Medical equipment performance indicator anomaly detection method and device based on PU learning

Abstract

This application proposes a method and device for detecting anomalies in medical device performance indicators based on PU learning. The method includes: using historical key performance indicator (KPI) streams as training data, clustering the training data according to similarity, obtaining the centroid curve of each cluster, and labeling the centroid curve of each cluster to obtain first anomaly labeled data and first unlabeled data; based on the first anomaly labeled data and the first unlabeled data, constructing a binary classifier through positive example unlabeled PU learning, and combining active learning to obtain anomaly labels and normal labels for the centroid curve of each cluster; obtaining the labels on the centroid curve of the cluster corresponding to the KPI stream to be detected, training the anomaly detection model corresponding to the KPI stream to be detected through semi-supervised learning, and detecting the KPI stream to be detected through the anomaly detection model. This method improves the accuracy of anomaly detection of medical device performance indicators while minimizing the labeling workload.

Description

Technical Field

[0001] The present application relates to the field of data detection technology, and in particular to a method and device for detecting abnormal performance indicators of medical equipment based on PU learning. Background Art

[0002] Currently, to ensure the reliability of medical equipment, medical professionals need to continuously collect and monitor a large amount of key performance indicator (KPI) data streams of medical equipment so that they can repair the equipment in a timely manner if an anomaly occurs. Therefore, KPI anomaly detection is crucial for medical equipment management.

[0003] In the medical field, KPI stream anomaly detection technology typically includes supervised, semi-supervised, and unsupervised algorithms. Supervised algorithms require manual labeling of all training set samples, using the features and labels of the KPI stream as input and utilizing machine learning algorithms to detect time series anomalies. Unsupervised algorithms do not require labeling and instead process the KPI stream and input it into a deep learning model to generate anomaly detection results. Semi-supervised algorithms combine the advantages of both algorithms, requiring only partially labeled data while utilizing both labeled and unlabeled data points to design anomaly detection algorithms.

[0004] However, the applicant found that among the above algorithms, the supervised algorithm requires a large amount of labeled data. However, in actual situations, KPI streams are often large-scale and diverse, and this labeling work requires a lot of time and energy costs, which is often difficult to achieve; the unsupervised algorithm has low accuracy and requires a large amount of training data. In actual scenarios, the pattern of KPI streams often changes dynamically, so this algorithm has poor practical applicability; although the semi-supervised algorithm reduces the labeling cost compared to the supervised algorithm and improves the detection accuracy of the model compared to the unsupervised algorithm, it still requires staff to accurately label all anomalies in a certain time period in a large number of KPI streams. Repeatedly confirming whether a KPI section has anomalies still brings a large workload. Therefore, the detection methods in the relevant technology, whether supervised learning methods, semi-supervised learning methods, or unsupervised learning methods, cannot achieve accurate anomaly detection for large-scale, diverse, and dynamically changing KPI streams with very little labeling work. Summary of the Invention

[0005] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] To this end, the first purpose of this application is to propose a medical equipment performance indicator anomaly detection method based on PU learning. This method is based on the PU learning method and can achieve accurate KPI anomaly detection with only a small number of labels. By integrating clustering, PU learning and semi-supervised learning, the accuracy of anomaly detection is improved while minimizing the labeling workload.

[0007] The second object of the present invention is to propose a device for detecting abnormalities in performance indicators of medical equipment based on PU learning.

[0008] A third object of the present invention is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, the first embodiment of the present invention proposes a method for detecting abnormalities in medical device performance indicators based on PU learning, comprising the following steps:

[0010] Using historical key performance indicator (KPI) streams as training data, clustering the training data according to similarity, obtaining a centroid curve for each cluster, and labeling the centroid curve for each cluster to obtain first abnormally labeled data and first unlabeled data;

[0011] Based on the first abnormal labeled data and the first unlabeled data, a binary classifier is constructed through positive unlabeled PU learning, and active learning is combined to obtain abnormal labels and normal labels for the centroid curve of each cluster;

[0012] The labels on the centroid curve of the cluster corresponding to the KPI flow to be detected are obtained, an anomaly detection model corresponding to the KPI flow to be detected is trained through semi-supervised learning, and the KPI flow to be detected is detected by the anomaly detection model.

