A long short-term memory that is used as a prediction measure for energy consumption in campus buildings.

The global energy demand according to the International Energy Agency mentions that there will be a 3% energy demand increase due to a continued economic growth which was witnessed in 2022. For instance, the Covid-19 pandemic had reduced carbon emissions significantly in 2022 since 1990 and there was a decline by 5.8%. Further, the article stated that “despite the decrease in greenhouse gases emission in 2020, the concentration of carbon dioxide reached the highest average annual concentration. The World Meteorological Organization (WMO) concluded that the global mean temperature of 2021 was 1.09 C higher compared to the era 1850 to 1900.” Energy consumption in buildings specifically contributes a vast amount in Malaysia. For example, commercial buildings and residential buildings energy consumption constituted 13%, while buildings in Malaysia made up 54% of total electricity usage in Malaysia. Globally, energy demand is on an increase due to economic growth and population growth. Additionally, a study has been out on university buildings being energy efficient and there were only a few institutions that contributed to energy consumption.  The article by Faiq, Tan and other researchers mentioned that the energy consumption could be predicted by computer modelling. This innovative idea led to a predictive model that aided to foresee  Multimedia University Malacca’s energy consumption.

The model is a data-driven model which is related to most predictive models. These models are comprehensive and complex in nature when it comes to analysing buildings through a physical model. The article however has a negative comment about physical models, “physical models are relatively complicated, yet they fail to accurately anticipate energy demand in buildings.” In data-driven models, SVR and Artificial Neural Networks (ANNs) were preferred as the complexity of a building's structure was excluded.  The structures in the predictive model were the thickness of the walls and the types of materials that were used on the buildings. The model however can be affected by other factors such as temperature which is an environmental variable. Whilst other models predicted other parameters such as solar irradiance prediction. Other environmental variables were collated from a Malaysian model, “weather data including temperature, wind velocity, humidity and air pressure were obtained from the Malaysian Meteorological Department (MET) in creating a predictive model here.” Further, there are many more external variables that can advance the accuracy of the model such as cooling and heating degree days, solar radiation, wind speed, the day of the week and year and there are many more variables.

The model that was eventually created is for institutional buildings to predict their daily day-ahead energy consumption. The model showed the following positive results, “it was noted that forecast and prediction were defined as the estimation of the future magnitude of a variable. These predictions can be benchmarked and used for system fault detection. Optimisations could be done by building retrofits to increase the efficiency of energy usage.” Although, in Malaysia the Gross Domestic Product is correlated with the amount of energy consumption which was connected to the assessment of the economic growth with a common metric. The article suggests that, “the proposed work could be helpful in implementing energy policies as accurate energy predictions may have a significant impact on capital expenditure”. 

The literature review of the study looked at the following themes: energy prediction, the energy prediction techniques through a Long Short-Term Memory (LSTM), which is a proposed prediction algorithm. The second prediction that was explored was a dropout for the proposed prediction algorithm and the important features. Further, evaluation metrics were analysed and experimental configuration pertaining to data, and experimental procedure. Moreover, the results and analysis showed the performance evaluation and the prediction of energy consumption from 1/8/2021 to 7/8/2021. 

The study concluded that the forecast of energy consumption of an institutional building can be predicted from an LSTM model, which gives a daily day-ahead forecast. The day, week, weekend or holiday affects the LSTM model. Further, to improve the accuracy of the model, environmental external variables are needed such as wind speed, solar radiation, temperature and so forth.

I think that gaining the predictive or forecasted view of energy consumption in institutional buildings is a very innovative idea. In this way, a university or any building can know their energy consumption and take action on how to be energy efficient.