Time sequence signal abnormity determination method and device based on multi-dimensional trajectory diagram, equipment and medium

Abstract

The invention discloses a time sequence signal abnormity determination method and device based on a multi-dimensional trajectory diagram, equipment and a medium, and relates to the technical field of deep learning, and the method comprises the steps: obtaining a time sequence signal; the types of the time sequence signals comprise physiological signals, physical signals, environment signals, seismic signals and sound wave signals; the time sequence signals are preprocessed; the preprocessing comprises interference signal removal and signal baseline stabilization; constructing a multi-dimensional trajectory diagram, and sending the multi-dimensional trajectory diagram to the client, so that the client returns an auxiliary detection result; performing feature extraction on the multi-dimensional trajectory diagram, inputting time sequence features into a deep learning model, and outputting classification and anomaly detection results; the feature extraction comprises geometric feature extraction, frequency domain feature extraction and time domain statistical feature extraction; according to the method and the device, the problems of insufficient analysis capability, insufficient information fusion and poor anomaly detection accuracy can be solved, and the accuracy, the stability and the robustness of the anomaly detection of the time sequence signal are improved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method, device, equipment and medium for determining a time series signal anomaly based on a multi-dimensional trajectory graph. Background Art

[0002] Currently, physiological signals such as ECG (electrocardiogram), EEG (electroencephalogram), and EMG (electroencephalogram) are increasingly being used in health monitoring and disease early warning. However, existing physiological signal analysis methods mostly focus on time series analysis of a single signal source, making it difficult to effectively handle complex and dynamically changing multidimensional time series signals. Existing technologies face several key technical challenges, particularly when it comes to fusion and anomaly detection involving multiple signal sources. They are particularly limited and typically rely on single-lead or simple time series analysis. For example, ECG and EEG often interact with each other in complex ways, and traditional methods often fail to accurately capture their dynamic changes and potential health anomalies. Due to the complex and multidimensional time series characteristics of these signals, existing single-signal analysis methods are unable to provide comprehensive and in-depth analysis. Another significant issue with existing technologies is the lack of multidimensional signal analysis capabilities. When processing multidimensional time series signals, existing methods are often unable to effectively fuse data from multiple signal sources, resulting in information loss and reduced signal analysis accuracy.

[0003] Existing feature extraction methods are often limited to simple time-domain statistical features (such as mean and standard deviation) or basic spectral analysis, failing to fully exploit the complex time series characteristics of signals. For example, signal characteristics such as periodicity, frequency content, and trend changes are often overlooked, yet these features are crucial for accurately identifying abnormal patterns. These shortcomings make traditional methods unable to effectively capture the dynamic changes of complex time series signals, limiting the accuracy of anomaly detection.

[0004] From the above, it can be seen that how to solve the problems of insufficient analytical capabilities, insufficient information fusion and poor anomaly detection accuracy, improve the accuracy, stability and robustness of time series signal anomaly detection, and effectively process and analyze multi-dimensional time series signals are issues to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, device, equipment, and medium for determining time series signal anomalies based on multidimensional trajectory graphs. This method can address the problems of insufficient analytical capabilities, insufficient information fusion, and poor anomaly detection accuracy, improve the accuracy, stability, and robustness of time series signal anomaly detection, and effectively process and analyze multidimensional time series signals. The specific solution is as follows:

[0006] In a first aspect, the present application discloses a method for determining a time series signal anomaly based on a multi-dimensional trajectory graph, which is applied to a computer device and includes:

[0007] Acquiring different types of time-series signals; the types of time-series signals include physiological signals, physical signals, environmental signals, seismic signals, and acoustic signals;

[0008] Preprocessing the time series signal to obtain the processed time series signal; the preprocessing includes removing interference signals and stabilizing signal baselines;

[0009] constructing a multi-dimensional trajectory graph based on the processed time series signal, and sending the multi-dimensional trajectory graph to the client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph;

[0010] Performing feature extraction on the multidimensional trajectory graph to obtain temporal features, and inputting the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction;

[0011] A time series signal diagnosis report is generated based on the auxiliary detection results and the classification and abnormality detection results.

[0012] Optionally, the acquiring of different types of timing signals includes:

[0013] Acquire physiological signals from wearable devices or medical monitoring instruments in real time at a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, electrooculogram, and galvanic skin response;

[0014] Acquire physical signals from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration;

[0015] Obtaining environmental signals from environmental monitoring sensors used for climate monitoring, smart home, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality;

[0016] Real-time acquisition of seismic signals from seismic instruments or vibration sensors;

[0017] Acquire sound wave signals; the sound wave signals include audio signals, sonar signals, and ultrasonic signals.

[0018] Optionally, preprocessing the timing signal to obtain the processed timing signal includes:

[0019] Using a Butterworth filter or wavelet denoising technology to perform an interference signal removal operation on the time series signal to obtain the time series signal after the removal;

[0020] A smooth signal baseline operation is performed on the removed time series signal based on filtering to obtain the processed time series signal.

[0021] Optionally, constructing a multidimensional trajectory graph based on the processed time series signal includes:

[0022] Using the global characteristic lead signal in the processed time series signal as the horizontal axis and the local lead signal as the vertical axis to construct a two-dimensional trajectory diagram;

[0023] On the basis of the two-dimensional trajectory diagram, a time axis or other lead signals are added to construct a multi-dimensional trajectory diagram.

