Process industry anomaly root diagnosis method based on causal circulation loop removal

Abstract

The invention provides a process industry anomaly root diagnosis method based on causal circulation loop removal, and relates to the technical field of fault diagnosis. The method comprises the following steps: offline modeling: calculating variable time delay in real time by adopting a method of combining dynamic time warping and a sliding window technology, and constructing a variable time delay matrix in a normal state; performing trend analysis on the variable time delay in a normal state, decomposing a circulation loop in the causal network into a time causal relationship, and removing the circulation loop existing in the causal network; establishing a Bayesian network based on variable time delay, and determining a conditional probability table in a normal state; and online diagnosis: constructing a variable time delay matrix in an abnormal state, updating the Bayesian network, determining a conditional probability table in the abnormal state, and comparing the conditional probability tables in the normal state and the abnormal state to locate an abnormal variable and deduce a source and a propagation path. In this way, the accuracy of dynamic process industrial anomaly root diagnosis can be improved.

Description

Technical Field

[0001] The invention relates to the field of fault diagnosis, and in particular to a method for diagnosing the root cause of anomalies in a process industry based on the elimination of a causal cycle. Background Art

[0002] Process industries are an important part of the manufacturing industry, and their safe and efficient operation is crucial to economic development. With the continuous advancement of technology, today's process industries are characterized by the ever-increasing scale and complexity of production. In these systems, the emergence of anomalies may trigger a cascading effect in various process units, posing a major challenge to maintenance teams. In order to mitigate or prevent the adverse consequences of such anomalies, it is necessary to identify the most likely anomaly propagation path, diagnose the root cause of the anomaly, and then take appropriate corrective measures to restore normal operation as soon as possible.

[0003] In process industries, causal cyclic loops are ubiquitous and unavoidable due to the existence of feedback control mechanisms. Causal cyclic loops (referred to as cyclic loops) are circular causal relationships caused by factors such as variable coupling, cyclic flow, and integration of materials and thermal processes in a causal network. The existence of these loops, whether from the physical structure of the system or the time-lag characteristics of the production process, will complicate the causal network and increase the uncertainty of abnormal diagnosis if not handled. Therefore, effectively handling these cyclic loops is crucial for accurate abnormal root cause diagnosis, which is a technology used to identify and trace the causes of abnormalities in the production process, and is essential to ensure the stability of the production process and improve product quality and safety.

[0004] In order to meet the above challenges, abnormal root cause diagnosis technology came into being. The concept of abnormal root cause diagnosis can be traced back to the 1990s, when researchers mainly relied on abnormal diagnosis or process mechanism knowledge to infer the root cause of abnormalities. However, due to the high cost of knowledge collection and complex logical operations, the abnormal root cause diagnosis technology at that time failed to be popularized in the complex process industry. In recent years, with the development and application of research such as machine learning and causal analysis, data-driven abnormal root cause diagnosis methods have brought new solutions to the intelligent diagnosis of complex processes and promoted the development of related research.

[0005] Generally speaking, abnormal root cause diagnosis methods can be divided into knowledge-based methods and data-based methods. Knowledge-based methods usually rely on connectivity or correlation and use piping and instrumentation diagrams, expert knowledge, and algebraic equations to construct process topology. These mainly include methods such as symbolic directed graphs, adjacency matrices, and fault trees, which are mostly used in chemical industrial processes with smaller scales and simpler operating mechanisms. These methods rely heavily on prior knowledge without considering the statistical information of real-time data. Therefore, they cannot explore the causal relationship between variables and are difficult to adapt to dynamic changes in process industries. With the accumulation and storage of a large amount of process data, data-based abnormal root cause diagnosis methods have developed rapidly.

[0006] Data-driven methods are more popular than traditional knowledge-based methods. This preference stems from their superior ability to analyze time series data, which is essential for revealing often obscured temporal dynamic characteristics. Data-driven methods are divided into deterministic methods and probabilistic methods. Deterministic data-driven methods are good at extracting features from process data sets, such as principal component analysis, independent component analysis, and canonical variable analysis. They determine whether anomalies occur by identifying patterns that deviate from normal operating conditions, and then infer the location of the anomaly. However, such methods usually require the screening of abnormal variable candidate sets in advance in abnormal root cause diagnosis in order to accurately identify the abnormal root cause. Therefore, this method faces many challenges in complex process industries, such as the mutual propagation of the influence between abnormal variables, resulting in the generation of smearing effects, that is, in the screening process of abnormal variable candidate sets, the existence of certain variables may mask the true causal relationship between other variables and anomalies. In response to this, probabilistic methods led by Bayesian networks (BN) have become the mainstream abnormal root cause diagnosis methods.

