A Personalized Prescription Generation Method Based on Sports Big Data

Abstract

This invention relates to the field of sports and health big data processing and personalized exercise prescription generation technology, and discloses a method for generating personalized prescriptions based on sports big data. This method covers the entire process of exercise data acquisition, time-series standardization, feature extraction, grading determination, and prescription output. By collecting continuous heart rate, displacement, and time information, an equidistant heart rate sequence is reconstructed and normalized. Combined with frequency domain transformation and integral analysis, the physiological and behavioral characteristics of the exerciser are accurately obtained. Based on low-frequency and high-frequency energy integrals and combined with average velocity, a training intensity index is constructed, and grading thresholds are set to achieve automated personalized exercise prescription output. The entire process supports historical data archiving and trend analysis. This method solves the problem of traditional methods relying on single indicators and subjective experience making precise evaluation difficult, thus improving the scientific rigor, relevance, and intelligence of exercise prescriptions.

Description

Technical Field

[0001] This invention relates to the field of sports and health big data processing and personalized exercise prescription generation technology, specifically a method for generating personalized prescriptions based on sports big data. Background Technology

[0002] With the popularization of health management and scientific fitness concepts, wearable device-based technologies for collecting and analyzing exercise physiological parameters have been widely applied. Existing big data analysis methods for sports and health mainly rely on basic data such as exercise heart rate, cadence, and calorie consumption. This data is aggregated and statistically analyzed on the device or cloud to provide users with general health advice and exercise interventions. However, traditional methods often use single or static indicators such as average heart rate, maximum heart rate, and energy consumption as the main criteria for judging individual exercise load and health levels, lacking in-depth analysis of dynamic characteristics such as time-varying heart rate sequences, frequency domain response, and the coupling of exercise intensity and cardiopulmonary regulation. Therefore, existing technologies have significant limitations in accurately reflecting the cardiopulmonary function status, training load changes, and individual differences during exercise.

[0003] Most current mainstream exercise prescription generation technologies rely on preset models, expert experience, or templates trained on historical data, lacking dynamic and personalized modeling capabilities for different users, different training cycles, heart rate response patterns, exercise intensity regulation, and phase trends. On the one hand, general templates cannot adapt to individual differences among athletes in terms of physiological age, health status, and training goals, resulting in insufficient accuracy and applicability of exercise recommendations. On the other hand, existing analysis methods generally lack dynamic big data processing algorithms based on multi-scale and multi-domain approaches, failing to fully utilize composite features such as time and frequency domains to systematically quantify exercise health status, easily leading to misjudgments of abnormal data, extreme conditions, or missing data. Furthermore, traditional methods in frequency domain analysis often employ qualitative descriptions or only focus on extreme points, failing to effectively integrate the spectral structure of continuous heart rate variability with actual training load, resulting in a lack of scientific and quantifiable decision-making basis for exercise prescription intensity grading and training plan parameter recommendations. With the increasing accuracy of wearable device data collection and the rapid growth of health data, existing methods for sports and health data analysis and personalized prescription generation urgently need to overcome the limitations of relying solely on static means or rules of thumb. They must introduce more rigorous, continuous, and fine-grained time-series modeling and frequency domain quantization mechanisms to achieve comprehensive extraction and structured output of key physiological information such as individual training response processes, dynamic heart rate variability, and the coupling of training intensity and cardiopulmonary load. Simultaneously, for complex scenarios such as multi-source heterogeneous big data, extreme or anomalous samples, and historical training trends, existing technologies need to improve the robustness and adaptability of algorithms to avoid analysis failures or unscientific recommendations due to special data conditions.

[0004] Therefore, this case aims to propose a personalized prescription generation method based on sports big data. Based on multi-source raw data such as heart rate changes, movement displacement, and time during exercise, the method automatically outputs refined, scientific, and dynamically adjustable personalized training prescriptions through data structuring, standardization, frequency domain feature extraction, hierarchical judgment, and parameterization. Summary of the Invention

[0005] This invention provides a method for generating personalized prescriptions based on sports big data, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for generating personalized prescriptions based on sports big data, comprising:

[0007] Collect continuous heartbeat time information, displacement, and time information during the movement, calculate the time difference between adjacent heartbeats to form an RR interval sequence, and calculate the average velocity to generate a structured data packet with unified fields.

[0008] The RR interval sequence is standardized, a fixed sampling step size is set to establish an equally spaced time axis, and the equally spaced RR estimation sequence is reconstructed by linear interpolation according to boundary and internal rules, and normalized according to the maximum and minimum values.

[0009] A frequency axis is established based on the sampling step size. A real cosine frequency domain transform is performed on the normalized RR estimation sequence to obtain the discrete power spectral density.

[0010] Two frequency bands, low frequency and high frequency, are set, and the power spectrum is discretely integrated within the corresponding frequency bands to obtain the low frequency integral value and the high frequency integral value respectively;

[0011] The bandwidth ratio is calculated based on the low-frequency integral value and the high-frequency integral value, and then combined with the average speed to form the training intensity index.

[0012] Set two thresholds for the training intensity index, and determine the training intensity index as one of three prescription levels according to the threshold range;

[0013] Based on the prescription level, the training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets are determined from the preset parameter table to form a structured prescription output;

[0014] The daily training intensity index is archived historically, and the average intensity is calculated by stage, generating stage trend output.

[0015] Optionally, the process of collecting continuous heartbeat time information, displacement, and time information during the movement, calculating the time difference between adjacent heartbeats to form an RR interval sequence, and simultaneously calculating the average velocity, generates a structured data packet with unified fields, specifically including:

[0016] During running training, a chest strap heart rate monitor was used to record the moments of continuous heartbeats.

