Fault probability monitoring and regulation system and method

Abstract

The application discloses a fault probability monitoring and regulation system and method, relates to the technical field of fault monitoring, and collects to-be-determined related parameters in each period through a collection module, determines the optimal order of an autoregressive model of each related parameter by using the Akaike information criterion, and verifies the to-be-determined related parameters in each period, decides whether to perform secondary collection and replacement based on a verification result, generates a related parameter vector of each period, and starts a window translation along a period in a reverse order from the related parameter vector of each period through a monitoring module, calculates and arranges a comprehensive feature vector in each window, extracts and generates a core feature vector set through a time series self-encoder, trains a hidden Markov model by using a forward-backward algorithm, and obtains the fault probability of each period based on Bayesian inference, compares the fault probability with a probability threshold value to determine whether a fault exists, and executes regulation when the fault exists, so that high-precision fault monitoring and regulation based on collection self-checking, multi-dimensional data fusion and probability inference are realized.

Description

Technical Field

[0001] This invention relates to the field of fault monitoring technology, specifically to a fault probability monitoring and control system and method. Background Technology

[0002] In industrial production and equipment maintenance, the stable operation of various systems and equipment is crucial. Failures can not only disrupt production and reduce efficiency, but also cause economic losses and even endanger personnel safety. Accurate and timely fault monitoring can detect potential anomalies in advance, buying time for preventative maintenance and fault repair, and ensuring reliable system operation. Therefore, conducting fault monitoring is of significant necessity.

[0003] However, existing fault monitoring technologies have the following shortcomings:

[0004] The lack of verification and analysis of the collected data during the data acquisition process makes it difficult to avoid misjudgments of equipment failure due to sensor malfunctions.

[0005] The monitoring process lacks in-depth feature mining of multidimensional data and efficient methods for integrating feature extraction and probabilistic reasoning, thus the accuracy and efficiency of fault identification need to be improved. Summary of the Invention

[0006] The purpose of this invention is to provide a fault probability monitoring and control system and method, and to provide a fault monitoring method and adaptive control method that realizes sensor acquisition self-testing, in-depth mining of multi-dimensional related data and probabilistic reasoning.

[0007] The technical solution to achieve the purpose of this invention is as follows:

[0008] The fault probability monitoring and control system includes a data acquisition module, a monitoring module, and a control module.

[0009] The acquisition module acquires M undetermined relevant parameters in the k-th period, and uses the Akaike information criterion to determine the autoregressive model Ar(q) for the m-th relevant parameter. m ) optimal order Calculate the m-th theoretical relevant parameter for the k-th period And verify the m-th undetermined relevant parameter in the k-th period. Based on the verification results, a decision is made on whether to perform a second data acquisition and replacement, generating the relevant parameter vector x for the kth period. Fr (k), k=1,2,3,…,m=1,…,M;

[0010] The monitoring module obtains the relevant parameter vector x from the k-th period. Fr(k) Start by translating the window N times in reverse order along the cycle, calculate and organize the comprehensive feature vector in each window, and extract the core feature vector set Y through a temporal autoencoder. Train the hidden Markov model using a forward-backward algorithm and obtain the fault probability p(S) of the k-th cycle based on Bayesian inference. k =2|Y,λ * The value of S is compared with a probability threshold to determine whether a fault exists. k =2 indicates that the state in the k-th cycle is a fault, λ and λ * These are the parameters and optimal parameters of the Hidden Markov Model, respectively.

[0011] When the control module determines a fault in the k-th cycle, it stops the equipment from running. Based on the error between each relevant parameter in the k-th cycle and the corresponding standard relevant parameter, it converts the error into the control quantity of the control device corresponding to each relevant parameter in the k-th cycle using the PID control algorithm and executes the control.

[0012] Furthermore, the acquisition module includes an acquisition control unit and an acquisition verification unit;

[0013] The acquisition and control unit acquires M undetermined correlation parameters in the k-th cycle and generates an undetermined correlation parameter vector. Based on the resampling annotation vector ξ(k) of the k-th cycle, the relevant parameters that need to be collected a second time are determined, and the results of the second collection are used to replace the corresponding undetermined relevant parameters to generate the relevant parameter vector x. Fr (k);

[0014] The data acquisition and verification unit extracts the m-th undetermined relevant parameter for the k-th period. Traversal order q m The set of values, based on the fitting order q of the m-th correlation parameter before the k-th period. m The autoregressive model Ar(q) corresponding to each value m The optimal order is chosen to minimize the Akaike information criterion. and the corresponding optimal autoregressive model Substitute into the k-th period The m-th correlation parameter of the k-th period is obtained by obtaining the m-th theoretical correlation parameter of the k-th period. And combined with the m-th undetermined relevant parameter of the k-th period Calculate the error of the m-th parameter in the k-th period, compare the error of the m-th parameter in the k-th period with the error threshold of the m-th parameter, and if it is greater than or equal to the error threshold of the m-th parameter, re-label the m-th parameter in the k-th period. m (k) is set to 1. If it is less than the error threshold of the m-th parameter, the m-th re-acquisition label ξ in the k-th period will be re-acquired and labeled. m(k) is set to 0, and the resampling annotation vector ξ(k) for the kth period is generated and fed back to the acquisition and control unit.

[0015] Furthermore, the data acquisition and verification unit determines the optimal order of the m-th relevant parameter. and the corresponding optimal autoregressive model Includes the following steps:

[0016] Traversal order q m The set of values ​​for q, based on the order q m A variational autoregressive model Ar(q) is constructed using a single value. m That is, the m-th relevant parameter in each cycle is equal to the q-th parameter in the previous cycle. m The weighted sum of the m-th relevant parameter over each period;

[0017] Before obtaining q in the kth period m The m-th relevant parameter of +1 period is fitted using the least squares method to obtain the variational autoregressive model Ar(q). m ) of q m One autoregressive coefficient;

[0018] Calculate the order q m A variational autoregressive model Ar(q) with a single value m Akaike Information Criterion AIC(q) m ); Select the optimal order that minimizes the Akaike information criterion. And obtain the corresponding optimal autoregressive model.

[0019] Furthermore, the monitoring module includes a feature processing unit and a fault determination unit;

[0020] The feature processing unit initializes the end period of the window to the kth period and shifts a single period N times in reverse order along the period. It calculates the time series statistics and cross-correlation matrix in each window and flattens them into vectors. It concatenates the vectors with the relevant parameter vectors of the end period in each window to generate the corresponding comprehensive feature vector. The fault-sensitive information of the comprehensive feature vector of each window is retained by the time series autoencoder and the period correlation is captured to generate the core feature vector set Y.