[0007] Unlike deterministic methods, probabilistic methods can better handle uncertainty and complexity, thereby improving the accuracy and reliability of diagnosis. BN can capture the dynamic relationship between variables by establishing a probabilistic relationship model between variables, thereby more accurately identifying potential abnormal root causes. However, BN is essentially acyclic, and various feedback mechanisms often exist in process industries, which may lead to the emergence of loops. Such loops make it more difficult to diagnose the root causes of abnormalities and identify propagation paths, because they may cause changes in the probability distribution in the model, thereby affecting the accuracy of the diagnostic results. Therefore, it is necessary to deal with the loops existing in the causal network accordingly in order to diagnose the root causes of abnormalities more accurately.

[0008] In order to deal with the ubiquitous loops in the process industry, methods such as virtual nodes, super nodes, and hierarchical diagnosis can be used to break the loops, or variables that directly cause loops or bidirectional edges can be omitted in the causal graph to eliminate the loops. Although these methods have shown effectiveness, they are limited to analyzing surface causal relationships and ignore the potential time priority in fault information. At the same time, for most process industries, the assumption of fixed and unchanging time delays between causal relationships is too rigid. In fact, time delays can change dynamically as the system operates, which violates the basic assumption of causal analysis models. Ignoring the variability of time delays may lead to inaccurate causal reasoning and pose a major obstacle to effective abnormal root cause diagnosis. Therefore, how to fully explore and use the deep information in the process industry operation data, improve the existing causal analysis model, and realize abnormal root cause diagnosis and propagation path identification has gradually become a scientific problem that needs to be studied urgently in the industry and academia. Summary of the invention

[0009] In order to solve the technical problem that the variability of time delay is ignored in the prior art, resulting in low accuracy in the diagnosis of the root causes of abnormalities in the process industry, an embodiment of the present invention provides a method for diagnosing the root causes of abnormalities in the process industry based on the elimination of causal loops. The technical solution is as follows:

[0010] On the one hand, a method for diagnosing the root cause of abnormality in a process industry based on the elimination of a causal loop is provided. The method is implemented by a device for diagnosing the root cause of abnormality in a process industry. The method comprises:

[0011] Offline Modeling:

[0012] Step S1, for process variable data generated under normal conditions in the process industry production process, a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under normal conditions is constructed;

[0013] Step S2, performing trend analysis on the variable time delay under normal conditions, decomposing the loops in the causal network into temporal causal relationships, and eliminating the loops existing in the causal network;

[0014] Step S3, establishing a Bayesian network based on variable time delay to determine a conditional probability table under a normal state;

[0015] Online Diagnosis:

[0016] Step S4, for process variable data generated under abnormal conditions in the process industry production process, a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under abnormal conditions is constructed;

[0017] Step S5, based on the constructed variable time delay matrix under abnormal state, update the Bayesian network, determine the conditional probability table under abnormal state, compare the conditional probability tables under normal and abnormal states to locate abnormal variables and infer the root cause and propagation path.

[0018] On the other hand, a device for diagnosing the root cause of anomalies in a process industry is provided, and the device for diagnosing the root cause of anomalies in a process industry comprises: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned methods for diagnosing the root cause of anomalies in a process industry based on the resolution of causal loops is implemented.

[0019] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned process industry abnormality root cause diagnosis methods based on causal loop release.

[0020] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0021] 1) Aiming at the problem of dynamic time delay in complex process industries, a method combining dynamic time warping and sliding window technology is used to realize real-time calculation of variable time delay and construct a variable time delay matrix to reveal the dynamic changes of time information;

[0022] 2) Aiming at the circular causal loop problem induced by factors such as variable coupling, cyclic flow, and integration of materials and thermal processes, a strategy for decomposing the circular loops in the causal network is designed by deeply mining the temporal information between process variables and combining the Mann-Kendall trend test. By analyzing the changing trend of variable time delays, the inherent causal loops are decomposed into temporal causal relationships, thereby eliminating the circular loops existing in the causal network.

[0023] 3) Based on the constructed variable time delay matrix, the traditional Bayesian network model is improved, and the limitation of the Bayesian network model on the fixed time delay assumption is lifted, so that it can handle the variable time delay between abnormal variables and improve the accuracy of abnormal root cause diagnosis in dynamic process industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1It is a flow chart of a method for diagnosing the root cause of abnormalities in process industries based on the elimination of causal loops provided by an embodiment of the present invention;

[0026] Figure 2 It is a structural schematic diagram of a process industry abnormality root cause diagnosis device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0028] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0029] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0030] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0031] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0032] In order to solve the problem of abnormal root cause diagnosis and propagation path identification in complex process industries, the embodiment of the present invention provides a method for abnormal root cause diagnosis in process industries based on the elimination of causal loops. The method can be implemented by a process industry abnormal root cause diagnosis device, which can be a terminal or a server. Figure 1 The flowchart of the abnormal root cause diagnosis method of process industry based on causal loop release is shown, the method includes three functional modules, the first functional module is a variable time delay matrix construction module, the second functional module is a causal loop release module, and the third functional module is an abnormal root cause diagnosis module. The processing flow of the method may include the following steps:

[0033] Offline Modeling:

[0034] Step S1, for process variable data generated under normal conditions in the process industry production process (referred to as normal data), a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under normal conditions is constructed;

[0035] In this embodiment, variable time lag: since process variables include various types, the time causal relationship shows different time lags in different states and changes with time.

