Causal modeling method and system for industrial production process system based on sequential structural equation model
Through the causal modeling method based on the time series structural equation model, the problem of insufficient explanatory and robustness of mathematical models in the industrial production process is solved, and the clear causal relationship between system variables is established, and the structural equation model is overcome for the time series The problem of data processing realizes causal modeling of the industrial production process.
Patent Information
- Application Number
- CN202210987440.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-17
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-08-17
AI Technical Summary
Due to its complexity and system characteristic parameter drift in industrial production process systems, it is difficult to establish accurate and stable mathematical models, machine learning models lack explanatory and robustness, and structural equation models are difficult to process time series data.
Using a causal modeling method based on the timing structural equation model, a linear or nonlinear timing structural equation model is established by collecting historical data, and the model is trained to determine the causal relationship between state variables, using the principle of transfer entropy and regression residual independence Obtain exogenous variables and construct quantitative causal relationships.
It overcomes the problems of explanatory inadequacy and robustness of machine learning models and is directly used for causal modeling of industrial production processes, providing a clear description of causal relationships between system variables, and improving the stability and long-term application of the model.
Smart Images

Figure CN115421451B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of artificial intelligence of industrial production processes, and specifically relates to a causal modeling method and system for an industrial production process system based on a temporal structural equation model. Background Art
[0002] In industrial production, it is usually difficult to use process mechanisms to establish accurate, stable and reliable mathematical models for the production process because the system mechanisms of the industrial production process are very complex, the internal parameters are difficult to measure, and there is a widespread drift in system characteristic parameters.
[0003] With the increasing maturity of technologies such as industrial communications, large-scale data storage, and cloud computing, the acquisition and storage of massive data is no longer a problem. Industrial control and information systems such as DCS, PLC, SCADA, MES, and real-time databases have been popularized in the industrial field. Industrial production has accumulated a huge amount of production process data, thus creating basic conditions for machine learning and data mining in the industrial production process.
[0004] With the rapid development of artificial intelligence (AI) technology and its deep integration into the industry, industry has entered the era of intelligent manufacturing. The industry generally uses artificial intelligence technologies represented by machine learning (including deep learning) to build data-driven models.
[0005] For industrial production processes, due to the stringent requirements for safe and stable production, the mathematical model needs to have clear and explicit interpretability and high reliability. At the same time, the mathematical model is required to be able to describe the causal relationship between system variables, so as to achieve reverse deduction based on the model, and further realize the application of fault diagnosis, operation optimization, quality improvement, cost reduction and consumption reduction, etc. However, the model established by the machine learning method lacks sufficient explanatory power and cannot describe the causal influence relationship between system variables. Therefore, it is difficult to use the machine learning regression model for inverse reasoning; in addition, the accuracy and effectiveness of the machine learning model are easily affected by the pollution, missing, noise, and unmeasurable interference of the training data, resulting in its long-term stability and lack of robustness in the application of industrial production processes.
[0006] The structural equation model (SEM) is a relatively mature model in the field of causal modeling research. It has a clear model structure, clear physical meaning and clear explanatory power. It is more suitable for the causal description of industrial production process systems. However, it lacks an effective method to process time delays in time series data, making it difficult to be directly used for causal modeling of industrial production process systems. Summary of the invention
[0007] Based on the above-mentioned shortcomings and deficiencies in the prior art, one of the objects of the present invention is to at least solve one or more of the above-mentioned problems in the prior art. In other words, one of the objects of the present invention is to provide a causal modeling method and system for an industrial production process system based on a time-series structural equation model that meets one or more of the above-mentioned needs.
[0008] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:
[0009] The causal modeling method of industrial production process system based on time-series structural equation model includes the following steps:
[0010] S1. Collecting historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and its data formation time; each state variable is sorted according to its data formation time;
[0011] S2. Using a time series structural equation model to establish a causal model structure for historical data information; wherein the causal model structure is a linear time series structural equation model or a nonlinear time series structural equation model;
[0012] S3. Train the causal model structure to obtain the causal relationship between the state variables of the industrial production process system.
[0013] As a preferred solution, in step S1, each state variable is sorted according to its data formation time and then aggregated into a wide table with time as row and state variable as column.
[0014] As a preferred solution, in step S1, the state variable is a variable related to production quality, operating parameters, production safety or operating stability.
