Gravity inversion can be used to obtain the spatial structure and physical properties of subsurface anomalies through gravity observation data. With the continuous development of machine learning, geophysical inversion methods based on deep learning have achieved good results. Geophysical inversion methods based on deep learning often employ large-scale data sets to obtain inversion networks with strong generalization. They are widely used but face a problem of lacking information constraints. Therefore, a self-constrained network is proposed to optimize the inversion results, composed of two networks with similar structures but different functions. At the same time, a fine-tuning strategy is also introduced. On the basis of data-driven deep learning, we further optimized the results by controlling the self-constrained network and optimizing fine-tuning strategy. The results of model testing show that the method proposed in this study can effectively improve inversion precision and obtain more reliable and accurate inversion results. Finally, the method is applied to the field data of Gonghe Basin, Qinghai Province, and the 3D inversion results are used to effectively delineate the geothermal storage area.

As a key component of remote-sensing technology, satellite gravity observation offers extensive coverage and high accuracy, effectively compensating for the shortcomings of terrestrial gravity measurements. Three-dimensional gravity data inversion can predict the physical property and spatial distribution of geological formations beneath the surface by analyzing the gravity data. In this paper, the heat source position within the Gonghe Basin’s geothermal system is identified through the analysis of satellite gravity data, and a constrained deep-learning inversion method is proposed. This method adds the fitting data constraints and depth-weighting function into the network model establishment of deep learning, and trains the network through a large number of datasets, so that the network is constrained by physical information in the training process to obtain the results with a better data-fitting accuracy and higher depth resolution. The proposed method is employed to verify the synthetic model data, and the inversion results indicated that, compared to other methods, the deep-learning gravity inversion method with the addition of physical information constraints has a higher inversion accuracy and depth resolution. Finally, the inversion results based on satellite gravity data revealed the presence of numerous low-density bodies in the underground range of 10–35 km in the research area. It is speculated that this part could be the heat source of the geothermal system in the Gonghe Basin. The findings from this study are expected to contribute to a deeper comprehension of the formation of the geothermal system in the region.

Three-dimensional magnetic inversion allows the distribution of magnetic parameters to be obtained, and it is an important tool for geological exploration and interpretation. However, because of the redundancy of the data obtained from large-scale investigations or high-density sampling, it is very computationally intensive to use these data for iterative inversion calculations. In this paper, we propose a method for compressing magnetic data by using an adaptive quadtree decomposition method, which divides the two-dimensional data region into four quadrants and progressively subdivides them by recursion until the data in each quadrant meets the regional consistency criterion. The method allows for dense sampling at the abnormal boundaries with large amplitude changes and sparse sampling at regions with small amplitude changes, and achieves the best approximation to the original data with the least amount of data, thus retaining more anomalous information while achieving the purpose of data compression. In addition, assigning values to the data in the quadrants using the averaging method is essentially equivalent to average filtering, which reduces the noise of the magnetic data. Testing the synthetic model and applying the method to mineral exploration a prove that it can effectively compress the magnetic data and greatly improve the computational efficiency.