Neural Network-Based Sleep Quality Monitoring and Management System and Method

Abstract

This application provides a sleep quality monitoring and management system and method based on a neural network, which relates to the field of intelligent monitoring. It uses a pulse monitoring device to collect the pulse data of the monitored object during sleep in real time, forms a time queue and transmits it to the intelligent terminal. The intelligent terminal first screens out abnormal data to ensure data accuracy; then classifies the screened pulse data, establishes a mapping relationship between the classification results and the sleep stages, and generates a sleep stage mapping result. Finally, the sleep quality of the monitored object is evaluated based on these mapping results, and targeted sleep improvement suggestions are provided. In this way, through precise data processing and analysis, the accuracy of sleep monitoring and the quality of personalized services can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring, and more particularly, to a sleep quality monitoring and management system and method based on a neural network in the embodiments of this application. Background Art

[0002] Sleep quality directly affects an individual's health and quality of life. With the accelerating pace of modern life, more and more people are facing sleep problems such as insomnia, sleep apnea syndrome, etc. Therefore, simple and efficient sleep quality monitoring methods have become particularly important. Patent CN103654744A discloses a sleep quality monitoring method and system, which obtains the pulse data of the monitored person during sleep through a pulse monitoring device and transmits this data to an intelligent terminal. The intelligent terminal classifies the pulse data using a clustering function algorithm and establishes a mapping relationship between the classification results and different stages of sleep. Based on this mapping relationship, the intelligent terminal can judge the sleep quality of the monitored person and extract corresponding suggestions from the sleep management database to provide to the user.

[0003] In practical applications, pulse data may have abnormal values due to various external or internal factors, such as poor sensor contact, sudden movements of the user, or environmental interference. If these abnormal data are directly used for analysis without being processed, it may lead to misjudgment of sleep staging, thereby affecting the accuracy of the final sleep quality assessment.

[0004] This patent uses a fuzzy algorithm to screen out abnormal data in the pulse data. However, the fuzzy algorithm relies on predefined rules or models when dealing with uncertainty problems, which limits its adaptability to unforeseen situations, especially in complex and changing practical application scenarios. And because parameter adjustment requires manual intervention, it is difficult for the system to automatically optimize to adapt to individual differences or environmental changes, resulting in limited flexibility and response speed. In addition, the static characteristics of the fuzzy logic system make it unable to effectively learn newly emerging patterns or trends, reducing the accuracy and reliability of long-term use. This limitation is particularly obvious when facing personalized needs and dynamic environments, and it is difficult to provide continuous and accurate services.

[0005] Therefore, an optimized sleep quality monitoring and management solution is desired. Summary of the Invention

[0006] To solve the above technical problems, this application is proposed.

[0007] According to one aspect of the present application, a neural network-based sleep quality monitoring and management method is provided, which includes: using a pulse monitoring device to collect the pulse data of the monitored object during sleep in real time to obtain a time queue of the pulse data, and transmitting the time queue of the pulse data to an intelligent terminal; in the intelligent terminal, screening abnormal data from the time queue of the pulse data to obtain a time queue of the pulse data after screening; in the intelligent terminal, classifying the time queue of the pulse data after screening to obtain a classification result, and establishing a mapping relationship between the classification result and the sleep stage to obtain a sleep stage mapping result; in the intelligent terminal, according to the sleep stage mapping result, obtaining the sleep quality of the monitored object, and providing sleep improvement suggestions for the monitored object according to the sleep quality; wherein, screening abnormal data from the time queue of the pulse data to obtain the pulse data after screening includes: extracting preliminarily screened abnormal pulse data from the time queue of the pulse data, and dividing the time queue of the pulse data to obtain a time queue of forward pulse data and a time queue of backward pulse data; according to the preliminarily screened abnormal pulse data, performing semantic association time series encoding on the time queue of the forward pulse data and the time queue of the backward pulse data respectively to obtain pulse data semantic association forward time series features and pulse data semantic association backward time series features; performing feature interaction of a dynamic memory mechanism on the pulse data semantic association forward time series features and the pulse data semantic association backward time series features to obtain pulse data semantic association forward and backward interaction fusion features, and obtaining an identification result according to the pulse data semantic association forward and backward interaction fusion features.

[0008] According to another aspect of the present application, a neural network-based sleep quality monitoring and management system is provided, which includes: a pulse data time queue extraction module, configured to extract preliminarily screened abnormal pulse data from the time queue of the pulse data, and divide the time queue of the pulse data to obtain a time queue of forward pulse data and a time queue of backward pulse data; a semantic association time series encoding module, configured to perform semantic association time series encoding on the time queue of the forward pulse data and the time queue of the backward pulse data respectively according to the preliminarily screened abnormal pulse data to obtain pulse data semantic association forward time series features and pulse data semantic association backward time series features; a dynamic memory mechanism feature interaction module, configured to perform feature interaction of a dynamic memory mechanism on the pulse data semantic association forward time series features and the pulse data semantic association backward time series features to obtain pulse data semantic association forward and backward interaction fusion features, and obtaining an identification result according to the pulse data semantic association forward and backward interaction fusion features.

[0009] Compared with the prior art, a sleep quality monitoring and management system and method based on a neural network provided by the present application extracts and divides the initially screened abnormal pulse data from the time queue of pulse data, and then performs semantic association time series encoding, so as to intelligently determine whether to determine the initially screened abnormal pulse data as abnormal pulse data according to the dynamic memory interaction representation between the forward time series features of the pulse data semantic association and the backward time series features of the pulse data semantic association. In this way, it can automatically adapt to individual physiological differences and environmental changes, reduce the dependence on manual intervention to adjust parameters, and enhance the robustness of the system when encountering unknown situations or newly emerging patterns. At the same time, it can continuously learn new patterns and trends from new data, and improve its performance in meeting personalized needs and adapting to dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 It is a flowchart of a method for monitoring and managing sleep quality based on a neural network according to an embodiment of the present application.

[0012] Figure 2 It is a flowchart of screening abnormal data from the time queue of the pulse data to obtain the pulse data after screening in the method for monitoring and managing sleep quality based on a neural network according to an embodiment of the present application.

[0013] Figure 3 It is a schematic diagram of data flow for screening abnormal data from the time queue of the pulse data to obtain the pulse data after screening in the method for monitoring and managing sleep quality based on a neural network according to an embodiment of the present application.

[0014] Figure 4 It is a flowchart of performing semantic association time series encoding on the time queue of the forward pulse data and the time queue of the backward pulse data respectively according to the initially screened abnormal pulse data to obtain the forward time series features of the pulse data semantic association and the backward time series features of the pulse data semantic association in the method for monitoring and managing sleep quality based on a neural network according to an embodiment of the present application.

[0015] Figure 5It is a flowchart for obtaining the forward and backward interaction fusion features of pulse data semantic association through feature interaction of the forward temporal features and backward temporal features of pulse data semantic association in the neural network-based sleep quality monitoring and management method according to an embodiment of the present application, and obtaining an identification result based on the forward and backward interaction fusion features of pulse data semantic association.

