A method, device and storage medium for monitoring faults in blast furnace ironmaking process

Abstract

The present invention extracts the cumulative error and the rate of change based on the sample observation information, divides the data set into three data sets for KPCA modeling in combination with the original observation information, and uses the Bayesian fusion strategy to construct process monitoring statistics to monitor process faults; fully considers the implicit information of the sample observations, and improves the monitoring performance of the KPCA fault monitoring algorithm for faults such as small offsets and pulse oscillations; according to the characteristics of the fault change curve, the present invention uses the internal connection between the sample observation time series from the perspective of observation information extraction, extracts the rate of change information and the cumulative error information reflecting the characteristics from the original observation information, and simultaneously uses KPCA to model each data set in combination with the observation information, and then constructs the BIC statistic for fault monitoring, which is more effective for monitoring small faults.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process fault monitoring, and in particular to a blast furnace ironmaking process fault monitoring method, device and computer storage medium. Background Art

[0002] The blast furnace is a vertical furnace used to produce molten iron and is an important part of the steelmaking process. In order to ensure the smooth and stable production of molten iron, it is necessary to monitor the status of each process variable in the entire blast furnace to maintain the stable operation of the blast furnace. In the traditional blast furnace ironmaking process fault monitoring, it is first necessary to obtain historical data under normal conditions during the ironmaking process, and then establish a global statistical analysis model, and then monitor the samples collected in real time online to determine whether they are in a normal state. However, on the one hand, due to the complex relationships between the variables involved in the blast furnace ironmaking process, and on the other hand, due to the variability of real-time conditions, the traditional model cannot adapt to current needs.

[0003] Since the blast furnace ironmaking production process has a large number of variables and complex variable relationships, traditional modeling strategies cannot accurately model the process. In order to better realize industrial process fault monitoring, Macgregor first proposed a multi-block fault detection method based on data, which has become a hot topic of research. The data set partitioning method is the core content of the multi-block modeling strategy. Constructing a data set can put variables with the same change characteristics together to highlight local characteristics, thereby improving the accuracy of fault monitoring. The quality of the data set partitioning method determines the quality of the final fault monitoring performance to a certain extent.

[0004] The problem in the current blast furnace ironmaking fault monitoring based on multi-block modeling is that a large number of methods only consider the linear relationship between variables when performing variable block division, and do not consider the nonlinear relationship between variables, which makes it difficult to ensure the rationality of the data set division, which will affect the accuracy of the final monitoring results. In addition, most of the current block modeling strategies only rely on the observed value information of the data for modeling, and do not explore the potential feature information hidden in the observed value information, so they are not effective when monitoring faults with slow, small changes or continuous oscillation changes. PCA or KPCA is often used in data set modeling, but PCA is constrained by the linear correlation of variables and cannot capture the nonlinear relationship in process variables. The multi-block modeling methods based on KPCA all use the observed values ​​of each process variable to construct latent variables for monitoring, without considering other effective information implied in the observed values. When the fault variable is in oscillation or the fault amplitude is small, if only the observed values ​​at each moment are used for process modeling, the effect is not significant. Therefore, how to accurately detect minor faults is a problem to be solved at present. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the problem of inaccurate detection of minor faults in the prior art due to the multi-block modeling method based on KPCA not taking into account other effective information implicit in the observation values.

[0006] In order to solve the above technical problems, the present invention provides a method for monitoring faults in a blast furnace ironmaking process, comprising:

[0007] The original data set under normal working conditions is standardized to obtain a standard data set;

[0008] Constructing a cumulative error data set and a change rate data set according to the standard data set;

[0009] Modeling the standard data set, the accumulated error data set and the change rate data set respectively, and calculating respective control limits;

[0010] Construct a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested;

[0011] Modeling the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively, and calculating the respective statistics;

[0012] A fusion statistic is calculated based on the control limit and the statistic using a Bayesian fusion strategy;

[0013] It is determined whether the fusion statistic is less than the verification control limit. If so, the sample set to be tested is normal and the next sample set to be tested is monitored.

[0014] Preferably, constructing a cumulative error data set according to the standard data set comprises:

[0015] Accumulating the differences between a preset number of sample values ​​in the standard data set and the set standard value to obtain the accumulated error data set;

[0016] The set standard value is the mean value of the standard data set.