[0021] The fault determination unit defines state S and uses a forward-backward algorithm to maximize the log-likelihood function logp(Y|λ). It then trains a hidden Markov model using the core feature vector set Y to obtain the optimal parameter set λ. * The failure probability p(S) in the k-th cycle is calculated using Bayesian inference. k =2|Y,λ *) and compare it with the probability threshold. If it is greater than or equal to the probability threshold, it is determined that the fault occurred in the k-th cycle. If it is less than the probability threshold, no action is taken. Here, λ is the parameter set of the Hidden Markov Model.

[0022] Specifically, the nth window includes the relevant parameter vectors from the (kn-N0+2)th period to the (k-n+1)th period, where N0 is the window size. The mth relevant parameter for the N0th period within the nth window is obtained, and the window mean μ of the mth relevant parameter in the nth window is calculated for each period. n,m Window standard deviation σ n,m Maximum window value Minimum window Window skewness Sk n,m Window kurtosis Ku n,m and the lag autocorrelation Af n,m Among them, window skewness Sk n,m Window kurtosis Ku n,m and the lag autocorrelation Af n,m The specific formula is as follows:

[0023]

[0024]

[0025] Among them, window skewness Sk n,m and window kurtosis Ku n,m Reflecting the symmetry and tail thickness of the numerical distribution of the m-th relevant parameter within the n-th window, respectively, hysteresis-autocorrelation Af n,m To measure the temporal linear correlation of the m-th relevant parameter within the n-th window, the relevant parameters within the n-th window are paired, and the covariance of each pair is calculated and divided by the product of the window standard deviations of the two relevant parameters in the pair. This yields the cross-correlation coefficient for each pair, which is then arranged into the cross-correlation coefficient matrix ρ for the n-th window. n Extract the cross-correlation matrix ρ of the nth window. n The upper triangular elements in the data are combined with the time series statistics of the M related parameters in the nth window and the related parameter vector in the (k-n+1)th period. The composite feature vector of the nth window is generated by concatenation and Z-score normalization. Obtain the comprehensive feature vector set

[0026] Furthermore, the temporal autoencoder includes a bidirectional encoding layer, an attention cohesion layer, and a reconstruction verification layer;

[0027] The bidirectional encoding layer uses forward and backward long short-term memory networks to process the comprehensive feature vector set respectively. The forward long short-term memory network integrates the feature vector of the nth window. Forward cell state of the (n+1)th window With forward hidden state Mapped to the forward cell state of the nth window With forward hidden state The process is repeated until the forward cell state of the first window is generated. and forward hidden state When the time stops, the backward long short-term memory network processes data in reverse order, starting from the first window and looping until the backward cell state of the Nth window is generated. and backward hidden state Stop when the time is up, and stitch together the forward hidden state of the nth window. With backward hidden state Generate the corresponding hidden state h n Organize into a hidden state sequence

[0028] The attention cohesion layer will hide the state h of the nth window. n The linear modulation, after being mapped by the Tanh function, is multiplied by the attention vector to obtain the attention score for the nth window. The attention weights are then normalized to the nth window using the Softmax function. Multiply the attention weight of each window by the hidden state and sum them to generate the context hidden state h. Ag It is then concatenated with the hidden state of each window, and mapped to the core feature vector of each window through linear modulation and ReLU function to generate the core feature vector set Y;

[0029] The reconstruction verification layer will use the core feature vector y of the nth window in the core feature vector set Y. n The reconstructed comprehensive feature vector of the nth window is generated by mapping linear modulation with the ReLU function. And organize them into a reconstructed comprehensive feature vector set Computational reconstruction of the comprehensive feature vector set With integrated feature vector set The reconstruction error is calculated and compared with a reconstruction error threshold. If and only if the reconstruction error is greater than or equal to the reconstruction error threshold, the parameters of the feature embedding layer, bidirectional encoding layer, and attention cohesion layer are updated in reverse based on the reconstruction error threshold, and the comprehensive feature vector set is reprocessed.

[0030] Furthermore, calculate the failure probability p(S) in the k-th cycle. k =2|Y,λ * This includes the following steps:

[0031] Define S = 1 and S = 2 to represent normal and fault states S, respectively. Initialize the parameter set λ, which includes the state probability vector π of the Nth window. N The state transition matrix A and the observation probability matrix B are defined as follows: A(j1,j2) in the j1st row and j2nd column of the state transition matrix A reflects the probability of transitioning from state S = j1 to state S = j2; B(j1,n) in the j1st row and nth column of the observation probability matrix B represents the core feature vector of the nth window appearing in state S = j1. The probability, j1,j2∈{1,2};

[0032] Define the state S of the nth window. k-n+1 =j1 and the probability of observing the core feature vectors from the Nth window to the (n+1th)th window is the forward variable α. n (j1), forward variable α n (j1) satisfies the forward recursion formula, as follows:

[0033]

[0034] Establish the forward variable α based on the forward recursion formula n (j1) and forward variable α N The linear relationship of (j3), the forward variable α N (j3) is as follows:

[0035] α N (j3)=π N (S k-N+1 =j3)·B(j3,N),

[0036] Where, π N (S k-N+1 =j3) is the state probability vector π of the Nth window. N Mid-state S k-N+1 = probability j3, j3∈{1,2};

[0037] Define the state S of the nth window. k-n+1 =j1 and the probability of observing the core feature vectors from the (n-1)th window to the 1st window is the backward variable β. n (j1), backward variable β n (j1) satisfies the backward recursion formula, as follows:

[0038]

[0039] Establish the backward variable β based on the backward recursion formula n (j1) and the linear relationship between the backward variable α1(j3)=1;

[0040] Define the state S of the nth window.k-n+1 The probability of j1 is the single-state probability γ. n (j1), as follows:

[0041]

[0042] Single-state probability γ n (j1) reflects the state S of the nth window. k-n+1 Expectations;

[0043] Define the state S of the nth window. k-n+1 =j1 and the state S of the (n-1)th window k-n+2 The probability of j2 is the joint probability γ of the two states. n (j1,j2), as follows:

[0044]

[0045] The joint probability γ of the two states n (j1,j2) reflects the state S from the nth window. k-n+1 The state S transitions to the (n-1)th window k-n+2 Expectations;

[0046] The objective is to maximize the log-likelihood function logp(Y|λ). In each iteration, the parameter set λ is updated until the change in the parameter set between adjacent iterations is less than the convergence threshold, thus obtaining the optimal parameter set λ. * , where, p(Y|λ)=α1(1)+α1(2)=β N (1)+β N (2);

[0047] Combining the optimal parameter set λ * The optimal state probability vector π of the Nth window * N Using the optimal state transition matrix A * and the optimal observation probability matrix B * By performing forward recursion, the failure probability p(S) of the k-th cycle is obtained. k =2|Y,λ * ).

