Wireless heart rate monitoring and short message alarm method based on Arduino single-chip microcomputer

Abstract

The invention discloses a wireless heart rate monitoring and short message alarm method based on an Arduino single-chip microcomputer, and particularly relates to the technical field of heart rate monitoring, which comprises the following steps: acquiring environment and user activity information through a multi-source sensor, acquiring a heart rate data priority according to an environment risk score and user activity intensity, and sending the heart rate data priority to the Arduino single-chip microcomputer; future environment and activity conditions are predicted based on the long short-term memory network, sampling frequency and transmission strategies are adjusted in advance, prospective intervention on potential risks is achieved, practicability, accuracy and reliability of the heart rate monitoring system are improved, and the heart rate monitoring system is suitable for wide health monitoring scenes; when the heart rate abnormity is compared with the threshold value, multi-level alarm is triggered, the high-frequency sampling energy consumption of the sensor can be reduced, the data transmission path is evaluated and optimized in combination with the network reliability, the abnormity detection precision and stability are remarkably improved, and the problems that the abnormity is difficult to recognize in time and the power consumption is too high in a single heart rate sensor in a complex environment are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of heart rate monitoring. More specifically, the present invention relates to a wireless heart rate monitoring and SMS alarm method based on an Arduino single-chip microcomputer. Background Art

[0002] With the wide application of wearable devices and intelligent sensor technologies, health management based on heart rate monitoring has formed a relatively mature application framework in fields such as telemedicine, rehabilitation care, and sports monitoring. Traditional solutions generally use wristbands, chest straps, or ear clip devices to continuously collect the user's heart rate and transmit the data to the backend system for analysis and storage via wireless communication. Through heart rate monitoring, not only can a preliminary warning of health risks be made, but it can also assist medical institutions in remotely tracking and intervening in the user's physiological state. However, with the diversification of people's daily activity scenarios and the complex and changeable environmental conditions, how to fully exploit various sensing information to improve the perception ability of the user's state and environmental changes has become a key requirement for current technological development.

[0003] Existing heart rate monitoring systems still face many technical challenges in practical applications. Most systems do not fully consider the impact of environmental variables and user personal activities on heart rate, which may lead to misinterpretation of heart rate data, resulting in false alarms or missed alarms of abnormal heart rate conditions. In addition, the frequent data transmission and high-frequency data sampling requirements increase the device power consumption, limiting the continuous usage time of the device. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a wireless heart rate monitoring and SMS alarm method based on an Arduino single-chip microcomputer. By integrating sensors, it can continuously monitor the environment and the user's state, dynamically adjust the priority of heart rate data for heart rate monitoring according to the data, and automatically send an SMS alarm when an abnormal heart rate is detected, so as to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A wireless heart rate monitoring and SMS alarm method based on an Arduino single-chip microcomputer, comprising the following steps:

[0006] Collect environmental conditions and the user's state through a wearable intelligent device and transmit them to the Arduino single-chip microcomputer; the environmental conditions include temperature and humidity, harmful gas concentration, and the user's state includes the user's physical activity information and heart rate; the physical activity information is obtained through a three-axis acceleration sensor, and the heart rate data is obtained through signal amplification and adaptive filtering processing;

[0007] Analyze the collected environmental conditions and user status to obtain the environmental risk score and user activity intensity; according to the environmental risk score and user activity intensity, obtain the priority of heart rate data, and adjust the heart rate data transmission priority and sampling frequency of the heart rate monitoring data based on the priority of heart rate data; fuse the non-linear cumulative effect of gas concentration, temperature and humidity deviating from the reference value to obtain the environmental risk score; calculate the user activity intensity based on the standard deviation within the time window of the sum of the three-axis vectors of acceleration;

[0008] When a heart rate abnormality is detected, trigger a hierarchical response according to the environmental risk coefficient and activity intensity level;

[0009] Analyze the preprocessed heart rate data according to the priority of heart rate data. After comparing with the set threshold, if there is an abnormality, send a text message alarm through the wireless network module;

[0010] For the missing heart rate data, it is complemented in real time through the embedded LSTM model, and the complemented result is collaboratively verified for abnormalities with the inference result of the cloud model; map the heart rate, acceleration, and temperature data into a spatio-temporal graph structure, fuse multi-modal features through dynamic edge weights, and when potential heart rate abnormalities are detected, activate the graph neural network model deployed on the Arduino side and processed by pruning and quantization, and perform more complex multi-modal data analysis to improve the detection accuracy.

[0011] Preferably, when potential abnormalities are detected, activate the graph neural network model deployed on the Arduino side and processed by pruning and quantization, and perform more complex multi-modal data analysis to improve the detection accuracy; the logic is: when the multi-modal data (heart rate, acceleration, temperature) does not trigger the abnormality threshold, the Arduino runs a lightweight detection module, and only quickly judges through a preset threshold (such as the heart rate exceeding the static range), occupying extremely low computing resources; once the simplified detection finds an abnormality (such as a short-term heart rate exceeding the standard), immediately activate the pruned and quantized GNN model and perform the following in-depth analysis:

[0012] Spatio-temporal correlation mining: Use the graph attention mechanism to dynamically update the weights between multi-modal nodes, and analyze whether the heart rate abnormality is associated with cross-modal factors such as sudden changes in environmental temperature and sudden increase in exercise intensity;

[0013] Temporal pattern recognition: Aggregate the features within the time window through a gated recurrent unit (GRU) to identify whether the abnormality is occasional noise or a continuous risk;

[0014] Pruning and quantization processing: Delete the redundant connections with edge weights less than 0.1 in the graph neural network, reduce the number of parameters while retaining the key modal interaction paths (such as the strongly correlated edge between heart rate and exercise intensity); compress the model weights from 32-bit floating point to 8-bit integer, and reduce the memory occupancy to below 50KB to ensure that it can run under the limited resources of the Arduino.