[0016] Figure 6 It is a flowchart for obtaining the forward and backward interaction fusion feature vector of pulse data semantic association through feature interaction of the forward temporal feature vector and backward temporal feature vector of pulse data semantic association in the neural network-based sleep quality monitoring and management method according to an embodiment of the present application.

[0017] Figure 7 It is a system block diagram of a neural network-based sleep quality monitoring and management system according to an embodiment of the present application. Detailed implementation manners

[0018] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0019] Sleep quality is crucial for health and quality of life. Especially in the context of the accelerating modern life rhythm, many people are facing sleep problems. To monitor sleep quality, Patent CN103654744A proposes a method for analyzing sleep quality through pulse data. The pulse monitoring device collects data and transmits it to the smart terminal, which classifies the data through a clustering algorithm and establishes a corresponding relationship with different sleep stages, thereby evaluating sleep quality and giving improvement suggestions. However, pulse data may be affected by various interferences, such as poor sensor contact or user activities, which may lead to abnormal data. If these abnormal data are not processed, it may affect the judgment of sleep stages and the accuracy of sleep quality assessment. This patent filters abnormal data through a fuzzy algorithm, but the fuzzy algorithm relies on preset rules and lacks flexibility, and cannot adapt to new situations. Especially in a dynamically changing environment, the fuzzy algorithm requires manual adjustment and cannot learn new patterns, which may affect accuracy and reliability during long-term use and is difficult to provide continuous and accurate sleep quality monitoring.

[0020] In view of the above technical problems, the present application proposes a neural network-based sleep quality monitoring and management method. Figure 1 It is a flowchart of the neural network-based sleep quality monitoring and management method according to an embodiment of the present application. As Figure 1As shown, the neural network-based sleep quality monitoring and management method according to an embodiment of the present application includes: S110, using a pulse monitoring device to collect the pulse data of the monitored object during sleep in real time to obtain a time queue of the pulse data, and transmitting the time queue of the pulse data to an intelligent terminal; S120, at the intelligent terminal, screening abnormal data from the time queue of the pulse data to obtain a time queue of the pulse data after screening; S130, at the intelligent terminal, classifying the time queue of the pulse data after screening to obtain a classification result, and establishing a mapping relationship between the classification result and the sleep stage to obtain a sleep stage mapping result; S140, at the intelligent terminal, obtaining the sleep quality of the monitored object according to the sleep stage mapping result, and providing sleep improvement suggestions for the monitored object according to the sleep quality.

[0021] In the above neural network-based sleep quality monitoring and management method, in step S110, a pulse monitoring device is used to collect the pulse data of the monitored object during sleep in real time to obtain a time queue of the pulse data, and the time queue of the pulse data is transmitted to an intelligent terminal. It should be understood that the pulse monitoring device generally includes a sensor module, a data acquisition and processing unit, and a communication module. During sleep, the device senses the pulse signal in real time by contacting the body of the monitored object (for example, through sensors worn on the wrist, fingertip, earlobe, or chest). Among them, the acquisition of the pulse signal is realized by sensors (such as photoplethysmography sensors, infrared sensors, or capacitance sensors, etc.), and these sensors can accurately detect the changes in the pulse wave and the pulsation frequency of blood flow. During this process, the pulse monitoring device continuously records the pulse wave generated by each heartbeat, usually sampling at a high frequency (such as multiple sampling points per second), thereby forming a time queue of the pulse data. This time queue represents the pulse changes of the monitored object during sleep, including various parameters such as pulse frequency, pulse interval, and waveform changes. These data can reflect the physiological state of the monitored person during sleep, such as whether there are arrhythmias, rapid heartbeats, or slow pulses, etc., and further provide in-depth analysis of the health status.

[0022] Once the pulse data is effectively collected and processed, the next task is to transmit this data to the intelligent terminal for further analysis and display. At this time, the pulse monitoring device needs to have a wireless communication function, and common methods include short-range wireless transmission technologies such as Bluetooth, Wi-Fi, or Zigbee. Through these wireless communication technologies, the monitoring device can transmit the time queue of the pulse data to the intelligent terminal in real time. After receiving these data, the intelligent terminal can display, store, and analyze them through the corresponding health management application.

[0023] In the above sleep quality monitoring and management method based on a neural network, in step S120, at the intelligent terminal, abnormal data in the time queue of the pulse data is screened to obtain the time queue of the pulse data after screening. It should be understood that considering that pulse data is usually recorded in the form of a time series, due to various external factors (such as measurement errors, equipment failures, interference, etc.), there may be outliers in this data. These outliers may be manifested as sudden changes in pulse frequency, unreasonable fluctuations, or values significantly deviating from the normal physiological range. If not screened out, it will affect the authenticity and effectiveness of the data, and even lead to incorrect health judgments. The screened data set can provide a more authentic basis for subsequent health analysis and help users better monitor their physical conditions.

[0024] Correspondingly, in screening abnormal data in the time queue of the pulse data to obtain the pulse data after screening, the technical concept of the present application is to use a neural network-based data analysis and processing algorithm to extract preliminarily screened abnormal pulse data from the time queue of the pulse data, and divide the time queue of the pulse data. Then, semantic association time series encoding is performed on the time queue of the forward pulse data and the time queue of the backward pulse data after division. Based on this, whether the preliminarily screened abnormal pulse data is determined as abnormal pulse data is intelligently judged according to the dynamic memory interaction representation between the semantic association forward time series feature of the pulse data and the semantic association backward time series feature of the pulse data. In this way, it can automatically adapt to individual physiological differences and environmental changes, reduce the dependence on manual intervention to adjust parameters, and enhance the robustness of the system when encountering unknown situations or newly emerging patterns. At the same time, it has a strong dynamic learning ability and can continuously learn new patterns and trends from new data, thereby improving its performance in meeting personalized needs and adapting to dynamic environments.

[0025] Figure 2 It is a flowchart for screening abnormal data in the time queue of the pulse data to obtain the pulse data after screening in the sleep quality monitoring and management method based on a neural network according to an embodiment of the present application. Figure 3 It is a schematic diagram of data flow for screening abnormal data in the time queue of the pulse data to obtain the pulse data after screening in the sleep quality monitoring and management method based on a neural network according to an embodiment of the present application. As Figure 2 and Figure 3As shown, in the step S120, screening abnormal data from the time queue of the pulse data to obtain the pulse data after screening includes: S121, extracting the preliminarily screened abnormal pulse data from the time queue of the pulse data, and dividing the time queue of the pulse data to obtain the time queue of the forward pulse data and the time queue of the backward pulse data; S122, respectively performing semantic association time series coding on the time queue of the forward pulse data and the time queue of the backward pulse data according to the preliminarily screened abnormal pulse data to obtain the semantic association forward time series feature of the pulse data and the semantic association backward time series feature of the pulse data; S123, performing feature interaction of the dynamic memory mechanism on the semantic association forward time series feature of the pulse data and the semantic association backward time series feature of the pulse data to obtain the semantic association forward and backward interaction fusion feature of the pulse data, and obtaining the recognition result according to the semantic association forward and backward interaction fusion feature of the pulse data.

