Support Vector Regression for Electricity Consumption Prediction in a Building in Japan

Background: Electricity is considered as the basic essential unit in today’s high-tech world. The electricity demand has been increased very rapidly due to increased urbanization, smart buildings, and usage of smart devices to a large extent. Building a reliable and accurate electricity consumption prediction model becomes necessary with the increase in building energy. From recent studies, prediction models such as support vector regression (SVR), gradient boosting decision tree (GBDT), artificial neural network (ANN), random forest (RF), and extreme gradient boosting (XGBoost) have been compared for the prediction of electricity consumption and XGBoost is found to be most efficient thus leading to the motivation for the proposed research work. Objective: The objective of this research is to propose a model that performs future electricity consumption prediction for different time horizons: short term prediction and long term prediction using extreme gradient boosting method and reduce the prediction errors. In addition to this based on the prediction, best and worst predicted days are also recognized. Methods: The method used in this research is the extreme gradient boosting for future building electricity consumption prediction. The extreme gradient boosting method performs prediction for the short term and long term for different seasons. The model is trained on a household building in Paris. Results: The model is trained and tested on the dataset and it predicts accurately with the lowest errors compared to other machine learning techniques. The model predicts accurately with RMSE of 140.45 and MAE of 28 which is the least errors when compared to the baseline prediction models. Conclusion: A model that is robust to all the conditions should be built by enhancing the prediction mechanism such that the model should be dependent on less factors to make electricity consumption prediction.

Download Full-text

Electricity Consumption Prediction of Solid Electric Thermal Storage with a Cyber–Physical Approach

Energies ◽

10.3390/en12244744 ◽

2019 ◽

Vol 12 (24) ◽

pp. 4744 ◽

Cited By ~ 1

Author(s):

Huichao Ji ◽

Junyou Yang ◽

Haixin Wang ◽

Kun Tian ◽

Martin Onyeka Okoye ◽

...

Keyword(s):

Physical Model ◽

Prediction Error ◽

Back Propagation ◽

Electricity Consumption ◽

Thermal Storage ◽

Support Vector ◽

Physical Approach ◽

Relative Errors ◽

Behavior Characteristics ◽

Consumption Prediction

This paper proposes a cyber–physical approach to enhance the prediction accuracy of electricity consumption of solid electric thermal storage (SETS) system, which integrates a physical model and a data-based cyber model. In the cyber–physical model, the prediction error of the physical model is used as an input of the cyber model to further calibrate the prediction error. Firstly, customers’ behavior characteristics are extracted by the integration of K-means and one-versus-one support vector machine. Secondly, based on the behavior characteristics and ambient temperature, the physical model is developed to predict daily electricity consumption. Finally, the error levels of physical model are classified, together with the temperature and prediction values of the physical model, are selected as the inputs of the cyber model using the back propagation (BP) neural network to calibrate the results of the physical model. The effectiveness of the proposed cyber–physical model (CPM) is verified by a 1 MW SETS system. The simulation results show that, compared with the physical model (PM) and cyber model (CM), the maximum relative errors (MRE) with the CPM are reduced to 25.4% and 4.8%, respectively.

Download Full-text