' static and dynamic physical characteristics, training goals, and corresponding treadmill settings. To improve data quality and usability, data cleaning and preprocessing techniques can be employed to remove outliers and noisy data, and the data can be normalized to allow for comparison and analysis of data from different dimensions on the same scale.

[0178] In some embodiments, the decision-making agent can process multi-dimensional user feature parsing information and comprehensively consider the impact of various factors on treadmill regulation. For example, when the parsed information includes data such as the user's maximum heart rate, target heart rate zone, and joint flexibility, the decision-making agent can extract key information through feature engineering and use multi-objective optimization algorithms to weigh the relationship between different factors. For an elderly user who wants to improve cardiopulmonary function but also has joint problems, the agent needs to limit the incline while increasing the heart rate to protect the joints. This can be achieved by constructing a decision model that considers multiple objective functions, such as minimizing heart rate deviation and maximizing comfort, while ensuring that the exercise intensity reaches a certain level.

[0179] In some more specific embodiments, to ensure the real-time nature and adaptability of the strategy, the decision-making agent needs to possess the ability to learn online and adjust dynamically. During actual operation, the decision-making agent can also dedicate a data receiving channel to directly and continuously receive the user's real-time dynamic vital sign data and adjust the treadmill's operating status in real time based on this data. For example, if the user's real-time heart rate exceeds the target range, the decision-making agent should immediately execute a preset emergency strategy, such as reducing the speed or pausing the treadmill, and promptly send a reminder message to the user. Furthermore, the decision-making agent can also dynamically adjust its internal model parameters based on the user's long-term exercise data and feedback to adapt to changes in the user's physical condition and exercise capacity. This can be achieved by implementing incremental learning algorithms or periodically retraining the model, ensuring that the agent's strategy always matches the user's latest needs and physical conditions.

[0180] Through the above implementation, the pre-trained decision-making agent can generate accurate, safe, and efficient treadmill control strategies based on a thorough understanding of user characteristics, meeting users' personalized fitness needs and improving the treadmill's intelligence level and user experience. The combination of a large language model and a decision-making agent represents a deep integration, fully leveraging the strengths of both in natural language processing and decision control. This combination demonstrates excellent performance in handling complex user data, utilizing domain knowledge, implementing personalized services, and continuously optimizing system performance.

[0181] According to some embodiments of this application, step S105, based on user feature parsing information, generates a treadmill control strategy adapted to the target user through a pre-trained decision-making agent, which may include:

[0182] Step S401: The decision-making agent retrieves knowledge entity information related to user feature parsing information from the pre-built knowledge graph database;

[0183] Step S402: The user feature parsing information and knowledge entity information are fused to obtain motion context representation information corresponding to the target user;

[0184] Step S403: Generate a control strategy based on motion context representation information to obtain a treadmill control strategy adapted to the target user.

[0185] In step S401 of some embodiments, the decision-making agent retrieves knowledge entity information associated with user feature parsing information from a pre-built knowledge graph database;

[0186] It's important to note that the decision-making agent first has access to a pre-built knowledge graph database, which contains rich knowledge in the sports and fitness domain. The knowledge entities in the knowledge graph cover various sports-related concepts and their interrelationships, such as different training goals (e.g., weight loss, muscle gain), health conditions (e.g., high blood pressure, joint problems), exercise types (e.g., treadmill training, strength training), exercise parameters (e.g., speed, incline), and exercise suggestions suitable for different user characteristics. When the decision-making agent receives user feature analysis information, it retrieves knowledge entities associated with these features from the knowledge graph. For example, if the user feature analysis information shows that the user's goal is weight loss and they have mild knee pain, the decision-making agent will search the knowledge graph for knowledge entities associated with "weight loss training" and "knee protection," such as low-impact treadmill training methods suitable for weight loss and exercise parameter settings that help protect the knees.

