Impact of nonintrusive load monitoring on CO2 emissions in Malaysia

Received Mar 17, 2021 Revised May 27, 2021 Accepted Jun 10, 2021 Nonintrusive load monitoring (NILM) based energy efficiency can conserve electricity by creating awareness with the behaviour change and shrinking CO2 emissions to the environment. However, the lack of effective models and strategies is problematic for policymakers to forecast quantitatively CO2 emissions. This paper aims to study the impact of NILM on CO2 emissions in Malaysia. Firstly, the predictive models were established based on Malaysia open data from 1996 to 2018. After that, scenario simulations were conducted to predict CO2 emissions and NILM impact on environmental degradation in 2019-2030. The results revealed that a 12% reduction in electricity consumption due to NILM could contribute to a 10.2% shrinkage of the total CO2 emissions. The result also statistically confirmed Malaysia to achieve a 45% reduction of CO2 intensity in 2030. With NILM, the carbon reduction can be further enhanced to 60.2%. The outcomes provide valuable references and supporting evidence for policymakers in planning effective carbon emission control policies and energy efficiency measures. The work can be extended by developing a decision support system and user interfaces access via the cloud.


INTRODUCTION
Sustainable development is a strategic approach to resolve the negative environmental consequences of economic growth and globalisation by finding possible solutions to remedy various problems caused by industrial and population growth [1]. As a developing nation, Malaysia gains rapid economic growth in industrialisation, urbanisation, and population growth since the 1980s. Undoubtedly, escalating demand for energy and electricity gives rise to energy demand and environmental degradation. Based on the energy commission (EC) of Malaysia [2], the total electricity consumption in 2018 marked 13,152 ktoe, or 4.33% higher than in 2017. The ever-growing demand for resources such as electricity escalating the global CO2 emissions with environmental deterioration [3]. As a result, it is challenging to attain sustainable development of the country without strategic planning.
Governments and organisations have introduced numerous global and public policies targeting greenhouse gases (GHG) emissions mitigation. The United Nations Framework Convention on Climate Change Conference (UNFCCC) in Paris, or commonly known as COP21, mandated each country's responsibility to reduce CO2 emission for heading towards a sustainable and low-carbon society [4]. To achieve the reduction target of CO2 emissions, Malaysia has implemented many policies to mitigate the environmental issue, particularly in terms of renewable energy and energy efficiencies (EE). In terms of residential and commercial buildings, the detailed information of electricity consumption using load monitoring, including the nonintrusive option (NILM) is a low-cost mechanism for analysing changes in single-source data from the metering and deducing individual electricity consumption of appliances [5]. Numerous use cases of NILM that can support EE include load demand forecasting, demand response and peak load shaving. As reported, a potential reduction of 12% of electricity consumption is achievable [6]. Ultimately, NILM opens up an opportunity to conserve electricity by creating awareness with the behaviour change and shrinking CO2 emissions to the environment [7].
To assist the policymakers in tracking future trends of CO2 emissions, the researchers carried out the forecasts from either regional or national level with different approaches [8], [9]. Many studies have revealed the relationship between several variables and energy consumption or energy-related CO2 emissions. Some authors suggested independent variables such as residential household, industrial, manufacturing, commercial and transportation in analysing the relationship with energy consumption [10] or CO2 emissions as responding variable [11], [12]. Apart from microscopic variables, the studies also conducted the relationship between the determinant variable with macroscopic variables such as gross domestic product GDP, population, urbanisation [10], and the single variable using GDP [13]. Several studies of scenario simulation method analyse the impacts of possible future CO2 emissions by considering several alternative settings of predictors [14]- [17] Previous literature provided good references to predicting CO2 emissions and the issues of electricity conservation due to NILM [18]- [21]. However, a literature gap still exists regarding the study of NILM impact on CO2 emissions, to the best of our knowledge. This paper aims to construct predictive models to evaluate the NILM impact on CO2 emissions. Firstly, two predictive models are established with Malaysia open data from 1996 to 2018. Secondly, scenario simulations were adopted to predict CO2 emissions in 2019-2030 with different scenarios to highlight the impact of NILM on CO2 emissions reduction. As the largest source of CO2 emissions, human activities with controlled energy consumption can directly affect emission mitigation [22]. Therefore, it is crucial to predict CO2 emissions quantitatively for policymakers to monitor and manage the target policies and practices. Furthermore, the result also provides empirical evidence of the carbon reduction as Malaysia pledged in COP21. Figure 1 depicts the forecasting procedure in this paper. The study starts with developing predictive models of CO2 emissions based on historical open data of the identified determinants derived from socioeconomic, demographic and technological innovations. Two regression models, namely multiple linear regression (MLR) and trend analysis (TA) models, are developed and eventually validated. Secondly, scenario simulation is conducted for CO2 emissions forecasting and investigate the impact of the reduced electricity consumption due to NILM on environmental degradation in 2019-2030. MS Excel with the plugin package of data analysis is used to conduct modelling, testing and simulations in this study.