[0017] Preferably, constructing a change rate data set according to the standard data set comprises:

[0018] The change rate of the sample variables in the standard data set at each moment is calculated to obtain a change rate data set.

[0019] Preferably, respectively modeling the standard data set, the accumulated error data set and the change rate data set, and calculating respective control limits comprises:

[0020] Performing KPCA modeling on the standard data set, the accumulated error data set and the change rate data set respectively;

[0021] Calculate the control limits of the standard data set and get the first data set T 2 Statistical Control Limits and SPE 2 Statistical control limits;

[0022] Calculate the control limits for the cumulative error data set to obtain the second data set T 2 Statistical Control Limits and SPE 2 Statistical control limits;

[0023] Calculate the control limits for the rate of change data set to obtain the third data set T 2 Statistical Control Limits and SPE 2 Statistical control limits.

[0024] Preferably, respectively modeling the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested, and calculating their respective statistics includes:

[0025] Performing KPCA modeling on the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively;

[0026] Calculate the statistics of the standardized sample set to be tested to obtain the first subset T 2 Statistics and SPE 2 Statistics;

[0027] Calculate the statistics of the cumulative error subset to be tested to obtain the second subset T 2 Statistics and SPE 2 Statistics;

[0028] Calculate the statistics of the change rate subset to be tested to obtain the third subset T 2 Statistics and SPE 2 Statistics.

[0029] Preferably, the method of calculating the fusion statistic according to the control limit and the statistic using the Bayesian fusion strategy includes:

[0030] Calculate the first likelihood function The second likelihood function The third likelihood function Fourth likelihood function in, represents the sample set to be tested in the i-th subset T 2Statistics, represents the original data set under normal conditions in the i-th data set T 2 Statistical control limits, represents the sample set to be tested in the i-th subset SPE 2 Statistics, represents the original data set under normal conditions in the i-th data set SPE 2 Statistical control limits;

[0031] Calculate the number of the sample set to be tested in the i-th subset T 2 Statistics and SPE 2 Conditional probability of failure of the statistic:

[0032]

[0033]

[0034]

[0035]

[0036] in, and is the prior probability of a normal sample, and its value is the confidence level β. and is the prior probability of the fault sample, and its value is (1-β);

[0037] Calculate Bayesian T 2 Fusion Statistics

[0038] Calculate Bayesian T 2 Fusion Statistics

[0039] Preferably, if the fusion statistic is not less than the validation control limit, the sample is faulty;

[0040] Calculate the contribution rate of each variable in the fault sample to the fault, and separate the variable with the largest contribution rate to the fault.

[0041] Preferably, calculating the contribution rate of each variable in the fault sample to the fault and separating the variable with the largest contribution rate to the fault comprises:

[0042] Calculate the T of the jth variable of the tth test sample 2 Contribution:

[0043]

[0044] Calculate the SPE of the jth variable of the tth sample to be tested 2 Contribution:

[0045]

[0046] where x tj is the jth variable of the tth sample to be tested, K t is the standardized kernel matrix, α is the transformation matrix, Λ is the weighting matrix, x t is the tth sample to be tested, θ is the median, k(x t ,x t ) is a nonlinear mapping;

[0047] Calculate the tth sample to be tested in the i-th subset T of the jth variable 2 Contribution rate:

[0048]

[0049] Calculate the tth sample to be tested in the i-th subset SPE of the jth variable 2 Contribution rate:

[0050]

[0051] Give each subset The corresponding weight values ​​are:

[0052]

[0053]

[0054] in, represents the i-th subset T 2 Statistics and SPE 2 The mean of a statistic;

[0055] And calculate the final contribution rate of the jth variable:

[0056]

[0057]

[0058] The present invention also provides a device for monitoring faults in a blast furnace ironmaking process, comprising:

[0059] A standard data set acquisition module is used to standardize the original data set under normal working conditions to obtain a standard data set;

[0060] A cumulative error data set and a change rate data set construction module, used to construct a cumulative error data set and a change rate data set according to the standard data set;

[0061] A control limit calculation module, used to respectively model the standard data set, the accumulated error data set and the change rate data set, and calculate respective control limits;

[0062] A module for constructing a cumulative error subset to be tested and a change rate subset to be tested, used for constructing a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested;

[0063] A statistics calculation module, used to respectively model the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested, and calculate the respective statistics;

[0064] A fusion statistics calculation module, used for obtaining a fusion statistics according to the control limit and the statistics by using a Bayesian fusion strategy;

[0065] The fault detection module is used to determine whether the fusion statistic is less than the verification control limit. If so, the sample set to be tested is normal and the next sample set to be tested is monitored.