[0036] Step S2, performing trend analysis on the variable time delay under normal conditions, decomposing the loops in the causal network into temporal causal relationships, and eliminating the loops existing in the causal network;

[0037] Step S3, establishing a Bayesian network based on variable time delay to determine a conditional probability table under a normal state;

[0038] Online Diagnosis:

[0039] Step S4, for process variable data (abnormal data for short) generated in abnormal conditions during the process industry production process, a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under abnormal conditions is constructed;

[0040] Step S5, based on the constructed variable time delay matrix under abnormal state, calculate the log-likelihood value to update the Bayesian network, determine the conditional probability table under abnormal state, compare the conditional probability tables under normal and abnormal state to locate the abnormal process variables and infer the root cause and propagation path.

[0041] In this embodiment, step S1 and step S4 are implemented by a variable time delay matrix construction module, step S2 is implemented by a causal loop release module, and step S3 and step S5 are implemented by an abnormality root cause diagnosis module.

[0042] In this embodiment, the specific technical solutions of the three functional modules are as follows:

[0043] In this embodiment, functional module one: variable time delay matrix construction module, the specific implementation steps are:

[0044] Step 1: Collect, transmit and store process variable data such as temperature, pressure, flow, etc. generated during the process industry production process through industrial sensors and data storage devices;

[0045] Step 2: Standardize the acquired process variables and adjust the data of different scales or distributions to a unified scale;

[0046] In this embodiment, Z-score standardization is used to process variable data, and the specific form is as follows:

[0047]

[0048] Among them, x' is the original data point, μ is the mean, and σ is the standard deviation.

[0049] Step 3: Determine the time delay between process variables using a dynamic time warping (DTW) method; specifically, the following steps may be included:

[0050] Step 31: Construct the DTW distance between process variables:

[0051] First, assume that the multivariate process variable data is:

[0052]

[0053] Where X is the process variable data; m is the number of measured process variables; n is the sampling time, also represents the number of samples; X m (n) is the data collected at the nth sampling moment of the mth process variable.

[0054] The DTW distance between process variables is constructed according to the following formula:

[0055]

[0056] Where D represents the DTW distance between two process variables, d i,j Represents the data x collected at the i-th sampling time corresponding to the first process variable i The data y collected at the jth sampling time corresponding to the second process variable j The Euclidean distance between i and j is any process variable data;

[0057] Step 32: Find the best DTW path to determine the time delay:

[0058] Then, the optimal path is determined by the matrix D to minimize the cumulative DTW distance. At this time, the difference between the two process variables is the time delay, which follows the following principles:

[0059] Within the matrix grid described by the two process variables, the DTW algorithm searches for the best trajectory from the start point to the end point, which is constrained to be monotonic.

[0060] Using the continuity principle in dynamic programming, the state transition equation of the algorithm for determining the shortest path from the starting point to the end point is formulated as follows:

[0061]

[0062] Where D(i,j) is the shortest DTW distance accumulated at positions i and j.

[0063] Step 4: Update DTW in real time by using sliding window technology to calculate the optimal time delay. Specifically, the following steps may be included:

[0064] Assume that the initial window of the i-th pair of process variables is W i,h , h is the number of windows, l is the window length, which determines the update speed. Then the initial window can be expressed as:

[0065] W i,h =[w i,t ,w i,t+1 ,…w i,t+l-1 ] (1.5)

[0066] Among them, every two process variables can be called a pair of process variables, w i,t Represents the data corresponding to the i-th paired process variable at time t, window W i,h Contains all process variable data from time t to time t+l-1. The transition window can be obtained by deleting k process variable data:

[0067] W i, ' h =[w i,t+k ,w i,t+k+1 ,…w i,t+l-1 ] (1.6)

[0068] By adding k new process variable data, a new window is obtained:

[0069] W i,h+1 =[w i,t+k ,w i,t+k+1 ,…w i,t+k+l-1 ] (1.7)

[0070] Through the sliding window technology, the DTW distances corresponding to different windows can be obtained. As the window continues to move forward, the DTW distances of all windows are recorded, realizing real-time update of the DTW distance to achieve the calculation of variable time delay.

[0071] Step 5: Save the optimal time delay corresponding to each window and construct a variable time delay matrix.

[0072] In this embodiment, the variable time delay matrix is ​​expressed as:

[0073] T'=[t1,t2,…,t k ,…,t p ] (1.8)

[0074]

[0075] Where T' represents the variable time delay matrix; t p Indicates the variable time delay corresponding to the pth window, specifically: t p is a matrix containing all the time delays between every two process variables in the pth window; t k The subscript k∈(1,p), p is the number of windows; represents the DTW distance between the mth variable and the m'th variable in the Wth window, and f(·) represents the time delay corresponding to the shortest DTW distance.

