Abstract: Ocean acidification forecasting has important implications for studying global carbon dioxide emissions reductions. However, due to seasonal and cyclical features, ocean acidification forecasting remains an extremely challenging task. Therefore, this paper proposes a grey wolf optimized fractional-order-accumulation discrete grey seasonal model (GFSM(1,1)). The GFSM(1,1) model improves the prediction of ocean acidification in two ways: The new information priority of seasonal data is improved by the fractional accumulation operator, and the adaptability of the grey model to seasonal data is increased by seasonal item parameters. The above two works have significantly improved the prediction accuracy of the grey prediction model for ocean acidification. The prediction results in practical cases prove that the prediction effect of the GFSM(1,1) model is not only better than the existing grey models (FMGM(1,N)、NSGM(1,N), and GM(1,1)) but also better than statistical models (Nonlinear regression and ARIMA), traditional neural network model (LSTM) and deep learning model (SVM). Finally, the GFSM(1,1) model is applied to the prediction of seawater acidification. The forecast results show that the ocean will be acidified at a rate of 0.001863 per year, and the pH of the ocean will decrease by about 0.03% per year compared to the same period in previous years. 

Key words: Ocean Acidification, GFSM(1,1) Model, Grey Wolf Optimization Algorithm, Fractional Accumulation Operator, Seasonal Model, CO2 Reduction