KVM seat system fault prediction method and device

Abstract

The invention discloses a KVM seat system fault prediction method and device. According to the method, the operation data of the KVM seat system are collected in real time and are respectively input into two specially-trained twin networks: one network is trained by using normal operation data, and the other network is trained by using fault data. The two networks output two sets of feature vectors respectively, and then the system state is evaluated by comparing the similarity of the two sets of feature vectors. If the similarity analysis shows that there is a significant difference, this indicates that the KVM seat system has a fault. According to the invention, the system abnormity can be identified in time, so that the problem that faults are difficult to predict quickly and accurately in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of KVM seat system fault prediction, and in particular to a KVM seat system fault prediction method and device. Background Art

[0002] In modern broadcast systems and large data center environments, KVM (Keyboard, Video, Mouse) agent systems play a vital role. As a core management and control platform, they are responsible for efficiently and stably connecting and operating multiple computers or servers. KVM agent systems enable operators to manage multiple devices in a single console environment, thereby improving work efficiency and reducing hardware redundancy.

[0003] However, with the development of technology and the growth of business needs, the scale and complexity of KVM seat systems are increasing, which brings new challenges. As a key component of the system, the stability of the KVM seat link is crucial to the performance of the entire system. Link failure not only causes service interruption, but also may cause data transmission delays or even data loss, which will have a negative impact on business continuity and end-user experience.

[0004] There are many reasons for KVM seat link failure, which may include hardware failure, software error, network problem, environmental factors, etc. In different application scenarios and system configurations, the characteristics of the failure may be different, which makes the prediction and diagnosis of the failure more complicated. Traditional manual troubleshooting methods are usually difficult to quickly and accurately identify and locate the fault, making it difficult for existing technologies to quickly and accurately predict the fault. Summary of the invention

[0005] The present invention provides a KVM seat system fault prediction method and device to solve the problem in the prior art that it is difficult to quickly and accurately predict faults.

[0006] In a first aspect, the present application provides a KVM seat system fault prediction method, comprising:

[0007] Get real-time operation data of KVM seat system;

[0008] Input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data, and the second twin network is a network trained with fault data;

[0009] Whether a fault occurs in the KVM agent system is determined according to the similarity between the first feature vector and the second feature vector.

[0010] By obtaining the real-time operating data of the KVM agent system, this application can ensure that the monitoring system can capture the latest status information, which is a prerequisite for accurate fault prediction. These data are input into the pre-trained twin network, and the extraction capability of the network can be used to generate two sets of feature vectors, representing the normal and abnormal states of the system. By comparing the similarity between the two sets of feature vectors, it is possible to effectively evaluate whether the current operating state of the system deviates from the normal range. If the similarity measurement result is greater than the preset threshold, this indicates that there is a significant difference between the actual operating state of the system and the normal state learned during training, so it can be inferred that the KVM agent system may have a fault. This fault prediction method based on real-time data and advanced algorithms can significantly improve the speed and accuracy of fault detection to solve the problem that it is difficult to quickly and accurately predict faults in the prior art.

[0011] As a preferred embodiment of the first aspect, judging whether the KVM agent system fails according to the similarity between the first feature vector and the second feature vector is specifically:

[0012] Calculate the Euclidean distance between the first eigenvector and the second eigenvector;

[0013] The Euclidean distance is expressed as

[0014] d(f j , n i )

[0015] In the formula, f j represents the first eigenvector, n i represents the second eigenvector;

[0016] Whether the KVM seat system fails is determined according to the size of the Euclidean distance.

[0017] The determining, based on the size of the Euclidean distance, whether the KVM seat system fails is specifically as follows:

[0018] If the Euclidean distance is greater than a preset threshold, it is determined that the KVM seat system fails.

[0019] In this preferred embodiment, the present application calculates the Euclidean distance between the first eigenvector and the second eigenvector output by the twin network, so that we can quantify and evaluate the difference between the current state and the normal state of the KVM agent system. When the Euclidean distance exceeds the preset threshold, it indicates that the deviation of the system state exceeds the normal fluctuation range, thereby accurately determining that the KVM agent system has a fault. The beneficial effect of this method is that it provides an efficient and accurate fault detection mechanism, which enables the system to respond quickly at the early stage of a fault, reduces the service interruption time caused by the fault, improves the reliability and operation and maintenance efficiency of the system, and ultimately ensures business continuity and user satisfaction.

