Multi-device collaborative ai personalized motion planning generation system

Abstract

Multi-device cooperative AI personalized exercise planning generation system. The present application belongs to the technical field of intelligent fitness and health management, and specifically relates to a multi-device cooperative AI personalized exercise planning generation method, which comprises the following steps: S1: obtaining a user basic parameter set and a current dynamic physiological parameter set; S2: performing preprocessing and feature extraction based on the basic parameter set and the current dynamic physiological parameter set, generating a multi-dimensional derived feature vector and a standardized user vector through a feature engineering module; S3: performing feature fusion using the standardized user vector and the multi-dimensional derived feature vector to generate a multi-dimensional user state portrait feature sequence; and S4: performing parallel analysis and decision on the multi-dimensional user state portrait feature sequence using a multi-source hybrid decision model to generate an exercise type preference vector. The present application can realize highly personalized, scientific and safe dynamic exercise planning which can be continuously executed across devices by fusing multi-source heterogeneous data, performing parallel multi-source hybrid decision, dynamically weighted fusion and exercise science knowledge graph mapping.

Description

Technical Field

[0001] This invention belongs to the field of intelligent fitness and health management technology, specifically involving an AI-powered personalized exercise planning and generation system that enables multi-device collaboration. Background Technology

[0002] With the rapid development of IoT and AI technologies, smart fitness equipment (such as smart treadmills, connected exercise bikes, and smart strength training machines) and wearable sensors (such as smart bracelets, heart rate monitors, and electromyography (EMG) armbands) are becoming increasingly popular. Utilizing these devices to provide users with automated and personalized exercise guidance has become a research hotspot. However, existing technological solutions have significant limitations: Most systems rely solely on simple user questionnaires (static) or a single machine learning model for recommendations, lacking a hybrid decision-making mechanism that organically combines expert knowledge, collective wisdom, and individual data models. This may result in shortcomings in the security, universality, or depth of personalization of the recommendation scheme. The generated plans are often a list of discrete sports events, lacking structured and parameterized generation based on deep sports science knowledge (such as physiological adaptation principles, movement correlation, and energy systems). The rationality and feasibility of the plans are highly dependent on the model training data, resulting in poor interpretability. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-powered personalized motion planning generation system that enables multi-device collaboration. This system can achieve highly personalized, scientifically safe, and continuously executable dynamic motion planning across devices by integrating multi-source heterogeneous data, parallel multi-source hybrid decision-making, dynamic weighted fusion, and motion science knowledge graph mapping.

[0004] The specific technical solution adopted by this invention is as follows: A multi-device collaborative AI-powered personalized motion planning generation method includes the following steps: S1: Obtain the user's basic parameter set and current dynamic physiological parameter set; S2: Based on the aforementioned basic parameter set and the current dynamic physiological parameter set, preprocessing and feature extraction are performed, and multi-dimensional derived feature vectors and standardized user vectors are generated through the feature engineering module; S3: Use the standardized user vector and multi-dimensional derived feature vector to perform feature fusion to generate a multi-dimensional user state profile feature sequence; S4: Use a multi-source hybrid decision model to perform parallel analysis and decision-making on the multi-dimensional user state profile feature sequence to generate a motion type preference vector, a collaborative filtering recommendation result vector, and a machine learning prediction result vector. S5: Through the weighted fusion module, configurable weight coefficients are assigned to the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector, and weighted fusion calculation is performed to generate a motion mode suitability score vector. S6: Based on the highest score in the scoring vector, the system maps and instantiates a specific, executable, and parameterized personalized motion planning scheme from the knowledge graph.

[0005] S7: When it is detected that the user switches from the first smart sports device to the second smart sports device, the user's training history data on the first smart sports device, the dynamic user state model, and the device capability description information of the second smart sports device are obtained. The AI ​​biological model then performs intelligent conversion and mapping across devices to generate a real-time motion plan adapted to the second intelligent motion device.

[0006] In step S2, the multi-dimensional derived feature vector includes: Body mass index is calculated as the ratio of weight to the square of height. Basal metabolic rate is estimated based on a preset formula that includes variables such as gender, age, height, and weight. Based on the range of body mass index values, body type classification labels are generated with reference to preset classification standards; Based on the user's age range, an age lifecycle label is generated with reference to a preset physiological stage classification standard.