[0015] Preferably, the wearable intelligent device is used to monitor environmental conditions and user status in real time, and is composed of a temperature and humidity sensor, a harmful gas detector, an accelerometer and a heart rate sensor; the sensors are installed on the wearable device to monitor environmental conditions, the physical activity information and heart rate of the user in real time, and transmit the data to the Arduino single-chip microcomputer; the heart rate data is processed by signal amplification and filtering.

[0016] Preferably, the process of obtaining the priority of the heart rate data is as follows:

[0017] Integrate the gas concentration, temperature change and humidity to calculate the environmental risk score index;

[0018] Use the acceleration data to calculate the user activity intensity;

[0019] According to the calculated environmental risk score and user activity intensity, calculate the priority of the heart rate data; dynamic adjustment is implemented in the Arduino single-chip microcomputer programming, and the priority of the heart rate data f is calculated through the following model s ,

[0020]

[0021] R e = w g ·log(1 + C g ) + w t ·|T - T ref | + w h ·H

[0022]

[0023] where f0 represents the basic heart rate data priority, which is set according to the design requirements of the device; HR_avg represents the average measured heart rate in the time window, HR tar represents the preset heart rate value, and the preset heart rate value is adjusted based on the adaptive threshold;

[0024] α represents the environmental risk score weight factor, which is used to adjust the influence degree of environmental risk on the heart rate data priority;

[0025] β represents the user activity intensity weight factor, which is used to adjust the influence degree of user activity on the heart rate data priority;

[0026] γ represents the power consumption factor, which represents the influence degree of the power consumption mode on the heart rate data priority;

[0027] R e represents the environmental risk score, C g represents the gas concentration, T represents the real-time temperature, T refis the reference temperature, H represents humidity, and w g , w t , w h respectively represent the weight factors of each item;

[0028] A u represents the user activity intensity, and a x , a y , a z represent the readings of the acceleration sensor in three axes, N represents the number of sampling points in the sliding window; P c represents the power consumption mode.

[0029] The power consumption mode is a power consumption mode dynamically determined based on the remaining battery power of the device, with a value range of 1 - 3. The higher the power consumption level, the more it indicates that the device enters the extreme power-saving state, and the corresponding heart rate data has a lower priority;

[0030] When the remaining battery power of the device ≤ 15%, the system automatically activates the power consumption mode 3. At this time, the device enters the extreme power-saving state, and the specific manifestations are:

[0031] Sensor sampling frequency degradation: The sampling interval of the heart rate sensor is extended from 1 second / time in the normal mode to 10 seconds / time, and the driving current of the optical module is reduced by 50%, sacrificing some data accuracy to give priority to ensuring battery life;

[0032] Calculation task dynamic offloading: Pause the real-time operation of the local LSTM prediction model, only retain the cache of the original heart rate data, and trigger delayed analysis after connecting to the charging device;

[0033] Communication protocol simplification: Turn off the Bluetooth BLE broadcast function, switch the alarm text message sending channel to NB-IoT low-frequency narrowband transmission, and compress the single data transmission volume to within 200 bytes.

[0034] Preferably, the acquisition mechanism of the heart rate abnormality probability is:

[0035] After the original photoplethysmogram signal is amplified and denoised by band-pass filtering, the heart rate variability index and the instantaneous heart rate trend are extracted; the normalized environmental parameters, user activity intensity and heart rate time series characteristics are aligned in time steps to form a multi-dimensional feature matrix;

[0036] The state prediction model receives the fusion features of the past T time steps, encodes the global state through the fully connected layer, and the global state at least includes the multiple of the gas concentration exceeding the standard and the trend of the user activity intensity. The local time series pattern is extracted through the convolutional layer, and the local time series pattern at least includes heart rate mutation and abnormal trend information of the beat interval;

[0037] The contribution degrees of the environmental conditions, user activity characteristics and heart rate characteristics are dynamically weighted through the cross-attention mechanism, and the heart rate abnormality probability is output;

[0038] If the confidence level of the heart rate abnormality probability meets the requirements, directly set the priority of the heart rate data based on the heart rate abnormality probability, and preempt the transmission bandwidth for the heart rate data with a high heart rate abnormality probability; the heart rate abnormality probability output by the state prediction model is a probability vector of discrete abnormality levels, and the confidence level of the heart rate abnormality probability is calculated through multi-class information entropy or differential entropy.

[0039] Preferably, the process of obtaining the priority of the heart rate data further includes: using a long short-term memory network to build a state prediction model, which is used to predict future environmental conditions and user states; training the state prediction model with historical data, the input includes past environmental parameters and user physical activity information, and the output is the predicted environmental risk score and the predicted user activity intensity; based on these prediction results, adjust the priority of the heart rate data monitored in advance.

[0040] Preferably, the calculation formula for the priority of the heart rate data based on the state prediction model is:

[0041]

[0042] where is the predicted environmental risk score, is the predicted user activity intensity.

[0043] Preferably, it further includes: triggering different levels of alarm modes according to the relationship between the heart rate abnormality index and the threshold, including:

[0044] Low-level alarm: When the heart rate abnormality index does not deviate significantly from the target range, prompt the user to pay attention to the heart rate change through vibration;

[0045] Medium-level alarm: When the heart rate abnormality index continuously deviates from the target range, indicating that it reaches a medium risk, remind the user of possible health hazards through sound and light alarm;

[0046] High-level alarm: When the heart rate abnormality index deviates from the target range and the environmental risk score or user activity intensity reaches the dangerous threshold, trigger a text message or remote push alarm, and at the same time increase the data collection and the priority of the heart rate data, support real-time monitoring and external intervention, and provide a real-time health report, including heart rate trend, environmental risk assessment and activity intensity analysis.

