Medical monitoring sensing data association storage method and system based on machine learning

Abstract

The invention provides a medical monitoring sensing data association storage method and system based on machine learning, and relates to the field of data processing. Medical monitoring data from different sensors are automatically identified, associated and stored by using a machine learning technology. By constructing the data association model, the relationship between the data can be accurately found, and a comprehensive data analysis view angle is provided for medical workers. According to the invention, the utilization efficiency and value of medical monitoring data are improved, and potential medical problems and risks can be found.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a medical monitoring sensor data correlation storage method and system based on machine learning. BACKGROUND

[0002] With the popularity of wearable devices, collecting and analyzing user sensor data has become an important means to improve the quality of medical services and conduct personalized diagnosis and treatment. User sensor data usually includes multiple categories of sensor signals, such as heart rate, blood pressure, step count, blood oxygen, HRV, etc. These data are of great significance to understanding the health status of patients, predicting disease risk, and developing personalized diagnosis and treatment plans.

[0003] In practical applications, how to effectively extract key data related to specific health events or symptoms from massive and complex sensor data remains a great challenge. Traditional data analysis methods often adopt a one-size-fits-all approach to processing all data, ignoring the differences between different categories of sensor signals and their different contributions to health events. Not only may this lead to the omission of key information, but it may also mask potential patterns and trends in the data due to homogenization. SUMMARY

[0004] The purpose of the present application is to provide a medical monitoring sensor data correlation storage method and system based on machine learning.

[0005] The present application is implemented as follows: In a first aspect, the present application provides a medical monitoring sensor data correlation storage method based on machine learning, comprising: obtaining target sensor signals related to a to-be-processed sensor signal, the target sensor signals covering different categories of sensor signals; obtaining the correlation between the different categories of sensor signals covered by the target sensor signals and the to-be-processed sensor signal according to the influence factors of the different categories of sensor signals; selecting a selected category of sensor signals from the different categories of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal; correlation storing the signal segments related to the to-be-processed sensor signal in the selected category of sensor signals; wherein the correlation storing the signal segments related to the to-be-processed sensor signal in the selected category of sensor signals comprises: retrieving multiple signal segments related to the to-be-processed sensor signal from the selected category of sensor signals; determining a continuity evaluation parameter between the plurality of signal segments according to signal coordinates of the plurality of signal segments in the selected category of sensor signals; when the continuity evaluation parameter between the plurality of signal segments reaches a preset standard, storing the signal segment related to the to-be-processed sensor signal in the selected category of sensor signals in association.

[0006] Optionally, the obtaining the correlation between the different categories of sensor signals covered by the target sensor signal and the to-be-processed sensor signal according to the influence factor of the different categories of sensor signals comprises: performing signal segment splitting on the to-be-processed sensor signal to obtain a to-be-processed signal segment; obtaining a basic correlation between the different categories of sensor signals and the to-be-processed sensor signal according to the to-be-processed signal segment; obtaining the correlation between the different categories of sensor signals and the to-be-processed sensor signal according to the basic correlation between the different categories of sensor signals and the to-be-processed sensor signal and the influence factor of the different categories of sensor signals.

[0007] Optionally, the obtaining the basic correlation between the different categories of sensor signals and the to-be-processed sensor signal according to the to-be-processed signal segment comprises: retrieving a target signal segment related to the to-be-processed signal segment from the different categories of sensor signals; determining the basic correlation between the different categories of sensor signals and the to-be-processed sensor signal according to a priority of the to-be-processed signal segment and a priority of the target signal segment related to the to-be-processed signal segment in the different categories of sensor signals.

[0008] Optionally, the target sensor signal is a plurality of target sensor signals, and the method further comprises: determining an influence factor of each target sensor signal according to a correlation mechanism between the to-be-processed sensor signal and the each target sensor signal; obtaining the correlation between the each target sensor signal and the to-be-processed sensor signal according to the influence factor of the each target sensor signal and the correlation between the different categories of sensor signals covered by the each target sensor signal and the to-be-processed sensor signal; sorting the plurality of target sensor signals according to the correlation between the each target sensor signal and the to-be-processed sensor signal; the obtaining the target sensor signal related to the to-be-processed sensor signal comprises: Splitting the to-be-processed sensor signal into signal segments to obtain a set of to-be-processed signal segments; Aggregating the signal segments in the set of to-be-processed signal segments to obtain a plurality of signal segment clusters; Respectively obtaining target sensor signals related to the plurality of signal segment clusters, one signal segment cluster corresponding to one related mechanism.

[0009] Optionally, the selected category of sensor signals is filtered from the different categories of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, and the selected category of sensor signals comprises: Filtering a target category of sensor signals and a non-selected category of sensor signals other than the target category from the different categories of sensor signals; Filtering a to-be-determined category of sensor signals from the non-selected category of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal; Taking the to-be-determined category of sensor signals and the target category of sensor signals as the selected category of sensor signals; The selected category of sensor signals is associatedly stored, and the method further comprises: Filtering a target category signal segment from the set of to-be-processed signal segments, the set of to-be-processed signal segments comprising signal segments obtained by splitting the to-be-processed sensor signal into signal segments; The signal segments in the selected category of sensor signals that are related to the target category signal segment are associatedly stored.

[0010] Optionally, the to-be-determined category of sensor signals is filtered from the non-selected category of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, and the method further comprises: Selecting a primary to-be-determined category of sensor signals having a correlation greater than or equal to a preset threshold from the non-selected category of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal; Taking the first N sensor signals having the greatest correlation in the primary to-be-determined category of sensor signals having a correlation greater than or equal to the preset threshold as the to-be-determined category of sensor signals, wherein N is a positive integer greater than or equal to 1.

[0011] Optionally, the selected category of sensor signals is associatedly stored, and the method further comprises: Determining non-selected category signal segments other than the target category signal segment in the set of to-be-processed signal segments; retrieve, from the selected category of sensor signals, signal segments related to the unselected category signal segment as a candidate signal segment set; select, from the candidate signal segment set, M signal segments for correlation storage, where M is a positive integer greater than or equal to 1.

[0012] Optionally, the retrieving, from the selected category of sensor signals, a plurality of signal segments related to the to-be-processed sensor signal comprises: acquiring, in the to-be-processed sensor signal, a first signal segment and a second signal segment adjacent in signal coordinates; retrieving, in the selected category of sensor signals, a third signal segment related to the first signal segment and a fourth signal segment related to the second signal segment; the determining, by the signal coordinates of the plurality of signal segments in the selected category of sensor signals, a continuity evaluation parameter between the plurality of signal segments comprises: determining, according to the signal coordinates of the third signal segment in the selected category of sensor signals and the signal coordinates of the fourth signal segment in the selected category of sensor signals, a continuity evaluation parameter between the third signal segment and the fourth signal segment; the correlation storage, in the selected category of sensor signals, of signal segments related to the to-be-processed sensor signal when the continuity evaluation parameter between the plurality of signal segments reaches a preset standard comprises: when the continuity evaluation parameter between the third signal segment and the fourth signal segment reaches a preset standard, selecting, in the selected category of sensor signals, signal segments related to the to-be-processed sensor signal and correlation storing the selected signal segments; wherein the determining, according to the signal coordinates of the third signal segment in the selected category of sensor signals and the signal coordinates of the fourth signal segment in the selected category of sensor signals, a continuity evaluation parameter between the third signal segment and the fourth signal segment comprises: determining, according to the signal coordinates of the third signal segment in the selected category of sensor signals and the signal coordinates of the fourth signal segment in the selected category of sensor signals, a signal interval between the third signal segment and the fourth signal segment; determining, according to the signal coordinate reference of the third signal segment and the fourth signal segment and the interval between the third signal segment and the fourth signal segment, a continuity evaluation parameter between the third signal segment and the fourth signal segment.

[0013] In a second aspect, the present application provides a computer system, comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method described above is implemented.

[0014] The present application has the following beneficial effects: The present application determines the correlation between the different categories of sensor signals covered by the target sensor signal and the to-be-processed sensor signal by combining the influence factors of different types of sensor signals, and the obtained correlation can represent the contribution of different categories of sensor signals to the to-be-processed sensor signal, which can improve the accuracy of the obtained correlation. The selected category of sensor signals is obtained by screening from the different categories of sensor signals covered by the target sensor signal based on the correlation between the different categories of sensor signals and the to-be-processed sensor signal, and the signal segments related to the to-be-processed sensor signal in the selected category of sensor signals are associated and stored, which can accurately find the connection between the data, provide a comprehensive data analysis perspective for medical workers, improve the utilization efficiency and value of medical monitoring data, and help to find potential medical problems and risks. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of a medical monitoring sensor data association storage method based on machine learning provided by an embodiment of the present application.

[0016] Figure 2 is a composition schematic diagram of a computer system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation manner part of the embodiments of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0018] The execution subject of the medical monitoring sensor data association storage method based on machine learning in the embodiments of the present application is a computer system, including but not limited to a server, a personal computer, a notebook computer, a tablet computer, a smart phone, etc. The computer system can be independently run to implement the present application, or can be connected to a network and interact with other computer systems in the network to implement the present application. The network in which the computer system is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, etc.

[0019] As shown in Figure 1 , the method comprises: Step S110: Obtain target sensor signals related to the sensor signal to be processed, which include different categories of sensor signals.

[0020] In step S110, the computer system obtains target sensor signals related to the sensor signal to be processed. First, the computer system identifies the specific content of the sensor signal to be processed. The sensor signal to be processed is, for example, one or more specific physiological indicators, behavioral activities, or medical events, such as changes in heart rate, daily activity levels, medication use, etc. of a certain patient. Taking changes in heart rate as an example, assuming that we want to analyze the reasons for the abnormal increase in heart rate of a patient within a certain period of time, then the heart rate data is the sensor signal to be processed.

