Petroleum refining device protection layer failure detection and diagnosis method and system based on multi-source data fusion

Abstract

The invention belongs to the technical field of industrial control system safety, and discloses a petroleum refining device protection layer failure detection method based on multi-source data fusion. According to the method, through normalization and alarm levels, data are collected on petroleum refining device simulation software, and a normalized process monitoring data training matrix and a statistical alarm information matrix are obtained; screening a variable set for failure detection and diagnosis of the protection layer according to a maximum correlation-minimum redundancy principle; and fusing and utilizing a discrete Bayesian classifier to obtain a protection layer failure detection and diagnosis result on statistical alarm information data, and utilizing a continuous Bayesian classifier to obtain a protection layer failure detection and diagnosis result on a process monitoring data training data matrix, so as to obtain a final protection layer failure detection and diagnosis result. Compared with an existing method, statistical alarm information data are creatively fused, and interference of noise in process monitoring data on protection layer failure detection and diagnosis is eliminated.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial control system safety, and in particular relates to a method and system for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion. Background Art

[0002] The petrochemical industry itself is a high-risk industry that is flammable and explosive. The production process involves complex processes and chemical reactions, which hide a variety of safety risks. These risks not only pose a great threat to the safety of personnel, but may also affect production efficiency. Therefore, ensuring the safe operation of petroleum refining equipment is the basis for improving its overall efficiency.

[0003] Protection layer analysis is a typical semi-quantitative hazard analysis method, which usually includes multiple independent protection layers, such as the basic process control system layer, the alarm and personnel intervention layer, the safety instrument system layer, the physical protection layer, the emergency response layer, etc. These protection layers are independent, effective and auditable, and have been widely used in petroleum refining plants. In particular, the basic process control system layer, the alarm and personnel intervention layer, and the safety instrument system layer are the three most critical independent protection layers in the operation of petroleum refining plants. Therefore, it is particularly important to design failure detection and diagnosis methods for these three layers.

[0004] At present, the failure detection and diagnosis method of the protective layer of the petroleum refining device mainly relies on process monitoring data. CN201610131396.2 uses process monitoring data to realize a failure detection and diagnosis method for the refining process. Although the process monitoring data can reflect the operating status of the device to a certain extent, during the occurrence and recovery of the protective layer failure and the switching between different protective layer failure states, the noise data may interfere with the detection and diagnosis of the protective layer failure, resulting in wrong judgments. Statistical alarm information can effectively reduce the interference of noise data due to its discrete characteristics and is often used to solve this problem. CN201610133660.6 provides a method for generating a dynamic alarm threshold. However, it is often difficult to accurately determine the specific type of protective layer failure by relying solely on statistical alarm information. And neither of the above two inventions conducts failure analysis of petroleum refining devices from the perspective of the protective layer. Therefore, relying solely on statistical alarm information cannot meet the needs of detecting and diagnosing the failure of the protective layer of petroleum refining devices.

[0005] Through the above analysis, the problems and defects of the prior art are as follows:

[0006] The detection and diagnosis methods of protective layer failure in petroleum refining units mainly rely on process monitoring data. Although process monitoring data can reflect the operating status of the unit to a certain extent, during the occurrence and recovery of protective layer failure and the switching between different protective layer failure states, noise data may interfere with the detection and diagnosis of protective layer failure, leading to wrong judgments. It is often difficult to accurately judge the specific type of protective layer failure by relying solely on statistical alarm information. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention provides a method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion.

[0008] The present invention is implemented as follows: a method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion comprises:

[0009] Step 1, determine the sliding window length and sampling interval;

[0010] Step 2, collecting process monitoring data and updating the process monitoring data matrix based on the step size;

[0011] Step 3, normalizing the process monitoring data by column vector;

[0012] Step 4, generating statistical alarm information by normalizing process monitoring data;

[0013] Step 5, filter the variable set according to the maximum correlation-minimum redundancy index;

[0014] Step 6: Utilize the statistical alarm information data and use the discrete Bayesian classifier to obtain the protection layer failure detection and diagnosis results c a ;

[0015] Step 7: Using the process monitoring data, a continuous Bayesian classifier is used to obtain the protection layer failure detection and diagnosis results c x ;

[0016] Step 8, merge c a and c x , and obtain the final protection layer failure detection and diagnosis results c f .

