Noise profoundly affects the quality of electromagnetic data, and selecting the appropriate hyperparameters for machine learning models poses a significant challenge. Consequently, the current machine learning denoising techniques fall short in delivering precise processing of Wide Field Electromagnetic Method (WFEM) data. To eliminate the noise, this paper presents an electromagnetic data denoising approach based on the improved dung beetle optimized (IDBO) gated recurrent unit (GRU) and its application. Firstly, Spatial Pyramid Matching (SPM) chaotic mapping, variable spiral strategy, Levy flight mechanism, and adaptive T-distribution variation perturbation strategy were utilized to enhance the DBO algorithm. Subsequently, the mean square error is employed as the fitness of the IDBO algorithm to achieve the hyperparameter optimization of the GRU algorithm. Finally, the IDBO-GRU method is applied to the denoising processing of WFEM data. Experiments demonstrate that the optimization capacity of the IDBO algorithm is conspicuously superior to other intelligent optimization algorithms, and the IDBO-GRU algorithm surpasses the probabilistic neural network (PNN) and the GRU algorithm in the denoising accuracy of WFEM data. Moreover, the time domain of the processed WFEM data is more in line with periodic signal characteristics, its overall data quality is significantly enhanced, and the electric field curve is more stable. Therefore, the IDBO-GRU is more adept at processing the time domain sequence, and the application results also validate that the proposed method can offer technical support for electromagnetic inversion interpretation.
Read full abstract