Artificial intelligence based forecasting of load demand and energy quality aspects for distribution network

In this thesis, load demand and power quality in distribution networks were analyzed using artificial intelligence (AI) methods, and forecasting models were developed with different data types. Harmonic forecasting in power systems was performed with GRU and the Bahdanau attention mechanism using active and reactive power data, yielding reliable results under complex load conditions. In lighting systems, harmonic forecasting relied solely on voltage data, with the LSTM model improved by hyperparameter optimization and reinforced with the dropout technique. For load demand, a GRU–Bahdanau hybrid model trained on active power data provided accurate predictions without additional inputs. The main contribution of the study is the combined evaluation of both harmonic and load forecasting within a single framework, validated with real field data. This holistic approach offers innovative contributions to the literature while serving as a strong reference for utilities. The findings show that AI-based models can support early detection of power quality issues, enhance planning and investment, and provide applicable solutions for improving energy efficiency in distribution networks.