[0017] The difference between two consecutive heartbeat times is used to obtain the RR interval sequence;

[0018] The process terminates when the number of RR interval samples is less than two, and outputs a structured prompt containing the reason and operation fields.

[0019] The total distance and total time were recorded synchronously using positioning and timing devices, and the average speed was calculated. The distance and time were data from the same training process.

[0020] The RR interval sequence, average speed, total distance, and total time are uniformly named and encapsulated to generate a structured data packet corresponding to one training session.

[0021] Optionally, the standardization of the RR interval sequence involves setting a fixed sampling step size to establish an equally spaced time axis, reconstructing the equally spaced RR estimation sequence using linear interpolation according to boundary and internal rules, and normalizing it based on the maximum and minimum values. Specifically, this includes:

[0022] Convert the time of each heartbeat to the relative time of the first heartbeat, and determine the total duration of the recording period;

[0023] Set the equal-interval sampling step size to 200 milliseconds, establish an equal-interval time axis from the start to the end of the recording period, and determine the total number of sampling points;

[0024] When the total number of sampling points is zero, it is adjusted to one sampling point and the first sampling time is set as the last relative heartbeat time;

[0025] In the case of only two heartbeats, the same RR estimate as the second heartbeat is directly assigned to all sampling times;

[0026] When the heart rate is no less than three times and any sampling time is within the interval defined by two adjacent heartbeats, the estimated RR value for that sampling time is calculated according to the linear interpolation rule of that interval;

[0027] When the sampling time is earlier than the relative time of the second heartbeat, the RR estimate is set to be the same as the second heartbeat.

[0028] When the sampling time is within the interval between the last two heartbeats, the estimated RR value is calculated according to the linear interpolation rule of that interval;

[0029] When the sampling time is aligned with the end of the record, the last RR interval is directly taken as the RR estimate.

[0030] Obtain the maximum and minimum values ​​of the RR estimation sequence. When the maximum value is greater than the minimum value, normalize it in an interval linear manner. When the two values ​​are equal, set the normalization result of each sampling point to zero.

[0031] Optionally, the step of establishing a frequency axis based on the sampling step size and performing a real cosine frequency domain transform on the normalized RR estimation sequence to obtain the discrete power spectral density specifically includes:

[0032] Convert the sampling step size to seconds;

[0033] The frequency resolution is set according to the number of samples and the sampling step size to generate a discrete frequency point sequence that starts from zero and increases with a fixed step size.

[0034] Perform a real cosine transform on the normalized RR estimation sequence to obtain the frequency domain coefficients corresponding to each discrete frequency point;

[0035] Calculate the power spectral density at each discrete frequency point based on the frequency domain coefficients.

[0036] Optionally, the step of setting two frequency bands, low frequency and high frequency, and performing discrete integration on the power spectrum within the corresponding frequency bands to obtain low-frequency integral values ​​and high-frequency integral values ​​respectively includes:

[0037] Set the low-frequency cutoff frequency to 0.15 Hz and the high-frequency cutoff frequency to 0.40 Hz, while satisfying the Nyquist condition related to the sampling frequency;

[0038] Determine the upper bound index of the two frequency bands on the discrete frequency axis based on the frequency resolution, and exclude the DC component;

[0039] When the upper bound index of the low-frequency band satisfies the integration condition, the low-frequency band power spectrum is discretely integrated to obtain the low-frequency integral value; otherwise, the low-frequency integral value is set to zero.

[0040] When the index relationship between the high-frequency band and the low-frequency band satisfies the integration condition, the high-frequency power spectrum is discretely integrated to obtain the high-frequency integral value; otherwise, the high-frequency integral value is set to zero.

[0041] Optionally, the step of calculating the frequency band ratio based on the low-frequency integral value and the high-frequency integral value, and combining it with the average velocity to form a training intensity index, specifically includes:

[0042] When the high-frequency integral value is greater than zero, calculate the ratio of the low-frequency integral value to the high-frequency integral value;

[0043] When the high-frequency integral value is equal to zero and the low-frequency integral value is greater than zero, the low-frequency integral value is scaled according to the frequency resolution and used as the ratio.

[0044] When both the low-frequency integral value and the high-frequency integral value are equal to zero, the ratio is recorded as one.

[0045] Multiplying the ratio by the average speed yields the training intensity index, measured in meters per second.

[0046] Optionally, setting two thresholds for the training intensity index and determining the training intensity index as one of three prescription levels according to the threshold range specifically includes:

[0047] Set the two segment thresholds for the training intensity index to 0.8 m / s and 1.8 m / s;

[0048] When the training intensity index is less than 0.8 meters per second, it is determined to be prescription level one;

[0049] When the training intensity index is between 0.8 m / s and 1.8 m / s, it is determined to be prescription level two;

[0050] When the training intensity index is not less than 1.8 meters per second, it is determined to be prescription level three.

[0051] Optionally, the step of determining training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets from a preset parameter table according to prescription level to form a structured prescription output specifically includes:

[0052] Establish a prescription parameter table:

[0053] Prescription Level 1: Training duration 30 minutes, recommended running speed is 70% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets;

[0054] Prescription Level 2: Training duration 45 minutes, recommended running speed is 90% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets;

[0055] Prescription Level 3: Training duration 40 minutes, recommended running speed equal to average speed, single interval distance 400 meters, number of interval sets 6, rest time between sets 90 seconds;

[0056] The prescription level, historical index, user ID, recommended running speed, training duration, interval distance, number of interval sets, and rest time between sets are encapsulated into a structured prescription output package.

