The power load prediction in gecol using machine learning methods

In this study, employs an artificial neural network to predict energy consumption, analysing SCADA (Supervisory Control and Data Acquisition) data with parameters including power generation, temperature, humidity, and wind speed. Pre-processing of the SCADA database optimizes data for the Neural Network algorithm. A series of experiments construct the prediction model, with evaluations based on accuracy and loss reduction, including an R2 value of 0,98 for training data and 0,95 for test data in 5-fold cross-validation. The study highlights the network's value for energy prediction, particularly in load shedding scenarios, aiming to assist decision-making within the General Electricity Company of Libya. Furthermore, the study uncovers energy consumption patterns using the Apriority algorithm and Weka tool, highlighting rules with 100% confidence. It associates high consumption with power exceeding 5800 MW and wind speeds of 3,19 to 8,45 km/hr, medium consumption with power between 4000,5 and 4269,5 MW and temperatures of 16,325 to 18,055°C, and low consumption with power between 2505,5 and 3080,5 MW, temperatures of 16,325 to 21,635°C, and the period of 4:00 to 6:00 am. In energy data classification analysis, the Random Forest algorithm as the top performer with a 94,651% accuracy rate in 9.56 seconds, followed by J48 and Classification Via Regression algorithms with accuracy rates of 93,046% and 92,707%, respectively. Quicker alternatives include One R, Random Tree, Decision Tree, and Attribute Selected Classifier with training times ranging from 0,08 to 0,88 seconds, contributing to a better understanding of energy management practices and providing insights for future research.