PTN: Crude oil price forecasting based on pyramidal attention mechanism

Crude oil, serving as the bedrock of the global industry, is a core resource for ensuring the stable operation of the economy. For a long time, the fluctuations in crude oil prices have attracted significant attention from scholars. However, existing daily forecasting models often perform poorly in multi-step forecasting tasks. In view of this, this paper innovatively constructs a time series network algorithm named PTN based on the pyramidal attention mechanism. By comprehensively considering various influencing factors, PTN conducts forecasts on the prices of Shanghai crude oil futures, aiming to effectively improve the accuracy of crude oil price forecasting.The PTN model constructs the pyramidal attention mechanism by leveraging multiple pooling layers and Transformer encoder layers. It gradually extracts multi-scale features, thereby accurately capturing the long-term dependency relationships within sequences. The data from comparative experiments indicate that under different prediction horizons of 7 days, 14 days, and 30 days, the PTN model outperforms traditional algorithms such as long short - term memory networks and temporal convolutional networks in terms of comprehensive performance. It not only has a relatively high prediction accuracy but also demonstrates good stability when the prediction window is extended. In addition, ablation experiments fully verify the efficiency of the pyramidal attention mechanism. The findings of this study can provide strong support for petroleum production enterprises to optimize their operational planning and offer practical references for investment traders to formulate strategies.