[0017] Further, the process monitoring data is collected based on the sliding window length and the sampling interval, including:

[0018] The process monitoring data training matrix for determining the initial sampling time is: X0∈R n×w ; Where n is the number of overall variables monitored in the process, and w is the sampling window length;

[0019] After k samplings, the process monitoring data training matrix is ​​updated to: XkΔt ; Where Δt is the sampling interval;

[0020] The process monitoring data training matrix is ​​centrally processed to generate a normalized process monitoring data training matrix, including:

[0021] By formula: Determine the normalized process monitoring data training matrix

[0022] The statistical alarm information data is calculated according to the normalized process monitoring data by the formula:

[0023] Generate; where σ j is the standard deviation of the jth variable, p is the alarm level constant, the larger the p is, the higher the alarm level is, that is, the greater the degree to which the current operating state of the petroleum refining unit deviates from the normal state.

[0024] Furthermore, the statistical alarm information data training matrix generated after the process monitoring data is updated k times is: kΔt ∈R n×w .

[0025] Furthermore, according to the maximum relevance-minimum redundancy guiding principle, variable selection is performed based on the normalized process monitoring data training matrix, including:

[0026] Determine the number m of features to be retained in the target set;

[0027] Based on the formula Determine the set S that is most relevant to the label m ,in

[0028]

[0029] The above formula is derived from the formula Approximate calculations;

[0030] Using the formula Determine the elements with the least redundancy within the set;

[0031] Combining the most relevant indicators with the label and the indicators with the least redundancy within the set,

[0032] Determine the feature set S for training m .

[0033] Furthermore, the discrete Bayesian classifier follows the formula:

[0034]

[0035] The probability of failure of each type of protective layer is calculated; thus, based on the formula: The case c with the highest probability of protective layer failure is obtained a .

[0036] Further, the continuous Bayesian classifier, according to the formula:

[0037] Calculate the probability of each type of protection layer failure;

[0038] Approximate P(x(t k )|c), use Instead of P(x(t k )|c),

[0039] in

[0040] Where K is the kernel function. This implementation adopts the multivariate normal kernel function, namely:

[0041] H is a symmetric positive definite matrix:

[0042]

[0043] get

[0044] then, Therefore, based on the formula:

[0045] The case c with the highest probability of protective layer failure is obtained x ;

[0046] Fusion of discrete Bayesian classifier and continuous Bayesian classifier results, based on the formula:

[0047]

[0048] Get the final protection layer failure detection result c f .

[0049] Another object of the present invention is to provide a petroleum refining device protective layer failure detection system based on multi-source data fusion, comprising:

[0050] Parameter setting module, used to determine parameters such as sliding window length and sampling interval;

[0051] A data acquisition module is used to collect the values ​​of control variables of each control loop from the petroleum refining device simulation software;

[0052] A data update module is used to update the process monitoring data training matrix based on the sampling interval to achieve the dynamic effect of the system;

[0053] A normalization processing module is used to perform column vector normalization on process monitoring data through a formula;

[0054] A statistical alarm information calculation module, used for generating statistical alarm information data using a normalized process monitoring data training matrix;

[0055] A variable selection module, used to select the variable set most relevant to the protection layer failure detection and diagnosis according to the maximum relevance-minimum redundancy index;

[0056] A discrete Bayesian classifier module is used to obtain a discrete Bayesian classifier protection layer failure detection and diagnosis result by using statistical alarm information data;

[0057] A continuous Bayesian classifier module is used to obtain the failure detection and diagnosis results of the continuous Bayesian classifier protection layer by using the process monitoring data;

[0058] The protection layer failure determination module is used to integrate module 7 and module 8 to obtain the final protection layer failure detection and diagnosis results.

[0059] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion.

[0060] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion.

[0061] Another object of the present invention is to provide an information data processing terminal, which is used to implement the petroleum refining device protective layer failure detection system based on multi-source data fusion.

[0062] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0063] First, the present invention collects data on the petroleum refining device simulation software through normalization and alarm level, obtains the normalized process monitoring data training matrix and the statistical alarm information matrix; screens the variable set for protection layer failure detection and diagnosis according to the maximum relevance-minimum redundancy principle; integrates the protection layer failure detection and diagnosis results obtained on the statistical alarm information data using a discrete Bayesian classifier, and obtains the protection layer failure detection and diagnosis results on the process monitoring data training data matrix using a continuous Bayesian classifier, and obtains the final protection layer failure detection and diagnosis results. Compared with the existing methods, this paper creatively integrates the statistical alarm information data, eliminating the interference of noise in the process monitoring data on the protection layer failure detection and diagnosis.

[0064] Second, the expected benefits and commercial value of the technical solution of the present invention after transformation are: it can provide protection layer failure detection and diagnosis technology for domestic oil refineries, reducing the major harm and economic losses caused by protection layer failure. The price of protection layer failure analysis varies from hundreds of thousands to millions depending on the number of on-site operation points. At present, there are hundreds of refineries in my country. Considering that the accuracy of the present invention in protection layer failure detection and diagnosis has reached a practical level and is more convenient, it can be used as an alternative solution, so it has great commercial value.

