A remote monitoring and warning system for arteriovenous fistula

Abstract

The application relates to the technical field of data processing, and discloses an arteriovenous fistula remote monitoring and warning system which comprises the following steps: collecting monitoring data of multiple dimensions of an arteriovenous fistula; performing graph clustering on all-dimensional monitoring data at each moment to obtain a category division result at each moment, obtaining a change point and a change abnormality rate according to the category division result, and obtaining an abnormal time period, abnormal data and an abnormal dimension sequence; performing EMD decomposition on each abnormal dimension sequence to obtain a plurality of IMF components, obtaining a cycle length of each IMF component, obtaining replacement data and a conversion vector of each segment of abnormal data according to the cycle length, and compressing the replacement data and the monitoring data to obtain compressed monitoring data; and completing the arteriovenous fistula remote monitoring and warning according to the compressed monitoring data and the conversion vector. The application aims to solve the problem that abnormal data destroys the regularity of monitoring data and affects the compression efficiency and transmission efficiency.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a remote monitoring and alerting system for arteriovenous fistulas. Background Technology

[0002] Arteriovenous fistulas (AVFs), as a surgical procedure used in hemodialysis, require real-time monitoring of various physiological states in patients. AVF monitoring devices monitor the AVF's status in real time, acquiring a large amount of data. This data is transmitted to a terminal server, which analyzes the data, identifies abnormalities, and alerts the patient about the AVF's condition. This enables remote monitoring and alerting of AVFs. Because AVF monitoring data includes multiple data types, and each type is large in volume, data compression is necessary before transmission to ensure real-time monitoring.

[0003] However, abnormal data often disrupts the regularity of monitoring data, thereby affecting the compression efficiency of monitoring data and reducing the transmission efficiency of monitoring data. This, in turn, affects the remote real-time monitoring of arteriovenous fistulas. Therefore, there is a need for a method that can convert abnormal data into corresponding regular data, improve the regularity and correlation of the data, and thus improve the compression effect. At the same time, by retaining the conversion vector of abnormal data into regular data, the remote terminal can restore the real data through the conversion vector. This allows the remote terminal to obtain the abnormal time period at the same time as obtaining the real data, thereby improving the monitoring efficiency and warning effect, timely detection of abnormalities and taking corresponding measures to prevent the patient's condition from deteriorating. Summary of the Invention

[0004] This invention provides a remote monitoring and alerting system for arteriovenous fistulas to solve the problem that abnormal data disrupts the regularity of monitoring data, thus affecting compression and transmission efficiency. The specific technical solution adopted is as follows:

[0005] One embodiment of the present invention provides a remote monitoring and alerting system for arteriovenous fistulas, the system comprising:

[0006] The monitoring data acquisition module collects monitoring data from multiple dimensions of the arteriovenous fistula;

[0007] Data analysis and compression module: Based on the monitoring data of each dimension at each time, obtain the graph structure of each time, perform graph clustering on the graph structure to obtain the category classification results of each time, obtain the change points and the broken and added edges between the change points and the previous time based on the category classification results of adjacent time, obtain the change anomaly rate of each change point based on the category classification results of the change points and adjacent time, as well as the broken and added edges, obtain the anomaly rate of each change point, obtain the anomaly points based on the change anomaly rate, and obtain several abnormal time periods and several abnormal data and corresponding abnormal dimension sequences in each abnormal time period based on the anomaly points, change points and monitoring data of each dimension;

[0008] EMD decomposition is performed on each abnormal dimension sequence to obtain several IMF components. The period length of each IMF component is obtained through Fourier transform. Based on the IMF components and period length, alternative data and transformation vectors for each abnormal data segment are obtained. Compressed monitoring data is obtained by compressing the alternative data and monitoring data.

[0009] The remote monitoring and alert module completes remote monitoring and alerting of arteriovenous fistulas based on compressed monitoring data and converted vectors.

[0010] Optionally, the specific method for obtaining the graph structure at each time step includes:

[0011] Using any dimension as the target dimension, normalize the monitoring data of all times in the target dimension, and record the result as the quantized data of the target dimension at each time. Obtain the quantized data of each dimension at each time from the monitoring data of all dimensions.