[0021] Next, the computer system searches its database for target sensor signals related to the sensor signal to be processed. These data are usually derived from the patient's health monitoring devices, electronic medical record systems, or other medical information systems. The target sensor signals not only contain information directly related to the sensor signal to be processed (such as the patient's electrocardiogram record), but also may include other factors that may affect the sensor signal to be processed, such as the patient's age, gender, weight, medical history, lifestyle, etc. These information are organized into different categories of sensor signals to facilitate subsequent analysis and processing.

[0022] To illustrate more specifically, assume that the sensor signal to be processed is the sudden increase in heart rate of a patient within a certain period of time. The computer system searches the database for the patient's electrocardiogram record as the directly related target sensor signal. At the same time, it also retrieves the patient's age (a physiological factor that may affect heart rate changes), gender (a physiological difference that may affect heart rate changes), weight (a factor that may affect heart burden and heart rate), medical history (such as heart disease, hypertension, etc. that may affect heart rate), lifestyle (such as whether or not to exercise regularly, whether or not to smoke, drink, etc. that may affect heart rate daily sensors) and other indirectly related sensor signals.

[0023] In the process of obtaining target sensor signals, the computer system can use machine learning models to optimize the search and screening process. For example, it can train a classification model that can automatically identify and retrieve the most relevant target sensor signal categories based on the input sensor signal to be processed (such as changes in heart rate). This model can be trained through historical data, where the input features include the type and value of the sensor signal to be processed, and the output is a probability distribution representing the degree of association between each sensor signal category and the sensor signal to be processed.

[0024] In training the classification model, the computer system can use supervised learning algorithms such as Support Vector Machine (SVM), Random Forest, or Gradient Boosting Machine, etc. These algorithms can learn a decision boundary or function from the labeled training data to distinguish the relevance of different categories of sensor signals to the to-be-processed sensor signal. During the training process, the computer system continuously adjusts the model parameters to minimize the prediction error and improve the classification accuracy. After the model training is completed, it is applied to the actual data retrieval task. For new to-be-processed sensor signals, the computer system inputs them into the model, and the model outputs a probability distribution indicating the relevance of each sensor signal category to the to-be-processed sensor signal. The computer system can select the top few sensor signal categories with the highest relevance as target sensor signals for subsequent analysis based on this probability distribution.

[0025] In addition to the classification model, the computer system can use clustering algorithms to further organize the target sensor signals. Clustering algorithms can group similar sensor signals into a class, making it easier for analysts to identify patterns and trends in the data. For example, the K-means clustering algorithm can divide sensor signals into K clusters based on their numerical characteristics, with sensor signals in each cluster being more numerically similar. In this way, the computer system can organize target sensor signals into different clusters, with each cluster representing a specific sensor pattern or patient population.

[0026] Step S120: According to the influence factors of different categories of sensor signals, obtain the relevance between different categories of sensor signals covered by the target sensor signals and the to-be-processed sensor signal.

[0027] In step S120, the computer system calculates the relevance between different categories of sensor signals in the target sensor signal and the to-be-processed sensor signal. This process is based on the influence factors (i.e. weights) of different categories of sensor signals, aiming to evaluate the contribution of each sensor signal to the to-be-processed sensor signal, and thus to select the most relevant sensor signals for analysis. First, the computer system determines the influence factors of different categories of sensor signals. These influence factors, also known as weights, reflect the importance of each sensor signal in a specific analysis scenario. The determination of influence factors can be based on various factors, including but not limited to statistical analysis of historical data, expert opinions, prediction results of machine learning models, etc.

[0028] Taking heart rate and blood pressure as an example, suppose a patient's heart rate abnormality is being analyzed. In this case, both heart rate signals and blood pressure signals are important sensor signal categories. However, their impact on the heart rate abnormality may be different. Through statistical analysis of historical data, it is found that the heart rate signal has a higher correlation with the heart rate abnormality, so the heart rate signal can be given a higher weight. On the contrary, although the blood pressure signal is also related to heart health, it may not be a major factor in the current analysis scenario, so it can be given a lower weight.

[0029] In addition to statistical analysis, the determination of influence factors can also rely on machine learning models. For example, a regression model can be trained, which takes various sensor signals as input features and the sensor signal to be processed as the target variable, and optimizes the model parameters by minimizing the prediction error. During training, the model automatically learns the impact of each input feature (i.e., sensor signal) on the target variable (i.e., sensor signal to be processed), and represents these impact degrees in the form of weights. After training, this model can be used to calculate the correlation between each sensor signal and the sensor signal to be processed in the new data set.

[0030] Once the influence factors of each sensor signal are determined, the computer system can enter the correlation calculation phase. This process usually involves segmenting the target sensor signal and calculating the correlation between each segment and the sensor signal to be processed. Correlation calculation can be based on various methods, such as cosine similarity, Euclidean distance, Pearson correlation coefficient, etc. Each of these methods has its own advantages and disadvantages, and the choice of method depends on the specific application scenario and data characteristics.

[0031] Taking cosine similarity as an example, suppose there are two vectors A and B, representing the sensor signal to be processed and a sensor signal segment, respectively. The formula for calculating cosine similarity is: ; where, represents the dot product of vectors A and B, represents the length of vectors A and B, respectively. The cosine similarity value ranges from -1 to 1, and the closer the value is to 1, the more similar the two vectors are.

[0032] In practical applications, the sensor signal to be processed and the sensor signal segment can be represented as feature vectors, where each feature corresponds to a numerical value of a sensor signal. Then, the cosine similarity between them is calculated using the above formula as a measure of correlation. It should be noted that since the influence factor (weight) has been determined for each sensor signal category, the corresponding feature values need to be weighted when calculating the dot product.

[0033] Specifically, assume that there are three sensor signal categories: heart rate, blood pressure, and activity, with their impact factors being 0.6, 0.3, and 0.1, respectively. Meanwhile, assume that the to-be-processed sensor signal and the sensor signal segment can be represented as the following feature vectors: .

[0034] When calculating the cosine similarity, the feature values of the sensor signal segment need to be weighted first: ; then, the cosine similarity between the weighted sensor signal segment and the to-be-processed sensor signal is calculated: ; By calculating the cosine similarity between each sensor signal segment and the to-be-processed sensor signal, a correlation matrix can be obtained, where each element represents the correlation degree between the corresponding sensor signal segment and the to-be-processed sensor signal. Finally, the computer system can filter out the most relevant sensor signal segment according to the correlation matrix for subsequent analysis.

[0035] Step S130: According to the correlation between the sensor signals of different categories and the to-be-processed sensor signal, the selected category of sensor signals is filtered out from the sensor signals of different categories.

[0036] In step S130, the computer system filters out the sensor signal category with the highest correlation degree from the to-be-processed sensor signal according to the correlation degrees between the sensor signals of different categories and the to-be-processed sensor signal calculated in step S120, and selects it as the selected category for subsequent analysis.

[0037] First, the computer system sorts the correlation degrees calculated in step S120. Assume that a correlation matrix has been obtained, where each row represents a sensor signal category, each column represents a to-be-processed sensor signal segment, and the elements in the matrix are the correlation degrees between the corresponding sensor signal category and the to-be-processed sensor signal segment. The computer system can sort this matrix by row (or column) so that the sensor signal category with the highest correlation degree is placed at the front. Taking the three sensor signal categories of heart rate, blood pressure, and activity as an example, assume that their correlation degrees with a to-be-processed sensor signal segment (such as an abnormal heart rate phenomenon within a certain period of time) have been calculated.

[0038] Next, the computer system sets a filtering threshold or selects a certain number of sensor signal categories as the selected category according to the sorted correlation matrix. The filtering threshold can be set according to actual analysis requirements and data characteristics, for example, all sensor signal categories with a correlation degree higher than a certain value can be selected, or the top N sensor signal categories (N is a preset integer) with the highest correlation degrees can be selected.

[0039] Continuing with the above example, suppose the filter threshold is set to 0.7. In this case, only the sensor signal class with a correlation higher than 0.7 will be selected as the selected class. In this scenario, only the heart rate signal meets the condition, so the heart rate signal will be selected as the selected class for further analysis.

[0040] Additionally, if the top N sensor signal classes with the highest correlation are selected as the selected classes, the computer system will select them in order of descending correlation until the number of selected sensor signal classes reaches N. For example, if the top two sensor signal classes with the highest correlation are selected, the heart rate signal and the blood pressure signal will be selected as the selected classes.

[0041] When setting the filter threshold or selecting the number of sensor signal classes, the computer system can consider factors such as data sparsity, noise interference, and their impact on correlation. If there are a large number of missing or abnormal values in the data, these values can reduce the accuracy of the correlation. To address this situation, the computer system can use data preprocessing techniques such as interpolation, smoothing, and denoising to improve the integrity and accuracy of the data, ensuring the reliability of the correlation calculation.

[0042] In addition, in some complex scenarios, the computer system can also use machine learning models to assist in screening the selected sensor signals of the selected classes. For example, a classification model can be trained to predict which sensor signal classes are most relevant to the sensor signal to be processed. This model can be trained based on historical data, where the input features include the correlation of each sensor signal class with the sensor signal to be processed, the statistical characteristics of the sensor signal class, etc., and the output is a binary classification label (indicating whether the sensor signal class is selected as the selected class). After the model is trained, the computer system can input new data into the model for prediction, and select the sensor signal of the selected class based on the prediction result.

[0043] Step S140: Store the signal segments of the selected class of sensor signals that are related to the sensor signal to be processed in association.

