Real-time health early warning system and method based on time perception auto-encoder

Abstract

The invention discloses a real-time health early warning system and method based on a time perception auto-encoder, belongs to the technical field of intelligent systems, and is applied to heart rate data analysis of wearable equipment for non-diagnosis or disease treatment. The implementation method comprises the following steps of: 1, fusing time sequence data of heart rate and step number by using a two-stage filtering method; 2, generating a resting heart rate matrix by using the maximum power of the resting heart rate power spectrum; 3, realizing real-time health early warning based on the resting heart rate matrix; compared with the prior art, the method has the advantages that the resting heart rate matrix is constructed, the specific heart rate cycle mode of the user can be captured by calculating the resting heart rate time domain sequence power spectrum, and the global time interval analysis method and the exponential decay time interval mapping method are proposed to be effectively suitable for processing non-equal interval time sequence data; and meanwhile, a resting heart rate error model based on a dynamic threshold value is provided to construct the health condition for monitoring the wearable equipment wearer in a personalized manner, so that the practicability of the health early warning model is improved.

Description

Technical Field

[0001] The present invention relates to a real-time health warning system and method based on a time-aware autoencoder, belonging to the technical field of intelligent systems and applied to heart rate data analysis of wearable devices for purposes other than diagnosis or treatment of diseases. Background Art

[0002] In the field of military medicine, the application of wearable technology is of great significance. Wearable devices enable all-weather, seamless tracking and recording of individual physiological states. Some wearable devices already support electrocardiogram functions, which can realize arrhythmia detection, heart rate monitoring, sinus rhythm detection, etc. Compared with the current conventional medical care that relies on breakpoint physiological measurements, continuous acquisition and analysis of physiological data can sensitively identify subtle changes and abnormal fluctuations under the normal baseline of an individual. Therefore, wearable devices are increasingly valued in real-time health data collection.

[0003] Real-time health warning data analysis based on wearable devices can be modeled as an anomaly detection model. Abnormal behavior in this context refers to data points that deviate significantly from normal observations, which are related to abnormal physiological states or the presence of external influencing factors. Autoencoders are a widely used wearable device data modeling method, in which the encoding module is dedicated to learning the deep implicit feature representation of the data, while the decoding module aims to accurately reconstruct the original physiological signals. Although autoencoders have shown significant potential in anomaly detection in health alert systems, the quality challenges of data collected by consumer-level wearable devices cannot be ignored. First, the data noise problem caused by hardware limitations and diverse usage scenarios of such devices is particularly prominent. For example, in daily activities, changes in the user's movement state and differences in the environment will inevitably introduce interference factors into the collected physiological indicators. Given the specificity of wearable data noise, existing denoising methods are not effective when directly applied to the analysis of wearable device heart rate warning data. Second, the generation of the resting heart rate matrix is ​​based on the duration of the periodic signal in the time series data. The resting heart rate matrix generated by a fixed period size will destroy the periodic characteristics of the time series data because the periodic characteristics of the time series data obtained by the wearable device have individual differences. In addition, the time series data generated by wearable devices often have irregular time interval characteristics. This is caused by uncertain factors such as the battery life limit of the device itself and the user's wearing habits. Obviously, the importance of data acquired a few hours and a few days ago for analyzing current health data is different. Although traditional preprocessing methods such as interpolation methods can fill in missing data, virtual time series cannot fully reflect the health status of individuals in real life.

[0004] Therefore, in the real-time health warning system of time-aware autoencoders, how to specifically denoise the wearable data, how to effectively use the periodicity of wearable data to generate a resting heart rate matrix, and how to efficiently model resting heart rate time series data have become urgent problems to be solved in view of the quality defects and time structure characteristics of the above-mentioned wearable data. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problems that traditional methods perform poorly in wearable data denoising, generating resting heart rate matrix, and modeling resting heart rate time series data, and propose a real-time health warning system and method based on time-aware autoencoder.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] The present invention discloses a real-time health warning method based on a time-aware autoencoder, comprising the following steps:

[0008] Step 1: Use a two-stage filtering method to fuse the time series data of heart rate and step count;