[0012] Taking any time point as the target time point, obtain the quantized data of each dimension of the target time point, treat each dimension as a node, and the quantized data of each dimension as the node value. The boundary value between any two nodes is represented by the absolute value of the difference between the node values ​​of the two nodes. Based on the nodes, node values ​​and boundary values, obtain the graph structure of the target time point for all dimensions; obtain the graph structure of each time point.

[0013] Optionally, the specific method for obtaining the change point and the broken and added edges between the change point and the previous time step based on the category classification results of adjacent time steps includes:

[0014] Obtain the classification results of any two adjacent time points. If the classification result of the later time point differs from that of the earlier time point, record the later time point as the point of change.

[0015] If the classification results at two time points are the same, continue to judge the change points; judge the classification results at any two adjacent time points to obtain the change points at all time points;

[0016] Taking any point of change as the target point of change, obtain the classification results of the target point of change and the adjacent previous time step. The edge between any two nodes in each category is recorded as a valid edge, and the edge between nodes of different categories is recorded as an invalid edge. The edge that changes from a valid edge to an invalid edge from the previous time step to the target point of change is recorded as a broken edge, and the edge that changes from an invalid edge to a valid edge from the previous time step to the target point of change is recorded as an added edge. Obtain the broken edges and added edges between each point of change and the previous time step.

[0017] Optionally, the method for obtaining the anomaly rate of each change point includes: {\mathcal{C}}_{\mathfrak{i}=\frac {\mathfrak{f}\mathfrak{i},\mathfrak{l}} {\mathfrak{f}\mathfrak{i},\mathfrak{r}}}\times \left [ {{\mathfrak{f}}_{\mathfrak{i},\mathfrak{l}}{\times \mathfrak{a}}_{\mathfrak{i}}+\left ( {1-{\mathfrak{f}}_{\mathfrak{i},\mathfrak{r}}} \right )\times {\mathfrak{b}}_{\mathfrak{i}}} \right ] ,in Indicates the first The rate of change anomalies at each point of change Indicates the first The frequency of the classification result of the previous time step adjacent to each change point in the classification results of all time steps. Indicates the first The frequency of the class classification at the next adjacent time step after each change point in the class classification results across all time steps. Indicates the first The sum of the normalized boundary values ​​of each point of change and the previous time step of the broken edge. Indicates the first The sum of the normalized edge values ​​of each change point and the increasing edge at the previous time step; the sum of the normalized edge values ​​represents the sum of the normalized edge values ​​obtained by linearly normalizing all edge values ​​in the graph structure at the same time step and accumulating the normalization results.

[0018] Optionally, the specific method for obtaining several abnormal time periods and several abnormal data and corresponding abnormal dimension sequences in each abnormal time period includes:

[0019] Take any anomaly as the target anomaly, and record the time period between the target anomaly and the next adjacent change point as the abnormal time period corresponding to the target anomaly. Record the dimension corresponding to the node whose category changes in the classification result of the target anomaly and the previous time as the abnormal dimension of the target anomaly. Record all the monitoring data of the abnormal dimension in the abnormal time period corresponding to the target anomaly as the abnormal data of the target anomaly in the abnormal dimension. Record all the monitoring data of the abnormal dimension as the abnormal dimension sequence.

[0020] Obtain the abnormal time period corresponding to each abnormal point, as well as several abnormal data and corresponding abnormal dimension sequences within each abnormal time period.

[0021] Optionally, the specific method for obtaining the alternative data and transformation vector for each segment of abnormal data includes:

[0022] Take any abnormal time period as the target abnormal time period, any abnormal data segment in the target abnormal time period as the target abnormal data, and the abnormal dimension sequence corresponding to the target abnormal data as the target abnormal dimension sequence. Obtain the period length of each IMF component in the target abnormal dimension sequence, obtain the ratio of the number of time periods included in the target abnormal time period to the period length of each IMF component, and record the IMF component corresponding to the period length with the largest ratio as the reference component of the target abnormal data.

