**4. Discussion**

Based on the results, our weighted proportion (Wepro) model can be applied to generate the residential electricity load profiles at the city level by utilising the exisiting household profile generators, either LPG or ALPG, which we have employed here, given that they both have specific behavioural profile models. The seasonal share analysis based on Wepro-LPG and Wepro-ALPG, shows each season's consumption share is in the range of 23% to 26%. The 1% share consists of approximately 80 h of load or about 3 days of load when calculated on the basis of the hourly dataset. For instance, if we compare the winter and summer seasons to the whole year in Wepro-LPG as shown in Figure 9, where winter is 26.02% and summer is 23.46%, it indicates that the electricity load in winter is almost 3% higher than in summer to the whole year, which is equal to approximately 240 h or about 9 days. In addition, both seasonal profiles indicate Winter as having the highest consumption share, which concurs with the known seasonal pattern in energy demands studies [60–62]. In addition, the seasonal analyses based on meteorological is important to be mentioned as some studies did

not mention which time-division concept they used for analysing the seasonal electricity profile. Furthermore, the monthly analysis results illustrate that December is having the highest consumption share, which accords with the result of some monthly electricity studies [61,63].

The hourly average share based on the days in each season show that the weekend days indicate a higher daylight share than the weekdays' daylight share in both models. The result of the daily share of the selected typical days for all models indicates that most weekend days have a higher consumption share than weekdays in the same season. It concurs with an analysis of weekday and weekend variation, where weekend days show slightly more electricity use than weekdays [64]. Exception found in the Wepro-ALPG model's selected days in winter, where weekday consumption is higher than at weekend.

The hourly average load profiles identify the morning and evening peaks in Wepro-ALPG and Wepro-LPG, where the Wepro-LPG model has a higher load than the Wepro-ALPG model for both peaks. It is also identified that the evening peak has a significant higher load value than the morning peak load value in both models. All the hourly average loads in a year and per season demonstrate a consistent curve shape within season and model, either Wepro LPG or Wepro ALPG. The consistency is also shown within the curve shape of the hourly seasonal average load share with the hourly seasonal load share based on the days within the model.

As a consequence, the application of our model requires a profile generator as an external tool to match the weighted city's profile with the representative occupants' profiles at the household level, since we are not building our own profile generator. It also influences the results of the generated load profiles where they will be based on the characteristics of the developed model in profile generator, include relying on the few selected input profiles as a result of the approach taken in this study. The issue of relying on the few selected input profiles may result in the less fluctuations load profiles as shown in the Wepro-ALPG load profiles for the morning curves. The main difference of the hourly average in a year between the models is shown in the morning curve, where for Wepro-LPG after reaches the peak on 6am, the load share is declined gradually until 4 pm, with some light peaks in between, while for Wepro-ALPG, the curve declines slightly until 9 am after reaching a peak at 7 am. It increases again at 10 am and remains stable until 1 pm. This issue is also has been initially identified in [16] where the generated profiles show less fluctuations on the single household level, while the fluctuations at the neighbourhood level matched with the measured values. We assume that the less fluctuations during the morning period generated in Wepro-ALPG might be caused by the consistent pre-defined profiles in ALPG, where they are developed based on the simple behavioural model of an occupancy profile. The occupancy model for general events in ALPG is configured using mean times to change the state of a person. In this case, it is limited to the three person's states: active (being home), inactive (e.g., sleeping) and away (e.g., to work) [16], while in the generated load profiles of Wepro-LPG, the fluctuations are obviously shown during the morning period which might be caused by the detailed behavioural model that emphasised on the person's desire developed in LPG model. Although, it requires a future analysis. In general, the Wepro-ALPG has more aligned curve shape with the average standard Dutch residential load profile as illustrated in Figure 14 than the Wepro-LPG, where it could be because ALPG model is built based on Dutch dwelling setting. Moreover, a measured dataset that adequately represents the case study is required for validation purposes although in general our generated hourly average load profiles have similar curve's trend with the standard Dutch residential load profile provided by NEDU.

In addition, our model is found to be more efficient in respect of its computational time. In this case processing the load profiles of the city's residential sector, which consists of a large number of households is more efficient rather than generating each household in LPG or a certain number of city's households in ALPG. It takes about ten minutes computation to generate a single-person household load profile in LPG and about fifteen minutes computation to generate a multi-occupants household load profile in LPG, for instance profile: Family, 3 children, both adults at work. Thus, in takes 60-min to generate the Wepro's selected five profiles of Table 3 in LPG, where we used LPG

version 8.9.0. Furthermore, the simulation of the current configuration that consists of five households from four types of pre-defined profiles in ALPG takes about eight minutes. We use Python 3.7 (64-bit) to run this configuration. All of these simulations either LPG or ALPG were conducted on a computer using an Intel core i5-5300U CPU processor @2.3 GHz and 8 GB of installed memory (RAM). Thus, the computation will take much longer than our approach to generate a single or the few load profiles at the city level. In this case, our approach to model the residential sector at the city level has also tackle the limitation addressed by the ALPG that the tool is aimed at small group of houses which is maximum about 100 households per simulation. Consequently, our approach also creates efficiencies in the size and storage of the generated files. For instance, the output folder of one "single with work" profile generated in LPG has 2.6 GB size and the output folder of our selected pre-defined profiles in ALPG has 1.5 GB size.
