High-dimensional time-varying reliability evaluation method based on active learning and Bayesian deep neural network

Abstract

The invention provides a high-dimensional time-varying reliability evaluation method based on active learning and a Bayesian deep neural network, which is used for evaluating the reliability of advanced equipment and comprises the following steps of: decomposing a random process variable involved in a time-varying reliability analysis problem into a plurality of random variables; preliminarily constructing a Bayesian deep neural network; monte Carlo sampling is carried out on an input variable, a predicted output response mean value and variance of the Bayesian deep neural network corresponding to the input variable Monte Carlo sample in a current iteration step are calculated, and a new active learning strategy of batch point adding is constructed according to the predicted output response mean value and variance. Active construction of the Bayesian deep neural network is carried out; the confidence degree of the failure probability is solved according to estimation of a Bayesian deep neural network on a predicted output response symbol, so that automatic convergence of a high-dimensional time-varying reliability modeling process is achieved, and construction of the Bayesian deep neural network is completed; and solving the time-varying reliability according to the constructed Bayesian deep neural network.

Description

Technical Field

[0001] The present invention belongs to the technical field of time-varying reliability, and in particular relates to a high-dimensional time-varying reliability evaluation method based on active learning and Bayesian deep neural network. Background Art

[0002] The large-scale, intelligent, integrated, and multidisciplinary nature of these advanced equipment makes them complex and sophisticated. Coupled with the extreme and diverse operating conditions, these equipment are subject to numerous dynamic and static uncertainties during their service life, with dimensions reaching tens or even hundreds. These uncertainties interact, propagate layer by layer, and are coupled and amplified, leading to fluctuations and dispersion in the performance of advanced equipment. This causes the reliability of these equipment to decline over time, creating a high-dimensional, time-varying reliability problem. For example, due to machining and assembly errors, the geometric characteristics of aircraft engine compressor blades are uncertain. Furthermore, due to the degradation of blade material properties caused by fouling, wear, and erosion, as well as the dynamic changes in geometric characteristics such as the blade leading edge and tip clearance, the operational stability of the aircraft engine is gradually reduced. The number of uncertain geometric characteristics for a single-stage blade alone can reach dozens. Considering the geometric differences between multiple blade stages, coupled with uncertainties such as material parameters and various environmental stresses, the dimensionality of this problem can reach hundreds of dimensions.

[0003] Advanced equipment has extremely high reliability requirements. Any failure or malfunction is highly likely to cause significant property damage or casualties. Therefore, conducting time-varying reliability analysis on advanced equipment—determining the cumulative failure probability and reliability variation of advanced equipment during service based on the uncertainty of input variables—is crucial for improving its reliability. While time-varying reliability analysis has been extensively studied, existing methods are mostly effective for low-dimensional problems. When dealing with high-dimensional problems, their accuracy and efficiency struggle to meet practical engineering requirements. On the one hand, when the input variables are high-dimensional, the strong coupling between them can lead to a strong nonlinearity in the target problem, resulting in large solution errors in existing dynamic reliability analysis methods. On the other hand, solving high-dimensional dynamic problems requires a larger sample size, resulting in low computational efficiency and even the curse of dimensionality.

[0004] The high-dimensionality, dynamic nature of variables, and the black-box nature of the problem make solving high-dimensional, time-varying reliability problems for advanced equipment extremely complex. Existing reliability analysis methods can encounter challenges such as large solution errors, low computational efficiency, the curse of dimensionality, or even inability to solve high-dimensional dynamic problems. Therefore, developing high-dimensional dynamic reliability analysis methods is crucial to providing technical support for reliability assessment and optimal design throughout the lifecycle of advanced equipment. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art and provide a high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network. The high-dimensional time-varying reliability problem is comprehensively modeled using Bayesian deep neural network through active learning, thereby achieving efficient solution of the time-varying reliability analysis problem directly in high-dimensional space.