[0023] If the number of moments included in the target abnormal time period is less than the period length of the reference component, then the first part of the reference component will be... One data point was used as a substitute data point, among which Indicates the number of moments included in the target abnormal time period;

[0024] If the number of moments in the abnormal time period is greater than the period length of the reference component, all data in the first period of the reference component are repeated periodically until the amount of data is equal to the number of moments in the target abnormal time period. The data obtained by periodic repetition is used as the replacement data to obtain the replacement data for the target abnormal data of the target abnormal time period.

[0025] The ratio of the substitute data to the target anomaly data at the same time is used as the transformation value of the target anomaly data at the corresponding time. The transformation values ​​of each time in the target anomaly data are arranged in time sequence to form the transformation vector of the target anomaly data.

[0026] Obtain alternative data and transformation vectors for each segment of abnormal data within each abnormal time period.

[0027] The beneficial effects of this invention are as follows: This invention obtains category classification results by performing graph clustering on monitoring data from multiple dimensions at each time point. Compared to directly classifying categories based on the similarity of node values ​​in the graph structure, this unifies the category classification standards at different times, making the category classification results at different times more comparable. By obtaining change points through changes in the category classification results and quantifying the degree of change based on the degree of change of these change points, the invention obtains abnormal time periods and abnormal data, making the changes in abnormal information more obvious and ensuring the accuracy of abnormal data acquisition. By replacing abnormal data with abnormal dimension sequences and incorporating the IMF components from EMD decomposition, the invention ensures consistent periodic changes in the replacement data, minimizing disruption to the regularity of data fluctuations after replacement, while also improving data redundancy and compression efficiency. Furthermore, by transforming vectors, this invention enables remote terminals to quickly acquire and analyze abnormal data, improving the efficiency of remote monitoring of arteriovenous fistulas. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a structural block diagram of a remote monitoring and warning system for arteriovenous fistulas provided in one embodiment of the present invention;

[0030] Figure 2 This diagram illustrates the changes in category classification between adjacent time points. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 The diagram illustrates a structural block diagram of a remote monitoring and alerting system for arteriovenous fistulas according to an embodiment of the present invention. The system includes:

[0033] The monitoring data acquisition module S101 collects monitoring data of arteriovenous fistulas from multiple dimensions.

[0034] The purpose of this embodiment is to compress and transmit the monitoring data of arteriovenous fistulas (AVFs). The remote terminal receives and analyzes the data to achieve remote monitoring and alerts of AVFs. Therefore, it is necessary to first collect the monitoring data of the patient's AVF. There are multiple types of data in AVFs that need to be monitored in real time. In this embodiment, each type of data is recorded as a dimension of monitoring data. The multiple dimensions of monitoring data in this embodiment include blood flow data, arterial blood pressure data, venous blood pressure data, fistula pressure data, and urea content data. Real-time monitoring of AVFs is achieved through these five dimensions of data. In this embodiment, the sampling time interval for each dimension of data is 5 minutes. The monitoring device starts collecting data immediately after the AVF surgery is completed.

[0035] Thus, multi-dimensional monitoring data of the arteriovenous fistula were obtained.

[0036] Data analysis and compression module S102:

[0037] (1) Perform graph clustering on all dimensions of the monitoring data at each time point to obtain the classification results at each time point. Based on the classification results, obtain the change points and change anomaly rate to obtain the abnormal time period, abnormal data and abnormal dimension sequence.

[0038] It should be noted that when identifying anomalous data, relying solely on fluctuations in data from a single dimension often leads to the omission of certain anomalies. Therefore, it is necessary to further identify anomalies by examining changes in the correlations between data, thereby improving the anomaly detection rate. Furthermore, since graph structures can represent the correlations between different data, anomalous data can be calculated by observing changes in clustering patterns within the graph structure.