[0024] Optionally, the extracting features from the multidimensional trajectory graph includes:

[0025] Performing geometric feature extraction on the multidimensional trajectory graph; the geometric feature extraction includes curvature, closure, and shape complexity;

[0026] Furthermore, frequency domain feature extraction is performed on the multidimensional trajectory graph using a fast Fourier transform method or a wavelet transform method;

[0027] Furthermore, time domain statistical features are extracted from the multi-dimensional trajectory graph; the time domain statistical features extracted include mean, standard deviation, kurtosis, and skewness.

[0028] Optionally, the step of inputting the temporal features into a preset deep learning model to output classification and anomaly detection results includes:

[0029] Building a deep learning model based on machine learning algorithms and using convolutional neural networks and long short-term memory networks; the machine learning algorithms include support vector machines and K-nearest neighbor algorithms;

[0030] Input temporal features into the deep learning model for training and classification to output classification and anomaly detection results.

[0031] In a second aspect, the present application discloses a device for determining a time series signal anomaly based on a multi-dimensional trajectory graph, comprising:

[0032] A time series signal acquisition module, configured to acquire different types of time series signals, including physiological signals, physical signals, environmental signals, seismic signals, and acoustic wave signals;

[0033] A preprocessing module, configured to preprocess the time series signal to obtain the processed time series signal; the preprocessing includes removing interference signals and stabilizing signal baselines;

[0034] an auxiliary detection result acquisition module, configured to construct a multi-dimensional trajectory graph based on the processed time series signal, and send the multi-dimensional trajectory graph to a client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph;

[0035] A feature extraction module is used to extract features from the multidimensional trajectory graph to obtain temporal features, and input the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction;

[0036] A diagnosis report generation module is used to generate a time series signal diagnosis report based on the auxiliary detection results and the classification and abnormality detection results.

[0037] Optionally, the timing signal acquisition module includes:

[0038] A physiological signal acquisition module is used to collect physiological signals from wearable devices or medical monitoring instruments in real time according to a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, electrooculogram, and galvanic skin response;

[0039] A physical signal acquisition module is used to acquire physical signals from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration;

[0040] Environmental signal acquisition module, used to obtain environmental signals from environmental monitoring sensors used for climate monitoring, smart home, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality;

[0041] A seismic signal acquisition module, used to collect seismic signals from seismic instruments or vibration sensors in real time;

[0042] The acoustic signal acquisition module is used to acquire acoustic signals; the acoustic signals include audio signals, sonar signals, and ultrasonic signals.

[0043] In a third aspect, the present application discloses an electronic device, comprising:

[0044] Memory, used to store computer programs;

[0045] A processor is used to execute the computer program to implement the aforementioned method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram.

[0046] In a fourth aspect, the present application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned method for determining a timing signal anomaly based on a multi-dimensional trajectory diagram are implemented.

[0047] It can be seen that the present application provides a method for determining the abnormality of a time series signal based on a multi-dimensional trajectory graph, including obtaining different types of time series signals; the types of the time series signals include physiological signals, physical signals, environmental signals, seismic signals and acoustic signals; preprocessing the time series signals to obtain the processed time series signals; the preprocessing includes interference signal removal and a stable signal baseline; constructing a multi-dimensional trajectory graph based on the processed time series signal, and sending the multi-dimensional trajectory graph to the client so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph; performing feature extraction on the multi-dimensional trajectory graph to obtain time series features, and inputting the time series features into a preset deep learning model to output classification and abnormality detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction and time domain statistical feature extraction; generating a time series signal diagnosis report based on the auxiliary detection results and the classification and abnormality detection results. This application obtains different types of time series signals to ensure the diversity, real-time and continuity of data, pre-processes the time series signals, removes interference signals, improves signal quality and accuracy, and constructs a multi-dimensional trajectory graph based on the processed time series signals. It can more comprehensively display the dynamic changes in space and time, improve the accuracy, stability and reliability of time series signal anomaly detection, obtain the auxiliary detection results returned by the client, extract features from the multi-dimensional trajectory graph, input the extracted features into a preset deep learning model to output classification and anomaly detection results, and generate a time series signal diagnostic report based on the auxiliary detection results and the classification and anomaly detection results, so as to solve the problems of insufficient analysis capabilities, insufficient information fusion and poor anomaly detection accuracy, and effectively process and analyze multi-dimensional time series signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0049] Figure 1 This is a flow chart of a method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram disclosed in this application;

[0050] Figure 2This is a structural diagram of a system for determining a time series signal anomaly based on a multi-dimensional trajectory diagram disclosed in this application;

[0051] Figure 3 This is a schematic structural diagram of a device for determining a time series signal anomaly based on a multi-dimensional trajectory diagram disclosed in this application;

[0052] Figure 4 This is a structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] See also Figure 1 As shown, an embodiment of the present invention discloses a method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram, which is applied to a computer device and may specifically include:

[0055] Step S11: Acquire different types of time-series signals; the types of time-series signals include physiological signals, physical signals, environmental signals, seismic signals, and acoustic wave signals.

[0056] In this embodiment, physiological signals are collected in real time from wearable devices or medical monitoring instruments according to a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, EOG (Electrooculogram), and EDA (Electrodermal Activity); physical signals are obtained from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration; environmental signals are obtained from environmental monitoring sensors used for climate monitoring, smart homes, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality; seismic signals are collected in real time from seismic instruments or vibration sensors; and sound wave signals are obtained; the sound wave signals include audio signals, sonar signals, and ultrasonic signals.

