The thesis is devoted to the development and research of a method for determining the coordinates of an acoustic signal source based on the time-difference method using machine learning technologies to improve the metrological characteristics of localization systems. This paper presents a review of existing active and passive methods for determining the coordinates of physical objects, in particular, time, phase (active and passive), frequency, Doppler, amplitude, and time-difference methods. For each of the methods, the basic information, its essence, and the structural scheme of the implementation of systems based on these methods are presented. Based on the analysis of the methods, it was found that the use of active methods to determine the coordinates of an acoustic signal source is practically impossible. The optimal method for solving the problem of determining the coordinates of an acoustic signal source is the time-difference method using a neural network as part of the computing component. The paper presents a block diagram of a system for determining the coordinates of an acoustic signal source based on the time-difference method and using a neural network. To generate a data set for training and testing the neural network, it was developed software and mathematical models that allow us to reproduce the process of passing an acoustic signal from the source to the sensors and calculate the time of its registration. Using the described software and mathematical models and algorithms, we investigated the errors of the system with default parameters. According to the results of the study, for the default system parameters, the maximum value of the absolute error does not exceed 16 m. Since this value is significantly higher than similar analogues, it was decided to optimize the system for determining the coordinates of the acoustic signal source according to the criterion of minimum error for such parameters as the number of training pairs, the number and structure of hidden layers of the neural network, the shape of the sensor arrangement and their number, the neural network training algorithm, the distance from the sensors to the perimeter of the location of the acoustic signal sources, and the method for determining the base sensor relative to which time differences are calculated. The optimization process was carried out iteratively, and the parameters optimized in previous iterations were used for subsequent iterations. Based on the results of the optimization, the optimal parameters of the system for determining the coordinates of the acoustic signal source were determined, at which the minimum error was obtained. The systems with optimized parameters and default parameters were compared. Without taking into account additional errors, the maximum value of the absolute error in determining the coordinates of the acoustic signal source for the system with optimized parameters is 4 orders of magnitude less. The maximum value is 4 orders of magnitude lower. In addition to the main sources of error, additional ones were investigated. One of the sources is the ambiguity of the results of determining the coordinates of the acoustic signal source, which is present in almost all object localization systems and leads to a significant increase in the error in determining the coordinates. According to the results of the study, it was found that in systems built on the basis of the time-difference method with a neural network, there is no ambiguity in the results of coordinate determination. The dependence of the absolute error in determining the coordinates of an acoustic signal source on temperature and humidity was obtained. It was found that the greatest increase in error is caused by temperature (range from -10°C to 50°C), and the maximum value of the absolute error in determining the coordinates is 110 m and 20 m, respectively. When humidity affects the change in the acoustic signal velocity, the maximum absolute error increases to 3.5 m. Given the significant increase in error, a correction was applied to reduce the influence of ambient temperature and humidity on the results of coordinate determination. The application of the correction made it possible to reduce the maximum value of the absolute error to 0.38 m under the influence of temperature and to 0.14 m under the influence of humidity. Taking into account the previous results of optimization and correction, the total error was investigated, taking into account the error of the neural network, the time of acoustic signal registration, as well as errors caused by changes in temperature and humidity. The maximum value of the total error does not exceed 1.5 m in the X coordinate, 0.25 m in the Y coordinate, 1.5 m for distance, and 0.031° for angle. The relative error in determining the distance and angle to the acoustic signal source is 5 times less compared to existing methods and does not exceed 0.08 %.