[0066] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned blast furnace ironmaking process fault monitoring method are implemented.

[0067] The above technical solution of the present invention has the following advantages compared with the prior art:

[0068] The present invention extracts the cumulative error and the rate of change based on the sample observation information, divides the data set into three data sets for KPCA modeling in combination with the original observation information, and uses the Bayesian fusion strategy to construct process monitoring statistics to monitor process faults; fully considers the implicit information of the sample observations, and improves the monitoring performance of the KPCA fault monitoring algorithm for faults such as small offsets and pulse oscillations; according to the characteristics of the fault change curve, the present invention uses the internal connection between the sample observation time series from the perspective of observation information extraction, extracts the rate of change information and the cumulative error information reflecting the characteristics from the original observation information, and simultaneously uses KPCA to model each data set in combination with the observation information, and then constructs the BIC statistic for fault monitoring, which is more effective for monitoring small faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0070] Figure 1 It is a flow chart of the implementation of the fault monitoring method of the blast furnace ironmaking process of the present invention;

[0071] Figure 2 It is a flow chart of multi-core principal component fault monitoring and diagnosis algorithm based on observation value information extraction;

[0072] Figure 3 It is the multi-block KPCA modeling method adopted by the present invention;

[0073] Figure 4 This is a diagram showing the monitoring results of faults during the actual blast furnace ironmaking process;

[0074] Figure 5 This is the monitoring result diagram of fault 11 in the TE process;

[0075] Figure 6 is the T of fault 11 during TE process 2 Diagnostic results diagram;

[0076] Figure 7 This is the SPE diagnosis result diagram of fault 11 in the TE process;

[0077] Figure 8 It is the variation curve of variable 32 before and after the failure in sub-block 1 during the TE process;

[0078] Fig. 9 It is the variation curve of variable 32 before and after the failure in sub-block 2 in the TE process;

[0079] Fig.10 A structural block diagram of a device for monitoring faults in a blast furnace ironmaking process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The core of the present invention is to provide a method, device and computer storage medium for fault monitoring in a blast furnace ironmaking process, thereby improving the detection effect of minor faults.

[0081] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0082] Please refer to Figure 1 and Figure 2 , Figure 1 This is a flow chart for implementing fault monitoring in the blast furnace ironmaking process provided by the present invention. Figure 2This is a flow chart of a multi-core principal component fault monitoring and diagnosis algorithm based on observation value information extraction; the specific operation steps are as follows:

[0083] The traditional multi-block modeling method splits the original data set into several subsets for modeling by finding the differences between variables, thereby obtaining local effective information between processes. This invention takes into account the diversity of information and the implicit internal connections between samples. Instead of splitting the original data set, the original data set is extracted from the entire original data set and then monitored in blocks. The modeling method of this paper is as follows: Figure 3 As shown;

[0084] S101: The original data set X under normal working conditions train Perform standardization to obtain the standard data set X0∈R n ×m ;

[0085] S102: constructing a cumulative error data set and a change rate data set according to the standard data set;

[0086] The difference between the preset number T of sample values ​​in the standard data set and the set standard value is accumulated to obtain the accumulated error data set X W ∈R (n-T)×m , the cumulative error at time t is:

[0087] The set standard value is the mean of the standard data set. The mean of each variable in the data set after data standardization is 0, so the accumulated error over a period of time can be directly added by adding the sample values ​​in X0.

[0088] The rate of change is defined as the rate of change of the sample variable before and after the observation value, reflecting the dynamic characteristics of the variable. When the fault that causes the system acts on a certain variable in the form of pulse oscillation, this modeling method can better monitor the fault;

[0089] Calculate the change rate of the sample variables in the standard data set at each moment to obtain the change rate data set X B ∈R (n -1)×m t , the variable change rate information at the moment is: x B (t) = x(t) - x(t-1).