[0013] Optionally, in one embodiment of the present application, the centroid curve of each cluster is labeled to obtain first abnormal labeled data and first unlabeled data, including: manually marking abnormal segments on the centroid curve of each cluster, using the labeled data in the abnormal segments as the first abnormal labeled data, and using the data on the centroid curve except the abnormal segments as the first unlabeled data.

[0014] Optionally, in one embodiment of the present application, obtaining the abnormal label and normal label of the centroid curve of each cluster includes: determining first normal labeled data in the first unlabeled data through the binary classifier; performing multiple iterative labeling on the first unlabeled data through an active learning method, and determining second normal labeled data and second abnormal labeled data in the first unlabeled data; using the first abnormal labeled data and the second abnormal labeled data as the abnormal labels, and using the first normal labeled data and the second normal labeled data as the normal labels.

[0015] Optionally, in one embodiment of the present application, obtaining the label on the centroid curve of the cluster corresponding to the KPI flow to be detected includes: determining the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster; reading the abnormal label and normal label on the centroid curve of the cluster corresponding to the KPI flow to be detected.

[0016] Optionally, in one embodiment of the present application, the anomaly detection model corresponding to the KPI stream to be detected is trained through semi-supervised learning, including: selecting a preset amount of data from the KPI stream to be detected as second unlabeled data; based on the anomaly label, normal label and second unlabeled data on the centroid curve of the cluster corresponding to the KPI stream to be detected, training the anomaly detection model corresponding to the KPI stream to be detected.

[0017] To achieve the above objectives, the second embodiment of the present application proposes a medical device performance indicator anomaly detection device based on PU learning, including the following modules:

[0018] A clustering module is configured to cluster the training data according to similarity using a historical key performance indicator (KPI) stream as training data, obtain a centroid curve for each cluster, and label the centroid curve for each cluster to obtain first anomaly labeled data and first unlabeled data;

[0019] an acquisition module, configured to construct a binary classifier through positive unlabeled PU learning based on the first abnormal labeled data and the first unlabeled data, and to obtain an abnormal label and a normal label of the centroid curve of each cluster in combination with active learning;

[0020] The training module is used to obtain the labels on the centroid curve of the cluster corresponding to the KPI flow to be detected, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning, and detect the KPI flow to be detected through the anomaly detection model.

[0021] Optionally, in one embodiment of the present application, the clustering module is specifically used to: manually mark abnormal segments on the centroid curve of each cluster, use the labeled data in the abnormal segments as the first abnormal labeled data, and use the data on the centroid curve except the abnormal segments as the first unlabeled data.

[0022] Optionally, in one embodiment of the present application, the acquisition module is specifically used to: determine first normal labeled data in the unlabeled data through the binary classifier; perform multiple iterative labeling on the first unlabeled data through an active learning method, and determine second normal labeled data and second abnormal labeled data in the first unlabeled data; use the first abnormal labeled data and the second abnormal labeled data as the abnormal labels, and use the first normal labeled data and the second normal labeled data as the normal labels.

[0023] Optionally, in one embodiment of the present application, the training module is specifically used to: determine the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster; read the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected.

[0024] Optionally, in one embodiment of the present application, the training module is also used to: select a preset amount of data from the KPI stream to be detected as second unlabeled data; and train an anomaly detection model corresponding to the KPI stream to be detected based on the abnormal labels, normal labels and the second unlabeled data on the centroid curve of the cluster corresponding to the KPI stream to be detected.

