**1. Introduction**

The residential energy sector plays a crucial role in achieving greater energy efficiency and emissions reduction goals. Studies have suggested that residential energy use is of great importance in ensuring global energy sustainability, given its energy-saving potential [1,2]. The International Energy Agency (IEA) has calculated that the residential sector contributes about 25% of energy consumption and 17% of carbon dioxide (CO2) emissions globally. It is therefore, essential to understand the residential energy consumption patterns locally to allow for an assessment of the energy-saving potential in the sector. However, lack of accessibility to measured high-resolution electricity consumption data at the city level such as smart-meter data and time use survey (TUS) data makes it difficult to understand the characteristics of electricity consumption locally. Research into this aspect will improve our understanding of residential electricity load profiles, which can be used to achieve improvements in energy efficiency as the residential sector has a major potential for energy savings [3]; to reduce CO2 emissions as extensive studies have identified that household behaviour has a significant impact on consumption and emission [3,4]; and to optimise energy management [5] as these types of studies have supported transmission grid planning for better energy management [5,6]. This suggests that energy policy should vary depending on local characteristics. Trends towards small scale renewable electricity generation and introduction of heat pumps and electric cars are changing the local energy system. Furthermore, policies towards developing Positive Energy Districts (PEDs) support the relevance of studying electricity load profiles at the district level [7]. Therefore, a computational method is required to handle a large number of population datasets and handle the granularity of the data. To scope the focus on end-user consumption, it is important to measure residential electricity consumption per unit accurately with respect to time, or so-called 'temporal resolution'. 'Temporal resolution' refers to the granularity of the data-sampling rate, which may be more or less equal to the acquisition rate by meter [8]. The key in temporal-resolution load-profile models of residential electricity consumption is to emphasise identification of the resolution that represents the essential local characteristics and consumption behaviour [8,9]. The importance of temporal resolution load profiles is that they ensure the accuracy of calculations of self-consumption and are able to optimise short-term fluctuations of electricity supply and demand [10]. The temporal-resolution load-profile method is the focus of our work.

We propose a simplified approach which uses a weighted proportion (Wepro) model to synthesise the residential electricity load profile at the city level, by utilising existing household load profile generators such as load profile generator (LPG) and artificial load profile generator (ALPG). The model requires some limited input parameters at the city level: the citizens's age groups (AG), gender (GD) structure, and labour force (LF) composition. This weighted method is widely used across many sectors to proportionately reweight values especially in relation with population statistics. This model can be applied for synthesising a residential electricity load profile by proportionally matching the city's main characteristics with the representative household profiles provided in the load profile generators. This simplified method can tackle the drawbacks of the existing studies that require many input parameters or disaggregated individual end-use smart meter data to generate the load profiles and the drawbacks of the standard load model used by the utilities. It is also mentioned in [11] that distribution system operators (DSOs) use rough estimations with respect to the worst-case situations for modelling the residential load models which are important in their network planning processes and in defining a standard daily load profile. Although in practice, it is challenging to validate our results with measured data, since the measured data at the city level are mostly unavailable.

#### *1.1. Load Profile Modelling Methods*

There are different methods for modelling load profiles with top-down [12,13] or bottom-up [3,14–22] approaches. As mentioned, extensive studies have shown that the data availability is the main drawback of the approaches as they both require many input parameters or detailed aggregated input data of homogeneous activities. Our work applies a different approach where it presents a combination of a top-down approach with a few input parameters, which use general statistics information of a city and a bottom-up load model with high temporal resolution data. It simply utilises the existing household load profile generators that have covered the detailed disaggregated input data in relation with behaviour, occupancy, time-use appliances and other related variables. The fixed input parameters of the city will be matched and adjusted with the representative household profiles proportionally.

Many load profile studies [3,14–17,19–21] have applied the occupancy model, behavioural aspect and time-use of electrical appliance in their methods, where certain studies [14,15] emphasize more the psychology model of individual behaviour, which makes the pre-defined household profile more detailed and provides vary profiles. Some models are simulated based on stochastic models [18–20]. Besides focusing on the household load profiles, some studies aim at generating the load profiles at the city level or a higher level than household level [12,20,21]. In this context, the load profiles researches can be expanded from the temporal analysis to the spatial analysis such as performed in these studies [12,21], which could be one of our future interests. In addition, another approach of modelling residential electricity demand is to use a microsimulation method. In this case, the shifting from aggregate distributions to decision making units at the individual level is the main core of microsimulation

modelling (MSM) [23]. MSM is characterised by a large-scale simulation, spatial behaviour in relation to energy consumption and interaction is the main feature of spatial environment. In consequence, the dynamic migration of the population will be simplified by the model [3]. While in our work, we model the population's variables: age group, labour force and gender structure. However, spatial interaction is not the main concern of our work.

The load profiles outputs are presented as high-resolution data. Existing energy studies were generating 60-min output data [6,21,22,24–29] and one-minute resolution data [14,16–20,30]. Some works [14,15] have provided a more detailed output in one-minute resolution at once generated 60-min report data. In our work, hourly temporal resolution data are provided to compare residential electricity consumption profiles based on seasonal variation, monthly variation and days variation. Seasonal variation in this case refers to the cycles of the season: Winter, Spring, Summer and Autumn. While the typical seasonal days are the selected days to be modelled in each season both weekdays and weekend. For example, we will select to model the one weekday and one weekend in Winter, Spring, Summer and Autumn seasons.