[0187] In step S402 of some embodiments, the user feature parsing information and knowledge entity information are fused to obtain motion context representation information corresponding to the target user;

[0188] It's important to note that after acquiring knowledge entity information related to user characteristics, the decision-making intelligence will fuse this information with the user characteristic parsing information. The purpose of this fusion is to integrate the user's specific characteristics with general knowledge from the knowledge graph, forming a comprehensive and personalized exercise context representation. This representation not only includes basic information such as the user's age, weight, and health status, but also incorporates professional exercise suggestions tailored to these characteristics, such as suitable exercise intensity, duration, and frequency. For example, for a user aiming to lose weight and experiencing knee pain, the fused exercise context representation might include: the user's basic body data, the calorie expenditure target for weight loss, a suitable low-impact running speed and incline range, and the recommended duration for each training session. This representation provides comprehensive contextual support for subsequently generating precise treadmill control strategies.

[0189] In step S403 of some embodiments, a control strategy is generated based on motion context representation information to obtain a treadmill control strategy adapted to the target user.

[0190] It's important to note that the decision-making agent uses its internal decision-making algorithm to generate a treadmill control strategy tailored to the target user, based on motion context representation information. This algorithm may be based on reinforcement learning, deep learning, or other advanced machine learning techniques. These algorithms can consider multiple objective functions and constraints, such as maximizing exercise effectiveness, ensuring exercise safety, and improving exercise comfort. When generating the control strategy, the algorithm determines the specific parameter settings of the treadmill based on information from the motion context representation. For example, for the user mentioned above, the decision-making agent might generate a treadmill control strategy with an initial speed of 4-5 km / h and an incline of 0%-3%, suggesting each training session last 30-45 minutes, 3-4 times per week. Furthermore, the agent dynamically adjusts the strategy based on the user's real-time dynamic vital signs data (such as heart rate and gait) to ensure a safe and effective training experience throughout the entire exercise process.

[0191] Through this process, the decision-making agent can fully utilize the professional sports knowledge in the knowledge graph, combined with the user's specific characteristics, to generate precise and personalized treadmill control strategies, thereby helping users achieve their fitness goals more scientifically.

[0192] According to some embodiments of this application, before step S105 generates a treadmill control strategy adapted to the target user based on user feature parsing information using a pre-trained decision agent, the method further includes pre-training the decision agent, which may specifically include:

[0193] Step S501: Construct a motion simulation process that matches multiple candidate users; wherein, the motion simulation process is used to simulate the environment of treadmill operation and candidate user movement.

[0194] Step S502: Obtain multiple user feature sample data and the original agent network;

[0195] Step S503: Process user feature sample data through the agent network to obtain the simulation strategy for this round of training;

[0196] Step S504: Interactive simulation is performed in the motion simulation process through the current training simulation strategy to determine the treadmill state simulation data and user state simulation data in the motion simulation process.

[0197] Step S505: Calculate the simulation reward signal based on the treadmill status simulation data and user status data;

[0198] Step S506: Update the agent network based on the simulation reward signal;

[0199] Step S507: Based on the updated agent network, return to the process of processing user feature sample data through the agent network until the simulation strategy of this round of training meets the preset training expectation conditions, and generate a decision agent according to the agent network.

[0200] In some embodiments, step S501 involves constructing a motion simulation process that matches multiple candidate users; wherein the motion simulation process is used to simulate the environment of treadmill operation and the movement process of candidate users.

[0201] It's important to note that building an environment that simulates treadmill operation and user movement is fundamental to training the decision-making agent. This motion simulation process needs to be able to simulate the movement responses of different types of users and various operating states of the treadmill. For example, it can simulate treadmill training at different speeds and inclines for users of different ages, weights, and health conditions, including their heart rate changes, fatigue levels, and movement trajectories. Such a simulation environment provides the agent with a safe and controllable training platform, allowing it to accumulate a wealth of experience in a virtual environment without the need for actual hardware testing.

[0202] In some embodiments, step S502 involves acquiring multiple user feature sample data and the original agent network;

[0203] It should be noted that multiple user feature sample data points are collected, including information such as age, weight, height, gender, health status, and training goals. Additionally, an initial agent network is obtained, which can be a deep neural network, as the initial model for the decision-making agent. For example, feature sample data from 1000 different users can be collected, covering individuals ranging from young to elderly, and from healthy people to those with minor health issues.

[0204] In some embodiments, step S503 involves processing user feature sample data through an agent network to obtain the simulation strategy for this round of training.

[0205] It's important to note that the agent network processes this user characteristic sample data and generates a current-round training simulation strategy. This strategy defines how the agent network adjusts the treadmill parameters based on the user's characteristics in the simulation environment. For example, for a young, healthy user, the strategy might suggest a higher speed and incline; while for an older user, it might suggest a lower speed and incline to ensure safety.

