Transmission losses—energy dissipated as heat in power lines and equipment during electricity delivery—represent a significant operational and economic challenge for grid operators worldwide. As renewable energy integration and variable demand patterns intensify, predicting these losses with higher accuracy has become essential for optimal dispatch, financial planning, and grid stability management.
Researchers have demonstrated that neural network models can significantly enhance transmission loss forecasting by incorporating correlation-derived input features alongside conventional operational and meteorological data. The study, validated using four years of operational data from Bosnia and Herzegovina's transmission system, introduced two key derived features: a 24-hour loss-pattern forecast that captures historical daily cycles, and a pattern-matched loss feature that identifies similar historical operating conditions.
The baseline multilayer perceptron (MLP) model achieved a mean absolute percentage error (MAPE) of 14.40% on test data. By integrating the 24-hour loss-pattern feature, prediction error dropped to 13.29%—approximately an 8% relative improvement. This enhancement works by explicitly capturing temporal behavior patterns that would otherwise require more complex neural network architectures.
The methodology's strength lies in its simplicity and computational efficiency. Rather than expanding the neural network itself—which increases computational burden and training complexity—the approach enriches input data with analytically derived features that encode domain knowledge about transmission system behavior. This strategy proves especially valuable for utilities operating under real-time dispatch requirements where computational speed matters.
For grid operators, improved loss forecasting translates directly to better resource allocation, more accurate economic dispatch calculations, and enhanced understanding of system efficiency metrics. The technique's validation on a full operational transmission system demonstrates practical applicability beyond academic settings. As renewable penetration continues climbing, utilities seeking to optimize system performance and reduce operational losses will find this approach particularly relevant for enhancing existing SCADA and EMS capabilities.
