utilizang multi agent machine learning technique to forecast domstic power demand response

نوع: Type: Thesis

مقطع: Segment: masters

عنوان: Title: utilizang multi agent machine learning technique to forecast domstic power demand response

ارائه دهنده: Provider: Murtadha Abdalsalam moahmood

اساتید راهنما: Supervisors: D SHABAZI

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: D.HATAMI & D TORKAMANI

زمان و تاریخ ارائه: Time and date of presentation: 2026

مکان ارائه: Place of presentation: 33

چکیده: Abstract: Modern electricity distribution networks are moving toward smart and active structures, as demand-side management in the residential sector has given rise to technical challenges such as voltage drops and increased losses due to the high fluctuation of loads and the large number of consumers. Traditional centralized control methods are not suitable for managing millions of consumers because of computational complexity, communication delays, and privacy concerns. Hence, decentralized management frameworks based on load forecasting using machine learning and multi-agent reinforcement learning (MARL) have attracted attention for managing demand response and improving network performance. However, these methods often suffer from low accuracy and have failed to address the issue of separating the load forecasting task from the load management task. In this research, to fill the gaps of previous studies, a method for estimating the power demand of residential consumers is proposed utilizing a multi-agent machine learning approach. In the proposed method, a Long Short-Term Memory (LSTM) network is used for load forecasting, and MARL is employed for decision-making and operational control of the network. Simulation results of the proposed method show that the LSTM model performs load forecasting with good accuracy, achieving an RMSE error of 0.612. Furthermore, the results demonstrate that using the MARL decision-making model based on LSTM load forecasting yields valuable outcomes in terms of reducing network losses, improving the voltage profile, and reducing operational costs, compared to traditional RL method