[0026] In the embodiment of the present application, the step S121, extracting the preliminarily screened abnormal pulse data from the time queue of the pulse data, and dividing the time queue of the pulse data to obtain the time queue of the forward pulse data and the time queue of the backward pulse data, includes: S1211, preliminarily determining the abnormal pulse data in the time queue of the pulse data to obtain the preliminarily screened abnormal pulse data; S1212, dividing the time queue of the pulse data with the preliminarily screened abnormal pulse data as the node to obtain the time queue of the forward pulse data and the time queue of the backward pulse data.

[0027] Specifically, in the step S1211, the abnormal pulse data in the time queue of the pulse data is preliminarily determined to obtain the preliminarily screened abnormal pulse data. It should be understood that abnormal data often has large volatility and uncertainty. If all the original pulse data is directly analyzed in depth, it may consume a large amount of computing resources and time, especially when the data volume is very large. Therefore, in the technical solution of this application, the abnormal pulse data in the time queue of the pulse data is preliminarily determined to obtain the preliminarily screened abnormal pulse data. In this way, the data volume for subsequent processing can be reduced, thereby improving the operating efficiency of the entire system. More specifically, when a time series of pulse data P = {p1, p2, p3, …, pn} is obtained, where each data point represents the pulse value measured at different time points. Among them, considering that the pulse data will not change violently in a short period of time. Therefore, the change rate Δpi = ∣pi−pi−1∣ between adjacent data points can be calculated. If this change rate exceeds a predetermined threshold (such as 20 beats per minute), then this data point is determined to be abnormal. For example, if a certain pulse value suddenly jumps from 80 to 150 and then back to 82, it may be due to measurement errors or sensor problems. Such data points should be marked as abnormal data. Further, in order to reduce misjudgments caused by short-term fluctuations, the sliding window method can be used to smooth the data. Among them, a sliding window with a size of 5 is used to calculate the mean value mean(pi−2, pi−1, pi, pi+1, pi+2) and the standard deviation of the data within each window. If the deviation of a certain data point pi from the window mean exceeds a certain threshold, then this point can also be regarded as abnormal data. In this way, in the case of a large amount of data, the obviously abnormal pulse data can be quickly identified and removed, thereby effectively reducing the data volume that needs to be further processed and improving the computing efficiency of the system.

[0028] Next, in order to more accurately identify and verify the preliminarily screened abnormal pulse data points, in the step S1212, the time queue of the pulse data is divided with the preliminarily screened abnormal pulse data as nodes to obtain the time queue of the forward pulse data and the time queue of the backward pulse data. In this way, the data change trends before and after the abnormal point can be observed more clearly, and the characteristics near the abnormal point can be analyzed in depth. Thereby to determine whether the abnormal point represents a real health problem, or whether it is only caused by measurement errors or short-term external interferences.

[0029] Figure 4 It is a flowchart of semantic association time series encoding for the time queue of the forward pulse data and the time queue of the backward pulse data respectively according to the preliminarily screened abnormal pulse data in the sleep quality monitoring and management method based on a neural network according to an embodiment of the present application to obtain the semantic association forward time series features of the pulse data and the semantic association backward time series features of the pulse data. AsFigure 4 As shown, in the embodiment of the present application, in the step S122, according to the initially screened abnormal pulse data, semantic association time series encoding is respectively performed on the time queue of the forward pulse data and the time queue of the backward pulse data to obtain the semantic association forward time series feature of the pulse data and the semantic association backward time series feature of the pulse data, including: S1221, taking the initially screened abnormal pulse data as the reference semantics, calculating the semantic association degrees between the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the semantic association degrees of the forward pulse data, and calculating the semantic association degrees between the reference semantics and each backward pulse data in the time queue of the backward pulse data to obtain the time queue of the semantic association degrees of the backward pulse data; S1222, respectively performing time series semantic association encoding on the time queue of the semantic association degrees of the forward pulse data and the time queue of the semantic association degrees of the backward pulse data to obtain the semantic association forward time series feature vector of the pulse data as the semantic association forward time series feature of the pulse data and the semantic association backward time series feature vector of the pulse data as the semantic association backward time series feature of the pulse data.

[0030] Specifically, in the step S1221, taking the initially screened abnormal pulse data as the reference semantics, calculating the semantic association degrees between the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the semantic association degrees of the forward pulse data, and calculating the semantic association degrees between the reference semantics and each backward pulse data in the time queue of the backward pulse data to obtain the time queue of the semantic association degrees of the backward pulse data. It should be understood that considering that abnormal points are usually accompanied by changes in physiological states, and these changes may have started to appear before the abnormal points and continue to have an impact for some time after that. Therefore, in order to capture and mine the correlations before and after the initially screened abnormal data with the initially screened abnormality, to help identify which pulse behaviors are closely related in the time series and which pulse behaviors may be isolated or abnormal, in the technical solution of the present application, taking the initially screened abnormal pulse data as the reference semantics, calculating the semantic association degrees between the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the semantic association degrees of the forward pulse data, and calculating the semantic association degrees between the reference semantics and each backward pulse data in the time queue of the backward pulse data to obtain the time queue of the semantic association degrees of the backward pulse data.

[0031] Specifically, in a specific embodiment of the present application, taking the calculation of the semantic association degree between the initially screened abnormal pulse data as the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the forward pulse data semantic association degree as an example: First, define the initial value of the window radius . Specifically, the initial is usually set to 1 to gradually expand the analysis scope. That is, define the window radius to control the analysis dynamic range. Then, for each initial window size , calculate each forward pulse data with a time span of before the initially screened abnormal pulse data , where is the mean value of the cosine similarity between ; as the semantic association degree of the forward pulse data within this window, where is and the cosine similarity between. Then gradually increase the size of the window radius , repeat the above steps, and record each corresponding semantic association degree until is the number of forward pulse data in the time queue of the forward pulse data . Each time a semantic association degree value is calculated to obtain the final time queue

[0032] of the forward pulse data semantic association degree, so that the change of the semantic association degree under different time windows can be evaluated, and thus the change trend of the pulse data can be understood more comprehensively. For example, if the semantic association degree gradually decreases, it may indicate the occurrence of an accelerated heartbeat or other abnormal conditions.