[0057] In this embodiment, physiological signals are collected in real time by various medical monitoring devices or wearable devices and are commonly used in fields such as health monitoring, disease diagnosis, and biofeedback. Physical signals include time-series data of physical quantities such as temperature, pressure, acceleration, and vibration. These signals can come from various sensors (such as temperature sensors, pressure sensors, accelerometers, gyroscopes, etc.) and are widely used in fields such as industrial equipment monitoring, smart hardware, and automated production lines. Environmental signals include time-series data related to the environment, such as noise, light, air pressure, humidity, and air quality. These data are acquired through environmental monitoring sensors and are commonly used in applications such as climate monitoring, smart homes, and environmental control. Seismic signals are used to monitor the propagation characteristics of seismic waves and play an important role in earthquake prediction and disaster warning systems. Acoustic signals include audio signals, sonar signals, and ultrasonic signals. Audio signals are commonly used in fields such as speech recognition, hearing testing, and environmental noise monitoring, while acoustic and ultrasonic signals are used in detection, imaging, and remote sensing technologies.

[0058] Taking ECG signals as an example, we focus on collecting ECG signals from limb leads (such as lead II) and precordial leads (such as leads V2, V4, and V5). Setting the sampling rate and resolution, for example, 360Hz and 11-bit, ensures the acquisition of high-quality, high-fidelity signals, providing a reliable data foundation for subsequent accurate analysis. Furthermore, in practical application scenarios, wearable devices or medical monitoring instruments can be used to achieve simultaneous acquisition of multiple lead signals, ensuring real-time and continuous data.

[0059] For example, EEG (electroencephalogram) signals can be used to construct a multidimensional feature space for monitoring neurological diseases. EEG signals contain rich information about brain activity, and different neurological diseases can cause changes in EEG signal characteristics such as frequency, amplitude, and phase. By collecting signals from multiple EEG electrodes, two-dimensional or multi-dimensional EEG signal feature maps can be constructed. Analyzing the morphology and variation patterns of these feature maps can be used to diagnose and monitor neurological diseases such as epilepsy and Parkinson's disease. Similarly, EMG (electromyography) signals can be used to diagnose muscle-related diseases. By collecting EMG signals from different muscle groups and constructing multidimensional EMG trajectory maps, and analyzing changes in EMG signals during muscle contraction and relaxation, this can help diagnose diseases such as muscle atrophy and weakness.

[0060] Step S12: pre-processing the time series signal to obtain the processed time series signal; the pre-processing includes removing interference signals and stabilizing signal baselines.

[0061] In this embodiment, a Butterworth filter or wavelet denoising technology is used to perform an interference signal removal operation on the timing series signal to obtain the timing series signal after removal; and a smooth signal baseline operation is performed on the timing series signal after removal based on filtering to obtain the processed timing series signal.

[0062] This application preprocesses the collected signal to remove noise components (such as baseline drift and power-frequency noise) to improve signal quality. Baseline drift, often caused by electrode movement or patient breathing, can affect signal accuracy. Filtering can stabilize the signal baseline. Power-frequency interference, which is 50 / 60Hz ambient electrical noise, can be removed to prevent interference with the valid signal.

[0063] In addition, other filtering algorithms, such as Kalman filtering, can also be used. Kalman filtering is an optimal linear filtering algorithm that is effective for processing noisy dynamic system data. While removing ECG signal noise, it can also predict and track signal trends. Deep learning-based detection algorithms, such as those based on CNN (Convolutional Neural Network), can also be used for QRS complex (ECG waveform characteristics of ventricular depolarization). CNNs can automatically learn the characteristics of QRS complexes and may have higher detection accuracy and robustness than traditional algorithms. This advantage may be particularly evident when detecting QRS waves in complex arrhythmias.

[0064] Step S13: constructing a multi-dimensional trajectory graph based on the processed time series signal, and sending the multi-dimensional trajectory graph to the client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph.

[0065] In this embodiment, the global characteristic lead signal in the processed timing signal is used as the horizontal axis, and the local lead signal is used as the vertical axis to construct a two-dimensional trajectory graph; on the basis of the two-dimensional trajectory graph, a time axis or other lead signals are added to construct a multi-dimensional trajectory graph, and the multi-dimensional trajectory graph is sent to the client and visualized so that the client can return auxiliary detection results based on the geometric features of the multi-dimensional trajectory graph; the geometric features include curvature, closure, and shape complexity.

[0066] For example, a two-dimensional ECG trajectory graph can be constructed, using each heartbeat as a unit, with a lead signal that reflects global characteristics (such as lead II) as the X-axis and a local lead signal (such as lead V2, V4, and V5) as the Y-axis. For example, when constructing a two-dimensional trajectory graph for leads II and V2, based on the manifestation of cardiac electrical activity in these two leads, a trajectory reflecting the spatiotemporal relationship of cardiac electrical activity is plotted on a two-dimensional plane. This allows the diagnostician to clearly observe abnormalities. This approach can demonstrate the spatiotemporal relationship between different signals and capture their common dynamic characteristics.