[0039] Specifically, firstly, taking monitoring data of any one dimension as an example, linear normalization is performed on the monitoring data of all times in that dimension, and the result is recorded as the quantized data of each time in that dimension. Quantized data for each dimension at each time is obtained for monitoring data of all dimensions. Taking any one time as an example, the quantized data of each dimension at that time is obtained, and each dimension is treated as a node. The quantized data of each dimension is the node value. The boundary value between any two nodes is represented by the absolute value of the difference between the node values ​​of the two nodes. Based on the nodes, node values, and boundary values, the graph structure for all dimensions at that time is obtained. The graph structure for each time is obtained using the above method.

[0040] Furthermore, taking the graph structure at any given time as an example, the Laplacian graph clustering method is used to obtain the category classification results for the nodes. For example, if there are five dimensions of monitoring data, namely A, B, C, D, and E, the category classification results at this time are (A, B, C) and (D, E), that is, A, B, and C are in one category, and D and E are in another category at this time. The category classification results at each time are obtained through graph clustering, where Laplacian graph clustering is a well-known technique and will not be described in detail in this embodiment.

[0041] It should be further explained that graph clustering is used instead of directly based on node value similarity for category classification because graph clustering considers not only the similarity between any two node values ​​but also the maximum separability of different categories during the category classification process. This results in significant differences between different categories. In addition, graph clustering performs clustering operations within the same structure, ensuring consistent clustering standards and high comparability of results. On the other hand, directly classifying categories based on node value similarity results in variations in the standards for node value similarity at different times. This leads to differences in the baseline of category classification results at different times, resulting in lower comparability and making it impossible to identify outliers based on changes in category classification results.

[0042] Specifically, for adjacent time points, the classification results of the two time points are obtained. If the classification result of the later time point differs from that of the earlier time point, the later time point is recorded as a change point. If the classification results of the two time points are the same, the change point judgment continues. The classification results of any two adjacent time points are judged according to the above method to obtain the change points in all time points. For any change point, the classification result of the change point and the adjacent previous time point are obtained. The edge between any two nodes in each category is recorded as a valid edge, that is, the edge between nodes within the same category is recorded as a valid edge, and the edge between nodes of different categories is recorded as an invalid edge. The edge from the previous time point to the change point that changes from a valid edge to an invalid edge is recorded as a broken edge, that is, the edge is valid at the previous time point and invalid at the change point. At the same time, the edge from the previous time point to the change point that changes from an invalid edge to a valid edge is recorded as an added edge. The broken edges and added edges between each change point and the previous time point are obtained according to the above method. Please refer to [link to relevant documentation]. Figure 2 It shows a schematic diagram of the changes in category classification between adjacent time points. Figure 2 In the left-hand category partitioning results, the valid edges are AB, AC, BC, and DE. In the right-hand category partitioning results, the valid edges are AB, CD, CE, and DE. Therefore, the broken edges are AC and BC, while the added edges are CD and CE.

[0043] Furthermore, based on the classification results of the change point and adjacent time points, as well as the breakage and addition of edges, the change anomaly rate of each change point is obtained, with the first... Taking a single point of change as an example, its abnormality rate The calculation method is: {\mathcal{C}}_{\mathfrak{i}=\frac {\mathfrak{f}\mathfrak{i}, \mathfrak{l}} {{\mathfrak{f}}_{\mathfrak{i}, \mathfrak{r}}}}\times \left [ {{\mathfrak{f}}_{\mathfrak{i},\mathfrak{l}}\times {\mathfrak{a}}_{\mathfrak{i}}+\left ( {1-{\mathfrak{f}}_{\mathfrak{i},\mathfrak{r}}} \right )\times {\mathfrak{b}}_{\mathfrak{i}}} \right ] ,in, Indicates the first The frequency of the classification result of the previous time step adjacent to each change point in the classification results of all time steps. Indicates the first The frequency of the class classification at the next adjacent time step after each change point in the class classification results across all time steps. Indicates the first The sum of the normalized boundary values ​​of each point of change and the previous time step of the broken edge. Indicates the first The sum of the normalized edge values ​​of the changing point and the increasing edge at the previous time step; it should be noted that, in order to avoid large differences in the dimensions of edge values ​​between different nodes, which would lead to large differences in the abnormality rate between different time steps, it is necessary to linearly normalize all edge values ​​in the graph structure at the same time step, and accumulate the normalization results to obtain the sum of the normalized edge values; the higher the frequency of the classification result, the greater the probability that the corresponding time step is a normal time step, and the larger the ratio of the two frequencies, the more it indicates that the first change point is normal. The greater the rate of change when each point of change transitions from a normal time to an abnormal time, the greater the cumulative sum of edge values ​​of broken edges, and the more the relationships between normal data are reduced. The greater the probability of an anomaly occurring at a given point of change, the more the frequency of occurrence at the previous time step is used to characterize normality, further amplifying the anomaly effect caused by edge breakage, resulting in a higher anomaly rate. Similarly, the larger the cumulative sum of edge values ​​of added edges, the more relationships between anomalous data are increased, and the greater the probability of anomalies occurring. Combining this with the frequency of occurrence at the next time step to characterize anomalies, the smaller the frequency table, the greater the anomaly and the higher the anomaly rate.