Data collection
The study employs the historical annual time series data from World Bank Open Data's World Development Indicators (WDIs) [ Malaysia (MEIH) [24], from 1996 to 2018. The total CO2 emissions (in thousand metric tons, kt) is chosen as the response variable. Several predictors are identified from socioeconomic, demography and technological innovation categories, namely per-capita gross domestic product GDP, per-capita electricity consumption (in kWh), R&D expenditure (in % of GDP), FDI net inflows (in Billion US$), generation (RE) (in % of total) and generation (fossil) (in % of total). The dataset is tested for normality before selecting significant predictors by the stepwise regression search method for modelling.

Predictive models
The MLR model is essentially a statistical regression technique that uses a group of predictors to predict the outcome of a response variable Y. It essentially gives more explanatory power to the regression model by including additional predictor variables [14]. The model is generally written as (1) as: where k is the number of independent variables, β0 is the intercept term, β1, ..., βk are regression coefficients, and ε is the random error term of the model. The derived MLR model is statistically tested and adopted for scenario simulation to forecast CO2 emissions in 2019-2030.
On the other hand, TA model is used to predict future CO2 emissions owing to its simplicity. The projections are based on what has happened in the past explains what will happen in the future [25], [26]. For forecasting of the CO2 emissions, the trends during 1996-2018 are first identified. According to the observed trend, the TA model is expressed as a regression function of the year as (2) to predict CO2 emissions.

Model validity
The regression models are rooted strongly in the field of statistical learning. Therefore, the model's goodness-of-fit tests are carried out to inspect the validity of the derived models, namely R-squared, T-test and F-test. Mean absolute percentage error (MAPE) is calculated as the deviation between the model's predicted values and actual values in the data population. It is calculated using (3) as a dimensionless index. The lower the value of MAPE, the better its performance [27].
where At is representing the actual value, n represents the number of observations, and Pt denotes the predicted value.

Prediction of CO2 emissions in 2019-2030
In order to forecast CO2 emissions and evaluate the NILM impact on CO2 emissions, both TA and MLR models are adopted in scenario simulations of the environmental degradation (2019-2030) in Malaysia. The MLR model predicts CO2 emissions, apart from evaluating CO2 emissions under the influence of electricity conservation due to NILM. On the other hand, the TA model predicts future CO2 emissions based on the historical trends under the compound effect of different factors. It is a so-called business-as-usual (BAU) model without considering the influence of NILM on CO2 emissions forecasting. By comparing the results from scenario simulations, the future CO2 emissions and the impact of NILM-based measure on environmental degradation in Malaysia are determined. Compared to [14] for technological innovation impact, this study emphasised the evaluation of the NILM impact on environmental degradation.

Modeling results and validity
Based on the regression results of the MLR model are listed in Table 1  The F-test of overall significance with a Signif-F value of 9.69E-18 (<0.05) in Table 1 (b) has confirmed the model's goodness-of-fit. The comparison of actual and predicted CO2 emissions values between 1996 and 2018 is depicted in Figure 2. The predicted CO2 emissions are very close to the actual ones, with a low MAPE value of 0.04%.
The F-test of overall significance with a Signif-F value of 3.08E-19 in Table 2 (b) has confirmed the model's goodness-of-fit. The comparison of actual and predicted CO2 emissions values between 1996 and 2018 is depicted in Figure 3. The predicted CO2 emissions are very close to the actual ones, with a low MAPE value of 0.05%. Chang et al. [28] suggested that the model performance can be classified according to MAPE metric. The model is excellent if MAPE is less than 10%, while MAPE between 10-20% is considered acceptable. Hence, the results have confirmed the creditability and well-fitted the models to forecast the CO2 emissions in Malaysia.

