Traffic signal control plans significantly impact transportation system  efficiency at intersections. Adaptive plans that adjust to real-time  conditions are more effective. Reinforcement learning (RL) adapts  strategies based on environmental feedback, making it proficient in  handling dynamic traffic scenarios. However, current RL methods have  long computational periods, hindering their adoption for new scenarios.  Another approach is to optimize the RL model itself for fast learning or  make it transferable with learned experience. The underlying control  algorithm should ensure convergence and minimize parameter sensitivity  in diverse migration scenarios. We propose MetaSignal, an efficient  meta-RL method for traffic signal control. Our approach uses Fourier  basis as the value function approximation in RL, offering advantages  like convergence facilitation, error bound achievement, and reduced  parameter dependence. The model-agnostic meta-learning framework allows  for effective adaptation to target scenarios with limited training cost.  Empirical evaluation shows promising and stable performance in  comprehensive experiments in synthetic and real-world traffic networks.