A blast furnace control parameter anomaly detection method, a terminal device, and a storage medium
By using normal distribution judgment, modified Laida's rule, DBSCAN clustering, and time-delay-based RRCF detection, the problem of detecting local outliers in blast furnace control parameters was solved, achieving accurate anomaly detection of blast furnace control parameters and improving the accuracy and stability of detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WISDRI ENG & RES INC LTD
- Filing Date
- 2022-09-21
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies cannot effectively detect local anomalies in blast furnace control parameters and easily overlook abnormal fluctuations caused by changes in furnace conditions, shutdown events, and measurement instruments or network communication, leading to inaccurate anomaly judgments.
A method for detecting anomalies in blast furnace control parameters is adopted, which includes normal distribution judgment, modified Laida's rule, DBSCAN clustering, and time-delay-based RRCF detection. Combined with hyperparameter adjustment, it automatically filters global and local outliers.
It enables precise anomaly detection of blast furnace control parameters, avoids misjudgments caused by experience-based threshold settings, and improves the accuracy and stability of anomaly detection.
Smart Images
Figure CN115496145B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blast furnace smelting, and in particular to a method, terminal equipment and storage medium for detecting abnormal blast furnace control parameters. Background Technology
[0002] Adjusting blast furnace conditions by modifying blast parameters such as blast volume, blast temperature, blast pressure, and injection rate is known as "bottom-level regulation." Traditionally, anomalies in control parameters are typically identified through experience. However, experience-based judgment usually involves setting thresholds to identify global anomalies, but it cannot automatically detect local anomalies. It usually requires observing trend curves and combining them with experience. Therefore, it is essential to establish an effective anomaly detection method to automatically detect both global and local anomalies to assist professionals in blast furnace control.
[0003] Current anomaly screening methods typically involve setting thresholds, which can only detect global anomalies and cannot identify localized anomalies. Alternatively, due to changes in furnace conditions, control parameters may exhibit significantly different values at different times, resulting in discontinuous anomalies. In such cases, it is usually necessary to adjust the thresholds based on the current furnace conditions. Furthermore, events such as furnace shutdowns can cause fluctuations in control parameter data. Additionally, due to limitations in measuring instruments or network communication, even during periods of stable furnace conditions, occasional anomalous values may appear and be easily overlooked. Summary of the Invention
[0004] To address the aforementioned problems, this invention proposes a method for detecting abnormal blast furnace control parameters, a terminal device, and a storage medium.
[0005] The specific plan is as follows:
[0006] A method for detecting abnormal blast furnace control parameters includes the following steps:
[0007] S1: Collect the values of various control parameters of the blast furnace at fixed sampling time intervals over a period of time, and determine whether the values of various control parameters within this period of time conform to a normal distribution;
[0008] S2: Extract control parameters that conform to a normal distribution and remove the extreme values. Based on the ratio of the standard deviation to the mean of the control parameters after removing the extreme values, correct the coefficient 3 of σ in the classical Laida's rule.
[0009] S3: Global outliers in the control parameters after removing extrema are eliminated using the modified Laida rule;
[0010] S4: Perform DBSCAN clustering on the control parameters processed in step S3;
[0011] S5: Perform RRCF detection based on time delay for each category of control parameters after clustering, and eliminate the local outliers therein;
[0012] S6: Evaluate the screening effect of the control parameters processed in step S5. When the evaluation result does not reach the expected goal, adjust the hyperparameters in steps S2, S4, and S5 according to the evaluation result and re-screen until the evaluation result reaches the expected goal.
[0013] Further, the method for judging whether it conforms to the normal distribution can be: calculate the standard deviation σ and the mean μ of the values of the control parameters during this period, draw a normal distribution graph based on the calculated standard deviation σ and the mean μ, and judge whether it conforms to the normal distribution based on the normal distribution graph.
[0014] Further, the correction method in step S2 is: set the ratio of the standard deviation to the mean of the control parameters collected in step S1 as r1, and the ratio of the standard deviation to the mean of the control parameters after removing extreme values in step S2 as r2, then:
[0015] If r2 ≤ A and r2 ≥ B, then modify the coefficient 3 of σ to 3 - r2; A and B respectively represent the upper and lower limits of the ratio;
[0016] If r2 < B, then do not modify the coefficient 3 of σ;
[0017] If r2 > A, then modify the coefficient 3 of σ to min(1 / r1, 1 / r2), where min represents taking the minimum value.