[0057] Optionally, the step of archiving the daily training intensity index historically, calculating the average intensity by stage, and generating stage trend output specifically includes:

[0058] Record training intensity indices daily and generate historical intensity sequences arranged chronologically by date;

[0059] When the number of historical days is not less than one day, the average training intensity of the stage shall be calculated using an equal-weighted average method.

[0060] When the number of historical days is zero, the average training intensity of the phase is set as the current training intensity index;

[0061] The output includes a stage trend structured result containing the stage average training intensity and the historical intensity sequence.

[0062] The present invention has the following beneficial effects:

[0063] This invention synchronously collects raw data on heart rate intervals, movement displacement, and duration, and encapsulates it into structured data packets, ensuring consistency and spatiotemporal coupling of exercise physiological and behavioral information. This standardized end-to-end collection mechanism ensures data integrity for subsequent analysis and facilitates data comparison and processing across devices and scenarios. Compared to previous practices that focused only on a single heart rate or displacement while ignoring the correspondence between collection standards and time sequences, this invention improves the scientific rigor and usability of exercise data. Especially in real-world exercise scenarios, data anomalies and missing data can easily lead to analytical distortion. The solution's specially designed anomaly detection and prompts (such as automatically terminating the process and providing a prompt when the number of heart rate samples is insufficient) enhance the system's fault tolerance and user experience, effectively avoiding the negative impact of non-standard data on personalized prescription generation.

[0064] This invention focuses on the standardized processing of continuous heart rate interval data. A fixed sampling step size is used to establish an equally spaced time axis, and linear interpolation reconstruction and normalization of the heart rate interval are achieved based on boundary and interval rules. The innovation lies in fully considering practical problems such as irregular sampling times, data sparsity, and missing data during exercise. Through boundary conditions and piecewise interpolation algorithms, high-quality continuous and normalized heart rate data is achieved, ensuring the accuracy of subsequent feature extraction. Compared to existing methods that directly utilize raw heart rate data or simply perform averaging, this invention more realistically reflects the dynamic changes in heart rate during exercise, providing a solid data foundation for frequency domain analysis and intensity discrimination. Furthermore, through normalization processing, the comparability and universality of data from different users, devices, and exercise states are enhanced, expanding the method's wide application capabilities in big data environments.

[0065] This method employs a real-cosine frequency domain transform to map the normalized RR estimation sequence to the frequency domain, calculating the discrete power spectral density and deeply exploring the frequency domain characteristics of heart rate variability. This approach breaks through the traditional approach of focusing only on superficial indicators such as time-domain mean or peak values, transforming the time-varying information of physiological signals into frequency-domain energy distribution, thus better revealing the complex mechanisms of autonomic nervous system regulation and cardiac response during exercise. Compared to existing methods that only analyze average or maximum heart rate, this method demonstrates higher sensitivity and scientific rigor in reflecting cardiac function and fatigue state, and capturing training load and recovery status. It lays a solid foundation for subsequent quantitative intensity discrimination based on frequency band energy and makes the intelligent generation of exercise prescriptions more closely aligned with individual physiological characteristics.

[0066] This invention clearly defines two classic frequency bands for heart rate variability analysis: low-frequency and high-frequency. It employs discrete integration to extract energy from each band, avoiding the limitations of relying on single indicators that easily lose information. By meticulously setting integration boundaries, excluding DC components, and dynamically adjusting the integration range according to sampling resolution, it effectively adapts to different data sampling conditions, ensuring the scientific rigor and comparability of the analysis results. Compared to previous methods that relied on subjective manual division or were susceptible to sampling anomalies during integration, this method offers improvements in both robustness and applicability. It can accurately capture exercise-induced changes in autonomic nerve activity, providing more objective data support for applications such as exercise load, training recovery, and health risk assessment.

[0067] This invention innovatively combines the ratio of low-frequency to high-frequency integrals with average speed, integrating heart rate variability characteristics and exercise behavior parameters to construct a training intensity index. This comprehensive index can more comprehensively and dynamically reflect training load and individual responses, avoiding the one-sidedness of relying solely on a single physiological signal or behavioral indicator. Compared with existing methods that rely solely on subjective scoring and single threshold determination, this training intensity index organically combines the complexity of heart rate variability with the objectivity of exercise performance, enhancing the scientific basis for personalized assessment and improving the quantification and refinement of individual prescriptions. Especially under big data platforms, this comprehensive index facilitates batch analysis, group intervention, and longitudinal trend tracking, expanding the depth and breadth of sports health management.

[0068] This invention sets dual thresholds based on a training intensity index to scientifically classify training load, automatically determining three exercise prescription levels corresponding to different training needs such as recovery, aerobic, and interval training. This classification is not only dynamically generated based on individual data but also quantitatively mapped using scientific principles of exercise physiology, overcoming the drawbacks of previous experience-based and subjective classifications. By setting reasonable segmentation intervals and classification labels, the solution provides clear data support for interventions targeting different training goals, effectively improving the targeting and operability of exercise prescriptions. Simultaneously, this classification system provides a standardized foundation for subsequent training plan parameterization, exercise risk control, and continuous tracking, promoting the intelligent, personalized, and standardized development of exercise intervention.

[0069] This solution, based on graded assessment results, intelligently matches specific parameters such as recommended training duration, running speed, interval distance, number of sets, and rest time from a preset parameter table, automatically generating a structured, personalized exercise prescription. Its innovation lies in the fact that the parameter configuration not only originates from exercise physiology theory but also corresponds to the individual's actual training intensity, achieving dynamic, data-driven, and automated prescription customization. Compared to traditional methods that rely on manual adjustments or a fixed template for a "one-size-fits-all" approach, this method improves the individual fit and scientific rigor of prescription recommendations, helps avoid the risks of overtraining or undertraining, and comprehensively enhances the actual effectiveness of exercise intervention and user satisfaction.

