<p>Technology fusion is a primary driver of modern innovation. Forecasting its emergence is strategically vital for identifying nascent technological opportunities and anticipating industrial transformations. However, capturing the higher-order dependencies and complex temporal evolution inherent in this process—which fundamentally differ from dyadic combinations—presents a unique challenge. To address this challenge, we construct a framework based on temporal hypergraph link prediction (THLP). Fine-grained technology domains are modeled as nodes, and their fusion events—the co-occurrence of multiple technologies in a single patent—are modeled as time-evolving hyperedges. We systematically design a multi-dimensional feature framework and propose a temporal hypergraph deep neural network to predict potential fusion events. Then, we validate our framework against multiple baselines using patent data from emerging technology fields. The superior predictive performance confirms the necessity of modeling higher-order dependencies within technology combinations. Beyond prediction, our framework reveals stable key drivers extracted from a hypergraph for technology fusion. Furthermore, it uncovers a U-shaped effect of sub-module maturity and an inverted U-shaped effect of external dependencies, providing deeper insights into the fusion process. This research offers a powerful analytical tool for managers, policymakers, and investors to identify emerging technological opportunities and formulate effective innovation strategies.</p>