In order to analyze and compare the differences in pore structures between shale gas and shale oil formations, a few samples from the Longmaxi and Bakken Formations were collected and studied using X-ray diffraction, LECO TOC measurement, gas adsorption and field-emission scanning electron microscope. The results show that samples from the Bakken Formation have a higher TOC than those from the Longmaxi Formation. The Longmaxi Formation has higher micropore volume and larger micropore surface area and exhibited a smaller average distribution of microsize pores compared to the Bakken Formation. Both formations have similar meso-macropore volume. The Longmaxi Formation has a much larger meso-macropore surface area, which is corresponding to a smaller average meso-macropore size. CO₂ adsorption data processing shows that the pore size of the majority of the micropores in the samples from the Longmaxi Formation is less than 1 nm, while the pore size of the most of the micropores in the samples from the Bakken Formation is larger than 1 nm. Both formations have the same number of pore clusters in the 2–20 nm range, but the Bakken Formation has two additional pore size groups with mean pore size diameters larger than 20 nm. Multifractal analysis of pore size distribution curves that was derived from gas adsorption indicates that the samples from the Longmaxi Formation have more significant micropore heterogeneity and less meso-macropore heterogeneity. Abundant micropores as well as meso-macropores exist in the organic matter in the Longmaxi Formation, while the organic matter of the Bakken Formation hosts mainly micropores.

In most cases, reservoir properties at one certain depth in the layer can be explicated by logging signals at just this depth point. In fact, the properties of complex reservoirs are often implicated in logging signals from the whole adjacent region of this certain depth point. So far, there is no effective way to solve this problem completely. For the first time, this letter tried to build a fully convolutional neural network (FCNN) to detect hydrocarbon from logging signals for the tight gas reservoir of Ordos Basin. The FCNN was based on a well-designed VGG-net. The prediction comparison between the empirical approach (EMA) and FCNN was implemented on 48 layers. The accuracy of FCNN was about 87.5%, which was higher than that of the EMA (75.0%). FCNN provided more reliable gas testing recommendations, especially when thin layers led to complex reservoir conditions. Deep learning (DL) has been proven to be an automatic feature extraction and direct hydrocarbon detection approach from logging signals. We are looking forward to its improvement and development in geophysics.