As an important energy and chemical raw material in the world, the price fluctuation of crude oil has been affecting the economic level of all countries in the world, however, the crude oil price has a strong nonlinear trend and chaotic characteristics, which makes the crude oil price prediction very challenging. Based on this, this paper proposes a three-stage forecasting model based on "two-mode decomposition-multi-scholar forecasting-nonlinear integration". In this study, we first use bimodal decomposition to obtain two different sets of multiscale components to show the detailed features of crude oil prices from different perspectives, then we use long-short-term neural network (LSTM), kernel extremum learning machine (KELM), and gated recursive unit (GRU) to learn the deeper features of the two sets of components from different aspects, and finally, we use the multi-objective viscosity optimization algorithm (MOSMA) and extreme gradient Boosting Tree (XGBoost) to integrate the six models and obtain point prediction and interval prediction results. The simulation experimental results show that compared with the comparison models, the bimodal decomposition can better reflect the detailed information of the original series of oil price and realize the complementary advantages between different decomposition methods; the nonlinear integration can effectively combine the advantages of different models and improve the prediction accuracy.