[0044] In step S140, the computer system stores the signal segments of the selected class of sensor signals that are related to the sensor signal to be processed in association. This step aims to help users (such as medical workers) quickly identify the most relevant and valuable sensor signal segments for the target analysis, so as to more efficiently obtain key information and make accurate judgments.

[0045] Firstly, the computer system determines which signal segments are relevant to the sensor signal to be processed. This usually involves further subdividing the selected category of sensor signals, dividing them into smaller signal segments, and evaluating the relevance of each signal segment to the sensor signal to be processed. This relevance can be measured by various indicators, such as synchronization in time, similarity in value, consistency in trend, etc. For example, in heart rate monitoring, if a signal segment has a heart rate variation pattern that matches the sensor signal to be processed (e.g., a patient suddenly feels palpitations), then this signal segment can be considered relevant.

[0046] In the process of determining relevant signal segments, the computer system can use some advanced signal processing techniques, such as filtering, smoothing, feature extraction, etc., to improve signal quality and reduce noise interference. These techniques help more accurately identify key features and trends in the signal.

[0047] Once the relevant signal segments are determined, the computer system stores them in association.

[0048] In summary, step S140 uses signal processing, machine learning, and other technical means to accurately find the relationship between data, providing medical workers with a comprehensive data analysis perspective.

[0049] In one implementation, step S140, the relevant signal segments in the selected category of sensor signals to the sensor signal to be processed are stored in association, including: Step S141: retrieving multiple signal segments related to the sensor signal to be processed from the selected category of sensor signals; Step S142: determining the continuity evaluation parameters between the multiple signal segments through the signal coordinates of the multiple signal segments in the selected category of sensor signals; Step S143: when the continuity evaluation parameters between the multiple signal segments reach the preset standard, the relevant signal segments in the selected category of sensor signals to the sensor signal to be processed are stored in association.

[0050] In step S141, the computer system identifies signal segments highly relevant to the sensor signal to be processed from the selected category of sensor signals. Relevance is usually evaluated based on the similarity of signal segments to the sensor signal to be processed in terms of value, trend, pattern, etc. To achieve this goal, the computer system can use various methods, including but not limited to signal matching algorithms, pattern recognition techniques, etc. For example, suppose we are analyzing electrocardiogram data of a heart disease patient, and the selected category of sensor signals is heart rate signals. The sensor signal to be processed is the heart rate variation during a heart attack of the patient. The computer system retrieves signal segments similar to the heart rate variation pattern during the heart attack from the heart rate signals as target signal segments.

[0051] First, the computer system divides the heart rate signal into multiple shorter signal segments, each containing consecutive heart rate data points. For each signal segment, the computer system extracts a series of features, such as mean, standard deviation, kurtosis, skewness, etc., which can summarize the main statistical properties of the signal segment. Then, the computer system calculates the similarity between each signal segment and the sensor signal to be processed. This can be achieved by calculating the distance (e.g., Euclidean distance) or correlation (e.g., Pearson correlation coefficient) between the signal segment features and the sensor signal features to be processed. A similarity threshold is set, and only those signal segments with a similarity exceeding the threshold are selected as target signal segments. Through the above steps, the computer system can identify multiple signal segments related to the sensor signal to be processed.

[0052] In step S142, after determining the signal segments, the computer system evaluates the continuity and closeness between these signal segments. The continuity evaluation parameter is a quantitative indicator that measures whether the signal segments are closely connected without significant discontinuities or jumps. This is crucial for understanding the continuity and stability of the patient's behavior.

[0053] The continuity evaluation parameter can be calculated by the following steps: for each signal segment, the computer system records its starting and ending coordinates (e.g., timestamps) in the selected category of sensor signals. For any two adjacent signal segments, the computer system calculates the time interval (or spatial distance if the signal is multi-dimensional) between them. One or more reference coordinates are selected, which can represent the importance of the signal segment or the focus of analysis. For example, in a heart rate signal, points with sharp changes in heart rate can be selected as reference coordinates. Then, the continuity evaluation parameter is calculated, for example, by first calculating the weighted sum of all adjacent signal segment intervals, with weights set according to the importance of the signal segment or analysis requirements. For example, if a signal segment contains critical information (such as a sudden change in heart rate), it can be given a higher weight. The influence of reference coordinates on signal segment continuity is considered. If a signal segment contains a reference coordinate or is very close to a reference coordinate, the value of the continuity evaluation parameter can be increased. The weighted sum of signal segment intervals and the influence of reference coordinates are combined to obtain the final continuity evaluation parameter. For example, the following formula can be used: ; where, is the weight of the i-th signal segment interval, is the interval between the i-th and i+1-th signal segments, and m is the number of reference coordinates, is the influence of the j-th reference coordinate on continuity.

[0054] For example, continuing with the heart rate signal, assume that three signal segments A, B, and C have been determined, with their starting and ending coordinates as follows: Signal segment A: ; Signal segment B: ; Signal segment C: .

[0055] wherein, It is assumed that signal segment B contains an important heart rate sudden change point, so it is given a higher weight.

[0056] The steps for calculating the continuity evaluation parameter are as follows: 1. Calculate the signal segment spacing: 2. Determine the reference coordinate influence: Since signal segment B contains a heart rate sudden change point, it is given an additional reference coordinate influence value.

[0057] 3. Calculate the continuity evaluation parameter: ; wherein, are the spacing weights between signal segments A and B, and between signal segments B and C, is the influence of the heart rate sudden change point contained in signal segment B on continuity.

[0058] In step S143, after determining the continuity evaluation parameter, the computer system compares it with the preset standard. If the continuity evaluation parameter reaches or exceeds the preset standard, it indicates that these signal segments have high continuity and closeness, and therefore are worthy of being associated and stored.

[0059] For example, continuing with the heart rate signal as an example, assume that the preset continuity evaluation parameter standard is 10. If the continuity evaluation parameter calculated through the above steps is greater than or equal to 10, the computer system will associate and store signal segments A, B, and C.

[0060] The purpose of associated storage is not only to store signal segments related to the sensor signal to be processed, but more importantly, by storing the continuity and closeness between these signal segments, it helps users understand the overall patterns and trends of patient behavior. This is particularly important for medical workers, as they need to develop personalized treatment plans and interventions based on patient behavior patterns.

[0061] In one implementation, step S120, according to the influence factors of different categories of sensor signals, the correlation between different categories of sensor signals covered by the target sensor signal and the sensor signal to be processed is obtained, including: Step S121: signal segment splitting is performed on the sensor signal to be processed to obtain a processed signal segment; Step S122: according to the processed signal segment, the basic correlation between different categories of sensor signals and the sensor signal to be processed is obtained; Step S123: Obtain the correlation between the sensor signals of different categories and the to-be-processed sensor signals according to the basic correlation between the sensor signals of different categories and the to-be-processed sensor signals and the influence factors of the sensor signals of different categories.

[0062] In step S121, the computer system splits the to-be-processed sensor signals into smaller signal segments to facilitate subsequent basic correlation calculation. The signal segment splitting can be based on a time window, a number of data points, or other custom rules. Each signal segment after splitting represents a subsequence in the to-be-processed sensor signals for correlation analysis with sensor signals of different categories.

[0063] For example, assume that the to-be-processed sensor signals are a continuous heart rate data with a time span of 24 hours. To perform signal segment splitting, the computer system can set a fixed length time window (e.g., 1 hour) to split the heart rate data into multiple 1-hour heart rate signal segments. Each signal segment contains heart rate data points within the time period.

[0064] In step S122, after signal segment splitting is completed, the computer system calculates the basic correlation between each to-be-processed signal segment and sensor signals of different categories. The basic correlation measures the similarity or correlation between signal segments and can be calculated by various methods such as cosine similarity, Euclidean distance, dynamic time warping (DTW), etc.

[0065] For example, continuing with the heart rate data example, assume that there are two categories of sensor signals: blood pressure signals and step signals. For each to-be-processed heart rate signal segment, the computer system can calculate the basic correlation with the blood pressure signal segment and the step signal segment within the corresponding time period, respectively.

[0066] Calculate the cosine similarity: Assume that the heart rate signal segment, the blood pressure signal segment, and the step signal segment can all be represented as vectors, such as the heart rate signal segment vector , the blood pressure signal segment vector , and the step signal segment vector . The cosine similarity can be calculated by the following formula: ; ; Where · denotes vector dot product, and | · | denotes vector length.

[0067] In some cases, some parts of the signal segment can be more important than others. For example, in a heart rate signal, abnormal values (such as sudden heart rate points) can be more analytically valuable than normal values. Therefore, when calculating the base correlation, the computer system can give these important parts a higher weight. This can be achieved by adjusting the element values of the signal segment vector or introducing an additional weight vector.

[0068] In step S123, after obtaining the base correlation, the computer system combines the influence factors of different categories of sensor signals to calculate the final correlation. The influence factor reflects the importance of each category of sensor signal in a specific analysis scenario, and is usually determined based on expert knowledge, historical data or machine learning models.

[0069] The way to determine the influence factor can be that a domain expert scores different categories of sensor signals according to his own experience, and the higher the score, the more important the category of signals. Or by analyzing historical data, the computer system can find the most relevant category of sensor signals to the sensor signal to be processed, and give it a higher influence factor. Or by training a classification or regression model, taking the category of sensor signals and the sensor signal to be processed as input, and taking the actual analysis result or expert evaluation as output, the influence factor of each category of sensor signals is learned through the model.

[0070] Once the influence factor is determined, the computer system can calculate the final correlation by the following formula: . Wherein, represents the i-th category of sensor signals, represents the base correlation between the category of signals and the sensor signal to be processed, represents the influence factor of the category of signals.