[0070] This invention archives daily training intensity indices, establishes historical intensity sequences, and dynamically calculates stage-specific average intensity and trends, forming a visualized stage-specific training performance assessment. This mechanism not only helps users and coaches scientifically track the effects of exercise interventions but also provides data support for subsequent personalized adjustments and risk warnings. Compared to traditional methods that only consider immediate indicators and lack historical longitudinal analysis, this invention can more systematically and dynamically reflect the changing trends of individual athletic performance, providing a solid data foundation and decision support for the scientific management of sports health and the optimization of training pathways. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example, refer to Figure 1 A personalized prescription generation method based on sports big data includes:

[0074] Collect continuous heartbeat time information, displacement, and time information during the movement, calculate the time difference between adjacent heartbeats to form an RR interval sequence, and calculate the average velocity to generate a structured data packet with unified fields.

[0075] The RR interval sequence is standardized, a fixed sampling step size is set to establish an equally spaced time axis, and the equally spaced RR estimation sequence is reconstructed by linear interpolation according to boundary and internal rules, and normalized according to the maximum and minimum values.

[0076] A frequency axis is established based on the sampling step size. A real cosine frequency domain transform is performed on the normalized RR estimation sequence to obtain the discrete power spectral density.

[0077] Two frequency bands, low frequency and high frequency, are set, and the power spectrum is discretely integrated within the corresponding frequency bands to obtain the low frequency integral value and the high frequency integral value respectively;

[0078] The bandwidth ratio is calculated based on the low-frequency integral value and the high-frequency integral value, and then combined with the average speed to form the training intensity index.

[0079] Set two thresholds for the training intensity index, and determine the training intensity index as one of three prescription levels according to the threshold range;

[0080] Based on the prescription level, the training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets are determined from the preset parameter table to form a structured prescription output;

[0081] The daily training intensity index is archived historically, and the average intensity is calculated by stage, generating stage trend output.

[0082] This system encompasses the entire process of data acquisition, time-series processing, feature extraction, grading and determination, and structured output. By collecting continuous heart rate intervals, displacement, and duration information during exercise, the system obtains fundamental data reflecting the exerciser's physiological and behavioral patterns. Subsequent standardization and interpolation steps reconstruct the heart rate interval sequence into equally spaced data, avoiding the impact of uneven data distribution on the accuracy of feature analysis. Frequency domain transformation is used to extract energy distribution from the time-series data, and integration across low- and high-frequency intervals better reveals the patterns of heart rate changes during exercise. By setting thresholds, training intensity is quantitatively graded, and further, combined with a preset parameter table, personalized prescriptions such as training duration and recommended running speed are automatically output, achieving fully automated processing from data to exercise intervention recommendations. This method overcomes the limitations of relying on single indicators or subjective experience in accurately assessing individual differences, improving the scientific rigor, relevance, and operability of exercise prescriptions. It also enables historical data archiving and trend tracking, facilitating subsequent personalized optimization and health management.

[0083] The process involves collecting continuous heartbeat time information, displacement, and time information during the movement; calculating the time difference between adjacent heartbeats to form an RR interval sequence; simultaneously calculating the average velocity; and generating a structured data packet with unified fields, specifically including:

[0084] During running training, a chest strap heart rate monitor was used to record the moments of continuous heartbeats.

[0085] The difference between two consecutive heartbeat times is used to obtain the RR interval sequence;

[0086] The process terminates when the number of RR interval samples is less than two, and outputs a structured prompt containing the reason and operation fields.

[0087] The total distance and total time were recorded synchronously using positioning and timing devices, and the average speed was calculated. The distance and time were data from the same training process.

[0088] The RR interval sequence, average speed, total distance, and total time are uniformly named and encapsulated to generate a structured data packet corresponding to one training session.

[0089] Further specific implementation steps include:

[0090] During running training, continuous heart rate interval data are collected using a chest strap heart rate monitor; the number of consecutive heartbeats recorded during the run is denoted as K, and the duration of each consecutive heartbeat is denoted as . ;in, The time of the kth heartbeat, in milliseconds; k is the heartbeat sequence index.

[0091] Set the RR interval between adjacent heartbeats as follows: ;in, The RR interval between the k-th and (k-1)-th heartbeats, in milliseconds;

[0092] Obtain the complete RR interval sequence: ;in, A set of RR interval sequences;

[0093] If K ≥ 2, continue; if K < 2, terminate the current prescription generation process and output a structured prompt:

[0094] ;in, This is a collection of abnormal outputs;

[0095] The total training distance was recorded synchronously using GPS and a timing device. Total time Calculate the average speed index ;in, The unit is meters. The unit is seconds;

[0096] All data will be uniformly named and stored in a structured format as follows: ;in, This is the structured data package for this training.

[0097] The system specifies a detailed workflow for data acquisition and structured processing. During exercise, continuous heartbeat intervals are collected using devices such as a chest strap, along with simultaneous acquisition of displacement and time, ensuring the consistency and integrity of all basic data. The system automatically calculates the time difference between adjacent heartbeats to obtain a continuous heartbeat interval sequence, while also calculating the average speed of the exercise. If the number of heartbeat samples is insufficient, the process is terminated directly, and a structured prompt is output to avoid interference from non-standard data in the analysis results. Finally, all raw data is standardized, named, and uniformly packaged into a structured data package for convenient subsequent processing. This approach not only effectively improves the scientific rigor and comparability of the data but also reduces the impact of abnormal data or acquisition errors on personalized prescription generation. Compared to existing methods that often neglect multi-source data synchronization and automatic handling of abnormal data, this method ensures the systematic nature and reliability of data processing, laying the foundation for high-quality subsequent analysis.