In generating the synthetic household load profile, extensive studies have proposed and demonstrated the models, and some of them [14–18,20,21] have also developed a simulator or generator. In this work, we focused on two household profile generators that have developed based on the closest dwelling profile to our case study: amsterdam (The Netherlands). The main reasons we selected to use LPG and ALPG in our model, because both of them are developed based on behavioural model, and having one detailed model as LPG and one simpler model as ALPG may represent the different variation.

Moreover, validating the accuracy of the generated load profiles is a challenging work due to the limited available measured data to compare with. ALPG compared it's synthesised load profiles data with measurement data over a year from transformers and households of 81 connected households located in Lochem (The Netherlands) [16,17]. Twenty two measured dwellings in United Kingdom were also used to validate a study of domestic electricity use [18]. LPG validated the generated load profiles data on different criteria: Plausibility check, yearly energy consumption and duration load curve value in comparison with smart meter data rollout in Germany by Institut für Zukunfts Energie Systeme (IZES). Some studies [19,20,30] used TUS data or other independent datasets as a measurement to validate the synthesised data. Most of the studies [14,16–18] presented matched results between the generated load profiles and the measured data.

Unfortunately, as our work is focused on the city level, it is more challenging to validate the synthesised data with the measured data because the measured data should be a comprehensive dataset that represents the city's data. Finding the available measured data of the case study is challenging, mostly due to the privacy issues, cost and the measured data should represent a city's residential sector by the households' amount in the city and to make sure that the residential dwellings are located inside the selected city. It easier to find the measured data of some households or residential data at the neighbourhood level as used in the validation of the mentioned studies [14–16,18], or if the TUS data at the city level has existed. As an overview, there are three available measured electricity consumption data at the national level or obtained from various locations in The Netherlands. A measured smart-meter data of 80-households in The Netherlands is available with hourly resolution at https: //www.liander.nl/partners/datadiensten/open-data/data. In fact, these data are not considerable enough to represent a real measured data for the amsterdam residential load profile. These 80 households' locations are also undefined and require a pre-processing task since missing values exist in the dataset. Moreover, the year we modelled is 2015 and in 2015, a large section of amsterdam still used traditional meters, therefore hourly data was not available. Besides the strict privacy laws in The Netherlands, time and cost are the main considerations in obtaining smart-meter data if they are not open data. The requires time and resources to approach every customer or household, which make the cost to obtain the city's measured data relatively high. A national time-series electricity consumption data is also available at Open Power System Data [31]. The source of the data is provided by ENTSO-E

Transparency platform [32]. The European Network of Transmission System Operators (ENTSO-E) represents most of the electricity Transmission System Operators (TSOs) across Europe. In fact, the data consists of all sectors: residential, industrial and others which is also required to be synthesised if we want to take the residential part of this national load profile. In fact, amsterdam might have a different residential profile load profile than the national's residential profile. The third dataset is the residential electricity load profiles dataset provided by NEDU [33], which will be presented in the Section 4. Therefore, a future study would be followed to improve current work when there is more data available. In addition, Table 1 provides an initial overview of the important categories in the load profiles studies based on the discussion in the related works.

**Table 1.** Overview of the detail load profile modelling methods based on the discussion provided in the related works'.


Furthermore, some case studies have employed data-mining techniques to identify residential electricity load profiles [4,8,34,35]. Recent studies have proposed data-mining-based methods such as K-means [4,29,34], hierarchical [29,35] and fuzzy algorithms for purposes of electricity load profile modelling [29]. A clustering-based framework to analyse household electricity consumption patterns using a k-means algorithm has been proposed for a study conducted in China. The clustering method was selected since the electricity consumption patterns in the data were relatively smooth. A k-means algorithm was applied because it works considerably faster than other cluster algorithms, and it was easier to interpret the clustering results. The analysis was conducted in three consecutive stages: holidays, seasonal and shifting phenomena [34]. Similarly, our study also clusters the load-profile analysis into three stages: seasonal, monthly and typical seasonal days. Another case study in China employed hierarchical clustering, which is widely recognised in the context of pattern recognition, because it is easy to operate, efficient and practical [35]. A quantitative analysis approach based on association rule mining (ARM) was proposed in [4] in order to identify the impacts of household characteristics (HCs) on residential electricity consumption patterns. In any case it is assumed that the load profile data on weekdays are somehow more typical and significant than those on weekend days, while our work has covered both the weekday load profiles and weekend day load profiles through selected typical days [4].

A range of statistical analysis methods have also been applied in order to model residential electricity load profiles [6,28,37–39], including determination of the key drivers of residential peak electricity demand. Some studies provided panel datasets including data from smart-meters [24,26,40]. A model was developed using Australian data for the greater Sydney region to analyse and model residential peak demand by providing both daily and seasonal patterns [37]. The analysis was in line with the results of multiple studies showing that peak residential electricity consumption was significantly influenced by the climate and the demand for cooling. In another study, hourly residential electricity consumption was used to estimate the Monte Carlo stochastic building-stock energy model of the dwellings in the sample and the climate data sources [28]. An error analysis was performed using normalised root mean square error (NRMSE), normalised mean absolute error (NMAE), maximum absolute difference (MAXAD) and maximum relative difference (MAXRD). The results from the modelling were validated using the hourly energy equations and electricity consumption data and the uncertainty of the Monte Carlo model was calculated using multiple runs as a sample. When combined with knowledge of user behaviour, this bottom-up building-stock approach, which uses energy performance certificate (EPC) databases, can be used to estimate aggregate mean hourly electricity consumption. In this case, calibration was required to develop urban energy models. This also indicated that the outdoor air temperature had a significant influence on the model [28].