[0071] For example, suppose the base correlation between the heart rate signal and the blood pressure signal is 0.8, and the base correlation with the step signal is 0.6. At the same time, suppose in the heart rate signal analysis scenario, the influence factor of the blood pressure signal is 1.2 (because blood pressure is directly related to heart rate in physiology), and the influence factor of the step signal is 0.8 (because although the step is related to health, it is less directly related to heart rate). The final correlation is calculated as follows: blood pressure signal correlation: 0.8x1.2=0.96. Step signal correlation: 0.6x0.8=0.48.

[0072] By comparing these final correlations, the computer system can determine which categories of sensor signals are most relevant to the sensor signal to be processed, thereby providing support for subsequent analysis and decision-making.

[0073] In an implementation, step S122, the basic correlation between the different categories of sensor signals and the to-be-processed sensor signal is obtained according to the to-be-processed signal segment, including: Step S1221: retrieving a target signal segment related to the to-be-processed signal segment from the different categories of sensor signals; Step S1222: determining the basic correlation between the different categories of sensor signals and the to-be-processed sensor signal according to the priority of the to-be-processed signal segment and the priority of the target signal segment related to the to-be-processed signal segment in the different categories of sensor signals.

[0074] In step S1221, the task of the computer system is to find target signal segments that are related to the to-be-processed signal segment in time, content, or features. These target signal segments will be used for subsequent basic correlation calculation. In order to achieve this goal, the computer system takes a series of measures to ensure the correlation of the signal segments.

[0075] The correlation judgment criteria may be, for example, time synchronization. For example, for time series data such as heart rate, blood pressure, etc., the computer system can check the degree of overlap in time between the target signal segment and the to-be-processed signal segment. If the two signal segments have significant overlap in time and the change trend is similar, they are, for example, related. Or it can be content similarity. For example, for non-time series data such as text, image, etc., the computer system can judge the correlation between signal segments by calculating the similarity (such as cosine similarity, Jaccard similarity, etc.) between them. The higher the similarity, the more likely the signal segments are related. Or it can be feature consistency. For example, in some cases, the computer system can extract the features (such as mean, variance, peak, etc.) of the signal segments and judge the correlation of the signal segments by comparing the consistency of these features.

[0076] For example, assume that the to-be-processed signal segment is a continuous heart rate data with a time span from 9 am to 10 am. The computer system finds signal segments related to this heart rate data from the blood pressure signal and the step signal.

[0077] For the blood pressure signal, the computer system can check the blood pressure data segment between 9 am and 10 am. If this blood pressure data has significant overlap with the heart rate data in time, and the blood pressure change trend is similar to the heart rate change trend (for example, when the heart rate rises, the blood pressure also rises), this blood pressure data segment can be considered as a related target signal segment.

[0078] For the step signal, the computer system can calculate the similarity between the step signal and the heart rate signal. For example, the dynamic time warping (DTW) algorithm can be used to compare the shape similarity between the step signal segment and the heart rate signal segment. If the similarity is high, this step data segment can also be considered as a related target signal segment.

[0079] In step S1222, after finding the relevant target signal segments, the next step for the computer system is to determine the base correlation degree according to the priority of the signal segments. The priority reflects the importance or criticality of the signal segment in the analysis, which can be determined based on the content, features or analysis results of the signal segment.

[0080] For example, the priority determination is based on the content of the signal segment. Some signal segments may contain key information such as abnormal values, peak values or valley values, which are very important for analysis. Therefore, the signal segments containing these key information can be given higher priority. Or it can be based on the features of the signal segment. Some features of the signal segment (such as mean, variance, kurtosis, etc.) may be more important for analysis. For example, in heart rate analysis, an abnormally high mean heart rate may be more valuable for analysis than a normal mean heart rate. Or it can be based on the analysis results. In some cases, the computer system may have already performed preliminary analysis on the signal segment and obtained some conclusions or findings. These conclusions or findings can guide the determination of priority. For example, if a signal segment is analyzed to be highly related to a certain disease state, it should be given a higher priority.

[0081] Once the priority of the signal segment is determined, the computer system can use the following formula to calculate the base correlation degree: ; where, represents the i-th sensor signal category, n represents the number of relevant target signal segments found in this category, represents the priority weight of the j-th relevant target signal segment, represents the j-th relevant target signal segment, represents the signal segment to be processed, represents a function that calculates the similarity between two signal segments.

[0082] For example, continuing with the heart rate data, suppose relevant target signal segments have been found in the blood pressure signal and the step count signal that are related to the signal segment to be processed. Now, the base correlation degree needs to be calculated according to the priority of these signal segments.

[0083] Suppose the priority weight of blood pressure signal segment 1 is 0.8 (because it contains a significant blood pressure peak), and the priority weight of blood pressure signal segment 2 is 0.6 (because it is just an ordinary blood pressure fluctuation). Suppose the priority weight of step count signal segment 1 is 0.5 (because it is not particularly similar to the heart rate signal segment), and the priority weight of step count signal segment 2 is 0.7 (because it is highly synchronized in time with the heart rate signal segment). Suppose the similarity between each relevant target signal segment and the signal segment to be processed has been calculated.

[0084] Now, the base correlation between the blood pressure signal and the step signal and the heart rate signal to be processed can be calculated using the above formula: .

[0085] By comparing the two base correlations, the computer system can determine which category of sensor signals is more relevant to the sensor signal to be processed, thereby providing support for subsequent analysis and decision-making.

[0086] In summary, steps S1221 and S1222 together constitute the process of calculating the base correlation between different categories of sensor signals and the sensor signal to be processed. Through correlation judgment and priority determination, the computer system can accurately evaluate the similarity and importance between signal segments, providing strong data support for subsequent analysis.

[0087] In one implementation, the number of target sensor signals is multiple, and the method further comprises: Step S150: Determine the influence factor of each target sensor signal according to the correlation mechanism between the sensor signal to be processed and each target sensor signal; Step S160: Obtain the correlation between each target sensor signal and the sensor signal to be processed according to the influence factor of each target sensor signal and the correlation between different categories of sensor signals covered by each target sensor signal and the sensor signal to be processed; Step S170: Sort the multiple target sensor signals according to the correlation between each target sensor signal and the sensor signal to be processed.

[0088] In step S150, the computer system determines the influence factor of each target sensor signal. The influence factor is a quantitative index for measuring the importance or weight of the target sensor signal in the overall analysis. In order to accurately determine the influence factor, the computer system considers the correlation mechanism between the sensor signal to be processed and each target sensor signal.

[0089] The correlation mechanism refers to the way or rule of establishing a connection between the sensor signal to be processed and the target sensor signal. These mechanisms can be based on various factors, including but not limited to time synchronization, content similarity, feature consistency, etc. For example, in terms of time synchronization, if two signals have significant overlap in time and similar change trends, they are, for example, relevant. In terms of content similarity, if two signals contain the same or similar information (such as keywords, patterns, etc.), they are also, for example, relevant.

[0090] The computer system can assign different weights to each target sensor signal according to the strength and complexity of the relevant mechanisms. For example, if the relevant mechanism between a certain target sensor signal and the sensor signal to be processed is very clear and strong (such as direct time synchronization and content correlation), a higher weight can be assigned to this signal.

[0091] In some cases, the computer system can use machine learning models to predict the impact factor of each target sensor signal. These models can be trained based on historical data and learn how to map relevant mechanisms to impact factors. For example, a regression model or a classification model can be used to predict the impact factor, where the input features may include multiple aspects of the relevant mechanism (such as time synchronization, content similarity, feature consistency, etc.), and the output is the impact factor of the target sensor signal.

[0092] In step S160, the computer system combines the correlation between different categories of sensor signals and the sensor signal to be processed calculated in step S120, and the impact factor determined in step S150, to calculate the overall correlation between each target sensor signal and the sensor signal to be processed.

[0093] The overall correlation can be calculated by the following formula: ; where, represents the i-th target sensor signal, n represents the number of different categories of sensor signals covered in this signal, represents the impact factor of the j-th category sensor signal in the i-th target sensor signal, represents the correlation between the j-th category sensor signal and the sensor signal to be processed (which has been calculated in step S120).

[0094] In step S170, the computer system sorts the multiple target sensor signals according to the overall correlation calculated in step S160, and lists them in order. This helps medical workers quickly identify the target sensor signals most relevant to the sensor signal to be processed, and thus more efficiently perform subsequent analysis.

[0095] The computer system can sort the overall correlation from high to low, so that the target sensor signal with the highest correlation is placed at the front. In this way, medical workers can first focus on those signals most relevant to the sensor signal to be processed.

[0096] In some cases, the computer system can also set a correlation threshold. Only those target sensor signals with an overall correlation exceeding the threshold will be listed. This helps to reduce noise interference and improve analysis efficiency.

[0097] For example, assume there are three target sensor signals (patient A, patient B, patient C), each of which contains two categories of sensor signals (such as heart rate signals and blood pressure signals). In step S120, the correlation between each category of sensor signal and the sensor signal to be processed has been calculated. In step S150, the influence factor is assigned to the category of sensor signal in each target sensor signal according to the correlation mechanism. Now, the overall correlation between each target sensor signal and the sensor signal to be processed is calculated and ranked: For patient A: Heart rate signal correlation: 0.8; blood pressure signal correlation: 0.6; heart rate signal influence factor: 0.7 (assigned based on the correlation mechanism); blood pressure signal influence factor: 0.5 (assigned based on the correlation mechanism); overall correlation (patient A): 0.7x0.8+0.5x0.6=0.86.

