Full text loading...
Shallow hydrocarbon accumulation in fault-bound traps is an important part of offshore exploration. However, shallow hydrocarbon accumulation is highly uncertain and unpredictable, which poses great challenges to high-cost offshore drilling. Taking the Pinghu Slope Belt of Xihu Depression in the East China Sea Basin as an example, this study proposes a systematic machine learning method for evaluating and predicting offshore shallow hydrocarbon accumulation in fault-bound traps by using neural network ensemble algorithm. Firstly, the main control factors affecting the shallow hydrocarbon accumulation were screened, quantitatively evaluated and dimensionally reduced as model input. Secondly, the bagging and boosting ensemble algorithms using BP neural network (BPNN) as the component were applied to construct shallow hydrocarbon accumulation prediction models with the hydrocarbon column height as the output. Then, the performance of different models was evaluated. Finally, the best-performing model was employed to the shallow hydrocarbon accumulation prediction in the area not involved in model construction. The results confirm that the boosting ensemble model is superior to bagging ensemble model and single BPNN model in evaluating shallow hydrocarbon accumulation in fault-bound traps. The methods proposed in this study can help petroleum explorers or companies reduce offshore hydrocarbon exploration risks and avoid costly drilling mistakes.