[0067] Based on the two-dimensional trajectory diagram, the time axis (t) or other related lead signals are added to construct a three-dimensional or higher-dimensional ECG trajectory. For example, using lead II (X-axis), lead V4 (Y-axis), and lead V2 (Z-axis) to construct a three-dimensional ECG trajectory diagram can more comprehensively display the dynamic changes in cardiac electrical activity in space and time, improving the ability to identify complex arrhythmias. In the case of complex arrhythmias, changes in the three-dimensional trajectory can more accurately reflect the abnormalities of electrical activity in different parts of the heart, providing richer information for diagnosis. Multi-dimensional trajectory diagrams can more comprehensively present the temporal and spatial relationships of signals, making them particularly suitable for analyzing complex signals.

[0068] For example, a multidimensional trajectory diagram can be presented to a doctor, who, based on their expertise, can make a diagnosis and then return the auxiliary test results. For example, abnormal changes in curvature in a two-dimensional trajectory diagram may indicate a problem with the conduction pathway of the heart's electrical activity; changes in closure may be associated with conditions such as myocardial ischemia. The doctor, combined with clinical experience, can make a comprehensive assessment of the patient's heart health.

[0069] This application transcends the limitations of traditional single-lead or simple multi-lead analysis by integrating the timing signals of individual leads (such as lead II and leads V2, V4, and V5) to construct a multidimensional ECG trajectory map. By mapping the signals from different leads onto multidimensional coordinate axes, the spatial relationships of cardiac electrical activity are intuitively displayed, fully capturing the dynamic changes in cardiac electrical activity across time and space. This multidimensional trajectory map construction technology fully exploits the nonlinear, time-varying, and non-causal characteristics of ECG signals, enhancing the detection of complex arrhythmias and other conditions.

[0070] Furthermore, in terms of trajectory feature extraction, in addition to using geometric features such as trajectory closure, curvature, and shape complexity, other feature extraction methods can also be employed. For example, frequency domain features can be used for analysis. By performing frequency domain transformations such as Fourier transforms on multidimensional trajectory graphs, features such as the energy distribution of the trajectory at different frequency components can be obtained. Different arrhythmia types may have unique characteristics in the frequency domain. By analyzing these frequency domain features, arrhythmias can be classified and diagnosed.

[0071] Step S14: extract features from the multidimensional trajectory graph to obtain temporal features, and input the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction.

[0072] In this embodiment, geometric features are extracted from the multidimensional trajectory graph; the geometric features extracted include curvature, closure, and shape complexity; and frequency domain features are extracted from the multidimensional trajectory graph using the FFT (Fast Fourier Transform) method or the wavelet transform method; and time domain statistical features are extracted from the multidimensional trajectory graph; the time domain statistical features extracted include mean, standard deviation, kurtosis, and skewness; a deep learning model is constructed based on a machine learning algorithm and using a convolutional neural network and a long short-term memory network; the machine learning algorithm includes a support vector machine and a K-nearest neighbor algorithm; the temporal features are input into the deep learning model for training and classification to output classification and anomaly detection results.

[0073] Geometric features reflect the geometric changes of signal trajectories in multidimensional space and can help identify abnormal changes in signals; frequency domain features can reveal the frequency change pattern of signals, which is especially important in periodic anomaly detection; time domain statistical features can reflect the fluctuation of signals and help identify sudden anomalies.

[0074] In this embodiment, deep learning models, such as CNNs and LSTMs (Long Short Term Memory Networks), are used to train and classify features. Deep learning models can automatically learn patterns that distinguish normal and abnormal signals from features, improving classification accuracy and robustness through training.

[0075] During the training process, the model learns the characteristic patterns of normal and abnormal trajectory maps, and improves the accuracy and reliability of classification by learning from a large amount of sample data.

[0076] To address the problem that traditional ECG recognition methods are susceptible to interference from baseline drift, electrode contact noise, and other factors, the present invention uses multidimensional trajectory morphology for classification, making it insensitive to short-term noise. By fusing multi-lead signals to construct a trajectory graph, multi-dimensional information is used to complement and verify each other, reducing the impact of noise on detection results. This application uses unique trajectory geometric features (such as trajectory closure, curvature changes, and shape complexity) to analyze different types of arrhythmias. These features can accurately depict the morphological changes of the multidimensional trajectory graph and reflect abnormalities in cardiac electrical activity. By deeply exploring trajectory features, normal and abnormal heart rhythms can be effectively distinguished, improving classification accuracy. For example, the smoothness of the trajectory can be used to infer the presence of abnormal heart rate.

[0077] Step S15: Generate a time series signal diagnosis report based on the auxiliary detection results and the classification and abnormality detection results.

[0078] In this embodiment, the auxiliary detection results, classification results, and anomaly detection results are collated and output to generate a detailed time series signal diagnosis report. The deep learning model can combine data from multiple signal sources and feature extraction to accurately determine whether there are potential abnormal patterns and indicate the type and severity of the anomaly. These detection results are combined with the auxiliary diagnosis results to generate a comprehensive time series signal diagnosis report.

