System fault automatic diagnosis and disposal method based on large model

Abstract

The invention discloses a system fault automatic diagnosis and disposal method based on a large model, and relates to the technical field of system fault diagnosis. Comprising the following steps: acquiring related data when a system fault occurs and establishing a database, encoding fault data by using a Grubrum angle field, dividing a sample data set into a training set and a test set, establishing a DRN fault diagnosis model, and performing automatic diagnosis and disposal of the system fault by using the optimal DRN fault diagnosis model. According to the system fault automatic diagnosis and disposal method based on the large model, fault occurrence reasons are accurately diagnosed according to the constructed database, a related fault reason set is created, fault diagnosis reasons are finally output, and fault diagnosis is performed by using the DRN model, so that the dependence on expert experience is reduced, and the fault diagnosis efficiency is improved. The automatic and intelligent fault diagnosis process is realized, and the operability and sustainability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of system fault diagnosis, and particularly to an automatic system fault diagnosis and disposal method based on a large model. Background Art

[0002] With the development of computer technology, people's demand for and dependence on intelligent terminals are increasing. Intelligent terminals such as mobile phones and tablets have gradually become an essential part of people's work and life. Once a fault occurs during use, it will directly affect people's normal work and life. However, since intelligent terminals are high-precision machines, system fault diagnosis and repair require relatively high professional knowledge. At present, there is no method in the industry that allows intelligent terminals with system faults to automatically perform system diagnosis and repair. When an intelligent terminal has a system fault, ordinary users can only hand the machine over to the agent to return it to the factory for repair, which has a long repair cycle and requires a certain repair cost.

[0003] The invention patent with the publication number CN105653389A discloses a data diagnosis and repair method and device. The method includes: when detecting that a data diagnosis and repair task is triggered, restarting the intelligent terminal and entering the data diagnosis mode; in the data diagnosis mode, creating an empty directory in the data partition; mounting the empty directory to the data partition so that the system starts by accessing the empty directory. This method realizes the self-automatic diagnosis and repair of the intelligent terminal system fault, can quickly solve the system fault problem in the use of the intelligent terminal, and saves the repair cost. However, this method cannot obtain the feedback status when the program gets stuck or an error occurs, so it is impossible to accurately judge the status of the program. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides an automatic system fault diagnosis and disposal method based on a large model, which solves the existing problems.

[0005] To achieve the above object, the present invention is realized through the following technical solutions: An automatic system fault diagnosis and disposal method based on a large model, including the following steps:

[0006] Step 1: Obtain the relevant data when a system fault occurs, and establish a database in combination with the relevant data for system fault diagnosis;

[0007] Step 2: Encode the fault data using Gramian angular field;

[0008] Step 3: Label the fault data according to the system fault type and its fault characteristics, and divide the obtained sample data set into a training set and a test set;

[0009] Step 4: Establish a DRN fault diagnosis model, train the DRN fault diagnosis model using the training data, and complete the verification of the DRN fault diagnosis model using the test data to obtain an optimal DRN fault diagnosis model;

[0010] Step 5: Use the optimal DRN fault diagnosis model to perform automatic system fault diagnosis and disposal.

[0011] Preferably, the relevant parameters include the status of the communication protocol connection between the monitored interface program and the acquisition service, the memory usage status of the monitored interface program, and the system events of the operating system.

[0012] Preferably, the three acquisition methods for the relevant parameters used in parallel in Step 1 include:

[0013] Adopt a timed polling method to collect the connection status of the communication protocol between the monitored interface program and the acquisition service, and obtain information on whether the communication of the interface program is disconnected;

[0014] Collect the memory usage status of the monitored interface program in real time, obtain the current memory usage in units of a set time, compare the memory usage at each set time with the memory usage at the previous set time to obtain information on whether the interface program is stuck. If the memory usage does not change within the set time, it proves that the interface program is in a stuck state;

[0015] Adopt a timed polling method to collect the system events of the operating system to obtain information on whether there are error events in the interface program.

[0016] Preferably, the relevant parameters further include the status of the communication protocol connection between the monitored interface program and the acquisition service, the CPU utilization rate of the monitored interface program, and the system events of the operating system.

[0017] Preferably, the three acquisition methods for the relevant parameters used in parallel in Step 1 include:

[0018] Adopt a timed polling method to collect the connection status of the communication protocol between the monitored interface program and the acquisition service, and obtain information on whether the communication of the interface program is disconnected;

[0019] Collect the CPU utilization rate of the monitored interface program in real time, obtain the current CPU utilization rate in units of a set time, compare the CPU utilization rate at each set time with the CPU utilization rate at the previous set time to obtain information on whether the interface program is stuck;

[0020] Adopt a timed polling method to collect the system events of the operating system to obtain information on whether there are error events in the interface program.