[0076] In this embodiment, during offline modeling, the functional module 1 processes the process variable data generated under normal conditions in the process industry production process according to the technical solutions described in steps 1 to 5, and constructs a variable time delay matrix under normal conditions; during online diagnosis, the functional module 1 processes the process variable data generated under abnormal conditions in the process industry production process according to the technical solutions described in steps 1 to 5, and constructs a variable time delay matrix under abnormal conditions.

[0077] In this embodiment, the functional module 2: the causal loop release module, has the following specific implementation steps:

[0078] Step 1: Develop a causal network for the process industry and construct an initial BN model; where BN stands for Bayesian network;

[0079] Step 2: Determine the loop in the causal network; specifically, the following steps may be included:

[0080] Step 21: Convert the causal network into a causal adjacency matrix, where the horizontal quantity in the causal adjacency matrix represents the cause variable, the vertical quantity represents the result variable, 0 represents the absence of a causal relationship, and 1 represents the existence of a causal relationship.

[0081] Step 22: Use the depth-first search algorithm to identify all paths and determine the loops in the causal network. Specifically, the following steps may be included:

[0082] Step 221: Initialization:

[0083] Create an access mark array to mark whether each node has been visited. The number of nodes is consistent with the dimension of the causal adjacency matrix, that is, each node corresponds to the horizontal and vertical coordinates of the causal adjacency matrix.

[0084] Create an array of recursive stacks to mark whether each node is on the current recursive path;

[0085] Create an empty list to store all the found cyclic loops;

[0086] Step Two Two Two, Depth-First Search:

[0087] Start from the current node, mark it as visited, and push the node onto the recursive stack;

[0088] Traverse all adjacent nodes of the current node. If an adjacent node has not been visited, recursively call the depth-first search. If an adjacent node is in the recursive stack, it means a cyclic loop has been found, and add it to the cyclic loop list.

[0089] Step Two Two Three, Backtracking:

[0090] When returning from the recursion, pop the current node from the recursive stack.

[0091] Step Three, Determine all cyclic loops in the causal network.

[0092] Step Three, Conduct variable time-delay trend analysis on the pairs of process variables involved in the cyclic loops;

[0093] In this embodiment, assume that the time-delay sequence (τ1, τ2, …, τ p ) consists of p independent and identically distributed random variables, where τ p represents the time delay obtained between every two process variables in the pth window. The Mann-Kendall trend test method is used for the test. The alternative hypothesis H1 is a two-sided test. Assume that for all i, where j ≤ p and i ≠ j, the calculation formula for the test statistic S is as follows:

[0094]

[0095] where the value of p is determined by the number of windows; x i and x j are the observed values of the ith and jth time-delay sequences (τ1, τ2, …, τ p ) respectively, and i < j. Sgn(·) is the sign function:

[0096]

[0097] where the corresponding values of (τ1, τ2, …, τ p ) are the observed values of the time-delay sequence. S follows a normal distribution with a mean of 0 and a variance of Var(S) = n(n - 1)(2n + 5) / 18. At this time, the calculation formula for the standardized test statistic Z is:

[0098]

[0099] In the bilateral trend test, if Z is given at the confidence level α, the null hypothesis is unacceptable, that is, at the confidence level α, the time delay series data has an obvious upward or downward trend. For the statistical variable Z, when it is greater than 0, the change trend of the time delay between process variables shows an upward trend, when it is less than 0, the change trend of the time delay between process variables is a downward trend, and if it is equal to 0, the change trend of the time delay between process variables is judged to be unchanged.

[0100] Step 4: By analyzing the changing trend of time delay, the causal relationship of the increasing lag trend inherent in the loop is transformed into a time causal relationship, thus eliminating the loop in the causal network.

[0101] In this embodiment, the basis for converting the loop into the time relationship used for abnormal root cause diagnosis is:

[0102] If the time delay is small, the effect of the cause variable on the result variable will be realized immediately; conversely, when the time delay is large, the result variable will take a long time to manifest. Therefore, the causal loop can be decomposed by judging the dynamic changes of the time delay of the process variable. The embodiment of the present invention proposes a logical rule for time causal relationship verification, which verifies the causal direction by analyzing the trend of time delay. The specific rules are as follows:

[0103] If the trend of the time delay between process variables is unchanged or decreasing, that is, the impact of the cause variable on the result variable will not change over time, which means that the time it takes for the impact to appear will not increase, then the causal relationship can be considered to remain unchanged.

[0104] If the changing trend of the time delay between process variables is upward, that is, the influence of the cause variable on the result variable will weaken over time, indicating that the time for the influence to appear is getting longer and longer, then the causal relationship needs to be converted into a time relationship.