[0025] The present application has the following technical effects: the present application uses historical key performance indicator KPI streams as training data, clusters the training data according to the degree of similarity, obtains the centroid curve of each cluster, and labels the centroid curve of each cluster to obtain the first abnormal labeled data and the first unlabeled data; based on the first abnormal labeled data and the first unlabeled data, a binary classifier is constructed through positive example unlabeled PU learning, and active learning is combined to obtain the abnormal label and normal label of the centroid curve of each cluster; the label on the centroid curve of the cluster corresponding to the KPI stream to be detected is obtained, and the anomaly detection model corresponding to the KPI stream to be detected is trained through semi-supervised learning, and the KPI stream to be detected is detected through the anomaly detection model. The present application is based on the PU learning method, which realizes accurate KPI anomaly detection with only a small number of labels. By integrating clustering, PU learning and semi-supervised learning, the accuracy of anomaly detection is improved while minimizing the labeling workload.

[0026] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the medical device performance indicator abnormality detection method based on PU learning described in the first aspect embodiment of the present application.

[0027] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0029] Figure 1 This is a flowchart of a method for detecting abnormal performance indicators of medical equipment based on PU learning, provided by one embodiment of the present application;

[0030] Figure 2 This is a flowchart of a specific method for detecting abnormal performance indicators of medical equipment based on PU learning, provided in one embodiment of the present application;

[0031] Figure 3 This is a structural diagram of a medical equipment performance indicator anomaly detection device based on PU learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0032] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0033] It should be noted that for a medical device, the number of KPI streams generated during operation is large and the patterns are diverse. On the one hand, if a PU learning model is trained for each KPI stream, the overall labeling workload is very large. On the other hand, if a PU learning model is trained for all KPI streams, since different KPI streams have different patterns, the most appropriate anomaly detectors and parameters for different KPI streams may be significantly different, so the model will suffer from low precision. To address the above technical problems, this application proposes a method and apparatus for detecting anomalies in medical device performance indicators based on PU learning.

[0034] The following describes the medical equipment performance indicator anomaly detection method and apparatus based on PU learning according to an embodiment of the present application with reference to the accompanying drawings.

[0035] Figure 1 This is a flowchart of a method for detecting abnormal performance indicators of medical equipment based on PU learning provided by an embodiment of the present application. Figure 1 As shown, the method includes the following steps:

[0036] Step S101: Using historical key performance indicator (KPI) streams as training data, clustering the training data according to similarity, obtaining a centroid curve for each cluster, and labeling the centroid curve for each cluster to obtain first abnormally labeled data and first unlabeled data.

[0037] Among them, the key performance indicator (KPI) stream is a data stream of key indicators reflecting the operating performance of medical equipment obtained when monitoring medical equipment. For example, the KPI stream may include the response delay of the medical equipment or the data stream of network throughput, etc. The KPI stream is usually a time series.

[0038] Specifically, in this application, pre-stored historical KPI streams can be obtained in different ways as training sets for the KPI stream anomaly detection model of this application. The training data is then clustered according to similarity, and the training data is divided into different clusters through clustering, and the centroid of each cluster is determined.

[0039] It should be noted that since the present application is aimed at clustering the KPI flow of time series, that is, the KPI flow is represented in the form of a line, therefore, in one embodiment of the present application, a method of clustering each discrete curve can be adopted to cluster the training data according to the similarity of the curves, and then obtain the centroid curve of each cluster after clustering.

[0040] Furthermore, the centroid curve of each cluster is annotated to obtain abnormal data and unannotated data in each centroid curve. In one embodiment of the present application, manual annotation can be used to allow staff to manually mark some abnormal segments on the centroid curve of each cluster. The number of marked abnormal segments can be set according to actual needs, and the annotated data in the abnormal segments is used as the first abnormal annotated data, and the data on each centroid curve other than the annotated abnormal segments is used as the first unannotated data.

[0041] It should also be noted that, since after clustering, each cluster corresponds to a centroid curve, and the number of clusters is far less than the number of time series curves in the training set, the workload of manual labeling by the staff in this application can be minimized.