[0015] As a preferred solution, in step S2, it is assumed that there are m state variables in the industrial production process system, and the causal model structure is:
[0016] x i (t): = f i (X,u i (t))
[0017] Where i = 1, 2, ..., m; x i (t) represents the i-th state variable at time t, u i (t) represents the external interference of the i-th state variable at time t; := represents that the right part is the cause of the left state variable;
[0018]
[0019] Among them, l is the longest lag time of the causal influence between each state variable.
[0020] As a preferred solution, in step S2, the sequential structural equation model of the linear structure is:
[0021]
[0022] Where k represents the lag time of the causal influence between state variables, in units of data sampling time; Represents x j (tk) for x i (t)’s causal influence; j = 1, 2, …, m, representing the sequence number of the state variable.
[0023] As a preferred solution, in step S3, training the linear structure temporal structural equation model includes the following steps:
[0024] S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable;
[0025] S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i (t) is the dependent variable, and the linear regression solution is obtained:
[0026]
[0027] x i The corresponding regression residual of (t) is:
[0028] S32. Calculate regression residuals and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as:
[0029] Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables
[0030] S33. Take the variable with the smallest sum of transfer entropy of each state variable, that is, get the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=arg min TE j ;
[0031] Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable;
[0032] S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s) (t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution;
[0033] If the number of elements that are 0 in array D is 1, stop iterating and go to step S35;
[0034] S35, state variable x i (t) can be expressed as:
[0035]
[0036]
[0037]
[0038] Among them, u=1, 2,...,m; v=1, 2,...,u-1.
[0039] As a preferred solution, in step S2, the temporal structural equation model of the nonlinear structure is:
[0040] x i (t): = f i (X,u i (t))
[0041] Where f is a nonlinear function.
[0042] As a preferred solution, in step S3, training the time series structural equation model of the nonlinear structure includes the following steps:
[0043] S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable;
[0044] S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i(t) is the dependent variable, and Gaussian process regression is performed to obtain:
[0045]
[0046]
[0047] x i The corresponding regression residual of (t) is:
[0048] S32, calculate the regression residual and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as:
[0049] Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables
[0050] S33. Take the variable with the smallest sum of transfer entropy of each state variable, that is, get the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=arg min TE j ;
[0051] Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable;
[0052] S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s) (t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution;
[0053] If the number of elements that are 0 in array D is 1, stop iterating and go to step S35;
[0054] S35, state variable x i (t) can be expressed as:
[0055]
[0056] The present invention also provides a causal modeling system for an industrial production process system based on a sequential structural equation model, applying the causal modeling method described in any of the above schemes, and the causal modeling system comprises:
[0057] The acquisition module is used to acquire historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and the time when the data is formed;
[0058] A modeling module is used to establish a causal model structure for historical data information using a time series structural equation model; wherein the causal model structure is a linear time series structural equation model or a nonlinear time series structural equation model;
[0059] The training module is used to train the causal model structure and obtain the causal relationship between the state variables of the industrial production process system.
[0060] Compared with the prior art, the present invention has the following beneficial effects:
[0061] The causal modeling method and system of the industrial production process system based on the time series structural equation model of the present invention establish a causal model for the time series data of a general industrial production process system as a description of the causal relationship between the variables in the system, overcome the problem of insufficient explanatory power of conventional data models based on machine learning, overcome the problem that conventional data models based on machine learning are difficult to support inverse reasoning in industrial production scenarios, and solve the problem that the structure of the structural equation model (SEM) is difficult to directly perform causal modeling for industrial production time series data. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 is a flow chart of a causal modeling method for an industrial production process system based on a sequential structural equation model according to Embodiment 1 of the present invention;
[0063] Figure 2 is a schematic diagram of a causal modeling method for an industrial production process system based on a sequential structural equation model according to Embodiment 1 of the present invention;
[0064] Figure 3 is a flow chart of the linear TS-SEM model training of Example 1 of the present invention;
[0065] Figure 4 It is a flow chart of the nonlinear TS-SEM model training of Example 1 of the present invention. DETAILED DESCRIPTION
[0066] In order to more clearly illustrate the embodiments of the present invention, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other accompanying drawings and other implementation methods can be obtained based on these accompanying drawings without creative work.
