This paper introduces a novel deterministic learning (DL)-based knowledge fusion neural control strategy tailored for unknown robot manipulators with predefined performance. For two different control training scenarios, online fusion and offline fusion control schemes are proposed respectively. In the online knowledge fusion scheme, a collaborative control methodology is embraced, integrating a mechanism for propagating weight update information into the neural network (NN) learning algorithm of DL. This integration facilitates the eventual convergence of system weights across all operational systems toward a shared optimal value. For the offline fusion control scheme, it transforms the fusion problem of multi-trajectory closed-loop dynamics knowledge learned by deterministic learning (DL) into the least squares solution problem of a system of linear equations. Moreover, leveraging the fused dynamic knowledge acquired through the aforementioned approaches, we construct a neural network (NN) learning controller based on integrated knowledge to realize a multi-task intelligent control for robotic manipulators in intricate scenarios. The simulation section provides empirical evidence of the efficacy of the proposed approach.