[0098] The standardization process for the RR interval sequence involves setting a fixed sampling step size to establish an equally spaced time axis, reconstructing the equally spaced RR estimation sequence using linear interpolation according to boundary and internal rules, and normalizing it based on the maximum and minimum values. Specifically, this includes:

[0099] Convert the time of each heartbeat to the relative time of the first heartbeat, and determine the total duration of the recording period;

[0100] Set the equal-interval sampling step size to 200 milliseconds, establish an equal-interval time axis from the start to the end of the recording period, and determine the total number of sampling points;

[0101] When the total number of sampling points is zero, it is adjusted to one sampling point and the first sampling time is set as the last relative heartbeat time;

[0102] In the case of only two heartbeats, the same RR estimate as the second heartbeat is directly assigned to all sampling times;

[0103] When the heart rate is no less than three times and any sampling time is within the interval defined by two adjacent heartbeats, the estimated RR value for that sampling time is calculated according to the linear interpolation rule of that interval;

[0104] When the sampling time is earlier than the relative time of the second heartbeat, the RR estimate is set to be the same as the second heartbeat.

[0105] When the sampling time is within the interval between the last two heartbeats, the estimated RR value is calculated according to the linear interpolation rule of that interval;

[0106] When the sampling time is aligned with the end of the record, the last RR interval is directly taken as the RR estimate.

[0107] Obtain the maximum and minimum values ​​of the RR estimation sequence. When the maximum value is greater than the minimum value, normalize it in an interval linear manner. When the two values ​​are equal, set the normalization result of each sampling point to zero.

[0108] Further specific implementation steps include:

[0109] Set relative heartbeat timing: ;in, The time between the k-th heartbeat and the first heartbeat, in milliseconds;

[0110] Total recording duration: ;in, This represents the total duration of the RR record interval, in milliseconds. The value at k=K at a relative time.

[0111] Set equal sampling step size Construct equidistant relative time axes:

[0112] ;in, This represents the nth equally spaced sampling time; n is the sampling point number. This represents the total number of sampling points;

[0113] If N=0, then let N=1. ;

[0114] For any sampling time Perform steps S201 to S205:

[0115] S201, Case with only two heartbeats: If K=2, then for all n, directly take... ;in, Sampling time The corresponding RR estimate;

[0116] S202. If K≥3 and there exists a unique [condition] make Then, it is calculated using linear interpolation of adjacent RR values. Where j is the left-end RR index of the interpolation interval;

[0117] S203, Initial Boundary: If ,but ;in, This refers to the moment relative to the second heartbeat;

[0118] S204, End-interval interpolation: If Substituting j=K-1 into the interpolation calculation is equivalent to ;in, , The relative moments of the last two heartbeats;

[0119] S205, End Boundary: If it appears For integer alignment cases, interpolation is directly taken. ;in, The last RR interval;

[0120] make ;in, RR estimation sequence The maximum value; RR estimation sequence The minimum value;

[0121] like Then normalization , ;in, This is the normalized RR value for the nth sampling point;

[0122] like Then let , .

[0123] This paper details the standardization and reconstruction methods for heart rate interval sequences. By converting the original heartbeat occurrence time into a relative time and setting a fixed, equidistant sampling step size, the system can uniformly reconstruct heart rate data sequences. For cases with insufficient data samples or special boundary conditions, the claims design boundary conditions and interpolation rules to ensure that all sampling points obtain appropriate RR interval estimates. Finally, normalization processing transforms data from different ranges into a unified interval, improving the comparability between different samples and the stability of subsequent analysis. This method eliminates the problems of uneven sampling and scale inconsistency in the original data through interpolation and normalization, improving the scientific rigor and accuracy of subsequent frequency domain analysis. Compared with existing methods that simply use the mean or ignore outliers, this method can more realistically reflect the continuous changes in heart rate during exercise, making subsequent feature extraction and personalized judgment more accurate.

[0124] The process of establishing a frequency axis based on the sampling step size and performing a real cosine frequency domain transform on the normalized RR estimation sequence to obtain the discrete power spectral density specifically includes:

[0125] Convert the sampling step size to seconds;

[0126] The frequency resolution is set according to the number of samples and the sampling step size to generate a discrete frequency point sequence that starts from zero and increases with a fixed step size.

[0127] Perform a real cosine transform on the normalized RR estimation sequence to obtain the frequency domain coefficients corresponding to each discrete frequency point;

[0128] Calculate the power spectral density at each discrete frequency point based on the frequency domain coefficients.

[0129] Further specific implementation steps include:

[0130] Convert the sampling step size to seconds: ;in, The sampling step size is expressed in seconds.

[0131] Set the frequency resolution to The unit is Hertz;

[0132] Set the m-th frequency point as Where m is the frequency index;

[0133] right Perform a real cosine transformation: m = 1, 2, ..., N; where, This is the m-th real frequency domain coefficient;

[0134] Calculate the m-th discrete power spectral density value , m=1,2,...,N.

