Modeling Energy Consumption Using Machine Learning

Abstract

Electrical, metal, plastic, and food manufacturing are among the major energy-consuming industries in the U.S. Since 1981, the U.S. Department of Energy Industrial Assessments Centers (IACs) have conducted audits to track and analyze energy data across several industries and provided recommendations for improving energy efficiency. In this article, we used statistical and machine learning techniques to draw insights from this IAC dataset with over 15,000 samples collected from 1981 to 2013. We developed predictive models for energy consumption using machine learning techniques such as Multiple Linear Regression, Random Forest Regressor, Decision Tree Regressor, and Extreme Gradient Boost Regressor. We also developed classifier models using Support Vector Machines, Random Forest, K-Nearest Neighbor (KNN), and deep learning. Results using this data set indicate that Random Forest Regressor is the best prediction technique with an R2 of 0.869, and the Random Forest classifier is the best technique with precision, recall, F1 score, and accuracy of 0.818, 0.884, 0.844, and 0.883, respectively. Deep learning also performed competitively with an accuracy of about 0.88 in training and testing after 10 epochs. The machine learning models could be useful in benchmarking the energy consumption of factories and identifying opportunities to improve energy efficiency.

Еlectricity Consumption Prediction Model for Improving Energy Efficiency Based on Artificial Neural Networks

“…With regard to the different types of clustering and the results obtained, model B is far better with clustering type III, and the value of AR2 for the non-household consumer cluster is better than the value presented by Sarswatula et al (2022), even though they observed industrial consumers only. Even though the average MAPE(%) given in Figure 4 indicates the better characteristics of model B, the power of this model becomes much clearer when a single, random consumer is observed (Figure 5).…”

“…On the other hand, most research is based on homogenous datasets, observing only one type of consumers in a residential building (Cai et al, 2019), public building (Zekić-Sušac et al, 2018), commercial building (Shapi et al, 2021) or a shopping mall (Kim et al, 2019). Moreover, there are studies in which the authors dealt exclusively with one type of consumers, for example household consumers (Le et al, 2019;Tso & Yau, 2007), industrial consumers (Sarswatula et al, 2022) or even with one type of electrical devices (Kumar et al, 2021).…”

Comparative Analysis of Machine Learning Models for Predictive Maintenance of Ball Bearing Systems

“…The algorithm then combines the predictions of the individual trees to make a final prediction. Once the XGBoost model is trained, it can be used to predict equipment failure based on new data [42]. Before the actual training starts, the data are split into training and test sets.…”

An Analysis of AI Models and Economic Applications

“…Even things like energy consumption were very accurate from certain neural networks as shown by Sarswatula and Pugh's study, as the researchers found that given weather conditions, historical data, and economic indicators, the amount of energy a certain household could be modeled and predicted. [7] Decision Trees: Decision trees are the final form of the model that I studied. Decision trees are also widely used for many different applications and can come in many different forms.…”