Transmission Loss Prediction Using Neural Networks With Correlation‐Derived Input Features

Accurate prediction of transmission losses is essential for efficient operation and planning of modern power systems under increasingly dynamic operating conditions driven by renewable energy integration and weather‐dependent variability. This paper proposes a neural network–based framework for transmission loss prediction that improves predictive performance through correlation‐derived input features. The methodology integrates operational power system measurements, meteorological variables, and analytically derived input features designed to capture temporal patterns and nonlinear dependencies in system behavior. Two derived input features are introduced: a 24‐h loss‐pattern forecast and a pattern‐matched loss feature, to capture historical pattern similarity directly in the model inputs. The framework is validated using 4 years of hourly operational and meteorological data from the transmission system of Bosnia and Herzegovina. The baseline MLP achieves a mean absolute percentage error (MAPE) of 14.40% on the test dataset. Incorporating the 24‐h loss‐pattern forecast feature reduces the prediction error to 13.29%, while the pattern‐matched loss feature provides a modest improvement relative to the baseline configuration. The results demonstrate that correlation‐derived inputs enhance prediction accuracy while maintaining model simplicity and computational efficiency.