[0090] S103: Modeling the standard data set, the accumulated error data set, and the change rate data set respectively, and calculating respective control limits;

[0091] Performing KPCA modeling on the standard data set, the accumulated error data set and the change rate data set respectively;

[0092] Calculate the control limits of the standard data set and get the first data set T 2 Statistical Control Limits and SPE 2 Statistical control limits;

[0093] Calculate the control limits for the cumulative error data set to obtain the second data set T 2 Statistical Control Limits and SPE 2 Statistical control limits;

[0094] Calculate the control limits for the rate of change data set to obtain the third data set T 2 Statistical Control Limits and SPE 2 Statistical control limits.

[0095] S104: constructing a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested;

[0096] Accumulating the differences between a preset number of sample values ​​in the standardized sample set to be tested and a set standard value to obtain the accumulated error subset;

[0097] The set standard value is the mean value of the sample set to be tested.

[0098] The change rate of the sample variables in the standard data set at each moment is calculated to obtain a change rate data set.

[0099] S105: Modeling the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively, and calculating the respective statistics;

[0100] Performing KPCA modeling on the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively;

[0101] Calculate the statistics of the standardized sample set to be tested to obtain the first subset T 2 Statistics and SPE 2 Statistics;

[0102] Calculate the statistics of the cumulative error subset to be tested to obtain the second subset T 2 Statistics and SPE 2 Statistics;

[0103] Calculate the statistics of the change rate subset to be tested to obtain the third subset T 2 Statistics and SPE 2 Statistics.

[0104] S106: Calculate a fusion statistic according to the control limit and the statistic using a Bayesian fusion strategy;

[0105] Calculate the first likelihood function The second likelihood function The third likelihood function Fourth likelihood function in, represents the sample set to be tested in the i-th subset T 2 Statistics, represents the original data set under normal conditions in the i-th data set T 2 Statistical control limits, represents the sample set to be tested in the i-th subset SPE 2 Statistics, represents the original data set under normal conditions in the i-th data set SPE 2 Statistical control limits;

[0106] Calculate the number of the sample set to be tested in the i-th subset T 2 Statistics and SPE 2 Conditional probability of failure of the statistic:

[0107]

[0108]

[0109]

[0110]

[0111] in, and is the prior probability of a normal sample, and its value is the confidence level β. and is the prior probability of the fault sample, and its value is (1-β);

[0112] Finally, the likelihood function is used as the weight to weight the probability of each subset of the tested sample set being a fault, and the Bayesian result is obtained. 2 The statistics are as follows:

[0113] Bayesian T 2 Fusion Statistics

[0114] Bayesian T 2 Fusion Statistics

[0115] S107: Determine whether the fusion statistic is less than the verification control limit. If so, the sample set to be tested is normal, and continue to monitor the next sample set to be tested.

[0116] Under the BIC monitoring index, the control limit of the statistic is (1-β). When the Bayesian statistic index is less than the confidence level β, the sample x to be tested n It is considered normal, otherwise it is considered a fault.

[0117] Based on the above embodiments, this embodiment further describes step S107 in detail, including:

[0118] When a system fault is detected, it is necessary to separate the source variables that cause the fault. The contribution graph method is a commonly used fault diagnosis method in traditional industrial monitoring processes. This paper uses a unified contribution framework to give the contribution graph of KPCA and proposes a weighted method to apply it to multi-block KPCA fault monitoring;

[0119] Calculate the contribution rate of each variable in the fault sample to the fault, and separate the variable with the largest contribution rate to the fault;

[0120] Calculate the T2 contribution of the jth variable of the tth sample to be tested:

[0121]

[0122] Calculate the SPE of the jth variable of the tth sample to be tested 2 Contribution:

[0123]

[0124] where x tj is the jth variable of the tth sample to be tested, K t is the standardized kernel matrix, α is the transformation matrix, Λ is the weighting matrix, x t is the tth sample to be tested, θ is the median, k(x t ,x t ) is a nonlinear mapping;

[0125] Calculate the tth sample to be tested in the i-th subset T of the jth variable 2 Contribution rate:

[0126]

[0127] Calculate the tth sample to be tested in the i-th subset SPE of the jth variable 2 Contribution rate:

[0128]

[0129] Give each subset The corresponding weight values ​​are:

[0130]

[0131]

[0132] in, represents the i-th subset T 2 Statistics and SPE 2 The mean of a statistic;

[0133] And calculate the final contribution rate of the jth variable:

[0134]

[0135]