[0009] Step 1.1: Obtain real-time heart rate and step count time series data as shown in formula (1) through wearable devices;

[0010] H={(h1,t1),(h1,t2),…,(h n ,t n )},S={(S1,t1),(S1,t2),…,(s n ,t n )} (1)

[0011] Among them, h i Indicates the tth i Heart rate value at the moment, s i Indicates the tth i Step count at the moment;

[0012] Step 1.2: Filter the heart rate time series data at night time using the method shown in formula (2);

[0013]

[0014] in, is the filtered heart rate sequence; the heart rate sequence is used to filter out the night heart rate with zero step count in the night time period [t0, t1);

[0015] Step 1.3: Merge the filtered heart rate time series data and step count time series data using the method shown in formula (3);

[0016]

[0017] Where X is the heart rate sequence and step count statistics s are merged into the formed sequence;

[0018] Step 1.4: Resample the merged time series data using the method shown in formula (4);

[0019]

[0020] Among them, I k ={i|t k ≤t i <t k+1} represents the time interval [t k , t k+1 ) is the index set of all data points in |I k | is a set I k Δt is the time interval of resampling, t k =t0+kΔt, where k=0, 1, 2, ...;

[0021] Step 1.5: Obtain the resampled time series data in a stationary state, and then obtain the resting heart rate data;

[0022]

[0023]

[0024] Among them, s j is the resampled sequence X r The number of steps in the time series data, W is the size of the sliding window, is the combined heart rate and step count time series data. k is the resting heart rate data. k Remove abnormal heart rate data caused by activities, resting heart rate h k The corresponding t k Also used for health monitoring heart rate data.

[0025] Step 2: Generate a resting heart rate matrix using the maximum power of the resting heart rate power spectrum;

[0026] Step 2.1: Divide the resting heart rate data into equal time series lengths, and then obtain the number of columns n of the resting heart rate matrix as shown in formula (6);

[0027]

[0028] f max = arg max f P(f) (6)

[0029] Among them, t s represents the start time of resting heart rate, t e represents the end time of resting heart rate, N represents the number of data points in the resting heart rate time series data, and f max represents the frequency corresponding to the maximum power in the resting heart rate power spectrum, P(f) represents the power spectrum corresponding to frequency f, and T represents the maximum power f max The corresponding cycle, Indicates a floor operation.

[0030] Step 2.2: Generate the resting heart rate matrix (H, T) as shown in formula (7);

[0031]

[0032] Among them, h i，j Indicates that at t i，j The resting heart rate at the moment, (N+n)×n represents the dimension of the matrix (H, T);

[0033] Step 3: Realize real-time health warning based on resting heart rate matrix;

[0034] Step 3.1: Build a time-aware autoencoder.

[0035] Step 3.1.1: Extract the resting heart rate matrix (H, T) to obtain the resting heart rate data H and its corresponding time data T as shown in formula (8);

[0036]

[0037]

[0038] Among them, h i，j Indicates that at t i，j The resting heart rate at the moment, (N+n)×n represents the dimension of the matrix (H, T);

[0039] Step 3.1.2: Obtain the time interval matrix ΔT as shown in formula (9);

[0040]

[0041] Among them, ΔT is the time interval matrix, and Δt is the time interval in the matrix.

[0042] Step 3.1.3: Use the exponential decay method to obtain the time mapping of the resting heart rate data as shown in formula (10);

[0043]

[0044] Among them, t iand t0 represent the current and starting time points of the resting heart rate sequence, respectively, and k is the decay coefficient, which controls the decay speed;

[0045] Step 3.2: Input the resting heart rate time series data and the time interval into the time-aware encoder, and train the time awareness unit in the time-aware encoder in the manner shown in formula (11);

[0046]

[0047] Among them, C t-1 is the short-term state at time t-1, W e and B e are the weight matrix and bias vector of the encoder respectively, and ReLU(·) represents the ReLU nonlinear rectification function, i.e. f(x) = max(0, x). C t-1 -ReLU(W e C t-1 +B e ) represents long-term memory, Represents short-term memory with time weight. Combine the short-term memory with time weight and the long-term memory to obtain the adjusted short-term state of the previous time step The specific short-term state at time t is shown in formula (12);

[0048]

[0049] where * is the element-wise product, Γ u and Γ f They represent the update and forget gate signals at time t related to the resting heart rate h, respectively.