[0021] Preferably, the specific steps of step two are as follows:

[0022] S21: Obtain one-dimensional time series data from the fault data;

[0023] S22: Perform normalization processing on the time series data to scale the amplitude range of the data between [0, 1];

[0024] S23: Convert the normalized time series data into a Gram matrix, where each element in the Gram matrix represents the inner product or correlation between two data points;

[0025] S24: Obtain eigenvectors and eigenvalues by performing eigenvalue decomposition on the Gram matrix;

[0026] S25: Convert the corresponding time and amplitude of each point in the time series data into the radius and angle of polar coordinates to form a Gram angle field;

[0027] S26: Input the generated Gram angle field as a two-dimensional image into a deep residual network for fault diagnosis.

[0028] Preferably, the establishment process of the DRN fault diagnosis model includes:

[0029] Construct multiple residual blocks; each residual block contains one or more convolutional layers, batch normalization layers, and activation function layers;

[0030] The output of the residual block is divided into a first part and a second part through a skip connection structure; the first part is the non-linear output F(x) after multiple convolutions; the second part is the identity mapping x that directly transmits the output to the output of the neural network through the skip connection structure as a part of the output result;

[0031] The residual block is expressed as:

[0032] y1 = h ( x1 ) + Γ ( x1, w1 ) ;

[0033] x 1+1 = f(y1);

[0034] Let h(x1) = x1, x 1+1 = y1;

[0035]

[0036] Wherein, y1 is the output of the l-th residual unit, x1 is the input of the l-th residual unit, and each residual structure is activated by a function; Γ is the residual function, h ( x 1) is a constant mapping layer, w1 is the weight value of the l-th layer residual unit, w1 = {w 1,k|1≤k≤K}, and K is the number of layers of the residual unit;

[0037] The training of the DRN fault diagnosis model using the training data is specifically as follows:

[0038] Initialize the parameters of the deep residual network;

[0039] Input the training data into the deep residual network, and calculate the output of the model through forward propagation;

[0040] Compare the output of the model with the true label, and calculate the value of the loss function ε:

[0041]

[0042] Use the chain rule to calculate the gradient of the model parameters, and transfer the gradient of the loss function from the output layer back to the input layer;

[0043] Update the model parameters using the gradient descent method according to the calculated gradient and learning rate.

[0044] Preferably, the verification of the DRN fault diagnosis model using the test data is specifically as follows:

[0045] Input the test data into the model, and calculate the output result of the model through forward propagation.

[0046] The present invention provides a method for automatic diagnosis and disposal of system faults based on a large model. Compared with the prior art, it has the following beneficial effects:

[0047] 1. The method for automatic diagnosis and disposal of system faults based on a large model detects the relevant parameters during the operation of the system, including the state of the communication protocol connection between the monitored interface program and the acquisition service, the memory usage status and / or CPU utilization rate of the monitored interface program, and the system events of the operating system, in three parallel ways. It fully detects the abnormal states of possible disconnection of the interface program communication, freezing of the interface program, and error reporting of the interface program during the operation of the system. Once any of the three detection methods detects fault information, the interface program is closed and then restarted to ensure the normal operation of the system.

[0048] 2. The automatic fault diagnosis and disposal method based on a large model accurately diagnoses the cause of a fault by constructing a completed database, creates a relevant set of fault causes, and finally outputs the fault diagnosis reason. Among them, this algorithm combines the information in the database, including system and inter-system relationship information, and uses the information in the database based on the output result of the model to iteratively construct a fault occurrence path through iteration to quickly lock the fault occurrence point. The DRN model is used for fault diagnosis, reducing the dependence on expert experience, realizing an automated and intelligent fault diagnosis process, and improving the operability and sustainability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a schematic flowchart of the method of the present invention;

[0050] Figure 2 is a schematic flowchart of using the Gram angular field to encode fault data in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] Refer to Figure 1-2 , the present invention provides two technical solutions:

[0053] The first embodiment: The automatic fault diagnosis and disposal method based on a large model includes the following steps:

[0054] Step 1: Obtain relevant data when the system fails, and establish a database in combination with the relevant data for system fault diagnosis.

[0055] Step 2: Use the Gram angular field to encode the fault data, specifically:

[0056] S21: Obtain one-dimensional time series data from the fault data;

[0057] S22: Perform normalization processing on the time series data to scale the amplitude range of the data to between [0, 1];

[0058] S23: Convert the normalized time series data into a Gram matrix, where each element in the Gram matrix represents the inner product or correlation between two data points;

[0059] S24: Obtain eigenvectors and eigenvalues by performing eigenvalue decomposition on the Gram matrix;

[0060] S25: Convert the corresponding time and amplitude of each point in the time series data into the radius and angle of polar coordinates to form a Gram angle field;

[0061] S26: Use the generated Gram angle field as a two-dimensional image and input it into a deep residual network for fault diagnosis.

