【摘要】 作为重力勘探领域的研究热点，三维重力反演能够为地质构造研究和矿产勘查提供丰富的信息.然而，传统三维重力反演方法在实际应用中常面临计算量大与存储成本高的问题.子空间反演方法作为一种广泛应用于地震与电法勘探等领域的反演方法，通过子空间线性表示模型项，改变优化目标，降低方程组维数，从而有效降低反演计算量.将子空间方法应用于三维重力反演，可提升初步地下密度建模的效率，并在后续研究中减少修正反演密度模型的试错成本.与以往将迭代过程中的梯度作为子空间的做法不同，本文提出了一种通过将二维平面数据分布特征投影至三维模型分布特征的子空间计算方法，并通过重力垂向一阶导数和重力倾斜角投影得出更优越的模型子空间，反演结果显示，该方法能更准确地反映物性高值区与场源地质体的边界，并且具有更强的聚焦能力.为了进一步降低子空间反演的计算成本，本文将其与频率域卷积快速算法相结合，有效提升了正演计算和矩阵运算的计算效率.在理论模型试验中，验证了该投影模型子空间反演方法的可行性和优越性，相较于传统反演方法计算效率整体提升97.33%～99.14%.在实际数据测试中，本文对奥林匹克坝研究区进行了重力子空间反演，于较短计算时间内获得了与矿区真实地质构造相符的反演结果.更多还原

【Abstract】 As a current research focus in the field of gravity exploration, three-dimensional(3D) gravity inversion can provide valuable information for geological structure research and mineral exploration. However,traditional 3D gravity inversion methods often face the challenges of large computational loads and high storage costs in practical applications. The subspace inversion method, widely used in seismic and electrical exploration,reduces the computational cost by using subspace linear representations of model terms, modifying the optimization objective, and lowering the dimensionality of the system. Applying the subspace method to 3D gravity inversion can improve the efficiency of preliminary underground density modeling and reduce trial-anderror costs in the subsequent refinement of the density model. Unlike previous methods that use the gradient from the iterative process as the subspace, this paper proposes a novel approach to compute the subspace by projecting the two-dimensional(2D) plane data distribution characteristics onto the 3D model distribution features. By using the vertical first-order derivative of gravity and the gravity tilt angle projection, a more optimal model subspace is obtained. The inversion results show that this method can more accurately reflect the boundaries between high-value physical property zones and the geological bodies of the source field, while also demonstrating a stronger focusing ability. To further reduce the computational cost of subspace inversion, this paper combines the method with a frequency-domain convolution fast algorithm, significantly improving the efficiency of forward modeling and matrix operations. In theoretical model tests, the feasibility and superiority of this projected model subspace inversion method were validated, with an overall improvement in computational efficiency of 97.33% to 99.14% compared to traditional inversion methods. In practical data tests, gravity subspace inversion was applied to the Olympic Dam study area, and inversion results consistent with the actual geological structure of the mining area were obtained in a much shorter computational time.更多还原