[0042] Step S102: Based on the first abnormal labeled data and the first unlabeled data, a binary classifier is constructed through positive unlabeled PU learning, and active learning is combined to obtain the abnormal label and normal label of the centroid curve of each cluster.

[0043] Among them, the binary classifier is a classifier used to mark abnormal data and normal data in unlabeled data, and the abnormal labels and normal labels are abnormal labeled samples and normal labeled samples with higher reliability.

[0044] In one embodiment of the present application, a binary classifier is first constructed based on the manually labeled first abnormal labeled data and the first unlabeled data through positive example unlabeled PU learning, and the first normal labeled data is determined in the first unlabeled data through the binary classifier. Then, normal data and abnormal data are continued to be labeled in the unlabeled data through active learning, and based on the results of the previous round of labeling, the labeling step is repeated, and the unlabeled data is iteratively labeled multiple times, so that samples that may be abnormal are selected from the unlabeled data through active learning for labeling, and a larger number of second normal labeled data and second abnormal labeled data with higher reliability are determined. Finally, the first abnormal labeled data and the second abnormal labeled data corresponding to each centroid curve are used as the abnormal label of the centroid curve, and the first normal labeled data and the second normal labeled data of each centroid curve are used as the normal label of the centroid curve.

[0045] Thus, the abnormal labels and normal labels of the centroid curve of each cluster are obtained.

[0046] Step S103: Obtain the label on the centroid curve of the cluster corresponding to the KPI flow to be detected, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning, and detect the KPI flow to be detected through the anomaly detection model.

[0047] In one embodiment of the present application, for any newly acquired KPI stream that requires anomaly detection, the cluster corresponding to the KPI stream to be detected can be determined by first calculating the shape similarity between the KPI stream to be detected and the centroid curve of each cluster determined after classification. That is, the KPI to be detected can be assigned to an existing cluster. Then, since step S102 has determined the anomaly label and normal label of the centroid curve of each cluster, the anomaly label and normal label on the centroid curve of the cluster corresponding to the KPI stream to be detected (i.e., the cluster to which it belongs) can be directly read.

[0048] Furthermore, based on the labels obtained on the centroid curve of the cluster corresponding to the KPI stream and the data of the KPI stream itself, the anomaly detection model corresponding to the KPI stream to be detected is trained through semi-supervised learning. As an example, training the anomaly detection model corresponding to the KPI stream to be detected through semi-supervised learning includes first selecting a preset amount of data from the KPI stream to be detected as the second unlabeled data, wherein the selected data can be set according to actual needs, for example, selecting the top 20% of the data from the KPI stream to be detected. That is, a preset amount of data is selected from the KPI stream to be detected as unlabeled data to participate in model training. Then, based on the abnormal labels, normal labels, and the second unlabeled data on the centroid curve of the cluster corresponding to the KPI stream to be detected, the anomaly detection model corresponding to the KPI stream to be detected is trained using a semi-supervised algorithm in the relevant technology.

[0049] Furthermore, the KPI stream to be detected is detected by the trained anomaly detection model. In an embodiment of the present application, the remaining data in the KPI stream to be detected is detected by the trained anomaly detection model corresponding to the KPI stream to be detected. Continuing with the above example, if the first 20% of the data from the KPI stream to be detected is selected as unlabeled data for model training, the trained anomaly detection model can be used to perform anomaly detection on the remaining 80% or 60% of the data in the KPI stream to be detected.

[0050] It should be noted that in some other embodiments of the present application, a preset number of second unlabeled data can be selected from the KPI stream to be detected and returned to the PU learning, that is, the first unlabeled data and the second unlabeled data are merged as unlabeled data. After the binary classifier is actively learned through PU learning, the abnormal label and normal label of the centroid curve of each cluster are obtained, that is, a preset number of data are selected from the KPI stream to be detected as unlabeled data to participate in PU learning and active learning to determine the abnormal label and normal label of the centroid curve of the cluster corresponding to the KPI stream to be detected, and then the determined abnormal label, normal label and second unlabeled data are combined to train the anomaly detection model corresponding to the KPI stream to be detected.