[0136] In order to solve the problem that it is difficult to effectively detect minor faults such as minor offsets and pulse oscillations in fault monitoring of blast furnace ironmaking process, the present invention proposes a multi-block kernel principal component fault monitoring and diagnosis method based on observation value information. The method extracts the cumulative error and change rate based on the sample observation value information, divides the data set into three data sets for KPCA modeling based on the original observation value information, and uses the Bayesian fusion strategy to construct process monitoring statistics to monitor process faults. At the same time, a fault diagnosis method based on a weighted unified contribution graph is designed for the proposed multi-block KPCA fault detection method. The proposed method is simulated with some similar fault detection methods for minor faults occurring in the actual blast furnace ironmaking process and the TE process, which shows that the method proposed in this article has good fault detection performance. According to the characteristics of the fault change curve, the present invention uses the internal connection between the sample observation value time series from the perspective of observation value information extraction, extracts the change rate information and cumulative error information reflecting the characteristics from the original observation value information, and simultaneously uses KPCA to model each data set in combination with the observation value information, and then constructs the BIC statistic for fault monitoring. Further, the weighted unified contribution diagram method is used to find out the source variables that cause the fault and isolate them, which is more effective in monitoring minor faults.

[0137] Based on the above embodiments, this embodiment is tested and compared in the actual process of blast furnace ironmaking and TE data set, as follows:

[0138] 1) Case study on the actual process of blast furnace ironmaking

[0139] The blast furnace is a vertical furnace used to produce molten iron and is an important part of the steelmaking production process. In order to ensure the smooth and stable production of molten iron, it is necessary to monitor and keep the gas flowing into the blast furnace stable. Suspension fault is a typical fault caused by abnormal rising gas. The continued action of the suspension fault will cause the pressure and heat on the furnace top to increase, and eventually damage the blast furnace structure. The data of this implementation case comes from a real blast furnace with a suspension fault, containing 3900 sample data, of which the first 2000 are normal samples and the last 1900 are samples under suspension fault. Similar to implementation case 1, the variable block is subjected to cumulative error and first-order difference information extraction, and finally 3 feature information sub-blocks are obtained. After completing the variable block division and sub-block expansion, the KPCA model is established to monitor each information sub-block, and then the monitoring results of all sub-blocks are fused through Bayesian inference to obtain the final monitoring index.

[0140] Table 1 shows the process variables involved in the blast furnace ironmaking suspension failure:

[0141] Table 1 Process variables involved in blast furnace ironmaking suspension failure

[0142]

[0143] Table 2 shows the monitoring results of blast furnace ironmaking suspension faults by the MBKPCA method described in the patent of this invention and three existing monitoring methods, namely kNN, PCA and MI-MBkNN.

[0144] Table 2 Comparison of monitoring method performance

[0145]

[0146] As can be seen from Table 2, the MBKPCA method described in the present invention has achieved the best monitoring results for blast furnace ironmaking suspension faults compared with the other three existing methods, with an alarm rate of 98.3% and a false alarm rate of 0%, with significant results. The following is a detailed analysis of the monitoring results of the MBKPCA method for blast furnace ironmaking suspension faults. (The red lines in each figure are calculated fault control limits).

[0147] Figure 4 The three sub-information sub-blocks and the monitoring results of the suspension fault after fusion are shown in turn. It can be seen that the patented method of the present invention has achieved good monitoring results. Sub-block 2 is the cumulative error information sub-block generated by expanding variable block 1. The blast furnace ironmaking suspension fault causes variable 8 to show a small amplitude fluctuation. If only relying on the observed value, this fault cannot be detected. By calculating the cumulative error of the observed value information, the amplitude fluctuation of the fault sample is more obvious, which is more conducive to detecting the occurrence of such faults.

[0148] 2) Case study implementation on TE process dataset

[0149] The TE process is mainly composed of five main units: reactor, condenser, compressor, separator and stripping tower. The process simulates a real chemical process operation. The data generated by it has the characteristics of time-varying, strong coupling and nonlinearity, and is widely used in testing the control and fault diagnosis models of complex industrial processes. The TE process contains 21 pre-set faults, corresponding to different fault types. 960 normal samples running smoothly in the TE process are used as training sets, and 960 samples under 21 fault conditions are used as fault test sets. Each sample contains 52 variables. 22 process measurement variables and 11 operating variables except stirring speed are selected for modeling and monitoring. In the simulated fault conditions, the faults are added at the 161st sample point, but because the cumulative error information sub-block data will lose the first T samples (T is 10 in the simulation), the actual fault occurrence points in the subsequent fault monitoring diagram are all located at the 151st sample. In the TE simulation experiment, the cumulative principal component contribution ratio method is used to determine the number of principal components, and the cumulative principal component contribution ratio is selected as 95%. The Gaussian kernel function σ is taken as 2000, T 2 The confidence level of the and SPE statistics is 95%. The alarm rates of each sub-block of TE process data and the fused statistics are shown in Table 3;

