Abstract
In this paper, an approach is presented for the demand-side management of residential loads in the urban areas of Pakistan using a battery storage system at the feeder level. The proposed storage system will be installed by a private distributor to supply affordable electricity during peak hours. The experimental data used to carry out this research work are the Pakistan Residential Energy Consumption (PRECON) data set. The households of the data set are categorized based on electric power usage through K-means clustering. The clusters are expanded for feeder synthesis to represent small-scale, medium-scale, and large-scale consumption. This expansion is performed through uniform distribution in a Monte Carlo simulation. The techno-economic analysis for the installation of a battery storage system is carried out for each feeder using SAM. The results of the research project elucidated that the load factors of the feeders representing small-scale, medium-scale, and large-scale consumption improved by 1%, 6%, and 7% by using the optimally sized batteries of 50 kW (670 kWh), 90 kW (1207 kWh), and 100 kW (1360 kWh), respectively. The distributor profit and the consumer utility bill savings ranged from US$12 k to US$25 k. The results proved the validity of the used approach to simultaneously reduce the consumer bill, maximize the distributor profit, and improve the feeder load factor. The novelty of this work lies in the location and in the way the system modeling has been performed with limited data.
1. Introduction
The management of residential loads at the utility, substation, and consumer levels has become an important issue with an increase in the number of residential users throughout the world during the last four decades [1]. The residential energy consumption of Pakistan has faced around a 60% increase during the decade 2009 to 2019, from 3.42 × 1010 kWh to 5.56 × 1010 kWh [2]. The number of residential consumers is the highest of all the electric power consumers in Pakistan. As of 2019, the domestic consumption of Pakistan is 48% of the total electricity consumption in the country [3]. These statistics elucidate that Pakistan has a history of continuous growth in the residential load, even though it is facing a supply-demand gap. The supply-demand gap faced by Pakistan amounted to 7 GW in 2019 [4]. This situation calls for a solution to deal with the demand-supply gap and less efficient energy usage in Pakistan and similar developing countries.
Demand-side management (DSM) by load shifting and peak shaving are two of the most promising solutions to the aforementioned problems [5–7]. Pimm et al. used the residential load profiles in a Monte Carlo simulation to determine the potential of peak shaving using battery storage on a low-voltage distribution network [8], to see if the network under study has the potential for peak shaving or if it requires the implementation of other DSM methods. Wang et al. succeeded in reducing the peak-to-valley ratio of the energy management system in a high-rise residential building by investigating its peak shaving and valley-filling potential through a multiagent system [9]. DSM is achievable under the control of utility-only, consumer-only, or by the interaction of both. A cosimulation framework for the utility-directed control of distributed energy storage in distribution networks through a heuristic algorithm is developed in [10]. Peak shaving with simultaneous valley filling can be performed for the residential load at the household level, community level, and feeder level. A power consumption schedule for smart homes having energy storage devices is presented in [11]. An optimal electricity and heat storage scheduling algorithm is presented in [12] to manage both household energy storage and community energy storage.
DSM can also be performed using renewable energy resources in combination with storage and intelligent systems [13–16]. An artificial intelligence-based DSM strategy is proposed in [17] for smart homes, which makes the use of K-means clustering to determine user comfort levels at reasonable energy prices. As in developing countries like Pakistan, the concept of smart homes is not common [18] and photovoltaic panels are not easily affordable [19], so only the storage bank is recommended in this research work for regular homes to provide the maximum benefit at a lower initial cost.
In a review of peak-shaving strategies, it is claimed by Moslem Uddin et al. that optimum scheduling and sizing of a battery energy storage system (BESS) are still a challenge while implementing the DSM through peak shaving [20]. After this claim, an optimal-sized BESS was designed by Ding et al. [21] for the industrial load and by Olivieri and McConky for the residential load with a major focus on emission reduction [22]. Chen et al. compared the optimal performance of four different energy storage batteries to generate economic revenue and provide auxiliary services to industrial users [23]. These research projects elucidate that the optimum scheduling and sizing of the BESS depend on the constraints and conditions for which the storage system is to be designed. This fact is taken care of in this research work by clearly defining constraints and priorities while sizing the battery bank.
In this research work, a unique approach of peak shaving is used for DSM of the residential sector representing the consumption patterns of the urban areas of Pakistan. The idea to follow is that utility is responsible for the supply of electric power during off-peak hours to charge the battery bank of the private distributor at off-peak time electricity charges. The distributor is supposed to dispatch electric power during peak hours to residential consumers. Electricity sold by the distributor is charged to the residential user at a rate less than the peak rate of electricity but greater than off-peak electricity charges. This approach will result in a win-win situation for the three parties included in the system topology, and this will be verified in this research work. The idea of a distributor’s battery bank was taken from the merchant owner storage concept explained by Opathella et al. [24]. The GlidePath independent storage developer in Texas [25] and Swell Energy in India [26] are practical demonstrations of the merchant-owned energy storage concept. A merchant-owned storage investment was also suggested by Siddiqui et al. for a restricted electrical industry [27].