[0051] Therefore, the medical equipment performance indicator anomaly detection method based on PU learning in this application reduces the workload of labeling while improving the accuracy of anomaly detection and improving the practicality in actual applications.

[0052] In summary, the medical equipment performance indicator anomaly detection method based on PU learning in the embodiment of the present application uses historical key performance indicator KPI streams as training data, clusters the training data according to the degree of similarity, obtains the centroid curve of each cluster, and labels the centroid curve of each cluster to obtain the first anomaly labeled data and the first unlabeled data; based on the first anomaly labeled data and the first unlabeled data, a binary classifier is constructed through positive example unlabeled PU learning, and active learning is combined to obtain the anomaly label and normal label of the centroid curve of each cluster; the label on the centroid curve of the cluster corresponding to the KPI stream to be detected is obtained, and the anomaly detection model corresponding to the KPI stream to be detected is trained through semi-supervised learning, and the KPI stream to be detected is detected through the anomaly detection model. This method is based on the PU learning method, which realizes accurate KPI anomaly detection with only a small number of labels. By integrating clustering, PU learning and semi-supervised learning, the accuracy of anomaly detection is improved while minimizing the labeling workload.

[0053] In order to more clearly illustrate the medical equipment performance indicator anomaly detection method based on PU learning of this application, the following Figure 2Let's take a specific example to illustrate.

[0054] like Figure 2 As shown, the PU learning-based anomaly detection framework PUAD includes an offline training process and an online detection process, which specifically includes three steps. First, clustering is performed: the time series of the training set are clustered according to their similarity. For each cluster, the staff will manually mark some abnormal segments for its centroid curve. Because the number of clusters is much smaller than the number of time series curves, the staff's labeling work can be minimized. In addition, a newly emerging KPI stream can be assigned to an existing cluster by calculating its shape similarity with the centroid of each cluster. Second, PU learning is performed: for each cluster centroid, PUAD applies PU learning to construct a binary classifier using existing anomaly annotations and unlabeled samples. Then, active learning methods are used to obtain reliable anomaly samples in multiple iterations. In this way, cluster centroids containing reliable anomaly and normal labels can be obtained. Third, semi-supervised learning is performed: for each KPI stream, an anomaly detection model will be trained based on the label on its cluster centroid. It should be noted that, Figure 2 It can be seen that before the second step of PU learning, feature extraction can be performed on the centroid curve and KPI flow to perform anomaly detection using the extracted features.

[0055] like Figure 3 As shown, in order to achieve the above-mentioned purpose, the second embodiment of the present application proposes a medical equipment performance indicator anomaly detection device based on PU learning of the present application, including: a clustering module 100, an acquisition module 200 and a training module 300.

[0056] Among them, the clustering module 100 is used to cluster the training data according to the degree of similarity using the historical key performance indicator KPI stream as training data, obtain the centroid curve of each cluster, and label the centroid curve of each cluster to obtain the first abnormal labeled data and the first unlabeled data.

[0057] The acquisition module 200 is used to build a binary classifier based on the first abnormal labeled data and the first unlabeled data through positive example unlabeled PU learning, and obtain the abnormal label and normal label of the centroid curve of each cluster in combination with active learning.

[0058] The training module 300 is used to obtain the labels on the centroid curve of the cluster corresponding to the KPI flow to be detected, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning, and detect the KPI flow to be detected through the anomaly detection model.

[0059] Optionally, in one embodiment of the present application, the clustering module 100 is specifically used to manually mark abnormal segments on the centroid curve of each cluster, with the labeled data in the abnormal segments as the first abnormal labeled data, and the data other than the abnormal segments on the centroid curve as the first unlabeled data.

