Lightning localization based on lightning-induced voltages on power network overhead lines using machine learning method

نوع: Type: Thesis

مقطع: Segment: PHD

عنوان: Title: Lightning localization based on lightning-induced voltages on power network overhead lines using machine learning method

ارائه دهنده: Provider: Mostafa Asadi

اساتید راهنما: Supervisors: Prof. M. H. Moradi

اساتید مشاور: Advisory Professors: Dr. Hamid Reza Karami , Prof. Farhad Rachidi, Prof. Marcos Rubinstein.

اساتید ممتحن یا داور: Examining professors or referees: Dr. Razini , prof. Vakilian , Dr. Naghizadeh

زمان و تاریخ ارائه: Time and date of presentation: 2025

مکان ارائه: Place of presentation: 33

چکیده: Abstract: Nowadays, lightning localization holds significant importance in various commercial and industrial domains, including power grids, transportation, and more. Consequently, lightning localization systems have been installed in numerous locations worldwide. These systems operate by directly detecting the electromagnetic waves emitted by lightning strikes through their sensors. Common localization methods employed in such systems include Magnetic Direction Finding (MDF), Time of Arrival/Time Difference of Arrival (ToA/TDoA), Interferometry (ITF), or a combination thereof. These methods typically require a minimum of four sensors for accurate lightning localization. However, due to the use of optimization techniques, a larger number of sensors are often utilized. This increases both the cost and structural complexity of lightning localization systems. Key challenges in conventional systems include the need for precise synchronization of all sensors to calculate the exact strike location and their sensitivity to noise. In addition to commercially established methods, various alternative approaches have been proposed in research literature, though they have not yet been adopted for commercial use. These methods primarily focus on enhancing the accuracy and performance of traditional techniques by integrating them with novel approaches, such as machine learning. However, most of these methods still face challenges related to the requirement for multiple sensors and their synchronization. In this thesis, for the first time, a single-sensor lightning localization method based on the lightning-induced voltage is proposed. This machine learning-based approach operates by processing the time-series waveform of lightning-induced voltage on overhead power lines. Key advantages of this method include its simplicity of implementation (due to the reduced number of required sensors), utilization of existing power grid infrastructure, elimination of sensor synchronization needs, as well as its high speed and accuracy. Among the key objectives of this thesis is the implementation of a fast-learning algorithm with appropriate complexity and high efficiency. To achieve this, diverse machine learning methods have been systematically evaluated to comprehensively analyze conventional and standard approaches, ultimately deriving a model that fulfills the intended objectives. The training datasets were generated through simulations of lightning-induced voltage at voltage sensor locations. The simulations employed multiple methodologies, including the simplified Rusck formula and the rigorous Agrawal field-to-line coupling method within the LIOV module. By incorporating various uncertainties—such as lightning location, peak of lightning current, return stroke velocity, and ground conductivity—alongside the implementation of power grid overhead line parameters (e.g., line circuit model parameters, towers, insulators, and line branches), the training data closely approximates real-world exprimental data. Furthermore, the simulated signals were aligned with real-world voltage sensor measurements through time-origin shifting operations (Cut-off), ensuring their practical relevance.

فایل: ّFile: Download فایل