[0206] In some embodiments, step S504 involves interactive simulation during the motion simulation process using the current training simulation strategy, in order to determine treadmill state simulation data and user state simulation data during the motion simulation process.

[0207] It's important to note that in the simulation environment, the agent interacts with the environment based on the generated simulation strategy. In this way, simulated data on treadmill status (such as speed, incline, and time) and simulated user status (such as heart rate, gait, and fatigue level) can be collected. This data reflects the actual effect of the agent's strategy in the simulation environment. For example, during the exercise simulation process, the agent network will try different combinations of speed and incline, observing the simulated user heart rate response and exercise performance.

[0208] In some embodiments, step S505 calculates a simulation reward signal based on treadmill state simulation data and user state data.

[0209] It's important to note that the simulation reward signal is calculated based on the collected treadmill and user state simulation data. This reward signal serves as a standard for evaluating the effectiveness of the agent's strategy and can be derived from the user's exercise results (such as calorie consumption and improvement in cardiopulmonary function) and safety (such as avoiding overexertion or injury). For example, if the user achieves the expected exercise intensity in the simulation and does not exhibit any abnormal physical responses, the agent network will receive a higher reward signal.

[0210] In some embodiments, step S506 involves updating the agent network based on the simulation reward signal.

[0211] It's important to note that the agent network is updated based on simulated reward signals, which can be achieved using reinforcement learning algorithms. This process adjusts the agent network's parameters to strengthen policies that yield high rewards and weaken those that result in low rewards. For example, if a policy causes the user's heart rate to rise too high, the agent network will reduce the weight of that policy; conversely, if a policy effectively helps the user achieve their exercise goals, the agent network will increase the weight of that policy.

[0212] In some embodiments, step S507 involves returning to the process of processing user feature sample data through the agent network based on the updated agent network until the current training simulation strategy meets the preset training expectation conditions, and generating a decision agent based on the agent network.

[0213] It's important to note that this training process is iterative, with the agent network continuously updated and optimized. After each update, the agent network reprocesses the user feature sample data and tests it in a simulation environment. This cycle continues until the agent network's policy meets preset training expectations, such as reaching a certain reward threshold or policy stability. Once these conditions are met, the training process ends, and the final agent network is determined as the decision-making agent, ready for use in actual treadmill control.

[0214] Through this pre-training process, the decision-making agent can generate accurate and safe treadmill control strategies based on user characteristics in practical applications, helping users effectively achieve their fitness goals.

[0215] According to some embodiments of this application, step S104, which involves reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information, may include:

[0216] The natural language instruction statement is uploaded to the target large language model deployed on the cloud server, and the natural language instruction statement is reasoned and parsed by the target large language model to obtain user feature parsing information.

[0217] In step S105, based on the user feature parsing information, a pre-trained decision-making agent generates a treadmill control strategy adapted to the target user, which may include:

[0218] Based on user feature analysis information, a decision-making intelligent agent deployed on a cloud server generates a treadmill control strategy adapted to the target user.

[0219] It's important to note that the natural language instruction statement is uploaded to the target large language model on the cloud server. This step ensures that the model receives a complete statement containing the user's training objectives, health data, and vital sign information. The cloud server provides powerful computing and storage resources, supporting the efficient operation and data processing of large-scale models. After receiving the instruction statement, the target large language model leverages its expertise in the sports and fitness field and its natural language processing capabilities to perform in-depth semantic analysis and feature extraction.

[0220] Furthermore, a decision-making agent on a cloud server is used to generate adaptive treadmill control strategies based on user characteristics. This decision-making agent, running on a cloud server, has access to and can process large amounts of user data and expertise in the field of exercise and fitness.

[0221] The deployment of cloud servers not only provides powerful computing support but also ensures data security and privacy protection. User data is encrypted during transmission and processing to prevent data leakage and unauthorized access. Furthermore, cloud servers can update and optimize models in real time, ensuring that the decision-making agent and large language model always make decisions based on the latest fitness knowledge and user feedback, thereby providing users with more accurate and personalized services. Through the above process, embodiments of this application demonstrate how to leverage the powerful computing capabilities of cloud servers and the intelligent processing of models to achieve efficient conversion from natural language instructions to personalized treadmill control strategies, helping users better achieve their fitness goals.

