Distributed photovoltaic power forecasting based on time series decomposition and deep learning

Abstract

This project aims to make a comparative analysis of forecasting models for improving the accuracy of short-term power output predictions in distributed photovoltaic (PV) systems. By combining time-series decomposition techniques with advanced deep learning architectures, the goal is to develop a robust model that effectively captures the complex patterns and volatility in modern solar PV power generation.

Specifically, this study obtained forecast results from real weather datasets and compared them with traditional PV power forecasting methods. The results show that the NeuralProphet model with weather inputs significantly improved forecasting accuracy, contributing to modern power forecasting techniques and the operational stability of electrical power systems.