[0006] In a first aspect, the present invention provides a high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network for assessing the reliability of advanced equipment, comprising the following steps:

[0007] Step 1: Use KL decomposition to decompose the random process variables involved in the time-varying reliability analysis problem of advanced equipment into several random variables;

[0008] Step 2: Use the orthogonal experimental design method to collect a small number of sample points based on the statistical characteristics of the input variables and solve their responses to obtain the initial training samples and preliminarily construct a Bayesian deep neural network;

[0009] Step 3: Perform Monte Carlo sampling on the input variables, calculate the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the Monte Carlo sample of the input variables in the current iteration step, and construct a new batch-added active learning strategy based on the mean and variance of the predicted output response to actively build the Bayesian deep neural network;

[0010] Step 4: Define a convergence criterion based on the confidence level of the failure probability. Calculate the confidence level of the failure probability based on the Bayesian deep neural network's estimate of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process and completing the construction of the Bayesian deep neural network.

[0011] Step 5: Based on the constructed Bayesian deep neural network, calculate the time trajectory corresponding to the Monte Carlo sample points of each input variable, determine whether the time trajectory is invalid, and solve the time-varying reliability based on the frequency of all failed sample trajectories.

[0012] Based on the above, in step 3, the new batch addition active learning strategy constructed is:

[0013] First, a prediction extreme value sample set and a prediction failure sample set are constructed based on the mean of the predicted output response. The prediction extreme value sample represents the input sample point corresponding to the minimum value of the time trajectory, and the prediction failure sample represents the input sample point that makes the function invalid.

[0014] Secondly, the K-means clustering method is used to perform cluster analysis on the predicted extreme value sample set and the predicted failure sample set. Then, based on the mean and variance of the output response predicted by the Bayesian deep neural network, the U function value corresponding to each type of sample is calculated, and then the point with the smallest U function value in each type of sample is selected as the new sample point for the next iteration step.

[0015] Based on the above, the number of clusters K1 and K2 of the predicted extreme value sample set and the predicted failure sample set are constrained so that they gradually decrease with the increase of the number of iteration steps n, which can be expressed as:

[0016] K1=10+int(n / 2)

[0017] K2=10+int(n / 2).

[0018] Based on the above, in step 5, the convergence criterion based on the failure probability confidence level is defined as:

[0019] N MCS =N1+N2+N3+N4

[0020] N f ≤N MCS -N1-N4

[0021]

[0022] In the above formula: N MCS Indicates the number of input variable samples obtained by Monte Carlo sampling of the input variable, N1 indicates the number of samples whose U function is greater than or equal to 2 and whose predicted mean is greater than or equal to 0, N2 indicates the number of samples whose U function is less than 2 and whose predicted mean is greater than or equal to 0, N3 indicates the number of samples whose U function is less than 2 and whose predicted mean is less than 0, N4 indicates the number of samples whose U function is greater than or equal to 2 and whose predicted mean is less than 0, N f Indicates the number of true failure samples;

[0023] If the convergence criterion based on the failure probability confidence level meets the convergence condition, it is considered that the model accuracy of the current Bayesian deep neural network has reached the requirement, and the iteration can be terminated and the next step can be entered; otherwise, the new samples in the current iteration step are added to the training samples, and steps 2 to 5 are continued to be iterated until the maximum number of iteration steps or the maximum number of sample points is reached.

[0024] In a second aspect, the present invention provides a high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network, comprising:

[0025] The first processing module is used to decompose the random process variables involved in the time-varying reliability analysis problem of advanced equipment into a plurality of random variables using KL decomposition;

[0026] The second processing module is used to: use the orthogonal experimental design method to collect a small number of sample points according to the statistical characteristics of the input variables and solve their responses, obtain initial training samples and preliminarily construct a Bayesian deep neural network;

[0027] The third processing module is used to perform Monte Carlo sampling on the input variables, calculate the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the Monte Carlo sample of the input variables in the current iteration step, and construct a new batch-added active learning strategy based on the mean and variance of the predicted output response to actively construct the Bayesian deep neural network;

[0028] The fourth processing module is used to define a convergence criterion based on the confidence level of the failure probability, and solve the confidence level of the failure probability based on the Bayesian deep neural network's estimation of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process and completing the construction of the Bayesian deep neural network;

[0029] The fifth processing module is used to calculate the time trajectory corresponding to the Monte Carlo sample points of each input variable based on the constructed Bayesian deep neural network, determine whether the time trajectory has failed, and solve the time-varying reliability based on the frequency of all failed sample trajectories.