[0222] According to some embodiments of this application, step S104, which involves reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information, may include:

[0223] The natural language instruction statement is uploaded to the target large language model deployed on the terminal device, and the natural language instruction statement is reasoned and parsed by the target large language model to obtain user feature parsing information.

[0224] In step S105, based on the user feature parsing information, a pre-trained decision-making agent generates a treadmill control strategy adapted to the target user, which may include:

[0225] Based on user feature analysis information, a decision-making intelligent agent deployed on the terminal device generates a treadmill control strategy adapted to the target user.

[0226] It should be noted that in this embodiment, the processing of natural language instructions is performed on the terminal device, which reduces reliance on network connectivity and improves the real-time performance and privacy of data processing. The target large language model is optimized and compressed to operate efficiently within the hardware limitations of the terminal device. For example, through model quantization and pruning techniques, the model size and computational requirements are significantly reduced, enabling it to run on smartphones or dedicated fitness equipment. This local processing method not only speeds up response times but also ensures that the user's health data does not leave the device, thereby enhancing data security.

[0227] Furthermore, based on user feature analysis information, the decision-making agent deployed on the terminal device can generate adaptive treadmill control strategies in real time. The implementation of the decision-making agent on the terminal device has also been optimized to ensure its efficient operation in resource-constrained environments. It utilizes the device's local computing power, combined with the user's real-time dynamic vital signs data (such as heart rate, gait, etc., which may come from connected smart wearable devices), to quickly formulate personalized exercise plans.

[0228] It should be understood that the advantage of performing these steps on the terminal device lies in its ability to provide a more personalized and immediate user experience. Users can obtain real-time exercise guidance and treadmill setting suggestions without waiting for a response from the cloud server. Furthermore, this localized processing method reduces data transmission costs and mitigates security risks that may arise from network latency. In summary, by deploying a target large language model and decision-making agent on the terminal device, the embodiments of this application achieve fast, secure, and efficient personalized treadmill control, providing users with a more intelligent fitness solution.

[0229] According to some embodiments of this application, step S104, which involves reasoning and parsing the natural language instruction statement based on a pre-adapted target large language model to obtain user feature parsing information, may include:

[0230] The natural language instruction statement is uploaded to the target large language model deployed on the treadmill, and the natural language instruction statement is reasoned and parsed by the target large language model to obtain user feature parsing information.

[0231] In step S105, based on the user feature parsing information, a pre-trained decision-making agent generates a treadmill control strategy adapted to the target user, which may include:

[0232] By deploying a decision-making agent on the treadmill, a treadmill control strategy adapted to the target user is generated based on user feature analysis information.

[0233] It's worth noting that the target large language model can also be built into the treadmill. This model has been specifically optimized and adapted for the sports and fitness field. Users can directly input natural language instructions on the treadmill's interface. Upon receiving this instruction, the large language model on the treadmill will immediately perform semantic analysis and feature extraction. Leveraging its expertise in the sports and fitness field, the model accurately parses key information such as the user's age, gender, weight, training goals, and health status, forming user feature analysis information. This localized processing method not only improves data processing speed but also enhances user data privacy protection, as the data does not need to be uploaded to the cloud but is processed directly within the treadmill.

[0234] Furthermore, the treadmill is equipped with a pre-trained decision-making agent that can generate personalized treadmill control strategies in real time based on user characteristic analysis. The deployment of this decision-making agent on the treadmill ensures that it can quickly respond to user needs and dynamically adjust based on the user's real-time dynamic vital signs data (such as heart rate and gait, which may come from the treadmill's built-in sensors).

[0235] It should be understood that deploying the target large language model and decision-making agent directly on the treadmill allows the entire system to operate independently without relying on external network connections. This design not only improves reliability and response speed but also reduces network latency and data transmission costs. Users can use the treadmill anytime, anywhere, to receive instant, personalized exercise guidance and control strategies. In summary, by deploying the target large language model and decision-making agent directly on the treadmill, the embodiments of this application achieve fast, safe, and efficient personalized treadmill control, providing users with a more intelligent fitness solution.

[0236] In some embodiments, step S106 controls the operating state of the treadmill based on a treadmill control strategy.