[0105] It should be noted that when there are multiple upward trends in the loop, the process variable with the most obvious upward trend should be selected for decomposition. This selection is crucial to ensure the most complete network structure.

[0106] In this embodiment, functional module three: abnormal root cause diagnosis module, the specific implementation steps are:

[0107] Offline Modeling:

[0108] Step 1: Use process variable data (i.e., normal data) generated in a normal state during the process industry production process to establish a hidden Markov model; specifically, the following steps are included:

[0109] Assume a set Q containing all potential hidden states:

[0110] Q=(q1,q2,…,q N ) (1.13)

[0111] In this embodiment, the normal and abnormal (referring to: fault) states are hidden states that cannot be observed;

[0112] V is the set of all possible observable states:

[0113] V=(v1,v2,…,v M ) (1.14)

[0114] The temperature, pressure, etc. displayed by each process variable are possible observable states; N is the number of possible hidden states, and M is the number of possible observed states. It is worth noting that N and M are not necessarily equal.

[0115] For a time series of length T, where I corresponds to the hidden state sequence and O corresponds to the observed state:

[0116] I=(i1,i2,…i t …,i T ) (1.15)

[0117] O=(o1,o2,…o t …,o T ) (1.16)

[0118] Among them, any hidden state is i t ∈Q, any observed state is o t ∈V.

[0119] The hidden Markov model is defined by the initial probability distribution of the hidden state Π, the state transition probability matrix A and the observed state probability matrix B; where Π and A determine the progression of the hidden state, and B specifies the order of observations; the state transition probability matrix A is:

[0120] A=(a ij ) (1.17)

[0121] Among them, a ij =P(i t+1 =q j |i t =q i ), i=1,2,…,N, j=1,2,…,N, a ij Indicates that it is in the hidden state q at time t i Under the condition of jThe probability of the observed state is:

[0122] B=(b j (k))(1.18)

[0123] Among them, b j (k) = P(o t =v k |i t =q i ), k=1,2,…,M, j=1,2,…,N, b j (k) represents the hidden state q at time t i Generate observation state v k The probability of the hidden state is:

[0124] Π=(π(i)) (1.19)

[0125] Where π(i)=P(i1=q i ), i = 1, 2, ..., N. π(i) represents the hidden state q at time t+1 i probability.

[0126] Therefore, the hidden Markov model λ can be represented by a triple model:

[0127] λ=(A,B,Π)(1.20)

[0128] Step 2: Based on the established hidden Markov model, the log-likelihood value under normal conditions is calculated using normal data with variable time delays; specifically, the following steps are included:

[0129] After the hidden Markov model is established, the log-likelihood value is calculated based on the observation sequence (referring to: all process variables), assuming that the probability density function P(X) of the Gaussian distribution of the process variable X is:

[0130]

[0131] Where τ is the time delay, e is a constant, μ(τ) and σ(τ) represent the mean and variance of the observed data after considering the variable time delay, respectively.

[0132] When the observed variables (i.e., all process variables) follow the mixed Gaussian model and the observation sequence is regarded as a mixture of multiple Gaussian distributions, the joint probability density function is equal to the product of the individual probability densities:

[0133]

[0134] Among them, μ m (τ), σ m(τ) represent the mean and variance of the mth variable respectively;

[0135] P(X) can be simplified to:

[0136]

[0137] Where Σ is the summation and T is the matrix transpose;

[0138] According to the joint probability density function, the likelihood function can be obtained as:

[0139]

[0140] Among them, L(μ(τ),σ(τ)) is the likelihood function, n is the number of samples, σ is the variance; ζ is the i is the i-th observation value;

[0141] Since the likelihood value is too small, it is not suitable to use multiplication. Usually the logarithm is taken. The calculation of the logarithmic likelihood value is as follows:

[0142]

[0143] Wherein, ln[L(μ(τ),σ(τ))] represents the log-likelihood value.

[0144] Step 3: Determine the prior probability and conditional probability based on the normal data and the calculated log-likelihood value under the normal state; specifically, the following steps are included:

[0145] First, the training data (i.e., normal data) is divided into two regions, namely "safe" and "abnormal". The lower and upper limits of the safe zone are calculated as follows:

[0146] a iL =μ i (τ)+(min i -μ i (τ)) / r (1.26)

[0147] a iH =μ i (τ)+(max i -μ i (τ)) / r (1.27)

[0148] Among them, a iL and a iH are the lower and upper limits of the ith process variable, μ i (τ) represents the mean value of the i-th process variable after considering the time delay τ. min i and max iare the minimum and maximum values ​​of the ith process variable, respectively. The value of r is chosen so that the data has an appropriate distribution in the safe and dangerous areas.