[0007] Step S4, which utilizes a multi-source hybrid decision-making model for parallel analysis and decision-making, specifically includes the following parallel execution process: The multidimensional user state profile feature sequence matches the feature sequence with a predefined structured expert rule base. Each rule contains a feature condition expression connected by logical operators and a corresponding rule-action weight vector. Calculate the degree to which the feature sequence satisfies the conditions of each rule. For rules whose satisfaction exceeds the threshold, activate their weight vectors and scale them according to the satisfaction to obtain the contribution vector of the rule. The contribution vectors of all activated rules are weighted and summed to generate a preliminary motion type preference vector; In the historical user-feature-behavior database, the similarity between the current user's feature sequence and the historical user's feature profile is calculated, and the K most similar users are selected to form a neighborhood. Analyze the long-term selection preferences, completion rates, and positive feedback data of this similar user community for various exercise modes; Based on neighborhood behavior data, the group preference score for each exercise mode is calculated, and after normalization, a collaborative filtering recommendation result vector is formed. The multidimensional user status profile feature sequence is input into a pre-trained multi-task learning model or multi-class prediction model, which has been trained on massive labeled data. The model directly outputs the predicted probability or potential benefit score for each preset exercise type that the current user is suited for; Normalize this probability distribution or rating sequence and output it as a vector of machine learning prediction results.

[0008] The weighted fusion module in step S5 has the following weight coefficient configuration strategy: Assign basic weight coefficients α, β, and γ to the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector, respectively, and α+β+γ=1; The weighting coefficients can be dynamically adjusted based on user identity attributes or usage stage. For new users, α > β > γ is set to prioritize security; for experienced users, β and γ weights are increased to enhance personalization and exploration. The fusion calculation is performed according to the formula V_final=α*V_rule+β*V_cf+γ*V_ml to generate the final motion mode fitness score vector.

[0009] The weighted fusion module in step S5 performs the following fusion calculation operation: it receives the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector generated by the multi-source hybrid decision model as input; according to the configured weight coefficients, it performs linear weighted summation calculation on the three input vectors to integrate the comprehensive decision information of rule reasoning, group preference and model prediction, and outputs a unified motion mode fitness score vector.

[0010] The sports science knowledge graph in step S6 is a structured semantic network containing multiple relationships, which includes at least movement patterns, specific events, motion parameters, target muscle groups, energy systems, and applicable equipment as entity nodes. The mapping and instantiation process is as follows: First, the recommended movement mode is determined based on the movement mode fitness score vector; Subsequently, in the sports science knowledge graph, along the sports pattern-inclusion-specific item relationship path, all specific sports item entities belonging to the pattern are retrieved; Next, combining the user's personalized parameters and the constraints of currently available equipment, specific intensity, number of sets, number of repetitions, and rest interval parameters are inferred and instantiated from the motion parameter attributes associated with the specific sports entity; finally, a complete, parameterized personalized exercise planning scheme is synthesized.

[0011] A multi-device collaborative AI-powered personalized motion planning and generation system includes: The parameter acquisition module is used to acquire the user's basic parameter set and current dynamic physiological parameter set; The feature engineering module is used to generate multi-dimensional derived feature vectors and standardized user vectors. The feature fusion module is used to generate a multi-dimensional user status profile feature sequence; The multi-source decision module is used to perform rule reasoning, collaborative filtering, and machine learning prediction in parallel, and output the corresponding vectors. The weighted fusion module is used to assign weights and perform fusion calculations on various types of vectors; The knowledge graph module is used to store motion-related entities and relationships, and supports solution mapping and instantiation; The device collaboration module is used to handle motion planning conversion and adaptation during device switching.

[0012] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned multi-device collaborative AI personalized motion planning generation method.

[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned multi-device collaborative AI personalized motion planning generation method.