[0020] As a preferred embodiment of the first aspect, if the Euclidean distance is greater than a preset threshold, after determining that the KVM agent system fails, the method further includes:

[0021] Get real-time operation data of KVM seat system;

[0022] According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category;

[0023] The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

[0024] In this preferred embodiment, after determining that the KVM seat system has failed, the present application further obtains real-time operating data and inputs the data into a pre-trained fully connected network based on preset judgment rules. This fully connected network is responsible for analyzing the data and outputting specific fault categories, including network failures, equipment failures, link failures, data transmission failures, and other failures. Such a processing flow not only improves the degree of automation of fault response, but also makes fault diagnosis more accurate and comprehensive, thereby providing strong support for quickly locating problems and taking corresponding repair measures, ensuring the stable operation of the system and the continuity of business.

[0025] As a preferred embodiment of the first aspect, if the Euclidean distance is greater than a preset threshold, after determining that the KVM agent system fails, the method further includes:

[0026] The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

[0027] In this preferred embodiment, the present application uses Euclidean distance, network topology information and multi-source data for comprehensive analysis to determine the precise location of the fault. The present application improves the accuracy and efficiency of fault location by combining real-time data analysis and network structure information. Specifically, the Euclidean distance provides a preliminary indication of the occurrence of a fault, while the network topology and multi-source data provide key information for further refining the fault location. Such an approach enables the operation and maintenance team to quickly identify and isolate the source of the fault, reduce the impact of the fault on the entire system, and accelerate the fault recovery process, thereby improving the reliability of the system and the continuity of business, ultimately optimizing the user experience and reducing operation and maintenance costs.

[0028] In a second aspect, the present application provides a KVM seat system fault prediction device. The KVM seat system fault prediction device includes an acquisition module, an input and output module, and a judgment module;

[0029] The acquisition module is used to obtain the real-time operation data of the KVM seat system;

[0030] The input-output module is used to input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data, and the second twin network is a network trained with fault data;

[0031] The judgment module 30 is used to judge whether the KVM seat system fails according to the similarity between the first feature vector and the second feature vector.

[0032] This device uses three modules to divide the work and coordinate work to better accurately predict faults. By obtaining the real-time operation data of the KVM agent system, this application can ensure that the monitoring system can capture the latest status information, which is a prerequisite for accurate fault prediction. These data are input into the pre-trained twin network, and the extraction ability of the network can be used to generate two sets of feature vectors, representing the normal and abnormal states of the system. By comparing the similarity between the two sets of feature vectors, it is possible to effectively evaluate whether the current operating state of the system deviates from the normal range. If the similarity measurement result is greater than the preset threshold, this indicates that there is a significant difference between the actual operating state of the system and the normal state learned during training, so it can be inferred that the KVM agent system may have a fault. This fault prediction method based on real-time data and advanced algorithms can significantly improve the speed and accuracy of fault detection to solve the problem that it is difficult to quickly and accurately predict faults in the prior art.

[0033] As a preferred embodiment of the second aspect, judging whether the KVM agent system fails according to the similarity between the first feature vector and the second feature vector is specifically:

[0034] Calculate the Euclidean distance between the first eigenvector and the second eigenvector;

[0035] The Euclidean distance is expressed as

[0036] d(f j , n i )

[0037] In the formula, f j represents the first eigenvector, ni represents the second eigenvector;

[0038] Whether the KVM seat system fails is determined according to the size of the Euclidean distance.

[0039] The determining, based on the size of the Euclidean distance, whether the KVM seat system fails is specifically as follows:

[0040] If the Euclidean distance is greater than a preset threshold, it is determined that the KVM seat system fails.

[0041] In this preferred embodiment, the present application calculates the Euclidean distance between the first eigenvector and the second eigenvector output by the twin network, so that we can quantify and evaluate the difference between the current state and the normal state of the KVM agent system. When the Euclidean distance exceeds the preset threshold, it indicates that the deviation of the system state exceeds the normal fluctuation range, thereby accurately determining that the KVM agent system has a fault. The beneficial effect of this method is that it provides an efficient and accurate fault detection mechanism, which enables the system to respond quickly at the early stage of a fault, reduces the service interruption time caused by the fault, improves the reliability and operation and maintenance efficiency of the system, and ultimately ensures business continuity and user satisfaction.

[0042] As a preferred embodiment of the second aspect, if the Euclidean distance is greater than a preset threshold, after determining that the KVM agent system fails, the method further includes:

[0043] Get real-time operation data of KVM seat system;

[0044] According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category;

[0045] The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

[0046] In this preferred embodiment, after determining that the KVM seat system has failed, the present application further obtains real-time operating data and inputs the data into a pre-trained fully connected network based on preset judgment rules. This fully connected network is responsible for analyzing the data and outputting specific fault categories, including network failures, equipment failures, link failures, data transmission failures, and other failures. Such a processing flow not only improves the degree of automation of fault response, but also makes fault diagnosis more accurate and comprehensive, thereby providing strong support for quickly locating problems and taking corresponding repair measures, ensuring the stable operation of the system and the continuity of business.