[0098] For patient B: Heart rate signal correlation: 0.7; blood pressure signal correlation: 0.5; heart rate signal influence factor: 0.8 (assigned based on the correlation mechanism); blood pressure signal influence factor: 0.4 (assigned based on the correlation mechanism); overall correlation (patient B): 0.8x0.7+0.4x0.5=0.76.

[0099] For patient C: Heart rate signal correlation: 0.6; blood pressure signal correlation: 0.4; heart rate signal influence factor: 0.6 (assigned based on the correlation mechanism); blood pressure signal influence factor: 0.3 (assigned based on the correlation mechanism); overall correlation (patient C): 0.6x0.6+0.3x0.4=0.48.

[0100] In order of overall correlation from high to low, the target sensor signals are listed as follows: 1, patient A (overall correlation: 0.86); 2, patient B (overall correlation: 0.76); 3, patient C (overall correlation: 0.48).

[0101] Through the above steps, the most relevant signal to the sensor signal to be processed can be accurately identified from multiple target sensor signals, and ordered data support is provided for medical workers.

[0102] In one implementation, step S110, obtaining target sensor signals related to the sensor signal to be processed, includes: Step S111: performing signal segment splitting on the sensor signal to be processed to obtain a set of processed signal segments; Step S112: clustering the signal segments in the set of processed signal segments to obtain a plurality of signal segment clusters; Step S113: Acquire the target sensor signals associated with multiple signal segment clusters respectively, with one signal segment cluster corresponding to one related mechanism.

[0103] In step S111, the computer system splits the sensor signal to be processed into multiple smaller signal segments to form a set of signal segments to be processed. Signal segment splitting is one of the key steps in data preprocessing, which helps the computer system to analyze the characteristics and patterns of the sensor signal to be processed in greater detail.

[0104] Signal segmentation methods include time-window-based segmentation. The computer system can set a fixed-length time window (such as 1 minute, 5 minutes, etc.) and then segment the sensor signal to be processed according to this time window. For example, if the sensor signal to be processed is a continuous heart rate data over a period of 24 hours, the computer system can segment it into 288 10-minute signal segments (assuming there are 6 10-minute signal segments per hour).

[0105] Alternatively, based on event-triggered segmentation, in some cases, the sensor signal being processed may contain specific event markers (such as outliers, peaks, etc.). The computer system can then segment the signal into different segments based on these event markers. For example, in heart rate data, the computer system can segment the signal into a new segment whenever the heart rate exceeds a certain threshold.

[0106] Alternatively, in addition to time windows, computer systems can also divide signals based on the number of data points. For example, each signal segment can be set to contain 100 data points, and the sensor signal to be processed can be divided according to this number.

[0107] For example, suppose the sensor signal to be processed is a continuous heart rate data segment spanning one hour. The computer system sets a 5-minute time window and divides the heart rate data into 12 signal segments. Each signal segment contains heart rate data points within that time period, forming a set of signal segments to be processed.

[0108] In step S112, after obtaining the set of signal segments to be processed, the next step of the computer system is to cluster these signal segments to form multiple signal segment clusters. The purpose of clustering is to group signal segments with similar characteristics or patterns into one category, so that the target sensor signal related to the sensor signal to be processed can be identified more accurately in the subsequent process.

[0109] Clustering can be achieved through features-based clustering. The computer system can extract features from each signal segment (such as mean, variance, kurtosis, skewness, etc.) and then use clustering algorithms (such as K-means, DBSCAN, etc.) to group these feature vectors into clusters. Signal segments with similar features will be grouped together.

[0110] Alternatively, clustering based on patterns, in addition to features, the computer system can also consider the pattern similarity between signal segments. For example, dynamic time warping (DTW) algorithm can be used to compare the shape similarity between signal segments, and signal segments with similar shape can be clustered into one cluster.

[0111] Alternatively, clustering based on expert knowledge, in some cases, domain experts can be able to divide signal segments into different clusters based on experience. The computer system can accept the input of experts and cluster accordingly. For example, continuing with the heart rate data, the computer system extracts the mean and variance of each signal segment as features, and then uses K-means clustering algorithm to cluster these feature vectors. Suppose that finally 3 clusters are obtained, representing low heart rate, normal heart rate and high heart rate signal segments respectively.

[0112] In step S113, after obtaining multiple signal segment clusters, the last step of the computer system is to obtain the target sensor signal related to each signal segment cluster respectively. The correlation mechanism refers to the way or rule of establishing the connection between the signal segment cluster and the target sensor signal. Different signal segment clusters may correspond to different correlation mechanisms.

[0113] For example, time synchronization is used, i.e. for time series data such as heart rate, blood pressure, etc., the time synchronization between the signal segment cluster and the target sensor signal is an important correlation mechanism, for example. If two signals have significant overlap in time and similar change trend, they are related, for example.

[0114] Or content similarity is used, in some cases, the content similarity between the signal segment cluster and the target sensor signal may be more important. For example, in text analysis, if two signal segment clusters contain the same keywords or phrases, they may be related to the same target sensor signal.

[0115] Or feature consistency is used, the feature consistency between the signal segment cluster and the target sensor signal is also a possible correlation mechanism. For example, in image analysis, if the texture, color and other features of two signal segment clusters are similar, they may be related to the same target image signal. For example, continuing with the heart rate data, for the signal segment cluster representing low heart rate, the computer system can find those target sensor signals that also contain low heart rate signal segments as related signals. For the signal segment cluster representing normal heart rate, it can find target sensor signals containing normal heart rate signal segments. And for the signal segment cluster representing high heart rate, it can find target sensor signals containing abnormal heart rate signal segments (such as sudden increase or decrease of heart rate). These correlation mechanisms may be based on factors such as time synchronization, content similarity or feature consistency.

[0116] Through the above steps, the computer system can accurately identify the signal segment most relevant to the to-be-processed sensor signal from a large number of target sensor signals, and provide strong support for subsequent data analysis and pattern recognition.

[0117] In an implementation, step S130, the selected category of sensor signals is filtered from the different categories of sensor signals according to the relevance between the different categories of sensor signals and the to-be-processed sensor signal, and includes: Step S131: filtering the target category of sensor signals and the unselected category of sensor signals from the different categories of sensor signals; Step S132: filtering the to-be-determined category of sensor signals from the unselected category of sensor signals according to the relevance between the different categories of sensor signals and the to-be-processed sensor signal; Step S133: taking the to-be-determined category of sensor signals and the target category of sensor signals as the selected category of sensor signals.

[0118] In the implementation of step S130, the target of the computer system is to filter the signal category most relevant to the to-be-processed sensor signal from the different categories of sensor signals as the selected category for subsequent analysis.

[0119] In step S131, the computer system performs preliminary filtering on all different categories of sensor signals to determine which categories are the target category and which are the unselected category. The target category is usually determined in advance according to domain knowledge, expert opinion or historical data, and has a high relevance or importance to the to-be-processed sensor signal.

[0120] The target category is determined in a manner such as based on domain knowledge. In some cases, domain experts may be able to directly determine which categories of sensor signals are most relevant to the to-be-processed sensor signal according to professional knowledge. For example, in heart rate monitoring, heart rate signals and blood pressure signals are usually the target category when analyzing heart rate abnormalities.

[0121] Or based on historical data, the computer system can determine which categories of sensor signals have a high relevance to the to-be-processed sensor signal by analyzing historical data. For example, an association rule mining algorithm (such as the Apriori algorithm) can be used to find frequent item sets between the categories of sensor signals and the to-be-processed sensor signal, thereby determining the target category.

[0122] Or based on a machine learning model, in some complex scenarios, the computer system can use a machine learning model to predict which sensor signal category is most likely related to the sensor signal to be processed. For example, a classification model can be trained to take sensor signal categories as input features and the presence or absence of the sensor signal to be processed as output labels, and determine the target category through model prediction. For example, suppose the electrocardiogram data of a heart disease patient is being analyzed, and the sensor signal to be processed is the patient's heart rate change during a heart attack. According to domain knowledge and historical data, it can be determined that heart rate signals and blood pressure signals are highly relevant target categories during a heart attack. Other sensor signals (such as step count, sleep quality, etc.) are considered as unselected categories other than the target category.

[0123] In step S132, after determining the target category and the unselected category, the next step of the computer system is to select the sensor signal category with a higher correlation degree from the unselected category as the pending category according to the correlation degree calculated in step S120. The pending category refers to those sensor signal categories that are not initially determined as the target category but show a higher correlation in actual data analysis.

[0124] The selection method of the pending category is, for example, to set a correlation degree threshold. The computer system can set a correlation degree threshold, and only the unselected category with a correlation degree higher than the threshold will be selected as the pending category. The threshold can be adjusted according to actual needs and data characteristics.

[0125] The computer system can sort all the correlation degrees of the unselected categories and select the top N categories with the highest correlation degrees as the pending categories (N is a preset integer). This method can ensure that the selected pending categories have high relevance and importance.

[0126] Or based on the machine learning model screening, in some cases, the computer system can use a machine learning model to assist in screening the pending category. For example, a regression model can be trained to predict the correlation between each unselected category and the sensor signal to be processed, and the pending category is selected according to the prediction result. For example, continuing with the example of the electrocardiogram data of a heart disease patient. Suppose the correlation between the heart rate signal, the blood pressure signal, and other unselected categories (such as step count, sleep quality, etc.) and the heart rate change signal to be processed has been calculated. By setting a correlation degree threshold or a sorting selection method, the step count signal can be selected as the pending category from the unselected categories, because the step count signal may have some relevance to the heart rate change during a heart attack (for example, the patient's activity decreases before a heart attack, which may lead to a decrease in step count).