[0237] It should be noted that the treadmill hardware system and the decision-making agent in this embodiment have an efficient communication mechanism. The treadmill can receive control signals from the agent and convert them into specific hardware operation instructions. For example, the speed adjustment signal sent by the decision-making agent based on the treadmill's control strategy will be received by the treadmill's microcontroller, which will then control the treadmill's motor drive module to achieve precise speed adjustment.

[0238] In some embodiments, a closed-loop control system can be added to achieve precise control of the treadmill's operating status. This closed-loop control system monitors the treadmill's actual operating status in real time using various sensors installed on the treadmill, such as speed and incline sensors, and feeds this data back to the decision-making agent. The agent compares the feedback data with the treadmill's control strategy, calculates the deviation, and adjusts the control signal accordingly to reduce the deviation. For example, if the target speed is 10 km / h and the actual speed is 9.8 km / h, the decision-making agent will calculate a corresponding compensation signal based on the deviation, increasing the motor's drive current to increase the speed until the actual speed matches the target speed. This closed-loop control method effectively addresses performance fluctuations caused by load changes, mechanical wear, and other factors during treadmill operation, ensuring the treadmill always operates stably according to the expected control strategy.

[0239] In some embodiments, the treadmill also incorporates an intelligent shock absorption system that automatically adjusts the hardness of the treadmill's shock-absorbing pads based on the user's weight and exercise intensity, providing a more comfortable exercise environment.

[0240] It is worth noting that the embodiments of this application embody a clever collaborative working approach, forming a closed-loop control circuit that includes data collection, semantic transformation, model parsing, strategy generation, and device control. During operation, the smart wearable device collects the user's dynamic vital signs data in real time and feeds it back to the control system, where it is fused with user feature analysis information. The decision-making agent adjusts the treadmill's control strategy in real time based on the updated user status information and precisely executes the corresponding operational status changes through the treadmill's hardware control system. For example, when the user's real-time heart rate display approaches the preset safety threshold of their maximum heart rate, the agent reacts quickly, reducing the treadmill's speed or incline to ensure the user's heart rate returns to a safe range. Furthermore, the treadmill's display screen or voice system can also provide timely feedback to the user on the current adjustments and subsequent training suggestions, enhancing the user's sense of control and participation in the training process and improving the overall exercise experience.

[0241] According to some embodiments of this application, after controlling the operating state of the treadmill based on the treadmill control strategy in step S106, the following may be included:

[0242] Step S601: Re-collect the target user's dynamic vital signs data through a smart wearable device;

[0243] Step S602: Based on the re-collected dynamic vital sign data, return to the execution of generating instructions based on training purpose information, user health data, static vital sign data, and dynamic vital sign data, so as to continuously update the treadmill control strategy.

[0244] In some embodiments of this application, the process of re-collecting dynamic vital sign data and continuously updating the treadmill control strategy through a smart wearable device after step S106 forms a closed-loop feedback mechanism to ensure that the user always gets the most suitable training experience throughout the exercise process.

[0245] In some embodiments, step S601 involves re-collecting the target user's dynamic vital signs data via a smart wearable device.

[0246] It's important to note that during treadmill operation, the user's wearable devices (such as smartwatches and heart rate monitors) continuously monitor their dynamic vital signs, including heart rate, blood pressure, gait, and calorie consumption. For example, a smartwatch can update heart rate data every second, while a heart rate monitor provides more accurate real-time heart rate monitoring. This dynamic data reflects the user's physical response and exercise status under the current treadmill settings and is a key indicator for evaluating training effectiveness and safety.

[0247] In some embodiments, step S602 involves returning to the execution of generating instructions based on training objective information, user health data, static vital signs data, and dynamic vital signs data, according to the re-collected dynamic vital signs data, in order to continuously update the treadmill control strategy.

[0248] It should be noted that after re-collecting dynamic vital sign data, natural language instructions can be regenerated by combining the original training objectives, user health data, static vital sign data, and the latest dynamic vital sign data. This process is similar to the initial instruction generation steps, but it is now based on existing exercise data, reflecting the real-time changes in the user's exercise state. For example, if the user's real-time heart rate is close to 85% of their maximum heart rate, and the current treadmill setting may be too high, this embodiment will generate a new instruction: "The user's current heart rate is too high; the speed should be reduced to 5 km / h, and the incline reduced to 2% to ensure safety." This instruction will trigger the target large language model to re-parse the user's features, and the decision-making agent will adjust the treadmill's operating state.