[0050] Step 3.3: Train the time-aware decoder and further reconstruct the resting heart rate using the method shown in formula (13);

[0051]

[0052] z t =f(W e , B e ;h t , t) = σ(W e h t +B e +V e t) (13)

[0053] Where f and g represent the time-aware encoder and decoder functions, respectively. t Represents the resting heart rate sequence data at time t. tIt represents the output of the time-aware encoder at time t and is also the input of the decoder. W represents the output of the time-aware decoder at time t, i.e., the reconstructed resting heart rate. e and B e are the weight matrix and bias vector of the encoder, W d and B d are the weight matrix and bias vector of the decoder respectively. V e is the time-dependent weight matrix. σ and φ are the activation functions in the encoder and decoder, respectively.

[0054] Step 3.4: Generate real-time health warning data using the original resting heart rate sequence and the reconstructed resting heart rate sequence through formula (14);

[0055]

[0056] Among them, θ represents the threshold value, represents the indicator function, which is specifically shown in formula (15):

[0057]

[0058] When e>θ, e-θ>0, so The value of is 1, which triggers a health alert. When e≤θ, e-θ≤0, so The value of is 0, and the heart rate and step count statistical time series are still collected; e represents the original resting heart rate h and the reconstructed resting heart rate The error loss is as shown in formula (16):

[0059]

[0060] Among them, δ is an adjustable hyperparameter;

[0061] The present invention discloses a real-time health warning system based on a time-aware autoencoder, which is used to implement the above method. The present invention discloses a real-time health warning system based on a time-aware autoencoder, which includes a heart rate and step time series data fusion module, a resting heart rate matrix generation module, and a real-time health warning module based on the resting heart rate matrix.

[0062] The heart rate and step count time series data fusion module is used to filter out noise caused by unhealthy conditions and obtain resting heart rate time series data;

[0063] The resting heart rate matrix generation module is used to convert the resting heart rate time series data into a resting heart rate matrix;

[0064] The real-time health warning module based on the resting heart rate matrix is ​​used to model the heart rate data and time data of the resting heart rate matrix and generate real-time health warning data.

[0065] Beneficial effects:

[0066] Compared with the prior art, it has the following beneficial effects:

[0067] The present invention proposes a two-stage filtering method to fuse heart rate and step count statistical data into a resting heart rate. The resting heart rate obtained by this method focuses on the heart rate of the wearer of the wearable device in a resting state at night.

[0068] The present invention innovatively constructs a personalized resting heart rate matrix. This personalized resting heart rate matrix plays an important role in health monitoring applications, especially considering that the physiological data collected by wearable devices have time period differences between individuals. By calculating the power spectrum of the resting heart rate time domain series, the present invention can capture the unique heart rate cycle pattern of each user, thereby greatly improving the accuracy of monitoring individual health status.

[0069] This paper proposes a global time interval analysis method and an exponential decay time interval mapping method, aiming to fully explore and integrate the time interval information in the resting heart rate time series. The resting heart rate time series and its corresponding time interval features are input into a time-aware autoencoder to effectively adapt to and accurately process non-equally spaced time series data, overcoming the limitations of traditional methods in analyzing such complex time series data.

[0070] The present invention proposes the construction of a resting heart rate error model based on dynamic thresholds. The resting heart rate error model can set threshold coefficients according to individual health needs to personalize the monitoring of health status data of wearers of wearable devices, thereby improving the practicality of the health warning model.