[0067] Embodiment 1:
[0068] like Figure 1 As shown, the causal modeling method of the industrial production process system based on the time series structural equation model of this embodiment includes the following steps:
[0069] S1. Data collection of industrial production process systems
[0070] Collecting historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and its data formation time; each state variable is sorted according to its data formation time;
[0071] Specifically, Figure 2 As shown, the historical production data of the system is derived and collected from the database of the control and information system of the specific production device (such as the control system DCS, PLC, etc. of the industrial production line, the information system MES, database, etc. of the industrial production line) through industrial communication protocols such as OPC. The process variables that need to collect data should be selected according to the general knowledge principles of the production process, and must include all variables related to production quality, operating parameters, production safety, and operating stability (including physical quantities measured by sensors in real time such as temperature, pressure, flow, current, power, and speed, as well as laboratory test data such as the content of each component in raw materials, intermediate materials, and products, and quality indicators). The data of all state variables collected should be sorted by time and aggregated into a wide table with time as rows and variables as columns.
[0072] S2. Use the Time Series Structure Equation Model (TS-SEM) to establish a causal model structure for historical data information; wherein the causal model structure is a linear time series structural equation model or a nonlinear time series structural equation model;
[0073] Specifically, suppose there are m state variables in the industrial production process system, and the causal model structure is:
[0074] x i (t): = f i (X,u i (t))
[0075] Where i = 1, 2, ..., m; x i (t) represents the i-th state variable at time t, u i (t) represents the external interference of the i-th state variable at time t; := indicates that the right part is the cause of the state variable on the left, representing the direction of causal influence between variables;
[0076]
[0077] Where X is an m*l dimensional matrix, containing the data of m state variables at the past times t-1, t-2, …, tl, and l is the longest lag time of the causal influence between the state variables.
[0078] Function f is a linear or nonlinear function, and the use of a linear or nonlinear function is determined based on a specific analysis of the characteristics of the system.
[0079] According to the characteristics of industrial production processes, generally unmeasurable factors or interference can be regarded as additive, so u in the above formula is i The term (t) can be used as an additive term. In addition, depending on the degree of linearity of the industrial production process system, TS-SEM with linear and nonlinear structures can be used.
[0080] Among them, when analyzing a specific production process object, if the production device process does not belong to a system with strong nonlinear characteristics from a mechanism analysis, a linear model structure is adopted to have stronger robustness and adaptability.
[0081] The sequential structural equation model of the linear structure is:
[0082]
[0083] Where k represents the lag time of the causal influence between state variables, with data sampling time as the unit; k≤l (the value of l is determined according to the actual production process and needs to be greater than the longest lag time of the influence relationship between the state variables in the system); Represents x j (tk) for x i (t)’s causal influence; j = 1, 2, …, m, representing the sequence number of the state variable.
[0084] In addition, when the nonlinear characteristics of the specific production process system are strong (such as a chemical reaction process) and the linear TS-SEM model has poor accuracy, the nonlinear TS-SEM model structure should be used, and the unmeasurable interference factor term should be used as an additive term. The nonlinear TS-SEM model can be expressed as follows:
[0085] x i (t): = f i (X,u i (t))
[0086] Where f is a nonlinear function.
[0087] S3. Train the causal model structure to obtain the causal relationship between the state variables of the industrial production process system.
[0088] According to the residual independence principle, the parameters of the linear TS-SEM model are obtained through an iterative method, that is, the variables that are most independent of the regression residuals of each state variable (called exogenous variables) are found in turn as the causal variable, and then the causal influence coefficients of the exogenous variables on the remaining state variables are obtained.
[0089] Specifically, Figure 3 As shown, the training of the linear structure sequential structural equation model includes the following steps:
[0090] S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable;
[0091] S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i (t)(i is not equal to j) is the dependent variable, and linear regression (least squares method) is performed to obtain:
[0092]
[0093] x i The corresponding regression residual of (t) is:
[0094] S32, calculate the regression residual and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as:
[0095] Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables
[0096] S33. Take the variable with the smallest sum of transfer entropy of each state variable, that is, get the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=arg min TE j ;
[0097] Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable; that is, let s = s + 1;
[0098] After finding the exogenous variables, the system can be expressed as follows:
[0099]
[0100] Among them, u in the first row D(s) (t) is external interference, exogenous variable x D(s) (t) is not affected by other state variables, but only by external interference factors. The second line of the above formula shows that for this system, the exogenous variable x D(s) (t) is the causal variable, while the remaining state variables are affected by the values of the exogenous variable in the past.