[0030] In a third aspect, the present invention provides a computer device, comprising:

[0031] one or more processors;

[0032] a memory for storing one or more programs,

[0033] When the one or more programs are executed by the one or more processors, the one or more processors are caused to perform the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network as described.

[0035] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network as described.

[0036] The present invention has outstanding substantive features and significant improvements over the prior art. Specifically:

[0037] (1) The method proposed in the present invention can efficiently solve the high-dimensional time-varying reliability analysis problem. By adopting the Bayesian deep neural network, the time-varying reliability analysis model is directly constructed in the high-dimensional space, avoiding the approximate error introduced by the dimensionality reduction analysis.

[0038] (2) The batch-based active learning strategy proposed in this invention can screen out key sample points in each iteration step, thereby enabling the rapid construction of a high-dimensional time-varying reliability analysis model. These key sample points come from the predicted extreme value sample set and predicted failure sample set obtained by solving the Bayesian deep neural network. At the same time, the batch-based feature also facilitates parallel computing, further improving the solution efficiency.

[0039] (3) The convergence criterion based on failure probability confidence proposed in the present invention can monitor the accuracy of the model in real time according to the estimation of the predicted output response symbol by the Bayesian deep neural network, thereby achieving adaptive convergence of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a 52-bar truss finite element model.

[0041] Figure 2 This is the distribution of various parameters of the 52-bar truss.

[0042] Figure 3 This is a schematic diagram of a Bayesian deep neural network.

[0043] Figure 4 It is a schematic diagram of the batch adding process.

[0044] Figure 5 This is a schematic diagram of the physical meaning of the U function.

[0045] Figure 6 It is a diagram of the confidence convergence criterion.

[0046] Figure 7 It is a schematic diagram of the convergence process of extreme value distribution.

[0047] Figure 8 It is the process of the extreme value distribution of the 52-bar truss vertex displacement converging with the iteration step.

[0048] Figure 9This is the convergence process of the predicted failure probability of the 52-bar truss vertex displacement with the iteration step. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0050] Example 1

[0051] Taking the finite element method of a rod truss as an example, this embodiment provides a high-dimensional time-varying reliability analysis method based on active learning and Bayesian deep neural network, which specifically includes the following steps:

[0052] Step 1: If Figure 1 As shown in the figure, the finite element model of the 52-bar truss is established using finite element analysis software. Consider the cross-sectional area A of the 52-bar truss i , elastic modulus E of each rod i And the six dynamic external loads P borne by the entire truss j A total of 110 parameters are random variables, and the distribution of each variable is as follows Figure 2 As shown. Among them, the 6 dynamic external loads P j Subject to random process variables, the time-varying reliability of a 52-bar truss is analyzed with the objective that the maximum displacement of its vertices does not exceed a predetermined value.

[0053] In order to facilitate model construction, the KL decomposition is first used to decompose the dynamic external load in the time-varying reliability analysis problem into several random variables, which can be specifically expressed as:

[0054]

[0055] Where: t represents time, θ represents random events, represents the mean of the random process, λ n ,n=1,2,...andφ n (t1) represent the eigenvalue and eigenfunction of the covariance function C(t1,t2), ξ n (θ) represents a random variable with mean 0 and variance 1.

[0056] Step 2: Use the orthogonal experimental design method according to Figure 2 The statistical characteristics of the input samples shown in the figure are collected with a small number of sample points, and their responses are solved by calling the finite element analysis software to obtain the initial training sample D initial :

[0057]

[0058] Where: Nt Indicates the number of discrete time parameters, and N is determined according to the minimum autocorrelation length of the random process variable. t =(4(t T -t0) / Δ min )+1. Then directly construct the Bayesian deep neural network model in high-dimensional space Thus, the function g(x,y(t),t) of the real and time-consuming high-dimensional time-varying reliability problem is approximated. The parameters of the Bayesian deep neural network are solved using the Hamiltonian Monte Carlo method. The Bayesian deep neural network (such as Figure 3 (shown) can be expressed as:

[0059]

[0060] Where: h(z; θ s ) represents the predicted output response of the Bayesian deep neural network, z=(x,y(t),t) represents the input variable vector, represents the weight and bias parameters of the Bayesian deep neural network, h L+1 represents the activation function of the output layer (i.e., the L+1 layer), h m ,m=1,2,...,L represents the activation function of the mth hidden layer, H m ,m=1,2,...,L represents the number of neurons in the mth hidden layer, i=1,2,...H L represents the connection weight parameter of the neurons in the output layer to the i-th neuron in the L-th hidden layer, i=1,2,...H m ,j=1,2,...H m-1 ,m=1,2,..,L represents the connection weight parameter of the i-th neuron in the m-th hidden layer to the j-th neuron in the m-1-th hidden layer,b L+1 represents the bias parameter of the output layer neurons, k=1,2,...H m ,m=1,2,..,L represents the bias parameter of the kth neuron in the mth hidden layer.

[0061] By using the Hamiltonian Monte Carlo method to solve formula (6), we can obtain the model parameter distribution of the Bayesian deep neural network, and then sample its model parameter distribution to solve the prediction mean and prediction variance of the Bayesian deep neural network:

[0062]

[0063] Where: represents the predicted mean of the Bayesian deep neural network, represents the predicted mean square error of the Bayesian deep neural network, and M' represents the number of samples of the Bayesian deep neural network parameters.

[0064] In this step, the initial sample size N = 1000, the time discrete number N t =10, the Bayesian deep neural network used contains 3 hidden layers, but the number of neurons in each hidden layer is 50, the activation function uses the ReLU function, and after the Hamiltonian Monte Carlo method is used to solve the posterior distribution of the Bayesian deep neural network parameters, the number of samples of each parameter is M'=10000.

[0065] Step 3: Perform Monte Carlo sampling on the input variable z = [x, y(t), t] to obtain N MCS =10 6 Input variable samples are taken and the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the input variable Monte Carlo sample in the current iteration step are calculated. Then a batch-added active learning strategy is proposed, such as Figure 4 As shown, the strategy first screens out the predicted extreme value sample set and predicted failure sample set The predicted extreme value samples represent the input sample points corresponding to the minimum value of the time trajectory, and the predicted failure samples represent the input sample points that can make the function fail;

[0066] Secondly, the K-means clustering method is used to perform cluster analysis on the predicted extreme value sample set and the predicted failure sample set. Then, according to the predicted mean and predicted variance of the Bayesian deep neural network, the U function value corresponding to the samples in each category is calculated, as follows: Figure 5 As shown:

[0067]

[0068] Where: Then, the point with the smallest U function value among all types of samples is selected as the new sample point for the next iteration step;

[0069] To balance the efficiency and accuracy of active modeling, the number of clusters K1 and K2 of the predicted extreme value sample set and the predicted failure sample set in the K-means clustering method are constrained so that the number of clusters gradually decreases with the increase of the iteration step. The process of cluster analysis of the predicted extreme value sample set can be expressed as follows:

[0070]

[0071] K1=10+int(n / 2) (15)

[0072] Similarly, the process of cluster analysis on the predicted failure sample set can be expressed as:

[0073]

[0074]

[0075] K2=10+int(n / 2) (21)

[0076] Where n represents the number of iteration steps.

[0077] Step 5: Propose a convergence criterion based on the failure probability confidence level (e.g. Figure 6 As shown in Figure 2), the confidence level of the failure probability is set according to the probability model's estimate of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process. The convergence criterion can be expressed as:

[0078] N MCS =N1+N2+N3+N4 (22)

[0079] N f ≤N MCS -N1-N4 (23)

[0080]

[0081] Where: N1 represents the number of samples whose U function is greater than or equal to 2 and whose predicted mean is greater than or equal to 0, N2 represents the number of samples whose U function is less than 2 and whose predicted mean is greater than or equal to 0, N3 represents the number of samples whose U function is less than 2 and whose predicted mean is less than 0, N4 represents the number of samples whose U function is greater than or equal to 2 and whose predicted mean is less than 0, N f Indicates the number of true failure samples.