[0047] Preferably, the calculation formula for the heart rate risk index is:

[0048]

[0049] where R trend represents the heart rate trend change rate, H var represents the heart rate fluctuation amplitude within the time window, T devIt represents the cumulative time when the heart rate deviates from the target value, and k represents the trend adjustment coefficient.

[0050] Preferably, it further includes a heart rate data transmission monitoring step, including:

[0051] Before the heart rate data is transmitted, network environment perception is performed to evaluate the signal strength, network congestion degree, and packet loss rate in real time, and the network status is collected in real time, including signal strength, current network congestion degree, packet loss rate, and round-trip delay; before the transmission starts (or periodic detection during continuous transmission), the network status is packaged and sent to the local (or gateway / cloud) for real-time evaluation; the network reliability evaluation model is used to analyze historical and real-time network data, and the input features include signal strength, historical packet loss rate, and network response time, and the network stability coefficient of each transmission path is output;

[0052] Compare the network stability coefficients of several optional transmission paths; select the path with the highest network stability coefficient for heart rate data transmission;

[0053] Compress and prioritize the heart rate data to reduce the delay caused by large data volumes;

[0054] After the data is successfully sent, use the end-to-end confirmation protocol to ensure that each data packet is confirmed; if the confirmation fails, trigger the retransmission mechanism; when a heart rate abnormality or data transmission failure is detected, immediately trigger an alarm.

[0055] Preferably, the method for obtaining the network stability coefficient is:

[0056] Construct a historical database: record the network status in different time periods and different geographical environments, as well as the actual transmission success rate and delay distribution;

[0057] Collect real-time data: the network status at the current moment and some short-term historical statistical indicators;

[0058] Feature extraction: perform missing value processing, outlier removal, and normalization preprocessing on the collected real-time data, and extract the statistical features within the window;

[0059] Model training: Select a random forest, gradient boosting machine, or lightweight neural network as the initial model; match the historical network status with the corresponding transmission success rate and network stability label; use regression metrics or classification metrics, combined with cross-validation, to evaluate the generalization performance of the model; perform model optimization through grid search and Bayesian optimization methods, and select the optimal parameters to obtain accurate and stable predictions;

[0060] On the Arduino side or the gateway / cloud side, obtain the latest network metrics in real time and input them into the deployed network reliability evaluation model, which outputs a value between 0 and 1, called the "network stability coefficient". The closer the value is to 1, the higher the reliability of the current network for transmitting heart rate data, meaning lower latency, fewer packet losses, and a more stable connection.

[0061] Preferably, predict the network stability coefficient before data transmission or during the sending interval (such as every few seconds to minutes) to balance real-time performance and resource consumption; if the network state changes drastically, shorten the inference cycle; if the network environment is relatively stable, reduce the prediction frequency of the network stability coefficient to reduce overhead.

[0062] Preferably, in medical-related scenarios, when collecting patients' physiological data and network environment data, relevant privacy regulations need to be complied with; encrypt the transmission path to ensure the security of heart rate data.

[0063] Preferably, the method further includes steps for sparse heart rate data completion and anomaly monitoring, including the following steps:

[0064] Collect heart rate time series data with missing segments, mark the missing positions, and extract heart rate variability features as physiological constraint conditions; the physiological constraint conditions include: obtaining the standard deviation of heart rate within a preset time window, mapping it to the [-1, 1] interval through the hyperbolic tangent function to generate a first sub-factor; performing a time integral on the absolute value of the heart rate change rate, taking the logarithm operation, and multiplying it by an age-related trend sensitivity coefficient to generate a second sub-factor; adding the first sub-factor and the second sub-factor as the output value of the physiological state factor; the physiological state factor is used to constrain the quantization index of the data completion process, and through the physiological characteristics of heart rate variability, ensure that the completed heart rate data conforms to the true physiological laws of the human body;

[0065] Construct a generator model based on bidirectional LSTM and multi-head attention mechanism, input sparse heart rate data, and output the completed continuous sequence, where the attention mechanism focuses on the temporal correlation before and after the missing segment;

[0066] Design a discriminator model to judge the authenticity of the completed data through a convolutional network and calculate the heart rate variability feature constraint loss to verify physiological rationality;

[0067] Adopt a hybrid strategy combining pre-training and adversarial training to optimize the generator. In the pre-training stage, simulate random missing and minimize the mean square error. In the adversarial training stage, dynamically adjust the loss weights of the generator and the discriminator;

[0068] Compress the trained generator into a lightweight LSTM model through knowledge distillation and deploy it to the Arduino side to achieve real-time missing completion. The completed data is linked with the inference results of the complex model on the cloud side to trigger anomaly grading alarms.

[0069] Preferably, the method further includes a health risk warning step based on multi-modal heterogeneous data fusion, including the following steps:

[0070] Synchronize the timestamps and align the sampling frequencies of the data from the heart rate sensor, acceleration sensor, and temperature sensor, and unify the low-frequency data to the high-frequency sampling rate through interpolation;

[0071] Map the multi-modal data into a graph structure, where each node represents a physiological or environmental parameter, and the edge weights between nodes are initialized based on the statistical correlation between modalities and dynamically updated through a spatio-temporal graph attention layer, specifically including:

[0072] Concatenate the feature representations of nodes i and j at time t;

[0073] Perform a linear transformation on the concatenated features with learnable parameters and apply an activation function;

[0074] Perform Softmax normalization on the calculation results to obtain the updated edge weights;

[0075] Use a gated recurrent unit to perform temporal aggregation on the dynamically updated multi-modal features to generate a fused health status code;

[0076] Prune and perform 8-bit fixed-point quantization processing on the graph neural network model, and deploy it to the Arduino side to perform hierarchical inference: run simplified threshold detection under normal conditions, and activate the complete model for in-depth analysis when an anomaly is detected;

[0077] Input the fused features into a random forest classifier, output low-risk, medium-risk, and high-risk levels, generate an alarm message in combination with semantic compression encoding, and push it to a preset terminal through a short message gateway.

