The Taipu Sluice is a key regulation structure in the Taihu Basin, and accurate forecasting of its discharge is crucial for coordinated flood control, water supply, and ecological regulation in the basin. However, conventional forecasting approaches suffered from issues such as fixed parameters, unclear response mechanisms to multi-source driving factors, and ambiguous nonlinear precipitation and water level thresholds. To address these limitations, this study integrated principal component analysis, corrected stochasticity rate method, and K-means clustering to develop a forecasting model for Taipu Sluice discharge driven by multiple factors and to establish a threshold identification system for the response of Pingwang water level to precipitation. The results showed that: (1) The key factors affecting Taipu Sluice discharge were summarized into three principal components: regional water-level synergy, rainfall-runoff effect, and upstream-downstream water-level antagonism, with a cumulative contribution rate of 93.47%. The variable importance ranking indicated that the 8:00 water level at Pingwang contributed the most. (2) A simplified multiple regression model based on the water levels at Pingwang, Wangjiangjing, and Chenmu stations achieved relatively high forecasting accuracy. Under high water level conditions in Taihu Lake (water level > 3.80 m), the discharge of Taipu Sluice exhibited a strong linear relationship with the Taihu-Pingwang water-level difference (R = 0.92). (3) Corrected stochasticity analysis demonstrated that when the daily variation amplitude of Pingwang water level was ≥0.10 m, deterministic processes dominated; otherwise, stochastic fluctuations prevailed. Through K-means clustering, a regional precipitation threshold of 25.0 mm, which triggered a significant change in Pingwang water level, was identified and could serve as an early-warning indicator for Pingwang water level, with high classification accuracy. The precipitation threshold and the water level variation threshold together formed a dual-threshold early-warning system, providing proactive decision support for adaptive regulation of Taipu Sluice and risk avoidance during heavy precipitation events. The coupled framework proposed in this study, which integrated principal component dimensionality reduction, regression modeling, and clustering-based threshold identification, offers a data-driven paradigm that can be referenced for hydrological forecasting and refined regulation under complex engineering control in plain tidal river network regions.