[0018] Further, step S5 specifically includes:
[0019] Set the set of control parameter sequences within a period of time as {x1, …, x n}, n represents the nth sampling moment, and x n represents the control parameter sequence collected at the nth sampling moment;
[0020] Divide the set of control parameter sequences into m windows, and set the size of each window w = n / m and round down;
[0021] For each window i, establish a standard normal distribution function f i , i ∈ [1, m], and set the standard deviation of the normal distribution function f i as the product of the window number and the window size, that is, μ i = i * w;
[0022] When constructing the RRCF, for each normal distribution function, construct the corresponding tree, and form a forest by constructing trees for m normal distribution functions. Based on the formed forest, construct an RRCF model to detect the data.
[0023] Furthermore, the method for constructing the corresponding tree for each normal distribution function is as follows: Set the tree size to t, and based on the normal distribution function f... i In the set of control parameter sequences {x1,…,x n According to the normal distribution function f i A tree is constructed by randomly selecting k subsets of size t without replacement, where k = n / t and rounded down.
[0024] Furthermore, the method for evaluating the screening effect in step S6 is as follows: scatter plots are drawn for the control parameters collected in step S1 and the control parameters processed in step S5, and the two scatter plots are compared to evaluate whether global outliers and spikes have been removed and whether step signals have been retained.
[0025] A blast furnace control parameter anomaly detection terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method described above in the embodiments of the present invention.
[0026] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above in the embodiments of the present invention.
[0027] The present invention adopts the above technical solution to realize the anomaly detection of blast furnace control parameters, and solves the problem that the existing system ignores fault changes, fluctuation changes or abnormal punctures when setting thresholds based on expert experience for anomaly screening. Attached Figure Description
[0028] Figure 1 The diagram shown is a flowchart of Embodiment 1 of the present invention.
[0029] Figure 2 The diagram shown illustrates the normal distribution of the control parameters in this embodiment.
[0030] Figure 3 The diagram shown illustrates the result of removing global outliers using the modified Laida rule in this embodiment.
[0031] Figure 4 The diagram shown is a schematic representation of the DBSCAN clustering results in this embodiment.
[0032] Figure 5 The figure shown is a schematic diagram of the anomaly scoring results based on time delay RRCF in this embodiment. Detailed Implementation
[0033] To further illustrate the embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, mainly used to illustrate the embodiments, and can be used in conjunction with the relevant descriptions in the specification to explain the operating principle of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention.
[0034] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation manners.
[0035] Embodiment 1:
[0036] The embodiment of the present invention provides a method for detecting abnormal blast furnace control parameters, as Figure 1 shown, the method includes the following steps:
[0037] S1: Collect the values of various blast furnace control parameters within a period of time at a fixed sampling time interval, and judge whether the values of each control parameter within this period conform to the normal distribution.
[0038] Those skilled in the art can set the size of the period of time and the size of the sampling time interval by themselves, and no limitation is made here.
[0039] The control parameters in this embodiment include blast furnace hot air temperature, oxygen enrichment rate, etc.
[0040] The method for judging whether it conforms to the normal distribution can be: calculate the standard deviation σ and the mean μ of the values of the control parameter within this period, and draw a normal distribution graph based on the calculated standard deviation σ and mean μ, as Figure 2 shown, judge whether it conforms to the normal distribution based on the normal distribution graph.
[0041] S2: Extract the control parameters that conform to the normal distribution and remove the extreme values therein, and modify the coefficient 3 of σ in the 3σ of the classical Ralida rule according to the ratio of the standard deviation and the mean of the control parameters after removing the extreme values.
[0042] The classical Ralida rule believes that |x j -μ|≥3σ are outliers, and the coefficient 3 before σ needs to be changed in the modified Ralida method.
[0043] The modification method adopted in this embodiment is: set the ratio of the standard deviation to the mean of the control parameters collected in step S1 as r1, and the ratio of the standard deviation to the mean of the control parameters after removing the extreme values in step S2 as r2, then:
[0044] If r2≤A and r2≥B, then modify the coefficient 3 of σ to 3 - r2;
[0045] If r2<B, then do not modify the coefficient 3 of σ;
[0046] If r2>A, then the coefficient 3 of σ is modified to min(1 / r1,1 / r2), where min means taking the minimum value.
[0047] A and B are both hyperparameters, representing the upper and lower limits of the ratio, respectively. A > B. In this embodiment, A and B are initially set to 1 and 0.5, respectively.
[0048] S3: Global outliers in the control parameters after removing extrema are eliminated using the modified Laida's rule. The result after elimination is as follows: Figure 3 As shown.
[0049] S4: Perform DBSCAN clustering on the control parameters processed in step S3.
[0050] Clustering can preserve step signals generated by control adjustments, preventing them from being rejected as anomalous signals. The clustering results are as follows: Figure 4 As shown.