[0150] Table 3 Alarm rate of each sub-block of TE process data and fused statistics

[0151]

[0152]

[0153] By comparing the alarm rate and false alarm rate of each sub-block for 21 faults in Table 3, it can be seen that the existence of sub-block 3 (accumulated error sub-block) can greatly improve the alarm rate of faults, but at the same time it also increases the false alarm rate of faults. Although the existence of sub-block 2 (change rate sub-block) fails to improve the alarm rate of faults, it also reduces the false alarm rate of faults. At the same time, it can be seen that in the case of fault 19, sub-block 2 can also improve the alarm rate of faults. In the final BIC fusion monitoring results, T 2 and the average alarm rate of SPE, and the average false alarm rate T 2 The average false alarm rate of SPE is improved, which is almost the same as the original sub-block and within the confidence limit, but is still lower than the false alarm rate of sub-block 3. Therefore, in summary, the overall performance of the fused sub-block is still better than the three sub-blocks.

[0154] Figure 5 The monitoring results of each sub-block of fault 11 are given. It can be seen that the fault alarm rate after fusion is greater than the alarm rate of each sub-block, among which sub-block 3 contributes more to the fault alarm rate. Figure 6Analysis of the diagnosis results of fault 11 shows that variables 9 and 32 are the source variables that cause fault 11. Taking variable 32 as an example, the change curve of variable 32 before and after the fault occurs in sub-block 1 and sub-block 3 is drawn. Figure 7 It can be seen that the offset of variable 32 in sub-block 1 (i.e., the original sub-block) increases before and after the failure. Figure 8 The errors of sub-block 1 are accumulated respectively. By comparison, it can be seen that the superposition of errors can increase the variable offset before and after the fault occurs, making the fault easier to detect, which verifies that the amount of information implicit in the original observation information can be obtained through information extraction, thereby improving the fault alarm rate.

[0155] Table 4 gives the alarm rates of 21 faults in the TE process under different fault monitoring methods, mainly including KPCA, MBIPCA, DWLOFKPCA, and MBIMDKNN.

[0156] Table 4 Alarm rates of 21 faults in TE process under different fault monitoring methods

[0157]

[0158] Among them, fault numbers 1 to 21 represent 21 types of faults, including 960 sets of training samples and 960 sets of test samples. Through simulation results and analysis, the average fault alarm rate of the proposed method in 21 fault conditions is high. Fig. 9 The comparison of the monitoring results of KPCA, MBIPCA and the proposed algorithm for fault 20 in the TE process shows that the proposed algorithm has good monitoring performance. Therefore, the KPCA based on observation value information extraction in this paper can make more full use of the implicit information inside the observation value information through the mining of implicit information, which helps to improve the fault monitoring performance.

[0159] In order to solve the problem that it is difficult to effectively detect small faults such as small offsets and pulse oscillations in fault monitoring of blast furnace ironmaking process, the patent of this invention proposes a multi-block kernel principal component fault monitoring and diagnosis method based on observation value information extraction. This method extracts the cumulative error and change rate of sample observation value information, divides the data set into three sub-blocks for KPCA modeling based on the original observation value information, and uses the Bayesian fusion strategy to construct process monitoring statistics to monitor process faults. At the same time, a fault diagnosis method based on a weighted unified contribution graph is designed for the proposed multi-block KPCA fault detection method. The proposed method is compared with some similar fault detection methods in actual application of blast furnace ironmaking and simulation of TE process, which shows that the method proposed in this article has good fault detection performance.

[0160] From the results of the above two implementation cases, it can be seen that the MBKPCA method described in the present invention has a better fault monitoring effect than the existing method.

