Use of Nerual Networks for Modeling Energy Consumption in the Residential Sector

Introduction

Energy use and associated greenhouse gas (GHG) emissions, and their potential effects on the global climate change have been the worldwide concern especially after the Kyoto Protocol. Improving the energy efficiency especially in the residential sector is one of the most effective ways to reduce end-use energy consumption and associated emissions. This is particularly effective for countries with high per capita energy consumptions, such as seen in Canada and USA.

In recent years, the residential energy use and associated GHG emissions in the developed countries account for almost one fifth of the total due to the significant increases in dwelling square footages and the number of appliances used in the households. Figure 1 shows sectoral percentage distributions of the energy consumption and associated GHG emissions between in 2005 for Canada and USA. Thus, reducing the end-use energy consumption and the associated emissions from the residential sector is one of the effective means of approaching the GHG emission reductions required by the Kyoto Protocol.

Figure 1.

Sectoral energy consumption and GHG emission distributions in 2005 in Canada and USA (EIA, 2008; OEE, 2008)

To reduce the end-use energy consumption and pollutant emissions from the residential sector, a large number of options need to be considered. These include improving the energy efficiency of dwellings through improving envelope characteristics; using higher efficiency heating and cooling equipment, household appliances and lighting; switching to less carbon-intensive fuels for space and domestic hot water heating (DHW); etc. Energy efficiency improvements have complex interrelated effects on the end-use energy consumption of households and the associated pollutant emissions (Ugursal & Fung, 1996). As a result, evaluating the effect of various energy efficiency improvement options on residential end-use energy consumption and associated emissions requires detailed mathematical models.

Recently, two types of modes have been used to model residential end-use energy consumption: the Engineering Model (EM) (Farahbakhsh et al., 1998; Larsen & Nesbakken, 2004) and the Conditional Demand Analysis (CDA) Model (Parti & Parti, 1980; Aigner et al., 1984; Caves et al., 1987; Fiebig et al., 1991; Bauwens et al., 1994; Hsiao et al., 1995; Bartels & Fiebig, 2000; Lins et al., 2002; Larsen & Nesbakken, 2004; Aydinalp & Ugursal, 2008).

The EM involves developing a housing database representative of the national housing stock and estimating the end-use energy consumption of the households in the database using a building simulation program. A building simulation program models the end-use energy consumption of a building based on thermodynamic and physics principles taking into consideration such factors as envelope characteristics, internal and solar heat gains, weather conditions, and occupant behavior. Thus, this approach requires a database representative of the housing stock with detailed household description data.

One of the most comprehensive national EM was developed Farahbakhsh et al. (1998) using the survey data of 8767 Canadian households and a building simulation software. The main advantage of this model is its capability to evaluate the impact of all types of potential energy saving measures on the residential end-use energy consumption. However, it requires extensive user expertise and lengthy input data preparation time for developing reliable end-use energy estimates. Another national EM was developed by Larsen & Nesbakken (2004) based on the survey data of over 2000 Norwegian households and a building simulation software. As stated by the authors, similar to the Canadian model developed by Farahbakhsh et al. (1998), the fundamental weakness of the Norwegian EM was the need of high number of numerical inputs required by the model.