[0048] Furthermore, the parameter set λ of the Hidden Markov Model is updated in the (l+1)th round as follows:

[0049] The state probability vector of the Nth window in the (l+1)th round Mid-state S k-N+1 The probability of =j1 Updated to be based on the parameter set λ of round l. l Calculated single-state probability

[0050] The state transition matrix A of the (l+1)th round l+1 A in row j1 and column j2 l+1 (j1,j2) is updated based on the parameter set λ of the l-th round. l The ratio of the expected number of transitions from state S=j1 to state S=j2 to the total expected number of transitions from state S=j1 is calculated as follows:

[0051]

[0052] in, and Based on the parameter set λ of the lth round respectively l Calculate the joint probability of the two states and the single state probability of the nth window;

[0053] Update the observation probability matrix B for the (l+1)th round based on the maximum likelihood estimation of the Gaussian distribution. l+1 B in row j1 and column n l +1 (j1,n), specifically as follows:

[0054] B l+1 (j1,n)=Gauss(μ l (j1,n),Σ l (j1,n)),

[0055] Where Gauss represents a Gaussian distribution, μ l (j1,n) and Σ l (j1,n) represent the parameter set λ based on the l-th round, respectively. l The state S of the nth window is calculated. k-n+1 The mean and variance of the distribution of j1 are calculated using the following formulas:

[0056]

[0057] The fault probability monitoring and control method includes the following steps:

[0058] Acquire the undetermined correlation parameter vector in the kth period. The optimal order that minimizes the Akaike Information Criterion is determined by fitting the m-th relevant parameter prior to the k-th cycle. and the corresponding optimal autoregressive model

[0059] Before the kth cycle Substituting the m-th relevant parameter of each period into the optimal autoregressive model Obtain the m-th theoretical correlation parameter of the k-th period Combined with the m-th undetermined relevant parameter of the k-th period Calculate the error of the m-th parameter in the k-th period and compare it with the error threshold of the m-th parameter;

[0060] If the error threshold of the m-th parameter is greater than or equal to the threshold value, a second data acquisition is performed, and the acquisition result is used as the m-th related parameter in the k-th period. If the error is less than the threshold of the m-th parameter, the m-th undetermined correlation parameter of the k-th period will be directly used as the m-th correlation parameter of the k-th period. Generate the relevant parameter vector x Fr (k);

[0061] Starting from the kth period, the window is shifted N times in reverse order along the period. The time series statistics and cross-correlation matrix in each window are calculated and flattened into vectors. These vectors are then concatenated with the relevant parameter vectors of the end period in each window to form the comprehensive feature vector in each window. The core feature vector set Y is then extracted and generated using a time series autoencoder.

[0062] Define the state S and use the forward-backward algorithm to maximize the log-likelihood function logp(Y|λ). Combine this with the core feature vector set Y to train the parameters λ of the Hidden Markov Model and obtain the optimal parameter set λ. * The failure probability p(S) in the k-th cycle is calculated using Bayesian inference. k =2|Y,λ * ) and compare with a probability threshold to determine whether there is a fault in the k-th cycle, where S k =2 indicates that the state in the k-th cycle is a fault;

[0063] If the error is greater than or equal to the probability threshold, it is determined that a fault has occurred in the k-th cycle and the equipment operation is stopped. Based on the error between each relevant parameter in the k-th cycle and the corresponding standard relevant parameter, the error is converted into the control quantity of the control device corresponding to each relevant parameter in the k-th cycle based on the PID control algorithm and the control is executed. If the error is less than the probability threshold, no action is taken.

[0064] Compared with existing technologies, this invention involves an acquisition module that collects undetermined relevant parameters in each cycle, uses the Akaike Information Criterion to determine the optimal order of the autoregressive model for each relevant parameter, and verifies the undetermined relevant parameters for each cycle. Based on the verification results, it decides whether to perform secondary acquisition and replacement, generating a relevant parameter vector for each cycle. The monitoring module starts from the relevant parameter vector of each cycle and moves the window in reverse order along the cycle, calculates and organizes the comprehensive feature vector within each window, and extracts the core feature vector set through a time-series autoencoder. It trains a hidden Markov model using a forward-backward algorithm and obtains the fault probability for each cycle based on Bayesian inference. The probability is compared with a probability threshold to determine whether a fault exists, thus achieving high-precision fault monitoring based on acquisition self-checking, multi-dimensional data fusion, and probabilistic inference. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of a fault probability monitoring and control system;

[0066] Figure 2 Flowchart for determining the optimal order of an autoregressive model;

[0067] Figure 3 Here is a flowchart of the timing autoencoder process;

[0068] Figure 4 This is a flowchart of a fault probability monitoring and control method. Detailed Implementation

[0069] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0070] Example 1:

[0071] like Figure 1 As shown in the figure, a specific embodiment of the present invention discloses a fault probability monitoring and control system, including a data acquisition module, a monitoring module and a control module;

[0072] The acquisition module collects undetermined relevant parameters through M main sensors in the k-th cycle, and uses the Akaike information criterion to determine the autoregressive model Ar(q) for the m-th relevant parameter. m ) optimal order Before combination The m-th correlation parameter of the k-th period is used to deduce the m-th theoretical correlation parameter of the k-th period. And verify the m-th undetermined relevant parameter in the k-th period. Based on the verification results, a decision is made on whether to activate and replace the corresponding secondary sensor for secondary acquisition, generating the relevant parameter vector x for the kth cycle. Fr (k), k=1,2,3,…,m=1,…,M;

[0073] The monitoring module obtains the relevant parameter vector x from the k-th period. Fr (k) Starting with N reverse translations of the window along the cycle, the comprehensive feature vector within each window is calculated and organized. A temporal autoencoder is used to retain fault-sensitive information and strengthen cycle correlation, generating a core feature vector set Y. A forward-backward algorithm is then used to train a hidden Markov model with the objective of maximizing the log-likelihood function logp(Y|λ). Bayesian inference is then combined to obtain the fault probability p(S) of the k-th cycle. k =2|Y,λ * ) and compare it with a probability threshold to determine whether a fault exists, where S k =2 indicates that the state in the k-th cycle is a fault, λ and λ *Let be the parameters and the optimal parameters of the Hidden Markov Model, respectively, and p(Y|λ) represent the probability of the core feature vector set Y appearing under the condition of parameter λ.

