Complex industrial process self-healing control method based on elastic time-varying Bayesian network

By adopting the self-healing control method of elastic time-varying Bayesian network in complex industrial processes, the control problems caused by non-stationary and time-varying characteristics are solved, and fast and effective control decision-making and automatic decision-making support are achieved, ensuring the industrial process Safe and stable operation.

CN118915653AActive Publication Date: 2024-11-08CHINA UNIV OF MINING & TECH
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202410959173.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-08
Estimated Expiration
2044-07-17

AI Technical Summary

Technical Problem

The non-stationarity and time-varying of complex industrial processes make traditional steady-state modeling and control methods difficult to apply, resulting in a decrease in control effect and decision-making accuracy. The existing methods are difficult to deal with multivariable coupling and rapidly changing working conditions.

Method used

Using a self-healing control method based on elastic time-varying Bayesian network, by establishing a heavy medium coal preparation system, collecting data for smooth processing, constructing a dynamic Bayesian network model, capturing the causal relationship between variables, and using conditional probability inference Control decision-making plans and update models in real time to adapt to process changes.

Benefits of technology

It realizes rapid and effective control decisions in complex industrial processes, ensures the safe and stable operation of industrial processes, provides automatic decision-making support, improves the sensitivity and reliability of control strategies, and adapts to changes in non-stationary and time-varying characteristics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915653A_ABST
    Figure CN118915653A_ABST
Patent Text Reader

Abstract

The invention discloses a complex industrial process self-healing control method based on an elastic time-varying Bayesian network. The method comprises the following steps: acquiring field data and analyzing and processing the field data; using an ADF square root algorithm to check the stability of the data, if the data is not stable, performing differential processing on the data, and stabilizing the data; fusing the stabilized data with the original stable data, and then performing data correlation analysis; through correlation analysis, determining the strength and direction of the relationship among the multiple variables; inputting the data into an elastic time-varying dynamic Bayesian network model, and learning a network structure and parameters; accurately positioning an abnormal time slice according to the learned structure and parameters in combination with a probability density function, and determining a process variable causing an anomaly; and inputting the abnormal data as evidence into the dynamic Bayesian network to deduce a decision scheme capable of eliminating the abnormal working condition, and converting the decision scheme into control operation. According to the method, an effective control decision can be quickly made after abnormity occurs in the coal slime flotation industrial process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of process industry self-healing control, and in particular is a complex industrial process self-healing control method based on elastic time-varying Bayesian network. Background Art

[0002] With the continuous improvement of industrial automation and intelligence, the challenges faced by process control in process industries are becoming more complex and diverse. Complex industrial processes are industrial production processes with high dynamics, nonlinearity, multivariable, time-varying characteristics and strong coupling. The process usually involves multiple interrelated subsystems and operating steps, and various factors inside and outside the system jointly affect its operating status and performance. Due to its dynamic and time-varying nature, most complex industrial processes are non-stationary industrial processes. They are affected by many factors during operation, causing system characteristics and behaviors to change over time. These factors include but are not limited to changes in environmental conditions, aging and wear of equipment, fluctuations in raw material characteristics, and adjustments to operating conditions. Non-stationarity is reflected in the dynamic changes of system parameters and states. When the system is abnormal, traditional steady-state modeling and control methods are difficult to apply.

[0003] Non-stationary industrial processes usually show the characteristics of parameters and structures changing over time. These characteristics cause the system state and parameters to change continuously, making it difficult for a single fixed model to accurately describe and judge the system behavior, which in turn affects the control effect and the accuracy of decision-making. In addition, the complexity and multivariate correlation characteristics of industrial processes also increase the difficulty of process control and safety management. How to achieve efficient monitoring and safety control of non-stationary industrial processes in a dynamic environment has become an important issue that needs to be solved in the current field of industrial process control. The existing safety control methods for non-stationary complex industrial processes can be divided into model-based methods, knowledge-driven methods, and data-driven methods. The model-driven method is based on physical and chemical principles, and the model is highly interpretable and suitable for specific process. However, the model establishment is complex and requires an in-depth understanding of the system mechanism. For nonlinear and non-stationary processes, modeling is too difficult, and model updating and maintenance are difficult, making it difficult to adapt to rapidly changing working conditions. Commonly used model-driven methods include first principle modeling, state space model, mechanism model, etc. The knowledge-driven method uses expert knowledge and experience, has clear rules, and a transparent reasoning process, which is suitable for known working conditions and fault types. However, it is difficult to establish and maintain the rule base, unable to handle new unknown faults, and limited in its ability to handle complex nonlinear relationships. Knowledge-driven methods include expert systems, fuzzy logic, rule reasoning, etc. Data-driven methods can process large amounts of data, capture complex patterns and nonlinear relationships, have strong adaptability, and can be applied to a variety of process flows. However, they rely on a large amount of high-quality data, are difficult to interpret and understand, and are less sensitive to emerging faults and operating conditions. Common data-driven methods include neural networks, machine learning, support vector machines, deep learning, etc. Therefore, it is necessary to combine the advantages of the above methods and make comprehensive use of expert knowledge, data analysis, and model simulation to make multi-level and multi-dimensional decisions to achieve safe control of non-stationary complex industrial processes. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a self-healing control method for complex industrial processes based on an elastic time-varying Bayesian network. The method has simple steps and high reliability. It can quickly make effective control decisions after an abnormality occurs in the coal slime flotation industrial process, thereby ensuring the safe and stable operation of the industrial process and providing automatic decision-making support for abnormal conditions of the flotation process.