[0071] Therefore, the real-time health warning system proposed by the present invention can accurately monitor changes in user health status data. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of a real-time health warning method based on time-aware autoencoder;

[0073] Figure 2 The original heart rate sequence diagram obtained by the wearable device;

[0074] Figure 3 The original step count statistics sequence diagram obtained by the wearable device;

[0075] Figure 4 Resting heart rate graph after fusion of heart rate sequence and step count statistics;

[0076] Figure 5 Visualization of the convergence process of the loss value of the training set;

[0077] Figure 6 Visualization of the convergence process of the loss value of the validation set;

[0078] Figure 7 Comparison of the original and reconstructed sequences for the first sample in the resting heart rate matrix;

[0079] Figure 8 Comparison of the original and reconstructed sequences for the second sample in the resting heart rate matrix;

[0080] Fig. 9 Scatter plot of abnormal ratings for real-time health alerts of subjects. DETAILED DESCRIPTION

[0081] In order to better illustrate the purpose and advantages of the present invention, the invention is further described below in conjunction with the accompanying drawings and examples. It should be noted that the implementation of the present invention is not limited to the following embodiments, and any form of modification or change made to the present invention will fall within the protection scope of the present invention.

[0082] Example

[0083] like Figure 1 As shown, a real-time health warning method based on a time-aware autoencoder of the present invention is specifically implemented in the following steps:

[0084] Step 1: Use a two-stage filtering method to fuse the time series data of heart rate and step count;

[0085] When consumer wearable data is used for real-time health warning, it is often subject to data anomalies that are not caused by changes in health status, such as strenuous physical exercise. Therefore, it is necessary to design an efficient data fusion strategy to obtain time series data caused by changes in health status from the heart rate and step count statistics collected by consumer wearables.

[0086] Step 1.1: Obtain real-time heart rate and step count time series data as shown in formula (1) through wearable devices;

[0087] H={(76,Jun-23 00:00:00),(74,Jun-23 00:03:00),…,(66,Sept-0506:08:00)},

[0088] S = {(0, Jun-23 00:00:00), (0, Jun-23 00:01:00), ..., (0, Sept-05 23:59:00)} (1) In this embodiment, the original heart rate sequence and the original step count sequence are respectively as follows: Figure 2 and Figure 3shown.

[0089] Step 1.2: Nighttime heart rate data can usually avoid the impact of strenuous exercise during the day on health. Therefore, in the first stage, the heart rate sequence is filtered out to find the nighttime heart rate data with zero step count in the nighttime period [t0, t1). In this embodiment, the night time period [t0, t1) is defined as 0:00 am to 7:00 am. It can be shown as formula (2):

[0090]

[0091] Step 1.3: After getting the heart rate sequence After that, it is combined with the step count statistics as shown in formula (3):

[0092]

[0093] Where X is the heart rate sequence and step count statistics S are merged in the formed sequence.

[0094] Step 1.4: Resample the merged sequence and set the resampling time interval to Δt. The time point after resampling can be expressed as t k =t0+kΔt, where k=0,1,2,…. The resampled sequence X r It can be expressed as formula (4):

[0095]

[0096] Among them, I k ={i|t k ≤t i <t k+1} represents the time interval [t k , t k+1 ) is the index set of all data points in |I k | is a set I k In this embodiment, the resampling time interval is Δt=1min.

[0097] Step 1.5: Although the first stage filtering operation obtains X r The influence of daily activities on health monitoring during the daytime period is filtered out, but it is necessary to further confirm that the wearer of the wearable device is in a stationary state to truly reflect the wearer's health status. Therefore, the second stage of filtering is introduced to obtain the resting heart rate in the present invention. In this embodiment, the size of the sliding window W = 12, so this step can be expressed as formula (5) as follows:

[0098]

[0099]

[0100] Among them, s j is the resampled sequence X r The step count statistics in , is the sequence obtained by the second stage filtering operation that combines heart rate and step count statistics. k The resting heart rate h is the resting heart rate of the present invention. k Filters out heart rate increases caused by activity, which helps to sense potential health conditions. k The corresponding t k It is also used for health monitoring. In this embodiment, the resting heart rate h k like Figure 4 shown.

[0101] Step 2: Generate a resting heart rate matrix using the maximum power of the resting heart rate power spectrum;

[0102] The generation of the resting heart rate matrix depends on the duration of the periodic signal in the time series data. The generation of the resting heart rate matrix with a fixed period size destroys the periodic characteristics of the time series, because the periodic characteristics of the time series of wearable data vary from person to person.

