Forecasting Power Quality Indicators Using Artificial Neural Networks in the Load Connection Process in Electric Power Systems

The integration of nonlinear loads modifies current waveforms and contributes to harmonic distortion, voltage deviations, and phase unbalance in power grid environments, increasing the complexity of Power Quality (PQ) assessment during planning stages. Existing approaches typically rely on detailed system modeling and repeated simulations, which may become impractical when multiple connection scenarios must be evaluated. This work proposes a predictive framework based on a reduced set of physically interpretable descriptors, combining indices derived from the Conservative Power Theory (CPT) with the Short-Circuit Ratio (SCR) to jointly represent load behavior and power grid conditions at the point of connection. These descriptors are used as inputs to an Artificial Neural Network (ANN) trained to approximate the relationship between electrical operating conditions and PQ indicators. The training dataset is generated from simulations of an IEEE benchmark system, covering multiple load configurations and operating scenarios. The model is used to estimate quantities such as voltage Total Harmonic Distortion and CPT-based performance indices. The obtained results show that the adopted descriptor space captures the main electrical behaviors governing PQ response across the analyzed conditions. Lower approximation errors are observed in scenarios dominated by a single physical mechanism, while increased deviations occur in cases involving combined distortion and unbalance effects, reflecting the underlying coupling between load characteristics and power grid properties. Within the considered domain, the proposed approach provides a physically grounded formulation for the preliminary evaluation of disturbing load integration, enabling the estimation of PQ indicators from a compact set of electrical descriptors without requiring repeated full-system simulations.