[0047] As a preferred embodiment of the second aspect, if the Euclidean distance is greater than a preset threshold, after determining that the KVM agent system fails, the method further includes:

[0048] The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

[0049] In this preferred embodiment, the present application uses Euclidean distance, network topology information and multi-source data for comprehensive analysis to determine the precise location of the fault. The present application improves the accuracy and efficiency of fault location by combining real-time data analysis and network structure information. Specifically, the Euclidean distance provides a preliminary indication of the occurrence of a fault, while the network topology and multi-source data provide key information for further refining the fault location. Such an approach enables the operation and maintenance team to quickly identify and isolate the source of the fault, reduce the impact of the fault on the entire system, and accelerate the fault recovery process, thereby improving the reliability of the system and the continuity of business, ultimately optimizing the user experience and reducing operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 : A flow chart of an embodiment of a KVM seat system fault prediction method provided by the present application;

[0051] Figure 2 : A structural diagram of an embodiment of a KVM seat system fault prediction device provided in this application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Embodiment 1

[0054] Please refer to Figure 1, which is a KVM seat system fault prediction method provided by an embodiment of the present invention.

[0055] In this embodiment, the process of the KVM seat system fault prediction method in this application is described in detail through steps S01-S03.

[0056] S01: Obtain real-time operation data of the KVM seat system.

[0057] S02: Input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data and the second twin network is a network trained with fault data.

[0058] As a preferred embodiment of the first embodiment, the training data acquisition and preprocessing steps of the first twin network and the second twin network are specifically as follows:

[0059] The operation data of the servers, network devices, sensors, etc. of the KVM agent system are collected in real time through the Internet of Things (IoT) technology and network monitoring tools, including link status data (such as bandwidth utilization, packet loss rate, delay), device performance data (such as CPU usage, memory occupancy), network traffic data, environmental parameters, etc. At the same time, historical fault records, system logs, configuration information, etc. are collected as training data and verification data for subsequent training of the twin network model. The collected raw data is cleaned, noise and outliers are removed, and normalization is performed to ensure data quality. The preprocessed data is converted into a format suitable for TCN model input, usually a two-dimensional matrix (number of samples, time step, number of features). This step allows the TCN model to effectively process time series data and extract useful features from it for prediction and classification. Multi-source heterogeneous data fusion technology is used to integrate data from different sources, and then the data divided into normal state and fault state are further placed in the twin network model.

[0060] Number of samples: Each sample represents a record of system operation status within a period of time. The time can be customized, such as collecting data within a day, an hour, or a minute.

[0061] Time step: refers to the number of time points contained in each sample. Each time point corresponds to a set of feature values. The choice of time step depends on the characteristics of the data and the requirements of the prediction task. If we want to capture the data changes within each hour, then the time step may be 60 (if the data is recorded once a minute).

[0062] Feature number: refers to the number of variables observed at each time point. Features may include multiple indicators such as link delay, packet loss rate, CPU usage, etc.

[0063] for example:

[0064] Sample 1 (data within a certain period of time):

[0065] 【10,0,1】#Sample 1, time step 1, feature 1;

[0066] The first sample has a link delay of 10 and a packet loss rate of 0.1% at the first time step;

[0067] 【12,0,2】#Sample 1, time step 2, feature 2;

[0068] The first sample has a link delay of 12 and a packet loss rate of 0.2% at the second time step;

[0069] Sample 2 (data from another time period).

[0070] As a preferred embodiment of the first embodiment, the construction process of the first twin network and the second twin network is specifically as follows:

[0071] The twin network consists of two sub-networks with the same structure and shared weights, and performs task judgment by comparing the similarity of the two inputs. In the KVM seat link fault identification, the data of the normal state and the fault state are input into the two sub-networks respectively, and the twin network is used to judge the difference between them to identify the fault.

[0072] It is necessary to prepare two identical sub-networks of the twin network, which extract features from the two samples in the input data pair respectively. At the same time, the two sub-networks need to share the same weight and bias parameters to ensure that they have the same processing capabilities for the data. The feature vectors extracted by the two sub-networks are compared for similarity on the fully connected layer to determine whether the link status is consistent or there is a fault, and the Euclidean distance is used to calculate the distance between the two feature vectors.