[0065] The technical solution of the present invention fills the technical gap in the industry at home and abroad: At present, the failure analysis of petroleum refining equipment at home and abroad focuses more on the failure of specific controllers, actuators and sensors in the basic control system layer, ignoring the failure state of other protection layers. The technical solution proposed by the present invention expands the failure analysis of petroleum refining equipment to the basic process control system layer, alarm and personnel intervention layer, and safety instrument system layer, filling the technical gap.

[0066] Does the technical solution of the present invention solve the technical problems that people have been eager to solve but have never been able to solve: (1) The existing technologies mostly focus on the failure of specific controllers, actuators and sensors in the basic control system layer, and ignore the failure states of other protection layers. The technical solution proposed in the present invention expands the failure analysis of petroleum refining equipment to the basic process control system layer, the alarm and personnel intervention layer, and the safety instrument system layer, filling the technical gap. (2) The failure analysis of existing petroleum refining equipment is mostly based on process monitoring data, which makes it difficult to deal with the impact of noise data on failure analysis during the failure state conversion process. The present invention uses a fusion strategy of process monitoring data and statistical alarm information to make up for this deficiency.

[0067] The technical solution of the present invention overcomes technical prejudice: the failure analysis of the protective layer of the petroleum refining device is mostly based on process monitoring data, because process monitoring data is the data that is relatively easy to obtain and explicitly reflects the operating status of the petroleum refining device. The present invention introduces statistical alarm information into the failure analysis of the protective layer, aiming to comprehensively obtain non-intuitive implicit features, so as to overcome the influence of noise data on failure analysis in the failure state conversion process that has been neglected for a long time. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of a method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion provided by an embodiment of the present invention.

[0069] Figure 2 It is a structural block diagram of a petroleum refining device protective layer failure detection system based on multi-source data fusion provided by an embodiment of the present invention.

[0070] Figure 3 This is an example diagram of the original collected data of the process monitoring of the control variable TI1150 in the protection layer of the basic control process system of the petroleum refining device provided by the embodiment of the present invention.

[0071] Figure 4 This is an example diagram of process monitoring data after normalization of the control variable TI1150 in the protection layer of the basic control process system of the petroleum refining device provided by the embodiment of the present invention.

[0072] Figure 5 This is an example diagram of statistical alarm information data generated by process monitoring data after normalization of the control variable TI1150 in the protection layer of the basic control process system of the petroleum refining device provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0074] like Figure 1 As shown, a method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion provided by an embodiment of the present invention comprises the following steps:

[0075] S101, determining the sliding window length and sampling interval;

[0076] S102, collecting process monitoring data, and updating the process monitoring data matrix based on the step size;

[0077] S103, performing column vector normalization on the process monitoring data;

[0078] S104, generating statistical alarm information by normalizing process monitoring data;

[0079] S105, screening the variable set according to the maximum correlation-minimum redundancy index;

[0080] S106, using the statistical alarm information data, using a discrete Bayesian classifier to obtain the protection layer failure detection and diagnosis results c a ;

[0081] S107, using process monitoring data, using continuous Bayesian classifier to obtain protection layer failure detection and diagnosis results c x ;

[0082] S108, fusion c a and c x , and obtain the final protection layer failure detection and diagnosis results c f .

[0083] In the first step (S101) of the embodiment of the present invention, by setting the length of the sliding window and the sampling interval, it is ensured that continuous process data can be obtained during the monitoring process. The length of the sliding window determines the time span of each data analysis, and the sampling interval determines the frequency of data collection. The reasonable selection of these two parameters is crucial for subsequent data updating, processing and analysis, and can ensure the real-time and continuity of the monitoring data.

[0084] Step S102 collects process monitoring data and updates the process monitoring data matrix based on the set step size. This step aims to dynamically acquire and update the monitoring data matrix so that the system can reflect the working status of the protective layer in the oil refining unit in real time. The updated data matrix will be used for subsequent data processing and analysis to provide a basis for protective layer failure detection.

[0085] In step S103, the process monitoring data is subjected to column vector normalization. The purpose of normalization is to unify the dimensions of each variable so that monitoring data of different magnitudes can be compared under the same standard. Through this process, the influence caused by the dimension difference in the original data can be eliminated, thereby providing a more reliable basis for subsequent data analysis.

[0086] Step S104 generates statistical alarm information based on the normalized process monitoring data. By statistically analyzing the monitoring data, the system can detect abnormal fluctuations or abnormal patterns and generate alarm signals. These alarm information can reflect the potential failure risk of the protective layer of the oil refining device in real time and provide preliminary warning for subsequent failure detection and diagnosis.