[0005] In order to achieve the above-mentioned purpose, the present invention provides a self-healing control method for a complex industrial process based on an elastic time-varying Bayesian network, including a heavy medium coal preparation system, wherein the heavy medium coal preparation system includes a conveyor, a heavy medium cyclone, an arc screen A, a clean coal de-mediation screen, an arc screen B, a gangue de-mediation screen, a dilution medium barrel, a dilute medium pump, a dilute medium magnetic separator, a water adding pump, a adding medium pump, a adding medium magnetic separator, a qualified medium barrel and a pressure pump; the feed port of the heavy medium cyclone is connected to the discharge end of the conveyor, the feed port of the arc screen A is connected to the overflow outlet of the heavy medium cyclone, the feed port of the clean coal de-mediation screen is connected to the discharge port of the arc screen A, the upper screen outlet of the clean coal de-mediation screen is used to output the clean coal, and its lower screen outlet is respectively connected to a first drainage pipeline and a second drainage pipeline; the feed port of the arc screen B is connected to the underflow outlet of the heavy medium cyclone, and the feed port of the gangue de-mediation screen is connected to the discharge port of the arc screen B The upper screen outlet of the gangue de-media screen is used to output the coal gangue, and its lower screen outlet is respectively connected to the third drainage pipeline and the fourth drainage pipeline; the feed port at the top of the dilution medium barrel is respectively connected to the outlet end of the first drainage pipeline and the outlet end of the third drainage pipeline; the inlet end of the dilution pump is connected to the dilution medium outlet at the bottom of the dilution medium barrel; the inlet end of the dilute medium magnetic separator is connected to the outlet end of the dilution pump; the water inlet end of the water pump is connected to the water source; the inlet end of the medium pump is connected to the medium source; the inlet end of the medium magnetic separator is connected to the outlet end of the medium pump; the feed port at the top of the qualified medium barrel is respectively connected to the outlet end of the second drainage pipeline, the outlet end of the fourth drainage pipeline, the outlet end of the dilute medium magnetic separator, the water outlet end of the water pump and the outlet end of the medium magnetic separator; the inlet end of the pressure pump is connected to the outlet end at the bottom of the qualified medium barrel, and its outlet end is connected to the feed port of the heavy medium cyclone;

[0006] The following steps are involved:

[0007] Step 1: Establish a heavy medium coal preparation system and determine the variables;

[0008] A heavy medium coal separation system is established by using a conveyor, a heavy medium cyclone, a curved screen A, a clean coal de-mediating screen, a curved screen B, a gangue de-mediating screen, a dilution medium barrel, a dilution medium pump, a dilution medium magnetic separator, a water pump, a medium pump, a medium magnetic separator, a qualified medium barrel and a pressure pump;

[0009] According to the operating characteristics of the heavy medium coal preparation system, the measurement variables and operation variables in the heavy medium coal preparation system are determined; there are 6 measurement variables, which are the double-layer screen discharge flow, single-layer screen discharge flow, mixed medium barrel slurry density, cyclone medium density, qualified medium barrel medium density, cyclone overflow ash content; there are 2 operation variables, which are the raw coal bunker coal input amount and cyclone medium inlet pressure;

[0010] Step 2: Collect coal preparation process data and use the ADF square root algorithm to test the stability of the industrial process;

[0011] S21: During the operation of the heavy medium coal preparation system, the coal preparation process data is collected by using the sensor group arranged in the heavy medium coal preparation system, and the data is output to the controller. The controller processes the coal preparation process data according to the internal clock module to obtain time series data;

[0012] S22: Use the ADF square root algorithm to test the stationarity of time series data;

[0013] S221: According to the characteristics of the time series, select the ADF test model with constant term and time trend, and use formula (1) to obtain the first-order difference Δy of the time series t ;

[0014]

[0015] In the formula, Δy t Represents y t -y t-1 ;y t-1 is the time series value lagged one order; μ is the constant term; λt is the time trend term; α is the most important coefficient in the ADF test, used to determine the existence of a unit root; β i is the coefficient of the difference term, which is used to eliminate the high-order autocorrelation in the sequence; ∈ t is the white noise error term; p is the lag order;

[0016] S222: Estimate the parameters in the ADF test model using the least squares method;

[0017] S223: Calculate the ADF statistic, which is the estimated α divided by its standard error;

[0018] S224: Construct hypothesis test, null hypothesis H 0 The unit root exists. In this case, the data is non-stationary. The alternative hypothesis H 1 The unit root does not exist, and the data is stable;

[0019] Compare the calculated ADF statistic with the corresponding critical value; the critical value comes from the Dickey-Fuller distribution table; if the ADF statistic is less than the critical value, the null hypothesis is rejected and the time series is considered to be stationary; if the ADF statistic is greater than the critical value, the null hypothesis cannot be rejected and the time series is considered to be non-stationary;

[0020] Step 3: Stabilize the non-stationary data;

[0021] For non-stationary data, the trend and periodic components in the time series are eliminated by difference until the data reaches a stable state. In this process, the unit root test or the time series diagram of the differenced data is used to check whether the differenced data reaches a stable state.

