Study on Conformable Kernel Regularized Grey Model Using Bayesian Optimization

When data is scarce, the grey prediction model is particularly valuable as it adeptly utilizes the available explicit and implicit information to uncover patterns within the data. The Kernel Regularized Nonhomogeneous Grey Model (KRNGM) is an advanced approach that incorporates kernel functions to manage the complexities of nonlinear time series data. In this paper, the KRNGM is further enhanced by integrating Conformable Fractional Accumulation (CFA), resulting in the Conformable Fractional Accumulation Kernel Regularized Nonhomogeneous Grey Model (CFAKRNGM). The CFAKRNGM is designed to improve the model's predictive capabilities by accounting for the nuances of data accumulation over time. To optimize the model's performance, we employ Bayesian optimization, a sophisticated method for tuning hyperparameters. This approach ensures that the CFAKRNGM is calibrated to achieve the best possible predictions. Our research shows CFAKRNGM offers a significant improvement over KRNGM in forecasting North American oil output, thanks to its superior ability to capture data dynamics for more accurate predictions. This success underscores CFAKRNGM's great potential in nonlinear time series forecasting.