[0073] Euclidean distance: d(υ1,υ2)=||υ1-υ2||2

[0074] The two sub-networks use a temporal convolutional neural network (TCN), which can effectively capture long-term dependencies in the data. The causal convolution layer in the TCN ensures that each time step of the model output depends only on past inputs, which helps maintain the causality and stability of the model. After a series of convolutional layers, pooling layers, and activation functions (Relu) for feature extraction, residual connections are used to help the model learn long-term dependencies and alleviate the gradient vanishing problem in deep networks. The output of the TCN is connected to a fully connected layer to convert the extracted features into the final classification results. The number of output nodes of the fully connected layer should match the number of fault types (if there are five fault types, the number of output nodes is 5).

[0075] At the same time, contrast loss is used as the loss function, which aims to minimize the feature distance between samples of the same category (normal state or fault state) and maximize the feature distance between samples of different categories (normal state and fault state).

[0076] Loss function:

[0077]

[0078] Use the label y (1 for similarity, -1 for dissimilarity), where m is the boundary interval and the similarity measure d, and v1 and v2 represent the feature vectors extracted from the two sub-networks in the twin network.

[0079] As a preferred embodiment of the first embodiment, the training process of the first twin network and the second twin network is specifically as follows:

[0080] The collected data is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the model performance. The twin network model is trained using the training set data, and the model parameters are optimized through the back propagation algorithm. During the training process, data enhancement, regularization and other techniques can be used to improve the generalization ability of the model. The trained model is evaluated using the test set data, and indicators such as recognition accuracy, recall rate, and F1 score are calculated to evaluate the model performance. The stability and reliability of the model are further verified through methods such as cross-validation.

[0081] S03: Determine whether the KVM agent system fails according to the similarity between the first feature vector and the second feature vector.

[0082] As a preferred embodiment of the first embodiment, judging whether the KVM agent system fails according to the similarity between the first feature vector and the second feature vector is specifically:

[0083] Deploy the trained twin network model to the KVM agent system to monitor the link status data in real time. Input the real-time monitored link status data into the twin network model to calculate the similarity between feature vectors.

[0084] There are n different positions in the normal state, and the feature vector of each position is represented by n i , where i=1, 2,…, n.

[0085] There are n different positions under the fault state, and the feature vector of each position is represented by f j , where j = 1, 2, ..., n.

[0086] Calculate the distance between the feature vector of each position in the fault state and the feature vector of each position in the normal state. d(f j , n i )

[0087] For each position j in the fault state, traverse all positions i in the normal state and compare the distance d(f j , n i ) and the size of the threshold T.

[0088] If d(f j , n i )>T, it is considered that there may be a fault at position j.

[0089] Whether the link is faulty is determined based on the similarity result. If the distance is less than a certain threshold T, the link is considered normal; otherwise, the link is considered faulty.

[0090] In this preferred embodiment, the present application calculates the Euclidean distance between the first eigenvector and the second eigenvector output by the twin network, so that we can quantify and evaluate the difference between the current state and the normal state of the KVM agent system. When the Euclidean distance exceeds the preset threshold, it indicates that the deviation of the system state exceeds the normal fluctuation range, thereby accurately determining that the KVM agent system has a fault. The beneficial effect of this method is that it provides an efficient and accurate fault detection mechanism, which enables the system to respond quickly at the early stage of a fault, reduces the service interruption time caused by the fault, improves the reliability and operation and maintenance efficiency of the system, and ultimately ensures business continuity and user satisfaction.

[0091] As a preferred embodiment of the first embodiment, if the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes:

[0092] Get real-time operation data of KVM seat system;

[0093] According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category;

[0094] The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

[0095] Network fault diagnosis:

[0096] If the link status data feature vector calculated by the twin network is significantly different from that in the normal state, especially if indicators such as packet loss rate and delay deviate significantly from the normal range, and this difference persists in multiple data inputs, it is judged as a network connection failure. 1. Monitor the packet loss rate: If the packet loss rate exceeds the set threshold, such as multiple consecutive data packets are lost or the packet loss rate continues to be higher than a certain proportion for a period of time, it is judged as a network failure. 2. Check the delay: When the delay time exceeds the normal range and lasts for a long time, it indicates that there is a network delay failure. 3. Analyze network traffic: If the network traffic fluctuates abnormally, such as a sudden surge or drop, and is inconsistent with the normal business pattern, it is a sign of network failure.