[0101] S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s) (t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution;
[0102] If the number of elements that are 0 in array D is 1, stop iterating and go to step S35;
[0103] S35. At this point, m-1 exogenous variables have been obtained. Combining the iterative calculations of the above steps, each state variable x i (t) can be expressed as:
[0104] x D(1) (t) = u D(1) (t)
[0105]
[0106]
[0107]
[0108]
[0109] Converted to matrix form:
[0110]
[0111]
[0112]
[0113] Among them, u=1, 2,...,m; v=1, 2,...,u-1.
[0114] The above formula is the TS-SEM model of linear structure. In this way, a clear quantitative causal relationship between the state variables within the system can be obtained:
[0115] x D(1) (t) is the causal variable for all other state variables, x D(2) (t) is divided by x D(1) (t) other state variables, x D(3) (t) is divided by x D(1) (t), x D(1) (t), and so on. D(m-1) (t) is only x D(m) (t) is one of the reasons, and x D(m) (t) is not the cause of any other state variable, but the result of other state variables. The quantitative causal relationship between state variables is contained in These coefficients.
[0116] In addition to the causal relationship between the state variables obtained above, the state variables are also affected by u D(1) (t),u D(2) (t),…,u D(m) (t) The influence of these external interference factors.
[0117] In addition, when the nonlinear characteristics of the specific production process system are strong (such as chemical reaction process) and the linear TS-SEM model has poor accuracy, a nonlinear TS-SEM model structure should be used. The training method of the nonlinear TS-SEM model is similar to that of the linear model, and is also based on the residual independence principle.
[0118] Specifically, Figure 4 As shown, the training of the sequential structural equation model of the nonlinear structure includes the following steps:
[0119] S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable;
[0120] S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i (t)(i is not equal to j) is the dependent variable, and Gaussian Process Regression is performed to obtain:
[0121]
[0122]
[0123] x i The corresponding regression residual of (t) is:
[0124] S32. Calculate regression residuals and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as:
[0125] Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables
[0126] S33. Take the variable with the smallest sum of transfer entropy of each state variable, that is, get the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=arg min TE j ;
[0127] Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable; that is, let s = s + 1;
[0128] After finding the exogenous variables, the system can be expressed as follows:
[0129]
[0130] Among them, u in the first row D(s) (t) is external interference, exogenous variable x D(s) (t) is not affected by other state variables, but only by external interference factors. The second line of the above formula shows that for this system, the exogenous variable x D(s) (t) is the causal variable, while the remaining state variables are affected by the values of the exogenous variable in the past.
[0131] S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s) (t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution;
[0132] If the number of elements that are 0 in array D is 1, stop iterating and go to step S35;
[0133] S35. At this point, m-1 exogenous variables have been obtained. Combining the iterative calculations of the above steps, each state variable x i (t) can be expressed as:
[0134]
[0135] x D(1) (t) is the causal variable for all other state variables, x D(2) (t) is divided by x D(1) (t) other state variables, x D(3) (t) is divided by x D(1) (t), x D(1) (t), and so on. D(m-1) (t) is only x D(m) (t) is one of the reasons, and x D(m) (t) is not the cause of any other state variable, but the result of the other state variables. The quantitative causal relationship between the state variables is included in the above Gaussian regression formula.
[0136] In addition to the causal relationship between the state variables obtained above, the state variables are also affected by u D(1) (t),u D(2) (t),…,u D(m) (t) The influence of these external interference factors.
[0137] The causal modeling method of the industrial production process system based on the time-series structural equation model of this embodiment needs to be solidified in the software or APP running on the industrial modeling in specific use. The server running on the industrial production site (usually installed in the computer room of the production control system) collects the historical and real-time data of the actual production line through communication protocols such as OPC in the industrial field. The algorithm can solidify the method in the software / APP through programming languages such as python and C++.
[0138] Based on the causal modeling method of the industrial production process system based on the time series structural equation model described above in this embodiment, this embodiment also provides a causal modeling system of the industrial production process system based on the time series structural equation model, including:
[0139] The acquisition module is used to collect historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and its data formation time; each state variable is sorted according to its data formation time;
[0140] The modeling module is used to establish a causal model structure for historical data information using a time series structural equation model; wherein the causal model structure is a linear time series structural equation model or a nonlinear time series structural equation model; the specific process can refer to the specific discussion of the above modeling method, which will not be repeated here.
[0141] The training module is used to train the causal model structure to obtain the causal relationship between the state variables of the industrial production process system. The specific training process can refer to the specific discussion of the above modeling method, which will not be repeated here.
[0142] The above description is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, according to the ideas provided by the present invention, there will be changes in the specific implementation methods, and these changes should also be regarded as the protection scope of the present invention.
