The simultaneous solution of multiple roots of nonlinear equation systems is a challenging task. Although differential evolution algorithms have been widely applied to solve such complex problems, the individual evolutionary information contained in the differential vectors generated during the evolutionary process is often not fully utilized. To address this, this paper proposes a meta-knowledge-based niching differential evolution algorithm. Its main features are as follows: (1) The differential vectors generated in the evolutionary process are regarded as “meta-knowledge” that contains search experience; (2) A neural network model is designed to learn and model meta-knowledge, with environmental feature vectors taken as the model input. This enables accurate perception of the current environment of individuals, thereby improving the differential vectors generated by prediction and efficiently guiding the subsequent evolution of the population; (3) A mutation method based on meta-knowledge is proposed to improve the search efficiency of the algorithm. Experimental results show that the proposed algorithm can effectively realize the simultaneous solution of multiple roots of nonlinear equation systems and perform excellently in terms of the peak rate (PR) and success rate (SR) indicators.