CN-122021993-A - Static and dynamic collaborative optimization method and system for KR desulfurization efficiency

Abstract

The invention discloses a static and dynamic collaborative optimization method and a system for KR desulfurization efficiency, which comprise the steps of collecting KR desulfurization process parameters in real time, constructing a high-quality data set, adopting Gaussian disturbance sensitivity evaluation to screen data set characteristics which are strongly related to a desulfurization target, clustering by K-means according to molten iron working conditions, establishing a deep neural network for improving the parameter adjustment of a gray wolf optimization algorithm for each cluster in parallel, outputting initial set values of the addition amount of a desulfurizing agent, the rotation speed of a stirring paddle and the insertion depth, collecting ladle liquid level images in real time in the stirring process, extracting vortex centers and vortex diameters, mapping the real-time vortex diameters to a 'desulfurization efficiency-vortex diameter' model to evaluate the current state, and carrying out closed-loop compensation adjustment on the initial set values according to the 'vortex diameter-stirring parameter' model. The invention can integrate multisource information, has static prediction capability and dynamic compensation mechanism, breaks through the bottleneck of traditional KR desulfurization control in the aspects of accuracy and robustness, and realizes the collaborative optimization of the whole process.

Inventors

• DAN BINBIN
• TANG BOWEN
• Chai Zhenjie
• Chen Yangtaidong
• Zhu Panlei
• DU LIPING
• ZHAO HAIJIE
• XIONG LING
• RONG ZHIJUN
• NIE QINWEN
• CHEN LI
• Feng Pengyun
• ZHOU CHUN

Assignees

• 武汉科技大学

Dates

Publication Date

20260512

Application Date

20251217

Claims (10)

1. 1. A static and dynamic collaborative optimization method for KR desulfurization efficiency is characterized by comprising the following steps of S1 collecting KR desulfurization process parameters in real time, cleaning, LOF outlier elimination and normalization to construct a high-quality data set, S2 adopting Gaussian disturbance sensitivity evaluation to screen data set characteristics strongly related to a desulfurization target, clustering by K-means according to molten iron working conditions, S3 establishing a deep neural network for improving gray wolf optimization algorithm parameters for each cluster in parallel, outputting initial set values of desulfurizing agent addition amount, stirring paddle rotating speed and insertion depth, S4 collecting ladle liquid level images in real time in a stirring process, extracting vortex centers and vortex paths through denoising, sharpening, geometric active contour segmentation and Canny-Radon algorithm, S5 mapping the real-time vortex paths to a 'desulfurization efficiency-vortex path' model to evaluate the current state, and carrying out closed-loop compensation adjustment on the initial set values according to the 'vortex path-stirring parameter' model to realize static prediction and dynamic feedback collaborative optimization.
2. 2. The method for the collaborative dynamic optimization of KR desulfurization efficiency according to claim 1, wherein in step S1, the KR desulfurization process parameters include at least one of a pre-desulfurization temperature, a pre-treatment sulfur content, a pre-treatment silicon content, a number of times of stirring paddle use, a stirring time, a stirring paddle insertion depth, a stirring paddle speed, a liquid level height, a pre-treatment molten iron mass, a target sulfur content, and a sulfur content difference before and after desulfurization.
3. 3. The method for collaborative dynamic optimization of KR desulfurization efficiency according to claim 1, wherein in step S1, the cleaned data set is processed by a local outlier factor algorithm, the local density deviation of each data point is calculated, and the data point with LOF score greater than a preset threshold is marked as an outlier.
4. 4. The method for collaborative dynamic optimization of KR desulfurization efficiency according to claim 1, wherein in step S2, the subset above the threshold is the final input feature strongly correlated to the desulfurization target in the high quality dataset of KR desulfurization process parameters.
5. 5. The method is characterized in that step S3 is used for establishing a depth neural network prediction model based on an improved gray-wolf optimization algorithm in parallel for different clusters, and the super parameters of the depth neural network are automatically optimized through the improved gray-wolf algorithm so as to accurately output initial set values of three key process parameters, namely the addition amount of a desulfurizing agent, the rotating speed of a stirring paddle and the insertion depth.
6. 6. The method for optimizing the KR desulfurization efficiency in a static and dynamic mode according to claim 1 is characterized in that in the step S4, a Canny operator is used for detecting streamline edges, a direction field is calculated, streamline intersection points are classified according to cell directions, and a center coordinate is obtained by averaging, namely a vortex center.
7. 7. The method for collaborative optimization of KR desulfurization efficiency according to claim 1, wherein in step S4, the center coordinates of the vortex are used as the origin, the pixel scale is scaled according to the physical size of the ladle opening, and the maximum radial distance of the vortex is measured to be the vortex diameter.
8. 8. The method for collaborative optimization of the static and dynamic KR desulfurization efficiency according to claim 1 is characterized in that a model of desulfurization efficiency-vortex diameter establishes a unitary regression relationship between the end point sulfur content and the real-time vortex diameter for online evaluation of whether the current desulfurization state is in an optimal interval.
9. 9. An electronic device comprising a processor, a memory and a program stored on the memory and executable on the processor, characterized in that the program when executed by the processor implements the steps of the method of any of the preceding claims 1-8.
10. 10. A computer-readable storage medium, wherein a program is stored on the storage medium, characterized in that the program, when executed by a processor, implements the steps of the method according to any of the claims 1-8.

