The explosive growth of IoT devices has spawned complex cybersecurity threats, yet traditional intrusion detection systems (IDS) face significant challenges in handling high-dimensional redundant data, protocol noise interference, and edge device resource constraints. This paper proposes a causal-driven multi-objective grey wolf optimization framework (Causal-MOGWO), which integrates the global search capability of grey wolf optimization (GWO) with the causal reasoning power of structural causal models (SCM) to achieve precise feature selection in IoT traffic. Causal-MOGWO consists of three core modules: (1) Causal Feature Selection Layer: Constructs a structural causal model of network traffic using the LiNGAM algorithm, quantifies the causal relationship between features and attack behaviors via average treatment effect (ATE), and generates causal weight vectors to suppress spurious correlations; (2) Improved Multi-Objective Grey Wolf Optimization Layer: Adopts hybrid encoding strategies and a quantum-based position update mechanism, combining three objectives of detection accuracy, feature dimensionality, and computational overhead to dynamically maintain Pareto front solutions, addressing premature convergence in traditional GWO; (3) Elastic Contractive Autoen-coder (ECAE): Achieves feature space compression through contraction regularization and ElasticNet regularization. Experiments demonstrate that Causal-MOGWO reduces feature dimensions from 49 to 10 (80% compression) on UNSW-NB15, and achieves >99.5% detection accuracy in multi-class classification on NSL-KDD, providing an effective solution for lightweight and high-precision intrusion detection in high-dimensional IoT traffic.