Scenario simulations
This section studied the impact of electricity conservation due to NILM on environmental degradation. The scenario simulations are conducted to reveal the insights of NILM impact. The scenario simulations are conducted by setting 5 (five) scenarios to analyse CO2 emissions from 2019 to 2030. Due to the pandemic of covid-19, Malaysia marked GDP growth of -6% but forecasted to recover after that. Malaysia GDP has an annual growth rate of 4.8% in 2019-2025 as forecasted by the International Monetary Fund (IMF) report [29]. In this study, we extend the same rate from 2026 to 2030 for scenario simulations. − Electricity consumption; the historical data was extracted from the Malaysia Energy Information Hub (MEIH) of Malaysia's Energy Commission. According to World Energy Markets Observatory (WEMO) 2017 report, electricity consumption is projected to increase by 4.8% annually right up to 2030 [30]. Scenario 1 with an annual growth rate of 4.8% of electricity consumption as the baseline for forecasting between 2019 and 2030. Scenarios 2-5 set the electricity consumption with the baseline and different reduction percentages (2~12%) due to the NILM effect.

Predicted CO2 emissions for 2019-2030
There are five scenario settings in the simulation set, as listed in Table 3. Designated as the baseline scenario, Scenario 1 employs the TA model to forecast CO2 emissions in 2019-2030 based on historical data without considering NILM impact. Scenarios 2-5 highlight the NILM-based EE with different reduction percentage in electricity consumption, on top of the annual base rate of 4.8% increment. Other variables set by the forecast data 2019-2030 as described. The five scenarios are simulated to forecast Malaysia's CO2 emission values in 2019-2030, as depicted in Figure 4. Baseline (+4.8%) with a reduction of 2% of electricity consumption 3.
Baseline with a reduction of 5% 4.
Baseline with a reduction of 8%

5.
Baseline with a reduction of 12% In this case, the TA model used in Scenario 1 serves as business-as-usual (BAU) by considering the compound effect of various predictor variables without any additional electricity conservation due to NILM. On the other hand, scenarios 2-5 forecast the CO2 emissions values by considering the impact of NILM. Therefore, the significance of the NILM on CO2 emissions in Malaysia can be highlighted. Compared with BAU (Scenario 1), the reduction of electricity consumption due to NILM-based EE (Scenario 2-5) has reduced the CO2 emissions from 351,739kt to 315,862kt. Although a 10.2% shrinkage of CO2 emissions, the trend shows the peaking of CO2 emissions is still far to be achieved by 2030.

Forecasting of carbon intensity forecasting of Malaysia by 2030
Malaysia has pledged to Nationally Determined Contribution (NDC) of UNFCCC for 45% reduction of GHG emissions, i.e. per-GDP emission intensity by 2030 (relative to 2005) [31]. More specifically, the pledge consists of a 35% reduction of unconditional basis and an additional 10% reduction on a conditional basis, i.e. upon receiving support from developed countries, including climate finance, technology transfer, and capacity building. Malaysia has implemented various policies and efforts to fulfil its commitment to the reduction of environmental degradation. Based on the performance of CO2 intensity per GDP shown in Figure 5, the projection with scenario 1 has shown a reduction of 55.3% (relative to 2005) in 2030. Furthermore, the CO2 emission intensity in 2030 can be further improved by an additional 5% due to the adoption of NILM-based measure. Therefore, Malaysia should employ various policies and mechanisms, such as NILM and its use cases, to substantially reduce electricity consumption and CO2 emissions for financial saving and environmental improvement.

CONCLUSION
As committed to UNFCCC of environmental improvement in 2030, it is crucial to translate the goal into effective planning with analytical and monitoring tools. The scenario simulation result revealed that a 12% reduction in electricity consumption due to the NILM-based EE had caused a 10.2% shrinkage of the total CO2 emissions. Meanwhile, the result also affirmed the possibility of achieving Malaysia's commitment to UNFCCC to lower CO2 intensity per unit GDP by 45% by 2030. More specifically, the reduction of 57.43%-61.40% due to the impact of NILM. Nevertheless, the trend shows that Malaysia's total CO2 emissions will continue to increase until 2030 without reaching the peak. Hence, Malaysia needs good policies to promote green technology and renewable energy and encourage Malaysians to be more aware of environmental sustainability and use energy efficiently. Apart from that, the generation of renewable energy and R&D expenditure of Malaysia is still far limited to bring any significant effect to reduce CO2 emissions of the country. The outcomes provide valuable references and supporting evidence for policymakers in planning effective carbon emission control policies and action plans. The work can be extended by developing a decision support system and user interface via the cloud.