[0078] Preferably, the pruning process is specifically: delete the redundant connections in the graph neural network with edge weights less than 0.1; the abnormal types of the semantic compression encoding include at least one of heart rate abnormality, environmental abnormality, and exercise intensity abnormality.

[0079] The technical effects and advantages of the present invention:

[0080] (1) The wireless heart rate monitoring and SMS alarm method based on Arduino single-chip microcomputer provided by the present invention acquires environmental conditions and user activity information by integrating multiple sensors, accurately evaluates environmental risks and user status in complex scenarios, and avoids the problem that a single heart rate sensor cannot comprehensively identify abnormalities in a high-noise or variable environment; with the help of dynamic priority scheduling and long short-term memory network state prediction model, the sampling and transmission frequency of heart rate data are prospectively adjusted, effectively reducing the energy consumption caused by invalid high-frequency sampling while promptly responding to potential risks.

[0081] (2) The wireless heart rate monitoring and SMS alarm method based on Arduino single-chip microcomputer provided by the present invention integrates network reliability evaluation and multi-level alarm mechanism in the heart rate monitoring and SMS alarm scheme, realizes real-time selection of different transmission paths and data compression confirmation, not only ensures the safe and stable transmission of key physiological signals during network fluctuations, but also significantly improves the adaptability and reliability of the system in sudden environmental changes and drastic fluctuations of user status. Brief Description of the Drawings

[0082] Figure 1 It is a flowchart of the wireless heart rate monitoring and SMS alarm method of the present invention.

[0083] Figure 2 It is a block diagram of the wearable intelligent device of the present invention. Detailed Embodiments

[0084] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0085] At the same time, it should be understood that for the convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship.

[0086] The following description of at least one exemplary embodiment is merely illustrative and in no way limits the application or use of the present application.

[0087] Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods, and devices should be considered as part of the specification.

[0088] Embodiment 1, referring to Figure 1 the flowchart of the wireless heart rate monitoring and SMS alarm method, the present invention provides as Figure 1A wireless heart rate monitoring and SMS alarm method based on Arduino single-chip microcomputer is shown as follows, including the following steps:

[0089] Step 1, environmental and user status monitoring: Collect environmental conditions and user status through a wearable intelligent device and transmit them to the Arduino single-chip microcomputer; the environmental conditions include temperature and humidity, harmful gas concentration, and the user status includes the user's physical activity information and heart rate;

[0090] Step 2, preprocessing operation: Denoise and normalize the heart rate data, use the advanced wavelet transform method to effectively remove high-frequency noise and maintain the essential characteristics of the heart rate data; apply the Z-score method to standardize the heart rate data to ensure the consistency of data analysis;

[0091] Step 3, analyze the preprocessed heart rate data, compare it with the set threshold, and if there is an abnormality, send an SMS alarm through the wireless network module.

[0092] In the embodiment of the present invention, it needs to be further explained that referring to Figure 2 the structural block diagram of the wearable intelligent device, the wearable intelligent device is used to monitor environmental conditions and user status in real time, and is composed of a temperature and humidity sensor, a harmful gas detector, an accelerometer and a heart rate sensor; the sensors are installed on the wearable device to monitor environmental conditions, the user's physical activity information and heart rate in real time, and transmit the data to the Arduino single-chip microcomputer; the heart rate data is processed by signal amplification and filtering; the Arduino single-chip microcomputer is connected to the data analysis platform, and the data analysis platform is connected to the feedback alarm device. The data analysis platform is responsible for receiving the data transmitted from the Arduino single-chip microcomputer, generating an environmental risk score and user activity intensity, and adjusting the priority of the heart rate data accordingly; the feedback alarm device is used to provide instant feedback to the user when detecting abnormal heart rate or other key health indicators. The feedback can be sound, light signal or vibration, and the feedback method will be different according to the preset alarm level.

[0093] In the embodiment of the present invention, it needs to be further explained that the environmental risk score and user activity intensity are obtained by analyzing the collected environmental conditions and user status; according to the environmental risk score and user activity intensity, the priority of the heart rate data is obtained, and the heart rate data transmission priority and / or sampling frequency of the heart rate monitoring data are adjusted based on the priority of the heart rate data.

[0094] Explanation: Heart rate data with high priority is preferentially sent to the monitoring center or medical service provider, which is particularly important for a real-time health monitoring system, especially in the case where the user's health condition may change rapidly; based on the priority of the heart rate data, the sampling frequency is increased to monitor the heart rate more densely, or the sampling frequency is reduced in the case of low priority to save resources.

[0095] In the embodiments of the present invention, it needs to be further explained that the process of obtaining the priority of the heart rate data is as follows:

[0096] Calculate the environmental risk score index by integrating the gas concentration, temperature change and humidity;

[0097] Calculate the user activity intensity using the acceleration data;

[0098] Calculate the priority of the heart rate data according to the calculated environmental risk score and user activity intensity; implement dynamic adjustment in Arduino single-chip microcomputer programming, for example: when the environmental risk or user activity intensity exceeds a certain threshold, increase the priority of the heart rate data; otherwise, decrease the priority of the heart rate data;

[0099] The priority of the heart rate data is calculated through the following model

[0100]

[0101] R e = w g ·log(1 + C g ) + w t ·|T - T ref | + w h ·H

[0102]

[0103] where f0 represents the basic heart rate data priority, which is set according to the design requirements of the device; HR_avg represents the average measured heart rate in the time window, HR tar represents the preset heart rate value, and the preset heart rate value is adjusted based on the adaptive threshold; calculate the average value of the heart rate by setting a time window to eliminate the influence of short-term fluctuations, so as to obtain more stable and representative heart rate data;

[0104] α represents the environmental risk score weight factor, which is used to adjust the influence degree of the environmental risk on the heart rate data priority;

[0105] β represents the user activity intensity weight factor, which is used to adjust the influence degree of the user activity on the heart rate data priority;

[0106] γ represents the power consumption factor, which represents the influence degree of the power consumption mode on the heart rate data priority;

[0107] R e represents the environmental risk score, C g represents the gas concentration (such as PM2.5 or other index concentrations), T represents the real-time temperature, T ref is the reference temperature, H represents the humidity, wg , w t , w h respectively represent various weight factors;

[0108] A u represents the user activity intensity, a x , a y , a z represents the readings of the acceleration sensor in three axes, N represents the number of sampling points in the sliding window; P c represents the power consumption mode.