Claims
1. A causal modeling method for industrial production process systems based on a sequential structural equation model, characterized in that: The following steps are involved: S1. Collecting historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and its data formation time; each state variable is sorted according to its data formation time; S2. Using a time series structural equation model to establish a causal model structure for historical data information; wherein the causal model structure is a time series structural equation model with a linear structure or a time series structural equation model with a nonlinear structure; S3, training the causal model structure to obtain the causal relationship between the state variables of the industrial production process system; In step S2, it is assumed that there are m state variables in the industrial production process system, and the causal model structure is: x i (t):=f i (X,u i (t)) Where i = 1, 2, ..., m; x i (t) represents the i-th state variable at time t, u i (t) represents the external interference of the i-th state variable at time t; := represents that the right part is the cause of the left state variable; Among them, l is the longest lag time of the causal influence between each state variable; In step S2, the sequential structural equation model of the linear structure is: Where k represents the lag time of the causal influence between state variables, in units of data sampling time; Represents x j (tk) for x i (t) causal influence; j = 1, 2, ..., m, represents the sequence number of the state variable; In step S3, training the linear structure temporal structural equation model includes the following steps: S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable; S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i (t) is the dependent variable, and the linear regression solution is obtained: x i The corresponding regression residual of (t) is: S32, calculate the regression residual and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as: Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables S33. Take the variable with the smallest sum of transfer entropies of each state variable, that is, obtain the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=argminTE j ; Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable; S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s )(t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution; If the number of elements that are 0 in array D is 1, stop iterating and go to step S35; S35, state variable x i (t) can be expressed as: Among them, u=1, 2,...,m; v=1, 2,..., u-1; In step S2, the temporal structural equation model of the nonlinear structure is: x i (t):=f i (X,u i (t)) Where, f is a nonlinear function; In step S3, training the time series structural equation model of the nonlinear structure includes the following steps: S30, define an array D of length m, for storing the serial numbers of exogenous variables, initially D(1), ..., D(m) = 0, and the number of exogenous variables s = 0; wherein the exogenous variable is the variable that is most independent of the regression residual of the state variable; S31, for each j, j≠D(p), p=1,2,…,m, take the data x of the jth state variable at the past moment j (t-1), x j (t-2), …, x j (tl) is the independent variable, and the state variable x i (t) is the dependent variable, and Gaussian process regression is performed to obtain: x i The corresponding regression residual of (t) is: S32, calculate the regression residual and x j The transfer entropy between (tl) is denoted as Then take the maximum transfer entropy of the j-th state variable's past data on the regression residual, recorded as: Calculate the sum of the transfer entropy of the jth state variable to the remaining state variables S33. Take the variable with the smallest sum of transfer entropies of each state variable, that is, obtain the sequence number of the exogenous variable and store it in array D, that is, D(s+1)=argminTE j ; Then the serial number in array D is increased by 1 to prepare for storing the serial number of the next exogenous variable; S34. If the number of elements in array D that are 0 is ≥ 2, use the current exogenous variable x D(s) (t) Regression residuals r on the remaining state variables i D(s) (t) replaces the original state variables x i (t), i≠D(s), and delete the state variable corresponding to the serial number D(s), and then return to step S31 for execution; If the number of elements that are 0 in array D is 1, stop iterating and go to step S35; S35, state variable x i (t) can be expressed as: x D(1) (t)=u D(1) (t) 2. The causal modeling method of an industrial production process system based on a sequential structural equation model according to claim 1 is characterized in that: In the step S1, each state variable is sorted according to the time when its data is formed and then aggregated into a wide table with time as the row and state variables as the column.
3. The causal modeling method of industrial production process system based on time series structural equation model according to claim 1 is characterized in that: In step S1, the state variables are variables related to production quality, operating parameters, production safety or operating stability.
4. A causal modeling system for an industrial production process system based on a time-series structural equation model, applying the causal modeling method according to any one of claims 1 to 3, characterized in that: The causal modeling system comprises: The acquisition module is used to acquire historical data information of the industrial production process system; wherein the historical data information includes each state variable in the industrial production process system and the time when the data is formed; A modeling module is used to establish a causal model structure for historical data information using a time series structural equation model; wherein the causal model structure is a time series structural equation model with a linear structure or a time series structural equation model with a nonlinear structure; The training module is used to train the causal model structure and obtain the causal relationship between the state variables of the industrial production process system.