[0082] If the confidence level of the failure probability meets the set requirements, the current model accuracy is considered to have met the requirements, and the iteration can be terminated and the next step can be entered. Otherwise, the new samples are added to the training samples, and steps 2 to 5 are iterated until the maximum number of iterations or the maximum number of sample points is reached. As the model is updated and iterated, the extreme value distribution of the time trajectory of the vertex displacement of the 52-bar truss will gradually converge to the exact solution (such as Figure 7 shown).

[0083] Step 6: Based on the constructed Bayesian deep neural network model, a subset simulation sampling scheme is used to sample the input variables, and the time trajectory corresponding to each sample point is calculated. The time-varying reliability is then solved based on the frequency of all failure sample trajectories:

[0084]

[0085] For the analysis of this embodiment, the time-varying reliability analysis results of the 52-bar truss vertex displacement converge as the iteration steps change. Figure 8 and Figure 9 As shown in the figure, it can be seen that with the continuous iteration of the active learning modeling process, the extreme value distribution of the vertex displacement of the 52-bar truss gradually converges to an exact solution, and the time-varying reliability analysis results solved by the Bayesian deep neural network are becoming more and more accurate, which proves the effectiveness of the method proposed in this invention.

[0086] Example 2

[0087] Based on the same inventive concept, the present application also provides a high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network. The implementation solution provided by the high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network is similar to the implementation solution described in the method of Example 1. Therefore, the specific limitations of one or more embodiments of the high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network provided below can be referred to the limitations of the method in Example 1 and will not be repeated here.

[0088] In an exemplary embodiment, a high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network is provided, characterized by including:

[0089] The first processing module is used to decompose the random process variables involved in the time-varying reliability analysis problem of advanced equipment into a plurality of random variables using KL decomposition;

[0090] The second processing module is used to: use the orthogonal experimental design method to collect a small number of sample points according to the statistical characteristics of the input variables and solve their responses, obtain initial training samples and preliminarily construct a Bayesian deep neural network;

[0091] The third processing module is used to perform Monte Carlo sampling on the input variables, calculate the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the Monte Carlo sample of the input variables in the current iteration step, and construct a new batch-added active learning strategy based on the mean and variance of the predicted output response to actively construct the Bayesian deep neural network;

[0092] The fourth processing module is used to define a convergence criterion based on the confidence level of the failure probability, and solve the confidence level of the failure probability based on the Bayesian deep neural network's estimation of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process and completing the construction of the Bayesian deep neural network;

[0093] The fifth processing module is used to calculate the time trajectory corresponding to the Monte Carlo sample points of each input variable based on the constructed Bayesian deep neural network, determine whether the time trajectory has failed, and solve the time-varying reliability based on the frequency of all failed sample trajectories.

[0094] Example 3

[0095] Each module in the above system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0096] In an exemplary embodiment, a computer device is provided, which may be a terminal. The computer device further includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is configured to exchange information between the processor and an external device. The communication interface of the computer device is configured to communicate with an external terminal via wired or wireless communication, where the wireless communication may be achieved via Wi-Fi, a mobile cellular network, NFC (near field communication), or other technologies. When executed by the processor, the computer program implements the steps of a high-dimensional time-varying reliability assessment method based on active learning and a Bayesian deep neural network. The display unit of the computer device is configured to produce a visually visible image, and may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0097] Those skilled in the art will understand that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different component arrangement.

[0098] In an exemplary embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network are implemented.

[0099] In an exemplary embodiment, a computer program product is also provided, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of a high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network are implemented.

[0100] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to these.