[0033] In the embodiment of the present application, in the step S1222, performing temporal semantic association encoding on the time queue of the forward pulse data semantic association degree and the time queue of the backward pulse data semantic association degree respectively to obtain a pulse data semantic association forward temporal feature vector as the pulse data semantic association forward temporal feature and a pulse data semantic association backward temporal feature vector as the pulse data semantic association backward temporal feature, including: S1222-1, passing the time queue of the forward pulse data semantic association degree through a forward pulse semantic association temporal encoder based on a forward LSTM model to obtain the pulse data semantic association forward temporal feature vector; S1222-2, passing the time queue of the backward pulse data semantic association degree through a backward pulse semantic association temporal encoder based on a backward LSTM model to obtain the pulse data semantic association backward temporal feature vector.

[0033] Specifically, in the step S1222-1, the time queue of the forward pulse data semantic correlation degree is passed through a forward pulse semantic correlation time sequence encoder based on a forward LSTM model to obtain the forward time sequence feature vector of the pulse data semantic correlation. It should be understood that considering that the semantic correlation degree at each time point in the time queue of the forward pulse data semantic correlation degree reflects the similarity between the pulse data at that moment and the preliminarily screened abnormal pulse data, therefore, in order to further explore and identify whether there are certain specific pulse behavior patterns or characteristics before the occurrence of the preliminarily screened abnormal event, and at the same time capture the changing trend of the pulse data over time, in the technical solution of this application, the time queue of the forward pulse data semantic correlation degree is passed through a forward pulse semantic correlation time sequence encoder based on a forward LSTM model to utilize the memory unit and gating mechanism inside the forward LSTM model to extract the complex dynamic semantic changes in the forward pulse semantic data, and obtain the forward time sequence feature vector of the pulse data semantic correlation.

[0034] Specifically, in the step S1222-2, the time queue of the backward pulse data semantic correlation degree is passed through a backward pulse semantic correlation time sequence encoder based on a backward LSTM model to obtain the backward time sequence feature vector of the pulse data semantic correlation. It should be understood that in order to further improve the semantic correlation degree analysis of the pulse data, in addition to encoding the time queue of the forward pulse data semantic correlation degree, it is also necessary to perform time sequence semantic correlation encoding on the time queue of the backward pulse data semantic correlation degree. Among them, the time queue of the backward pulse data semantic correlation degree reflects the semantic similarity between the pulse data and the abnormal data after the occurrence of the preliminarily screened abnormal event, so it can help the system capture the potential trend of the pulse data change, especially the pulse behavior pattern after the occurrence of the abnormal event. The backward time sequence encoder utilizes a backward LSTM (Long Short-Term Memory) model, which can trace back from future time points to the current moment, so as to learn the reverse dynamic information in the time series data. The time queue of the backward pulse data semantic correlation degree is input into the backward pulse time sequence encoder based on the backward LSTM model. The structure of the model is similar to that of the standard LSTM, but it encodes the time series from backward to forward. Specifically, the backward LSTM passes the data input at future moments reversely through the internal memory unit of the recurrent neural network, and controls the flow and memory of information through gating mechanisms (such as forget gate, input gate and output gate), so that the sequence information of the backward pulse data can be effectively stored and extracted. Since the backward LSTM can capture the time dependence from future to past, it can identify the potential trends and change patterns in the pulse data after the occurrence of the abnormal event.

[0035] Figure 5The flowchart shows the process of obtaining the forward-backward interaction fusion feature of the pulse data semantic association through feature interaction with a dynamic memory mechanism for the forward temporal feature of the pulse data semantic association and the backward temporal feature of the pulse data semantic association in the neural network-based sleep quality monitoring and management method according to the embodiments of the present application, and obtaining the recognition result based on the forward-backward interaction fusion feature of the pulse data semantic association. As Figure 5 shown, in the embodiment of the present application, step S123, which performs feature interaction with a dynamic memory mechanism on the forward temporal feature of the pulse data semantic association and the backward temporal feature of the pulse data semantic association to obtain the forward-backward interaction fusion feature of the pulse data semantic association, and obtains the recognition result based on the forward-backward interaction fusion feature of the pulse data semantic association, includes: S1231, performing feature interaction with a dynamic memory mechanism on the forward temporal feature vector of the pulse data semantic association and the backward temporal feature vector of the pulse data semantic association to obtain the forward-backward interaction fusion feature vector of the pulse data semantic association as the forward-backward interaction fusion feature of the pulse data semantic association; S1232, obtaining the recognition result based on the forward-backward interaction fusion feature vector of the pulse data semantic association, where the recognition result is used to indicate whether the preliminarily screened abnormal pulse data is determined as abnormal pulse data.

[0036] Specifically, in step S123, feature interaction with a dynamic memory mechanism is performed on the forward temporal feature of the pulse data semantic association and the backward temporal feature of the pulse data semantic association to obtain the forward-backward interaction fusion feature of the pulse data semantic association, and the recognition result is obtained based on the forward-backward interaction fusion feature of the pulse data semantic association. It should be understood that considering that the forward temporal feature vector of the pulse data semantic association and the backward temporal feature vector of the pulse data semantic association respectively express important information about the dynamic semantic changes of the pulse data in different time series. Therefore, in order to achieve deeper time series feature fusion, thereby obtaining a richer and more accurate feature representation to more comprehensively understand the relationship between the pulse data at each time point and its front and back contexts, and thus providing rich data support for the subsequent judgment of abnormal points, in the technical solution of the present application, the forward temporal feature vector of the pulse data semantic association and the backward temporal feature vector of the pulse data semantic association are processed through feature interaction with a dynamic memory mechanism to obtain the forward-backward interaction fusion feature vector of the pulse data semantic association.

[0037] Figure 6 The flowchart shows the process of performing feature interaction with a dynamic memory mechanism on the forward temporal feature vector of the pulse data semantic association and the backward temporal feature vector of the pulse data semantic association to obtain the forward-backward interaction fusion feature vector in the neural network-based sleep quality monitoring and management method according to the embodiments of the present application. As Figure 6As shown, in the embodiment of the present application, in step S1231, feature interaction of the dynamic memory mechanism is performed on the semantic association forward temporal feature vector of the pulse data and the semantic association backward temporal feature vector of the pulse data to obtain the semantic association forward-backward interaction fusion feature vector of the pulse data, including: S1231-1, performing collaborative implicit feature extraction on the semantic association forward temporal feature vector of the pulse data and the semantic association backward temporal feature vector of the pulse data to obtain the implicit collaborative coding vector between the forward-backward temporal features of the semantic association of the pulse data; S1231-2, according to the implicit collaborative coding vector between the forward-backward temporal features of the semantic association of the pulse data, performing dynamic attention modulation on the semantic association forward temporal feature vector of the pulse data and the semantic association backward temporal feature vector of the pulse data to obtain the semantic association forward-backward interaction fusion feature vector of the pulse data.