[0051] The specific method of DBSCAN clustering is as follows:
[0052] Given a dataset consisting of the values of control parameters over a given period of time, a neighborhood radius Eps and the minimum number of points within the neighborhood radius that become the core object: MinPts.
[0053] Starting from any point p, mark it as "visited" and check if it is a core point (i.e., p's Eps neighborhood has at least MinPts objects). If it is not a core point, mark it as a noise point. Otherwise, create a new cluster C for p and add all objects in p's Eps neighborhood to the candidate set N.
[0054] Iteratively add objects from N that do not belong to other clusters to C. During this process, for an object p' in N marked as "unvisited", mark it as "visited" and check its Eps neighborhood. If p' is also a core object, then all objects in p''s Eps neighborhood are added to N. Continue adding objects to C until C cannot expand further, i.e., until N is empty. At this point, cluster C is completely generated.
[0055] Randomly select the next unvisited object from the remaining objects, and repeat the iteration process until all objects have been visited.
[0056] S5: Perform time-delay-based RRCF detection on each clustered control parameter to remove local outliers.
[0057] The Robust Random Cut Forest (RRCF) anomaly detection algorithm for data streams, based on random forests, primarily optimizes and improves the iForest model, enabling more efficient representation and anomaly detection of input data streams. Experiments demonstrate that RRCF is more efficient and accurate than iForest. RRCF's complexity is defined by the depth of the tree; it first calculates the depth of each node and then sums the results to obtain the tree's complexity. For a forest, the complexity is averaged based on the number of trees. Therefore, calculating the complexity change caused by a particular anomaly node represents the complexity change for all other nodes.
[0058] This embodiment improves the traditional RRCF method by proposing a time-delay RRCF method, specifically as follows:
[0059] Let the set of control parameter sequences over a period of time be {x1,…,x}. n}, where n represents the nth sampling time, x n This represents the sequence of control parameters acquired at the nth sampling time. The set of control parameter sequences is divided into m windows, and the size of each window is set to w = n / m and rounded down.
[0060] Establish a standard normal distribution function f for each window i. i Let i ∈ [1, m], and set the normal distribution function f. i The standard deviation is the product of the window number and the window size, i.e., μ. i = i*w, where the variance of all windows is the same and is a fixed value.
[0061] During the construction of RRCF, a corresponding tree is built for each normal distribution function. The trees of m normal distribution functions are combined into a forest, and the RRCF model is built based on the forest to detect the data.
[0062] The method for constructing a corresponding tree for each normal distribution function is as follows: Set the tree size to t, and based on the normal distribution function f... i In the set of control parameter sequences {x1,…,x n According to the normal distribution function f i A tree is constructed by randomly selecting k subsets of size t without replacement, where k = n / t and rounded down.
[0063] In the set of control parameter sequences {x1,…,x} n According to the normal distribution function f i When randomly selecting k subsets of size t without replacement, samples are preferentially selected from the peaks of the normal distribution function. The following example illustrates the specific selection process. Assume n = 100, t = 5, and k = 20:
[0064] For a normal distribution function f1, since f1 has the smallest standard deviation, the peak is concentrated in the initial region of the set. Assuming it is concentrated at x1, the process of randomly drawing a subset of samples is: K1={x1,x2,x3,x7,x...} 31 K2 = {x4, x5, x6, x 11 ,x 52}……
[0065] ...
[0066] For the normal distribution function f5, assume the peaks are concentrated at x. 10 At this point, the process of randomly selecting a subset of samples is as follows: K1={x2,x9,x... 10 ,x 11 ,x 81 K2 = {x8, x 12 ,x 13 ,x 34 ,x 86}……
[0067] S6: Evaluate the screening effect of the control parameters after step S5. If the evaluation result does not meet the expected goal, adjust the hyperparameters in steps S2, S4 and S5 according to the evaluation result and re-screen them until the evaluation result meets the expected goal.
[0068] In this embodiment, the method for evaluating the screening effect is as follows: Scatter plots are drawn for the control parameters collected in step S1 and the control parameters processed in step S5, and the two scatter plots are compared to evaluate whether global outliers and glitches have been eliminated, and whether the step signal generated by control adjustment has been retained. The scoring results are as follows: Figure 5 As shown.
[0069] The hyperparameters in step S4 include neighborhood radius and minimum number of samples in the DBSCAN clustering algorithm, while the hyperparameters in step S5 include tree size, forest size, and window size in the RRCF algorithm.
[0070] Hyperparameters can be tuned based on experience using evaluation results, or the optimal solution in the parameter space can be selected by setting a grid search or random search. There are no restrictions on this.