[0109] In the embodiments of the present invention, it should be further explained that the process of obtaining the heart rate data priority further includes: building a state prediction model using a long short-term memory network, and the state prediction model is used to predict future environmental conditions and user states; training the state prediction model using historical data, the input includes past environmental parameters and user physical activity information, and the output is the predicted environmental risk score and the predicted user activity intensity; based on these prediction results, adjusting the heart rate data priority of heart rate monitoring in advance to respond to potential risks or increased activities in advance, including the following steps:

[0110] Step 101, data preparation and preprocessing:

[0111] Extract environmental parameters and user physical activity information from the historical data stored in the system, and perform necessary data cleaning and preprocessing, such as filling missing values, data standardization, etc., to meet the input requirements of the long short-term memory network model;

[0112] Furthermore, segment the historical data, perform short-time Fourier transform on each time window using the Hanning window method, and extract the real and imaginary states in the frequency domain; then use linear embedding (similar to the formula in the attachment) to fuse the time-domain and frequency-domain features to form a richer feature representation; it can capture periodic or instantaneous dynamic changes, improve the model's perception ability of subtle state changes, and thus provide higher-quality input for subsequent predictions;

[0113] Step 102, initialization and training of the state prediction model:

[0114] Initialize the state prediction model, set the number of input nodes, hidden layers, output nodes, and training parameters (such as learning rate, batch size, number of iterations); divide the historical data set into a training set and a validation set, use the training set to train the LSTM model, and at the same time use the validation set to adjust the model parameters to prevent overfitting;

[0115] Step 103, real-time prediction and adjustment of heart rate data priority: Real-time input the latest environmental and user state data into the trained state prediction model, and obtain the prediction results, that is, the predicted environmental risk score and the predicted user activity intensity;

[0116] Furthermore, based on the LSTM structure, a cross-weight attention layer based on the frequency domain is added:

[0117] Using the real and imaginary states in the frequency domain of the input data, query, key, and value matrices are constructed respectively;

[0118] The multi-head attention mechanism (such as the cross-attention formula described in the appendix) is used to achieve information exchange between the real state enhancing the imaginary state and the imaginary state enhancing the real state;

[0119] The output attention-weighted features are concatenated or fused with the original LSTM hidden layer output as the final prediction features;

[0120] Furthermore, in the real-time prediction stage, a multi-scale information fusion device is introduced:

[0121] Perform multi-scale convolution processing on the features output by the attention module, and use convolution kernels of different sizes to extract local (short-term) and global (long-term) features;

[0122] The multi-scale features are formed into a joint representation through max pooling and concatenation operations;

[0123] Combined with the latest online data, an incremental or online update algorithm is used to fine-tune the model parameters to cope with environmental changes;

[0124] Step 104. According to the model prediction results, adjust the heart rate data priority of the heart rate sensor according to the predicted environmental risk score and the predicted user activity intensity.

[0125] In the embodiments of the present invention, it needs to be further explained that if the credibility of the heart rate abnormality probability does not meet the requirements, the calculation formula for the heart rate data priority based on the state prediction model is:

[0126]

[0127] where, is the predicted environmental risk score, which is predicted by a long short-term memory network (LSTM) based on historical environmental parameters (gas concentration, temperature, humidity, etc.); is the predicted user activity intensity, which is predicted by LSTM based on historical acceleration data.

[0128] In the embodiments of the present invention, it needs to be further explained that the energy consumption mode of the device is automatically adjusted according to the monitoring urgency and environmental risk; for example, when the environment is stable and the user activity is low, reduce the data collection and processing frequency and reduce the number of wireless transmissions to extend the battery life; the power consumption mode values satisfy the following:

[0129]

[0130] Among them, R low represents the minimum value of the environmental risk score, and R high represents the maximum value of the environmental risk score; A low represents the minimum value of the user activity intensity, and A high represents the maximum value of the user activity intensity;

[0131] The low-power mode P c = 1 indicates that it is applicable when the environment is stable and the user activity is low;

[0132] The medium-power mode P c = 2 indicates that it is applicable when the environment has a slight risk or the user activity is moderate;

[0133] The high-power mode P c = 3 indicates that it is applicable when the environmental risk is high or the user activity is intense.

[0134] It should be further explained in the embodiments of the present invention that a multi-level alarm mechanism is further included, and different levels of alarm modes are triggered according to the relationship between the heart rate abnormality index and the threshold, including:

[0135] Low-level alarm: When the heart rate abnormality index does not deviate seriously from the target range, the user is reminded to pay attention to the heart rate change through vibration;

[0136] Medium-level alarm: When the heart rate abnormality index continuously deviates from the target range, indicating that a medium risk is reached, the user is reminded of possible health hazards through audible and visual alarms (such as LED flashing or sound prompts);

[0137] High-level alarm: When the heart rate abnormality index deviates from the target range and the environmental risk score or the user activity intensity reaches the danger threshold, a text message or remote push alarm is triggered, and at the same time, the data collection and the priority of the heart rate data are increased to support real-time monitoring and external intervention, and a real-time health report is provided, including heart rate trend, environmental risk assessment and activity intensity analysis.

