An Online Integrated Learning Prediction Method for Hardness of Continuous Annealed Products

Abstract

The invention discloses an online integrated learning prediction method for the hardness of continuous annealed products, which belongs to the technical field of automatic control of the continuous annealing production process of iron and steel enterprises. Using the enterprise's historical continuous annealing production data samples, using the integrated learning modeling method using LSSVM as a sub-learning machine, the offline product hardness prediction models were established for strip steel with different quenching and tempering degrees; in actual production, real-time reading of continuous annealing production process data, and use the off-line product hardness prediction model established through integrated learning to carry out real-time prediction of the current strip steel product hardness; through the inspection of actual production data, the method of the present invention can significantly improve the accuracy and robustness of the continuous annealing product hardness prediction result Rod, so that the on-site operators can grasp the quality of the current strip products in real time, and make timely adjustments according to the situation, making up for the lack of large lag in offline detection, thereby helping the continuous annealing production line to improve product quality, improve production operation level, and increase economic benefits .

Description

technical field [0001] The invention belongs to the technical field of automatic control of the continuous annealing production process of iron and steel enterprises, and in particular relates to an online integrated learning and prediction method for the hardness of continuous annealing products. Background technique [0002] In the actual production process of the continuous annealing unit of the cold rolling plant of the iron and steel enterprise, the hardness of the strip steel is the core index to measure the product quality and guide the production. In the actual production process, the hardness of the strip steel cannot be detected online. The on-site method is to measure the hardness of the strip steel by intercepting the head and tail parts of the annealed steel strip, and then conduct offline experimental analysis, so as to judge the product quality. However, obtaining product quality information through off-line experimental analysis generally has a certain time l...

AI Technical Summary

Problems solved by technology

However, obtaining product quality information through off-line experimental analysis generally has a certain time lag, that is to say, only when the strip steel is produced for a period of time, can its specific quality information be obtained. The speed is very fast, and the annealing treatment can be completed within a few minutes, so that the hardness of the strip often fluctuates greatly, resulting in quality problems such as substandard hardness or even scrap, which seriously affects the economic benefits of the cold rolling plant

[0003] The paper "Design and Implementation of Strip Quality Prediction and Process Monitoring System of Continuous Annealing Unit Based on PLS [D]" (Wang Yuan, Northeastern University, 2009) although proposed a method based on Partial Least Squares (Partial Least Squares) for the hardness of strip products ,PLS) data-driven modeling method, but the method proposed in this paper cannot meet the needs of the actual production process. The main reasons are: (1) The information related to the strip hardness considered in this paper is less , there are only about 20 pieces, but there are as many as 51 pieces of process information related to strip hardness in the actual production process; (2) The PLS method proposed in this document is mainly aimed at the monitoring and fault diagnosis of the continuous annealing production process, And the PLS method is a kind of linear regression method, but the actual production process is nonlinear, which leads to the low accuracy of the PLS method; (3) the sample data has many input items and mutual coupling, but the output range is narrow The problem is that there are large differences in input items between samples, but the output results are the same or similar, making it difficult for traditional data modeling methods to obtain high prediction accuracy and robustness.

Images