[0062] Step 3: Label the fault data according to the system fault types and their fault characteristics, and divide the obtained sample data set into a training set and a test set.

[0063] Step 4: Establish a DRN fault diagnosis model, train the DRN fault diagnosis model with the training data, and use the test data to complete the verification of the DRN fault diagnosis model to obtain an optimal DRN fault diagnosis model.

[0064] Step 5: Use the optimal DRN fault diagnosis model to perform automatic diagnosis and disposal of system faults.

[0065] In an embodiment of the present invention, the relevant parameters include the connection status of the communication protocol between the monitored interface program and the acquisition service, the memory usage status of the monitored interface program, and the system events of the operating system.

[0066] Specifically, the acquisition methods of the three relevant parameters adopted in parallel include:

[0067] Adopt a timed polling method to collect the connection status of the communication protocol between the monitored interface program and the acquisition service, and obtain information on whether the communication of the interface program is disconnected;

[0068] Collect the memory usage status of the monitored interface program in real time, obtain the current memory usage in units of a set time, compare the memory usage at each set time with the memory usage at the previous set time to obtain information on whether the interface program is stuck. If the memory usage does not change within the set time, it proves that the interface program is in a stuck state;

[0069] Adopt a timed polling method to collect the system events of the operating system to obtain information on whether there are error events in the interface program.

[0070] In an embodiment of the present invention, the relevant parameters further include the connection status of the communication protocol between the monitored interface program and the acquisition service, the CPU utilization rate of the monitored interface program, and the system events of the operating system.

[0071] Specifically, the acquisition methods of the three relevant parameters adopted in parallel include:

[0072] Adopt the method of timed polling to collect the connection status of the communication protocol between the monitored interface program and the collection service, and obtain the information on whether the communication of the interface program is disconnected;

[0073] Real-time collect the CPU utilization rate of the monitored interface program, obtain the current CPU utilization rate in units of set time, and compare the CPU utilization rate of each set time with that of the previous set time to obtain the information on whether the interface program is stuck;

[0074] Adopt the method of timed polling to collect the system events of the operating system to obtain the information on whether there are error events in the interface program.

[0075] The second implementation method: The process of establishing the DRN fault diagnosis model includes:

[0076] Construct multiple residual blocks; each residual block contains one or more convolutional layers, batch normalization layers and activation function layers;

[0077] The output of the residual block is divided into a first part and a second part through a skip connection structure; the first part is the non-linear output F(x) after multiple convolutions; the second part is the identity mapping x that directly transmits the output to the output of the neural network through the skip connection structure as a part of the output result;

[0078] The residual block is expressed as:

[0079] y1 = h ( x1 ) + Γ ( x1,w1 ) ;

[0080] x 1+1 = f ( y1 ) ;

[0081] Let h ( x 1) = x1,x 1+1 = y1;

[0082]

[0083] In the formula, y1 is the output of the l-th residual unit, x1 is the input of the l-th residual unit, and each residual structure is activated through a function; Γ is the residual function, h ( x 1) is the constant mapping layer, w1 is the weight value of the l-th layer residual unit, w1 = {w 1,k|1≤k≤K}, and K is the number of residual unit layers;

[0084] The training of the DRN fault diagnosis model using the training data is specifically as follows:

[0085] Initialize the parameters of the deep residual network;

[0086] Input the training data into the deep residual network, and calculate the output of the model through forward propagation;

[0087] Compare the output of the model with the true label, and calculate the value of the loss function ε:

[0088]

[0089] Use the chain rule to calculate the gradient of the model parameters, and transfer the gradient of the loss function from the output layer back to the input layer;

[0090] Update the parameters of the model using the gradient descent method according to the calculated gradient and learning rate.

[0091] The verification of the DRN fault diagnosis model is completed using the test data, specifically as follows:

[0092] Input the test data into the model, and calculate the output result of the model through forward propagation.

[0093] Compare the output result of the DRN fault diagnosis model with the true label of the test data, and determine whether the performance index of the DRN fault diagnosis model is qualified. If it is qualified, output the optimal DRN fault diagnosis model; otherwise, retrain the model.

[0094] Meanwhile, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited, and conventional equipment can be used.