[0079] The timing signal diagnostic report mainly includes the following core contents:

[0080] 1. Diagnostic results of abnormal signals: Based on the classification results of the model, the report will list in detail the abnormalities detected in the signal. Specific examples include but are not limited to the following: physiological abnormalities, such as arrhythmias (atrial fibrillation, ventricular premature beats, etc.), epileptic seizures, muscle atrophy, and neurological diseases (such as Parkinson's disease). Potential health problems can be identified through analysis of physiological signals such as electrocardiograms, electroencephalograms, and electromyograms. Equipment failures, such as sensor failures and signal acquisition errors, can be detected through analysis of irregular changes in signals to detect possible equipment failures or data loss. Environmental abnormalities, such as excessive ambient noise, abnormal temperature and humidity, can be detected through environmental monitoring signals to detect the impact of potential external environmental factors on system or equipment operation. Physical signal abnormalities, such as abnormal vibrations and temperature fluctuations, can help identify faults in industrial equipment or sensors.

[0081] 2. Diagnostic recommendations and action guidance: Medical recommendations: Based on the diagnostic results, the system will provide specific medical recommendations to doctors. For example, if arrhythmia is detected, the report may recommend further ECG monitoring or drug treatment; if an epileptic seizure is detected, the report may recommend hospitalization for observation or adjustment of the treatment plan; Further examination recommendations: For some uncertain abnormalities, the system will recommend more examinations or tests, such as more detailed cardiac or EEG examinations, or laboratory tests to confirm the diagnosis; Early warning and emergency treatment recommendations: In some emergency situations, such as detecting abnormal fluctuations in vital signs (such as the risk of cardiac arrest), the system will issue an emergency warning and recommend emergency medical intervention measures;

[0082] 3. Personalized health management recommendations: Based on abnormal test results, the report can also provide personalized health management recommendations. These recommendations may include lifestyle adjustments, regular check-up recommendations, exercise and diet guidance, etc., to help patients better manage their health and prevent the occurrence or recurrence of diseases;

[0083] 4. Data visualization and interactive functions: The report will provide clear charts and visual analysis to help doctors intuitively understand abnormal patterns in the signal. For example, abnormal waveforms in the electrocardiogram or electroencephalogram can be displayed through multi-dimensional trajectory diagrams or feature change diagrams, allowing doctors to quickly determine the type of abnormality and the scope of impact. At the same time, the report also provides interactive functions, allowing doctors to conduct in-depth analysis of the test results, such as zooming in to view abnormal signals and comparing signal changes in different time periods, further improving diagnostic accuracy;

[0084] 5. Historical Data Comparison and Trend Analysis: The system can also compare a patient's historical signal data with current data, generating trend analysis reports to help doctors identify long-term health trends. For example, if a patient's electrocardiogram frequently shows certain waveform changes, the report will indicate whether these changes are consistent with long-term health trends or are related to certain known conditions.

[0085] This application provides a means of visualizing multidimensional ECG trajectory graphs. While traditional ECG analysis relies on numerical calculations, this invention intuitively displays trajectory graphs, allowing doctors to directly observe the trajectory characteristics of arrhythmias and quickly identify abnormal rhythms. During the visualization process, an interactive design facilitates doctors to scale, annotate, and analyze the trajectory graphs, enhancing their clinical application value. Furthermore, combining visualization technology with an expert system provides doctors with diagnostic advice and reference information, assisting them in making more accurate diagnostic decisions.

[0086] The structure of the system for determining abnormality of time series signals based on multi-dimensional trajectory diagram in this application is as follows: Figure 2As shown, it includes a data acquisition module, a signal processing module, a trajectory calculation module, a classification module, and a result output module. This application protects a method of constructing a multidimensional trajectory diagram based on each lead, such as a method of constructing an ECG two-dimensional trajectory diagram based on limb leads (I, II, III, avL, avR, AVF) and precordial leads (V2, V4, V5). The method of constructing the trajectory diagram is clear, that is, the X1 axis is the limb lead signal value, and the X2 axis is the precordial lead signal value; the ECG abnormality features are extracted through the geometric morphological features of the two-dimensional or multi-dimensional trajectory (curvature, closure, fractal dimension, etc.), and are used for the automatic classification of arrhythmias. This method is not only applicable to common types of arrhythmias such as atrial fibrillation, atrial flutter, and ventricular arrhythmias, but also includes the classification and diagnosis of complex arrhythmia combinations. It uses traditional machine learning (SVM, KNN) and deep learning (CNN, LSTM) models to classify heart rhythms using two-dimensional or multi-dimensional ECG trajectory data, and protects the training, optimization, and application processes based on these models, including selecting appropriate model parameters, preprocessing trajectory data to meet model requirements, and improving classification accuracy through model fusion. It intuitively displays arrhythmia characteristics through two-dimensional or multi-dimensional trajectory graphs to enhance visual analysis capabilities, and protects methods and processes for using trajectory graphs to assist doctors in quickly determining the type and severity of arrhythmias and providing an intuitive basis for diagnosis, including trajectory graph display methods, annotation methods, and methods for interacting with doctors. It protects methods for diagnosing equipment faults by fusing multi-dimensional periodic time series signals. It protects methods for analyzing human sleep and rhythms using multi-dimensional time-series physiological signals (including but not limited to electrodermal, electroencephalographic, electrooculographic, and electromyographic signals). It also provides real-time ECG trajectory detection suitable for various scenarios.