[0038] Specifically, in step S1231-1, collaborative implicit feature extraction is performed on the semantic association forward temporal feature vector of the pulse data and the semantic association backward temporal feature vector of the pulse data to obtain the implicit collaborative coding vector between the forward-backward temporal features of the semantic association of the pulse data. It should be understood that the collaborative feature extraction network is composed of a point convolutional layer and an activation function, aiming to extract high-level abstract features from the original feature vector. Through a carefully designed network architecture, the collaborative feature extraction network can capture the deep association between the two feature vectors, namely the so-called "implicit collaboration". This implicit collaboration is not limited to simple linear combinations, but involves complex non-linear interaction patterns.

[0039] More specifically, the process of the above step S1231-1 can be expressed by the formula: ; where and respectively represent the semantic association forward temporal feature vector of the pulse data and the semantic association backward temporal feature vector of the pulse data, and are pointwise addition, pointwise subtraction, and pointwise multiplication by position respectively, is concatenation processing, is the point convolutional coding operation, is the Leaky ReLU activation function, is the implicit collaborative coding vector between the forward-backward temporal features of the semantic association of the pulse data. Through the convolutional operation, the network can identify local association patterns, and the use of activation functions, such as ReLU or Leaky ReLU, ensures that the network can learn non-linear feature representations, which is crucial for capturing the complex relationships between features.

[0040] In the embodiments of the present application, in step S1231-2, according to the implicit co-coding vector between the forward-backward temporal features of the pulse data semantic association, dynamic attention modulation is performed on the forward temporal feature vector of the pulse data semantic association and the backward temporal feature vector of the pulse data semantic association to obtain the forward-backward interaction fusion feature vector of the pulse data semantic association, including: S1231-21, writing the implicit co-coding vector between the forward-backward temporal features of the pulse data semantic association into a dynamic memory unit to obtain a forward-backward temporal dynamic key vector of the pulse data semantic association; S1231-22, extracting the forward-backward temporal dynamic key vector of the pulse data semantic association from the dynamic memory unit, and performing feature optimization based on a first transformer on the forward temporal feature vector of the pulse data semantic association and the forward-backward temporal dynamic key vector of the pulse data semantic association to obtain an optimized forward temporal feature vector of the pulse data semantic association; S1231-23, extracting the forward-backward temporal dynamic key vector of the pulse data semantic association from the dynamic memory unit, and performing feature optimization based on a second transformer on the backward temporal feature vector of the pulse data semantic association and the forward-backward temporal dynamic key vector of the pulse data semantic association to obtain an optimized backward temporal feature vector of the pulse data semantic association; S1231-24, performing feature linear interaction on the optimized forward temporal feature vector of the pulse data semantic association and the optimized backward temporal feature vector of the pulse data semantic association to obtain the forward-backward interaction fusion feature vector of the pulse data semantic association.

[0041] Specifically, in step S1231-21, the implicit co-coding vector between the forward-backward temporal features of the pulse data semantic association is written into a dynamic memory unit to obtain a forward-backward temporal dynamic key vector of the pulse data semantic association. It should be understood that the dynamic memory unit is a special type of memory structure, and its working principle is similar to the memory unit in a recurrent neural network, but it has stronger flexibility and adaptability. The dynamic memory unit can dynamically adjust its own memory content according to the current input state.

[0042] More specifically, the process of step S1231-21 can be represented by the following formula: ; where is a 1×1 convolution operation, is the forward-backward temporal dynamic key vector of the pulse data semantic association. This means that when a new feature vector is input, the dynamic memory unit can adjust its memory according to the newly input data to better adapt to the context requirements of the current scenario.

[0043] Specifically, in the steps S1231-22 and S1231-23, the pulse data semantic association forward-backward temporal dynamic key vectors are extracted from the dynamic memory unit, and the pulse data semantic association forward temporal feature vectors and the pulse data semantic association forward-backward temporal dynamic key vectors are input into the feature attention modulation module based on the first Transformer structure to obtain the optimized pulse data semantic association forward temporal feature vectors. At the same time, the pulse data semantic association forward-backward temporal dynamic key vectors are extracted from the dynamic memory unit, and the pulse data semantic association backward temporal feature vectors and the pulse data semantic association forward-backward temporal dynamic key vectors are input into the feature attention modulation module based on the second Transformer structure to obtain the optimized pulse data semantic association backward temporal feature vectors. The core of the feature attention modulation module based on the first Transformer structure and the feature attention modulation module based on the second Transformer structure is the self-attention mechanism, which allows the model to consider the influence of all other elements when processing each input element, rather than just adjacent elements. This global perspective helps the model capture long-range dependencies, which is particularly important for understanding complex data.

[0044] More specifically, the process of the above step S1231-22 can be expressed by the formula: ; where is the pulse data semantic association forward query embedding matrix, is the first pulse data semantic association forward bias vector, is matrix multiplication, is the pulse data semantic association forward query vector, is the pulse data semantic association forward value embedding matrix, is the second pulse data semantic association forward bias vector, is the pulse data semantic association forward value vector, is transpose vector of, is length of, is function, is the optimized pulse data semantic association forward temporal feature vector.

[0045] More specifically, the process of the above step S1231-23 can be expressed by the formula: ; where is the pulse data semantic association backward query embedding matrix, is the first pulse data semantic association backward bias vector, is the pulse data semantic association backward query vector, is the pulse data semantic association backward value embedding matrix, is the second pulse data semantic association backward offset vector, is the pulse data semantic association backward value vector, is the optimized pulse data semantic association backward time series feature vector.

[0046] In the technical solution of the present application, by using the feature attention modulation module based on the first transformer structure and the feature attention modulation module based on the second transformer structure, the forward and backward time series feature vectors of the pulse data semantic association can be modulated based on the query attention with the forward-backward time series dynamic key vector of the pulse data semantic association. This is equivalent to using the forward-backward time series dynamic key vector of the pulse data semantic association as a common attention space, which not only enhances the expression of key information in the feature vector, reduces the influence of noise and irrelevant information, but also aligns the optimized forward time series feature vector of the pulse data semantic association and the optimized backward time series feature vector of the pulse data semantic association in the semantic space, and utilizes the feature interaction and fusion between the two.

[0047] Specifically, in the step S1231-24, a feature linear interaction is performed on the optimized forward time series feature vector of the pulse data semantic association and the optimized backward time series feature vector of the pulse data semantic association to obtain the forward-backward interaction fusion feature vector of the pulse data semantic association. It should be understood that the main task of the feature linear interaction network is to further integrate these two optimized feature vectors through a linear transformation to generate the final forward-backward interaction fusion feature vector of the pulse data semantic association.

[0048] More specifically, the process of the above step S1231-24 can be expressed by the formula: ; where, is the weighting coefficient, is the forward-backward interaction fusion feature vector of the pulse data semantic association. In this way, by performing a feature linear interaction on the optimized forward time series feature vector of the pulse data semantic association and the optimized backward time series feature vector of the pulse data semantic association, the forward and backward information can be combined to comprehensively capture the time dynamics and potential laws of the pulse data.

