采用动力学模态分解和长短期记忆网络的 压缩空气储气装置流场动态预测方法

To address the challenge of rapidly and accurately predicting the transient evolution of operational parameters in within the air storage tanks of compressed air energy storage (CAES) systems, this study proposes a reduced-order flow field prediction model based on dynamic mode decomposition (DMD)and long short-term memory (LSTM)network. This model achieves decoupled modeling of the spatiotemporal evolution features of the complex nonlinear flow fields inside storage tanks and establishes a method for accurate global parameter characterization and rapid flow field prediction. DMD decouples and reduces the full-order spatiotemporal datasets of key physical quantities (e. g., velocity, temperature, and pressure)within the storage tanks, enabling accurate extraction of dominant modes governing flow field evolution. An LSTM network model is then constructed to predict the temporal evolution characteristics of the low-dimensional dynamical system, reconstruct high-dimensional spatiotemporal flow fields, and achieve dynamic and precise prediction of full-order flow parameters. The experimental results demonstrate that the average relative errors for pressure field prediction is 0.05%, 0.19% for the velocity field, and 0.83% for the volume fraction field. The model reduces computational cost by four orders of magnitude while maintaining prediction accuracy. This method enables high-precision real-time flow field prediction for the air storage tanks of CAES systems, significantly improving computational efficiency without sacrificing prediction accuracy, and provides valuable insights for flexible and efficient regulation of such systems under all operating conditions.