Abstract: Keyhole tungsten inert gas welding (K-TIG) offers the advantage of single-sided welding with double-sided formation without the need for groove preparation, making it an ideal choice for automated welding processes. However, due to the influence of gravity, the dynamic balance of forces required to sustain the keyhole is disrupted during the horizontal position welding of circumferential welds in large vertical stainless steel storage tanks using K-TIG, which adversely affects weld quality. In order to expand its application scenarios and achieve real-time adjustment of penetration state during the horizontal position welding of circumferential welds with K-TIG, the impact of varying welding current and welding speed with different amplitudes on both the penetration state and its recognition in K‑TIG horizontal position welding was investigated based on an optical character recognition-support vector machine (OCR-SVM) penetration recognition model. Key geometric features influencing penetration recognition were identified by the recognition capability of the OCR‑SVM model under variable parameter conditions, thereby providing a foundation for the automated control of K‑TIG horizontal position welding. Experimental results indicate that variations in welding current and welding speed directly affect the penetration state in K-TIG horizontal position welding, with changes in welding current exerting a more pronounced influence. The penetration recognition model based on OCR‑SVM achieves recognition accuracies of 92.29% and 91.50% under conditions of varying current and varying speed, respectively. Key geometric features for recognition are identified as the width and area of the weld pool and the keyhole.