[0095] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0096] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automatic fault diagnosis and disposal method for a large model-based system, characterized in that, It includes the following steps: Step 1: Obtain relevant data during system failures and establish a database in combination with relevant data for system fault diagnosis; Step 2: Encode the fault data using the Gram angle field; Step 3: Label the fault data according to the system fault types and their fault characteristics, and divide the obtained sample data set into a training set and a test set; Step 4: Establish a DRN fault diagnosis model, train the DRN fault diagnosis model using the training data, and verify the DRN fault diagnosis model using the test data to obtain the optimal DRN fault diagnosis model; Step 5: Use the optimal DRN fault diagnosis model to perform automatic diagnosis and disposal of system faults.

2. The method for automatically diagnosing and disposing of system failures based on a large model according to claim 1, wherein: The relevant parameters include the connection status of the communication protocol between the monitored interface program and the acquisition service, the memory usage status of the monitored interface program, and the system events of the operating system.

3. The method for automatic diagnosis and disposal of system faults based on a large model according to claim 2, wherein: The three acquisition methods for the relevant parameters adopted in parallel in Step 1 include: Adopt a timed polling method to collect the connection status of the communication protocol between the monitored interface program and the acquisition service, and obtain information on whether the communication of the interface program is disconnected; Collect the memory usage status of the monitored interface program in real time, obtain the current memory usage in units of a set time, and compare the memory usage at each set time with the memory usage at the previous set time to obtain information on whether the interface program is stuck; Adopt a timed polling method to collect the system events of the operating system to obtain information on whether there are error events in the interface program.

4. The method for automatic diagnosis and disposal of system faults based on a large model according to claim 1, wherein: The relevant parameters also include the connection status of the communication protocol between the monitored interface program and the acquisition service, the CPU utilization rate of the monitored interface program, and the system events of the operating system.

5. The method for automatic diagnosis and disposal of system faults based on a large model according to claim 4, wherein: The three acquisition methods for the relevant parameters adopted in parallel in Step 1 include: Adopt a timed polling method to collect the connection status of the communication protocol between the monitored interface program and the acquisition service, and obtain information on whether the communication of the interface program is disconnected; Collect the CPU utilization rate of the monitored interface program in real time, obtain the current CPU utilization rate in units of a set time, and compare the CPU utilization rate at each set time with the CPU utilization rate at the previous set time to obtain information on whether the interface program is stuck; Adopt a timed polling method to collect the system events of the operating system to obtain information on whether there are error events in the interface program.

6. The method for automatic diagnosis and disposal of system failures based on a large model according to claim 1, characterized in that: Step 2 is specifically as follows: S21: Obtain one-dimensional time series data from the fault data; S22: Perform normalization processing on the time series data to scale the amplitude range of the data to between [0, 1]; S23: Convert the normalized time series data into a Gram matrix, where each element in the Gram matrix represents the inner product or correlation between two data points; S24: Obtain eigenvectors and eigenvalues by performing eigenvalue decomposition on the Gram matrix; S25: Convert the corresponding time and amplitude of each point in the time series data into the radius and angle of polar coordinates to form a Gram angle field. S26: Input the generated Gram angular field as a two-dimensional image into the deep residual network for fault diagnosis.

7. The method for automatic diagnosis and disposal of system failures based on a large model according to claim 1, characterized in that: The process of establishing the DRN fault diagnosis model includes: Construct multiple residual blocks; each residual block contains one or more convolutional layers, batch normalization layers, and activation function layers; The output of the residual block is divided into a first part and a second part through a skip connection structure; the first part is the non-linear output F(x) after multiple convolutions; the second part is the identity mapping x that directly transmits the output to the output of the neural network through the skip connection structure as a part of the output result; The residual block is expressed as: y1 = h(x1) + Γ(x1, w1); x 1+1 = f(y1); Let h(x1) = x1, x 1+1 = y1; where y1 is the output of the l-th residual unit, x1 is the input of the l-th residual unit, and each residual structure is activated by a function; Γ is the residual function, h ( x 1) is the identity mapping layer, w1 is the weight value of the l-th layer residual unit, w1 = { w 1,k|1≤k≤K} , K is the number of residual unit layers; Training the DRN fault diagnosis model using the training data is specifically as follows: Initialize the parameters of the deep residual network; Input the training data into the deep residual network, and calculate the output of the model through forward propagation; Compare the output of the model with the true label, and calculate the value of the loss function ε: Use the chain rule to calculate the gradient of the model parameters, and transmit the gradient of the loss function from the output layer back to the input layer; Update the model parameters using the gradient descent method according to the calculated gradient and learning rate.

8. The method for automatically diagnosing and disposing of system faults based on a large model according to claim 1, wherein: Verifying the DRN fault diagnosis model using the test data is specifically as follows: Input the test data into the model, and calculate the output result of the model through forward propagation.

Images