[0127] In step S133, the computer system merges the sensor signals of the pending categories with the sensor signals of the target categories to form a final selected category set. The selected category set contains all the sensor signal categories that are highly relevant to the sensor signal to be processed, and they will be the focus of the subsequent data analysis.

[0128] The way of category merging is, for example, to directly merge the pending categories and the target categories into a new set as the selected categories. Alternatively, in some cases, the computer system can assign different weights to different categories of signals according to their importance or relevance, and consider these weights in the merging process. For example, a weighted average method can be used to calculate the overall relevance or importance score of the selected categories.

[0129] For example, continuing with the electrocardiogram data of a heart disease patient, in step S132, the step signal has been screened as a pending category. Now, the step signal is merged with the heart rate signal and the blood pressure signal (target categories) to form a final selected category set. This set will be the focus of the subsequent data analysis to further explore the relationship between heart attack, heart rate, blood pressure and step count.

[0130] Through the above steps, the computer system can accurately screen out the signal categories that are most relevant to the sensor signal to be processed from different categories of sensor signals, providing strong support for subsequent data analysis and pattern recognition.

[0131] In one implementation, step S132, the sensor signals of the pending categories are screened from the unselected categories of sensor signals according to the relevance between the different categories of sensor signals and the sensor signal to be processed, comprising: Step S1321: selecting the original pending category sensor signals with a relevance higher than a relevance threshold value from the unselected categories of sensor signals according to the relevance between the different categories of sensor signals and the sensor signal to be processed; Step S1322: selecting the top N sensor signals with the highest relevance from the original pending category sensor signals with a relevance greater than or equal to a preset threshold value as the sensor signals of the pending categories, wherein N is a positive integer greater than or equal to 1.

[0132] In step S1321, the computer system determines a relevance threshold value and screens the signal categories with higher relevance from the unselected categories of sensor signals according to this threshold value. These signal categories will be considered as the original pending category sensor signals, and they may be further confirmed as the pending categories in the subsequent steps.

[0133] The way to determine the relevance threshold can be, for example, a statistical distribution-based method. The computer system can calculate the statistical distribution (e.g. mean, standard deviation, etc.) of the relevance between all unselected category sensor signals and the to-be-processed sensor signal, and determine a reasonable relevance threshold based on this distribution. For example, the mean plus one or two standard deviations can be selected as the threshold to ensure that the selected signal categories have a certain statistical significance.

[0134] Or a domain knowledge-based method. In some cases, domain experts can directly give a reasonable relevance threshold based on professional knowledge. This threshold can be based on past research experience, experimental results or industry standards.

[0135] Alternatively, a machine learning model-based method. The computer system can also use a machine learning model to predict a reasonable relevance threshold. This model can be trained based on historical data to learn the relationship between relevance and signal category importance, and output a predicted threshold. For example, suppose we are analyzing the blood glucose data of a diabetic patient, and the to-be-processed sensor signal is an event of abnormal elevation of the patient's blood glucose. There is a series of unselected category sensor signals, including diet records, exercise records, sleep records, etc. First, the computer system calculates the relevance between these signals and the to-be-processed sensor signal. Then, based on the statistical distribution method, the mean plus one standard deviation is selected as the relevance threshold. In this way, all signal categories with a relevance higher than this threshold (e.g. diet records) are selected as the original pending category sensor signals.

[0136] In step S1322, after determining the original pending category sensor signals, the next step of the computer system is to sort these signal categories according to their relevance and select the top N signal categories with the highest relevance as the final pending categories. This process helps to ensure that the selected pending categories have the highest relevance to the to-be-processed sensor signal.

[0137] The computer system sorts all original pending category sensor signals in descending order of relevance. In this way, the signal categories with the highest relevance will be ranked first. After sorting, the computer system selects the top N signal categories with the highest relevance as the pending categories. The value of N can be determined according to actual needs and data characteristics. For example, if there are many signal categories in the data set, but only a few important categories are needed for analysis, then N can be set small; conversely, if a more comprehensive understanding of the information in the data set is desired, then N can be set large.

[0138] For example, continuing with the blood glucose data of a diabetes patient as an example. In step S1321, the diet record has been filtered as a sensor signal of a primary pending category. Assume that there are other sensor signals of unselected categories (such as exercise record, sleep record, etc.) that also reached the relevance threshold and are filtered as sensor signals of a primary pending category. Now, the computer system ranks these signal categories in order of relevance from high to low. Assume that the ranking result is: diet record (relevance 0.9), exercise record (relevance 0.7), sleep record (relevance 0.5). If N = 2 is selected, then the diet record and the exercise record will be selected as the final pending categories because they have the highest relevance to the sensor signal to be processed.

[0139] Through the above steps, the computer system can filter out the signal categories with high relevance to the sensor signal to be processed from the sensor signals of unselected categories as the sensor signals of a pending category. This process not only considers the relevance between the signal category and the sensor signal to be processed, but also ensures that the selected pending categories have the highest importance through ranking and selection.

[0140] In an implementation, step S140, the signal segments of the selected category of sensor signals that are relevant to the sensor signal to be processed are stored in association, further includes: Step S144: filtering target category signal segments from the set of signal segments to be processed, the set of signal segments to be processed including the signal segments obtained by signal segment splitting on the sensor signal to be processed; Step S145: storing in association the signal segments of the selected category of sensor signals that are relevant to the target category signal segments.

[0141] In step S144, the computer system filters the target category signal segments from the set of signal segments to be processed. The set of signal segments to be processed is obtained by signal segment splitting on the sensor signal to be processed in step S111, and it contains multiple signal segments, each of which represents a subsequence of the sensor signal to be processed. The target category signal segments refer to those signal segments that are closely related to a specific analysis target or focus.

[0142] The computer system can extract the features (such as mean, variance, kurtosis, skewness, etc.) of each signal segment and filter the target category signal segments according to these features. For example, in heart rate monitoring, if the analysis target is to identify abnormal heart rate events, the computer system can filter out those signal segments with heart rate values outside the normal range (such as less than 60 times per minute or more than 100 times per minute) as target category signal segments.

[0143] In addition to the feature-based screening method, the computer system can also use a pattern correlation algorithm to identify the target category signal segments. For example, a dynamic time warping (DTW) algorithm can be used to compare the shape similarity between signal segments, and screen out the signal segments with high correlation to a preset pattern or template as the target category signal segments.

[0144] In some cases, domain experts can directly determine which signal segments belong to the target category based on their professional knowledge. The computer system can accept the input of the experts and screen accordingly.

[0145] For example, suppose that the electrocardiogram data of a patient with heart disease is being analyzed, and the sensor signal to be processed is the heart rate variation of the patient during a certain heart attack. In step S111, the heart rate data has been split into multiple signal segments, and a set of signal segments to be processed has been formed. Now, the analysis target is to identify the abnormal events of heart rate during a heart attack. Based on this target, the feature-based screening method can be used to screen out the signal segments with heart rate values outside the normal range (e.g., less than 60 or more than 120 per minute) as the target category signal segments.

[0146] In step S145, after screening out the target category signal segments, the computer system finds the signal segments in the selected category of sensor signals that are related to these signal segments and stores them in association. The correlation is usually evaluated based on the similarity or association between signal segments, and can be achieved by various methods. The computer system can calculate the similarity (e.g., cosine similarity, Euclidean distance, etc.) between the signal segments in the selected category of sensor signals and the target category signal segments, and evaluate the correlation based on the similarity. The higher the similarity, the more relevant the two signal segments are. For time series data such as heart rate, blood pressure, etc., the computer system can check the degree of overlap in time between the signal segments in the selected category of sensor signals and the target category signal segments. If two signal segments have significant overlap in time and similar change trends, they are, for example, relevant. In addition to similarity and time synchronization, the computer system can also extract features (e.g., mean, variance, etc.) of the signal segments and compare the consistency of these features to evaluate the correlation. The higher the feature consistency, the more likely the two signal segments are related.

[0147] The computer system can store the related signal segments in association by changing the visual attributes of the signal segments such as color, thickness, flashing frequency, etc. For example, different colors can be used to distinguish signal segments of different categories, or bold lines can be used to highlight the related signal segments.

[0148] In addition to visual emphasis, the computer system can also provide interactive display functions, allowing users to view details of relevant signal segments in detail through operations such as zooming, panning, etc. For example, interactive chart libraries such as D3.js, ECharts, etc. can be used to create dynamic and interactive signal display interfaces.

[0149] To further help users understand the meaning and importance of relevant signal segments, the computer system can also generate corresponding text descriptions and explanations. These descriptions and explanations can be written based on the characteristics, patterns or analysis results of the signal segments, and presented on the user interface.

[0150] For example, continuing with the example of electrocardiogram data of a heart disease patient, in step S144, the target category signal segments, i.e. the signal segments of heart rate abnormal events during a heart attack, have been filtered out. Now, it is necessary to find the signal segments related to these target category signal segments in the selected category sensor signals (such as heart rate signals, blood pressure signals, etc.) and store them in association.

[0151] In one implementation, step S140, the signal segments in the selected category sensor signals related to the to-be-processed sensor signal are stored in association, further comprising: Step S146: determining unselected category signal segments in the to-be-processed signal segment set other than the target category signal segments; Step S147: retrieving signal segments related to the unselected category signal segments in the selected category sensor signals to obtain a candidate signal segment set; Step S148: filtering M signal segments in the candidate signal segment set for association storage, where M is a positive integer greater than or equal to 1.

[0152] In step S146, the computer system identifies from the to-be-processed signal segment set those signal segments that do not belong to the target category signal segments, which are referred to as unselected category signal segments. Unselected category signal segments may contain information related to the to-be-processed sensor signal but not explicitly classified as target category, so they also have potential analysis value.