[0249] According to some embodiments of this application, step S602, based on the re-collected dynamic vital sign data, returns to the execution of generating instructions based on training purpose information, user health data, static vital sign data, and dynamic vital sign data to continuously update the treadmill control strategy, and may include:

[0250] Step S701: Based on the re-collected dynamic vital signs data, return to the execution of generating instructions based on training objective information, user health data, static vital signs data, and dynamic vital signs data to obtain updated natural language instructions.

[0251] Step S702: Based on the target large language model, perform reasoning and parsing on the updated natural language instruction statement to obtain the updated user feature parsing information;

[0252] Step S703: The decision-making agent updates the treadmill control strategy for the target user based on the updated user feature analysis information and the previous treadmill control strategy.

[0253] Step S704: Control the operating status of the treadmill based on the updated treadmill control strategy.

[0254] It should be noted that after the dynamic vital signs data is re-collected, the updated natural language instructions will be re-inputted into the target large language model for inference and parsing. The target large language model will then re-parse the user's feature information, including their current exercise state and health status, based on the new instructions. For example, the target large language model might determine that the user's heart rate is close to 85% of their maximum heart rate, requiring a reduction in exercise intensity to ensure safety. The parsing result will generate new user feature parsing information, such as: "The user's heart rate is high; the current exercise intensity is too high; it is recommended to reduce speed and incline."

[0255] Furthermore, after receiving updated user feature analysis information, the decision-making agent can combine the previous treadmill control strategy to generate a new control strategy. For example, if the previous strategy was a speed of 6 km / h and an incline of 3%, the new strategy might be adjusted to a speed of 5 km / h and an incline of 2%. The agent will comprehensively consider the user's training goals, health status, and real-time dynamic vital sign data to ensure that the new strategy is both safe and effective. For example, for a user who wants to lose weight but has a high heart rate, the decision-making agent might further reduce the speed and incline while increasing the interval time to ensure that the user can continue training within a safe heart rate range.

[0256] Furthermore, the treadmill can adjust its operating status based on updated control strategies. For example, the treadmill speed might automatically adjust from 6 km / h to 5 km / h, and the incline from 3% to 2%. Simultaneously, the treadmill's display screen or voice system will provide feedback to the user regarding the adjustment and new training suggestions, such as, "Your heart rate is too high; we have reduced the speed and incline. Please maintain your current state and continue training." This real-time feedback and adjustment mechanism ensures that the user remains in an optimal training state throughout the exercise, while avoiding the risks of overexertion.

[0257] According to some embodiments of this application, step S703, by updating the treadmill control strategy for the target user based on the updated user feature parsing information and the previous treadmill control strategy by the decision-making agent, may include:

[0258] Step S801: Based on the updated user feature parsing information, the decision-making agent assesses the current motion state of the target user.

[0259] Step S802: Based on the current exercise state, the updated user feature parsing information, and the treadmill control strategy of the previous round, calculate the change in speed and the change in incline for the current round.

[0260] Step S803: Generate a smoothing control strategy based on the change in speed and the change in incline in this round, so as to update the treadmill control strategy.

[0261] It should be noted that in this embodiment, the decision-making agent first comprehensively assesses the target user's current exercise state based on the updated user feature parsing information. This includes dynamic vital sign data such as the user's real-time heart rate, blood pressure, gait, speed, and incline, as well as static vital sign data such as the user's age, weight, and health status. For example, if the user's real-time heart rate is 130 beats per minute, and their maximum safe heart rate is 150 beats per minute, the agent will determine that the user is currently exercising at a moderate intensity. Simultaneously, the agent will also consider the user's training goals, such as weight loss, muscle gain, or improved cardiopulmonary function, as well as the user's health status, such as a history of heart disease or joint problems. For example, for a user who wants to lose weight and has mild knee pain, the agent will pay particular attention to their heart rate and gait to ensure that the exercise intensity is moderate and does not put excessive pressure on the knees.

