Abstract:
The establishment of a 3D urban geological model requires standardized information on borehole stratigraphy. However, due to the cognitive differences among professional and technical personnel as well as the massive volume of data, it is difficult to efficiently develop regularized and unified stratigraphic information. An artificial neural network (ANN) classification model is constructed to achieve efficient processing of massive non-standard borehole stratigraphic data. By selecting key feature variables based on geological experience, applying one-hot encoding technology to efficiently convert text variables into numerical variables, constructing a multi-layer neural network, using genetic algorithms to optimize the number of network layers and neurons, and introducing the AM-softmax function in ANN output layer, the performance of stratigraphic classification is improved. The model achieved a classification accuracy of 95.3%, significantly reducing manual labor. Compared with traditional classification model (Random Forest, Support Vector Machine, Logistic Regression, and K-Nearest Neighbors), it demonstrated superior accuracy and faster computation speed. The model provides critical pre-work support for 3D geological modeling and offers a new framework for selecting influential factors in stratigraphic standardization. The AM-softmax function is applied to ANN classification model for the first time, aiming to improve the recognition ability of drilling strata identification. By integrating geological experience with coding technology, the efficient digitization of textual variable is achieved. The proposed classification model significantly reduces the cost of manual standardization of borehole strata, and compared to traditional algorithms, it offers both high accuracy and efficiency, providing a new paradigm for 3D geological modeling and classification factor screening.