[0103] Step 2.1: The number of columns n of the personalized resting heart rate matrix is ​​calculated according to formula (6):

[0104]

[0105] f max = arg max f P(f) (6)

[0106] Among them, t s represents the start time of resting heart rate, t e represents the end time of resting heart rate, N represents the number of data points in the resting heart rate time series data, and f max represents the frequency corresponding to the maximum power in the resting heart rate power spectrum, P(f) represents the power spectrum corresponding to frequency f. T represents the maximum power f max The corresponding cycle, Indicates a floor operation.

[0107] In this embodiment, the number of columns n of the personalized resting heart rate matrix used for model training is 2, which is obtained by the following calculation:

[0108]

[0109] After converting the timestamp to a time interval in seconds, the specific value can be calculated:

[0110]

[0111] Step 2.2: The present invention improves the traditional method of generating a resting heart rate matrix with a fixed cycle size to generate a personalized resting heart rate matrix. max The dimension of the obtained personalized resting heart rate matrix (H, T) with timestamp for model training is 86×2, as shown in formula (7):

[0112]

[0113] Where H is the preprocessed resting heart rate.

[0114] Step 3: Realize real-time health warning based on resting heart rate matrix;

[0115] Different from modeling only time series data, the present invention inputs resting heart rate time series data and time interval into a time-aware encoder. The autoencoder represents the inherent rules of normal resting heart rate data through learned low-dimensional features. When abnormal resting heart rate samples are input, the reconstructed resting heart rate has a larger error.

[0116] Step 3.1: Build a time-aware autoencoder.

[0117] Step 3.1.1: To construct the autoencoder, extract the resting heart rate data H and its corresponding time data T from the personalized resting heart rate matrix (H, T) defined above as shown in formula (8):

[0118]

[0119]

[0120] Step 3.1.2: Calculate the time interval matrix ΔT by equation (9) to describe the time relationship between resting heart rate data:

[0121]

[0122] Step 3.1.3: Calculate the time map using the exponential decay method as follows:

[0123]

[0124] Among them, t i and t0 represent the current and starting time points of the resting heart rate sequence respectively, and k is the attenuation coefficient, which controls the attenuation speed. In this embodiment, k = 1, then This time interval based on the entire time series can obtain the global temporal information of resting heart rate.

[0125] Step 3.2: Input the resting heart rate time series data and the time interval into the time-aware encoder. In this embodiment, the short-term state of the previous time step in the time-aware unit in the encoder It can be expressed as follows:

[0126]

[0127] Among them, C t-1 is the short-term state at time t-1, W e and B e are the weight matrix and bias vector of the encoder respectively, and ReLU(·) represents the ReLU nonlinear rectification function, i.e. f(x) = max(0, x). C t-1 -ReLU(W e C t-1 +B e ) represents long-term memory, Represents short-term memory with time weight. Combine the short-term memory with time weight and the long-term memory to obtain the adjusted short-term state of the previous time step

[0128] In this embodiment, the short-term state C of the encoder with time awareness at time t t for:

[0129]

[0130] where * is the element-wise product, Γ u and Γ f They represent the update and forget gate signals at time t related to the resting heart rate h, respectively.

[0131] Step 3.3: The time-aware encoder learns the resting heart rate data itself and fully models the timestamps of the resting heart rate data in the process of building the health warning model. The decoder part of the autoencoder is similar to the encoder part in structure and is responsible for reconstructing the encoded learning expression back to the resting heart rate data to support the subsequent error model construction. In this embodiment, for the first and second samples in the resting heart rate matrix, the time-aware autoencoder can be simplified by formula (13):

[0132]

[0133]

[0134]

[0135] in, Indicates time t i Resting heart rate series data on. represents the time-aware encoder time t i The output of the decoder is also the input of the decoder. Denotes the time-aware decoder time t i The output on is the reconstructed resting heart rate. and are the weight matrix and bias vector of the encoder, respectively. and are the weight matrix and bias vector of the decoder respectively. is the time-related weight matrix. σ and φ are the activation functions in the encoder and decoder, respectively. In this embodiment, the time-aware autoencoder is trained for 20 cycles, and the loss values ​​of the training set and the validation set change during the training process as follows: Figure 5 and Figure 6 In this embodiment, the original sequence and the reconstructed sequence of the first sample and the second sample in the resting heart rate matrix are as follows: Figure 7 and Figure 8 The original resting heart rate series is the pink solid line, and the reconstructed resting heart rate series is the blue dotted line.