Description

Static and dynamic collaborative optimization method and system for KR desulfurization efficiency Technical Field The invention relates to the technical field of automatic control in a ferrous metallurgy process, in particular to a static and dynamic collaborative optimization method and system for KR desulfurization efficiency. Background In the steel industry, precise control of the sulfur content of molten iron is critical to the production of high quality steels. KR (Kambara Reactor) desulfurization process is an important link for guaranteeing the cleanliness and performance index of steel, and promotes the full contact and chemical reaction of desulfurizing agent and molten iron by means of mechanical stirring, so as to reduce the sulfur content in molten iron. However, currently, operational control of the process still relies primarily on human experience to adjust the relevant process parameters. The empirically guided mode exposes the limitations of poor adaptability, adjustment lag and the like when facing complex working conditions such as fluctuation of molten iron components, dynamic change of flow fields and the like, and not only affects the desulfurization reaction uniformity and stable control of the end point sulfur content, but also is easy to cause low utilization rate of the desulfurizing agent and increase of production cost. Along with popularization of intelligent manufacturing concepts, a prediction method based on data driving is gradually applied to desulfurization process optimization. For example, by adopting BP neural network, support vector regression and other algorithms to construct a process parameter prediction model, the rationality of initial setting is improved to a certain extent. However, most of the methods perform offline training based on historical data, the output of the method is preset as a static parameter under a fixed working condition, and the method lacks the capability of real-time sensing and feedback adjustment of dynamic information such as fluid state, reaction interface update and the like in the actual stirring process, so that the method is difficult to realize online self-correction and dynamic optimization of the parameter when the method faces sudden working condition fluctuation. Disclosure of Invention The invention mainly aims to solve the problems, and provides a static and dynamic collaborative optimization method and system for KR desulfurization efficiency. Through integrating a static prediction model and a dynamic feedback mechanism based on visual images, ladle liquid level vortex images are acquired and analyzed in real time, vortex diameter characteristics reflecting stirring intensity and reaction conditions are extracted, unstructured visual information is converted into quantifiable process parameters, and real-time closed-loop optimization of desulfurizing agent addition amount and stirring operation parameters is further realized. The invention can integrate multisource information, has static prediction capability and dynamic compensation mechanism, breaks through the bottleneck of traditional KR desulfurization control in the aspects of accuracy and robustness, and realizes the collaborative optimization of the whole process. On the other hand, the invention is beneficial to reducing the consumption of desulfurizing agent and equipment loss while improving the desulfurizing efficiency and the quality of molten steel, and provides a reliable technical path for intelligent upgrading of the ferrous metallurgy process. In order to solve the technical problems, the invention adopts the following technical scheme: A static and dynamic collaborative optimization method for KR desulfurization efficiency comprises the following steps of S1 collecting KR desulfurization process parameters in real time, conducting cleaning, LOF outlier rejection and normalization to construct a high-quality data set, S2 adopting Gaussian disturbance sensitivity evaluation to screen data set characteristics strongly related to a desulfurization target, clustering by using K-means according to molten iron working conditions, S3 establishing a deep neural network for improving a gray wolf optimization algorithm to tune parameters for each cluster in parallel, outputting initial set values of desulfurizing agent addition amount, stirring paddle rotating speed and insertion depth, S4 collecting ladle liquid level images in real time in a stirring process, conducting denoising, sharpening, geometric active contour segmentation and Canny-Radon algorithm to extract vortex centers and vortex paths, S5 mapping the real-time vortex paths to a 'desulfurization efficiency-vortex path' model to evaluate the current state, and conducting closed-loop compensation adjustment on the initial set values according to the 'vortex path-stirring parameter' model to achieve static prediction and dynamic feedback collaborative optimization. In the technical scheme, in the step S