Abstract: To address the challenges in controlling hydraulic fracture propagation trajectories and stimulated volumes within multi-lithologic coal measure strata - particularly those caused by formation interfaces and interlayer strength variations - this study employs a combined finite-discrete element method (FDEM) to investigate hydraulic fracture behavior. The research reveals twelve distinct propagation modes under multifactor coupling conditions, classifiable into four categories: interface penetration, penetration with interface extension, interface tracking, and interface arrest. We developed a novel hybrid artificial intelligence model integrating BP neural networks with differential evolution (DE) and grey wolf optimization (GWO) algorithms (BP-DEGWO) to predict fracture trajectories and stimulation effectiveness. Systematic analysis identified key controlling factors - including rock strength contrast coefficient, interface dip angle, interfacial strength, injection rate, and perforation angle - and quantified their relative importance under varying in-situ stress conditions. The study further proposes an intelligent optimization framework for hydraulic fracturing design in stratified formations based on the BP-DEGWO model. This approach provides a predictive tool for fracture geometry in heterogeneous coal-bearing strata, and a methodological reference for AI-assisted fracturing optimization. The findings offer significant implications for enhancing stimulation efficiency in multi-lithologic unconventional reservoirs.