OPTIMIZATION OF RANDOM FOREST FOR ACTIVE POWER PREDICTION BASED ON THREE-PHASE VOLTAGE AND CURRENT PARAMETERS Muhammad Dani Solihin, Erita Astrid, Muchsin Harahap, Mhd Ikhsan Rifki, M. Khalil Gibran, Amir Saleh
Universitas Negeri Medan
Abstract
The study aims to implement Random Forest Regression in predicting total active power in a three-phase power system based on voltage and current parameters for each phase. The dataset used consists of measurement data compiled using a power quality meter, totaling 2,343 data points. The target output of the study focuses on total active power values, with a data splitting ratio of 80 percent and 20 percent for testing. The model scenario was configured using a random forest, which produced evaluation metrics of Mean Absolute Error (MAE) of 3,495.70 Watts, Root Mean Squared Error (RMSE) of 38,712.37 Watts, and a coefficient of determination (R-squared) of 0.9709. The model scenario was then optimized using the GridSearchCV method, with the optimization results describing an escalation in model performance. The performance escalation was marked by an increase in the evaluation matrix results, with the MAE decreasing to 3.390.99 Watts, the RMSE decreasing to 36.676.96 Watts, and the R(2) increasing to 0.9739. The model visualization is represented in a scatter plot, error histogram, and a graph comparing actual and predicted active power, showing that the model provides a stable error distribution and has a high accuracy rate. The research results describe that the random forest model using hyperparameter optimization can be used to model non-linear patterns and characteristics of data between voltage parameters and total active power current
Keywords: Random Forest Regression, Active Power Prediction, Voltage and Current Features , Three-Phase Electrical System, Model Optimization