[0101] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network for evaluating the reliability of advanced equipment, characterized by: The following steps are involved: Step 1: Use KL decomposition to decompose the random process variables involved in the time-varying reliability analysis problem of advanced equipment into several random variables; Step 2: Use the orthogonal experimental design method to collect a small number of sample points based on the statistical characteristics of the input variables and solve their responses to obtain the initial training samples and preliminarily construct a Bayesian deep neural network; Step 3: Perform Monte Carlo sampling on the input variables, calculate the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the Monte Carlo sample of the input variables in the current iteration step, and construct a new batch-added active learning strategy based on the mean and variance of the predicted output response to actively build the Bayesian deep neural network; Step 4: Define a convergence criterion based on the confidence level of the failure probability. Calculate the confidence level of the failure probability based on the Bayesian deep neural network's estimate of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process and completing the construction of the Bayesian deep neural network. Step 5: Based on the constructed Bayesian deep neural network, calculate the time trajectory corresponding to the Monte Carlo sample points of each input variable, determine whether the time trajectory is invalid, and solve the time-varying reliability based on the frequency of all failed sample trajectories.

2. The high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network according to claim 1 is characterized in that: In step 3, the new batch-adding active learning strategy constructed is: First, a prediction extreme value sample set and a prediction failure sample set are constructed based on the mean of the predicted output response. The prediction extreme value sample represents the input sample point corresponding to the minimum value of the time trajectory, and the prediction failure sample represents the input sample point that makes the function invalid. Secondly, the K-means clustering method is used to perform cluster analysis on the predicted extreme value sample set and the predicted failure sample set. Then, based on the mean and variance of the output response predicted by the Bayesian deep neural network, the U function value corresponding to each type of sample is calculated, and then the point with the smallest U function value in each type of sample is selected as the new sample point for the next iteration step.

3. The high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network according to claim 2 is characterized by: The number of clusters for the predicted extreme value sample set and the predicted failure sample set K 1 and K 2 is constrained so that it increases with the number of iterations n It gradually decreases with the increase of ， 。 4. The high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network according to any one of claims 1 to 3, characterized in that: In step 5, the convergence criterion based on the failure probability confidence level is defined as: ， ， ， In the above formula: represents the number of input variable samples obtained by Monte Carlo sampling of the input variable, N 1 indicates the number of samples whose U function is greater than or equal to 2 and whose predicted mean is greater than or equal to 0. N 2 represents the number of samples whose U function is less than 2 and whose predicted mean is greater than or equal to 0. N 3 represents the number of samples whose U function is less than 2 and whose predicted mean is less than 0. N 4 represents the number of samples whose U function is greater than or equal to 2 and whose predicted mean is less than 0. N f Indicates the number of true failure samples; If the convergence criterion based on the failure probability confidence level meets the convergence condition, it is considered that the model accuracy of the current Bayesian deep neural network has reached the requirement, and the iteration can be terminated and the next step can be entered; otherwise, the new samples in the current iteration step are added to the training samples, and steps 2 to 5 are continued to be iterated until the maximum number of iteration steps or the maximum number of sample points is reached.

5. A high-dimensional time-varying reliability assessment system based on active learning and Bayesian deep neural network, characterized by: include: The first processing module is used to decompose the random process variables involved in the time-varying reliability analysis problem of advanced equipment into a plurality of random variables using KL decomposition; The second processing module is used to: use the orthogonal experimental design method to collect a small number of sample points according to the statistical characteristics of the input variables and solve their responses, obtain initial training samples and preliminarily construct a Bayesian deep neural network; The third processing module is used to perform Monte Carlo sampling on the input variables, calculate the mean and variance of the predicted output response of the Bayesian deep neural network corresponding to the Monte Carlo sample of the input variables in the current iteration step, and construct a new batch-added active learning strategy based on the mean and variance of the predicted output response to actively construct the Bayesian deep neural network; The fourth processing module is used to define a convergence criterion based on the confidence level of the failure probability, and solve the confidence level of the failure probability based on the Bayesian deep neural network's estimation of the predicted output response sign, thereby achieving automatic convergence of the high-dimensional time-varying reliability modeling process and completing the construction of the Bayesian deep neural network; The fifth processing module is used to calculate the time trajectory corresponding to the Monte Carlo sample points of each input variable based on the constructed Bayesian deep neural network, determine whether the time trajectory has failed, and solve the time-varying reliability based on the frequency of all failed sample trajectories.

6. A computer device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the high-dimensional time-varying reliability assessment method based on active learning and Bayesian deep neural network are implemented.

Images