[0161] Please refer to Fig.10 , Fig.10 A structural block diagram of a device for monitoring faults in a blast furnace ironmaking process provided by an embodiment of the present invention; the specific device may include:

[0162] The standard data set acquisition module 100 is used to standardize the original data set under normal working conditions to obtain a standard data set;

[0163] The cumulative error data set and change rate data set construction module 200 is used to construct the cumulative error data set and the change rate data set according to the standard data set;

[0164] A control limit calculation module 300, used to respectively model the standard data set, the accumulated error data set and the change rate data set, and calculate respective control limits;

[0165] A module 400 for constructing a cumulative error subset to be tested and a change rate subset to be tested, for constructing a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested;

[0166] A statistics calculation module 500 is used to respectively model the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested, and calculate the respective statistics;

[0167] A fusion statistics calculation module 600 is used to calculate the fusion statistics according to the control limit and the statistics using a Bayesian fusion strategy;

[0168] The fault detection module 700 is used to determine whether the fusion statistic is less than the verification control limit. If so, the sample set to be tested is normal and the next sample set to be tested is monitored.

[0169] The blast furnace ironmaking process fault monitoring device of the present embodiment is used to implement the aforementioned blast furnace ironmaking process fault monitoring method. Therefore, the specific implementation method of the blast furnace ironmaking process fault monitoring device can be seen in the implementation method part of the blast furnace ironmaking process fault monitoring method mentioned above, for example, the standard data set acquisition module 100, the cumulative error data set and the change rate data set construction module 200, the control limit calculation module 300, the cumulative error subset to be measured and the change rate subset to be measured construction module 400, the statistic calculation module 500, the fusion statistic calculation module 600, the fault detection module 700, are respectively used to implement steps S101, S102, S103, S104, S105, S106, and S107 in the above-mentioned blast furnace ironmaking process fault monitoring method. Therefore, its specific implementation method can refer to the description of the corresponding each part of the implementation method, which will not be repeated here.

[0170] A specific embodiment of the present invention also provides a device for monitoring faults in a blast furnace ironmaking process, comprising: a memory for storing a computer program; and a processor for implementing the steps of the above-mentioned method for monitoring faults in a blast furnace ironmaking process when executing the computer program.

[0171] A specific embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for monitoring faults in a blast furnace ironmaking process are implemented.

[0172] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0173] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0176] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for monitoring faults in a blast furnace ironmaking process, characterized in that: include: The original data set under normal working conditions is standardized to obtain a standard data set; Constructing a cumulative error data set and a change rate data set according to the standard data set; Modeling the standard data set, the accumulated error data set and the change rate data set respectively, and calculating respective control limits; Construct a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested; Modeling the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively, and calculating the respective statistics; A fusion statistic is calculated based on the control limit and the statistic using a Bayesian fusion strategy; Determine whether the fused statistic is less than the verification control limit. If so, the sample set to be tested is normal, and continue to monitor the next sample set to be tested. If the fused statistic is not less than the verification control limit, the sample is faulty. Calculate the contribution rate of each variable in the fault sample to the fault, and separate the variable with the largest contribution rate to the fault, which specifically includes: Calculate the T of the jth variable of the tth sample to be tested 2 Contribution: Calculate the SPE of the jth variable of the tth sample to be tested 2 Contribution: where x tj is the jth variable of the tth sample to be tested, K t is the standardized kernel matrix, α is the transformation matrix, Λ is the weighting matrix, x t is the tth sample to be tested, θ is the median, k(x t ,x t ) is a nonlinear mapping; Calculate the tth sample to be tested in the i-th subset T of the jth variable 2 Contribution rate: Calculate the tth sample to be tested in the i-th subset SPE of the jth variable 2 Contribution rate: Give each subset The corresponding weight values ​​are: in, represents the i-th subset T 2 Statistics and SPE 2 The mean of a statistic; And calculate the final contribution rate of the jth variable:

2. The method for monitoring faults in a blast furnace ironmaking process according to claim 1, characterized in that: The constructing a cumulative error data set according to the standard data set comprises: Accumulating the differences between a preset number of sample values ​​in the standard data set and the set standard value to obtain the accumulated error data set; The set standard value is the mean value of the standard data set.

3. The method for monitoring faults in a blast furnace ironmaking process according to claim 1, characterized in that: The constructing a rate of change data set according to the standard data set comprises: The change rate of the sample variables in the standard data set at each moment is calculated to obtain a change rate data set.