[0074] When the control module determines a fault in the k-th cycle, it stops the equipment from running. Based on the error between each relevant parameter in the k-th cycle and the corresponding standard relevant parameter, it converts the error into the control quantity of the control device corresponding to each relevant parameter in the k-th cycle based on the existing PID control algorithm and executes the control.

[0075] Furthermore, the acquisition module includes an acquisition control unit and an acquisition verification unit;

[0076] The acquisition and control unit acquires different unknown correlation parameters through M main sensors in the k-th cycle, and obtains the unknown correlation parameter vector. Based on the resampling annotation vector ξ(k)=[ξ1(k),…,ξ in the k-th cycle m (k),…,ξ M [k] Select the secondary sensor to perform secondary data acquisition in the kth cycle, and prompt the staff to replace or repair the corresponding primary sensor when the equipment is idle, and replace the vector of undetermined relevant parameters with the secondary acquisition results. The corresponding undetermined related parameters are treated as undetermined related parameters, and a related parameter vector is generated. in, and Let ξ be the m-th undetermined correlation parameter and the correlation parameter of the k-th period. m (k) represents the m-th re-acquisition annotation in the k-th period, when ξ m When (k) = 1, it indicates that the corresponding secondary sensor is activated in the k-th cycle to replace the faulty primary sensor. When ξ m When (k) = 0, it means that the secondary sensor does not need to be activated in the kth cycle. m = 1, ..., M, and the relevant parameters are the operating parameters that are closely related to the fault in the actual operation of the equipment.

[0077] The data acquisition and verification unit obtains data from the vector of undetermined relevant parameters. Extract the m-th undetermined correlation parameter of the k-th period sequentially. Traversal order q m The set of values, based on the order q of the fitting of the m-th related parameter vector before the k-th period. m The autoregressive model Ar(q) corresponding to each value m And select the optimal order that minimizes the Akaike information criterion. and the corresponding optimal autoregressive model Before the kth cycle Substituting the m-th relevant parameter of each period into the optimal autoregressive model Obtain the m-th theoretical correlation parameter of the k-th period And combined with the m-th undetermined relevant parameter of the k-th period Calculate the error of the m-th parameter in the k-th period, and determine whether the error of the m-th parameter in the k-th period is greater than or equal to the error threshold of the m-th parameter. If it is greater than or equal to the error threshold of the m-th parameter, re-acquire and label the m-th parameter ξ in the k-th period. m If (k) is set to 1, and it is less than the error threshold of the m-th parameter, then the m-th re-acquisition label ξ in the k-th period will be re-acquired and labeled. m (k) is set to 0, and the re-acquisition labels of the M related parameters in the k-th period are arranged into the re-acquisition label vector ξ(k) of the k-th period and fed back to the acquisition control unit.

[0078] like Figure 2 As shown, further, the acquisition and verification unit determines the optimal order of the m-th relevant parameter. and the corresponding optimal autoregressive model Includes the following steps:

[0079] The order q of the m-th related parameter m Treat it as a variable and iterate through the order q m The set of possible values;

[0080] Based on order q m A variational autoregressive model Ar(q) is constructed using a single value. m That is, the m-th relevant parameter in each cycle is equal to the q-th parameter in the previous cycle. m The weighted sum of the m-th relevant parameter over each period;

[0081] Call q from the kth cycle onwards m The m-th relevant parameter in the +1 period is obtained by adjusting the variational autoregressive model Ar(q) using the least squares method. m ) of q m The autoregressive coefficients make q before the (k-1)th period m The weighted sum of the m-th correlation parameter of each period and the m-th correlation parameter x of the (k-1)-th period. m By minimizing the error of (k-1), the variational autoregressive model Ar(q) is obtained. m The autoregressive coefficient vector;

[0082] Calculate the order q m A variational autoregressive model Ar(q) with a single value m Akaike Information Criterion AIC(q) m The specific formula is as follows:

[0083]

[0084] in, For the variational autoregressive model Ar(q) m The calculated m-th theoretical correlation parameter for the (k-1)-th period;

[0085] The optimal order that minimizes the Akaike information criterion is selected. Obtain the corresponding optimal autoregressive model

[0086] Furthermore, the monitoring module includes a feature processing unit and a fault determination unit;

[0087] The feature processing unit initializes the end period of the window to the k-th period and shifts it N times in reverse order along the period, moving forward one period each time. It calculates the time series statistics and cross-correlation matrix within each window and flattens them into vectors. These vectors are then concatenated with the relevant parameter vectors of the end period within each window to generate a comprehensive feature vector for each window. The time series autoencoder retains the fault-sensitive information of the comprehensive feature vector for each window and captures period correlations, generating a core feature vector set.

[0088] The fault determination unit defines state S and uses a forward-backward algorithm to maximize the log-likelihood function logp(Y|λ). It then trains a hidden Markov model using the core feature vector set Y to obtain the optimal parameter set λ. * The failure probability p(S) in the k-th cycle is calculated using Bayesian inference. k =2|Y,λ * The result is compared with the probability threshold. If it is greater than or equal to the probability threshold, the device is determined to have failed in the k-th cycle. If it is less than the probability threshold, no action is taken.

[0089] Specifically, the end period of the nth window is the (k-n+1)th period. Since the window size is fixed at N0, the nth window includes the relevant parameter vector from the (kn-N0+2)th period to the (k-n+1)th period. The mth relevant parameter within the N0th period of the nth window is obtained, and the time-series statistics of the mth relevant parameter in the nth window are calculated, including the window mean μ of the mth relevant parameter in the nth window. n,m Window standard deviation σ n,m Maximum window value Minimum window Window skewness Sk n,m Window kurtosis Ku n,m and the lag autocorrelation Af n,m Among them, window skewness Sk n,m Window kurtosis Ku n,m and the lag autocorrelation Af n,m The specific formula is as follows:

[0090]

[0091] Among them, window skewness Sk n,m and window kurtosis Ku n,m Reflecting the symmetry and tail thickness of the numerical distribution of the m-th relevant parameter within the n-th window, respectively, hysteresis-autocorrelation Af n,m This is used to measure the linear correlation between the m-th relevant parameter within the n-th window and the time series, and to calculate the m1-th relevant parameter within the n-th window. and the m2th related parameter covariance Divide by the m1th relevant parameter Window standard deviation With the m2th related parameter Window standard deviation The product of these components yields the cross-correlation matrix ρ of the nth window. n Cross-correlation coefficient of row m1 and column m2 From the cross-correlation matrix ρ of the nth window n Extract the upper triangular elements containing diagonal elements, the time series statistics of the M related parameters of the nth window, and the related parameter vector of the (k-n+1)th period. The vectors are concatenated column-wise and then normalized using Z-Score to generate the comprehensive feature vector for the nth window. Arrange the comprehensive feature vectors of N windows in periodic order to obtain the comprehensive feature vector set. Since the windows are moved in reverse order, the Nth window has the earliest end period.