[0149] Since the process variables are not probability values, they can be normalized according to the following mathematical procedure and generate conditional probabilities, defined as:

[0150]

[0151] Among them, α i is the conditional probability, which represents the probability of the normal state and fault state of the i-th process variable, which is equivalent to converting numbers into probabilities; b is a constant, ε is the setting accuracy, ρ is the number of possible states (in the embodiment of the present invention, there are only two possibilities: normal and fault), λ i =ln[L(μ(τ),σ(τ))] is the log-likelihood value, λ n =max(λ i ). Then, the conditional probability value is normalized:

[0152]

[0153] in, Ξ represents the sum of all α, N(α) represents its normalization, i,j = 1, 2, …, γ, γ is the number of log-likelihood values.

[0154] In this embodiment, equations (1.28) and (1.29) use the log-likelihood value obtained by the hidden Markov model to calculate the probability of the initial node (i.e., the node without a parent node) in the Bayesian network model. This is not sufficient to construct a conditional probability table for all process variables. It is necessary to continue to execute step four to obtain the conditional probability of all process variables, thereby constructing a conditional probability table for all process variables.

[0155] Step 4: Based on the determined prior probability and conditional probability, an improved BN model is established; specifically, the following steps are included:

[0156] BN describes the logical relationships between variables through the graph of conditional probabilities. Its description usually contains three terms: prior probability, conditional probability, and posterior probability. "Prior" refers to the initial information inherent in the variable. "Conditional" refers to the mutual information shared between the variables. "Posterior" reflects the degree of trust in the variable determined by the evidence. The prior information can be updated based on the collected information (also called evidence). Bayes' theorem is stated as follows:

[0157]

[0158] Among them, θ represents the parameter or hypothesis that we want to infer, which is usually called "unknown parameter" or "parameter to be estimated".

[0159] X represents observation data or evidence, which is the information we collect in practical applications. This data can be used to update the belief about θ. P(θ) is the prior probability, and P(X) is the probability of observation or evidence. P(X|θ) is the conditional probability of X given θ, also known as the likelihood. P(θ|X) is the conditional probability of θ given X, which can also be called the posterior probability because it contains the degree of belief that depends on the observation. The prior probability is the probability of normality and failure calculated from normal data, the evidence probability is the probability of normality and failure obtained through the hidden Markov model, and the conditional probability is the probability of evidence / parameter occurring given the parameter / evidence.

[0160] For a fixed value of X, this can be written as:

[0161] P(θ|X)=P(X|θ)×P(θ) (1.31)

[0162] The normalization constant P(X) depends on the existing evidence for a particular piece of evidence for X, and is also known as the Bayesian update equation. In general, at a given node (referring to the process variable) X = (X1, X2, ... X m )In a directed acyclic graph, the joint probability function of BN is:

[0163]

[0164] Among them, X i-1 Yes X i The parent node of P(X i |X i-1 ) indicates that in X i-1 In the event of X i Probability of occurrence.

[0165] Step 5: According to the improved BN model, construct a conditional probability table under normal conditions;

[0166] In this embodiment, by updating the probability formulas of equations (1.30), (1.31) and (1.32), the fault state probabilities of all process variables can be obtained, thereby constructing a conditional probability table under normal conditions.

[0167] Online Diagnosis:

[0168] Step 6: Use process variable data generated under abnormal conditions in the process industry production process to update the hidden Markov model;

[0169] Step 7: Calculate the log-likelihood value under abnormal conditions;

[0170] Step 8: Use the log-likelihood value under abnormal conditions as evidence to update the improved BN model;

[0171] Step nine, calculate the conditional probability table under abnormal conditions;

[0172] Step 10: Compare the conditional probability tables under normal and abnormal conditions to locate abnormal process variables and infer the root cause and propagation path.

[0173] In this embodiment, steps six to nine are online steps of steps one to five, with abnormal data as input. Step ten is to discover abnormal variables through changes in the conditional probability table and infer the abnormal root cause and propagation path.

[0174] The process industry abnormality root cause diagnosis method based on causal loop release provided by the embodiment of the present invention only uses process variable data collected in the process industry. Without relying on accurate mathematical models and sufficient expert knowledge, it deeply mines time series information and explores the dynamic changes of time delays, thereby effectively decomposing the loops in the causal network. At the same time, by relaxing the restrictions on the fixed time delay assumption, the Bayesian network model is improved to achieve abnormality root cause diagnosis and propagation path identification.

[0175] In summary, the process industry abnormality root cause diagnosis method based on causal loop elimination provided by the embodiment of the present invention has at least the following beneficial effects:

[0176] 1) Aiming at the problem of dynamic time delay in complex process industries, a method combining dynamic time warping and sliding window technology is used to realize real-time calculation of variable time delay and construct a variable time delay matrix to reveal the dynamic changes of time information;

[0177] 2) Aiming at the circular causal loop problem induced by factors such as variable coupling, cyclic flow, and integration of materials and thermal processes, a strategy for decomposing the circular loops in the causal network is designed by deeply mining the temporal information between process variables and combining the Mann-Kendall trend test. By analyzing the changing trend of variable time delays, the inherent causal loops are decomposed into temporal causal relationships, thereby eliminating the circular loops existing in the causal network.

