*1.2. Electricity Consumption Studies in The Netherlands*

As an overview, some studies in relation with the electricity consumption in the case study's country are provided. The household electricity consumption constitutes approximately twenty percent of the total energy consumption in The Netherlands [41]. Behavioural profiles of electricity consumption can be determined according to Dutch household and dwelling characteristics [16,17,42]. A study based on collected questionnaires relating to the dwellings above in winter 2008 showed that household size, dwelling type, use of dryers, washing cycles and number of showers influence electricity consumption significantly [43]. Furthermore, a model-based analysis [41] has been performed to explore the effects of smart-meter adoption, occupant behaviour and appliance efficiency on reducing electricity consumption in relation to CO2 emissions in The Netherlands. The paper looked at electricity consumption by end-users, projecting the best- and worst-case scenarios for carbon intensity annually. All cases assumed that carbon intensity would not increase in the future under current Dutch and European policies [41]. A real-life assessment of the effect of smart electrical appliances was conducted among Dutch households with a dynamic electricity tariff, an energy management system and a smart washing machine [29]. The results showed changes in laundry behaviour and thus electricity usage. The households regularly used the automation that came with smart washing machines [44]. The results of the study are interesting and could be a focus in our future work.

In relation to the residential Dutch load profiles, a recent study includes the local impact of an increasing penetration of photovoltaic (PV) panels and heat pumps (HPs) using the load measurements from three Dutch areas. It shows that the average daily load profile, without photovoltaics (PVs) and heatpumps (HPs) in all areas resembles the standard residential load profile. However, because of a shift from gas to electric stoves the time of peak load occurs earlier in the day [11]. We have also mentioned another profile generator ALPG [16,17] in our review, which is applied in our model. It is an open source load profile generator developed based on Dutch dwelling setting. The generated load profile is compared with measured data in Lochem (The Netherlands) over one year. It indicates a similar statistical trend, although some minor differences were identified, for instance the static standby power usage from the ALPG is too flat [16].

In brief, our contributions in this paper are the following: (1) We have developed a simplified method for modelling residential electricity load profiles in cities using Weighted proportion (Wepro) model that reflects local characteristics. (2) We introduce a practical and efficient approach to synthesize electricity load profiles, which does not require many input parameters or disaggregated individual end-uses input data to generate the load profiles. (3) We assess residential electricity load profiles based on time-division concepts: seasonal variation, monthly variation, typical seasonal days and hourly variation. The approach adopted here is illustrated the application in the case study with simple examples of the proportion adjustments of the city's profiles and household's profiles.

The rest of the paper is structured as follows: Section 2 describes the research design; Section 3 presents the results, which is the application of the method for the amsterdam case; Section 4 evaluates and discusses the results; and Section 5 concludes the paper and present the research implications for future work.

## **2. Materials and Methods**

The proposed method consists of four phases: data collection, data pre-processing, data-modelling and load-profile analyses. Data collection can be challenging, frustrating and time-consuming, especially when we want to acquire high- resolution time series data. In order to generate the hourly profile of residential electricity consumption in cities, it is required to provide city's main input data on population information such as on gender, age groups and labour force. Furthermore, it is essential to identify the required dataset or information such as national holidays per year, solar irradiation dataset and outdoor temperature dataset. All these data should cover the same periods of time. In this work, the proposed model is validated by the case-study city of the H2020 ClairCity project presented here, namely amsterdam (The Netherlands). ClairCity is a research project modelling air pollution and carbon emissions. The project identifies current air emissions or pollutant concentrations by technology and citizens' activities, behaviour and practices in six pilot cities or regions: amsterdam, Bristol, Aveiro, Liguaria, Ljubljana and Sosnowiec. The aim is to develop locally specific policy packages in which clean-air, low-carbon, healthy futures are quantified, modelled and analysed [45–51].

In data collection and pre-processing phases, it is important to study the latter comprehensively, as it can improve data quality and the accuracy of the result [52]. Data corruption, missing values and outliers are the commonest problems in data-processing [52,53]. In general, there are four tasks in data pre-processing: cleaning, transformation, integration and reduction [52,54,55]. Table 2 summarises the common problems of data pre-processing tasks and their solutions:

In this work, the data collected from amsterdam (The Netherlands), are in the form of a panel dataset, which is a cross-sectional data sample at specific point in time [52]. The panel dataset consists of information on age groups, the gender structure of each age group, the labour force, national holidays, solar irradiation and temperature datasets. The information on age groups, gender and the labour force are obtained indirectly [56–58] from Central Bureau Statistics (CBS, The Netherlands). In this case, we have elected to model the load profile for 2015. The population age is grouped into three groups: 0–17 years old, 18–64 years old and above 64 years old. The unemployment rate is recorded as 6.7% [56]. The labour force and age groups data are not in the form of datasets. Both of them provide information on the share of employment and unemployment, and the share of population's age groups and gender structure in the city, during the selected period. Therefore, there is no pre-processing technique is required in this case as well as for the solar irradiation dataset provided in ALPG. Data on public holidays are integrated into LPG's model as one of the independent inputs, like the temperature dataset. The temperature dataset and solar irradiation dataset are retrieved from

the Royal Netherlands Meteorological Institute (KNMI), the official Dutch national weather service. More specifically for temperature, we selected the data from the 240 Schiphol weather station, which is the nearest station to amsterdam and is in the same region of Noord-Holland. In this dataset, there is no missing values, noisy or inconsistent data. A reduction technique is applied, since the station code variable is not required in the modelling tool. Furthermore, due to the different standards between the data source and LPG's format. We transformed the dataset from .txt to .csv by reducing the first variable, station code, and normalising the temperature value. As mentioned, we have done data pre-processing tasks and documenting our specific work in relation to the data pre-processing steps in more details is in preparation.

**Table 2.** Data pre-processing: The tasks perform in data pre-processing include their common problems and solutions of these problems [52,55].