[0049] In an embodiment of the present application, the step S1232 of obtaining the recognition result by semantically associating the forward and backward interaction fusion feature vectors according to the pulse data includes: passing the semantically associated forward and backward interaction fusion feature vectors of the pulse data through a pulse anomaly data recognizer based on a classifier to obtain the recognition result. That is, the semantically associated forward and backward interaction fusion feature vectors of the pulse data obtained by performing dynamic memory interaction using the semantically associated forward temporal feature vector and the semantically associated backward temporal feature vector of the pulse data are classified, so as to intelligently determine whether the preliminarily screened abnormal pulse data is determined as abnormal pulse data. In this way, it can automatically adapt to individual physiological differences and environmental changes, reduce the dependence on manual intervention to adjust parameters, and enhance the robustness of the system when encountering unknown situations or newly emerging patterns. At the same time, it has a strong dynamic learning ability and can continuously learn new patterns and trends from new data, thereby improving its performance in meeting personalized needs and adapting to dynamic environments.

[0050] In a preferred example, considering that the semantically associated forward temporal feature vector and the semantically associated backward temporal feature vector of the pulse data respectively represent the short-range-long-range bidirectional temporal association features of the forward pulse data semantic association degree and the backward pulse data semantic association degree, when performing feature-interaction attention based on the dynamic memory mechanism, the multi-dimensional temporal association feature differences caused by the temporal distribution differences of data sources in different time domains will lead to insufficient correspondence of the dynamic mechanism, resulting in the lack of instance judgment of the attention interaction features in the semantically associated forward and backward interaction fusion feature vectors of the pulse data, thereby affecting the accuracy of the recognition result obtained by passing it through the pulse anomaly data recognizer based on the classifier.

[0051] Based on this, when passing the semantically associated forward and backward interaction fusion feature vectors of the pulse data through the pulse anomaly data recognizer based on the classifier, the semantically associated forward and backward interaction fusion feature vectors of the pulse data are first optimized, including the steps of: calculating the distance between the and the eigenvalues of the semantically associated forward and backward interaction fusion feature vectors of the pulse data, for example L 2 distance, and taking the square root of the distance to obtain a semantically associated forward and backward interaction fusion distance representation matrix of the pulse data, that is: ; where represents the th eigenvalue of the semantically associated forward and backward interaction fusion feature vectors of the pulse data, represents the th eigenvalue of the semantically associated forward and backward interaction fusion feature vectors of the pulse data, represents the semantically associated forward and backward interaction fusion feature vectors of the pulse data, Indicates the distance between eigenvalues, Indicates the eigenvalue at the th position in the forward and backward interactive fusion distance representation matrix of the semantic association of the pulse data.

[0052] Obtain the self - association matrix of the forward and backward interactive fusion of the semantic association of the pulse data, which is the pulse data semantic association forward and backward interactive fusion eigenvector as a row vector, that is ; where, Indicates the transposed vector of the forward and backward interactive fusion eigenvector of the semantic association of the pulse data, Indicates matrix multiplication, Indicates the self - association matrix of the forward and backward interactive fusion of the semantic association of the pulse data.

[0053] Multiply the forward and backward interactive fusion eigenvector of the semantic association of the pulse data with the forward and backward interactive fusion distance representation matrix of the semantic association of the pulse data to obtain the first - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data, that is ; where, Indicates the forward and backward interactive fusion distance representation matrix of the semantic association of the pulse data, Indicates the first - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data.

[0054] Multiply the first - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data with the matrix product of the forward and backward interactive fusion distance representation matrix of the semantic association of the pulse data and the self - association matrix of the forward and backward interactive fusion of the semantic association of the pulse data to obtain the multi - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data ; where, Indicates the multi - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data.

[0055] Dot - multiply the multi - level mapping vector of the forward and backward interactive fusion of the semantic association of the pulse data with the eigenvector of the forward and backward interactive fusion of the semantic association of the pulse data composed of the eigenvalues of the self - association matrix of the forward and backward interactive fusion of the semantic association of the pulse data to obtain the optimized forward and backward interactive fusion eigenvector of the semantic association of the pulse data, where interpolation or zero - filling is performed when the eigenvalues are insufficient.

[0056] That is, through the linear objective mapping representation of the similarity distance representation matrix of the forward and backward interactive fusion feature vectors semantically associated with the pulse data, the complete similarity instantiation of the self-association of the forward and backward interactive fusion feature vectors semantically associated with the pulse data is represented by the quadratic objective mapping based on the multi-level distribution hierarchy, and the associated fusion kernel bias is used to compensate for the negative influence factor of the association mismatch, so as to improve the eigenvalue instance judgment degree of the forward and backward interactive fusion feature vectors semantically associated with the pulse data under the similarity constraint, that is, the significance degree of the eigenvalue as an instance for the classification regression judgment, and improve the accuracy of the recognition result obtained by the forward and backward interactive fusion feature vectors semantically associated with the pulse data through the pulse anomaly data recognizer based on the classifier.

[0057] In summary, step S120 is clarified, which extracts and divides the preliminarily screened abnormal pulse data from the time queue of the pulse data, and then performs semantic association time series coding, so as to intelligently judge whether to determine the preliminarily screened abnormal pulse data as abnormal pulse data according to the dynamic memory interaction representation between the forward time series features of the pulse data semantic association and the backward time series features of the pulse data semantic association. In this way, it can automatically adapt to individual physiological differences and environmental changes, reduce the dependence on manual intervention to adjust parameters, and enhance the robustness of the system when encountering unknown situations or newly emerging patterns. At the same time, it can continuously learn new patterns and trends from new data, and improve its performance in meeting personalized needs and adapting to dynamic environments.

[0058] In the above-mentioned neural network-based sleep quality monitoring and management method, in step S130, on the intelligent terminal, classify the time queue of the filtered pulse data to obtain a classification result, and establish a mapping relationship between the classification result and the sleep stage to obtain a sleep stage mapping result. It should be understood that the pulse data reflects the changes in the human physiological state, and different sleep stages (such as light sleep, deep sleep, rapid eye movement sleep, etc.) are accompanied by different physiological characteristics, including pulse frequency, rhythm, and volatility. Therefore, by classifying the filtered pulse data, the pulse characteristics corresponding to different sleep stages can be identified, thereby effectively inferring the sleep state. Among them, the time queue of the filtered pulse data usually contains cleaner and more effective pulse information, removing obvious outliers or noise. Next, use machine learning or deep learning algorithms (such as support vector machines, random forests, or neural networks, etc.) to classify these pulse data. The goal of the classification process is to identify the time series patterns of the pulse data, such as the fluctuation frequency and change trend of the pulse, and then classify it into specific sleep stages. The model uses the labeled pulse data and the corresponding sleep stage labels during training to learn the implicit relationship between the pulse data and the sleep stage. Once the pulse data is classified and the classification result is obtained, the next step is to map these classification results to the actual sleep stages. Usually, the division of sleep stages is comprehensively judged through multiple physiological indicators (such as electroencephalogram, electrocardiogram, etc.), and the pulse data provides auxiliary information. Therefore, by establishing a mapping relationship between the pulse data classification result and the sleep stage, the pulse signal can be associated with specific sleep states (such as deep sleep, light sleep, wakefulness, etc.). Finally, the sleep stage mapping result output by the intelligent terminal according to this mapping relationship can provide users with important health information such as their sleep quality and sleep depth.

