Abstract: 【Purpose/significance】Knowledge search is a core component of invention process, and reducing the knowledge search costs in invention holds paramount significance for enhancing innovation performance. However, there is currently a lack of quantita⁃ tive models on knowledge search rules and search costs in complex invention processes, which is an urgent research gap that needs to be addressed.【Method/process】Based on constructing a knowledge network for invention problems, this study models common heuris⁃ tic knowledge search rules and search costs, aiming to explore search rules with lower knowledge search costs, and estimates idealized knowledge search costs using the Q-learning algorithm.【Result/conclusion】Simulating the proposed knowledge search model using patent data in the lithography technology domain, the results show that prioritizing the search for more familiar or commonly used do⁃ main knowledge generally effectively reduces search costs. However, there remains a substantial gap between the knowledge search costs obtained using all heuristic search rules and the ideal search cost.【Innovation/limitation】This study firstly proposes a quantita⁃ tive knowledge search model for solving specific invention problems, which can provide specific recommendations for knowledge search decisions in technical research. The main limitation is the assumption that the constructed prior knowledge network contains all solution knowledge elements, which needs to be relaxed in future research.