[0022] Step 4: Conduct structural learning and establish a flexible time-varying dynamic Bayesian network model;

[0023] S41: Fusing the stabilized data with the original stabilized data to obtain fused data, performing correlation analysis on the fused data to determine the strength and direction of the relationship between multiple variables, and presenting the relationship between the variables in the form of a graph;

[0024] S42: The key variables that affect the coal preparation effect are used as nodes in the elastic time-varying dynamic Bayesian network, and an elastic time-varying dynamic Bayesian network model that reflects the causal relationship between industrial process variables is constructed to capture the changes in the dependency relationship of the heavy medium coal preparation system in different time periods;

[0025] S43: Use fused data to train a flexible time-varying dynamic Bayesian network model to mine the causal relationship behind the data and determine the dynamic dependency between nodes;

[0026] Step 5: elastic time-varying dynamic Bayesian network parameter learning;

[0027] Given the conditional probability table of each node, the elastic time-varying dynamic Bayesian network is used to learn parameters, and a non-stationary elastic time-varying dynamic Bayesian network model is obtained, which provides a basis for the inference control decision plan; when an abnormal condition occurs in the coal selection process, the probability density function is used to infer the abnormal time slice; then, the conditional probability distribution is used to determine the variables causing the abnormal condition, and the state of each variable at the current moment is inferred, and the optimal safety control decision plan is inferred;

[0028] Step 6: Determine the time slice where the abnormality occurs and the variables that cause the abnormal condition;

[0029] The online abnormal data phenomenon variables are input as evidence into the non-stationary elastic time-varying dynamic Bayesian network model. The non-stationary elastic time-varying dynamic Bayesian network model accurately locates the time slices of the heavy medium coal preparation process where the abnormality occurs based on the learned structure and parameters, and determines the process variables that cause the abnormality.

[0030] Step 7: Infer the control decision plan;

[0031] The expected quality index and the state value of the variable are input into the local Bayesian network structure for reasoning, and the posterior probability of the operation node is obtained. According to the principle of maximum posterior probability, the adjustment plan of the heavy medium coal preparation process decision, i.e., the control decision, is obtained;

[0032] Step 8: Implement control decisions;

[0033] The controller adjusts the operating parameters of the heavy medium coal preparation system according to the obtained control decision. If the abnormal condition is eliminated, it enters the normal working mode. If the abnormal condition is not eliminated, the online abnormal data is collected again and step seven is re-executed. The online abnormal data phenomenon variables are continued to be used as evidence information to infer the control decision plan until the abnormal condition is eliminated.

[0034] In the coal preparation process, the characteristics of raw materials, the performance of equipment and environmental conditions may affect the entire production process. The dependencies and parameters of the system may change over time, and the entire process presents non-stationary characteristics. Self-healing control methods based solely on data or expert knowledge are difficult to adapt to non-stationary complex industrial production processes with high dynamics, multiple variables, time-varying characteristics and strong coupling. The elastic time-varying dynamic Bayesian network model is based on the vector autoregression model and can be used to describe the time dynamics between multiple variables. In order to deal with the self-healing control problem after abnormal working conditions occur in non-stationary complex industrial processes, the present invention provides a self-healing control decision model based on an elastic time-varying dynamic Bayesian network. The elastic time-varying dynamic Bayesian network decision model combines data-driven and mechanism knowledge for modeling, and is a modeling method jointly driven by model and data. When data is abundant, the model can automatically adjust parameters by learning data; when data is scarce, it can be supplemented with prior knowledge to ensure the effectiveness of the model. And it can update and adapt to changes in non-stationary processes in real time. By dynamically adjusting the model parameters and structure, the state changes of the system in different time periods can be accurately reflected, ensuring accurate description of the system state and predicting the behavior of the system. The elastic time-varying dynamic Bayesian network can handle the complex dependencies between multiple variables at the same time, which is particularly important when dealing with multivariable coupling problems in complex industrial processes, because multiple process variables are usually mutually influential and correlated. When dealing with the time-varying parameters and structures of industrial processes, the model can capture the long-term trend changes in the industrial process and the time-varying characteristics such as sudden abnormal conditions, making the model more flexible and accurate. The dynamic Bayesian network itself has the advantage of handling uncertainty and can quantify and manage uncertainty through probability distribution. In the elastic time-varying dynamic Bayesian network model, uncertainty is described by conditional probability distribution, which can effectively deal with randomness and noise in industrial processes. In order to realize anomaly detection and decision reasoning of non-stationary complex industrial processes, the main process of the elastic time-varying dynamic Bayesian network model is divided into three parts: the first part is data preprocessing. The collected field data is analyzed and processed, and the stationarity of the data is tested using the ADF square root algorithm. If the data is non-stationary, the data is differentially processed to stabilize the data. In the coal preparation process, understanding the stationarity of data is crucial to establishing effective control strategies and improving sorting efficiency. Trends in non-stationary data can interfere with the decision-making ability of the model and easily lead to overfitting. Converting non-stationary data processes into stationary data processes has significant benefits for control decisions. This not only improves the accuracy and robustness of the prediction model, but also simplifies the calculation and analysis process. Through stabilization processing, the interpretability and operability of the data are enhanced, and the sensitivity and reliability of the control strategy are improved. Therefore, through the technology of data stabilization, not only the effectiveness of the control strategy is ensured, but also it is conducive to improving the sorting efficiency.The stabilized data is fused with the original stable data. After data fusion, correlation analysis is performed. Through correlation analysis, the relationship and dependency between different variables are evaluated, which can provide an intuitive basis for control decisions. The second part is the elastic time-varying dynamic Bayesian network structure learning and parameter learning. Construct a network structure that reflects the causal relationship between industrial process variables to capture the changes in the dependency of the system in different time periods. Accurately locate the abnormal time slice based on the learned structure and parameters, and determine the process variables that cause the abnormality. The third part is decision reasoning. The abnormal data is input as evidence into the elastic time-varying dynamic Bayesian network to infer a decision plan that can eliminate abnormal conditions, and the decision plan is converted into a robust control operation for application in the coal preparation process.