[0135] This paper proposes specific steps to obtain the discrete power spectral density from the reconstructed normalized heart rate interval sequence through a real-cosine frequency domain transform. By mapping the equally spaced sampling sequence to the frequency domain and calculating the energy distribution at each frequency point, the system can uncover the time-varying patterns and frequency domain characteristics behind the heart rate signal. This processing step can identify physiological phenomena such as heart rate fluctuations during exercise, fatigue states, and autonomic nervous system regulation, providing a scientific basis for the quantitative assessment of exercise intensity and recovery status. Unlike traditional approaches that only focus on time-domain characteristics or simple maximum / minimum value statistics, this method can comprehensively reflect the dynamic trends and complexity of heart rate changes during exercise, improving the depth and accuracy of sports and health data analysis.

[0136] The process involves setting up two frequency bands, low frequency and high frequency, and then performing discrete integration on the power spectrum within the corresponding frequency bands to obtain the low-frequency integral value and the high-frequency integral value, specifically including:

[0137] Set the low-frequency cutoff frequency to 0.15 Hz and the high-frequency cutoff frequency to 0.40 Hz, while satisfying the Nyquist condition related to the sampling frequency;

[0138] Determine the upper bound index of the two frequency bands on the discrete frequency axis based on the frequency resolution, and exclude the DC component;

[0139] When the upper bound index of the low-frequency band satisfies the integration condition, the low-frequency band power spectrum is discretely integrated to obtain the low-frequency integral value; otherwise, the low-frequency integral value is set to zero.

[0140] When the index relationship between the high-frequency band and the low-frequency band satisfies the integration condition, the high-frequency power spectrum is discretely integrated to obtain the high-frequency integral value; otherwise, the high-frequency integral value is set to zero.

[0141] Further specific implementation steps include:

[0142] Set the low-frequency band cutoff frequency to Set the high-frequency band cutoff frequency to and satisfy ;

[0143] The corresponding upper bounds of the indices are as follows: , ;in, This is the upper bound of the frequency point index for the low-frequency band; This serves as the upper bound for the frequency point index in the high-frequency band.

[0144] It also displays the exclusion of the DC component m=1;

[0145] like ,set up Otherwise ;in, This is the integral value of the power spectrum in the low-frequency band, in Hertz.

[0146] like ,set up ,otherwise ;in, This is the integral value of the power spectrum in the high-frequency band.

[0147] This paper specifies the detailed integration operations for extracting energy from two typical physiological frequency bands—low-frequency and high-frequency—based on frequency domain features. By scientifically setting frequency band boundaries, excluding DC components, and adapting the integration range according to the sampling resolution, the system can accurately calculate the low-frequency and high-frequency energies reflecting autonomic nerve activity. This method solves the one-sidedness of previous methods that directly extract total energy or analyze only a single interval. It can more meticulously reflect the dynamic response of exercise heart rate under different physiological regulatory mechanisms, providing stronger quantitative support for training load, recovery level, and health risk analysis. Unlike existing analytical methods that rely solely on subjective experience or crude statistics, this method enhances the scientific rigor, robustness, and applicability of frequency domain feature extraction, improving the objective basis for subsequent personalized prescription generation.

[0148] The calculation of the bandwidth ratio based on the low-frequency integral value and the high-frequency integral value, and the combination with the average velocity to form the training intensity index, specifically includes:

[0149] When the high-frequency integral value is greater than zero, calculate the ratio of the low-frequency integral value to the high-frequency integral value;

[0150] When the high-frequency integral value is equal to zero and the low-frequency integral value is greater than zero, the low-frequency integral value is scaled according to the frequency resolution and used as the ratio.

[0151] When both the low-frequency integral value and the high-frequency integral value are equal to zero, the ratio is recorded as one.

[0152] Multiplying the ratio by the average speed yields the training intensity index, measured in meters per second.

[0153] Further specific implementation steps include:

[0154] Constructing the frequency domain ratio index ;

[0155] Synthetic training intensity index The unit is m / s.

[0156] This paper proposes a method to construct a ratio index from the integral values ​​of low and high frequencies, and then combine it with average velocity to form a training intensity index. By fusing the frequency domain ratio reflecting heart rate variability with the average velocity reflecting actual exercise performance, a more comprehensive and dynamic evaluation of the exerciser's training load can be achieved. This method solves the problem of distortion or incompleteness when using only a single physiological signal or behavioral indicator, improving the scientific rigor and precision of training load assessment. Through the systematic integration of physiological and behavioral data, it not only enhances individualized assessment capabilities but also facilitates batch, trend, and cluster analysis. Compared with existing experience-based scoring or item-based judgment methods, this invention achieves automatic quantification of exercise load, reduces subjectivity, and provides a data foundation for precise exercise prescription formulation.

[0157] The setting of two thresholds for the training intensity index, and the determination of the training intensity index into one of three prescription levels according to the threshold range, specifically includes:

[0158] Set the two segment thresholds for the training intensity index to 0.8 m / s and 1.8 m / s;

[0159] When the training intensity index is less than 0.8 meters per second, it is determined to be prescription level one;

[0160] When the training intensity index is between 0.8 m / s and 1.8 m / s, it is determined to be prescription level two;

[0161] When the training intensity index is not less than 1.8 meters per second, it is determined to be prescription level three.

[0162] Further specific implementation steps include:

[0163] Setting the segment boundaries for the training intensity index is divided into: , ;in, , This refers to the prescription segmentation threshold.

[0164] Establish classification mapping rules:

[0165] Among them, A, B, and C are category labels.