[0014] The technical effects achieved by this invention are as follows: This invention generates a comprehensive, dynamic, multi-dimensional user profile by integrating basic static parameters of the user with dynamic physiological parameters from multiple devices. Combined with a sports science knowledge graph, it can generate personalized solutions that not only recommend exercise types but also detail specific movement parameters, offering a high degree of personalization.

[0015] This invention employs a parallel multi-source hybrid decision-making model, simultaneously fusing rule-based reasoning based on prior safety and performance knowledge, collaborative filtering based on verification of similar user groups, and data-driven machine learning prediction. Through a dynamically configurable weight fusion mechanism, the decision-making focus can be flexibly adjusted according to user identity type and usage stage, ensuring both safe onboarding for new users and meeting the personalized and advanced exploration needs of experienced users, resulting in more scientific and reasonable recommendation results.

[0016] This invention, through a unique device collaboration module and AI biological model, can intelligently map and generate equivalent training plans based on the physiological stimulus vectors of completed training, the user's real-time status, and the capabilities of the target device when a user switches between different smart fitness devices. This solves the problem of fragmented training between devices in existing technologies, ensuring the continuity and consistency of long-term training plans, and improving training effectiveness and user experience.

[0017] This invention can periodically update the user's status profile based on the user's latest physiological data and training results, and re-make decisions and generate plans, enabling exercise planning to adaptively adjust as the user's status evolves. This helps users break through training plateaus and achieve continuous progress. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a block diagram of the electronic device of the present invention. Detailed Implementation