[0136] Step 3.4: Generate real-time health warning data using the original resting heart rate sequence and the reconstructed resting heart rate sequence through formula (14);

[0137]

[0138] Among them, θ represents the threshold value, represents the indicator function, which is specifically shown in formula (15):

[0139]

[0140] When e>θ, e-θ>0, so The value of is 1, which triggers a health alert. When e≤θ, e-θ≤0, so The value of is 0, and the heart rate and step count statistical time series are still collected; e represents the original resting heart rate h and the reconstructed resting heart rate The error loss is as shown in formula (16):

[0141]

[0142] Among them, δ is an adjustable hyperparameter;

[0143] The threshold θ is calculated as follows:

[0144]

[0145] Wherein, M is the number of samples in the resting heart rate matrix generated in the second step. λ is a coefficient that can be set according to individual health parameters. In the construction of the error model, it is usually a 3-times standard deviation model, that is, λ is 3. When λ takes a larger value, it means that the threshold is higher and it is not easy to generate a health warning. When λ takes a smaller value, it means that the threshold is lower and it is easy to generate a health warning. In this embodiment:

[0146]

[0147] In this embodiment, λ=3. Then the threshold is as follows:

[0148] θ=0.1070+3×0.0670=0.3080

[0149] In this embodiment, the health warning generated by the resting heart rate error model is as follows: Fig. 9 The anomaly threshold is shown as a horizontal yellow dashed line. Anomaly scores above the threshold are shown as red diamonds, and anomaly scores below the threshold are shown as solid black dots. The self-reported symptom onset date is shown as a solid blue line, and the infectious period is marked in gray.

Claims

1. A real-time health warning method based on time-aware autoencoder, characterized by: The following steps are included: Step 1: Use a two-stage filtering method to fuse the time series data of heart rate and step count; Step 2: Generate a resting heart rate matrix using the maximum power of the resting heart rate power spectrum; Step 2.1: Divide the resting heart rate data into equal time series lengths, and then obtain the number of columns n of the resting heart rate matrix as shown in formula (6); f max =arg max f P(f) (6) Among them, t s represents the start time of resting heart rate, t e represents the end time of resting heart rate, N represents the number of data points in the resting heart rate time series data, and f max represents the frequency corresponding to the maximum power in the resting heart rate power spectrum, P(f) represents the power spectrum corresponding to frequency f, and T represents the maximum power f max The corresponding cycle, Indicates a round-down operation; Step 2.2: Generate the resting heart rate matrix (H, T) as shown in formula (7); Among them, h i,j Indicates that at t i,j The resting heart rate at the moment, (N+n)×n represents the dimension of the matrix (H,T); Step 3: Realize real-time health warning based on resting heart rate matrix; Step 3.1: Build a time-aware autoencoder. Step 3.2: Input the resting heart rate time series data and the time interval into the time-aware encoder, and train the time awareness unit in the time-aware encoder using the method shown in formula (11); Among them, C t-1 is the short-term state at time t-1, W e and B e are the weight matrix and bias vector of the encoder respectively, ReLU(·) represents the ReLU nonlinear rectification function, i.e. f(x)=max(0,x); C t-1 -ReLU(W e C t-1 +B e ) represents long-term memory, Represents short-term memory with time weight; combines short-term memory with time weight and long-term memory to obtain the adjusted short-term state of the previous time step The specific short-term state at time t is shown in formula (12); where * is the element-wise product, Γ u and Γ f They represent the update and forget gate signals at time t related to the resting heart rate h, respectively; Step 3.3: Train the time-aware decoder and further reconstruct the resting heart rate using the method shown in formula (13); z t =f(W e ,B e ;h t ,t)=σ(W e h t +B e +V e t) (13) Where f and g represent the time-aware encoder and decoder functions, respectively; h t represents the resting heart rate sequence data at time t; z t represents the output of the time-aware encoder at time t, and is also the input of the decoder; represents the output of the time-aware decoder at time t, i.e., the reconstructed resting heart rate; W e and B e are the weight matrix and bias vector of the encoder, W d and B d are the weight matrix and bias vector of the decoder respectively; V e is the time-dependent weight matrix; σ and φ are the activation functions in the encoder and decoder respectively; Step 3.4: Generate real-time health warning data using the original resting heart rate sequence and the reconstructed resting heart rate sequence through formula (14); Among them, θ represents the threshold value, represents the indicator function, which is specifically shown in formula (15): When e>θ, e-θ>0, so The value of is 1, which triggers a health alert; when e≤θ, e-θ≤0, so The value of is 0, and the heart rate and step count statistical time series are still collected; e represents the original resting heart rate h and the reconstructed resting heart rate The error loss is shown in formula (16): Here, δ is an adjustable hyperparameter.