[0092] like Figure 3 As shown, the temporal autoencoder further includes a bidirectional encoding layer, an attention cohesion layer, and a reconstruction verification layer;

[0093] The bidirectional encoding layer utilizes parallel forward long short-term memory (LSTM) and backward long short-term memory (LSTM) networks to process the integrated feature vector set, respectively. The forward long short-term memory network discards the forward cell state of the (n+1)th window according to the periodic order using the forget gate. Part of the information is obtained from the comprehensive feature vector of the nth window using the input gate. With the forward hidden state of the (n+1)th window Select a portion of the information and map it to the forward cell state of the nth window. The output gate is used to concatenate the processing results of the forget gate and the input gate, and then mapped to generate the forward hidden state of the nth window. Repeat this process until the forward cell state of the first window is generated. and forward hidden state When the loop stops, the backward Long Short-Term Memory network processes data in reverse order, starting from the first window and repeatedly using forget gates, input gates, and output gates until the backward cell state of the Nth window is generated. and backward hidden state Stop when the timer stops. Concatenate the forward and backward hidden states of the same window to generate the hidden states of N windows, and arrange them in periodic order to form a hidden state sequence.

[0094] The attention cohesion layer hides the state sequence. The hidden state h of the nth window n The attention score for the nth window is obtained by linear modulation and Tanh function mapping to the range [-1, 1], and then multiplied by a preset attention vector. The attention score vectors are arranged in cyclical order. At And it is normalized to the attention weight vector w using the Softmax function. At The attention weight vector w At With hidden state sequence The attention weights of each window are multiplied by the hidden states and summed to generate the context hidden state h. Ag Hide the context state h Ag The hidden states of each window are concatenated and mapped to the core feature vectors of each window using linear modulation and the ReLU function to generate a core feature vector set.

[0095] The reconstruction verification layer will use the core feature vector y of the nth window in the core feature vector set Y. n Dimensionality is expanded through linear modulation and nonlinear expression is enhanced using the ReLU function, thus generating the reconstructed comprehensive feature vector for the nth window. Generate a reconstructed comprehensive feature vector set Computational reconstruction of the comprehensive feature vector set With integrated feature vector set The reconstruction error is compared with a reconstruction error threshold. If the reconstruction error is less than the reconstruction error threshold, the core feature vector set Y is deemed qualified. If the reconstruction error is greater than or equal to the reconstruction error threshold, the core feature vector set Y is deemed over-condensed to the point of losing the comprehensive feature vector set. The fault-sensitive information in the core feature vector set Y makes it impossible to effectively reconstruct the data. Therefore, the parameters of the feature embedding layer, bidirectional encoding layer, and attention cohesion layer are updated in reverse based on the reconstruction error threshold, and the comprehensive feature vector set is reprocessed.

[0096] Furthermore, calculate the failure probability p(S) in the k-th cycle. k =2|Y,λ *This includes the following steps:

[0097] Define the state S = {1, 2}, where S = 1 and S = 2 represent the state S as normal and faulty, respectively;

[0098] Initialize the state probability vector π of the Nth window N State probability vector π N This includes the probabilities of state S being normal and faulty, where the Nth window corresponds to the (k-N+1)th period, which is the earliest period in the period sequence;

[0099] Initialize the state transition matrix A and the observation probability matrix B. The state transition matrix A, row j1, column j2, A(j1,j2), reflects the probability of transitioning from state S = j1 to state S = j2. The observation probability matrix B, row j1, column n, B(j1,n), represents the core feature vector of the nth window appearing in state S = j1. The probability, j1,j2∈{1,2};

[0100] The parameter set λ of the Hidden Markov Model is defined as including the state probability vector π of the Nth window. N State transition matrix A and observation probability matrix B;

[0101] Define the state S of the nth window. k-n+1 =j1 and the probability of observing the core feature vectors from the Nth window to the (n+1th)th window is the forward variable α. n (j1), and the forward variable α n (j1) satisfies the forward recursion formula, as follows:

[0102]

[0103] According to the forward recursion formula, the forward variable α n (j1) can be rearranged into the forward variable α N (j3) is a linear combination of the elements in the state transition matrix A and the observation probability matrix B, with the forward variable α N (j3) is as follows:

[0104] α N (j3)=π N (S k-N+1 =j3)·B(j3,N),

[0105] Among them, S k-N+1 =j3 reflects the state of the Nth window in the periodic sequence, π N (S k-N+1 =j3) is the state probability vector π of the Nth window. N Mid-state S k-N+1= probability j3, j3∈{1,2};

[0106] Define the state S of the nth window. k-n+1 =j1 and the probability of observing the core feature vectors from the (n-1)th window to the 1st window is the backward variable β. n (j1), and the backward variable β n (j1) satisfies the backward recursion formula, as follows:

[0107]

[0108] According to the backward recursion formula, the backward variable β n (j1) can be rearranged into a linear combination of the backward variable α1(j3) and the elements in the state transition matrix A and the observation probability matrix B, where the backward variable α1(j3) = 1 and j3 ∈ {1,2}. This is because the k-th period is the end period in the period sequence, and the state and core feature vector of each window have been determined.

[0109] Define the state S of the nth window. k-n+1 The probability of j1 is the single-state probability γ. n (j1), as follows:

[0110]

[0111] Single-state probability γ n (j1) reflects the state S of the nth window. k-n+1 Expectations;

[0112] Define the state S of the nth window. k-n+1 =j1 and the state S of the (n-1)th window k-n+2 The probability of j2 is the joint probability γ of the two states. n (j1,j2), as follows:

[0113]

[0114] The joint probability γ of the two states n (j1,j2) reflects the state S from the nth window. k-n+1 The state S transitions to the (n-1)th window k-n+2 Expectations;

[0115] With the objective of maximizing the log-likelihood function logp(Y|λ), the parameter set λ of the Hidden Markov Model is updated in each iteration until the change in the parameter set between adjacent iterations is less than the convergence threshold. The parameter set λ of the last iteration is then taken as the optimal parameter set λ. * , where, p(Y|λ)=α1(1)+α1(2)=β N (1)+βN (2);

[0116] Combining the optimal parameter set λ * The optimal state probability vector π of the Nth window in * N Using the optimal state transition matrix A * and the optimal observation probability matrix B * By performing forward recursion, the failure probability p(S) of the k-th cycle is obtained. k =2|Y,λ * ).