[0087] In step S105, the maximum relevance-minimum redundancy (mRMR) index is used to screen the variable set, select the characteristic variables with the greatest correlation with the failure of the protective layer, and remove redundant information to improve data processing efficiency. Then, in step S106, the failure of the protective layer is detected and diagnosed using a discrete Bayesian classifier using these characteristic variables and statistical alarm information. Through the discrete Bayesian classifier, the probability of failure can be modeled and a preliminary diagnosis result can be obtained.

[0088] In step S107, a continuous Bayesian classifier is used to perform further failure detection and diagnosis based on process monitoring data. This classifier takes into account the continuity of the data and can more accurately identify potential failure risks. Finally, in step S108, the detection results of the discrete Bayesian classifier and the continuous Bayesian classifier are fused to obtain the final protection layer failure detection and diagnosis results. Through the fusion of multi-source data, the accuracy and robustness of the detection can be further improved to ensure the safety and stability of the oil refining unit.

[0089] The method for collecting process monitoring data based on sliding window length and sampling interval provided by the embodiment of the present invention includes:

[0090] The process monitoring data training matrix for determining the initial sampling time is: X0∈R n×w ; Where n is the number of overall variables monitored in the process, and w is the sampling window length;

[0091] After k samplings, the process monitoring data training matrix is ​​updated to: X kΔt ; Where Δt is the sampling interval;

[0092] The process monitoring data training matrix is ​​centrally processed to generate a normalized process monitoring data training matrix, including:

[0093] By formula: Determine the normalized process monitoring data training matrix

[0094] The statistical alarm information data is calculated according to the normalized process monitoring data by the formula:

[0095] Generate; where σ j is the standard deviation of the jth variable, p is the alarm level constant, the larger the p is, the higher the alarm level is, that is, the greater the degree to which the current operating state of the petroleum refining unit deviates from the normal state.

[0096] The statistical alarm information data training matrix generated after the process monitoring data provided by the embodiment of the present invention is updated k times of sampling is: kΔt ∈R n×w .

[0097] According to the maximum correlation-minimum redundancy guiding principle provided by the embodiment of the present invention, variable selection is performed on the basis of the normalized process monitoring data training matrix, including:

[0098] Determine the number m of features to be retained in the target set;

[0099] Based on the formula Determine the set S that is most relevant to the label m ,in

[0100]

[0101] The above formula is derived from the formula Approximate calculations;

[0102] Using the formula Determine the elements with the least redundancy within the set;

[0103] Combining the most relevant indicators with the label and the indicators with the least redundancy within the set,

[0104] Determine the feature set S for training m .

[0105] The discrete Bayesian classifier provided in the embodiment of the present invention is according to the formula:

[0106]

[0107] The probability of failure of each type of protective layer is calculated; thus, based on the formula: The case c with the highest probability of protective layer failure is obtained a .

[0108] The continuous Bayesian classifier provided in the embodiment of the present invention is according to the formula:

[0109] Calculate the probability of each type of protection layer failure;

[0110] Approximate P(x(t k )|c), use Instead of P(x(t k )|c),

[0111] in

[0112] Where K is the kernel function. This implementation adopts the multivariate normal kernel function, namely:

[0113] H is a symmetric positive definite matrix:

[0114]

[0115] get

[0116] then, Therefore, based on the formula:

[0117] The case c with the highest probability of protective layer failure is obtained x ;

[0118] Fusion of discrete Bayesian classifier and continuous Bayesian classifier results, based on the formula:

[0119]

[0120] Get the final protection layer failure detection result c f .

[0121] like Figure 2 As shown, a petroleum refining device protective layer failure detection system based on multi-source data fusion provided by an embodiment of the present invention includes:

[0122] Parameter setting module, used to determine parameters such as sliding window length and sampling interval;

[0123] A data acquisition module is used to collect the values ​​of control variables of each control loop from the petroleum refining device simulation software;

[0124] A data update module is used to update the process monitoring data training matrix based on the sampling interval to achieve the dynamic effect of the system;

[0125] A normalization processing module is used to perform column vector normalization on process monitoring data through a formula;

[0126] A statistical alarm information calculation module, used for generating statistical alarm information data using a normalized process monitoring data training matrix;

[0127] A variable selection module, used to select the variable set most relevant to the protection layer failure detection and diagnosis according to the maximum relevance-minimum redundancy index;

[0128] A discrete Bayesian classifier module is used to obtain a discrete Bayesian classifier protection layer failure detection and diagnosis result by using statistical alarm information data;

[0129] A continuous Bayesian classifier module is used to obtain the failure detection and diagnosis results of the continuous Bayesian classifier protection layer by using the process monitoring data;

[0130] The protection layer failure determination module is used to integrate module 7 and module 8 to obtain the final protection layer failure detection and diagnosis results.

[0131] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion.

[0132] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion.