[0138] It should be further explained in the embodiments of the present invention that the calculation formula of the heart rate risk index is:

[0139]

[0140] Among them, R trend represents the heart rate trend change rate, H var represents the heart rate fluctuation amplitude within the time window, T dev represents the cumulative time when the heart rate deviates from the target value, and k represents the trend adjustment coefficient.

[0141] Explanation, exp(|HR_avg - HR tar|) is used to amplify the impact of heart rate deviation values, making the formula more sensitive when dealing with large deviation values and enabling quick response to potential health risks; cosh(R trend ) is used to smooth out rapid changes in the heart rate trend and prevent short-term fluctuations from overly affecting the index calculation; ln(1 + H var ) is used to smooth the amplitude of fluctuations, ensuring that small fluctuations have less impact and large fluctuations have a significant impact; arctan(T dev ) is used to limit the impact of cumulative time within a controllable range and prevent misjudgment caused by slightly abnormal conditions accumulated over a long period.

[0142] Summary: In the embodiments of the present invention, by integrating multiple sensors to obtain environmental risk and user activity information, and using dynamic priority scheduling and adaptive sampling strategies, precise monitoring of heart rate and effective energy consumption management are achieved; on the one hand, through comprehensive analysis of multi-source data, the state of the user can be comprehensively captured in a complex environment and the recognition accuracy of heart rate abnormalities can be improved; on the other hand, for the problem of high-frequency sampling of sensors, a hierarchical transmission and power consumption optimization mechanism is adopted, which significantly reduces energy consumption while ensuring the real-time nature of key data, effectively solving the technical problems that a single heart rate sensor cannot fully identify environmental and activity information and that continuous high-frequency sampling leads to excessive energy consumption, and significantly improving the reliability and battery life performance of the system.

[0143] Embodiment 2. The difference between the embodiment of the present invention and Embodiment 1 is that the present invention provides a wireless heart rate monitoring and SMS alarm method based on an Arduino single-chip microcomputer, which further includes a heart rate data transmission monitoring step, including:

[0144] Before the heart rate data is transmitted, network environment perception is carried out to real-time evaluate the signal strength, network congestion degree, and packet loss rate, and the network status is real-time collected, including signal strength, current network congestion degree, packet loss rate, and round-trip delay; before the transmission starts (or periodic detection during continuous transmission), the network status is packaged and sent to the local (or gateway / cloud) for real-time evaluation; a network reliability evaluation model is used to analyze historical and real-time network data, and the input features include signal strength, historical packet loss rate, and network response time, and the network stability coefficient of each transmission path is output;

[0145] The network stability coefficients of several optional transmission paths (such as different operators, different base stations, or different network systems) are compared; the path with the highest network stability coefficient is selected for the transmission of heart rate data;

[0146] The heart rate data is compressed and priority sorted to reduce the delay caused by a large amount of data;

[0147] After the data is successfully sent, an end-to-end confirmation protocol is adopted to ensure that each data packet is confirmed; if the confirmation fails, a retransmission mechanism is triggered; if a heart rate abnormality or data transmission failure is detected, an alarm is immediately triggered.

[0148] In the embodiments of the present invention, it needs to be further explained that the method for obtaining the network stability coefficient is as follows:

[0149] Construct a historical database: record the network status under different time periods and different geographical environments, as well as the actual transmission success rate and delay distribution;

[0150] Collect real-time data: the network status at the current moment (such as signal strength, network congestion, packet loss rate, RTT, bandwidth, jitter), and statistical indicators of part of the short-term history (such as the recent few minutes or recent transmissions);

[0151] Feature extraction: perform missing value processing, outlier removal, and normalization preprocessing on the collected real-time data, and extract statistical features (mean, variance, maximum value, minimum value, quantile) within the window;

[0152] Model training: select a random forest, gradient boosting machine, or lightweight neural network as the initial model; match the historical network status with the corresponding transmission success rate and network stability label; use regression metrics or classification metrics, combined with cross-validation, to evaluate the generalization performance of the model; perform model optimization through grid search and Bayesian optimization methods, and select the optimal parameters to obtain accurate and stable predictions;

[0153] On the Arduino side or gateway / cloud, obtain the latest network metrics in real time, input them into the deployed network reliability assessment model, and output a value between 0 and 1, called the "network stability coefficient". The closer the value is to 1, the higher the reliability of the current network for transmitting heart rate data, which means lower latency, fewer packet losses, and a more stable connection.

[0154] In the embodiments of the present invention, it needs to be further explained that a network stability coefficient prediction is performed before data transmission or during the sending interval (such as every few seconds to minutes) to balance real-time performance and resource consumption; if the network status changes violently, shorten the inference cycle; if the network environment is relatively stable, reduce the network stability coefficient prediction frequency to reduce overhead;

[0155] In the embodiments of the present invention, it needs to be further explained that in medical-related scenarios, when collecting patients' physiological data and network environment data, relevant privacy regulations need to be complied with; the transmission path is encrypted to ensure the security of heart rate data.

[0156] In summary, in the embodiments of the present invention, by adding a network stability coefficient evaluation step in wireless heart rate monitoring and SMS alarm, by collecting and analyzing multi-dimensional indicators such as signal strength, network congestion, and packet loss rate in real time, a comprehensive score in the range of 0 to 1 is provided for multiple optional transmission paths; combined with data compression, priority sorting, and end-to-end confirmation protocol, it can effectively reduce the delay and packet loss risks brought by large amounts of data, and switch to the optimal path in a timely manner when the network state fluctuates violently, significantly improving the reliability and real-time performance of heart rate data transmission. Compared with the existing methods that only rely on a single routing strategy, this solution can significantly reduce the ineffective energy consumption in complex environments, ensure the integrity of key physiological data and remote alarm in the first time, thereby greatly enhancing the stability and adaptability of the system.

