Autonomous construction method of fault diagnosis model based on meta-Bayesian optimization and meta-features

Abstract

A method for self-building a fault model based on Bayesian optimization and meta-features belongs to the field of electrical data processing technology. The method includes: constructing a mapping between each meta-feature in the data set meta-features to the meta-features in the modeling meta-feature set on historical tasks to form a modeling meta-data set; constructing a meta-agent model to measure each data in the modeling meta-feature set; designing a ranking loss according to the measurement results of each data in the modeling meta-feature set, and obtaining the parameters of the optimal meta-agent model according to the ranking loss; inputting the data meta-feature set on the current task into the meta-agent model, and the meta-agent model gives the modeling result. The present invention constructs a mapping between the data set meta-features to the modeling decision meta-features on a large number of historical tasks in the field to form a modeling meta-data set, relaxes the measurement result loss to a pairwise ranking loss, realizes automatic modeling of small sample diagnosis tasks on new data sets, and quickly obtains competitive initial diagnosis capabilities.

Description

Technical Field

[0001] The invention relates to an autonomous construction method of a fault diagnosis model based on meta-Bayesian optimization and meta-features, and belongs to the technical field of data electrical data processing. Background Art

[0002] The publication number CN113315661A discloses a carrier network fault diagnosis method based on a dynamic Bayesian network. The method includes constructing a fault propagation model according to the relationship between faults and symptoms; constructing a dynamic Bayesian fault propagation model according to an abnormal symptom node set and an abnormal symptom node set, and obtaining a fault diagnosis model on two time slices according to the dynamic Bayesian fault propagation model; obtaining the maximum possible set of fault node states in the previous stage according to the fault diagnosis model on the two time slices; and obtaining the maximum possible set of fault node states in the current stage according to the maximum possible set of fault node states in the previous stage.

[0003] Although the invention models the fault diagnosis problem in a dynamic network environment as a dynamic Bayesian fault diagnosis model and proposes a carrier network fault diagnosis method based on a dynamic Bayesian network, which can effectively reduce the noise problem caused by the dynamic network environment to the fault diagnosis model, it cannot achieve automatic modeling of small sample diagnosis tasks on new data sets and quickly obtain competitive initial diagnosis capabilities. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features. The present invention constructs a mapping between dataset meta-features and modeling decision meta-features on a large number of domain historical tasks to form a modeling meta-data set, relaxes the measurement result loss to a pairwise ranking loss, and realizes automatic modeling of small sample diagnosis tasks on new datasets, thereby quickly acquiring competitive initial diagnostic capabilities.

[0005] To achieve the above-mentioned object, the present invention provides a method for self-building a fault model based on Bayesian optimization and meta-features, which is characterized by comprising:

[0006] Step 1: Construct a dataset meta-feature set based on historical tasks Each meta-feature in the modeling fault meta-feature set The mapping between the meta-features in the modeling metadata set , I and J are positive integers greater than or equal to 2;

[0007] Step 2: Build a meta-agent model Modeling Metadata Set Each data in is measured, where are the parameters of the meta-agent model; is a hyperparameter;

[0008] Step 3: Based on the modeling metadata set The measurement results of each data in the design ranking loss, according to the ranking loss to obtain the parameters of the optimal meta-proxy model ;

[0009] Step 4: Set the data element feature set for the current task Input to the meta-agent model, Give fault diagnosis results .

[0010] Compared with the prior art, the method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features provided by the present invention has the following beneficial effects:

[0011] The present invention constructs a mapping between dataset meta-features and modeling decision meta-features on a large number of historical tasks in the field to form a modeling meta-data set, relaxes the measurement result loss to a pairwise ranking loss, realizes automatic modeling of small sample diagnosis tasks on new datasets, and quickly obtains competitive initial diagnostic capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features provided by the present invention. DETAILED DESCRIPTION

[0013] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0014] In the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0015] Figure 1 is a flow chart of the method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features provided by the present invention, such as Figure 1 As shown, the method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features provided by the present invention includes:

[0016] Step 1: Construct a dataset meta-feature set based on historical tasks Each meta-feature in the modeling fault meta-feature set The mapping between the meta-features in the modeling metadata set , I and J are positive integers greater than or equal to 2;

[0017] Step 2: Build a meta-agent model Modeling Metadata Set Each data in is measured, where are the parameters of the meta-agent model; is a hyperparameter;

[0018] Step 3: Based on the modeling metadata set The measurement results of each data in the design ranking loss, according to the ranking loss to obtain the parameters of the optimal meta-proxy model ;

[0019] Step 4: Set the data element feature set for the current task Input to the Meta-Proxy Model , by the meta-agent model Give fault diagnosis results .

[0020] In the present invention,

[0021] Where, k = 1,…,J, is the activation function.

[0022] The present invention constructs a mapping between dataset meta-features and modeling decision meta-features on a large number of historical tasks in a field through the above-mentioned technical solution, constitutes a modeling meta-data set, relaxes the measurement result loss to a pairwise ranking loss, realizes automatic modeling of small sample diagnosis tasks on new datasets, and quickly obtains competitive initial diagnostic capabilities.

[0023] The method for self-building a fault model based on Bayesian optimization and meta-features provided by the present invention further includes, before step 1: completing the learning of meta-knowledge of hyperparameter optimization through a Bayesian proxy model to achieve hyperparameter optimization under a small number of optimization iterations. Seek excellence.

