Abstract: Deep coal seams in complex geological environments are characterized by extremely high in-situ stress, severe surrounding rock deformation, and considerable variations in both strike and dip angles. These factors cause pronounced spatial heterogeneity in surrounding rock stress distribution and structural morphology, resulting in highly variable and dynamically evolving support system requirements. Existing support systems, constrained by fixed design parameters and limited functional adaptability, are incapable of meeting the requirements of intelligent mining in such coal seams. In this study, taking a kilometer-deep ultra-long working face in a demonstration mine of the Huainan mining area as the research object, a comprehensive analytical framework for the force coupling (magnitude, direction, and point of application) and spatial configuration-displacement coupling between the surrounding rock and the support system was established to reveal the spatiotemporal evolution mechanisms of overburden zonal fracturing and dynamic stress redistribution in ultra-long working faces. On this basis, a deep-learning neural network-based coupled prediction model was developed to achieve intelligent forecasting and real-time assessment of the stress states and positional behavior of both the surrounding rock and the support system. Furthermore, based on a non-parametric clustering algorithm, the working face was divided into distinct zones according to variations in support resistance and spatial position, and corresponding support and position control strategies were proposed for each zone. Subsequently, an integrated intelligent analysis and regulation system, termed "Three Measurements, Two Controls, and One Platform", was constructed for working faces in complex geological environments. This system enables comprehensive perception of surrounding rock mechanical behavior, support system stress responses, and spatial configurations, as well as integrated decision-making and control, significantly enhancing the ability of the support system to adapt to both gradual geological evolution and abrupt dynamic disturbances. Field application results verify its effectiveness in improving the adaptability and flexibility of the support system under complex conditions. In the ultra-long working face configuration, with a mining height of 5.0–6.2 m, an average dip angle of 14°, and a maximum strike angle of 17°, the cutting cycle frequency was increased to five passes per day, representing a 38.29% improvement compared with conventional mining methods. During a three-month demonstration period, the working face advanced 210.2 m and produced nearly 600 000 tons of coal. The system realizes safe and highly efficient mining of deep "three-soft" coal seams with a large mining height in the Huainan mining area. The intelligent mining approach adapted to the characteristics of deep and complex coal seams proposed in this study is expected to provide technical support for the safe and efficient exploitation of deep coal resources.