[0097] Equipment failure judgment:

[0098] When the characteristic vector of device performance data (such as CPU usage, memory occupancy) has a low similarity with the normal state and exceeds the preset threshold range, combined with changes in other related indicators (such as device temperature, fan speed, etc.), it can be judged as a device performance failure. 1. Pay attention to CPU usage: If the CPU usage is high for a long time, close to or exceeds the warning threshold, and there is a significant difference from the load conditions during normal operation, it is judged as a device failure. 2. Check memory usage: If the memory usage rate increases abnormally, causing the system to run slowly or freeze, it indicates a problem with the device memory. 3. Check the device temperature: Some sensors can monitor the device temperature. If the temperature is too high and exceeds the normal operating range, it means that the device is at risk of failure.

[0099] Link fault judgment:

[0100] Comprehensively consider bandwidth utilization, packet loss rate, and latency: If the bandwidth utilization is too low or too high, accompanied by an increase in packet loss rate and latency, it is judged as a link failure. At the same time, check the link connection status: Monitor the physical connection status of the link, such as loose plugs, damaged cables, etc., which can also cause link failures. If the bandwidth utilization in the link status data fluctuates greatly, or short-term link interruption signals appear frequently, the twin network model can capture these abnormal features and judge it as a link stability failure.

[0101] Data transmission fault judgment:

[0102] By monitoring the characteristic vectors of network traffic data, if the data transmission rate is found to be abnormally reduced, or data loss occurs, and the characteristic vectors are significantly different from the normal state in the twin network, it is judged as a data transmission failure.

[0103] Other fault diagnosis:

[0104] Analyze system logs: Check whether there are any abnormal records in the system logs, such as error messages, warning messages, etc., which may indicate other types of failures. Consider environmental factors: Abnormal changes in environmental parameters such as temperature and humidity may also affect system operation and cause failures.

[0105] In this preferred embodiment, after determining that the KVM seat system has failed, the present application further obtains real-time operating data and inputs the data into a pre-trained fully connected network based on preset judgment rules. This fully connected network is responsible for analyzing the data and outputting specific fault categories, including network failures, equipment failures, link failures, data transmission failures, and other failures. Such a processing flow not only improves the degree of automation of fault response, but also makes fault diagnosis more accurate and comprehensive, thereby providing strong support for quickly locating problems and taking corresponding repair measures, ensuring the stable operation of the system and the continuity of business.

[0106] As a preferred embodiment of the first embodiment, if the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes:

[0107] The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

[0108] In this preferred embodiment, the present application uses Euclidean distance, network topology information and multi-source data for comprehensive analysis to determine the precise location of the fault. The present application improves the accuracy and efficiency of fault location by combining real-time data analysis and network structure information. Specifically, the Euclidean distance provides a preliminary indication of the occurrence of a fault, while the network topology and multi-source data provide key information for further refining the fault location. Such an approach enables the operation and maintenance team to quickly identify and isolate the source of the fault, reduce the impact of the fault on the entire system, and accelerate the fault recovery process, thereby improving the reliability of the system and the continuity of business, ultimately optimizing the user experience and reducing operation and maintenance costs.

[0109] By obtaining the real-time operating data of the KVM agent system, this application can ensure that the monitoring system can capture the latest status information, which is a prerequisite for accurate fault prediction. These data are input into the pre-trained twin network, and the extraction capability of the network can be used to generate two sets of feature vectors, representing the normal and abnormal states of the system. By comparing the similarity between the two sets of feature vectors, it is possible to effectively evaluate whether the current operating state of the system deviates from the normal range. If the similarity measurement result is greater than the preset threshold, this indicates that there is a significant difference between the actual operating state of the system and the normal state learned during training, so it can be inferred that the KVM agent system may have a fault. This fault prediction method based on real-time data and advanced algorithms can significantly improve the speed and accuracy of fault detection to solve the problem that it is difficult to quickly and accurately predict faults in the prior art.

[0110] Embodiment 2

[0111] Please refer to Figure 2 , which is a KVM seat system fault prediction device provided in an embodiment of the present application.

[0112] In this embodiment, the KVM seat system fault prediction device includes an acquisition module 10 , an input and output module 20 and a determination module 30 .

[0113] The acquisition module 10 is used to acquire the real-time operation data of the KVM seat system.

[0114] The input-output module 20 is used to input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data and the second twin network is a network trained with fault data.

[0115] As a preferred embodiment of the second embodiment, the training data acquisition and preprocessing steps of the first twin network and the second twin network are specifically as follows:

[0116] The operation data of the servers, network devices, sensors, etc. of the KVM agent system are collected in real time through the Internet of Things (IoT) technology and network monitoring tools, including link status data (such as bandwidth utilization, packet loss rate, delay), device performance data (such as CPU usage, memory occupancy), network traffic data, environmental parameters, etc. At the same time, historical fault records, system logs, configuration information, etc. are collected as training data and verification data for subsequent training of the twin network model. The collected raw data is cleaned, noise and outliers are removed, and normalization is performed to ensure data quality. The preprocessed data is converted into a format suitable for TCN model input, usually a two-dimensional matrix (number of samples, time step, number of features). This step allows the TCN model to effectively process time series data and extract useful features from it for prediction and classification. Multi-source heterogeneous data fusion technology is used to integrate data from different sources, and then the data divided into normal state and fault state are further placed in the twin network model.