[0019] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0020] like Figure 1 As shown, the AI-based personalized motion planning generation method for multi-device collaboration includes the following steps: S1: Obtain the user's basic parameter set and dynamic physiological parameter set; the basic parameter set includes at least gender, age, height, and weight; the dynamic physiological parameter set is collected in real time through at least one wearable smart device or biosensor used by the user, including heart rate, blood pressure, blood oxygen saturation, pulse rate, and pulse pressure difference; S2: Preprocessing and feature extraction based on the basic parameter set and the current dynamic physiological parameter set; Normalization is performed based on age, height, and weight to eliminate the influence of dimensions, and all continuous parameter values ​​are mapped to the [0, 1] interval to form a standardized vector. Body Mass Index (BMI) is calculated as the ratio of weight to the square of height. Based on the user's gender, age, height, and weight, the basal metabolic rate (BMR) is calculated using the Mifflin-StJeor equation. For males: BMR = 10 * W + 6.25 * H - 5 * A + 5; For women: BMR = 10 * W + 6.25 * H - 5 * A - 161; Where W is weight (kg), H is height (cm), and A is age (years), the calculated basal metabolic rate characterizes the user's static energy expenditure level; Based on the range of body mass index values, body type classification labels are generated with reference to preset classification standards. According to the mapping of body mass index values, BMI < 18.5 is "underweight" (code 0), 18.5 ≤ BMI < 24 is "normal" (1), 24 ≤ BMI < 28 is "overweight" (2), and BMI ≥ 28 is "obese" (3). Based on the numerical range of the user's age, age life cycle labels are generated with reference to the preset physiological stage division standard. According to the age, the labels are generated as follows: age <18 is "teenager" (0), age 18≤age <45 is "youth" (1), age 45≤age <65 is "middle-aged" (2), and age ≥65 is "elderly" (3). These derived features are also normalized to form a multi-dimensional derived feature vector. S3: Use standardized user vectors and multi-dimensional derived feature vectors to perform feature fusion and generate a multi-dimensional user state profile feature sequence, which is a mathematical representation of the user's current comprehensive state. S4: Utilize a multi-source hybrid decision model to perform parallel analysis and decision-making on the feature sequences of multi-dimensional user state profiles, generating motion type preference vectors, collaborative filtering recommendation result vectors, and machine learning prediction result vectors; specifically: Motion type preference vector: Parallel analysis and decision-making using a multi-source hybrid decision model specifically includes the following parallel execution processes: The multidimensional user state profile feature sequence matches the feature sequence with a predefined structured expert rule base. Each rule contains a feature condition expression connected by logical operators and a corresponding rule-action weight vector. The degree to which the feature sequence satisfies the conditions of each rule is calculated. For rules whose satisfaction exceeds the threshold, their weight vectors are activated and scaled according to the satisfaction level to obtain the contribution vector of the rule. This vector quantifies the strength of type recommendation based on prior knowledge such as safety and efficiency. The contribution vectors of all activated rules are weighted and summed to generate a preliminary motion type preference vector; Collaborative filtering recommendation result vector: In the historical user-feature-behavior database, the similarity between the current user's feature sequence and the historical user's feature profile is calculated, and the K most similar users are selected to form a neighborhood; Analyze the long-term selection preferences, completion rates, and positive feedback data of this similar user community for various exercise modes; Based on neighborhood behavior data, the group preference score for each exercise mode is calculated and normalized to form a collaborative filtering recommendation result vector. This vector reflects the exercise mode preference that has been verified to be effective for groups similar to the current user's physiological state and basic attributes. Machine learning prediction result vector: Input the multi-dimensional user status profile feature sequence into a pre-trained multi-task learning model or multi-class prediction model, which has been trained on massive labeled data; The model directly outputs the predicted probability or potential benefit score for each preset exercise type that the current user is suited for; Normalize this probability distribution or rating sequence and output it as a vector of machine learning prediction results, which represents the direct prediction of the data-driven model's fit to an individual user. The above three decision paths are executed in parallel within the system without blocking each other, and finally output a vector with the same three dimensions synchronously. S5: The weighted fusion module performs the following fusion calculation operations: It receives the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector generated by the multi-source hybrid decision model as input; according to the configured weight coefficients, it performs linear weighted summation on the three input vectors to integrate the comprehensive decision information of rule reasoning, group preference and model prediction, and outputs a unified motion mode fitness score vector. The weighting coefficient configuration strategy adopted by the weighted fusion module includes: Assign basic weight coefficients α, β, and γ to the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector, respectively, and satisfy α+β+γ=1; The weighting coefficients can be dynamically adjusted based on user identity attributes or usage stage. For new users, α > β > γ is set to prioritize security; for experienced users, the weights of β and γ are increased to enhance personalization and exploration. Based on the assigned weight coefficients, a fusion calculation is performed to generate the final motion pattern fitness score vector; The calculation formula is: V_final=α*V_rule+β*V_cf+γ*V_ml, where V_rule is the motion type preference vector, V_cf is the collaborative filtering recommendation result vector, and V_ml is the machine learning prediction result vector; S6: Based on the highest score in the scoring vector, the system maps and instantiates a specific, executable, and parameterized personalized motion planning scheme from the knowledge graph; The sports science knowledge graph is a structured semantic network containing multiple relationships, which includes at least movement patterns, specific events, motion parameters, target muscle groups, energy systems, and applicable equipment as entity nodes; The mapping and instantiation process specifically includes: First, determining the recommended movement mode based on the movement mode fitness score vector; Subsequently, in the sports science knowledge graph, along the relationship path of sports mode-containing-specific items, all specific sports items belonging to the mode are retrieved; Next, combining the user's personalized parameters and the constraints of currently available equipment, specific intensity, number of sets, number of repetitions, and rest interval parameters are inferred and instantiated from the motion parameter attributes associated with the specific sports entity; finally, a complete parameterized personalized exercise planning scheme is synthesized. S7: When a user switches from the first smart sports device to the second smart sports device, the system acquires the training history data, dynamic user state model, and device capability description information of the second smart sports device. The device capability description information is stored in a structured format, including device type, supported sports modes, resistance / load range, monitorable sports parameters, and unique device identification. Based on historical training data, dynamic user state models, and device capability descriptions, an AI biological model generates a real-time motion plan adapted to a second intelligent motion device and coordinated with the previous training phase. During the generation of the real-time motion plan, the AI ​​biological model performs intelligent conversion and mapping functions. Specifically, based on the incomplete parts of the first real-time motion plan, the remaining progress of the preset training goal, and the specific capabilities of the second intelligent motion device, the original plan's exercise load and exercise mode are converted into equivalent values ​​in physiological and mechanical terms. This process ensures that the training load and physiological stimulation remain consistent across devices and maintain continuity with the long-term training plan in terms of progress and intensity, thereby achieving cross-device collaborative training. Specifically, the intelligent conversion and mapping function is performed in the following steps: Training stimulus vectorization involves parsing and extracting features from historical training data from the first intelligent motion device to generate a standardized training stimulus vector. This vector is generated by multi-dimensional feature extraction and fusion of the user's historical training data on the first intelligent motion device. It represents the core physiological and mechanical stimuli of the completed training, including at least: dominant physiological pattern, quantified metabolic load, quantified mechanical load, encoding of major activated muscle groups, and cumulative fatigue information. First, determine the dominant physiological pattern of the previous training stage, including but not limited to aerobic endurance, strength, hypertrophy, or mixed types. Secondly, load quantization calculations are performed: For equipment primarily used for aerobic training, calculate its total training impulse. This calculation is not based on a simple summation of time, but rather on a weighted integral based on the relationship between real-time heart rate and individual resting heart rate and maximum heart rate. Its formula can be simplified as: Total training impulse = Σ (exercise duration × intensity coefficient × heart rate ratio), where the intensity coefficient is non-linearly determined based on the user's heart rate reserve range. For equipment primarily used for strength training, calculate its total mechanical work and metabolic equivalent; the total mechanical work is Σ (load weight × displacement distance × number of repetitions); at the same time, combine parameters such as movement speed to estimate its metabolic consumption and neuromuscular fatigue. Finally, the training in the previous stage is quantitatively represented as a multidimensional stimulus vector, whose data structure can be expressed as: {mode type, total TRIMP, total mechanical work, main activated muscle groups, neural fatigue index, remaining uncompleted percentage}. Target device capability analysis; read the device capability description information of the second intelligent motion device, and analyze its set of available motion modes, supported resistance or load range, and monitorable motion parameters, etc. The intelligent mapping decision, based on the following priority rules, maps multidimensional stimulus vectors to the capability space of the second intelligent motion device: Rule 1: Pattern-priority mapping prioritizes finding and mapping movement patterns that are the same as or similar to the dominant physiological patterns in historical training, which can be provided by the second intelligent movement device. Rule 2: Cross-mode equivalent conversion; When a second intelligent motion device cannot provide the same or similar physiological mode, "cross-mode equivalent conversion" is performed; This rule is used to handle scenarios with significant differences in device capabilities; Continuous calibration and planning generation; the preliminary plan generated by mapping needs to be calibrated by the dynamic user state model; the dynamic user state model evaluates the user's current real-time fatigue level based on the data monitored in real time in step S2, and checks whether the initial load of the new planning plan matches; the plan is fine-tuned according to the calibration results to ensure that the user can execute it safely and effectively. Ultimately, the generated real-time motion planning information is presented to the user through at least one of the following methods: speech synthesis, device display, or haptic feedback from wearable devices.

