Machine Learning Based Electrical Load Prediction Using Regression Learners: A Performance Evaluation

The primary intent of the present research was to design and execute an electrical load forecasting system using machine learning techniques. Implementing an advanced predictive method, specifically a machine learning algorithm, helped in accurate load forecasting which is crucial for efficient power grid management, avoiding blackouts, and optimizing resource allocation. Electricity load fluctuates due to complex factors like weather, holidays, and human behavior, making traditional forecasting methods struggle. This is where machine learning (ML) shines. ML algorithms can learn from historical data, identifying complex patterns and relationships that influence electricity demand. This allows them to make more accurate predictions than static models. In this work, regression learning models in machine learning are used with the MATLAB platform. Three years of real-time data from the Wavi substation in India is used in the work. Considering day, date, hour of day max & min temperature of the day, and input voltage and current are taken as input parameters to test fourteen different models of assorted regression algorithms, namely, linear regression, SVM, decision tree model, and neural networks. The performance of these models is evaluated using commonly used metrics, RMSE, MSE & MAE along with a few other parameters. An optimized trained model is then tested with real data to obtain a forecasted load for one week. The correlation between the Actual load and forecasted load is found to be 0.999962 showing strong relation between actual and forecasted values.