Low earth orbit (LEO) satellite networks (LSNs) are regarded as a critical component of next-generation space-air-ground integrated networks. However, such networks remain vulnerable to cascading failures triggered by natural or anthropogenic factors, so potential attackers may strategically compromise critical nodes to induce systemic network collapse. Critical node detection (CND) through simulation studies enables proactive vulnerability analysis and targeted protection, which is recognized as a vital approach to enhance the stability of LSNs and communication efficiency. Existing simulation models fail to adequately capture the dynamic nature and heterogeneity of LSNs, resulting in unreliable performance evaluations. To overcome limitations, we propose a dynamic heterogeneous cascading failure model specifically designed for LSNs. Then, we reformulate the CND problem as a multi-constrained optimization problem and develop a Critical Satellite nodes Detection Evolutionary Algorithm called CSDEA. Extensive experimental results validate the superior effectiveness of the proposed method in balancing constraint satisfaction and solution optimality.