[0133] Another object of the present invention is to provide an information data processing terminal, which is used to implement the petroleum refining device protective layer failure detection system based on multi-source data fusion.

[0134] The present invention is specifically implemented:

[0135] The embodiment of the present invention provides a method for detecting and diagnosing failure of a protective layer of a petroleum refining device based on multi-source data fusion, comprising:

[0136] Step 1: Determine the sliding window length and sampling interval. Combined with the specific petroleum refining unit simulation software, determine the sliding window length and sampling interval based on the data update frequency. The window length indicates how many sampling data the process monitoring data matrix contains, and the sampling interval indicates how often the data acquisition script performs the data acquisition operation.

[0137] Step 2: Collect process monitoring data and update the process monitoring data matrix based on the sampling step. Taking the sampling results of some control variables in the protection layer of the basic control process system of the reaction regeneration module of the petroleum refining unit as an example, the process monitoring data collected in this step is Each column of data represents: main riser outlet temperature measurement point TI1149 (degrees Celsius), main riser outlet temperature measurement point TI1150 (degrees Celsius), main riser outlet temperature measurement point TI1151 (degrees Celsius) and regeneration slide valve opening TV1101 (percentage), 5 rows of data represent a sliding window size of 5. This is only used to help the technical staff to better understand the implementation of the present invention, not a fixed sliding window size and variable set. In actual applications, the sliding window size will be much larger than 5. In process acquisition, all controlled variables in the control loop are collected as much as possible for screening.

[0138] According to the data updating method of claim 2, the process monitoring data matrix after the 45th update is: Where X 45 Rows 1 to 5 are the data obtained from sampling 221 to 225.

[0139] Step 3: Normalize the column vector of the process monitoring data. In this step, the column vector is normalized by the formula: Determine the normalized process monitoring data training matrix, with X 45 For example, after normalization:

[0140]

[0141] Step 4: Generate statistical alarm information through normalized process monitoring data. In this step, the statistical alarm information data is calculated by the formula: Generate; where σ j is the standard deviation of the jth variable, p is the alarm level constant; 45 Taking the generated statistical alarm data as an example, the statistical alarm information matrix obtained is: The value of p is 3, which is consistent with the value of practical application.

[0142] Step 5: Filter the variable set according to the maximum correlation-minimum redundancy index. In this step, the maximum correlation index is calculated based on the formula: Determine the set S that is most relevant to the label m ,in Since the joint probability density of continuous variables is usually difficult to obtain, the above formula is replaced by Approximate calculation; the minimum redundancy index calculation method is: using the formula Determine the element with the least redundancy within the set; combine the most relevant indicators for the label and the indicators with the least redundancy within the set, based on Determine the feature set S for training m .

[0143] Step 6: Utilize the statistical alarm information data and use the discrete Bayesian classifier to obtain the protection layer failure detection and diagnosis results c a In this step, the statistical alarm information discrete Bayesian classifier uses the statistical alarm information training matrix to calculate the probability of each type of protection layer failure: Therefore, based on the formula: The case c with the highest probability of protective layer failure is obtained a Here c a This is a type of protection layer failure condition, such as "Basic control process protection layer failure - main lift pipe regeneration slide valve is fixed at a larger opening", "Basic control process protection layer failure - smoke exhaust fan is fixed at a smaller speed", etc.

[0144] Step 7: Using the process monitoring data, the continuous Bayesian classifier obtains the protection layer failure detection and diagnosis result F2. In this step, the process monitoring data training matrix is ​​used to calculate the probability of each type of protection layer failure: Since in reality P(x(t k )|c) is often not available, so in this step, the multivariate kernel density estimation method is used to approximate P(x(t k )|c), use Instead of P(x(t k )|c), where Where K is the kernel function. This implementation adopts the multivariate normal kernel function, namely: H is a symmetric positive definite matrix: get

[0145] Finally, the formula: The classification results of the continuous Bayesian classifier of process monitoring data are obtained. Therefore, based on the formula: The case c with the highest probability of protective layer failure is obtained x , here c x This is a type of protection layer failure condition, such as "Basic control process protection layer failure - main lift pipe regeneration slide valve is fixed at a larger opening", "Basic control process protection layer failure - smoke exhaust fan is fixed at a smaller speed", etc.

[0146] Step 8: Integrate the protection layer failure detection and diagnosis results obtained in step 6 a and the protective layer failure detection and diagnosis result c obtained in step 7 x , and obtain the final protection layer failure detection and diagnosis results c f In this step, based on the formula:

[0147]

[0148] Get the final protection layer failure detection result c f Furthermore, c is the search space for protection layer failure detection and diagnosis, c f It is a possible value state of c, such as: "The basic control process protection layer fails - the main lift pipe regeneration slide valve is fixed at a larger opening", "The basic control process protection layer fails - the smoke exhaust fan is fixed at a smaller speed", etc.