[0153] If each signal segment in the to-be-processed signal segment set has been assigned a corresponding label (such as "target category", "non-target category", etc.), the computer system can simply identify the unselected category signal segments by checking the labels.

[0154] If the signal segments do not have explicit labels, the computer system can analyze the characteristics of the signal segments (e.g., mean, variance, kurtosis, skewness, etc.) to determine whether they belong to the target category. Those signal segments that are significantly different from the characteristics of the target category signal segments will be considered as non-selected category signal segments. In some cases, the computer system can use a machine learning model to predict whether each signal segment belongs to the target category. This model can be trained based on historical data and learn how to distinguish the target category signal segments from the non-selected category signal segments.

[0155] For example, assume that the blood glucose data of a diabetic patient is being analyzed, and the target category signal segments are those with abnormally high or low blood glucose values. In the set of signal segments to be processed, there can be other signal segments with relatively small fluctuations in blood glucose values, which are considered as non-selected category signal segments.

[0156] In step S147, after determining the non-selected category signal segments, the computer system retrieves those signal segments in the selected category sensor signals that are related to these signal segments. The relevance is usually evaluated based on the similarity or correlation between the signal segments, and a set of candidate signal segments can be obtained, which can contain similar information or patterns as the non-selected category signal segments.

[0157] The computer system can calculate the similarity (e.g., cosine similarity, Euclidean distance, etc.) between each signal segment in the selected category sensor signals and the non-selected category signal segments, and select those signal segments with similarity exceeding a certain threshold as candidate signal segments.

[0158] For time series data, such as heart rate, blood pressure, etc., the computer system can check the degree of temporal overlap between the signal segments in the selected category sensor signals and the non-selected category signal segments. If two signal segments have significant overlap in time and similar change trends, they are, for example, relevant. In addition to similarity and temporal synchronization, the computer system can also extract features (e.g., mean, variance, etc.) of the signal segments and compare the consistency of these features to evaluate the relevance. The higher the feature consistency, the more likely the two signal segments are related.

[0159] For example, continuing with the blood glucose data of a diabetic patient. Assume that the selected category sensor signals include diet records, exercise records, etc. For each non-selected category signal segment (i.e., signal segment with relatively small fluctuations in blood glucose values), the computer system can retrieve those signal segments in the diet records and exercise records that are time-synchronized and similar in characteristics as candidate signal segments. For example, if the patient's diet and exercise habits are relatively stable during the time period corresponding to a non-selected category signal segment, the computer system can find similar signal segments in these records as candidate signal segments.

[0160] In step S148, after obtaining the candidate signal segment set, the last step of the computer system is to filter out M signal segments for association storage. The value of M can be determined according to actual needs and analysis goals. The filtering process may involve comprehensive consideration of various factors, such as signal segment similarity, time synchronization, feature consistency, etc.

[0161] The computer system can sort the candidate signal segments according to indicators such as similarity, time synchronization, or feature consistency, and select the top M signal segments for association storage.

[0162] For example, continuing with the blood glucose data of a diabetic patient. Suppose M = 3 is set, i.e. it is desired to associate store 3 candidate signal segments. The computer system can sort the candidate signal segments according to similarity and time synchronization, and select the top 3 signal segments for association storage. These signal segments may represent the most relevant and important dietary record or exercise record information to the signal segments of the unselected category, helping the user to better understand the potential causes and influencing factors of blood glucose fluctuations.

[0163] In one implementation, step S141, retrieving a plurality of signal segments related to the sensor signal to be processed from the sensor signals of the selected category, includes: Step S1411: obtaining a first signal segment and a second signal segment adjacent in signal coordinates in the sensor signal to be processed; Step S1412: retrieving a third signal segment related to the first signal segment and a fourth signal segment related to the second signal segment from the sensor signals of the selected category.

[0164] Step S142, determining a continuity evaluation parameter between the plurality of signal segments in the signal coordinates of the sensor signals of the selected category, includes: Step S1421: determining a continuity evaluation parameter between the third signal segment and the fourth signal segment according to the signal coordinates of the third signal segment in the sensor signals of the selected category and the signal coordinates of the fourth signal segment in the sensor signals of the selected category.

[0165] Step S143, when the continuity evaluation parameter between the plurality of signal segments reaches a preset standard, the signal segments related to the sensor signal to be processed in the sensor signals of the selected category are associated and stored, including: Step S1431: when the continuity evaluation parameter between the third signal segment and the fourth signal segment reaches a preset standard, the signal segments related to the sensor signal to be processed are filtered out from the sensor signals of the selected category, and the filtered signal segments are associated and stored.

[0166] In step S1411, the computer system finds two signal segments in the sensor signal to be processed that are adjacent in the signal coordinates, referred to as the first signal segment and the second signal segment. These two signal segments should be closely connected so that the signal segments related to them respectively in the selected category of sensor signals can be found subsequently.

[0167] If the sensor signal to be processed is time series data (e.g. heart rate, blood pressure, etc.), the computer system can set a fixed length of time window to divide the signal into multiple signal segments. Then, the signal segments in two adjacent time windows are selected as the first signal segment and the second signal segment. In some cases, the sensor signal to be processed can contain specific event markers (e.g. outliers, peaks, etc.). The computer system can divide the signal into different segments according to these event markers, and select adjacent event segments as the first signal segment and the second signal segment. If the characteristics (e.g. mean, variance, etc.) of the sensor signal to be processed change significantly in a short period of time, the computer system can divide the signal into different segments according to these change points, and select adjacent characteristic change segments as the first signal segment and the second signal segment.

[0168] For example, assume that the sensor signal to be processed is a continuous heart rate data with a time span of 1 hour. The computer system sets a time window of 5 minutes to divide the heart rate data into 12 signal segments. Now, the two adjacent signal segments (e.g. the 5th signal segment and the 6th signal segment) are selected as the first signal segment and the second signal segment.

[0169] In step S1412, after determining the first signal segment and the second signal segment, the next step of the computer system is to retrieve the signal segments related to these two signal segments respectively in the selected category of sensor signals, referred to as the third signal segment and the fourth signal segment. The relevance is usually evaluated based on the similarity or correlation between the signal segments.

[0170] The computer system can calculate the similarity (e.g. cosine similarity, Euclidean distance, etc.) between each signal segment in the selected category of sensor signals and the first signal segment and the second signal segment, and select the signal segments with similarity exceeding a certain threshold as the third signal segment and the fourth signal segment.

[0171] For time series data, the computer system can check the degree of temporal overlap between the signal segments in the selected category of sensor signals and the first signal segment and the second signal segment. If the two signal segments have significant overlap in time and similar change trends, they are, for example, relevant.

[0172] In addition to similarity and temporal synchronization, the computer system can also extract the features (e.g. mean, variance, etc.) of the signal segments and compare the consistency of these features to evaluate the relevance. The higher the feature consistency, the more likely the two signal segments are related.

[0173] For example, continuing with the heart rate data example. Assume that the selected category of sensor signals includes a blood pressure signal and a step count signal. For the first signal segment (the 5th heart rate signal segment), the computer system can find a similar time segment in the blood pressure signal as the third signal segment; for the second signal segment (the 6th heart rate signal segment), the computer system can find a similar time segment in the step count signal as the fourth signal segment.

[0174] In step S1421, the computer system evaluates the continuity between the third signal segment and the fourth signal segment to determine whether they are closely connected to form a meaningful signal segment sequence. The continuity evaluation parameter is a quantitative indicator that measures the closeness and continuity between the signal segments.

[0175] For time series data, the computer system can calculate the time interval between the third signal segment and the fourth signal segment. The shorter the time interval, the better the continuity between the two signal segments. In addition to the time interval, the computer system can also consider the reference of the signal coordinates. For example, if the third signal segment and the fourth signal segment both contain important feature change points or event markers, their continuity may be more important. The computer system can combine the time interval and the reference of the signal coordinates to calculate the continuity evaluation parameter through weighted summation or multiplication, etc. The weights can be determined according to actual needs and analysis goals.

[0176] For example, continuing with the heart rate data example. Assume that the third signal segment is a time segment in the blood pressure signal and the fourth signal segment is a time segment in the step count signal. The computer system can calculate the time interval between the two time segments and consider whether they both contain important feature change points (such as sudden rise in blood pressure, sudden decrease in step count, etc.). Then, the continuity evaluation parameter is calculated according to the time interval and the reference of the signal coordinates.

[0177] In step S1431, the computer system determines whether the continuity evaluation parameter between the third signal segment and the fourth signal segment meets the preset standard. If it meets the standard, the signal segments related to the sensor signal to be processed in the selected category of sensor signals are stored in association; otherwise, the signal segments may need to be further screened or adjusted.

[0178] For example, continuing with the heart rate data example. Assume that the continuity evaluation parameter between the third signal segment and the fourth signal segment meets the preset standard, indicating that the two signal segments are closely connected to form a meaningful signal segment sequence. At this time, the computer system can store the signal segments related to the sensor signal to be processed (such as the third signal segment and the fourth signal segment) in the selected category of sensor signals in association.

[0179] Through the above steps, the computer system can accurately retrieve the plurality of signal segments related to the to-be-processed sensor signal from the selected category of sensor signals, and determine which signal segments should be stored in association by evaluating the continuity between the signal segments.

[0180] In an implementation, at step S1421, the continuity evaluation parameter between the third signal segment and the fourth signal segment is determined according to the signal coordinates of the third signal segment in the selected category of sensor signals and the signal coordinates of the fourth signal segment in the selected category of sensor signals, including: At step S14211, the signal interval between the third signal segment and the fourth signal segment is determined according to the signal coordinates of the third signal segment in the selected category of sensor signals and the signal coordinates of the fourth signal segment in the selected category of sensor signals. At step S14212, the continuity evaluation parameter between the third signal segment and the fourth signal segment is determined according to the signal coordinate reference of the third signal segment and the fourth signal segment, and the interval between the third signal segment and the fourth signal segment.