[0087] The present invention constructs a multi-dimensional trajectory diagram based on the fusion of time series signals for physiological abnormality monitoring, which has significant advantages over traditional methods. In terms of arrhythmia detection, traditional methods rely on single leads or simple time series analysis, which makes it difficult to identify complex abnormalities. The present invention fuses multi-lead signals to construct a multi-dimensional trajectory diagram, making full use of the nonlinear, time-varying and non-causal characteristics of electrocardiogram signals. Taking the two-dimensional trajectory diagram as an example, the combination of lead II and precordial leads can reflect the electrical activity of different parts of the heart. When the heart has arrhythmia, such as atrial fibrillation, the complex relationship between the signals will cause the shape complexity, closure and other characteristics of the trajectory to change, thereby accurately identifying abnormalities and improving detection accuracy.

[0088] From a visualization perspective, traditional ECG analysis relies on numerical calculations and lacks intuitiveness. The multidimensional trajectory diagram of the present invention can intuitively display abnormal patterns, allowing doctors and AI (Artificial Intelligence) systems to directly observe and analyze them. For example, when diagnosing myocardial ischemia, changes in closure and curvature on the two-dimensional trajectory diagram can be clearly presented, reducing diagnostic difficulty and improving diagnostic efficiency. Furthermore, the present invention is highly robust to noise, while traditional methods are susceptible to noise interference such as baseline drift. The multidimensional trajectory diagram integrates multi-lead signals, so even if some leads are affected by noise, the information from other leads can still ensure the accuracy of the overall analysis. Furthermore, the present invention has low computational complexity, while traditional deep learning models are computationally complex and difficult to apply in real time. By optimizing the computational process, the present invention is suitable for portable devices and cloud-based AI analysis, enabling real-time monitoring, providing patients with timely diagnosis and health management, and promoting the development of telemedicine.

[0089] In the hospital's ECG monitoring system, the technical solution of the present invention has been preliminarily applied in the monitoring of some patients with arrhythmias. For example, when using traditional ECG monitoring methods for a patient suffering from paroxysmal atrial fibrillation, it is sometimes difficult to accurately capture the abnormal signals during an attack due to the intermittent and transient nature of atrial fibrillation attacks. However, after adopting a monitoring method based on a multi-dimensional trajectory graph, by continuously monitoring the patient's multi-lead ECG signals and constructing two-dimensional and three-dimensional trajectory graphs, the characteristic changes of the trajectory graph during an atrial fibrillation attack can be clearly observed. In the two-dimensional trajectory graph, the shape complexity of the trajectory during an atrial fibrillation attack increases significantly, and the closure also changes; in the three-dimensional trajectory graph, the disordered changes in the space of the cardiac electrical activity can be more intuitively seen. These characteristics provide an important basis for doctors to promptly detect atrial fibrillation attacks and formulate treatment plans.

[0090] In telemedicine scenarios, the technical solution of the present invention also has broad application prospects. For example, the patient's ECG signal is collected through a wearable ECG acquisition device, and then the data is transmitted to the cloud for analysis. In a telemedicine practice, a patient in a remote area used a smart bracelet to collect ECG signals at home, and the bracelet transmitted the collected data to the cloud in real time. The cloud system used the technology of the present invention to construct a multi-dimensional trajectory map and perform analysis and diagnosis. The results showed that the patient's ECG trajectory map was abnormal, and ventricular premature beats were suspected. Based on the diagnosis results, the doctor contacted the patient in time and recommended that the patient go to the local hospital for further examination. In the end, the patient was diagnosed and received timely treatment at the local hospital. This case fully demonstrates the feasibility and effectiveness of the technology of the present invention in telemedicine, and can provide timely medical services to patients in remote areas.

[0091] In this embodiment, different types of timing signals are obtained; the types of the timing signals include physiological signals, physical signals, environmental signals, seismic signals and acoustic signals; the timing signals are preprocessed to obtain the processed timing signals; the preprocessing includes interference signal removal and a stable signal baseline; a multidimensional trajectory graph is constructed based on the processed timing signals, and the multidimensional trajectory graph is sent to the client so that the client returns auxiliary detection results based on the multidimensional trajectory graph; feature extraction is performed on the multidimensional trajectory graph to obtain timing features, and the timing features are input into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction and time domain statistical feature extraction; a timing signal diagnosis report is generated based on the auxiliary detection results and the classification and anomaly detection results. This application obtains different types of time series signals to ensure the diversity, real-time and continuity of data, pre-processes the time series signals, removes interference signals, improves signal quality and accuracy, and constructs a multi-dimensional trajectory graph based on the processed time series signals. It can more comprehensively display the dynamic changes in space and time, improve the accuracy, stability and reliability of time series signal anomaly detection, obtain the auxiliary detection results returned by the client, extract features from the multi-dimensional trajectory graph, input the extracted features into a preset deep learning model to output classification and anomaly detection results, and generate a time series signal diagnostic report based on the auxiliary detection results and the classification and anomaly detection results, so as to solve the problems of insufficient analysis capabilities, insufficient information fusion and poor anomaly detection accuracy, and effectively process and analyze multi-dimensional time series signals.

[0092] See also Figure 3 As shown, an embodiment of the present invention discloses a device for determining a time series signal anomaly based on a multi-dimensional trajectory diagram, which may specifically include:

[0093] A time series signal acquisition module 11 is used to acquire different types of time series signals; the types of time series signals include physiological signals, physical signals, environmental signals, seismic signals and acoustic signals;

[0094] A preprocessing module 12 is used to preprocess the time series signal to obtain the processed time series signal; the preprocessing includes removing interference signals and stabilizing signal baselines;

[0095] An auxiliary detection result acquisition module 13 is configured to construct a multi-dimensional trajectory map based on the processed time series signal, and send the multi-dimensional trajectory map to the client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory map;

[0096] A feature extraction module 14 is configured to extract features from the multidimensional trajectory graph to obtain temporal features, and input the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction;

[0097] The diagnosis report generating module 15 is used to generate a time series signal diagnosis report based on the auxiliary detection results and the classification and abnormality detection results.