[0117] Furthermore, the parameter set λ of the Hidden Markov Model is updated in the (l+1)th round as follows:

[0118] The state probability vector of the Nth window in the (l+1)th round Mid-state S k-N+1 The probability of =j1 Updated to be based on the parameter set λ of round l. l Calculated single-state probability

[0119] The state transition matrix A of the (l+1)th round l+1 A in row j1 and column j2 l+1 (j1,j2) is updated based on the parameter set λ of the l-th round. l The ratio of the expected number of transitions from state S=j1 to state S=j2 to the total expected number of transitions from state S=j1 is calculated as follows:

[0120]

[0121] in, and Based on the parameter set λ of the lth round respectively l Calculate the joint probability of the two states and the single state probability of the nth window;

[0122] Update the observation probability matrix B for the (l+1)th round based on the maximum likelihood estimation of the Gaussian distribution. l+1 B in row j1 and column n l +1 (j1,n), specifically as follows:

[0123] B l+1 (j1,n)=Gauss(μ l (j1,n),Σ l (j1,n)),

[0124] Where Gauss represents a Gaussian distribution, μ l (j1,n) and Σ l(j1,n) represent the parameter set λ based on the l-th round, respectively. l The state S of the nth window is calculated. k-n+1 The mean and variance of the distribution of j1 are calculated using the following formulas:

[0125]

[0126] Example 2

[0127] like Figure 4 As shown, a specific embodiment of the present invention discloses a fault probability monitoring and control method, which is executed based on the fault probability monitoring and control system and includes the following steps:

[0128] In the k-th cycle, different undetermined correlation parameters are collected by M main sensors to obtain the undetermined correlation parameter vector. The optimal order that minimizes the Akaike Information Criterion is determined by fitting the m-th relevant parameter prior to the k-th cycle. and the corresponding optimal autoregressive model

[0129] Before the kth period Substituting the m-th relevant parameter of each period into the optimal autoregressive model Obtain the m-th theoretical correlation parameter of the k-th period Combined with the m-th undetermined relevant parameter of the k-th period Calculate the error of the m-th parameter in the k-th period and compare it with the error threshold of the m-th parameter;

[0130] If the error is greater than or equal to the error threshold of the m-th parameter, then the secondary sensor is activated to perform a second data acquisition in the k-th cycle, and the acquisition result is used as the m-th related parameter in the k-th cycle. If the error is less than the threshold of the m-th parameter, then the m-th undetermined correlation parameter of the k-th period is directly used as the m-th correlation parameter of the k-th period. Generate the relevant parameter vector x Fr (k);

[0131] Starting from the kth period, the window is shifted N times in reverse order along the period. The time series statistics and cross-correlation matrix in each window are calculated and flattened into vectors. These vectors are then concatenated with the relevant parameter vectors of the end period in each window to form the comprehensive feature vector in each window. The fault-sensitive information of the comprehensive feature vector of each window is retained and the period correlation is captured using a time series autoencoder to generate the core feature vector set Y.

[0132] Define the state S and use the forward-backward algorithm to maximize the log-likelihood function logp(Y|λ). Combine this with the core feature vector set Y to train the parameters λ of the Hidden Markov Model and obtain the optimal parameter set λ.* The failure probability p(S) in the k-th cycle is calculated using Bayesian inference. k =2|Y,λ * And compare it with the probability threshold, where S k =2 indicates that the state in the k-th cycle is a fault;

[0133] If the error is greater than or equal to the probability threshold, the equipment is determined to have malfunctioned in the k-th cycle and the equipment operation is stopped. Based on the error between each relevant parameter in the k-th cycle and the corresponding standard relevant parameter, the PID control algorithm is used to convert the error into the control quantity of the control equipment corresponding to each relevant parameter in the k-th cycle and the control is executed. If the error is less than the probability threshold, no action is taken.

[0134] Example 3:

[0135] As an example, the fault probability monitoring and control system interfaces with a pressure sensor deployed on the filter media of the intelligent filter machine via a PLC controller. In the k-th cycle, the main pressure sensor collects the differential pressure of the filter media to be determined. The optimal autoregressive model is fitted and verified to decide whether to activate the secondary pressure sensor to collect data again. The differential pressure of the filter media in the k-th cycle and the previous N-1 cycles is integrated and calculated using a reverse-shifted window to obtain a comprehensive feature vector. The core feature vector set Y is generated by processing the data using a time-series autoencoder. A hidden Markov model is trained and the probability of filter media blockage in the k-th cycle is calculated based on Bayesian inference. When blockage is determined to occur in the k-th cycle, the cleaning flow rate and cleaning time of the cleaning component are determined by a PID control algorithm based on the differential pressure of the filter media and the standard differential pressure of the filter media when there is no blockage. That is, when applied to the filter media of the intelligent filter machine, the relevant parameter is the differential pressure of the filter media, the fault is the blockage of the filter media, the control device is the cleaning device, and the control quantities include the cleaning flow rate and the cleaning time.

[0136] Example 4:

[0137] As an example, the fault probability monitoring and control system interfaces with the temperature and speed sensors deployed on the intelligent filter via a PLC controller. In the k-th cycle, the system collects the undetermined temperature and motor speed through the main temperature and main speed sensors. It then fits and determines the optimal autoregressive model for verification to decide whether to activate the corresponding secondary sensors for re-collection. Using a reverse-shifted window, it integrates and calculates the temperature and motor speed of the k-th cycle and the preceding N-1 cycles to obtain a comprehensive feature vector. This vector is then processed using a time-series autoencoder to generate a core feature vector set Y. A hidden Markov model is trained, and based on Bayesian inference, the probability of abnormal operation of the intelligent filter in the k-th cycle is calculated. If an abnormality is determined, based on the error between the temperature and the standard temperature in the k-th cycle, and the error between the motor speed and the standard motor speed, a PID control algorithm is used to determine the power adjustment amount of the temperature control device and the motor, respectively. In other words, when applied to an intelligent filter, the relevant parameters include temperature and motor speed, the fault is abnormal operation, the control devices are the temperature control device and the motor, and the control amount includes the power adjustment amount of the temperature control device and the motor.