[0021] like Figure 2 As shown, a multi-device collaborative AI personalized motion planning and generation system, used to implement the aforementioned method, includes: The parameter acquisition module is used to acquire the user's basic parameter set and current dynamic physiological parameter set; The feature engineering module is used to generate multi-dimensional derived feature vectors and standardized user vectors. The feature fusion module is used to generate a multi-dimensional user status profile feature sequence; The multi-source decision module is used to perform rule reasoning, collaborative filtering, and machine learning prediction in parallel, and output the corresponding vectors. The weighted fusion module is used to assign weights and perform fusion calculations on various types of vectors; The knowledge graph module is used to store motion-related entities and relationships, and supports solution mapping and instantiation; The device collaboration module is used to handle motion planning conversion and adaptation during device switching.

[0022] like Figure 3 As shown, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method. Meanwhile, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the above-described method.

[0023] Example 1: New User Initialization and First Personalization Scheme Generation Scenario: New user A registers for the first time and uses a smart fitness system that integrates a smart treadmill, a smart strength training rack, and a smart bracelet.

[0024] Implementation process: Basic parameter collection: User A inputs the following information through the APP: gender male, age 35, height 175cm, weight 85kg.

[0025] Dynamic parameter acquisition: User A wears a smart bracelet, and the system obtains in real time that his resting heart rate is 72 beats / min and his blood pressure is 128 / 85 mmHg.

