Petrochemical Commodity Price Prediction Model Based on Wavelet Decomposition and Bayesian Optimization

The current predictions of petrochemical commodity prices are generally inadequate in terms of effective time and frequency domain modeling, and are plagued by issues such as lagging predicted values and memory dependence. To address these problems, this paper proposes an attentional neural network model that utilizes wavelet decomposition and Bayesian optimization. Initially, the intricate petrochemical commodity price data is decomposed into sub-series of varying frequencies, utilizing wavelet decomposition in a divide-and-conquer strategy to extract longitudinal frequency domain features. Subsequently, a neural network that has been improved by an attention mechanism is utilized to extract the transversal time domain features. Throughout this process, a Bayesian computing approach is proposed to optimize the hyperparameters. In order to investigate the forecasting performance of the proposed model, extensive experiments were conducted utilizing the prices of propylene, butadiene, phenol, and toluene, with six other state-of-the-art methods included for comparison purposes. The experimental results demonstrate that the proposed model has superior performance when it comes to petrochemical commodity price forecasting.