综放开采煤矸微波成像机理及混合度识别方法研究

煤矸混合度识别是实现综放智能化开采的核心技术，已成为煤炭开采领域亟需解决的技术难题。本项目以非接触式微波探测技术为手段，综合运用电磁逆散射理论、非线性反演算法、深度学习等理论和方法，开展综放开采过程中的煤矸微波成像机理及混合度识别方法研究。在掌握不同工况下煤矸微波散射机理及多重散射效应的基础上，构建煤矸电磁逆散射非线性模型并进行仿真分析；设计融合多重散射效应物理机制的深度神经网络架构，建立全变分正则化约束下的深度神经网络电磁反演模型，揭示煤矸本构参数与散射场之间的非线性关系，实现煤矸精准超分辨成像；分析微波成像的原位跟踪体系，掌握煤矸混合体宏观结构和背景噪声的演化规律，设计无监督条件下煤矸混合度的快速识别方法，并开展实验验证和评估，为实现综放工作面的自动精准放煤提供新的理论和方法。

The recognition of coal-and-gangue mixing degree is the core technology to realize the intelligent mining of fully mechanized caving face, which has become an urgent-needed technical problem in the field of coal mining. In this project, the non-contact microwave detection technology is adapted as a means, and the electromagnetic inverse scattering theory, nonlinear inversion algorithm and depth learning method are comprehensive used to study the microwave imaging mechanism and mixing degree identification method of coal-and-gangue in the process of top coal caving. On the basis of mastering the microwave scattering mechanism and multiple scattering effects of the coal-and-gangue under different working conditions, the electromagnetic inverse scattering nonlinear model of coal-and-gangue is constructed and simulated. The deep neural network architecture is designed to integrate the physical mechanism of multiple scattering effects, and the electromagnetic inversion model is established under the constraints of full variational regularization to reveal the nonlinear relationship between the constitutive parameters and its scattering field of coal-and-gangue, so as to achieve accurate super-resolution imaging. By analyzing the in-situ tracking system of microwave imaging and grasping the evolution law of the macroscopic structure of coal-and-gangue and background noise, a fast recognition method of coal-and-gangue mixing degree under unsupervised conditions is designed, and some experimental verifications and evaluations are conducted. The research provides a new theory and method for realizing automatic and accurate coal caving in fully mechanized caving face.