In the application of detecting and locating low-altitude drones using acoustic signals, the existing planar microphone arrays face problems such as low directional resolution and poor interference resistance. The research focuses on addressing these challenges by establishing a stochastic three-dimensional(3D) array optimization design model with multiple constraints, suitable for drone direction finding. Additionally, an optimization-solving method based on an elite-tournament selection strategy is proposed. Based on a four-ring 3D array configuration, the model minimizes the peak side-lobe level as the optimization objective while setting array structural constraints and limiting the beam main lobe width. A multi-parameter constraint optimization model for stochastic 3D arrays is constructed. Furthermore, an elite-tournament selection strategy is proposed for optimizing the solution process. The elite strategy and tournament strategy are combined into a multi-fusion selection strategy, which is applied during the iterative process of genetic algorithm optimization. This combination enhances the convergence rate of the algorithm and its global search capability, leading to the achievement of the optimized array configuration. Simulation and experimental results show that the direction finding pattern of the target array exhibits fewer false detection points, demonstrating improved noise immunity and spatial resolution. Compared to the four-ring 3D array, the target array reduces the detection failure rate for low-altitude drones by 4.33%, and the azimuth and elevation angle errors decrease by 1.54° and 0.73°, respectively. The maximum detection distance is improved by 12 m.