[0149] like Figure 2 As shown, an embodiment of the present invention provides a petroleum refining device protective layer failure detection and diagnosis system based on multi-source data fusion, comprising:

[0150] Module 1, parameter setting module, is used to determine parameters such as sliding window length and sampling interval.

[0151] Module 2, data acquisition module, is used to collect the values ​​of control variables of each control loop from the petroleum refining device simulation software.

[0152] Module 3, data update module, is used to update the process monitoring data training matrix based on the sampling interval to achieve the dynamic effect of the system.

[0153] Module 4, a normalization processing module, is used to perform column vector normalization on process monitoring data through calculation.

[0154] Module 5, a statistical alarm information calculation module, is used to generate statistical alarm information data using a normalized process monitoring data training matrix.

[0155] Module 6, a variable selection module, is used to select a set of variables most relevant to the protection layer failure detection and diagnosis according to the maximum relevance-minimum redundancy index.

[0156] Module 7, a discrete Bayesian classifier module, is used to obtain the failure detection and diagnosis results of the discrete Bayesian classifier protection layer by using statistical alarm information data.

[0157] Module 8, a continuous Bayesian classifier module, is used to obtain protection layer failure detection and diagnosis results of a continuous Bayesian classifier using process monitoring data.

[0158] Module 9, a protection layer failure determination module, is used to integrate module 7 and module 8 to obtain the final protection layer failure detection and diagnosis results.

[0159] In addition, module 3, parameter update module is specifically used to: determine the process monitoring data training as: X0∈R w×n ; Where n is the number of overall variables for process monitoring, and w is the length of the sampling window; the process monitoring data is updated once every sampling interval Δt, and the process monitoring data matrix after k updates is:

[0160] In order to enable those skilled in the art to better understand the present invention, an example of process monitoring data of a variable is listed below. Here, the exemplary implementation mode of the present invention and its description are only used to explain the present invention, but not to limit the present invention. Figure 3 As shown, this embodiment takes the variable petroleum refining unit reaction regeneration module basic control process protective layer main riser temperature TI1150 as a sampling example of the monitoring variable, where the sliding window size is 4496. At the 11th, 1609th and 3006th moments, the regeneration slide valve TV1101 that controls the variable is manually disabled. Reflected in the process monitoring data curve, the data at these locations deviate significantly from the mean. Furthermore, the process monitoring data within the failure time is significantly different from the process monitoring data within the normal operating time period, so the data can be used to detect and diagnose the protective layer failure scenario of the petroleum refining unit.

[0161] In addition, module 4, the normalization processing module is specifically used to: Determine the normalized process monitoring data training matrix

[0162] In order to enable those skilled in the art to better understand the present invention, an example of normalizing process monitoring data of a variable is listed below. Here, the exemplary implementation of the present invention and its description are only used to explain the present invention, but not to limit the present invention. Figure 4 As shown, this embodiment normalizes the process monitoring data values ​​of TI1150 to a distribution centered at 0. Figure 4 and Figure 3 ,The normalization processing does not lose the abnormal performance of the process monitoring data within the three protection layer failure time, but the data after normalization of each process monitoring variable has smaller numerical differences, which will reduce the interference to the calculation caused by the large numerical differences of different variables due to different units.

[0163] In addition, module 5, the statistical alarm information calculation module is specifically used for: Generate statistical alarm information data; where σ j is the standard deviation of the jth variable, p is the alarm level constant; the generated statistical alarm information data is: A k ∈R w×n .

[0164] In order to enable those skilled in the art to better understand the present invention, an example of generating statistical alarm information based on process monitoring data is listed below. Here, the exemplary implementation of the present invention and its description are only used to explain the present invention, but not to limit the present invention. Figure 5 As shown in the figure, TI1150 obtains discretized statistical alarm information through function calculation. Compared with process monitoring data, the discretized statistical alarm information does not have the noise data brought by the protection layer failure process, which alleviates the impact of noise data in the protection layer failure generation and recovery stages on the protection layer failure detection and diagnosis.

[0165] In addition, module 6, the variable selection module is specifically used to: Based on the formula: Determine the set S that is most relevant to the label m ,in Since the joint probability density of continuous variables is usually difficult to obtain, the above formula is replaced by Approximate calculation; the minimum redundancy index calculation method is: using the formula Determine the element with the least redundancy within the set; combine the indicator most relevant to the label and the indicator with the least redundancy within the set, using Determine the feature set S for training m .