[0035] The method has simple steps and high reliability. It can quickly make effective control decisions after an abnormality occurs in the coal slime flotation industrial process, ensure the safe and stable operation of the industrial process, and provide automatic decision-making support for abnormal conditions in the flotation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the present invention;

[0037] Figure 2 This is an example diagram of the process of heavy medium coal preparation in the present invention;

[0038] Figure 3 It is the ADF non-stationary test result diagram in the present invention;

[0039] Figure 4 It is a visual representation diagram after data stabilization in the present invention;

[0040] Figure 5 It is a structural diagram of the elastic time-varying dynamic Bayesian network in the present invention;

[0041] Figure 6 It is a scatter plot of abnormal data distribution of the heavy medium coal preparation process in the present invention;

[0042] Figure 7 is a histogram of abnormal data distribution of the heavy medium coal preparation process in the present invention;

[0043] Figure 8 Changes in overflow ash curve of heavy medium model. DETAILED DESCRIPTION

[0044] The present invention will be further described below in conjunction with the accompanying drawings.

[0045] Heavy medium coal preparation is a commonly used coal washing method. In the heavy medium coal preparation process, the raw coal is first crushed and screened to remove large pieces of coal and gangue to ensure that the particle size of the selected coal is uniform. The pre-treated coal enters the heavy medium separator, which is filled with a heavy medium suspension. In the heavy medium suspension, the coal with a lower density will float on the surface of the suspension, while the gangue with a higher density will sink to the bottom of the suspension. In this way, the difference in density between coal and gangue can be utilized, and the coal and gangue can be separated through the flotation effect of the heavy medium suspension to achieve the purpose of improving coal quality and reducing impurity content. The clean coal floating on the surface of the suspension and the gangue sinking to the bottom are dehydrated by the dewatering screen respectively, and the clean coal and gangue are obtained after removing the surface suspension. Finally, the magnetite powder in the suspension is recovered by the magnetic separator, and then the magnetite powder is added back into the suspension for recycling. With the advancement of science and technology and the continuous improvement of environmental protection requirements, the development trend of heavy medium coal preparation is moving towards automation and intelligence. It is very important to adopt automation and safety control technology to improve the separation efficiency and ensure the safe and stable operation of the industrial process.

[0046] like Figures 1 to 8 As shown, the present invention provides a self-healing control method for a complex industrial process based on an elastic time-varying Bayesian network, including a heavy medium coal preparation system, wherein the heavy medium coal preparation system includes a conveyor, a heavy medium cyclone, an arc screen A, a clean coal de-mediation screen, an arc screen B, a gangue de-mediation screen, a dilution medium barrel, a dilute medium pump, a dilute medium magnetic separator, a water adding pump, a adding medium pump, a adding medium magnetic separator, a qualified medium barrel and a pressure pump; the feed port of the heavy medium cyclone is connected to the discharge end of the conveyor, the feed port of the arc screen A is connected to the overflow outlet of the heavy medium cyclone, the feed port of the clean coal de-mediation screen is connected to the discharge port of the arc screen A, the upper screen outlet of the clean coal de-mediation screen is used to output the clean coal, and the lower screen outlet is respectively connected to the first drainage pipeline and the second drainage pipeline; the feed port of the arc screen B is connected to the underflow outlet of the heavy medium cyclone, the feed port of the gangue de-mediation screen is connected to the discharge port of the arc screen B, and the gangue The outlet on the screen of the de-media screen is used to output coal gangue, and its outlet under the screen is respectively connected to the third drainage pipeline and the fourth drainage pipeline; the feed port on the top of the dilution medium barrel is respectively connected to the outlet end of the first drainage pipeline and the outlet end of the third drainage pipeline; the inlet end of the dilution pump is connected to the dilution medium outlet at the bottom of the dilution medium barrel; the inlet end of the dilute medium magnetic separator is connected to the outlet end of the dilution pump; the water inlet end of the water pump is connected to the water source; the inlet end of the medium pump is connected to the medium source; the inlet end of the medium magnetic separator is connected to the outlet end of the medium pump; the feed port on the top of the qualified medium barrel is respectively connected to the outlet end of the second drainage pipeline, the outlet end of the fourth drainage pipeline, the outlet end of the dilute medium magnetic separator, the water outlet end of the water pump and the outlet end of the medium magnetic separator; the inlet end of the pressure pump is connected to the outlet end at the bottom of the qualified medium barrel, and its outlet end is connected to the feed port of the heavy medium cyclone;