[0044] Further, following the above method, the anomaly rate of each change point is obtained. All anomaly rates are linearly normalized, and a preset first threshold is given to determine anomalies. In this embodiment, the preset first threshold is calculated using 0.7. Change points whose normalized anomaly rate is greater than the preset first threshold are recorded as anomalies, and several anomalies are obtained. The time period between each anomaly point and the next adjacent change point is recorded as the anomaly time period. For any anomaly point, the dimension corresponding to the node whose category changed in the category classification result of the anomaly point and the previous time is recorded as the anomaly dimension of the anomaly point. All monitoring data of the anomaly dimension in the anomaly time period corresponding to the anomaly point are recorded as the anomaly data of the anomaly point in the anomaly dimension. All monitoring data of the anomaly dimension are recorded as the anomaly dimension sequence. Since the monitoring data is time-series data, all monitoring data of the anomaly dimension is also time-series data. The anomaly time period, anomaly data, and anomaly dimension sequence corresponding to each anomaly point are obtained according to the above method. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 The dimension that undergoes category change is dimension C. Figure 2 If the latter category classification result is the category classification result of the outlier, then dimension C is the outlier dimension of that outlier.

[0045] At this point, several abnormal time periods, abnormal data, and abnormal dimension sequences have been obtained.

[0046] (2) Perform EMD decomposition on each abnormal dimension sequence to obtain several IMF components, obtain the period length of each IMF component through Fourier transform, obtain the alternative data and transformation vector of each abnormal data according to the IMF component and its period length, and compress the monitoring data according to the alternative data and monitoring data to obtain the compressed monitoring data.

[0047] It should be noted that normal data has strong regularity and is easy to compress, while abnormal data will disrupt the regularity of the data and reduce the compression rate. Therefore, we consider replacing the abnormal data with a portion of the data in the abnormal dimension sequence corresponding to the abnormal data to improve the redundancy rate and thus ensure the compression effect. At the same time, in order to ensure that the replacement data conforms to the regularity, we consider performing EMD decomposition on the abnormal dimension sequence and replacing the IMF component with a period length similar to the abnormal time period to ensure that the regularity of data fluctuation is not destroyed, thereby ensuring a high redundancy rate of the replaced data.

[0048] Specifically, for any segment of abnormal data within any abnormal time period, the abnormal dimension sequence corresponding to the abnormal data is obtained. EMD decomposition is performed on this abnormal dimension sequence to obtain several IMF components. A Fourier transform is performed on each IMF component to convert it to the frequency domain. The frequency of each IMF component is obtained based on the transformation result. The reciprocal of the frequency yields the period of each IMF component, denoted as the period length of each IMF component. The period length represents the periodic data fluctuation of the IMF component over a period of time. Calculating the period using Fourier transform is existing technology and will not be elaborated upon in this embodiment. The abnormal time period includes several time points. It should be noted that the abnormal time period includes the abnormal point but does not include the next adjacent change point. The ratio of the number of time points included in the abnormal time period to the period length of each IMF component is obtained, where the ratio is calculated by dividing the smaller value by the larger value. The IMF component corresponding to the period length with the largest ratio is denoted as the reference component of the abnormal data. Replacement data for the abnormal data is obtained based on the reference component, the period length, and the abnormal time period. If the number of time points included in the abnormal time period is less than the period length of the reference component, then the first... One data point was used as a substitute data point, among which This refers to the number of moments included in the abnormal time period. If the number of moments included in the abnormal time period is greater than the period length of the reference component, then all data within the first period of the reference component are repeated periodically until the data volume equals the number of moments included in the abnormal time period. The data obtained from this periodic repetition is used as replacement data, thus obtaining the replacement data for that segment of abnormal data in the abnormal time period. For the replacement data and the abnormal data, the ratio of the data at the same moment in the replacement data and the abnormal data is used as the transformation value of that moment in the abnormal data. The transformation values ​​of each moment in the abnormal data are arranged in chronological order to form the transformation vector of the abnormal data. The replacement data and transformation vector of each segment of abnormal data in each abnormal time period are obtained according to the above method.