2. A real-time health warning method based on a time-aware autoencoder as claimed in claim 1, characterized in that: Step 1 is implemented as follows: Step 1.1: Obtain real-time heart rate and step count time series data as shown in formula (1) through wearable devices; H={(h1,t1),(h1,t2),…,(h n ,t n )},S={(s1,t1),(s1,t2),…,(s n ,t n )} (1) Among them, h i Indicates the tth i Heart rate value at the moment, s i Indicates the tth i Step count at the moment; Step 1.2: Filter the heart rate time series data at night time using the method shown in formula (2); in, is the filtered heart rate sequence; the heart rate sequence is used to filter out the night heart rate with zero step count in the night time period [t0, t1); Step 1.3: Merge the filtered heart rate time series data and step count time series data using the method shown in formula (3); Where X is the heart rate sequence and the step count statistics S are merged into the formed sequence; Step 1.4: Resample the merged time series data using the method shown in formula (4); Among them, I k ={i|t k ≤t i <t k+1 } represents the time interval [t k ,t k+1 ) is the index set of all data points in |I k | is a set I k The size of; Δt is the time interval of resampling, t k =t0+kΔt, where k=0, 1, 2, ...; Step 1.5: Obtain the resampled time series data in a stationary state, and then obtain the resting heart rate data; Among them, s j is the resampled sequence X r The number of steps in the time series data, W is the size of the sliding window, is the combined heart rate and step count time series data; where h k is the resting heart rate data; resting heart rate h k Remove abnormal heart rate data caused by activities, resting heart rate h k The corresponding t k Also used for health monitoring heart rate data.

3. A real-time health warning method based on a time-aware autoencoder as claimed in claim 1, characterized in that: Step 3.1 is implemented as follows: Step 3.1.1: Extract the resting heart rate matrix (H, T) to obtain the resting heart rate data H and its corresponding time data T as shown in formula (8); Among them, h i,j Indicates that at t i,j The resting heart rate at the moment, (N+n)×n represents the dimension of the matrix (H,T); Step 3.1.2: Obtain the time interval matrix ΔT as shown in formula (9); Where ΔT is the time interval matrix, and Δt is the time interval in the matrix; Step 3.1.3: Use the exponential decay method to obtain the time mapping of the resting heart rate data as shown in formula (10); Among them, t i and t0 represent the current and starting time points of the resting heart rate sequence respectively, and k is the attenuation coefficient, which controls the attenuation speed.

4. A real-time health warning system based on a time-aware autoencoder implementing the method of claim 1, characterized in that: It includes a heart rate and step count time series data fusion module, a resting heart rate matrix generation module, and a real-time health warning module based on the resting heart rate matrix; The heart rate and step count time series data fusion module is used to filter out noise caused by unhealthy conditions and obtain resting heart rate time series data; The resting heart rate matrix generation module is used to convert the resting heart rate time series data into a resting heart rate matrix; The real-time health warning module based on the resting heart rate matrix is ​​used to model the heart rate data and time data of the resting heart rate matrix and generate real-time health warning data.

Images