Multiples prediction and adaptive subtraction are two important aspects in the wave‐equation‐based surface‐related multiples suppression method. In this work, GPU with parallel computation advantage is exploited to accelerate surface‐related multiples prediction, which improves calculation efficiency greatly. In the adaptive subtraction algorithm, multiples modeling traces contain predicted multiple trace, high‐frequency reconstruction trace from predicted multiple trace, Hilbert transform trace from predicted multiple trace, and their translation trace. Hilbert transform trace will compensate phase information of multiples, and high‐frequency reconstruction trace improves high‐frequency information of predicted multiples, which compensate energy difference in the frequency band. Iterative algorithms are used during prediction and subtraction. Iterative prediction improves kinematic properties and iterative subtraction improves dynamic properties, which make predicted multiples match actual multiples well. The proposed method is applied to the theoretical SMAART model and real marine data. The test results show that the method can predict multiples of seismic data efficiently, also suppress surface‐related multiples without damaging primary effectively.