[0157] In a possible embodiment, the method further includes a sparse heart rate data completion and anomaly monitoring step, including the following steps:

[0158] Collect heart rate time series data with missing segments, mark the missing positions and extract heart rate variability features as physiological constraint conditions; the physiological constraint conditions include: obtaining the standard deviation of heart rate within a preset time window, mapping it to the interval [-1, 1] through the hyperbolic tangent function to generate a first sub-factor; performing time integration on the absolute value of the heart rate change rate, taking the logarithm operation and multiplying by an age-related trend sensitivity coefficient to generate a second sub-factor; adding the first sub-factor and the second sub-factor as the output value of the physiological state factor; the physiological state factor is used to constrain the quantization index of the data completion process, and through the physiological characteristics of heart rate variability, ensure that the completed heart rate data conforms to the real physiological laws of the human body;

[0159] Construct a generator model based on bidirectional LSTM and multi-head attention mechanism, input sparse heart rate data, and output the completed continuous sequence, where the attention mechanism focuses on the temporal correlation before and after the missing segment;

[0160] Design a discriminator model, judge the authenticity of the completed data through a convolutional network, and calculate the heart rate variability feature constraint loss to verify the physiological rationality;

[0161] Adopt a hybrid strategy combining pre-training and adversarial training to optimize the generator. In the pre-training stage, simulate random missing and minimize the mean square error. In the adversarial training stage, dynamically adjust the loss weights of the generator and the discriminator;

[0162] Compress the trained generator into a lightweight LSTM model through knowledge distillation, deploy it to the Arduino side to achieve real-time missing completion, and link the completed data with the inference results of the complex model in the cloud to trigger anomaly classification alarm;

[0163] In a possible embodiment, the method further includes a health risk warning step based on multi-modal heterogeneous data fusion, including the following steps:

[0164] Synchronize the timestamps and align the sampling frequencies of the data from the heart rate sensor, acceleration sensor, and temperature sensor, and unify the low-frequency data to the high-frequency sampling rate through interpolation;

[0165] Map the multi-modal data into a graph structure, where each node represents a physiological or environmental parameter, the edge weights between nodes are initialized based on the statistical correlation between modalities, and the weights are dynamically updated through a spatio-temporal graph attention layer, specifically including:

[0166] Concatenate the feature representations of node i and node j at time t,

[0167] Perform a linear transformation on the concatenated features with learnable parameters and apply an activation function,

[0168] Perform Softmax normalization on the calculation result to obtain the updated edge weights;

[0169] Adopt a gated recurrent unit to perform temporal aggregation on the dynamically updated multi-modal features to generate a fused health status encoding;

[0170] Prune and perform 8-bit fixed-point quantization on the graph neural network model, compress the model size to less than 50KB, and deploy it to the Arduino side to perform hierarchical inference: run a simplified threshold detection under normal conditions, and activate the complete model for in-depth analysis when an anomaly is detected;

[0171] Input the fused features into a random forest classifier, output low-risk, medium-risk, and high-risk levels, generate an alarm message in combination with semantic compression encoding, the format of the alarm message is "abnormal type + duration + intensity", and push it to a preset terminal through a short message gateway.

[0172] Furthermore, the pruning process is specifically: delete the redundant connections in the graph neural network with edge weights less than 0.1; the abnormal types of the semantic compression encoding include at least one of heart rate abnormality, environmental abnormality, and exercise intensity abnormality.

[0173] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A wireless heart rate monitoring and SMS alarm method based on Arduino microcontroller, characterized in that: include: Environmental conditions and user status are collected through wearable smart devices and transmitted to the Arduino microcontroller; the environmental conditions include temperature, humidity, and harmful gas concentrations, and the user status includes the user's physical activity information and heart rate; the physical activity information is obtained through a three-axis acceleration sensor, and the heart rate data is obtained through signal amplification and adaptive filtering; Analyze the collected environmental conditions and user status to obtain the environmental risk score and user activity intensity; obtain the heart rate data priority based on the environmental risk score and user activity intensity, and adjust the heart rate data transmission priority and sampling frequency of the heart rate monitoring data based on the heart rate data priority; integrate the nonlinear cumulative effect of gas concentration, temperature and humidity deviating from the reference value to obtain the environmental risk score; Calculate the user activity intensity based on the standard deviation of the acceleration three-axis vector sum within the time window; When an abnormal heart rate is detected, a graded response is triggered based on the environmental risk factor and activity intensity level; The pre-processed heart rate data is analyzed according to the heart rate data priority, and after comparing with the set threshold, if there is any abnormality, a text message alarm is sent through the wireless network module; For missing heart rate data, the embedded LSTM model is used to complete it in real time, and the completion results are coordinated with the cloud model inference results to verify the anomalies; the heart rate, acceleration, and temperature data are mapped to a spatiotemporal graph structure, and multimodal features are fused through dynamic edge weights. When potential heart rate anomalies are detected, the pruned and quantized graph neural network model deployed on the Arduino side is activated to improve detection accuracy.

2. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 1, characterized in that, The wearable smart device is used to monitor environmental conditions and user status in real time, and is composed of a temperature and humidity sensor, a harmful gas detector, an accelerometer and a heart rate sensor; the sensor is installed on the wearable device, monitors environmental conditions and the user's physical activity information and heart rate in real time, and transmits the data to the Arduino microcontroller; the heart rate data is processed by signal amplification and filtering.

3. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 1, characterized in that, The process of obtaining the heart rate data priority is as follows: Calculate the environmental risk score index by integrating gas concentration, temperature change and humidity; Use acceleration data to calculate the intensity of user activity; According to the calculated environmental risk score and user activity intensity, the heart rate data priority of the heart rate data is calculated; dynamic adjustment is implemented in Arduino microcontroller programming, and the heart rate data priority f is calculated through the following model s , Among them, f0 represents the priority of basic heart rate data, which is set according to the design requirements of the device; HR_avg represents the average heart rate measured in the time window, HR tar Indicates a preset heart rate value, and adjusts the preset heart rate value based on an adaptive threshold; α represents the environmental risk score weight factor, which is used to adjust the impact of environmental risk on the priority of heart rate data; β represents the user activity intensity weight factor, which is used to adjust the impact of user activity on the priority of heart rate data; γ represents the power consumption factor, which indicates the degree of influence of the power consumption mode on the priority of heart rate data; R e represents the environmental risk score, C g represents gas concentration, T represents real-time temperature, T ref is the reference temperature, H represents humidity, w g ,w t ,w h Represent various weight factors respectively; A u Indicates the user activity intensity, a x ,a y ,a z represents the readings of the acceleration sensor in three axes, N represents the number of sampling points in the sliding window; P c Indicates the power consumption mode.

4. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 3, characterized in that, The process of obtaining the heart rate data priority also includes: using a long short-term memory network to build a state prediction model, the state prediction model is used to predict future environmental conditions and user status; using historical data to train the state prediction model, the input includes past environmental parameters and user physical activity information, and the output is a predicted environmental risk score and a predicted user activity intensity; based on these prediction results, the heart rate data priority of heart rate monitoring is adjusted in advance.

5. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 4, characterized in that: The heart rate data priority calculation formula based on the state prediction model is: in, Score the predicted environmental risk, is the predicted user activity intensity.

6. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 1, characterized in that: Also includes: According to the relationship between the abnormal heart rate index and the threshold, different levels of alarm modes are triggered, including: Low-level alarm: When the heart rate abnormality index deviates from the target range but is not serious, the user is reminded to pay attention to the heart rate change through vibration; Medium-level alarm: When the heart rate abnormality index continues to deviate from the target range, indicating that it has reached a medium risk, the user will be reminded through an audible and visual alarm; High-level alarm: When the heart rate abnormality index deviates from the target range and the environmental risk score or user activity intensity reaches the danger threshold, a text message or remote push alarm is triggered.

7. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 6, characterized in that: The calculation formula of the heart rate risk index is: Among them, R trend Indicates the rate of change of heart rate trend, H var Indicates the heart rate fluctuation amplitude within the time window, T dev It represents the cumulative time that the heart rate deviates from the target value, and k represents the trend adjustment coefficient.

8. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to any one of claims 1-7, characterized in that: It also includes heart rate data transmission monitoring steps, including: Before the heart rate data is transmitted, the network environment is sensed to evaluate the signal strength, network congestion and packet loss rate in real time. The network reliability evaluation model is used to analyze historical and real-time network data. The input features include signal strength, historical packet loss rate, and network response time, and the network stability coefficient of each transmission path is output. Compare the network stability coefficients of several optional transmission paths; select the path with the highest network stability coefficient to transmit the heart rate data; Compress and prioritize heart rate data to reduce latency caused by large amounts of data; After the data is successfully sent, an end-to-end confirmation protocol is used to ensure that each data packet is confirmed; if the confirmation fails, the retransmission mechanism is triggered; if an abnormal heart rate or data transmission failure is detected, an alarm is triggered immediately.

9. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 7, characterized in that: The method also includes sparse heart rate data completion and abnormality monitoring steps, including the following steps: Collect heart rate time series data with missing segments, mark the missing positions and extract heart rate variability characteristics as physiological constraints; the physiological constraints include: obtaining the heart rate standard deviation within a preset time window, mapping it to the [-1,1] interval through a hyperbolic tangent function to generate a first sub-factor; performing time integration on the absolute value of the heart rate change rate, taking a logarithmic operation and multiplying it by an age-related trend sensitivity coefficient to generate a second sub-factor; adding the first sub-factor to the second sub-factor as the output value of the physiological state factor; the physiological state factor is used to constrain the quantitative index of the data completion process, and through the physiological characteristics of heart rate variability, ensure that the completed heart rate data conforms to the real physiological laws of the human body; A generator model based on bidirectional LSTM and multi-head attention mechanism is constructed to input sparse heart rate data and output the completed continuous sequence, where the attention mechanism focuses on the temporal correlation between the missing segments. Design a discriminator model to determine the authenticity of the completed data through a convolutional network, and calculate the heart rate variability feature constraint loss to verify physiological rationality; A hybrid strategy combining pre-training and adversarial training is used to optimize the generator. In the pre-training phase, random deletions are simulated and the mean square error is minimized. In the adversarial training phase, the loss weights of the generator and the discriminator are dynamically adjusted. The trained generator is compressed into a lightweight LSTM model through knowledge distillation and deployed to the Arduino side to achieve real-time missing completion.

10. A wireless heart rate monitoring and SMS alarm method based on Arduino single chip microcomputer according to claim 9, characterized in that: The method also includes a health risk early warning step based on multimodal heterogeneous data fusion, including the following steps: The data of the heart rate sensor, acceleration sensor and temperature sensor are synchronized with the timestamp and sampling frequency, and the low-frequency data is unified to the high-frequency sampling rate through interpolation; Multimodal data are mapped into a graph structure, where each node represents a physiological or environmental parameter. The edge weights between nodes are initialized based on the statistical correlation between modalities and dynamically updated through the spatiotemporal graph attention layer, including: Concatenate the feature representations of node i and node j at time t; Perform a linear transformation on the concatenated features through learnable parameters and apply an activation function; Perform Softmax normalization on the calculation results to obtain the updated edge weights; The gated recurrent unit is used to perform temporal aggregation on the dynamically updated multimodal features to generate the fused health status code; Prune and quantize the graph neural network model with 8-bit fixed-point numbers, and deploy it to the Arduino terminal to perform hierarchical reasoning: Run simplified threshold detection in normal state, and activate the full model for in-depth analysis when anomalies are detected; The fused features are input into the random forest classifier to output low risk, medium risk, and high risk levels. The alarm information is generated by combining semantic compression coding and pushed to the preset terminal through the SMS gateway.

Images