Research on Shale TOC Prediction Method Based on Improved BP Neural Network

With the increasing attention to shale oil and gas in the field of oil and gas exploration and development, accurate prediction of TOC content has become the key to evaluating shale gas sweet spots. This paper studies a method for predicting shale TOC content using a BP neural network optimized by an improved cuckoo search algorithm. First, for the Longmaxi Formation shale, through logging sensitivity analysis, seven logging parameters sensitive to TOC content were determined: DEN, AC, RT, U, K, GR, and CNL. Using these parameters, a CSBP model was established and compared with the traditional BP neural network, multiple linear fitting method, and extended ∆lgR method. The results show that the CSBP model has higher prediction accuracy and generalization ability, with the mean absolute error and mean absolute percentage error being 0.38 and 15.00% respectively, which are significantly better than other methods. Further, the CSBP model was applied to predict the TOC content of Well W16 in the study area and verified by comparing with the measured TOC values. The correlation between the predicted and measured values is 0.89, and the change trends are consistent, confirming the applicability of the CSBP model. Finally, combined with the seismic waveform - guided simulation inversion technology, the planar and spatial distribution of TOC in the study area was predicted. The correlations between the predicted and measured values of four wells in the study area are all greater than 0.89. This method has high accuracy in the three - dimensional TOC content prediction of shale reservoirs and provides technical support for the evaluation of shale gas sweet spots in the work area.