[0059] In the above-mentioned neural network-based sleep quality monitoring and management method, in step S140, on the intelligent terminal, according to the sleep stage mapping result, the sleep quality of the monitored object is obtained, and based on the sleep quality, sleep improvement suggestions are provided for the monitored object. It should be understood that considering that the evaluation of sleep quality depends not only on the total sleep time but also on multiple factors such as the distribution, depth, and continuity of sleep stages. Therefore, in the technical solution of this application, the intelligent terminal obtains the scoring result of sleep quality (for example, good, average, poor, etc.) and provides specific sleep improvement suggestions for the user. For different sleep problems, the terminal can recommend a series of personalized suggestions. For example, if the user has too few deep sleep stages, the system may recommend increasing pre-sleep relaxation activities, avoiding excessive use of electronic devices, or suggesting adjusting the sleep environment (such as temperature, light, noise, etc.). If the user has insufficient REM sleep, it may be recommended to do appropriate exercise or adjust the evening eating habits. In this way, the intelligent terminal can not only accurately evaluate the user's sleep quality but also provide a personalized sleep improvement plan according to the actual situation, helping the user optimize the sleep structure, improve the overall health level, and thus achieve a better quality of life and health management effect.

[0060] In summary, the neural network-based sleep quality monitoring and management method according to the embodiments of the present application is elucidated. It uses a pulse monitoring device to collect the pulse data of the monitored object during sleep in real time, forms a time queue, and transmits it to the intelligent terminal. The intelligent terminal first screens out abnormal data to ensure data accuracy; then classifies the screened pulse data and establishes a mapping relationship between the classification results and sleep stages to generate a sleep stage mapping result. Finally, based on these mapping results, the sleep quality of the monitored object is evaluated, and targeted sleep improvement suggestions are provided. In this way, through precise data processing and analysis, the accuracy of sleep monitoring and the quality of personalized services can be effectively improved.

[0061] Figure 7 The system block diagram of the neural network-based sleep quality monitoring and management system according to the embodiments of the present application. As Figure 7As shown, the neural network-based sleep quality monitoring and management system 100 according to an embodiment of the present application includes: a pulse data time queue extraction module 110, configured to extract preliminarily screened abnormal pulse data from the time queue of the pulse data, and divide the time queue of the pulse data to obtain a time queue of forward pulse data and a time queue of backward pulse data; a semantic association time series encoding module 120, configured to perform semantic association time series encoding on the time queue of the forward pulse data and the time queue of the backward pulse data respectively according to the preliminarily screened abnormal pulse data to obtain a pulse data semantic association forward time series feature and a pulse data semantic association backward time series feature; a dynamic memory mechanism feature interaction module 130, configured to perform feature interaction of the dynamic memory mechanism on the pulse data semantic association forward time series feature and the pulse data semantic association backward time series feature to obtain a pulse data semantic association forward and backward interaction fusion feature, and obtain an identification result according to the pulse data semantic association forward and backward interaction fusion feature.

[0062] Here, those skilled in the art can understand that the specific operations of each step in the above neural network-based sleep quality monitoring and management system have been described in detail above with reference to Figures 1 to 6 the description of the neural network-based sleep quality monitoring and management method, and therefore, the repeated description thereof will be omitted.

[0063] As described above, the neural network-based sleep quality monitoring and management system 100 according to an embodiment of the present application can be implemented in various terminal devices. In one example, the neural network-based sleep quality monitoring and management system 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the neural network-based sleep quality monitoring and management system 100 can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the neural network-based sleep quality monitoring and management system 100 can also be one of the many hardware modules of the terminal device.

[0064] Alternatively, in another example, the neural network-based sleep quality monitoring and management system 100 and the terminal device can also be separate devices, and the neural network-based sleep quality monitoring and management system 100 can be connected to the terminal device through a wired and / or wireless network and transmit interaction information according to a predefined data format.

[0065] In the embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0066] The module described as a separation component may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place or distributed over multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0067] In addition, in each embodiment of the present invention, the functional modules can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware, or in the form of hardware plus software functional modules.

[0068] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0069] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any associated drawing marks in the claims should not be regarded as limiting the claimed rights.

[0070] In addition, obviously, the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. Words such as "second" are used to denote names and do not denote any particular order.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit of the technical solutions of the present invention.

Claims

1. A sleep quality monitoring and management method based on a neural network, comprising: Using a pulse monitoring device to collect pulse data of a monitored subject in real time during sleep to obtain a time queue of the pulse data, and transmitting the time queue of the pulse data to an intelligent terminal; at the intelligent terminal, filtering abnormal data from the time queue of the pulse data to obtain a time queue of filtered pulse data; at the intelligent terminal, classifying the time queue of filtered pulse data to obtain a classification result, and establishing a mapping relationship between the classification result and the sleep stage to obtain a sleep stage mapping result; In the intelligent terminal, the sleep quality of the monitored object is obtained according to the sleep stage mapping result, and sleep improvement suggestions are provided to the monitored object according to the sleep quality; characterized in that abnormal data is screened out from the time queue of the pulse data to obtain the screened pulse data, including: Extracting primary screening abnormal pulse data from the time queue of the pulse data, and dividing the time queue of the pulse data to obtain a time queue of forward pulse data and a time queue of backward pulse data; According to the primary screening abnormal pulse data, respectively, the time queue of the forward pulse data and the time queue of the backward pulse data are semantically associated with time series encoding to obtain pulse data semantically associated forward time series features and pulse data semantically associated backward time series features; Performing feature interaction of a dynamic memory mechanism on the pulse data semantically associated forward time series features and the pulse data semantically associated backward time series features to obtain pulse data semantically associated forward and backward interactive fusion features, and obtaining a recognition result based on the pulse data semantically associated forward and backward interactive fusion features; According to the initial screening abnormal pulse data, the time queue of the forward pulse data and the time queue of the backward pulse data are respectively semantically associated with time series encoding to obtain pulse data semantically associated forward time series features and pulse data semantically associated backward time series features, including: Taking the initial screening abnormal pulse data as the reference semantics, calculating the semantic association between the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the semantic association of the forward pulse data, and calculating the semantic association between the reference semantics and each backward pulse data in the time queue of the backward pulse data to obtain the time queue of the semantic association of the backward pulse data; The time queue of the forward pulse data semantic association and the time queue of the backward pulse data semantic association are respectively subjected to temporal semantic association encoding to obtain a pulse data semantic association forward temporal feature vector as the pulse data semantic association forward temporal feature and a pulse data semantic association backward temporal feature vector as the pulse data semantic association backward temporal feature.