[0026] Feature engineering and portrait generation: The system normalizes height, weight, and age.

[0027] The calculated BMI is approximately 27.8, which falls into the "overweight" category (label 2). BMR is approximately 1806 kcal / day; Based on the age of 35, the age lifecycle label is generated as "youth" (label 1).

[0028] By integrating all the above features (standardized basic parameters, BMI, BMR, body type label, and age label), a multidimensional user status profile feature sequence for user A is generated.

[0029] Multi-source hybrid decision-making: Rule-based reasoning: A rule in the system's rule base is activated: "IF body type classification == 'overweight' AND age label == 'youth' THEN recommend increasing the weights of 'moderate intensity aerobic exercise' and 'full-body resistance training'." This rule outputs a preference vector for exercise types that emphasize safety and basal metabolic rate enhancement; Collaborative filtering: Find K user groups with similar characteristics to user A (overweight, youth) in the historical database. It is found that this group has the highest completion rate and positive review rate for "treadmill interval walking and running" and "circuit strength training". Generate collaborative filtering recommendation result vector. Machine learning prediction: User profile features are input into a pre-trained model. The model predicts that user A has a high potential benefit score for training in the categories of "aerobic endurance" and "hypertrophy". The output is a vector of machine learning prediction results. Weighted fusion: Since user A is a new user, the system uses a weight configuration of α > β > γ (e.g., α = 0.5, β = 0.3, γ = 0.2) to prioritize safety. After fusion, the "aerobic-resistance hybrid mode" scores the highest in the final scoring vector.

[0030] Knowledge graph instantiation: The system locks the "Aerobic-Resistance Hybrid Mode". A query of the knowledge graph reveals that this mode includes sub-items such as "Treadmill Aerobic Zone Running" and "Weighted Equipment Circuit Training".

[0031] Based on User A's BMR and heart rate data, and the available equipment (treadmill, power rack), the following specific plans are instantiated: "Plan A: 5-minute warm-up jog on the treadmill (1% incline, 6km / h speed); Plan B: Circuit training (seated row - 15kg*12 reps, leg press - 40kg*15 reps, plank for 30 seconds), 3 sets in total, with 45 seconds of rest between sets; Plan C: 5-minute cool-down walk on the treadmill (5.5km / h speed). Output: This solution guides user A to execute the procedure via the treadmill display screen and wristband vibration prompts.

[0032] Example 2: Cross-device training collaboration and real-time planning conversion Scenario: Experienced user B originally planned to complete a 45-minute high-intensity interval training (HIIT) session on a smart exercise bike at the gym, but after 20 minutes, due to equipment malfunction, he needed to switch to a smart rowing machine to continue training.

[0033] Implementation process: Triggering and Information Acquisition: The system detects that user B has disconnected from the bicycle and reconnected to the rowing machine, and immediately acquires: Historical data: Completed 20-minute HIIT cycling data (heart rate, power, cadence); Dynamic status: Current real-time heart rate (high), estimated fatigue level; Equipment capabilities: Rowing machine description file (supports endurance and interval modes, adjustable resistance, and monitors paddle frequency and power); AI biological model conversion mapping: Stimulus vectorization: The completed training was analyzed and its dominant mode was determined to be "aerobic high-intensity interval training"; the total training impulse (TRIMP) was calculated to be 35 (hypothetical value); the main activated muscle group was the lower limb; a stimulus vector was generated {mode: aerobic HIIT, TRIMP: 35, main muscle group: lower limb, fatigue index: medium to high, remaining progress: 55%}; Equipment Analysis: The rowing machine offers "aerobic endurance" and "aerobic interval" modes; Intelligent mapping: Apply rule 1 (mode priority) and select "aerobic interval" mode as the mapping target; Equivalent Conversion: The AI ​​model converts the remaining HIIT load on the bicycle (based on heart rate and power goals) into an equivalent program on the rowing machine: "Using interval mode, perform 8 rounds: each round (high resistance stroke rate 28 spm, lasting 1 minute; low resistance stroke rate 22 spm, recovery 45 seconds), total duration approximately the remaining planned duration;" Continuous calibration: Based on user B's current high real-time heart rate, the model will fine-tune the first round of high resistance intensity by 5% to ensure a safe start; Output and Execution: The new rowing machine interval training program is immediately displayed on the rowing machine screen, and the user is prompted by voice to start; the training load and physiological stimulation goals are consistent with the original cycling program, achieving seamless coordination.