[0181] At step S14211, the computer system calculates the signal interval between the third signal segment and the fourth signal segment. The signal interval is a quantitative indicator for measuring the distance between two signal segments in time or space. For time series data, the signal interval usually represents the interval between two signal segments on the time axis; for non-time series data, the signal interval may represent the distance between two signal segments in the feature space.

[0182] If the third signal segment and the fourth signal segment are two time segments in time series data, the computer system can calculate the difference between the start time and the end time of the two time segments to obtain the signal interval. For example, if the start time of the third signal segment is t1, the end time is t2, the start time of the fourth signal segment is t3, and the end time is t4, the signal interval can be calculated as |t3-t2| (assuming that the two signal segments are continuous in time and do not overlap).

[0183] For non-time series data such as images, texts, etc., the computer system can first convert the signal segments into feature vectors, and then use metrics such as Euclidean distance, Manhattan distance, or cosine similarity to calculate the signal interval. For example, if the third signal segment and the fourth signal segment are converted into feature vectors v1 and v2 respectively, the signal interval can be calculated as ‖v1-v2‖ (Euclidean distance).

[0184] For example, suppose the heart rate data and blood glucose data of a diabetic patient are being analyzed. The third signal segment is a time period in the heart rate data representing a rapid change in the patient's heart rate during that time period; the fourth signal segment is a time period in the blood glucose data representing a corresponding change in the patient's blood glucose level. If both signal segments are time series data, the computer system can calculate the time interval between them as the signal interval. For example, if the third signal segment has a start time of 9:00 AM and an end time of 9:10 AM, and the fourth signal segment has a start time of 9:05 AM and an end time of 9:15 AM, then the signal interval is 5 minutes.

[0185] After determining the signal interval, the computer system considers the signal coordinate reference of the third signal segment and the fourth signal segment, as well as the signal interval itself, to comprehensively evaluate the continuity between the two signal segments. The computer system can assign a reference score to each signal segment, which reflects the importance of the signal segment in the analysis. The reference score can be determined based on the characteristics (such as mean, variance, kurtosis, etc.), patterns (such as periodicity, trend, etc.), or outliers (such as extreme values, abrupt points, etc.) of the signal segment. For example, if the third signal segment contains a significant heart rate peak, and the fourth signal segment also shows a corresponding blood glucose peak, then the reference scores of the two signal segments can be higher.

[0186] In calculating the continuity evaluation parameter, the computer system can weight the signal interval to consider the reference of the signal coordinates. Specifically, if the reference scores of the two signal segments are both high, then even if the signal interval between them is slightly larger, the continuity evaluation parameter can still be high; conversely, if the signal interval is very small but the reference scores are low, then the continuity evaluation parameter can be low.

[0187] The continuity evaluation parameter can be calculated by the following formula: ; where the reciprocal of the signal interval is used to convert the interval into a quantity that is positively correlated with continuity (because the smaller the interval, the better the continuity), and the sum of the reference scores reflects the importance of the two signal segments in the analysis. This formula can be adjusted according to actual needs, such as by introducing a weight factor to balance the influence of the signal interval and the reference score.

[0188] For example, continuing with the heart rate and blood glucose data of a diabetic patient. Suppose the reference scores of the third signal segment and the fourth signal segment are 8 and 9 (out of a maximum of 10), respectively, and the signal interval is 5 minutes. According to the above formula, the continuity evaluation parameter can be calculated as: ; This value indicates that there is good continuity between the third signal segment and the fourth signal segment, although the time interval between them is 5 minutes, but since their reference scores are high (indicating that these two signal segments are very important in the analysis), the continuity evaluation parameter is still high.

[0189] Through the above steps, the computer system can accurately calculate the signal interval between the third signal segment and the fourth signal segment, and comprehensively consider the reference of the signal coordinates and the signal interval to determine the continuity evaluation parameter. This not only helps to evaluate the closeness and continuity between the two signal segments, but also provides strong support for subsequent data analysis and pattern recognition.

[0190] The embodiments of the present application provide a computer system, which comprises one or more processors, a memory, and one or more computer programs. Figure 2 As shown in the figure, the computer system 100 comprises a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, through a bus 102. Optionally, the computer system 100 can further comprise a transceiver 104. It should be noted that in actual application, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present application.

[0191] The embodiments of the present application provide a computer system, which comprises one or more processors, a memory, and one or more computer programs.

[0192] The above only describes some embodiments of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, some improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A method for storing medical monitoring sensor data based on machine learning, characterized in that, The method comprises: acquiring a target sensor signal related to a to-be-processed sensor signal, the target sensor signal covering different categories of sensor signals; acquiring, according to an influence factor of the different categories of sensor signals, a correlation degree between the different categories of sensor signals covered by the target sensor signal and the to-be-processed sensor signal; screening a selected category of sensor signals from the different categories of sensor signals according to the correlation degree between the different categories of sensor signals and the to-be-processed sensor signal; associatively storing a signal segment related to the to-be-processed sensor signal in the selected category of sensor signals; wherein the associatively storing the signal segment related to the to-be-processed sensor signal in the selected category of sensor signals comprises: retrieving a plurality of signal segments related to the to-be-processed sensor signal from the selected category of sensor signals; determining a continuity evaluation parameter between the plurality of signal segments through signal coordinates of the plurality of signal segments in the selected category of sensor signals; when the continuity evaluation parameter between the plurality of signal segments reaches a preset standard, associatively storing the signal segment related to the to-be-processed sensor signal in the selected category of sensor signals.

2. The method of claim 1, wherein, The acquiring, according to the influence factor of the different categories of sensor signals, the correlation degree between the different categories of sensor signals covered by the target sensor signal and the to-be-processed sensor signal comprises: performing signal segment splitting on the to-be-processed sensor signal to obtain a to-be-processed signal segment; acquiring a basic correlation degree between the different categories of sensor signals and the to-be-processed sensor signal according to the to-be-processed signal segment; acquiring the correlation degree between the different categories of sensor signals and the to-be-processed sensor signal according to the basic correlation degree between the different categories of sensor signals and the to-be-processed sensor signal and the influence factor of the different categories of sensor signals.

3. The method of claim 2, wherein, The acquiring, according to the to-be-processed signal segment, the basic correlation degree between the different categories of sensor signals and the to-be-processed sensor signal comprises: retrieving a target signal segment related to the to-be-processed signal segment from the different categories of sensor signals; determining the basic correlation degree between the different categories of sensor signals and the to-be-processed sensor signal according to a priority of the to-be-processed signal segment and a priority of the target signal segment related to the to-be-processed signal segment in the different categories of sensor signals.

4. The method of claim 1, wherein: the target sensor signal is a plurality of target sensor signals, and the method further comprises: determining an influence factor of each target sensor signal according to a correlation mechanism between the to-be-processed sensor signal and each target sensor signal; acquiring the correlation degree between each target sensor signal and the to-be-processed sensor signal according to the influence factor of each target sensor signal and the correlation degree between the different categories of sensor signals covered by each target sensor signal and the to-be-processed sensor signal. According to the correlation between each target sensor signal and the to-be-processed sensor signal, the plurality of target sensor signals are sorted; The target sensor signal related to the to-be-processed sensor signal comprises: The to-be-processed sensor signal is segmented into signal segments to obtain a to-be-processed signal segment set; The signal segments in the to-be-processed signal segment set are aggregated to obtain a plurality of signal segment clusters; Respectively obtain the target sensor signal related to the plurality of signal segment clusters, and one signal segment cluster corresponds to one correlation mechanism.

5. The method of claim 1, wherein, The selected category of sensor signals is filtered from the different categories of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, comprising: Filtering target category sensor signals and unselected category sensor signals other than the target category from the different categories of sensor signals; According to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, the to-be-determined category of sensor signals is filtered from the unselected category of sensor signals; The to-be-determined category of sensor signals and the target category of sensor signals are used as the selected category of sensor signals; The selected category of sensor signals related to the to-be-processed sensor signal is associatedly stored, and further comprising: Filtering target category signal segments from the to-be-processed signal segment set, wherein the to-be-processed signal segment set comprises signal segments obtained by segmenting the to-be-processed sensor signal into signal segments; The selected category of sensor signals related to the target category signal segment is associatedly stored.

6. The method of claim 5, wherein, The to-be-determined category of sensor signals is filtered from the unselected category of sensor signals according to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, comprising: According to the correlation between the different categories of sensor signals and the to-be-processed sensor signal, the original to-be-determined category of sensor signals with a correlation greater than or equal to a preset threshold is selected from the unselected category of sensor signals; The first N sensor signals with the largest correlation in the original to-be-determined category of sensor signals with a correlation greater than or equal to a preset threshold are used as the to-be-determined category of sensor signals, wherein N is a positive integer greater than or equal to 1.

7. The method of claim 6, wherein, The selected category of sensor signals related to the to-be-processed sensor signal is associatedly stored, further comprising: Determine the unselected category signal segments in the to-be-processed signal segment set other than the target category signal segments; The selected category of sensor signals related to the unselected category signal segments is used as a candidate signal segment set; M signal segments are filtered from the candidate signal segment set for associated storage, wherein M is a positive integer greater than or equal to 1.

8. A computer system, characterized by Comprise: One or more processors; Memory; One or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs, when executed by the processors, implement the method according to any one of claims 1-7.

Images