[0047] The following steps are involved:

[0048] Step 1: Establish a heavy medium coal preparation system and determine the variables;

[0049] A heavy medium coal separation system is established by using a conveyor, a heavy medium cyclone, a curved screen A, a clean coal de-mediating screen, a curved screen B, a gangue de-mediating screen, a dilution medium barrel, a dilution medium pump, a dilution medium magnetic separator, a water pump, a medium pump, a medium magnetic separator, a qualified medium barrel and a pressure pump;

[0050] As shown in the analysis cases in Table 1 and Table 2, the measured variables and operating variables in the heavy medium coal preparation system are determined according to the operating characteristics of the heavy medium coal preparation system; there are 6 measured variables, which are the double-layer screen discharge flow, the single-layer screen discharge flow, the mixed medium barrel slurry density, the cyclone medium density, the qualified medium barrel medium density, and the cyclone overflow ash content; there are 2 operating variables, which are the raw coal bunker coal input amount and the cyclone medium inlet pressure;

[0051] Table 1: Variables measured during heavy medium coal preparation

[0052]

[0053]

[0054] Table 2: Heavy medium coal preparation process operating variables

[0055]

[0056]

[0057] Step 2: Collect coal preparation process data and use the ADF square root algorithm to test the stability of the industrial process;

[0058] The ADF (Augmented Dickey-Fuller) test is a statistical test method used to detect whether time series data has a unit root. The presence of a unit root indicates that the time series data is non-stationary, that is, the mean, variance, and autocorrelation structure change over time. The stationarity of time series is a basic assumption of many statistical and econometric models, so the ADF test is of great significance in time series analysis. The ADF test determines the stationarity of the time series by testing the unit root in the model. It is an extended form of the Dickey-Fuller test that takes into account more complex situations, including high-order autocorrelation structures in the time series. The ADF test eliminates high-order autocorrelations in the series by introducing lagged difference terms, thereby improving the reliability of the test.

[0059] S21: During the operation of the heavy medium coal preparation system, the coal preparation process data is collected by using the sensor group arranged in the heavy medium coal preparation system, and the data is output to the controller. The controller processes the coal preparation process data according to the internal clock module to obtain time series data;

[0060] S22: Use the ADF square root algorithm to test the stationarity of time series data;

[0061] S221: According to the characteristics of the time series, select the ADF test model with constant term and time trend, and use formula (1) to obtain the first-order difference Δy of the time series t ;

[0062]

[0063] In the formula, Δy t Represents y t -y t-1 ;y t-1 is the time series value lagged one order; μ is the constant term; λt is the time trend term; α is the most important coefficient in the ADF test, used to determine the existence of a unit root; β i is the coefficient of the difference term, which is used to eliminate the high-order autocorrelation in the sequence; ∈ t is the white noise error term; p is the lag order;

[0064] S222: Estimate the parameters in the ADF test model using the least squares method;

[0065] S223: Calculate the ADF statistic, which is the estimated α divided by its standard error;

[0066] S224: Construct hypothesis test, null hypothesis H 0 The unit root exists. In this case, the data is non-stationary. The alternative hypothesis H 1 The unit root does not exist, and the data is stable;

[0067] Compare the calculated ADF statistic with the corresponding critical value; the critical value comes from the Dickey-Fuller distribution table; if the ADF statistic is less than the critical value, the null hypothesis is rejected and the time series is considered to be stationary; if the ADF statistic is greater than the critical value, the null hypothesis cannot be rejected and the time series is considered to be non-stationary;

[0068] If a time series is non-stationary, it means that its statistical characteristics change over time. The existence of unit roots is one of the main reasons for the non-stationarity of time series. The ADF test method constructs a hypothesis test to determine whether the time series contains a unit root. The original hypothesis H 0 If a unit root exists, the data is non-stationary; the alternative hypothesis H 1If the unit root does not exist, the data is stationary. In the coal preparation process, understanding the stationarity of the data is crucial to establishing an effective control strategy and improving the sorting efficiency. In the coal preparation process, key variables such as density and flow rate may fluctuate over time, and this fluctuation may lead to instability in process control and reduced sorting efficiency. Experiments have shown that key variables in the coal preparation process show non-stationarity. Automatic control systems based on traditional control methods may not be able to effectively adapt to these changes, resulting in a decrease in control performance.

[0069] like Figure 3 As shown, the present invention determines the non-stationary process variables by performing ADF test on the time series data in the heavy medium coal preparation process, thereby providing a reliable basis for further data processing and control strategy adjustment. In the heavy medium coal preparation process, the coal input amount of the raw coal bin (A), the double-layer screen discharge flow rate (B), the single-layer screen discharge flow rate (C), the cyclone medium density (F), the cyclone inlet pressure (G), and the cyclone overflow ash (H) are non-stationary variables. In the coal slime thickening and flotation process, the thickener underflow flow (L), the thickener medium density (M), the slurry preprocessor medium density (N), and the flotation tank stirring speed (P) are non-stationary variables.