4. The method for monitoring faults in a blast furnace ironmaking process according to claim 1, characterized in that: The respectively modeling the standard data set, the accumulated error data set and the change rate data set, and calculating the respective control limits comprises: Performing KPCA modeling on the standard data set, the accumulated error data set and the change rate data set respectively; Calculate the control limits of the standard data set and get the first data set T 2 Statistical Control Limits and SPE 2 Statistical control limits; Calculate the control limits for the cumulative error data set to obtain the second data set T 2 Statistical Control Limits and SPE 2 Statistical control limits; Calculate the control limits for the rate of change data set to obtain the third data set T 2 Statistical Control Limits and SPE 2 Statistical control limits.

5. The method for monitoring faults in a blast furnace ironmaking process according to claim 4, characterized in that: The respectively modeling the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested, and calculating the respective statistics includes: Performing KPCA modeling on the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested respectively; Calculate the statistics of the standardized sample set to be tested to obtain the first subset T 2 Statistics and SPE 2 Statistics; Calculate the statistics of the cumulative error subset to be tested to obtain the second subset T 2 Statistics and SPE 2 Statistics; Calculate the statistics of the change rate subset to be tested to obtain the third subset T 2 Statistics and SPE 2 Statistics.

6. The method for monitoring faults in a blast furnace ironmaking process according to claim 5, characterized in that: The method of using the Bayesian fusion strategy to calculate the fusion statistic according to the control limit and the statistic includes: Calculate the first likelihood function The second likelihood function The third likelihood function Fourth likelihood function Among them, T i 2 represents the sample set to be tested in the i-th subset T 2 Statistics, represents the original data set under normal conditions in the i-th data set T 2 Statistical control limits, represents the sample set to be tested in the i-th subset SPE 2 Statistics, represents the original data set under normal conditions in the i-th data set SPE 2 Statistical control limits; Calculate the number of the sample set to be tested in the i-th subset T 2 Statistics and SPE 2 Conditional probability of failure of the statistic: in, and is the prior probability of a normal sample, and its value is the confidence level β. and is the prior probability of the fault sample, and its value is (1-β); Calculate Bayesian T 2 Fusion Statistics Calculate Bayesian T 2 Fusion Statistics 7. A device for monitoring faults in a blast furnace ironmaking process, characterized in that: include: A standard data set acquisition module is used to standardize the original data set under normal working conditions to obtain a standard data set; A cumulative error data set and a change rate data set construction module, used to construct a cumulative error data set and a change rate data set according to the standard data set; A control limit calculation module, used to respectively model the standard data set, the accumulated error data set and the change rate data set, and calculate respective control limits; A module for constructing a cumulative error subset to be tested and a change rate subset to be tested, used for constructing a cumulative error subset to be tested and a change rate subset to be tested according to the standardized sample set to be tested; A statistics calculation module, used to respectively model the standardized sample set to be tested, the cumulative error subset to be tested, and the change rate subset to be tested, and calculate the respective statistics; A fusion statistics calculation module, used for obtaining a fusion statistics according to the control limit and the statistics by using a Bayesian fusion strategy; A fault detection module is used to determine whether the fused statistic is less than the verification control limit. If so, the sample set to be tested is normal and the next sample set to be tested is monitored. If the fused statistic is not less than the verification control limit, the sample is faulty. Calculate the contribution rate of each variable in the fault sample to the fault, and separate the variable with the largest contribution rate to the fault, which specifically includes: Calculate the T of the jth variable of the tth sample to be tested 2 Contribution: Calculate the SPE of the jth variable of the tth sample to be tested 2 Contribution: where x tj is the jth variable of the tth sample to be tested, K t is the standardized kernel matrix, α is the transformation matrix, Λ is the weighting matrix, x t is the tth sample to be tested, θ is the median, k(x t ,x t ) is a nonlinear mapping; Calculate the tth sample to be tested in the i-th subset T of the jth variable 2 Contribution rate: Calculate the tth sample to be tested in the i-th subset SPE of the jth variable 2 Contribution rate: Give each subset The corresponding weight values ​​are: in, represents the i-th subset T 2 Statistics and SPE 2 The mean of a statistic; And calculate the final contribution rate of the jth variable:

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for monitoring faults in a blast furnace ironmaking process as claimed in any one of claims 1 to 6 are implemented.

Images