[0166] The system clearly defines the segmented thresholds and classification rules for the training intensity index, quantitatively dividing training load into three prescription levels. By establishing scientifically sound thresholds and classification labels, the system can automatically map individual training intensity to different intervention recommendations such as recovery, aerobic, and interval training. This step solves the problems of inconsistent grading standards and implementation difficulties caused by previous subjective grading or reliance on expert experience. Quantitative grading not only provides a clear scientific basis for exercise intervention but also facilitates the parameterization and standardized management of subsequent training plans. Compared with traditional empirical or manual adjustments, this method improves the scientific rigor, consistency, and operability of grading, contributing to the automatic generation and mass promotion of exercise prescriptions.

[0167] The process of determining training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets from a preset parameter table according to prescription level to form a structured prescription output specifically includes:

[0168] Establish a prescription parameter table:

[0169] Prescription Level 1: Training duration 30 minutes, recommended running speed is 70% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets;

[0170] Prescription Level 2: Training duration 45 minutes, recommended running speed is 90% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets;

[0171] Prescription Level 3: Training duration 40 minutes, recommended running speed equal to average speed, single interval distance 400 meters, number of interval sets 6, rest time between sets 90 seconds;

[0172] The prescription level, historical index, user ID, recommended running speed, training duration, interval distance, number of interval sets, and rest time between sets are encapsulated into a structured prescription output package.

[0173] Further specific implementation steps include:

[0174] Construct a prescription generation table:

[0175] ;in, Recommended training duration; Recommended running speed; G represents the recommended distance for a single interval; G represents the recommended number of interval groups. The resting time between the two groups;

[0176] Encapsulate the training plan into structured output:

[0177] ;in, This is a prescription output package.

[0178] This invention proposes a method that automatically matches and outputs complete exercise prescription parameters, including training duration, recommended running speed, interval distance, number of sets, and resting time, from a preset parameter table based on graded results, and outputs them in structured data format. Through this automated parameter configuration, the most suitable exercise prescription can be automatically recommended for different training levels, achieving intelligent customization and dynamic adjustment of exercise intervention. This method overcomes the shortcomings of traditional methods, such as reliance on human experience, lack of individualized prescription content, and inadequate dynamic adaptability. Compared to a fixed template approach, this invention effectively improves personalization and scientific rigor, reduces the risk of human intervention and operational errors, and makes the execution of exercise programs more targeted and efficient.

[0179] The process of archiving daily training intensity indices historically, calculating average intensity by stage, and generating stage trend output specifically includes:

[0180] Record training intensity indices daily and generate historical intensity sequences arranged chronologically by date;

[0181] When the number of historical days is not less than one day, the average training intensity of the stage shall be calculated using an equal-weighted average method.

[0182] When the number of historical days is zero, the average training intensity of the phase is set as the current training intensity index;

[0183] The output includes a stage trend structured result containing the stage average training intensity and the historical intensity sequence.

[0184] Further specific implementation steps include:

[0185] Let the training intensity index of the user on day d be . Making history days Where d is the day index; It is a non-negative integer;

[0186] Then the historical intensity sequence ;

[0187] like ,set up ;in, This represents the average training intensity during the phase.

[0188] like Then let ;

[0189] Forming structured output ;in, Output package for phased trends.

[0190] This invention clarifies the process of historical archiving and phased trend analysis of training intensity indices. By recording daily training intensity, it dynamically calculates phased averages and historical change sequences, outputting phased trend results. This approach not only facilitates users and professionals in scientifically tracking training effects but also provides data support for continuous optimization of exercise programs and health risk monitoring. Through longitudinal trend analysis, it can dynamically reflect changes in athletic ability and the effectiveness of interventions, enabling personalized adjustments and intelligent risk warnings. Compared to previous practices that only focused on immediate indicators and ignored historical changes, this invention achieves a systematic, scientific, and forward-looking approach to sports health management, effectively improving the accuracy and depth of exercise prescription management.

[0191] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0192] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating personalized prescriptions based on sports big data, characterized in that, include: Collect continuous heartbeat time information, displacement, and time information during the movement, calculate the time difference between adjacent heartbeats to form an RR interval sequence, and calculate the average velocity to generate a structured data packet with unified fields. The RR interval sequence is standardized, a fixed sampling step size is set to establish an equally spaced time axis, and the equally spaced RR estimation sequence is reconstructed by linear interpolation according to boundary and internal rules, and normalized according to the maximum and minimum values. A frequency axis is established based on the sampling step size. A real cosine frequency domain transform is performed on the normalized RR estimation sequence to obtain the discrete power spectral density. Two frequency bands, low frequency and high frequency, are set, and the power spectrum is discretely integrated within the corresponding frequency bands to obtain the low frequency integral value and the high frequency integral value respectively; The frequency band ratio is calculated based on the low-frequency integral value and the high-frequency integral value, and then combined with the average velocity to form the training intensity index, which specifically includes: When the high-frequency integral value is greater than zero, calculate the ratio of the low-frequency integral value to the high-frequency integral value; When the high-frequency integral value is equal to zero and the low-frequency integral value is greater than zero, the low-frequency integral value is scaled according to the frequency resolution and used as the ratio. When both the low-frequency integral value and the high-frequency integral value are equal to zero, the ratio is recorded as one. Multiply the ratio by the average speed to obtain the training intensity index, in meters per second; Set two thresholds for the training intensity index, and determine the training intensity index as one of three prescription levels according to the threshold range; Based on the prescription level, the training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets are determined from the preset parameter table to form a structured prescription output; The daily training intensity index is archived historically, and the average intensity is calculated by stage, generating stage trend output.

