Improved Adaptive DBSCAN for Data Cleaning in Molten Iron Weighing Process

In the intricate process of steel manufacturing, the precise measurement of molten iron is a pivotal procedure, which directly impacts the quality of steel production. Rail weighbridges are commonly deployed in the steel industry for molten iron measurement. Therefore, the accuracy of rail weighbridges weighing data is extremely important. However, due to process intricacies and equipment nuances, weighbridge weight data often exhibits outliers. This problem can potentially hinder data-centric modeling and predictive tasks. To solve this problem, this paper introduces a data cleaning methodology based on an Improved Adaptive Density-Based Spatial Clustering of Applications with Noise (IA-DBSCAN). The proposed data cleaning method is rooted in an improved adaptive DBSCAN approach, which is particularly effective in identifying clusters of arbitrary shapes and can differentiate noise from actual clusters. The application of the IA-DBSCAN algorithm facilitates the identification and cleaning of concentrated outliers. The algorithm dynamically seeks eps and MinPts parameters to achieve optimal clustering outcomes. Constraints have been incorporated into the improved algorithm, enabling it to achieve data cleaning more rapidly and accurately. After data cleaning, based on the characteristics of the selected dataset in this paper, data imputation using the linear regression method can align with the original data characteristics. Experimental results indicate that the proposed method can effectively identify and clean outliers data in the dataset, supplement missing values based on the original data characteristics, and construct a complete and anomaly-free dataset, establish a robust foundation for modeling and prediction.

1 Introduction

Amid the rapid and steadfast evolution of the economy, the increasing integration of industrialization and informatization has spurred a heightened demand for integrating foundational data across diverse application systems[1]. This positions the establishment of the digital economy as a pivotal catalyst for fostering effective internal development within enterprises. Within the realm of steel enterprises, the process of molten iron measurement emerges as an indispensable and pivotal procedure[2]. When molten iron enters the steelmaking plant, measurement becomes an essential step. Currently, molten iron measurement in the steel industry primarily relies on rail weighbridges or crane scales[3], [4]. However, owing to procedural and equipment-related factors, instances of inaccurate measurement data frequently manifest[5]. Such discrepancies inevitably give rise to variations in measurement data, exerting an impact on the quality of the resulting steel products. Consequently, the purification of outliers within the weighing data becomes imperative to avert more severe disastrous. Subsequent to outlier cleaning, corresponding imputation is imperative for the refined data, laying the groundwork for its subsequent utilization. This multifaceted process is integral for ensuring the accuracy and reliability of molten iron measurement data within the steel industry.