Inversion of self-potential data based on improved sparrow search algorithm

Self-potential (SP) method is a natural source geophysical prospecting technology with low cost and simple operation for field observation. Generally, the inversion of SP data has the characteristics of ill-posedness and non-linearity. The conventional inversion algorithms are classified into the types of local optimization and global search. However, it is hard for the gradient-based local optimization algorithm to obtain the global optimal solution, and the inversion results are dependent on the construction of initial model. The traditional global search algorithm has the disadvantages of low convergence rate, insufficient ability of escaping local minima and instability. Thus, a new global optimization strategy (sparrow search algorithm, SSA) was improved on this basis. Specifically, the sparrow swarm was initialized with the chaotic map superimposed reverse learning strategy, and the locations of sparrows were updated using the Levy flight strategy based on the random probability, so as to further improve the possibility of the algorithm to explore the solution space and escape the local minima in terms of the enhanced algorithm. Then, the sparrow search algorithm (SSA) and improved sparrow search algorithm (ISSA) were applied to the inversion interpretation of the synthetic SP data (without and with 10% and 30% of random noise contamination) and the field data measured from India and France, so as to comparatively verify the inversion results of the improved algorithm. As shown in the theoretical test results, the inversion errors of SSA in the vertical cylinder and inclined sheet models without noise interference are 0.42% and 0.25% respectively. In contrast, the inversion errors of ISSA are 0.06% and 0.07%, respectively. Therefore, the fitting accuracy of the improved algorithm is increased to 3-7 times of SSA. Besides, the convergence speed and accuracy of ISSA are obviously better than SSA in the convergence curve of the objective function. Moreover, the stability and accuracy of SSA inversion parameters, as well as its fitting degree to abnormal response curve, will become worse with the presence of random noise, and the erosion will become more drastic as the amplitude of the random noise increases. However, ISSA can not only maintain its fast convergence behavior, but also obtain smaller fitting errors in the abnormal response curve. Actual data inversion results further verifies that the ISSA has faster convergence speed, stronger anti-interference capability, higher precision and more robustness, which could be effectively utilized to the quantitative interpretation of SP data, and promoted to solve other geophysical inversion problems.