[0262] After assessing the user's current exercise status, the decision-making agent combines updated user feature analysis information with the previous treadmill control strategy to calculate the changes in speed and incline for the current round. This step requires the agent to consider multiple factors, including the user's training goals, health condition, current exercise intensity, and the effectiveness of the previous strategy. For example, if the user's heart rate is slightly above the target range, the agent might calculate a small speed reduction (e.g., -0.5 km / h) and a small incline reduction (e.g., -1%) to gently reduce exercise intensity. Conversely, if the user's heart rate is below the target range and there are no signs of fatigue, the agent might calculate an appropriate speed increase (e.g., +0.5 km / h) and an appropriate incline increase (e.g., +1%) to improve exercise effectiveness.

[0263] Finally, the decision-making agent generates a smooth adjustment strategy based on the calculated changes in speed and incline to update the treadmill's operating status. The purpose of this smooth adjustment strategy is to ensure that changes in speed and incline are not too drastic, thus avoiding discomfort or increased risk of injury for the user. For example, if the speed change is -0.5 km / h and the incline change is -1%, the agent will generate a gradual adjustment strategy to smoothly transition the speed and incline to the new settings over a certain period. Specifically, the agent might set a transition time of 30 seconds, reducing the speed by 0.1 km / h and the incline by 0.2% every 5 seconds until the target speed and incline are reached. This smooth adjustment not only improves user comfort but also helps the user better adapt to new exercise intensity, thereby improving training effectiveness.

[0264] Through the above process, the decision-making agent can dynamically adjust the treadmill's speed and incline based on the user's real-time movement status and characteristics, ensuring that the user remains in optimal training condition throughout the exercise. This dynamic adjustment mechanism not only improves the safety and effectiveness of training but also provides users with a personalized exercise experience, helping them achieve their fitness goals more scientifically.

[0265] It is worth noting that the closed-loop feedback mechanism ensures that the treadmill's control strategy can be dynamically adjusted based on the user's real-time physical condition. For example, in a training session, the user's initial settings might be a speed of 6 km / h and an incline of 3%. As the training progresses, the user's heart rate gradually increases. When the heart rate reaches a preset safe upper limit, the smart wearable device transmits data to the treadmill, which then reassesses and generates a new control strategy, reducing the speed and incline to ensure user safety. Furthermore, embodiments of this application may explain the reasons for the adjustment to the user through the treadmill's display screen or voice system and provide subsequent training suggestions. This real-time feedback and adjustment mechanism not only improves training safety but also helps users achieve their fitness goals more scientifically and optimize overall exercise results.

[0266] The treadmill according to embodiments of this application is controlled by any of the treadmill control methods described in the embodiments of this application. By integrating a large language model and a decision-making intelligent agent, combined with real-time data monitoring and feedback mechanisms, the treadmill of this application provides users with a personalized, intelligent, safe, and efficient exercise solution. It not only helps users conduct fitness training more scientifically but also enhances the comfort and enjoyment of exercise, while simultaneously promoting users' long-term health management and the achievement of their exercise goals.

[0267] Reference Figure 2 , Figure 2 This illustration shows the hardware structure of an electronic device according to another embodiment. The electronic device may include:

[0268] The processor 21 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, 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 this application.

[0269] The memory 22 can be implemented in the form of read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 22 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 22 and is called and executed by the processor 21 to execute the treadmill control method of the embodiments of this application.

[0270] Input / output interface 23 is used to implement information input and output;

[0271] The communication interface 24 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0272] Bus 25 transmits information between various components of the device (e.g., processor 21, memory 22, input / output interface 23, and communication interface 24);

[0273] The processor 21, memory 22, input / output interface 23 and communication interface 24 are connected to each other within the device via bus 25.

[0274] This application also provides a computer program product, which includes a computer program. The processor of a computer device reads and executes the computer program, causing the computer device to perform the treadmill control method described above.

[0275] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “including,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0276] It should be understood that in this disclosure, "at least one item" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer 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 represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0277] The technical solutions disclosed herein, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0278] It should also be understood that the various implementation methods provided in this application can be combined arbitrarily to achieve different technical effects.

[0279] The above is a detailed description of the embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.