[0060] Optionally, in one embodiment of the present application, the acquisition module 200 is specifically used to: determine first normal labeled data in the unlabeled data through a binary classifier; perform multiple iterative labeling on the first unlabeled data through an active learning method, and determine second normal labeled data and second abnormal labeled data in the first unlabeled data; use the first abnormal labeled data and the second abnormal labeled data as abnormal labels, and use the first normal labeled data and the second normal labeled data as normal labels.

[0061] Optionally, in one embodiment of the present application, the training module 300 is specifically used to: determine the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster; read the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected.

[0062] Optionally, in one embodiment of the present application, the training module 300 is further used to: select a preset amount of data from the KPI stream to be detected as the second unlabeled data; and train the anomaly detection model corresponding to the KPI stream to be detected based on the abnormal labels, normal labels and second unlabeled data on the centroid curve of the cluster corresponding to the KPI stream to be detected.

[0063] It should be noted that the above description of the embodiment of the medical equipment performance indicator abnormality detection method based on PU learning is also applicable to the embodiment of the device, and the implementation principle is the same, which will not be repeated here.

[0064] In summary, the medical equipment performance indicator anomaly detection device based on PU learning in the embodiment of the present application, based on the PU learning method, can achieve accurate KPI anomaly detection with only a small number of labels. By integrating clustering, PU learning and semi-supervised learning, it improves the accuracy of anomaly detection while minimizing the labeling workload.

[0065] In order to implement the above-mentioned embodiments, the present invention also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements a medical device performance indicator abnormality detection method based on PU learning described in the first aspect of the embodiment of the present application.

[0066] Although the present application is disclosed in detail with reference to the accompanying drawings, it should be understood that these descriptions are merely exemplary and are not intended to limit the application of the present application. The scope of protection of the present application is defined by the appended claims and may include various modifications, alterations and equivalents made to the invention without departing from the scope and spirit of the present application.

[0067] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0068] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0069] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0070] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0071] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0072] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0073] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0074] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and are not to be construed as limiting the present invention. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting abnormal performance indicators of medical equipment based on PU learning, characterized in that: The following steps are involved: Using historical key performance indicator (KPI) streams as training data, clustering the training data according to similarity, obtaining a centroid curve for each cluster, and labeling the centroid curve for each cluster to obtain first abnormally labeled data and first unlabeled data; Based on the first abnormal labeled data and the first unlabeled data, a binary classifier is constructed through positive example unlabeled PU learning, and the abnormal label and normal label of the centroid curve of each cluster are obtained in combination with active learning, wherein the binary classifier is constructed through positive example unlabeled PU learning, the first normal labeled data is determined in the first unlabeled data through the binary classifier, the normal data and abnormal data are continued to be labeled in the unlabeled data through active learning, the unlabeled data is iteratively labeled multiple times, and samples that may be abnormal are selected from the unlabeled data through active learning for labeling, and the second normal labeled data and the second abnormal labeled data are determined, and the first abnormal labeled data and the second abnormal labeled data corresponding to each centroid curve are used as the abnormal label of the centroid curve, and the first normal labeled data and the second normal labeled data of each centroid curve are used as the normal label of the centroid curve; Obtain the labels on the centroid curve of the cluster corresponding to the KPI flow to be detected, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning, and detect the KPI flow to be detected through the anomaly detection model, wherein, for any newly obtained KPI flow that needs to be detected for anomaly, first determine the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster determined after classification, based on the abnormal labels and normal labels of the centroid curve of each cluster that have been determined, read the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected, and based on the labels on the centroid curve of the cluster corresponding to the obtained KPI flow and the data of the KPI flow itself, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning.

2. The method according to claim 1, characterized in that The step of labeling the centroid curve of each cluster to obtain first abnormal labeled data and first unlabeled data includes: An abnormal segment is manually marked on the centroid curve of each cluster, the labeled data in the abnormal segment is used as the first abnormal labeled data, and the data on the centroid curve except the abnormal segment is used as the first unlabeled data.