[0178] 3) Based on the constructed variable time delay matrix, the traditional Bayesian network model is improved, and the limitation of the Bayesian network model on the fixed time delay assumption is lifted, so that it can handle the variable time delay between abnormal variables and improve the accuracy of abnormal root cause diagnosis in dynamic process industries.

[0179] Figure 2 FIG. 1 is a schematic diagram of a process industry abnormality root cause diagnosis device provided by an embodiment of the present invention. Figure 2 Optionally, the process industry abnormality root cause diagnosis device 210 may include a first processor 2001 .

[0180] Optionally, the process industry abnormality root cause diagnosis device 210 may further include a memory 2002 and a transceiver 2003 .

[0181] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0182] Combine the following Figure 2 The various components of the process industry abnormality root cause diagnosis device 210 are specifically introduced:

[0183] The first processor 2001 is the control center of the process industry abnormality root cause diagnosis device 210, which can be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiment of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (FPGAs).

[0184] Optionally, the first processor 2001 may execute various functions of the process industry abnormality root cause diagnosis device 210 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0185] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 2 CPU0 and CPU1 are shown in FIG.

[0186] In a specific implementation, as an embodiment, the process industry abnormality root cause diagnosis device 210 may also include multiple processors, such as Figure 2 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0187] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0188] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently, and may be connected to the first processor 2001 through the interface circuit ( Figure 2 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0189] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0190] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 2 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0191] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 2 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0192] It should be noted that Figure 2 The structure of the process industry abnormality root cause diagnosis device 210 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure identification device may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0193] In addition, the technical effects of the process industry abnormality root cause diagnosis device 210 can refer to the technical effects of the process industry abnormality root cause diagnosis method based on causal loop release described in the above method embodiment, which will not be repeated here.

[0194] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0195] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0196] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0197] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0198] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0199] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0200] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0201] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0202] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0203] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0204] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for diagnosing the root cause of abnormalities in process industries based on the elimination of causal loops, characterized in that: The method comprises: Offline Modeling: Step S1, for process variable data generated under normal conditions in the process industry production process, a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under normal conditions is constructed; Step S2, performing trend analysis on the variable time delay under normal conditions, decomposing the loops in the causal network into temporal causal relationships, and eliminating the loops existing in the causal network; Step S3, establishing a Bayesian network based on variable time delay to determine a conditional probability table under a normal state; Online Diagnosis: Step S4, for process variable data generated under abnormal conditions in the process industry production process, a method combining dynamic time warping and sliding window technology is used to calculate variable time delays in real time, and a variable time delay matrix under abnormal conditions is constructed; Step S5, based on the constructed variable time delay matrix under abnormal state, calculate the log-likelihood value to update the Bayesian network, determine the conditional probability table under abnormal state, compare the conditional probability tables under normal and abnormal states to locate the abnormal process variables and infer the root cause and propagation path.

2. The process industry abnormality root cause diagnosis method based on causal loop elimination according to claim 1 is characterized in that: The step S1 comprises: Step S11, obtaining process variable data generated in a normal state during the process industry production process; Step S12, standardizing the acquired process variable data generated under the normal state; Step S13, determining the time delay between process variables using the DTW method; wherein DTW stands for Dynamic Time Warping; Step S14, updating DTW in real time by combining sliding window technology to calculate the optimal time delay; Step S15, saving the optimal time delay corresponding to each window, and constructing a variable time delay matrix under normal conditions.

3. The process industry abnormality root cause diagnosis method based on causal loop release according to claim 2 is characterized in that: The step S13 comprises: Step S131, constructing the DTW distance between process variables; wherein the DTW distance is expressed as: Where D represents the DTW distance between two process variables, d i,j Represents the data x collected at the i-th sampling time corresponding to the first process variable i The data y collected at the jth sampling time corresponding to the second process variable j The Euclidean distance between them; n is the sampling time, which also represents the number of samples; Step S132, determining the best path through the matrix D so as to minimize the accumulated DTW distance, and the difference between the two corresponding process variables is the time delay.

4. The process industry abnormality root cause diagnosis method based on causal loop release according to claim 2 is characterized in that: The step S14 comprises: Assume that the initial window of the i-th pair of process variables is W i,h , then the initial window W i,h It is expressed as: IN i,h =[in i,t ,In i,t+1 ,…In i,t+l-1 ] Among them, h is the number of windows; l is the window length, which determines the update speed, and w i,t Represents the data corresponding to the i-th paired process variable at time t; Window W i,h Contains all process variable data from time t to time t+l-1, and the transition window is obtained by deleting k process variable data: IN i, ′ h =[in i,t+k ,In i,t+k+1 ,…In i,t+l-1 ] By adding k new process variable data, a new window is obtained: IN i,h+1 =[in i,t+k ,In i,t+k+1 ,…In i,t+k+l-1 ] Through the sliding window technology, the DTW distances corresponding to different windows are obtained. As the window moves forward, the DTW distances of all windows are recorded, realizing real-time update of the DTW distance to achieve the calculation of variable time delay.