2. The sleep quality monitoring and management method based on neural network according to claim 1, characterized in that: Extracting the primary screening abnormal pulse data from the time queue of the pulse data, and dividing the time queue of the pulse data to obtain a time queue of forward pulse data and a time queue of backward pulse data, including: Preliminarily determining abnormal pulse data in the time queue of the pulse data to obtain the primary screening abnormal pulse data; The time queue of the pulse data is divided with the initial screening abnormal pulse data as a node to obtain the time queue of the forward pulse data and the time queue of the backward pulse data.

3. The sleep quality monitoring and management method based on neural network according to claim 2 is characterized in that: The time queue of the forward pulse data semantic association and the time queue of the backward pulse data semantic association are respectively subjected to temporal semantic association coding to obtain a pulse data semantic association forward temporal feature vector as the pulse data semantic association forward temporal feature and a pulse data semantic association backward temporal feature vector as the pulse data semantic association backward temporal feature, including: The time queue of the semantic association of the forward pulse data is passed through a forward pulse semantic association time series encoder based on a forward LSTM model to obtain a forward time series feature vector of the semantic association of the pulse data; The time queue of the backward pulse data semantic association is passed through a backward pulse semantic association time series encoder based on a backward LSTM model to obtain the pulse data semantic association backward time series feature vector.

4. The sleep quality monitoring and management method based on neural network according to claim 3 is characterized in that: Performing feature interaction of the pulse data semantic association forward time series feature and the pulse data semantic association backward time series feature using a dynamic memory mechanism to obtain pulse data semantic association forward and backward interactive fusion features, and obtaining a recognition result based on the pulse data semantic association forward and backward interactive fusion features, including: Performing feature interaction of a dynamic memory mechanism on the pulse data semantic association forward time series feature vector and the pulse data semantic association backward time series feature vector to obtain a pulse data semantic association forward and backward interactive fusion feature vector as the pulse data semantic association forward and backward interactive fusion feature; The recognition result is obtained by interactively fusing feature vectors forward and backward according to the semantic association of the pulse data. The recognition result is used to indicate whether the initially screened abnormal pulse data is determined as abnormal pulse data.

5. The sleep quality monitoring and management method based on neural network according to claim 4 is characterized in that: Performing feature interaction of a dynamic memory mechanism on the pulse data semantically associated forward time series feature vector and the pulse data semantically associated backward time series feature vector to obtain a pulse data semantically associated forward and backward interactive fusion feature vector, including: Performing collaborative implicit feature extraction on the pulse data semantically associated forward time series feature vector and the pulse data semantically associated backward time series feature vector to obtain an implicit collaborative coding vector between the pulse data semantically associated forward-backward time series features; According to the implicit collaborative coding vector between the pulse data semantically associated forward-backward temporal features, dynamic attention modulation is performed on the pulse data semantically associated forward temporal feature vector and the pulse data semantically associated backward temporal feature vector to obtain the pulse data semantically associated forward-backward interactive fusion feature vector.

6. The sleep quality monitoring and management method based on neural network according to claim 5 is characterized in that: According to the implicit collaborative coding vector between the pulse data semantically associated forward-backward temporal features, the pulse data semantically associated forward temporal feature vector and the pulse data semantically associated backward temporal feature vector are dynamically attention modulated to obtain the pulse data semantically associated forward-backward interactive fusion feature vector, including: Writing the implicit collaborative coding vector between the pulse data semantically associated forward-backward time series features into a dynamic memory unit to obtain a pulse data semantically associated forward-backward time series dynamic key vector; Extracting the pulse data semantic association forward-backward time series dynamic key vector from the dynamic memory unit, and performing feature optimization based on the first converter on the pulse data semantic association forward time series feature vector and the pulse data semantic association forward-backward time series dynamic key vector to obtain an optimized pulse data semantic association forward time series feature vector; Extracting the pulse data semantic association forward-backward temporal dynamic key vector from the dynamic memory unit, and performing feature optimization based on the second converter on the pulse data semantic association backward temporal feature vector and the pulse data semantic association forward-backward temporal dynamic key vector to obtain an optimized pulse data semantic association backward temporal feature vector; The optimized pulse data semantic association forward time series feature vector and the optimized pulse data semantic association backward time series feature vector are subjected to feature linear interaction to obtain the pulse data semantic association forward and backward interaction fusion feature vector.

7. The sleep quality monitoring and management method based on neural network according to claim 6 is characterized in that: The recognition result is obtained according to the pulse data semantically associated forward and backward interactive fusion feature vector, including: passing the pulse data semantically associated forward and backward interactive fusion feature vector through a classifier-based pulse abnormality data identifier to obtain the recognition result.

8. A sleep quality monitoring and management system based on neural network, characterized in that: include: A pulse data time queue extraction module, used to extract primary screening abnormal pulse data from the pulse data time queue, and divide the pulse data time queue to obtain a forward pulse data time queue and a backward pulse data time queue; A semantic association temporal coding module, for performing semantic association temporal coding on the time queue of the forward pulse data and the time queue of the backward pulse data respectively according to the primary screening abnormal pulse data to obtain a pulse data semantic association forward temporal feature and a pulse data semantic association backward temporal feature; A dynamic memory mechanism feature interaction module is used to perform a dynamic memory mechanism feature interaction on the pulse data semantic association forward time series feature and the pulse data semantic association backward time series feature to obtain the pulse data semantic association forward and backward interaction fusion feature, and obtain a recognition result according to the pulse data semantic association forward and backward interaction fusion feature; According to the initial screening abnormal pulse data, the time queue of the forward pulse data and the time queue of the backward pulse data are respectively semantically associated with time series encoding to obtain pulse data semantically associated forward time series features and pulse data semantically associated backward time series features, including: Taking the initial screening abnormal pulse data as the reference semantics, calculating the semantic association between the reference semantics and each forward pulse data in the time queue of the forward pulse data to obtain the time queue of the semantic association of the forward pulse data, and calculating the semantic association between the reference semantics and each backward pulse data in the time queue of the backward pulse data to obtain the time queue of the semantic association of the backward pulse data; The time queue of the forward pulse data semantic association and the time queue of the backward pulse data semantic association are respectively subjected to temporal semantic association encoding to obtain a pulse data semantic association forward temporal feature vector as the pulse data semantic association forward temporal feature and a pulse data semantic association backward temporal feature vector as the pulse data semantic association backward temporal feature.

9. The sleep quality monitoring and management system based on neural network according to claim 8, characterized in that: The pulse data time queue extraction module comprises: Preliminarily determining abnormal pulse data in the time queue of the pulse data to obtain the primary screening abnormal pulse data; The time queue of the pulse data is divided with the initial screening abnormal pulse data as a node to obtain the time queue of the forward pulse data and the time queue of the backward pulse data.

Images