Claims

1. A treadmill control method, characterized in that, The method, applied to controlling a treadmill equipped with a smart wearable device, includes: Acquire training objective information, user health data, and static vital sign data of the target users; The target user's dynamic vital signs data are collected through the smart wearable device; Key information is extracted from the training objective information, the user health data, the static vital signs data, and the dynamic vital signs data to obtain key motion indication information corresponding to the target user; Natural language is generated based on the key motion instruction information to obtain intermediate instruction statements in natural language form, and the intermediate instruction statements are then checked. In response to the intermediate instruction statement passing the statement check, the intermediate instruction statement is determined to be a natural language instruction statement; The natural language instruction statement is reasoned and parsed based on the pre-adapted target large language model to obtain user feature parsing information; Based on the user feature parsing information, a pre-trained decision-making agent generates a treadmill control strategy adapted to the target user. The treadmill's operating status is controlled based on the aforementioned treadmill control strategy; The dynamic vital signs data of the target user are re-collected through the smart wearable device; Based on the re-collected dynamic vital sign data, the process returns to generate an instruction statement based on the training objective information, the user health data, the static vital sign data, and the dynamic vital sign data, resulting in an updated natural language instruction statement. Based on the analysis of the updated natural language instruction statement through the target large language model, the updated user feature analysis information is obtained. Based on the updated user feature parsing information and the previous treadmill control strategy, the decision-making agent updates the treadmill control strategy for the target user. The treadmill's operating status is controlled based on the updated treadmill control strategy.

2. The method according to claim 1, characterized in that, Before performing reasoning and parsing on the natural language instruction statement based on the pre-adapted target large language model to obtain user feature parsing information, the method further includes pre-adapting and adjusting the target large language model, specifically including: Acquire knowledge data in the sports domain, sample data for training objectives, user sample data, and the original large-scale natural language model; Based on the sports domain knowledge data, the training objective sample data, and the user sample data, instruction texts are generated to obtain multiple instruction text sample data. The indicator sample data is labeled to obtain indicator label information corresponding to each indicator sample data; Based on multiple sample data of the indicated terms and the indicated term annotation information corresponding to each sample data of the indicated terms, an adaptive adjustment dataset is constructed. The large natural language model is trained using the adaptive dataset. In response to the natural language large model meeting the preset adjustment expectation conditions during adaptive training, the natural language large model is identified as the target large language model.

3. The method according to claim 1, characterized in that, Before the pre-trained decision-making agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, the process further includes pre-training the decision-making agent, specifically including: A motion simulation process is constructed to match multiple candidate users; wherein, the motion simulation process is used to simulate the environment of treadmill operation and the exercise process of the candidate users; Acquire multiple user feature sample data and the original agent network; The user feature sample data is processed by the intelligent agent network to obtain the simulation strategy for this round of training; The training simulation strategy described above is used to conduct interactive simulation in the motion simulation process in order to determine the treadmill state simulation data and the user state simulation data in the motion simulation process. The simulation reward signal is calculated based on the treadmill status simulation data and the user status data; The agent network is updated based on the simulated reward signal; Based on the updated agent network, the process returns to processing the user feature sample data through the agent network until the current training simulation strategy meets the preset training expectation conditions, and the decision agent is generated according to the agent network.

4. The method according to claim 1, characterized in that, The pre-trained decision-making agent generates a treadmill control strategy adapted to the target user based on the user feature parsing information, including: The decision-making agent retrieves knowledge entity information related to the user feature parsing information from a pre-built knowledge graph database. The user feature parsing information and the knowledge entity information are fused together to obtain motion context representation information corresponding to the target user; A control strategy is generated based on the motion context representation information to obtain a treadmill control strategy adapted to the target user.

5. The method according to claim 1, characterized in that, The step of updating the treadmill control strategy for the target user based on the updated user feature parsing information and the previous treadmill control strategy by the decision-making intelligent agent includes: The decision-making agent evaluates the current motion state of the target user based on the updated user feature parsing information. Based on the current exercise state, the updated user feature parsing information, and the treadmill control strategy of the previous round, calculate the speed change and the incline change for this round. A smoothing control strategy is generated based on the changes in speed and slope in the current cycle to update the treadmill control strategy.

6. A treadmill, characterized in that, The treadmill is used to be controlled by the treadmill control method according to any one of claims 1 to 5.

7. An electronic device, characterized in that, include: The device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the treadmill control method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the treadmill control method as described in any one of claims 1 to 5.

Images