2. The personalized prescription generation method based on sports big data according to claim 1, characterized in that, The process involves collecting continuous heartbeat time information, displacement, and time information during the movement; calculating the time difference between adjacent heartbeats to form an RR interval sequence; simultaneously calculating the average velocity; and generating a structured data packet with unified fields, specifically including: During running training, a chest strap heart rate monitor was used to record the moments of continuous heartbeats. The difference between two consecutive heartbeat times is used to obtain the RR interval sequence; The process terminates when the number of RR interval samples is less than two, and outputs a structured prompt containing the reason and operation fields. The total distance and total time were recorded synchronously using positioning and timing devices, and the average speed was calculated. The distance and time were data from the same training process. The RR interval sequence, average speed, total distance, and total time are uniformly named and encapsulated to generate a structured data packet corresponding to one training session.

3. The personalized prescription generation method based on sports big data according to claim 2, characterized in that, The standardization process for the RR interval sequence involves setting a fixed sampling step size to establish an equally spaced time axis, reconstructing the equally spaced RR estimation sequence using linear interpolation according to boundary and internal rules, and normalizing it based on the maximum and minimum values. Specifically, this includes: Convert the time of each heartbeat to the relative time of the first heartbeat, and determine the total duration of the recording period; Set the equal-interval sampling step size to 200 milliseconds, establish an equal-interval time axis from the start to the end of the recording period, and determine the total number of sampling points; When the total number of sampling points is zero, it is adjusted to one sampling point and the first sampling time is set as the last relative heartbeat time; In the case of only two heartbeats, the same RR estimate as the second heartbeat is directly assigned to all sampling times; When the heart rate is no less than three times and any sampling time is within the interval defined by two adjacent heartbeats, the estimated RR value for that sampling time is calculated according to the linear interpolation rule of that interval; When the sampling time is earlier than the relative time of the second heartbeat, the RR estimate is set to be the same as the second heartbeat. When the sampling time is within the interval between the last two heartbeats, the estimated RR value is calculated according to the linear interpolation rule of that interval; When the sampling time is aligned with the end of the record, the last RR interval is directly taken as the RR estimate. Obtain the maximum and minimum values ​​of the RR estimation sequence. When the maximum value is greater than the minimum value, normalize it in an interval linear manner. When the two values ​​are equal, set the normalization result of each sampling point to zero.

4. The personalized prescription generation method based on sports big data according to claim 3, characterized in that, The process of establishing a frequency axis based on the sampling step size and performing a real cosine frequency domain transform on the normalized RR estimation sequence to obtain the discrete power spectral density specifically includes: Convert the sampling step size to seconds; The frequency resolution is set according to the number of samples and the sampling step size to generate a discrete frequency point sequence that starts from zero and increases with a fixed step size. Perform a real cosine transform on the normalized RR estimation sequence to obtain the frequency domain coefficients corresponding to each discrete frequency point; Calculate the power spectral density at each discrete frequency point based on the frequency domain coefficients.

5. The personalized prescription generation method based on sports big data according to claim 4, characterized in that, The process involves setting up two frequency bands, low frequency and high frequency, and then performing discrete integration on the power spectrum within the corresponding frequency bands to obtain the low-frequency integral value and the high-frequency integral value, specifically including: Set the low-frequency cutoff frequency to 0.15 Hz and the high-frequency cutoff frequency to 0.40 Hz, while satisfying the Nyquist condition related to the sampling frequency; Determine the upper bound index of the two frequency bands on the discrete frequency axis based on the frequency resolution, and exclude the DC component; When the upper bound index of the low-frequency band satisfies the integration condition, the low-frequency band power spectrum is discretely integrated to obtain the low-frequency integral value; otherwise, the low-frequency integral value is set to zero. When the index relationship between the high-frequency band and the low-frequency band satisfies the integration condition, the high-frequency power spectrum is discretely integrated to obtain the high-frequency integral value; otherwise, the high-frequency integral value is set to zero.

6. The personalized prescription generation method based on sports big data according to claim 1, characterized in that, The setting of two thresholds for the training intensity index, and the determination of the training intensity index into one of three prescription levels according to the threshold range, specifically includes: Set the two segment thresholds for the training intensity index to 0.8 m / s and 1.8 m / s; When the training intensity index is less than 0.8 meters per second, it is determined to be prescription level one; When the training intensity index is between 0.8 m / s and 1.8 m / s, it is determined to be prescription level two; When the training intensity index is not less than 1.8 meters per second, it is determined to be prescription level three.

7. The personalized prescription generation method based on sports big data according to claim 6, characterized in that, The process of determining training duration, recommended running speed, interval distance, number of interval sets, and rest time between sets from a preset parameter table according to prescription level to form a structured prescription output specifically includes: Establish a prescription parameter table: Prescription Level 1: Training duration 30 minutes, recommended running speed is 70% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets; Prescription Level 2: Training duration 45 minutes, recommended running speed is 90% of average speed, no interval distance set, 1 set of intervals, and 0 seconds of rest time between sets; Prescription Level 3: Training duration 40 minutes, recommended running speed equal to average speed, single interval distance 400 meters, number of interval sets 6, rest time between sets 90 seconds; The prescription level, historical index, user ID, recommended running speed, training duration, interval distance, number of interval sets, and rest time between sets are encapsulated into a structured prescription output package.

8. The method for generating personalized prescriptions based on sports big data according to claim 7, characterized in that, The process of archiving daily training intensity indices historically, calculating average intensity by stage, and generating stage trend output specifically includes: Record training intensity indices daily and generate historical intensity sequences arranged chronologically by date; When the number of historical days is not less than one day, the average training intensity of the stage shall be calculated using an equal-weighted average method. When the number of historical days is zero, the average training intensity of the phase is set as the current training intensity index; The output includes a stage trend structured result containing the stage average training intensity and the historical intensity sequence.

Images