3. The method according to claim 1 or 2, characterized in that The obtaining of abnormal labels and normal labels of the centroid curve of each cluster includes: Determining first normal labeled data from the first unlabeled data by the binary classifier; Iteratively labeling the first unlabeled data multiple times using an active learning method to determine second normal labeled data and second abnormal labeled data from the first unlabeled data; The first abnormal labeled data and the second abnormal labeled data are used as the abnormal labels, and the first normal labeled data and the second normal labeled data are used as the normal labels.

4. The method according to claim 1, wherein The step of obtaining a label on a centroid curve of a cluster corresponding to a KPI flow to be detected includes: Determining the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster; Read the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected.

5. The method according to claim 4, characterized in that The training of the anomaly detection model corresponding to the KPI flow to be detected by semi-supervised learning includes: Selecting a preset amount of data from the KPI stream to be detected as second unlabeled data; An anomaly detection model corresponding to the KPI flow to be detected is trained based on the anomaly label, the normal label and the second unlabeled data on the centroid curve of the cluster corresponding to the KPI flow to be detected.

6. A device for detecting abnormal performance indicators of medical equipment based on PU learning, characterized in that: include: A clustering module is configured to cluster the training data according to the degree of similarity using the historical key performance indicator (KPI) stream as training data, obtain a centroid curve for each cluster, and label the centroid curve for each cluster to obtain first anomaly labeled data and first unlabeled data; An acquisition module is used to construct a binary classifier based on the first abnormal labeled data and the first unlabeled data through positive example unlabeled PU learning, and obtain the abnormal label and normal label of the centroid curve of each cluster in combination with active learning, wherein the binary classifier is constructed through positive example unlabeled PU learning, the first normal labeled data is determined in the first unlabeled data through the binary classifier, the normal data and abnormal data are continued to be labeled in the unlabeled data through active learning, the unlabeled data is iteratively labeled multiple times, and samples that may be abnormal are selected from the unlabeled data through active learning for labeling, and second normal labeled data and second abnormal labeled data are determined, and the first abnormal labeled data and the second abnormal labeled data corresponding to each centroid curve are used as the abnormal label of the centroid curve, and the first normal labeled data and the second normal labeled data of each centroid curve are used as the normal label of the centroid curve; A training module is used to obtain the labels on the centroid curve of the cluster corresponding to the KPI flow to be detected, train the anomaly detection model corresponding to the KPI flow to be detected through semi-supervised learning, and detect the KPI flow to be detected through the anomaly detection model. For any newly acquired KPI flow that needs to be detected for anomaly, the cluster corresponding to the KPI flow to be detected is determined by first calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster determined after classification. Based on the abnormal labels and normal labels of the centroid curve of each cluster that have been determined, the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected are read. Based on the labels on the centroid curve of the cluster corresponding to the KPI flow obtained, and the data of the KPI flow itself, the anomaly detection model corresponding to the KPI flow to be detected is trained through semi-supervised learning.

7. The detection device according to claim 6, characterized in that The clustering module is specifically used to: manually mark abnormal segments on the centroid curve of each cluster, use the labeled data in the abnormal segments as the first abnormal labeled data, and use the data on the centroid curve except the abnormal segments as the first unlabeled data.

8. The detection device according to claim 6 or 7, characterized in that: The acquisition module is specifically used to: Determining first normal labeled data from the unlabeled data by the binary classifier; Iteratively labeling the first unlabeled data multiple times using an active learning method to determine second normal labeled data and second abnormal labeled data from the first unlabeled data; The first abnormal labeled data and the second abnormal labeled data are used as the abnormal labels, and the first normal labeled data and the second normal labeled data are used as the normal labels.

9. The detection device according to claim 6, characterized in that: The training module is specifically used to: Determining the cluster corresponding to the KPI flow to be detected by calculating the shape similarity between the KPI flow to be detected and the centroid curve of each cluster; Read the abnormal labels and normal labels on the centroid curve of the cluster corresponding to the KPI flow to be detected.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting abnormalities in performance indicators of medical equipment based on PU learning is implemented.

Images