[0071] This invention determines the σ coefficient in the Laida rule by removing the standard deviation and mean deviation after removing extreme values, thus avoiding the problem of sub-outliers being undetectable due to global extreme outliers; at the same time, it uses RRCF with time delay to screen for local anomalies in the control parameters, so as to consider the problem of time lag correlation when processing time series data.
[0072] Example 2:
[0073] The present invention also provides a blast furnace control parameter anomaly detection terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the method embodiment described above in Embodiment 1 of the present invention.
[0074] Furthermore, as an executable solution, the blast furnace control parameter anomaly detection terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The blast furnace control parameter anomaly detection terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described composition of the blast furnace control parameter anomaly detection terminal device is merely an example and does not constitute a limitation on the blast furnace control parameter anomaly detection terminal device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the blast furnace control parameter anomaly detection terminal device may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.
[0075] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the blast furnace control parameter anomaly detection terminal equipment, connecting all parts of the equipment via various interfaces and lines.
[0076] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the blast furnace control parameter anomaly detection terminal device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0077] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the embodiments of the present invention.
[0078] If the modules / units integrated in the blast furnace control parameter anomaly detection terminal equipment are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc.
[0079] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.
Claims
1. A method for detecting abnormal blast furnace control parameters, characterized in that, It includes the following steps: S1: Collect the values of various control parameters of the blast furnace within a period of time at a fixed sampling time interval, and determine whether the values of the control parameters within this period conform to the normal distribution; the control parameters include: blast furnace hot air temperature and oxygen enrichment rate; S2: Extract the control parameters that conform to the normal distribution and remove the extreme values therein. According to the ratio of the standard deviation to the mean of the control parameters after removing the extreme values, correct the coefficient 3 of σ in the 3σ of the classical Leida method; S3: Remove the global outliers in the control parameters after removing the extreme values through the corrected Leida method; S4: Perform DBSCAN clustering on the control parameters processed in step S3; S5: Perform RRCF detection based on time delay on each category of control parameters after clustering, and remove the local outliers therein; S6: Evaluate the screening effect of the control parameters processed in step S5. When the evaluation result does not reach the expected goal, adjust the hyperparameters in steps S4 and S5 according to the evaluation result and re-screen until the evaluation result reaches the expected goal; The correction method in step S2 is: Set the ratio of the standard deviation to the mean of the control parameters collected in step S1 as r1, and the ratio of the standard deviation to the mean of the control parameters after removing the extreme values in step S2 as r2, then: If r2 ≤ A and r2 ≥ B, then modify the coefficient 3 of σ to 3 - r2; A and B respectively represent the upper and lower limits of the ratio; If r2 < B, then do not modify the coefficient 3 of σ; If r2 > A, then modify the coefficient 3 of σ to min(1 / r1, 1 / r2), where min represents taking the minimum value.
2. The method for detecting abnormal blast furnace control parameters according to claim 1, characterized in that: The method for judging whether it conforms to the normal distribution is: Calculate the standard deviation σ and the mean μ of the values of the control parameters within this period, draw a normal distribution graph based on the calculated standard deviation σ and the mean μ, and judge whether it conforms to the normal distribution based on the normal distribution graph.
3. The method for detecting abnormal blast furnace control parameters according to claim 1, characterized in that: Step S5 specifically includes: Let the set of control parameter sequences over a period of time be {x1,…,x}. n }, where n represents the nth sampling time, x n This represents the sequence of control parameters collected at the nth sampling time. Divide the control parameter sequence set into m windows, and set the size of each window w = n / m and round down; Establish a normal distribution function f for each window i i Let i ∈ [1, m], and set the normal distribution function f. i The standard deviation is the product of the window number and the window size, i.e., μ. i = i * w; When constructing the RRCF, construct a corresponding tree for each normal distribution function, form a forest by constructing trees for m normal distribution functions, and construct an RRCF model based on the formed forest to detect the data.
4. The method for detecting abnormal blast furnace control parameters according to claim 3, characterized in that: The method for constructing a corresponding tree for each normal distribution function is as follows: Set the tree size to t, and based on the normal distribution function f... i In the set of control parameter sequences {x1,…,x n According to the normal distribution function f i A tree is constructed by randomly selecting k subsets of size t without replacement, where k = n / t and is rounded down.
5. The method for detecting abnormal blast furnace control parameters according to claim 1, characterized in that: The method for evaluating the screening effect in step S6 is: Draw scatter plots for the control parameters collected in step S1 and the control parameters processed in step S5 respectively, and compare the two scatter plots to evaluate whether the global outliers and spikes are removed and the step signals are retained.
6. A terminal device for detecting abnormal blast furnace control parameters, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 5.