[0098] In this embodiment, different types of timing signals are obtained; the types of the timing signals include physiological signals, physical signals, environmental signals, seismic signals and acoustic signals; the timing signals are preprocessed to obtain the processed timing signals; the preprocessing includes interference signal removal and a stable signal baseline; a multidimensional trajectory graph is constructed based on the processed timing signals, and the multidimensional trajectory graph is sent to the client so that the client returns auxiliary detection results based on the multidimensional trajectory graph; feature extraction is performed on the multidimensional trajectory graph to obtain timing features, and the timing features are input into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction and time domain statistical feature extraction; a timing signal diagnosis report is generated based on the auxiliary detection results and the classification and anomaly detection results. This application obtains different types of time series signals to ensure the diversity, real-time and continuity of data, pre-processes the time series signals, removes interference signals, improves signal quality and accuracy, and constructs a multi-dimensional trajectory graph based on the processed time series signals. It can more comprehensively display the dynamic changes in space and time, improve the accuracy, stability and reliability of time series signal anomaly detection, obtain the auxiliary detection results returned by the client, extract features from the multi-dimensional trajectory graph, input the extracted features into a preset deep learning model to output classification and anomaly detection results, and generate a time series signal diagnostic report based on the auxiliary detection results and the classification and anomaly detection results, so as to solve the problems of insufficient analysis capabilities, insufficient information fusion and poor anomaly detection accuracy, and effectively process and analyze multi-dimensional time series signals.

[0099] In some specific embodiments, the timing signal acquisition module 11 may specifically include:

[0100] A physiological signal real-time acquisition module is used to collect physiological signals from wearable devices or medical monitoring instruments in real time according to a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, electrooculogram, and skin galvanic response;

[0101] A physical signal acquisition module is used to acquire physical signals from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration;

[0102] Environmental signal acquisition module, used to obtain environmental signals from environmental monitoring sensors used for climate monitoring, smart home, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality;

[0103] A real-time seismic signal acquisition module is used to acquire seismic signals from seismic instruments or vibration sensors in real time;

[0104] The acoustic signal acquisition module is used to acquire acoustic signals; the acoustic signals include audio signals, sonar signals, and ultrasonic signals.

[0105] In some specific embodiments, the pre-processing module 12 may specifically include:

[0106] An interference signal removal module is used to perform an interference signal removal operation on the time series signal using a Butterworth filter or a wavelet denoising technique to obtain the time series signal after the removal;

[0107] The steady signal baseline module is used to perform a steady signal baseline operation on the removed time series signal based on filtering to obtain the processed time series signal.

[0108] In some specific embodiments, the auxiliary detection result acquisition module 13 may specifically include:

[0109] a two-dimensional trajectory diagram construction module, configured to use the global characteristic lead signal in the processed time series signal as the horizontal axis and the local lead signal as the vertical axis to construct a two-dimensional trajectory diagram;

[0110] The multi-dimensional trajectory diagram construction module is used to add a time axis or other lead signals on the basis of the two-dimensional trajectory diagram to construct a multi-dimensional trajectory diagram.

[0111] In some specific embodiments, the feature extraction module 14 may specifically include:

[0112] A geometric feature extraction module is used to extract geometric features from the multi-dimensional trajectory graph; the geometric feature extraction includes curvature, closure, and shape complexity;

[0113] A frequency domain feature extraction module is used to extract frequency domain features from the multidimensional trajectory graph using a fast Fourier transform method or a wavelet transform method;

[0114] The time domain statistical feature extraction module is used to extract time domain statistical features from the multidimensional trajectory graph; the time domain statistical features extracted include mean, standard deviation, kurtosis, and skewness.

[0115] In some specific embodiments, the feature extraction module 14 may specifically include:

[0116] A deep learning model building module, for building a deep learning model based on machine learning algorithms and using convolutional neural networks and long short-term memory networks; the machine learning algorithms include support vector machines and K-nearest neighbor algorithms;

[0117] The classification and anomaly detection module is used to input temporal features into the deep learning model for training and classification, and output classification and anomaly detection results.

[0118] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram, performed by the electronic device, as disclosed in any of the aforementioned embodiments.

[0119] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0120] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.

[0121] Among them, the operating system 221 is used to manage and control the various hardware devices and computer programs 222 on the electronic device 20 to realize the calculation and processing of the data 223 in the memory 22 by the processor 21, which can be Windows, Unix, Linux, etc. In addition to including a computer program that can be used to complete the method for determining the timing signal anomaly based on the multi-dimensional trajectory diagram performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks. In addition to including data transmitted from an external device received by the timing signal anomaly determination device based on the multi-dimensional trajectory diagram, the data 223 can also include data collected by its own input and output interface 25, etc.

[0122] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0123] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium, in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the method for determining a timing signal anomaly based on a multi-dimensional trajectory diagram disclosed in any of the aforementioned embodiments are implemented.

