An Optimal Neural Network for Hourly and Daily Energy Consumption Prediction in Buildings

Introduction

The energy prediction models can bring a considerable amount of improvement in overall energy savings and reduction in environmental impacts. In 2013, almost 57% of total energy was consumed by the residential and tertiary sector in the Metropolitan City of Turin (Pavone, 2014). A higher accurate energy consumption prediction is a very complicated and difficult to formulate the task but indeed very important and therefore is the focus of many researchers. There are many parameters that have a direct influence on energy consumption of residential buildings. The important of these are the structure of the building, the external weather conditions, the behaviour of residents, and the usage of home appliances and so on. Over the last many years, different mathematical models for energy consumption prediction have been proposed in the literature. In order to find the future demands of energy, the energy consumption prediction at various scales is very important. The accurate energy consumption prediction also plays a key role in efficient production, distribution, selling, operation, planning and management in the energy management system. Power generation systems can be made more reliable if with the accurate energy consumption prediction. Further, if efficient energy consumption is carried out, the economy, fuel utilization, and other sectors dependent on energy management systems can be managed in a better way.

The energy consumption prediction can be divided into, short term, medium term, and long-term categories based on the time in which the energy has been utilized. Different techniques have been proposed in the literature for the energy consumption prediction with various evaluation criteria, parameters and the varying degree of accuracies (Wahid and Kim, 2016). Likewise, the application of different types of AI techniques for energy consumption and other types of predictions was highlighted by Wahid et al. (2017), Shah and Ghazali (2011) and Shah et al. (2012).

A linear regression model was developed by Catlina et al. (2008) for the monthly energy consumption prediction of residential buildings in some areas of France. The authors have considered different parameters for an accurate prediction. The parameters considered were the compactness of buildings, the ratio of opaque and glass surfaces, the thermal diffusion of roof and walls and time constant of buildings. A maximum deviation of 5.1% between the actual and calculated consumption was observed whereas the average deviation observed in the simulation was 2%. The authors have concluded that the energy consumption can be optimized by considering the design of the building in terms of the strong relationship between the building compactness and the heat consumption. Some other parameters e.g. thermal inertia has also noticeable impact. Although, the authors have obtained a reasonable accuracy but still there is much room for improvement. The less accuracy is due to the fact that the authors have not considered the internal patterns of the consumption.

Different types of features have been explored by Zhou and Zhu (2013) that have a relationship with the energy consumption in residential buildings. The study was focused on residential buildings located in some parts of China and different seasons e.g. summer, winter and environmental conditions were considered in the study. The study was conducted for the annual energy consumption regression model was developed for simulation. The thermal diffusion of walls and the outdoor temperature were considered as independent variables. A maximum deviation of 15% was observed between the actual and calculated energy consumption values. Here, again the authors have focused more on the external parameters and have not considered the internal parameters representing valuable pattern in the energy consumption and hence the accuracy the found can further be improved. To emphasize on the importance of renewable energy systems and awareness of dangers by carbon dioxide, Wang and Liu (2017) proposed an energy optimization model based on particles swarm optimization (PSO) for efficient energy management in wireless sensor networks.