[0070] Step 3: Stabilize the non-stationary data;

[0071] Difference is a method of calculating the difference between adjacent data points in a time series, which is used to eliminate trends and cyclical components in the time series, thereby making the data closer to a stationary state. For non-stationary data, differences are used to eliminate trends and cyclical components in the time series until the data reaches a stationary state; in this process, the unit root test or the time series diagram of the differenced data is used to check whether the differenced data has reached a stationary state; a single difference can remove the linear trend in the time series by subtracting the value of the previous time point. If the series is a random walk and the data is still not stationary, a second difference or a higher-order difference may be required. If the data is still not stationary after the second difference, a third difference or a higher-order difference can be performed until the data reaches a stationary state;

[0072] Trends in non-stationary data can interfere with the decision-making ability of the model and easily lead to overfitting. Converting a non-stationary data process into a stationary data process has significant benefits for control decisions. This not only improves the accuracy and robustness of the prediction model, but also simplifies the calculation and analysis process. Through the stationary processing, the interpretability and operability of the data are enhanced, and the sensitivity and reliability of the control strategy are improved. After using the above method to perform a non-stationary test, the present invention uses a data difference method to convert the non-stationary data process into a stationary data process. On the basis of stationary data, any abnormal points that deviate from stationarity are more easily detected, which helps to promptly discover and handle abnormal situations in the coal preparation process, and can learn more robust elastic time-varying dynamic Bayesian network structures and parameters. Figure 4 It is a visual representation of the data after smoothing.

[0073] Step 4: Conduct structural learning and establish a flexible time-varying dynamic Bayesian network model;

[0074] Due to the complexity of the production environment, the heavy medium coal preparation process has the characteristics of non-stationary complex industrial processes such as multivariables, time-varying characteristics and strong coupling. The elastic time-varying dynamic Bayesian network model can adapt to parameters and structures that change over time, which is crucial for dealing with the non-stationary characteristics of the coal preparation process; the causal relationship between variables is captured according to the model, and these relationships are presented in the form of a graph, thereby providing an intuitive basis for control decisions. After determining the key variables that affect the coal preparation effect, these variables are regarded as nodes in the network; after the non-stationary data is stabilized, the elastic time-varying dynamic Bayesian network model is trained through the data to explore the causal relationship behind the data and determine the dynamic dependencies between nodes. As time goes on, the model can update these dependencies in real time, reflecting the real-time state changes in the coal preparation process. When an abnormality is detected in a key variable or its dependency, corresponding adjustments can be made in time to ensure the safe and stable operation of the coal preparation process. According to the network structure Figure 5 It is obvious that the dynamic changes of the heavy medium coal preparation process and the causal relationship between variables can be clearly seen. As time goes by, the structure and edges of the graph are updated in real time.

[0075] S41: Fusing the stabilized data with the original stabilized data to obtain fused data, performing correlation analysis on the fused data to determine the strength and direction of the relationship between multiple variables, and presenting the relationship between the variables in the form of a graph;

[0076] S42: The key variables that affect the coal preparation effect are used as nodes in the elastic time-varying dynamic Bayesian network, and an elastic time-varying dynamic Bayesian network model that reflects the causal relationship between industrial process variables is constructed to capture the changes in the dependency relationship of the heavy medium coal preparation system in different time periods;

[0077] S43: Use fusion data to train a flexible time-varying dynamic Bayesian network model, explore the causal relationship behind the data, determine the dynamic dependency between nodes, and provide an intuitive basis for control decision-making;

[0078] Step 5: elastic time-varying dynamic Bayesian network parameter learning;

[0079] After learning the elastic time-varying dynamic Bayesian network structure of the heavy medium coal preparation process, parameter learning is performed; given the conditional probability table of each node, the elastic time-varying dynamic Bayesian network is used to perform parameter learning to obtain a non-stationary elastic time-varying dynamic Bayesian network model, which provides a basis for the inference control decision-making plan; Table 3 is the conditional probability table of the quality variable overflow ash H when abnormal conditions occur in the heavy medium coal preparation process. Process variable 1 indicates normal, 2 indicates that the abnormal value is too small, and 3 indicates that the abnormal value is too large. In the reasoning, the abnormal evidence status is set to 2. When abnormal conditions occur in the coal preparation process, the probability density function is used to infer the abnormal time slice, and the scatter plot and histogram of the abnormal data distribution are drawn, as shown in Figure 3. Figure 6 and Figure 7 As shown; then, the conditional probability distribution is used to determine the variables that cause abnormal conditions, infer the state of each variable at the current moment, and infer the optimal safety control decision plan;

[0080] Table 3: Conditional probability table of overflow ash H in heavy medium coal preparation process

[0081]

[0082]

[0083] Step 6: Determine the time slice where the abnormality occurs and the variables that cause the abnormal condition;

[0084] The online abnormal data phenomenon variables are input as evidence into the non-stationary elastic time-varying dynamic Bayesian network model. The non-stationary elastic time-varying dynamic Bayesian network model accurately locates the time slices of the heavy medium coal preparation process where the abnormality occurs based on the learned structure and parameters, and determines the process variables that cause the abnormality.

[0085] Step 7: Infer the control decision plan;

[0086] The expected quality index and the state value of the variable are input into the local Bayesian network structure for reasoning, and the posterior probability of the operation node is obtained. According to the principle of maximum posterior probability, the adjustment plan of the heavy medium coal preparation process decision, that is, the control decision, is obtained; as shown in Table 4, four process cases are selected for verification. Among them, "↑" indicates that the variable adjustment direction increases, "ˉ" indicates that the adjustment direction decreases, and "-" indicates that it remains unchanged.