[0166] In addition, module 7, the discrete Bayesian classifier module is specifically used to calculate the probability of each type of protection layer failure: Based on the formula: The case c with the highest probability of protective layer failure is obtained a , here c a is a mapping of a type of protection layer failure condition, such as c a =1 is the protection layer failure condition "Basic control process protection layer failure - main lift pipe regeneration slide valve is fixed at a large opening", c a =2 indicates the protection layer failure condition "Basic control process protection layer failure - the range hood is fixed at a lower speed" etc.

[0167] In addition, module 8, the continuous Bayesian classifier module is specifically used to calculate the probability of each type of protection layer failure: Since in reality P(x(t k )|c) is often not available, so in this step, the multivariate kernel density estimation method is used to approximate P(x(t k )|c), use Instead of P(x(t k )|c), where Where K is the kernel function. This implementation adopts the multivariate normal kernel function, namely: H is a symmetric positive definite matrix: get The probability calculation formula for each type of protection layer failure is obtained: Based on the formula: The case c with the highest probability of protective layer failure is obtained x , here c x is a mapping of a type of protection layer failure condition, such as c x =1 is the protection layer failure condition "Basic control process protection layer failure - main lift pipe regeneration slide valve is fixed at a large opening", c x =2 indicates the protection layer failure condition "Basic control process protection layer failure - the range hood is fixed at a lower speed" etc.

[0168] In addition, module 9, the protection layer failure determination module is specifically used to: based on the formula: Get the final protection layer failure detection result c f , where c f is a mapping of a type of protection layer failure condition, such as c f =1 is the protection layer failure condition "Basic control process protection layer failure - main lift pipe regeneration slide valve is fixed at a large opening", such as c f =2 indicates the protection layer failure condition "Basic control process protection layer failure - the range hood is fixed at a lower speed" etc.

[0169] Based on the above method, the present invention also provides a program, a computer device, and a non-transitory computer-readable storage medium storing computer instructions for a petroleum refining device protective layer failure detection and diagnosis method based on multi-source data fusion, which are used to execute the above method, which will not be repeated here.

[0170] 1. Specific application fields or related products of the present invention.

[0171] Example 1: Failure analysis of protective layer of temperature control module in reaction and regeneration unit of petroleum refining device

[0172] 1. System Overview

[0173] The reaction regeneration unit is the core part of the residual oil catalytic cracking. Its stable operation directly affects the quality and output of the product. Temperature is an important control variable of the reaction regeneration unit. The task of the temperature control module is to maintain the reaction temperature stable.

[0174] 2. Protection layer model structure

[0175] The correlation between each protective layer and temperature is as follows:

[0176] The basic process control system layer includes the main riser horizontal section temperature control loop, the auxiliary riser horizontal section temperature control loop, the external heat return pipe slide valve control loop, etc.

[0177] The alarm and personnel intervention layer functions include over-temperature alarm and under-temperature alarm;

[0178] The safety instrumented system layer is designed to trigger interlocking conditions and perform key variable recovery when a temperature alarm signal is received.

[0179] 3. Type of protective layer failure

[0180] The most common types of failures occur at the basic process control system level, including failures of specific controllers, actuators, and sensors, which cause key control variables to exceed safety thresholds;

[0181] There are two main types of failures at the alarm and personnel intervention layer: false alarms and missed alarms. A false alarm means that the basic process control system layer has not failed, but the alarm and personnel intervention layer has an alarm. A missed alarm means that the basic process control system layer has failed, but the alarm and personnel intervention layer has not alarmed.

[0182] The main failures of the safety instrument system layer include: interlock trigger failure, interlock action execution failure and interlock reset failure. Interlock trigger failure means that after the alarm is triggered, the safety instrument system layer fails to effectively start the safety interlock; interlock action execution failure means that the safety instrument system layer fails to effectively execute the safety interlock action; interlock reset failure means that the key variables of the safety instrument system layer fail to return to normal after executing the safety interlock action.

[0183] 2. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0184] Example 1: Failure analysis of protective layer of temperature control module in reaction and regeneration unit of petroleum refining device

[0185] Figure 3 is the process monitoring value of the control variable, the main riser outlet temperature TI1150. The simulation system manually sets the failure of the regeneration slide valve TV1101 at the 11th, 1609th and 3306th moments to obtain the process monitoring data curve. The protective layer failure detection method of the petroleum refining device based on multi-source data fusion is used to cooperate with the opening of the reaction regeneration unit regeneration slide valve TV1101 after variable screening and the temperature values ​​of multiple measuring points to perform the protective layer failure detection. Figure 4 It is shown that the difference between the values ​​in the failure state and the normal state can be magnified by normalization. Figure 5 The discrete statistical alarm information shows that there is no noise data generated during the switching of the failure state. Finally, by fusing the results of two Bayesian classifiers, two types of failures are obtained: "Basic control process protection layer failure-main lift pipe regeneration slide valve fixed at a larger opening" and "Basic control process protection layer failure-main lift pipe regeneration slide valve fixed at a smaller opening". In addition, since TI1150 can automatically return to the normal temperature range every time this failure occurs, the classifier can determine whether the alarm and personnel intervention layer and the safety instrument system layer have failed or not.