[0138] This invention discloses a fault probability monitoring and control system and method, including an acquisition module, a monitoring module, and a control module. The acquisition module acquires undetermined relevant parameters in each cycle, uses the Akaike Information Criterion to determine the optimal order of the autoregressive model for each relevant parameter, and verifies the undetermined relevant parameters for each cycle. Based on the verification results, it decides whether to perform secondary acquisition and replacement, generating a relevant parameter vector for each cycle. The monitoring module starts from the relevant parameter vector of each cycle and moves the window backward along the cycle, calculates and organizes the comprehensive feature vector within each window, and extracts the core feature vector set through a time-series autoencoder. It trains a hidden Markov model using a forward-backward algorithm and obtains the fault probability for each cycle based on Bayesian inference. The probability is compared with a probability threshold to determine whether a fault exists. When a fault exists, the control module activates the corresponding control equipment for control based on a PID control algorithm, realizing high-precision fault monitoring and control based on acquisition self-checking, multi-dimensional data fusion, and probabilistic inference.

[0139] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A fault probability monitoring and regulation system comprising a regulation module that stops the operation of a device upon determining a device fault and performs regulation based on a PID control algorithm, characterized in that, Also comprising a collection module and a monitoring module; The acquisition module is in the first Collection in each cycle The first undetermined relevant parameter is determined using the Akaike Information Criterion. The optimal order of the autoregressive model with the relevant parameters is calculated to deduce the th... The first cycle The theoretically relevant parameters were verified. The first cycle Based on the verification results, a decision is made regarding whether to perform a second data collection and replacement of the undetermined relevant parameters, generating the first... The relevant parameter vector for each period, , ; The monitoring module starts from the first The relevant parameter vector for each period begins to shift the window in reverse periodic order. Next, the comprehensive feature vector within each window is calculated and organized, and the core feature vector set is generated by refining it through a temporal autoencoder. A hidden Markov model is trained using a forward-backward algorithm, and the result is obtained based on Bayesian inference. The failure probability of each cycle is compared with the probability threshold to determine whether a failure exists. The collection module comprises a collection control unit and a collection verification unit The acquisition control unit acquires different to-be-determined related parameters through the first main sensor in the first cycle to obtain a to-be-determined related parameter vector, selects a secondary sensor based on the re-sampling and labeling vector of the first cycle to perform secondary acquisition in the second cycle, replaces corresponding to-be-determined related parameters in the to-be-determined related parameter vector with the secondary acquisition result, regards the to-be-determined related parameters that are not replaced as to-be-determined related parameters, and generates a related parameter vector, wherein the related parameter is an operating parameter that is closely related to a fault in actual operation of the equipment. The data acquisition and verification unit sequentially extracts the first relevant parameter from the vector of undetermined parameters. The first cycle There are several undetermined related parameters, and the order of traversal is... The set of values, based on the first The first cycle before The order of vector fitting for each relevant parameter For each value, an autoregressive model is selected, and the optimal order that minimizes the Akaike Information Criterion is chosen. and the corresponding optimal autoregressive model, the first One cycle ago The first cycle Substituting the relevant parameters into the optimal autoregressive model yields the first... The first cycle The relevant theoretical parameters, combined with the first theoretical parameter The first cycle The calculation of the first undetermined relevant parameter The first cycle The error of the parameter is used to determine the first parameter error. The first cycle Is the error of each parameter greater than or equal to the first parameter? If the error threshold of the parameter is greater than or equal to the first parameter, then... The error threshold of the first parameter will be... The first cycle The individual weight collection label is set to 1; if it is less than the first... The error threshold of the first parameter will be... The first cycle Set the individual collection label to 0, and the first Each cycle The re-collection and labeling of the relevant parameters are arranged as follows: The resampled and labeled vectors are fed back to the acquisition and control unit for each cycle; wherein the first cycle computes a summary feature vector for the first window includes the following steps: Get the Within each window The first cycle The relevant parameters are calculated respectively. The first window The time series statistics of the relevant parameters, including the first one. The first window Window mean of each relevant parameter Window standard deviation Maximum window value Minimum window value Window skew Window kurtosis and lag autocorrelation Among them, window skewness Window kurtosis and lag autocorrelation The specific formula is as follows: ， ， ， wherein, is the first periodic parameter, is the first , is the window size; Calculate the first The first window The relevant parameters and the first The covariance of the relevant parameter divided by the first parameter The window standard deviation of the first relevant parameter and the second The product of the window standard deviations of the relevant parameters yields the product of the nth parameter. In the cross-correlation matrix of the nth window Line 1 The cross-correlation coefficient of the columns, starting from the first... Extract the upper triangular elements containing diagonal elements from the cross-correlation matrix of the nth window, and compare them with the nth window. window The time series statistics of the relevant parameters and the first The relevant parameter vectors of the nth period are concatenated column-wise and then normalized using Z-Score to generate the nth period. The combined feature vector of each window, .

2. The failure probability monitoring and regulation system of claim 1, wherein, The monitoring module comprises a feature processing unit and a fault determination unit; The feature processing unit initializes the end period of the window as the first period and shifts the window by one period in reverse order times, calculates the timing statistics and the cross-correlation coefficient matrix in each window and flattens them into vectors, splices the relevant parameter vectors of the end period in each window to generate the corresponding comprehensive feature vector, retains the fault-sensitive information of the comprehensive feature vector of each window through the timing autoencoder and captures the period correlation to generate the core feature vector set;​ The fault determination unit defines the state, uses a forward-backward algorithm to maximize the log-likelihood function, trains a hidden Markov model using the core feature vector set to obtain the optimal parameter set, and uses Bayesian inference to calculate the first... The failure probability of the i-th cycle is compared with the probability threshold if and only if the i-th cycle... When the failure probability of the first cycle is greater than or equal to the probability threshold, the first cycle is determined. There is a fault in one cycle; wherein calculating the failure probability for the first period comprises the steps of: Definition of states , and represent the states normal and fault, respectively; Initialize the first The state probability vector of each window State probability vector Including status Let be the probabilities of normal operation and failure, where is the probability of failure. The window corresponds to the first One cycle; Initialize the state transition matrix and observation probability matrix , where the state transition matrix The Line 1 List Reflected by state Transition to state The probability of observation, the probability matrix The Line 1 List Indicates the state When the first The core feature vector of each window The probability, ; Defining a parameter set for a hidden markov model Including a state probability vector Of a first Window And a state transition matrix And an observation probability matrix Definition of the state of the first window and the core feature vector of the first window and the core feature vector of the first window and the core feature vector of the first window and the core feature vector of the first window and the core feature vector of the first window and the core feature vector of the first window ， The forward variables are arranged in accordance with the forward recursion formula into forward variables and the state transition matrix and the observation probability matrix The forward variables are arranged as follows: ， wherein, reflecting the state of the first window in the sequence of cycles, reflecting the state of the first window in the sequence of cycles, reflecting the probability of the state of the first window in the sequence of cycles, ;​​​ Definition of the state of the first window Definition of the state of the first window Definition of the state of the first window The probability of observing the first window to the core feature vector of the first window is a backward variable The backward variable satisfies the backward recursive formula, as follows: ， According to the backward recursion formula, the backward variable is arranged as the backward variable is a linear combination of elements in the state transition matrix and the observation probability matrix , the backward variable , ; Definition of the first The state of each window The probability is the single-state probability. The details are as follows: ， Single-state probability Reflecting on the first The state of each window Expectations The core feature vector set; Definition of the first The state of each window And the first The state of each window The probability is the joint probability of the two states. The details are as follows: ， Dual-state joint probability reflecting the state of the first window transitioning to the state of the first window ​ maximizing a log-likelihood function updating the parameter set of the hidden Markov model in each iteration until the parameter set of the adjacent iteration changes by an amount less than a convergence threshold, the parameter set of the last iteration as the optimal parameter set wherein ; optimal parameter set the optimal state probability vector of the first window in the optimal state transition matrix and the optimal observation probability matrix is performed forward recursion to obtain the failure probability of the first cycle.