[0049] Furthermore, abnormal data is replaced by alternative data, and combined with other data in the monitoring data to obtain transformed monitoring data. In this embodiment, the transformed monitoring data of each dimension is compressed using LZ77 to obtain compressed monitoring data. At the same time, abnormal data is marked with its abnormal points and abnormal dimensions.

[0050] This completes the compression of the monitoring data, resulting in compressed monitoring data.

[0051] The remote monitoring and warning module S103 completes remote monitoring and warning of arteriovenous fistulas based on compressed monitoring data and converted vectors.

[0052] The compressed monitoring data, transformation vectors, and marked outliers and outlier dimensions are transmitted to a remote terminal. Each transformation vector corresponds to a set of outliers and outlier dimensions. After receiving the transmitted data, the remote terminal first decompresses the data according to the LZ77 compression method. At the same time, it restores the corresponding outlier data according to the transformation vectors, outliers, and outlier dimensions. Based on the outlier data, remote monitoring of arteriovenous fistulas can be achieved. The monitoring results can be used to determine whether an abnormal state has occurred in the arteriovenous fistula. If an abnormal state has occurred, the patient will be alerted.

[0053] Thus, remote monitoring and alerting of arteriovenous fistulas were completed. Furthermore, due to the existence of the transformation vector, the remote terminal can quickly obtain and analyze abnormal data, thereby improving the efficiency of remote monitoring.

[0054] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A remote monitoring and alert system for arteriovenous fistulas, characterized in that, The system includes: The monitoring data acquisition module collects monitoring data from multiple dimensions of the arteriovenous fistula; Data analysis and compression module: Based on the monitoring data of each dimension at each time, obtain the graph structure of each time, perform graph clustering on the graph structure to obtain the category classification results of each time, obtain the change points and the broken and added edges between the change points and the previous time based on the category classification results of adjacent time, obtain the change anomaly rate of each change point based on the category classification results of the change points and adjacent time, as well as the broken and added edges, obtain the anomaly rate of each change point, obtain the anomaly points based on the change anomaly rate, and obtain several abnormal time periods and several abnormal data and corresponding abnormal dimension sequences in each abnormal time period based on the anomaly points, change points and monitoring data of each dimension; EMD decomposition is performed on each abnormal dimension sequence to obtain several IMF components. The period length of each IMF component is obtained through Fourier transform. Based on the IMF components and period length, alternative data and transformation vectors for each abnormal data segment are obtained. Compressed monitoring data is obtained by compressing the alternative data and monitoring data. The remote monitoring and alert module completes remote monitoring and alerting of arteriovenous fistulas based on compressed monitoring data and converted vectors; The specific methods for obtaining the alternative data and transformation vector for each segment of abnormal data are as follows: Take any abnormal time period as the target abnormal time period, any abnormal data segment in the target abnormal time period as the target abnormal data, and the abnormal dimension sequence corresponding to the target abnormal data as the target abnormal dimension sequence. Obtain the period length of each IMF component in the target abnormal dimension sequence, obtain the ratio of the number of time periods included in the target abnormal time period to the period length of each IMF component, and record the IMF component corresponding to the period length with the largest ratio as the reference component of the target abnormal data. If the number of moments included in the target abnormal time period is less than the period length of the reference component, the first n data in the reference component will be used as replacement data, where n represents the number of moments included in the target abnormal time period. If the number of moments in the abnormal time period is greater than the period length of the reference component, all data in the first period of the reference component are repeated periodically until the amount of data is equal to the number of moments in the target abnormal time period. The data obtained by periodic repetition is used as the replacement data to obtain the replacement data for the target abnormal data of the target abnormal time period. The ratio of the substitute data to the target anomaly data at the same time is used as the transformation value of the target anomaly data at the corresponding time. The transformation values ​​of each time in the target anomaly data are arranged in time sequence to form the transformation vector of the target anomaly data. Obtain alternative data and transformation vectors for each segment of abnormal data within each abnormal time period.

