Performance Evaluation of Machine Learning Techniques on the microgrid and weather data

Authors

  • Ronak Solanki Author
  • Devang Pandya Author
  • Ketan Sarvakar Author

Keywords:

Microgrid, Support Vector Regression (SVR), Gradient Boosting Regression (GBR), Non-linear Relationships, Ensemble Learning

Abstract

Microgrid uses distributed power generation. Renewable resources have a discontinuous pattern of power generation. Hence, it is difficult to forecast wind or solar power accurately. Forecasting methods of renewable energy sources and their consumption have been investigated deeply for the prediction of demand in the near future using conventional machine learning techniques.

We attempt to analyze the performance of a few well-known machine learning techniques for the prediction of target variables. This paper explores the overall performance of Support Vector Regression (SVR) with linear and non-linear kernels and Gradient Boosting Regression (GBR) in predictive modeling on microgrid and weather data.

SVR with a linear kernel achieves awesome accuracy (99.9%) on a dataset in which a robust linear relationship exists between input functions and the target variable. This high accuracy highlights SVR’s effectiveness for troubles characterized via clear linear trends.

In assessment, non-linear SVR kernels, which include polynomial and Radial Basis Function (RBF), carry out less correctly because of the lack of non-linear relationships among the participating variables. Gradient Boosting Regression, even while showing lower accuracy in a few cases, excels in coping with complicated and non-linear interactions, making it mainly well-acceptable for tasks related to difficult relationships, which include climate forecasting.

GBR’s ensemble learning approach, which integrates multiple vulnerable decision trees, gives stepped-up performance through decreasing bias and variance and taking complicated feature interactions. The outcomes imply that even while SVR with a linear kernel is good for easy, linear datasets, GBR is ultimate for situations with non-linear and multifaceted data relationships. Thus, model selection must align with the dataset’s characteristics and the specific necessities of the predictive project.

Downloads

Published

2025-01-01