[0186] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0187] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion, characterized in that: The following steps are involved: Step 1, determine the sliding window length and sampling interval; Step 2, collecting process monitoring data and updating the process monitoring data matrix based on the step size; Step 3, normalizing the process monitoring data by column vector; Step 4, generating statistical alarm information by normalizing process monitoring data; Step 5, filter the variable set according to the maximum correlation-minimum redundancy index; Step 6: Utilize the statistical alarm information data and use the discrete Bayesian classifier to obtain the protection layer failure detection and diagnosis results c a ; Step 7: Using the process monitoring data, a continuous Bayesian classifier is used to obtain the protection layer failure detection and diagnosis results c x ; Step 8, merge c a and c x , and obtain the final protection layer failure detection and diagnosis results c f .

2. The method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion according to claim 1, characterized in that: The process monitoring data is collected based on the sliding window length and the sampling interval, including: The process monitoring data training matrix for determining the initial sampling time is: X0∈R n×w ; Where n is the number of overall variables monitored in the process, and w is the sampling window length; After k samplings, the process monitoring data training matrix is ​​updated to: X kΔt ; Where Δt is the sampling interval; The process monitoring data training matrix is ​​centrally processed to generate a normalized process monitoring data training matrix, including: By formula: Determine the normalized process monitoring data training matrix The statistical alarm information data is calculated according to the normalized process monitoring data by the formula: Generate; where σ j is the standard deviation of the jth variable, p is the alarm level constant, the larger the p is, the higher the alarm level is, that is, the greater the degree to which the current operating state of the petroleum refining unit deviates from the normal state.

3. The method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion as claimed in claim 2, characterized in that: The statistical alarm information data training matrix generated after the process monitoring data is updated k times is: kΔt ∈R n×w .

4. The method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion as claimed in claim 2, characterized in that: According to the maximum relevance-minimum redundancy guiding principle, variable selection is performed based on the normalized process monitoring data training matrix, including: Determine the number m of features to be retained in the target set; Based on the formula Determine the set S that is most relevant to the label m ,in The above formula is derived from the formula Approximate calculations; Using the formula Determine the elements with the least redundancy within the set; Combining the most relevant indicators with the label and the indicators with the least redundancy within the set, Determine the feature set S for training m .

5. The method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion as claimed in claim 4, characterized in that: The discrete Bayesian classifier follows the formula: The probability of failure of each type of protective layer is calculated; thus, based on the formula: The case c with the highest probability of protective layer failure is obtained a .

6. The method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion as claimed in claim 5, characterized in that: Continuous Bayesian classifier, according to the formula: Calculate the probability of each type of protection layer failure; Approximate P(x(t k )|c), use Instead of P(x(t k )|c), where Where K is the kernel function. This implementation adopts the multivariate normal kernel function, namely: H is a symmetric positive definite matrix: get then, Therefore, based on the formula: The case c with the highest probability of protective layer failure is obtained x ; Fusion of discrete Bayesian classifier and continuous Bayesian classifier results, based on the formula: Get the final protection layer failure detection result c f .

7. A system for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion, which implements the method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion as claimed in any one of claims 1 to 6, characterized in that: The petroleum refining device protective layer failure detection system based on multi-source data fusion includes: Parameter setting module, used to determine parameters such as sliding window length and sampling interval; A data acquisition module is used to collect the values ​​of control variables of each control loop from the petroleum refining device simulation software; A data update module is used to update the process monitoring data training matrix based on the sampling interval to achieve the dynamic effect of the system; A normalization processing module is used to perform column vector normalization on process monitoring data through a formula; A statistical alarm information calculation module, used for generating statistical alarm information data using a normalized process monitoring data training matrix; A variable selection module, used to select the variable set most relevant to the protection layer failure detection and diagnosis according to the maximum relevance-minimum redundancy index; A discrete Bayesian classifier module is used to obtain a discrete Bayesian classifier protection layer failure detection and diagnosis result by using statistical alarm information data; A continuous Bayesian classifier module is used to obtain the failure detection and diagnosis results of the continuous Bayesian classifier protection layer by using the process monitoring data; The protection layer failure determination module is used to integrate module 7 and module 8 to obtain the final protection layer failure detection and diagnosis results.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for detecting failure of protective layer of petroleum refining equipment based on multi-source data fusion as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for detecting failure of a protective layer of a petroleum refining device based on multi-source data fusion as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the petroleum refining device protective layer failure detection system based on multi-source data fusion as described in claim 7.

Images