地质大数据的发展促进了地质勘探和人工智能的结合。在地质勘探及储层评价中，岩性识别是一项重要的工作，然而传统的岩性识别方法周期长、效率低、精度有限，使得地质工作人员难以及时并准确的识别岩性，已不能满足实际生产需要。以往相关研究和实验表明，人工智能神经网络可对测井数据进行自动预测识别，具有较强的容错能力，能够提高识别的自动化程度和解释精度。此外，神经网络模型还可以与随钻测量装备配合实现实时预测识别，为钻探工程提供技术支撑。

针对传统岩性识别方法的不足，本文通过理论分析、样本测试以及现场验证等方法，基于SVND-PNN神经网络模型对煤矿井下岩性识别进行研究，研究表明该模型对煤矿井下岩性识别的准确率得到提升。本次研究结果对煤矿井下智能岩性识别提供了一定的理论依据和参考意义。主要研究成果如下：

建立了SVND-PNN神经网络模型。该模型将SVND算法引入到PNN中，以克服PNN在样本数量增加时训练复杂度呈指数级增长的问题，从而限制其在大规模问题上的应用。该算法通过选择代表性样本来减小训练数据的维度，从而提高了训练速度。此方法可以提高岩性识别的效率，为其在实际应用中的表现提供支持。进行了仿真实验，将SVND-PNN模型与其他传统的方法进行对比，该模型在岩性识别方面有更高的准确率，验证了该模型的有效性和优越性。为了展现该模型在现场应用的效果，对高家堡矿区的标准井段岩性样本进行实验，综合比较四种识别方法的实验结果，发现SVND-PNN模型相对于传统的模型而言，具有更高的岩性识别准确率，因此该模型在岩性识别方面具有一定的可行性。

The development of geological big data has facilitated the integration of geological exploration and artificial intelligence. In geological exploration and reservoir evaluation, rock identification is an important task. However, traditional rock identification methods are time-consuming, inefficient, and have limited accuracy, making it difficult for geological workers to identify rock types in a timely and accurate manner, which cannot meet the needs of actual production. Artificial intelligence neural networks can automatically predict and identify logging data, with strong fault tolerance, and can improve the automation and interpretation accuracy of identification. In addition, neural network models can be combined with downhole measurement equipment to achieve real-time prediction and identification, providing technical support for drilling geological exploration.

This study aims to improve the accuracy of underground rock recognition in coal mines through theoretical analysis, sample testing, and on-site verification methods based on the SVND-PNN neural network model. The research shows that this model improves the accuracy of rock recognition in coal mines. The results of this study provide a theoretical basis and reference for intelligent rock recognition in coal mines. The research The main results are as follows:

Established an SVND-PNN neural network model. This model introduces the SVND algorithm into PNN to overcome the exponential increase in training complexity as the number of samples increases, thereby limiting its application in large-scale problems. This algorithm reduces the dimensionality of training data by selecting representative samples, thereby improving training speed. This method can improve the efficiency of lithology recognition and provide support for its performance in practical applications. Simulation experiments were conducted to compare the SVND-PNN model with other traditional methods. The model showed higher accuracy in lithology recognition, verifying its effectiveness and superiority. In order to demonstrate the effectiveness of this model in on-site application, experiments were conducted on standard well section lithology samples in the Gaojiabao mining area. By comprehensively comparing the experimental results of four recognition methods, it was found that the SVND-PNN model has a higher accuracy in lithology recognition compared to traditional models. Therefore, this model has certain feasibility in lithology recognition.