[0087] Table 4: Strategy for adjusting control variables in heavy medium coal preparation process

[0088]

[0089]

[0090] Step 8: Implement control decisions;

[0091] The controller adjusts the operating parameters of the heavy medium coal preparation system according to the obtained control decision. If the abnormal condition is eliminated, it enters the normal working mode. If the abnormal condition is not eliminated, the online abnormal data is collected again and step seven is re-executed. The online abnormal data phenomenon variables are continued to be used as evidence information to infer the control decision plan until the abnormal condition is eliminated.

[0092] Furthermore, the method of the present invention can be verified by observing the change of overflow ash after implementing the control decision:

[0093] In order to verify the effectiveness of the elastic time-varying dynamic Bayesian network modeling method and case adjustment strategy, four cases can be simulated on the heavy medium coal preparation process simulation platform. When abnormal conditions occur, it is observed whether the cyclone overflow ash content can be restored to the normal range. Figure 8 As shown in the figure, the threshold of the fine ore position of the heavy medium process is set to 0.11. When the threshold is exceeded, it is input as evidence into the established elastic time-varying dynamic Bayesian network model to infer the control decision plan. Figure 8 It can be seen that in the study of each case, when an abnormality occurs, the decision-making plan inferred by the model proposed in the present invention can quickly eliminate the abnormality and ensure that the non-stationary heavy medium coal preparation process can operate safely and stably.

[0094] The present invention aims at the problem that the non-stationary and time-varying characteristics in the heavy medium coal preparation process lead to frequent abnormal working conditions. A self-healing control decision method for non-stationary complex industrial processes based on an elastic time-varying dynamic Bayesian network is proposed. The model is based on a vector autoregression model and is used to describe the time dynamics between multiple variables. By dynamically adjusting the model structure and parameters, the changes in the system are reflected in real time to ensure that the model accurately describes the system behavior. By collecting data from the coal preparation plant, the stationarity of the data is tested using the ADF square root algorithm. For non-stationary data, the difference method is used to stabilize it, and the stationary data is fused with the original stationary data for correlation analysis. The strength and direction of the relationship between the variables are determined to understand the relationship between the variables, so as to make more accurate decisions when anomalies occur. After data preprocessing, the data is input into the elastic time-varying dynamic Bayesian network model to learn the network structure and parameters. According to the learned structure and parameters, the abnormal time slice is accurately located in combination with the probability density function, and the process variables causing the abnormality are determined. The abnormal data is input into the network model as evidence to infer a decision plan that can eliminate the abnormal working condition and convert it into a robust control operation. Finally, the effectiveness of the proposed method is verified in the heavy medium coal preparation process, demonstrating the application potential of the elastic time-varying dynamic Bayesian network model in non-stationary industrial processes.