2. The remote monitoring and warning system for arteriovenous fistulas according to claim 1, characterized in that, The specific method for obtaining the graph structure at each time step is as follows: Using any dimension as the target dimension, normalize the monitoring data of all times in the target dimension, and record the result as the quantized data of the target dimension at each time. Obtain the quantized data of each dimension at each time from the monitoring data of all dimensions. Taking any time point as the target time point, obtain the quantized data of each dimension of the target time point, treat each dimension as a node, and the quantized data of each dimension as the node value. The boundary value between any two nodes is represented by the absolute value of the difference between the node values ​​of the two nodes. Based on the nodes, node values ​​and boundary values, obtain the graph structure of the target time point for all dimensions; obtain the graph structure of each time point.

3. The remote monitoring and warning system for arteriovenous fistulas according to claim 1, characterized in that, The specific method for obtaining the change points and the broken and added edges between the change points and the previous time step based on the classification results of adjacent time steps includes: Obtain the classification results of any two adjacent time points. If the classification result of the later time point differs from that of the earlier time point, record the later time point as the point of change. If the classification results at two time points are the same, continue to judge the change points; judge the classification results at any two adjacent time points to obtain the change points at all time points; Taking any point of change as the target point of change, obtain the classification results of the target point of change and the adjacent previous time step. The edge between any two nodes in each category is recorded as a valid edge, and the edge between nodes of different categories is recorded as an invalid edge. The edge that changes from a valid edge to an invalid edge from the previous time step to the target point of change is recorded as a broken edge, and the edge that changes from an invalid edge to a valid edge from the previous time step to the target point of change is recorded as an added edge. Obtain the broken edges and added edges between each point of change and the previous time step.

4. The remote monitoring and warning system for arteriovenous fistulas according to claim 1, characterized in that, The specific method for obtaining the anomaly rate of each change point is as follows: in, Indicates the first The rate of change anomalies at each point of change Indicates the first The frequency of the classification result of the previous time step adjacent to each change point in the classification results of all time steps. Indicates the first The frequency of the class classification at the next adjacent time step after each change point in the class classification results across all time steps. Indicates the first The sum of the normalized boundary values ​​of each point of change and the previous time step of the broken edge. Indicates the first The sum of the normalized edge values ​​of each point of change and the increasing edge at the previous time step; The sum of edge value normalization represents the linear normalization of all edge values ​​in the graph structure at the same time, and the sum of the normalization results is obtained by accumulating the edge value normalization.

5. The remote monitoring and warning system for arteriovenous fistulas according to claim 1, characterized in that, The specific method for obtaining several abnormal time periods and several abnormal data and corresponding abnormal dimension sequences in each abnormal time period is as follows: Take any anomaly as the target anomaly, and record the time period between the target anomaly and the next adjacent change point as the abnormal time period corresponding to the target anomaly. Record the dimension corresponding to the node whose category changes in the classification result of the target anomaly and the previous time as the abnormal dimension of the target anomaly. Record all the monitoring data of the abnormal dimension in the abnormal time period corresponding to the target anomaly as the abnormal data of the target anomaly in the abnormal dimension. Record all the monitoring data of the abnormal dimension as the abnormal dimension sequence. Obtain the abnormal time period corresponding to each abnormal point, as well as several abnormal data and corresponding abnormal dimension sequences within each abnormal time period.

Images