[0034] Example 3: Long-term User Status Evolution and Dynamic Replanning of the Solution Scenario: User C uses this system for a 3-month weight loss and health improvement training program. The system periodically (e.g., every two weeks) reassesses and adjusts the plan based on the latest data.

[0035] Implementation process: Periodic data updates: After 3 months, User C's weight dropped from the initial 95kg to 82kg; the latest physical examination data also showed improvement in blood pressure; the system collected new sets of basic and dynamic physiological parameters; Feature recalculation and profile update: BMI decreased from 33.2 (obese) to 26.7 (overweight), and body type classification labels were updated from 3 to 2; BMR was recalculated due to weight loss; As my birthday approaches, the age lifecycle label has been updated from "youth" to "middle age"; Generate entirely new, multi-dimensional user profiles that reflect their improved state; Decision-making and integration strategy adjustment: rule reasoning: the new profile activates different rules, such as "IF body type classification == 'overweight' AND age label == 'middle-aged' THEN add consideration of 'joint flexibility training' and 'core stability training' in the recommendation". Collaborative filtering: When the neighborhood of similar users changes, more attention is paid to the behavior of users whose BMI has dropped to overweight or who are in middle age; Machine learning prediction: Based on the new data, the model predicts that user C has increased potential benefits from "strength endurance" and "high-intensity functional training"; Weight Adjustment: Given that User C has become a senior user, the system automatically adjusts the fusion weights to β≈γ>α (e.g., α=0.2, β=0.4, γ=0.4), placing greater emphasis on personalization (collaborative filtering) and exploration (model prediction); after fusion, the score of "functional strength and metabolic regulation complex pattern" jumps; Solution upgrade instantiation: Based on the new model, the system retrieves and instantiates more complex solutions from the knowledge graph, such as: "Dynamic warm-up (bear crawl, lunge walk) → superset training (kettlebell swing + explosive push-ups) → aerobic circuit (30 seconds of battle rope + 30 seconds of rowing machine sprint) → flexibility and balance training (yoga ball exercise)"; the movement parameters (such as kettlebell weight, number of sets) are all reset based on user C's current strength level (estimated from historical training data); Results: Through continuous status tracking and dynamic replanning, the system ensures that the exercise program is always highly matched with the user's current actual ability, physiological stage, and health goals, avoiding plateaus and promoting continuous progress.

[0036] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A multi-device collaborative AI-based personalized motion planning generation method, characterized in that, Includes the following steps: S1: Collect user's basic parameter set and current dynamic physiological parameter set; S2: Preprocess and extract features from the basic parameter set and the current dynamic physiological parameter set, and generate multi-dimensional derived feature vectors and standardized user vectors through the feature engineering module; S3: Perform feature fusion on the standardized user vector and the multi-dimensional derived feature vector to generate a multi-dimensional user state profile feature sequence; S4: A multi-source hybrid decision-making model is used to perform parallel analysis and decision-making on the multi-dimensional user state profile feature sequence to generate a motion type preference vector, a collaborative filtering recommendation result vector, and a machine learning prediction result vector. S5: Through the weighted fusion module, configurable weight coefficients are assigned to the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector, and weighted fusion operation is performed to generate a motion mode suitability score vector. S6: Based on the highest score in the scoring vector, map and instantiate a specific, executable, and parameterized personalized motion planning scheme from the knowledge graph.

2. The method according to claim 1, characterized in that: In step S2, the multi-dimensional derived feature vector includes: Body mass index (BMI) is calculated as the ratio of weight to the square of height. Basal metabolic rate is estimated based on a preset formula that includes variables such as gender, age, height, and weight. Based on the range of body mass index values, body type classification labels are generated with reference to preset classification standards; Based on the user's age range, an age lifecycle label is generated with reference to a preset physiological stage classification standard.