Claims

1. A self-healing control method for a complex industrial process based on an elastic time-varying Bayesian network, comprising a heavy medium coal preparation system, the heavy medium coal preparation system comprising a conveyor, a heavy medium cyclone, an arc screen A, a clean coal de-mediation screen, an arc screen B, a gangue de-mediation screen, a dilution medium barrel, a dilute medium pump, a dilute medium magnetic separator, a water adding pump, a adding medium pump, a adding medium magnetic separator, a qualified medium barrel and a pressure pump; the feed port of the heavy medium cyclone is connected to the discharge end of the conveyor, the feed port of the arc screen A is connected to the overflow outlet of the heavy medium cyclone, the feed port of the clean coal de-mediation screen is connected to the discharge port of the arc screen A, the screen outlet of the clean coal de-mediation screen is used to output clean coal, and the screen outlet is respectively connected to a first drainage pipeline and a second drainage pipeline; the feed port of the arc screen B is connected to the underflow outlet of the heavy medium cyclone, the feed port of the gangue de-mediation screen is connected to the discharge port of the arc screen B, and the gangue de-mediation screen The outlet on the screen is used to output coal gangue, and its outlet under the screen is respectively connected to the third drainage pipeline and the fourth drainage pipeline; the feed port on the top of the dilution medium barrel is respectively connected to the outlet end of the first drainage pipeline and the outlet end of the third drainage pipeline; the inlet end of the dilution pump is connected to the dilution medium outlet at the bottom of the dilution medium barrel; the inlet end of the dilute medium magnetic separator is connected to the outlet end of the dilution pump; the water inlet end of the water pump is connected to the water source; the inlet end of the medium pump is connected to the medium source; the inlet end of the medium magnetic separator is connected to the outlet end of the medium pump; the feed port on the top of the qualified medium barrel is respectively connected to the outlet end of the second drainage pipeline, the outlet end of the fourth drainage pipeline, the outlet end of the dilute medium magnetic separator, the water outlet end of the water pump and the outlet end of the medium magnetic separator; the inlet end of the pressure pump is connected to the outlet end at the bottom of the qualified medium barrel, and its outlet end is connected to the feed port of the heavy medium cyclone; It is characterized in that The following steps are involved: Step 1: Establish a heavy medium coal preparation system and determine the variables; A heavy medium coal separation system is established by using a conveyor, a heavy medium cyclone, a curved screen A, a clean coal de-mediating screen, a curved screen B, a gangue de-mediating screen, a dilution medium barrel, a dilution medium pump, a dilution medium magnetic separator, a water pump, a medium pump, a medium magnetic separator, a qualified medium barrel and a pressure pump; According to the operating characteristics of the heavy medium coal preparation system, the measurement variables and operation variables in the heavy medium coal preparation system are determined; there are 6 measurement variables, which are the double-layer screen discharge flow, single-layer screen discharge flow, mixed medium barrel slurry density, cyclone medium density, qualified medium barrel medium density, cyclone overflow ash content; there are 2 operation variables, which are the raw coal bunker coal input amount and cyclone medium inlet pressure; Step 2: Collect coal preparation process data and use the ADF square root algorithm to test the stability of the industrial process; S21: During the operation of the heavy medium coal preparation system, the coal preparation process data is collected by using the sensor group arranged in the heavy medium coal preparation system, and the data is output to the controller. The controller processes the coal preparation process data according to the internal clock module to obtain time series data; S22: Use the ADF square root algorithm to test the stationarity of time series data; S221: According to the characteristics of the time series, select the ADF test model with constant term and time trend, and use formula (1) to obtain the first-order difference Δy of the time series t ; In the formula, Δy t Represents y t -y t-1 ;y t-1 is the time series value lagged one order; μ is the constant term; λt is the time trend term; α is the most important coefficient in the ADF test, which is used to determine the existence of a unit root; β i is the coefficient of the difference term, which is used to eliminate the high-order autocorrelation in the sequence; ∈ t is the white noise error term; p is the lag order; S222: Estimate the parameters in the ADF test model using the least squares method; S223: Calculate the ADF statistic, which is the estimated α divided by its standard error; S224: Construct hypothesis test, null hypothesis H 0 The unit root exists. In this case, the data is non-stationary. The alternative hypothesis H 1 The unit root does not exist, and the data is stable; Compare the calculated ADF statistic with the corresponding critical value; the critical value comes from the Dickey-Fuller distribution table; if the ADF statistic is less than the critical value, the null hypothesis is rejected and the time series is considered to be stationary; if the ADF statistic is greater than the critical value, the null hypothesis cannot be rejected and the time series is considered to be non-stationary; Step 3: Stabilize the non-stationary data; For non-stationary data, the trend and periodic components in the time series are eliminated by difference until the data reaches a stable state. In this process, the unit root test or the time series diagram of the differenced data is used to check whether the differenced data reaches a stable state. Step 4: Conduct structural learning and establish a flexible time-varying dynamic Bayesian network model; S41: Fusing the stabilized data with the original stabilized data to obtain fused data, performing correlation analysis on the fused data to determine the strength and direction of the relationship between multiple variables, and presenting the relationship between the variables in the form of a graph; S42: The key variables that affect the coal preparation effect are used as nodes in the elastic time-varying dynamic Bayesian network, and an elastic time-varying dynamic Bayesian network model that reflects the causal relationship between industrial process variables is constructed to capture the changes in the dependency relationship of the heavy medium coal preparation system in different time periods; S43: Use fused data to train a flexible time-varying dynamic Bayesian network model to mine the causal relationship behind the data and determine the dynamic dependency between nodes; Step 5: elastic time-varying dynamic Bayesian network parameter learning; Given the conditional probability table of each node, the elastic time-varying dynamic Bayesian network is used to learn parameters, and a non-stationary elastic time-varying dynamic Bayesian network model is obtained, which provides a basis for the inference control decision plan; when an abnormal condition occurs in the coal selection process, the probability density function is used to infer the abnormal time slice; then, the conditional probability distribution is used to determine the variables causing the abnormal condition, and the state of each variable at the current moment is inferred, and the optimal safety control decision plan is inferred; Step 6: Determine the time slice where the abnormality occurs and the variables that cause the abnormal condition; The online abnormal data phenomenon variables are input as evidence into the non-stationary elastic time-varying dynamic Bayesian network model. The non-stationary elastic time-varying dynamic Bayesian network model accurately locates the time slices of the heavy medium coal preparation process where the abnormality occurs based on the learned structure and parameters, and determines the process variables that cause the abnormality. Step 7: Infer the control decision plan; The expected quality index and the state value of the variable are input into the local Bayesian network structure for reasoning, and the posterior probability of the operation node is obtained. According to the principle of maximum posterior probability, the adjustment plan of the heavy medium coal preparation process decision, i.e., the control decision, is obtained; Step 8: Implement control decisions; The controller adjusts the operating parameters of the heavy medium coal preparation system according to the obtained control decision. If the abnormal condition is eliminated, it enters the normal working mode. If the abnormal condition is not eliminated, the online abnormal data is collected again and step seven is re-executed. The online abnormal data phenomenon variables are continued to be used as evidence information to infer the control decision plan until the abnormal condition is eliminated.

Citation Information

Patent Citations

  • Dense medium coal dressing process safety operation control method based on active learning and BN

    CN112415894A

  • Multi-step prediction method for clean coal quality in dense medium coal separation process based on LSTM (Long Short Term Memory)

    CN114330930A

  • Inter-batch quality control method for flexible manufacturing of aerospace initiating explosive device charging and pressing

    CN115016421A

  • Intrinsic viscosity on-line soft measurement method based on double Gaussian Bayesian network

    CN116127841A

  • Coal slime flotation safe operation control method based on distributed dynamic Bayesian network

    CN116899758A