3. The failure probability monitoring and regulation system of claim 2, wherein, The time series autoencoder comprises a bidirectional encoding layer and an attention aggregation layer; The bidirectional coding layer utilizes forward and backward long short-term memory networks to process the comprehensive feature vector set separately. The forward long short-term memory network processes the first... The comprehensive feature vector of the first window, the first The forward cell state and forward hidden state of the nth window are mapped as follows: The forward cell state and forward hidden state of each window are processed cyclically until the forward cell state and forward hidden state of the first window are generated, at which point the process stops. The backward long short-term memory network processes the forward cell state and forward hidden state of the first window in reverse chronological order, starting from the first window and processing until the forward cell state and forward hidden state of the second window are generated. Stop when the window reaches its backward cell state and backward hidden state, and stitch together the first window. The forward and backward hidden states of each window generate corresponding hidden states, which are then organized into a hidden state sequence. The attention aggregation layer multiplies the linear modulation of the hidden state of the first window by the attention vector after mapping through a Tanh function to obtain the attention score of the first window, and normalizes the attention score of the first window through a Softmax function to obtain the attention weight of the first window. The attention aggregation layer multiplies the linear modulation of the hidden state of the first window by the attention vector after mapping through a Tanh function to obtain the attention score of the first window, and normalizes the attention score of the first window through a Softmax function to obtain the attention weight of the first window. The attention aggregation layer multiplies the linear modulation of the hidden state of the first window by the attention vector after mapping through a Tanh function to obtain the attention score of the first window, and normalizes the attention score of the first window through a Softmax function to obtain the attention weight of the first window. The attention aggregation layer multiplies the linear modulation of the hidden state of the first window by the attention vector after mapping through a Tanh function to obtain the attention score of the first window, and normalizes the attention score of the first window through a Softmax function to obtain the attention weight of the first window. The attention aggregation layer multiplies the linear modulation of the hidden state of the first window by the attention vector after mapping through a Tanh 4. The failure probability monitoring and regulation system of claim 3, wherein, The time series autoencoder further comprises a reconstruction verification layer; The reconstruction verification layer will set the core feature vectors as follows: The core feature vector of the first window is generated by mapping linear modulation and the ReLU function. The reconstructed comprehensive feature vectors of each window are organized into a reconstructed comprehensive feature vector set. The reconstruction error between the reconstructed comprehensive feature vector set and the comprehensive feature vector set is calculated and compared with the reconstruction error threshold. If and only if the reconstruction error is greater than or equal to the reconstruction error threshold, the parameters of the feature embedding layer, bidirectional coding layer and attention cohesion layer are updated in reverse based on the reconstruction error threshold and the comprehensive feature vector set is reprocessed.

5. The failure probability monitoring and regulation system of claim 1, wherein, determining the optimal order of the number of relevant parameters and the corresponding optimal autoregressive model comprises the steps of: traversing the order of the set of values based on the order of the single value to build a varying autoregressive model, i.e. each th correlation parameter of a period equals a weighted sum of the th correlation parameter of the periods preceding it. Get the One cycle ago The first cycle The variational autoregressive model is obtained by fitting the relevant parameters using the least squares method. One autoregressive coefficient; Computing the order Akaike's information criterion for the individual value of the VAR model The optimal order is selected as the order with the minimum Akaike information criterion and the corresponding optimal autoregressive model is obtained.

6. The method of failure probability monitoring and regulation, performed by the failure probability monitoring and regulation system according to any one of claims 1-5, characterized in that, Comprising the following steps: In the first cycle, the to-be-determined correlation parameter vector is acquired, and the best order and the corresponding best autoregressive model are determined based on the Akaike information criterion minimum of the first correlation parameter fitting before the first cycle. The first One cycle ago The first cycle Substituting the relevant parameters into the optimal autoregressive model yields the first... The first cycle The relevant theoretical parameters, combined with the first... The first cycle The calculation of the first undetermined relevant parameter The first cycle The parameter error and the first Comparison of error thresholds for each parameter; If greater than or equal to the first Each parameter error threshold is used for secondary data acquisition, and the acquisition result is taken as the first... The first cycle If the relevant parameter is less than the first one, The error threshold of the first parameter will be... The first cycle The undetermined relevant parameter is directly used as the first... The first cycle One relevant parameter is used to generate a relevant parameter vector; Move the window from the first Starting from the cycle, shift the single cycle in reverse order. Next, calculate the temporal statistics and cross-correlation matrix in each window and flatten them into vectors. Then, concatenate them with the relevant parameter vector of the end period in each window to form the comprehensive feature vector in each window. Use the temporal autoencoder to extract and generate the core feature vector set. Defining states And use forward-backward algorithm to maximize the log-likelihood function as the goal, combined with the core feature vector set training hidden Markov model to get the optimal parameter set, using Bayesian inference to calculate the probability of failure of the period and compared with the probability threshold to determine whether the period is failure; If the probability is greater than or equal to the probability threshold, a fault is determined to have occurred and the equipment operation is stopped, according to the... The error between each relevant parameter and the corresponding standard relevant parameter in each cycle is transformed into the error of the first cycle based on the PID control algorithm. The control device adjusts the amount of each relevant parameter in each cycle and performs the control. If the amount is less than the probability threshold, no action is taken.

Images