[0124] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0125] The above is a detailed introduction to the method, device, equipment and storage medium for determining a time-series signal anomaly based on a multi-dimensional trajectory diagram provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for determining time series signal anomalies based on a multi-dimensional trajectory graph, characterized in that: Applicable to computer devices, including: Acquiring different types of time-series signals; the types of time-series signals include physiological signals, physical signals, environmental signals, seismic signals, and acoustic signals; Preprocessing the time series signal to obtain the processed time series signal; the preprocessing includes removing interference signals and stabilizing signal baselines; constructing a multi-dimensional trajectory graph based on the processed time series signal, and sending the multi-dimensional trajectory graph to the client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph; Performing feature extraction on the multidimensional trajectory graph to obtain temporal features, and inputting the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction; A time series signal diagnosis report is generated based on the auxiliary detection results and the classification and abnormality detection results.

2. The method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to claim 1, characterized in that: The acquisition of different types of timing signals includes: Acquire physiological signals from wearable devices or medical monitoring instruments in real time at a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, electrooculogram, and galvanic skin response; Acquire physical signals from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration; Obtaining environmental signals from environmental monitoring sensors used for climate monitoring, smart home, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality; Real-time acquisition of seismic signals from seismic instruments or vibration sensors; Acquire sound wave signals; the sound wave signals include audio signals, sonar signals, and ultrasonic signals.

3. The method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to claim 1, characterized in that: The preprocessing of the timing signal to obtain the processed timing signal includes: Using a Butterworth filter or wavelet denoising technology to perform an interference signal removal operation on the time series signal to obtain the time series signal after the removal; A smooth signal baseline operation is performed on the removed time series signal based on filtering to obtain the processed time series signal.

4. The method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to claim 1, wherein: The constructing of a multi-dimensional trajectory graph based on the processed time series signal includes: Using the global characteristic lead signal in the processed time series signal as the horizontal axis and the local lead signal as the vertical axis to construct a two-dimensional trajectory diagram; On the basis of the two-dimensional trajectory diagram, a time axis or other lead signals are added to construct a multi-dimensional trajectory diagram.

5. The method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to claim 1, characterized in that: The extracting features from the multi-dimensional trajectory graph includes: Performing geometric feature extraction on the multidimensional trajectory graph; the geometric feature extraction includes curvature, closure, and shape complexity; Furthermore, frequency domain feature extraction is performed on the multidimensional trajectory graph using a fast Fourier transform method or a wavelet transform method; Furthermore, time domain statistical features are extracted from the multi-dimensional trajectory graph; the time domain statistical features extracted include mean, standard deviation, kurtosis, and skewness.

6. The method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to any one of claims 1 to 5, characterized in that: Inputting the temporal features into a preset deep learning model to output classification and anomaly detection results includes: Building a deep learning model based on machine learning algorithms and using convolutional neural networks and long short-term memory networks; the machine learning algorithms include support vector machines and K-nearest neighbor algorithms; Input temporal features into the deep learning model for training and classification to output classification and anomaly detection results.

7. A device for determining abnormality of a time series signal based on a multi-dimensional trajectory diagram, characterized in that: include: A timing signal acquisition module is used to acquire different types of timing signals; The types of the time series signals include physiological signals, physical signals, environmental signals, seismic signals and acoustic wave signals; A preprocessing module, configured to preprocess the time series signal to obtain the processed time series signal; The preprocessing includes removing interference signals and stabilizing signal baselines; an auxiliary detection result acquisition module, configured to construct a multi-dimensional trajectory graph based on the processed time series signal, and send the multi-dimensional trajectory graph to a client, so that the client returns an auxiliary detection result based on the multi-dimensional trajectory graph; A feature extraction module is used to extract features from the multidimensional trajectory graph to obtain temporal features, and input the temporal features into a preset deep learning model to output classification and anomaly detection results; the feature extraction includes geometric feature extraction, frequency domain feature extraction, and time domain statistical feature extraction; A diagnosis report generation module is used to generate a time series signal diagnosis report based on the auxiliary detection results and the classification and abnormality detection results.

8. The device for determining anomalies of time series signals based on a multi-dimensional trajectory diagram according to claim 7, characterized in that: The timing signal acquisition module includes: A physiological signal acquisition module is used to collect physiological signals from wearable devices or medical monitoring instruments in real time according to a preset sampling rate and resolution; the physiological signals include electrocardiogram, electroencephalogram, electromyogram, electrooculogram, and galvanic skin response; A physical signal acquisition module is used to acquire physical signals from various sensors; the sensors include temperature sensors, pressure sensors, accelerometers, and gyroscopes; the physical signals include temperature, pressure, acceleration, and vibration; Environmental signal acquisition module, used to obtain environmental signals from environmental monitoring sensors used for climate monitoring, smart home, and environmental control; the environmental signals include noise, light, air pressure, humidity, and air quality; A seismic signal acquisition module, used to collect seismic signals from seismic instruments or vibration sensors in real time; The acoustic signal acquisition module is used to acquire acoustic signals; the acoustic signals include audio signals, sonar signals, and ultrasonic signals.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for determining a time series signal anomaly based on a multi-dimensional trajectory diagram according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that Used to store a computer program; wherein, when the computer program is executed by a processor, the method for determining a timing signal anomaly based on a multi-dimensional trajectory diagram according to any one of claims 1 to 6 is implemented.

Images