Efficient marine prediction data correction method based on CPU-GPU collaborative computing
Jiahao Zhang, Li Ma, Yang Li · IET conference proceedings. · 2025
In marine environmental forecasting, accurate correction of marine prediction data is crucial for improving forecast accuracy. However, with the rapid increase in marine data volume and the diversification of data sources, traditional data processing and correction methods are struggling to meet the demands of real-time performance and accuracy. This challenge is particularly pronounced in the context of high-dimensional, multi-source data. This paper proposes a "G-M-C" marine data correction method based on CPU-GPU collaborative computing, aiming to overcome the bottlenecks in processing large-scale marine data. The method is implemented in three steps, focusing on data grid interpolation, Ensemble Empirical Mode Decomposition (EEMD) processing, and neural network training. Specifically, grid interpolation is moved from the CPU to the GPU, utilizing intelligent GPU resource scheduling to fully exploit GPU's parallel computing capabilities; EEMD is optimized to accelerate the data analysis process; and neural network training is restructured to leverage the multi-core architecture of GPUs, significantly improving training efficiency. Experimental results demonstrate that the optimized model achieves an 80% efficiency improvement, significantly reducing computation time while enhancing prediction accuracy. This method provides strong support for the real-time processing and analysis of marine data.