[0117] Number of samples: Each sample represents a record of system operation status within a period of time. The time can be customized, such as collecting data within a day, an hour, or a minute.

[0118] Time step: refers to the number of time points contained in each sample. Each time point corresponds to a set of feature values. The choice of time step depends on the characteristics of the data and the requirements of the prediction task. If we want to capture the data changes within each hour, then the time step may be 60 (if the data is recorded once a minute).

[0119] Feature number: refers to the number of variables observed at each time point. Features may include multiple indicators such as link delay, packet loss rate, CPU usage, etc.

[0120] for example:

[0121] Sample 1 (data within a certain period of time):

[0122] 【10,0,1】#Sample 1, time step 1, feature 1;

[0123] The first sample has a link delay of 10 and a packet loss rate of 0.1% at the first time step;

[0124] 【12,0,2】#Sample 1, time step 2, feature 2;

[0125] The first sample has a link delay of 12 and a packet loss rate of 0.2% at the second time step;

[0126] Sample 2 (data from another time period).

[0127] As a preferred embodiment of the second embodiment, the construction process of the first twin network and the second twin network is specifically as follows:

[0128] The twin network consists of two sub-networks with the same structure and shared weights, and performs task judgment by comparing the similarity of the two inputs. In the KVM seat link fault identification, the data of the normal state and the fault state are input into the two sub-networks respectively, and the twin network is used to judge the difference between them to identify the fault.

[0129] It is necessary to prepare two identical sub-networks of the twin network, which extract features from the two samples in the input data pair respectively. At the same time, the two sub-networks need to share the same weight and bias parameters to ensure that they have the same processing capabilities for the data. The feature vectors extracted by the two sub-networks are compared for similarity on the fully connected layer to determine whether the link status is consistent or there is a fault, and the Euclidean distance is used to calculate the distance between the two feature vectors.

[0130] Euclidean distance: d(υ1,υ2)=||υ1-υ2||2

[0131] The two sub-networks use a temporal convolutional neural network (TCN), which can effectively capture long-term dependencies in the data. The causal convolution layer in the TCN ensures that each time step of the model output depends only on past inputs, which helps maintain the causality and stability of the model. After a series of convolutional layers, pooling layers, and activation functions (Relu) for feature extraction, residual connections are used to help the model learn long-term dependencies and alleviate the gradient vanishing problem in deep networks. The output of the TCN is connected to a fully connected layer to convert the extracted features into the final classification results. The number of output nodes of the fully connected layer should match the number of fault types (if there are five fault types, the number of output nodes is 5).

[0132] At the same time, contrast loss is used as the loss function, which aims to minimize the feature distance between samples of the same category (normal state or fault state) and maximize the feature distance between samples of different categories (normal state and fault state).

[0133] Loss function:

[0134]

[0135] Use the label y (1 for similarity, -1 for dissimilarity), where m is the boundary interval and the similarity measure d, and v1 and v2 represent the feature vectors extracted from the two sub-networks in the twin network.

[0136] As a preferred embodiment of the second embodiment, the training process of the first twin network and the second twin network is specifically as follows:

[0137] The collected data is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the model performance. The twin network model is trained using the training set data, and the model parameters are optimized through the back propagation algorithm. During the training process, data enhancement, regularization and other techniques can be used to improve the generalization ability of the model. The trained model is evaluated using the test set data, and indicators such as recognition accuracy, recall rate, and F1 score are calculated to evaluate the model performance. The stability and reliability of the model are further verified through methods such as cross-validation.

[0138] The judgment module 30 is used to judge whether the KVM seat system fails according to the similarity between the first feature vector and the second feature vector.

[0139] As a preferred embodiment of the second embodiment, judging whether the KVM seat system fails according to the similarity between the first feature vector and the second feature vector is specifically:

[0140] Deploy the trained twin network model to the KVM agent system to monitor the link status data in real time. Input the real-time monitored link status data into the twin network model to calculate the similarity between feature vectors.

[0141] There are n different positions in the normal state, and the feature vector of each position is represented by n i , where i=1, 2,…, n.