[0024] In the fault model self-building method based on Bayesian optimization and meta-features provided by the present invention, the meta-knowledge of hyperparameter optimization is learned through the meta-learning Bayesian proxy model to achieve hyperparameter optimization under a small number of optimization iterations. Optimization includes:

[0025] S1: Input a small amount of observation data , from the observation data Sampling subset , T is a positive integer less than or equal to K, K is a positive integer greater than or equal to 10;

[0026] S2: Calculate the loss function according to the following formula: ,

[0027] In the formula, is the i-th data set combination, Combination of the i-th data set Evaluation results of cross-task meta-learning; Bayesian surrogate model Parameters;

[0028] S3: The Bayesian surrogate model is made by of :

[0029] ,

[0030] In the formula, is the learning coefficient; For The gradient of

[0031] S4: The Bayesian surrogate model is made by Parameters :

[0032] ,

[0033] In the formula, is the learning coefficient; For The gradient of

[0034] S5: Determine whether the loss function L is the minimum. If the loss function L is not the minimum, return to step S2. If the loss function L is the minimum, terminate the training and output the Bayesian proxy model with better parameters. .

[0035] The present invention uses the above technical solution to train the Bayesian proxy model on a small amount of observation data of fixed length to obtain a Bayesian proxy model with better parameter super-meta-learning.

[0036] The cross-task data set fusion method based on meta-Bayesian optimization provided by the present invention also includes the following steps:

[0037] S6: Make ;

[0038] S7: From observational data Sampling subset ;

[0039] S8: Calculate the loss function according to the following formula: ;

[0040] S9: Update the Bayesian surrogate model by Parameters :

[0041] ,

[0042] In the formula, is the learning coefficient; For The gradient of

[0043] S10: The Bayesian surrogate model is made by Parameters :

[0044] ,

[0045] In the formula, is the learning coefficient; For The gradient of

[0046] S11: Determine the loss function Is it the minimum? If the loss function If the loss function does not reach the minimum, assign M+1 to M and return to step S7. If the loss function reaches the minimum, the training is terminated and the optimal parameter super-element learning Bayesian proxy model is output. , thus obtaining the hyperparameters: .

[0047] The present invention uses the above technical solution to train the Bayesian proxy model on a small amount of observation data of variable length to obtain the Bayesian proxy model for optimal parameter super-meta-learning.

[0048] The present invention also provides a system, which includes a storage medium and one or more processors, wherein the storage medium is used to store a computer program, and the computer program can be called by one or more processors to implement the above method.

[0049] The present invention also provides a computer program product, which is a program code compiled by a computer language according to the above method, and the program code can be called and executed by one or more processors.

[0050] The present invention constructs a mapping between dataset meta-features and modeling decision meta-features on a large number of historical tasks in different fields to form a modeling meta-dataset. By relaxing the measurement result loss to a pairwise ranking loss, the present invention realizes automatic modeling of small sample diagnosis tasks on new datasets and quickly acquires competitive initial diagnostic capabilities. The method is mainly used in electromechanical products.

[0051] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for autonomously constructing a fault diagnosis model based on meta-Bayesian optimization and meta-features, characterized in that: include: Step 1: Construct a dataset meta-feature set based on historical tasks Each meta-feature in the modeling fault meta-feature set The mapping between the meta-features in the modeling metadata set , I and J are positive integers greater than or equal to 2; Step 2: Build a meta-agent model Modeling Metadata Set Each data in is measured, where are the parameters of the meta-agent model; is a hyperparameter; Step 3: Based on the modeling metadata set The measurement results of each data in the design ranking loss, according to the ranking loss to obtain the parameters of the optimal meta-proxy model ; Step 4: Set the data element feature set for the current task Input to the Meta-Proxy Model , by the meta-agent model Give fault diagnosis results .

2. The method according to claim 1, characterized in that ， Where k = 1,…,J, is the activation function.

3. The method according to claim 1, characterized in that Before step 1, it also includes: using the Bayesian proxy model to complete the learning of meta-knowledge of hyperparameter optimization to achieve hyperparameter optimization in a small number of optimization iterations Seek excellence.

4. The method according to claim 3, characterized in that Through the Bayesian proxy model, the meta-knowledge of hyperparameter optimization is learned to achieve hyperparameter optimization with a small number of optimization iterations. Optimization includes: S1: Input a small amount of observation data , from the observation data Sampling subset , T is a positive integer less than or equal to K, K is a positive integer greater than or equal to 10; S2: Calculate the loss function according to the following formula: , In the formula, is the i-th data set combination, Combination of the i-th data set Evaluation results of cross-task meta-learning; Bayesian surrogate model Parameters, S3: The Bayesian surrogate model is made by of : ， In the formula, is the learning coefficient; For The gradient of S4: The Bayesian surrogate model is made by Parameters : , where is the learning coefficient; For The gradient of S5: Determine whether the loss function L is the minimum. If the loss function L is not the minimum, return to step S2. If the loss function L is the minimum, terminate the training and output a better meta-learning Bayesian proxy model. .

5. The method according to claim 4, characterized in that The following steps are also included: S6: Make ; S7: From observational data Sampling subset ; S8: Calculate the loss function according to the following formula: ; S9: Update the Bayesian surrogate model by Parameters : ， In the formula, is the learning coefficient; For The gradient of S10: The Bayesian surrogate model is made by Parameters : ， In the formula, is the learning coefficient; For The gradient of S11: Determine the loss function Is it the minimum? If the loss function If the loss function does not reach the minimum, assign M+1 to M and return to step S7. If the loss function reaches the minimum, the training is terminated and the optimal parameter super-element learning Bayesian proxy model is output. , thus obtaining the hyperparameters: .

Images