5. The process industry abnormality root cause diagnosis method based on causal loop release according to claim 4 is characterized in that: The variable time delay matrix under normal conditions is expressed as: Where T' represents the variable time delay matrix; t p Indicates the variable time delay corresponding to the pth window, specifically: t p is a matrix containing all the time delays between every two process variables in the pth window; t k The subscript k∈(1,p), p is the number of windows; represents the DTW distance between the mth variable and the m'th variable in the Wth window, and f(·) represents the time delay corresponding to the shortest DTW distance.

6. The method for diagnosing the root cause of abnormalities in process industries based on the elimination of causal loops according to claim 4, characterized in that: The step S2 comprises: Step S21, developing a causal network of the process industry and constructing an initial BN model; wherein BN represents a Bayesian network; Step S22, determining the loops existing in the causal network; Step S23, performing variable time delay trend analysis on the paired process variables involved in the loop; Step S24, by analyzing the changing trend of time delay, the causal relationship of the increasing hysteresis trend inherent in the loop is converted into a time causal relationship, and the loop existing in the causal network is eliminated.

7. The process industry abnormality root cause diagnosis method based on causal loop release according to claim 6 is characterized in that: The step S22 comprises: Step S221, converting the causal network into a causal adjacency matrix, wherein the horizontal quantity in the causal adjacency matrix represents the cause variable, the vertical quantity represents the result variable, 0 represents the absence of a causal relationship, and 1 represents the presence of a causal relationship; Step S222, using a depth-first search algorithm to identify all paths and determine loops existing in the causal network, includes: Step S2221, initialization: Create an access mark array to mark whether each node has been visited. The number of nodes is consistent with the dimension of the causal adjacency matrix, that is, each node corresponds to the horizontal and vertical coordinates of the causal adjacency matrix. Create a recursive stack array to mark whether each node is on the current recursive path; Create an empty list to store all found loops; Step S2222, depth-first search: Starting from the current node, mark it as visited and push the node into the recursive stack; Traverse all adjacent nodes of the current node. If the adjacent node has not been visited, recursively call the depth-first search. If the adjacent node is in the recursive stack, it means that a loop is found and it is added to the loop list. Step S2223, backtracking: When returning from the recursion, pop the current node from the recursive stack; Step S223, determining all loops in the causal network.

8. The process industry abnormality root cause diagnosis method based on causal loop release according to claim 6 is characterized in that: The step S23 comprises: Assume that the time delay sequence (τ1,τ2,…,τ p ) consists of p independent and identically distributed random variables, among which τ p represents the time delay between each two process variables obtained in the pth window; the Mann-Kendall trend test method is used for testing. The alternative hypothesis H1 is a bilateral test. Assume that for all i, where j≤p and i≠j, the test statistic S is calculated as follows: Among them, the value of p is determined by the number of windows; x i and x j are the observed values of the i-th and j-th time-delay sequences (τ1, τ2, …, τ p ), respectively, and i < j, and Sgn(·) is the sign function: Where S is a normal distribution with a mean of 0 and a variance of Var(S)=n(n-1)(2n+5) / 18; at this time, the calculation formula for the standardized test statistic Z is: For the statistic Z, when it is greater than 0, the changing trend of the time delay between process variables shows an upward trend. When it is less than 0, the changing trend of the time delay between process variables is a downward trend. If it is equal to 0, the changing trend of the time delay between process variables is judged to be unchanged.

9. The method for diagnosing the root cause of abnormalities in process industries based on the elimination of causal loops according to claim 6, characterized in that: The step S24 comprises: If the trend of the time delay between process variables is constant or decreasing, that is, the effect of the cause variable on the result variable will not change over time, indicating that the time for the effect to appear will not increase, then the causal relationship is considered to remain unchanged; If the trend of the time delay between process variables is increasing, that is, the influence of the cause variable on the result variable will weaken over time, indicating that the time for the influence to appear is getting longer and longer, then the causal relationship needs to be converted into a time relationship; Among them, when there are multiple rising trends in the loop, the process variable with the most obvious rising trend is selected for decomposition.

10. The process industry abnormality root cause diagnosis method based on causal loop elimination according to claim 1, characterized in that: The step S3 comprises: Step S31, using normal data in the process industry production process to establish a hidden Markov model; wherein the normal data is: process variable data generated in the process industry production process under normal conditions; Step S32, based on the established hidden Markov model, using normal data considering variable time delay, calculate the log-likelihood value under normal conditions; Step S33, determining the prior probability and conditional probability under the normal state according to the normal data and the calculated log-likelihood value under the normal state; Step S34, establishing an improved BN model based on the determined prior probability and conditional probability under the normal state; Step S35: construct a conditional probability table under normal conditions according to the improved BN model.

Images