[0142] There are n different positions under the fault state, and the feature vector of each position is represented by f j , where j = 1, 2, ..., n.

[0143] Calculate the distance between the feature vector of each position in the fault state and the feature vector of each position in the normal state. d(f j , n i )

[0144] For each position j in the fault state, traverse all positions i in the normal state and compare the distance d(f j , n i ) and the size of the threshold T.

[0145] If d(f j , n i )>T, it is considered that there may be a fault at position j.

[0146] Whether the link is faulty is determined based on the similarity result. If the distance is less than a certain threshold T, the link is considered normal; otherwise, the link is considered faulty.

[0147] In this preferred embodiment, the present application calculates the Euclidean distance between the first eigenvector and the second eigenvector output by the twin network, so that we can quantify and evaluate the difference between the current state and the normal state of the KVM agent system. When the Euclidean distance exceeds the preset threshold, it indicates that the deviation of the system state exceeds the normal fluctuation range, thereby accurately determining that the KVM agent system has a fault. The beneficial effect of this method is that it provides an efficient and accurate fault detection mechanism, which enables the system to respond quickly at the early stage of a fault, reduces the service interruption time caused by the fault, improves the reliability and operation and maintenance efficiency of the system, and ultimately ensures business continuity and user satisfaction.

[0148] As a preferred embodiment of the second embodiment, if the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes:

[0149] Get real-time operation data of KVM seat system;

[0150] According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category;

[0151] The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

[0152] Network fault diagnosis:

[0153] If the link status data feature vector calculated by the twin network is significantly different from that in the normal state, especially if indicators such as packet loss rate and delay deviate significantly from the normal range, and this difference persists in multiple data inputs, it is judged as a network connection failure. 1. Monitor the packet loss rate: If the packet loss rate exceeds the set threshold, such as multiple consecutive data packets are lost or the packet loss rate continues to be higher than a certain proportion for a period of time, it is judged as a network failure. 2. Check the delay: When the delay time exceeds the normal range and lasts for a long time, it indicates that there is a network delay failure. 3. Analyze network traffic: If the network traffic fluctuates abnormally, such as a sudden surge or drop, and is inconsistent with the normal business pattern, it is a sign of network failure.

[0154] Equipment failure judgment:

[0155] When the characteristic vector of device performance data (such as CPU usage, memory occupancy) has a low similarity with the normal state and exceeds the preset threshold range, combined with changes in other related indicators (such as device temperature, fan speed, etc.), it can be judged as a device performance failure. 1. Pay attention to CPU usage: If the CPU usage is high for a long time, close to or exceeds the warning threshold, and there is a significant difference from the load conditions during normal operation, it is judged as a device failure. 2. Check memory usage: If the memory usage rate increases abnormally, causing the system to run slowly or freeze, it indicates a problem with the device memory. 3. Check the device temperature: Some sensors can monitor the device temperature. If the temperature is too high and exceeds the normal operating range, it means that the device is at risk of failure.

[0156] Link fault judgment:

[0157] Comprehensively consider bandwidth utilization, packet loss rate, and latency: If the bandwidth utilization is too low or too high, accompanied by an increase in packet loss rate and latency, it is judged as a link failure. At the same time, check the link connection status: Monitor the physical connection status of the link, such as loose plugs, damaged cables, etc., which can also cause link failures. If the bandwidth utilization in the link status data fluctuates greatly, or short-term link interruption signals appear frequently, the twin network model can capture these abnormal features and judge it as a link stability failure.

[0158] Data transmission fault judgment:

[0159] By monitoring the characteristic vectors of network traffic data, if the data transmission rate is found to be abnormally reduced, or data loss occurs, and the characteristic vectors are significantly different from the normal state in the twin network, it is judged as a data transmission failure.

[0160] Other fault diagnosis:

[0161] Analyze system logs: Check whether there are any abnormal records in the system logs, such as error messages, warning messages, etc., which may indicate other types of failures. Consider environmental factors: Abnormal changes in environmental parameters such as temperature and humidity may also affect system operation and cause failures.

[0162] In this preferred embodiment, after determining that the KVM seat system has failed, the present application further obtains real-time operating data and inputs the data into a pre-trained fully connected network based on preset judgment rules. This fully connected network is responsible for analyzing the data and outputting specific fault categories, including network failures, equipment failures, link failures, data transmission failures, and other failures. Such a processing flow not only improves the degree of automation of fault response, but also makes fault diagnosis more accurate and comprehensive, thereby providing strong support for quickly locating problems and taking corresponding repair measures, ensuring the stable operation of the system and the continuity of business.