3. The method according to claim 1, characterized in that: Step S4, which utilizes a multi-source hybrid decision-making model for parallel analysis and decision-making, specifically includes the following parallel execution process: Rule-based reasoning process: The feature sequence is matched with a structured expert rule base, where each rule contains a logical condition expression and a corresponding rule-action weight vector; the condition satisfaction degree is calculated, and rules that satisfy the condition are activated and their weight vectors are scaled to generate a contribution vector; all contribution vectors are weighted and summed to generate a motion type preference vector. Collaborative filtering process: In the historical user database, the similarity between the current user feature sequence and each historical user profile is calculated, and the K most similar users are selected to form a neighborhood; the historical preferences and positive feedback data of this neighborhood for various exercise modes are aggregated and analyzed; based on the data, the group preference score is calculated and normalized to generate a collaborative filtering recommendation result vector; Machine learning prediction process: The feature sequence is input into a pre-trained multi-class prediction model; the model outputs the user's adaptation probability or benefit score distribution for each preset exercise type; the distribution is normalized and output as a machine learning prediction result vector.

4. The method according to claim 1, characterized in that, The weighting coefficient configuration strategy adopted by the weighted fusion module in step S5 includes: Assign basic weight coefficients α, β, and γ to the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector, respectively, and satisfy α+β+γ=1; The weighting coefficients can be dynamically adjusted based on user identity attributes or usage stage. For new users, α > β > γ is set to prioritize security; for experienced users, the weights of β and γ are increased to enhance personalization and exploration. Based on the assigned weight coefficients, a fusion calculation is performed to generate the final motion pattern fitness score vector.

5. The method according to claim 1, characterized in that: The weighted fusion module in step S5 performs the following fusion calculation operation: it receives the motion type preference vector, collaborative filtering recommendation result vector, and machine learning prediction result vector generated by the multi-source hybrid decision model as input; according to the configured weight coefficients, it performs linear weighted summation calculation on the three input vectors to integrate the comprehensive decision information of rule reasoning, group preference and model prediction, and outputs a unified motion mode fitness score vector.

6. The method according to claim 1, characterized in that: The sports science knowledge graph in step S6 is a structured semantic network containing multiple relationships, which includes at least movement patterns, specific events, motion parameters, target muscle groups, energy systems, and applicable equipment as entity nodes. The mapping and instantiation process specifically includes: First, determining the recommended movement mode based on the movement mode fitness score vector; Subsequently, in the sports science knowledge graph, along the relationship path of sports mode-containment-specific item, all specific sports item entities belonging to the mode are retrieved; Next, combining the user's personalized parameters and the constraints of currently available equipment, specific intensity, number of sets, number of repetitions, and rest interval parameters are inferred and instantiated from the motion parameter attributes associated with the specific sports project entity; finally, a complete parameterized personalized sports planning scheme is synthesized.

7. The method according to claim 1, characterized in that: It also includes step S7, which, when the user is detected to switch from the first smart sports device to the second smart sports device, obtains the user's training history data on the first smart sports device, the dynamic user state model, and the device capability description information of the second smart sports device; performs intelligent conversion and mapping across devices through the AI ​​biological model to generate a real-time exercise plan that is adapted to the second smart sports device and coordinated with the previous stage of training, so as to ensure the equivalence and continuity of training load and physiological stimulation.

8. A multi-device collaborative AI personalized motion planning and generation system, characterized in that, For implementing the method as described in any one of claims 1-7, comprising: The parameter acquisition module is used to collect the user's basic parameter set and current dynamic physiological parameter set; The feature engineering module is used to generate multi-dimensional derived feature vectors and standardized user vectors. The feature fusion module is used to generate a multi-dimensional user status profile feature sequence; The multi-source decision module is used to perform rule reasoning, collaborative filtering, and machine learning prediction in parallel, and output the corresponding vectors. The weighted fusion module is used to assign weights and perform fusion calculations on various types of vectors; The knowledge graph module is used to store motion-related entities and relationships, and supports solution mapping and instantiation; The device collaboration module is used to handle motion planning conversion and adaptation during device switching.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images