نوع مقاله : مقاله پژوهشی
1 آمار، دانشکده علوم ریاضی، دانشگاه یزد، یزد، ایران
2 دانشکده مهندسی هوافضا، دانشگاه صنعتی خواجه نصیرالدین طوسی، تهران، ایران
عنوان مقاله [English]
Submarine robots or autonomous underwater vehicles (AUVs) are one of the most important tools for identifying, monitoring and inspecting the marine environment and the oceans. In addition, it is used for applications such as tracking surface targets. In the process of tracking a target by an autonomous underwater vehicle, designing the most efficient guidance law is of particular importance. In order to evaluate the efficiency of the tracking process, various criteria such as ease of implementation, less need for target data and the probability of hitting the target must be considered. Among these factors and other effective factors in evaluating tracking performance, hit probability is the most important and telling variable. in complex situations, the most common way for calculating this parameter is the Monte Carlo method. This method is based on performing multiple simulations of the AUV and target motion for various uncertainties in the problem. The ratio of the number of times that the tracking process is successful provides an estimate of the hit probability. However, in order to achieve good accuracy, it is necessary to select a sufficiently large number of repetitions in the Monte Carlo method and therefore the computational cost of calculating the hit probability will be high. In this paper, first, using machine learning methods and in particular the gradient boosting method, a model for predicting the hit probability is presented with the appropriate accuracy. Then, using this model and by geometric calculations, the tracking parameters in the preset phase are determined in such a way that maximizes the hit probability. The efficiency of this method will be demonstrated through the simulation of different scenarios. In the end, by considering the randomness along the path, the AUV and target dynamic system is modeled as a stochastic process using the Ornstein-Olenbeck process. Then, the Monte Carlo simulation is described and similarly, previous works can be repeated.