[0163] As a preferred embodiment of the second embodiment, if the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes:

[0164] The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

[0165] In this preferred embodiment, the present application uses Euclidean distance, network topology information and multi-source data for comprehensive analysis to determine the precise location of the fault. The present application improves the accuracy and efficiency of fault location by combining real-time data analysis and network structure information. Specifically, the Euclidean distance provides a preliminary indication of the occurrence of a fault, while the network topology and multi-source data provide key information for further refining the fault location. Such an approach enables the operation and maintenance team to quickly identify and isolate the source of the fault, reduce the impact of the fault on the entire system, and accelerate the fault recovery process, thereby improving the reliability of the system and the continuity of business, ultimately optimizing the user experience and reducing operation and maintenance costs.

[0166] This device uses three modules to divide the work and coordinate work to better accurately predict faults. By obtaining the real-time operation data of the KVM agent system, this application can ensure that the monitoring system can capture the latest status information, which is a prerequisite for accurate fault prediction. These data are input into the pre-trained twin network, and the extraction ability of the network can be used to generate two sets of feature vectors, representing the normal and abnormal states of the system. By comparing the similarity between the two sets of feature vectors, it is possible to effectively evaluate whether the current operating state of the system deviates from the normal range. If the similarity measurement result is greater than the preset threshold, this indicates that there is a significant difference between the actual operating state of the system and the normal state learned during training, so it can be inferred that the KVM agent system may have a fault. This fault prediction method based on real-time data and advanced algorithms can significantly improve the speed and accuracy of fault detection to solve the problem that it is difficult to quickly and accurately predict faults in the prior art.

[0167] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A KVM seat system fault prediction method, characterized in that: include: Get real-time operation data of KVM seat system; Input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data, and the second twin network is a network trained with fault data; Whether a fault occurs in the KVM agent system is determined according to the similarity between the first feature vector and the second feature vector.

2. The KVM seat system fault prediction method according to claim 1, characterized in that: The determining, based on the similarity between the first feature vector and the second feature vector, whether the KVM agent system fails is specifically: Calculate the Euclidean distance between the first eigenvector and the second eigenvector; The Euclidean distance is expressed as d(f j ,n i ) In the formula, f j represents the first eigenvector, ni represents the second eigenvector; Whether the KVM seat system fails is determined according to the size of the Euclidean distance.

3. The KVM seat system fault prediction method according to claim 2, characterized in that: The determining, based on the size of the Euclidean distance, whether the KVM seat system fails is specifically as follows: If the Euclidean distance is greater than a preset threshold, it is determined that the KVM seat system fails.

4. The KVM seat system fault prediction method according to claim 3, characterized in that: If the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes: Get real-time operation data of KVM seat system; According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category; The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

5. The KVM seat system fault prediction method according to claim 3, characterized in that: If the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes: The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

6. A KVM seat system fault prediction device, characterized in that: It includes an acquisition module, an input and output module, and a judgment module; The acquisition module is used to obtain the real-time operation data of the KVM seat system; The input-output module is used to input the real-time operation data into the pre-trained first twin network and the second twin network, so that the first twin network outputs a first feature vector and the second twin network outputs a second feature vector; wherein the first twin network is a network trained with normal data, and the second twin network is a network trained with fault data; The judgment module is used to judge whether the KVM seat system fails according to the similarity between the first feature vector and the second feature vector.

7. The KVM seat system fault prediction device according to claim 6, characterized in that: The determining, based on the similarity between the first feature vector and the second feature vector, whether the KVM agent system fails is specifically: Calculate the Euclidean distance between the first eigenvector and the second eigenvector; The Euclidean distance is expressed as d(f j ,n i ) In the formula, f j represents the first eigenvector, n i represents the second eigenvector; Whether the KVM seat system fails is determined according to the size of the Euclidean distance.

8. The KVM seat system fault prediction device according to claim 7, characterized in that: The determining, based on the size of the Euclidean distance, whether the KVM seat system fails is specifically as follows: If the Euclidean distance is greater than a preset threshold, it is determined that the KVM seat system fails.

9. The KVM seat system fault prediction device according to claim 8, characterized in that: If the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes: Get real-time operation data of KVM seat system; According to a preset judgment rule, the real-time operation data of the KVM agent system is input into a pre-trained fully connected network, so that the fully connected network outputs a fault category; The fault categories include network faults, equipment faults, link faults, data transmission faults and other faults.

10. The KVM seat system fault prediction device according to claim 8, characterized in that: If the Euclidean distance is greater than a preset threshold, after determining that the KVM seat system fails, the method further includes: The location where the fault occurs is determined based on the Euclidean distance, network topology and multi-source data.

Images