The data set for this research work is Pakistan Residential Energy Consumption (PRECON) [28]. These data are categorized based on electric power usage using K-means clustering [29] to estimate the electricity consumption by residential users on a small, medium, and large scale. The clusters are expanded by using a stochastic approach for feeder synthesis [30] through uniform distribution in a Monte Carlo Simulation. The feeders are subjected to peak shaving for DSM with the help of a distributor’s battery bank. The techno-economic analysis for the optimal BESS is performed in the system advisor model (SAM). An idea resembling this one was presented in [31] for the store-on grid scheme to facilitate the implementation of photovoltaic panels with battery backup in developing countries, for which prosumers had to pay additional charges to the grid to utilize the stored energy. Consumers do not have to pay a penny in the DSM scheme presented in this work.
The net present value (NPV) rule is used to ensure that the proposed strategy is a go for the DSM of the residential load at the feeder level. The comparison of the performance of battery types is not conducted in this research work. This can be a future work for researchers interested in the domain of DSM to deal with residential load problems. The results of this research work elucidate a considerable improvement in the load factors (LFs) of the synthesized feeders. The percentage improvement of the LF in this research work is much higher than that obtained in a recent previous work by Sonet et al. [32], in which they considered DSM by shifting only flexible loads of the residential consumers in urban areas of Bangladesh.
This research work is novel in certain aspects. The novelty of this research work lies in the location for which it is performed, i.e., urban areas of Pakistan. No case study for DSM has been conducted before for this location to the best of our knowledge. Also, the PRECON data set is the first of its type and the most recent original per-minute load data that were gathered using smart meters specifically installed for this purpose in 42 houses in Lahore, Pakistan [28], and it has not been used before to perform any kind of techno-economic analysis. The modeling of feeders is performed by using a very limited amount of data. The method devised for the data extension for feeder synthesis by uniform distribution in a Monte Carlo simulation is also unique. In developing countries like Pakistan, it is more common to have a BESS at the household level, but this research work suggests a centralized systematic approach of installing the BESS at the feeder level to benefit the utility, consumer, and private investor. In developing countries, the household-level BESS installation process is decentralized [33] and not specifically meant for improving grid LF. Moreover, the BESS size is just estimated. This research work aims at finding the optimal-sized BESS for managing the residential load, improving grid LFs, profiting the distributor, and reducing consumer bills simultaneously.
The remainder of this research paper is laid out as follows: Section 2 describes the methodology followed to carry out this research work. Section 3 describes data analysis, preparation, and categorization, which includes the analysis of the PRECON data set to make sure that it holds enough multiplicity to be generalized for the representation of urban areas of Pakistan. Section 4 consists of the details of developing the feeders to represent the communities with small, mid-range, and large values of electric power consumption. Section 5 includes the details of BESS selection. Section 6 includes the results obtained from the proposed strategy. The analyses and discussion of the results are also included in Section 6, followed by the last section that concludes the research work.
2. Methodology
The test network/system topology considered in this research includes the grid, distributor’s battery bank, distribution feeders, and residential loads. For residential loads in Pakistan, the voltage rating of a distribution transformer under normal conditions is usually 11 kV/430 V between phases, according to the recommended planning criteria voltages presented in the Distribution Planning Code (DPC 3.4) by the National Electric Power Regulatory Authority (NEPRA) [34]. An illustration of the idea followed in this research work is given in Figure 1. According to Figure 1, the distributor will buy (store) cheap electricity from the grid during off-peak hours and sell it to residential consumers during peak hours. The peak hours in Pakistan are 5 p.m. to 9 p.m. from December to February, 6 p.m. to 10 p.m. from March to May, 7 p.m. to 11 p.m. from June to August, and 6 p.m. to 10 p.m. from September to November [35].
The methodology followed to carry out this research work is illustrated in Figure 2. The initial step was to obtain the PRECON data set in its original form [28]. The second step was data preparation by dealing with the missing entries of data through imputation. The third step was to perform K-means clustering of the data to categorize them into three categories. The fourth step was to synthesize three feeders by using the load data of the households included in the three clusters as the initial sample. These three feeders represented three communities. The fifth step was to determine the LF of the three feeders before the virtual implementation of the peak-shaving strategy. The sixth step was to perform the techno-economic analysis of the BESS meant for the peak shaving of the residential load at the feeder level. The seventh step was to determine the LF of the three feeders after the implementation of the peak-shaving strategy. The last step was to compare the technical and economic aspects of the results for the feeders representing small-scale, medium-scale, and large-scale loads, concluding the research.
3. Data Analysis, Preparation, and Categorization
3.1. Data Analysis
The PRECON data set was recorded for a year from June 1, 2018, to May 31, 2019. The owners of the PRECON data set claim it to be valid also for developing countries other than Pakistan where the power market is not yet completely flourished [28]. It is evident from the research carried out for DSM in developing countries like India, Brazil, Nigeria, and South Africa that their peak demand time is in the evening time like that of Pakistan [36–39]. Based on their claim, the PRECON data set was analyzed based on the metadata provided by its developers to see the diversity in the demographics of its households. The metadata of the PRECON data set provides information on kW usage of electricity, property areas, number of floors, ceiling insulation, ceiling height, number of adults and children, number of various electronic appliances, number of rooms, and similar information for each household included in the data set [28]. Two of these parameters are most important to estimate the diversity of data: kW usage and the property area of each household. These parameters are elucidated in Figures 3 and 4, respectively. The peak load of the houses included in the data set ranges from 2 kW to 15 kW, and the average load ranges from 0.5 kW to 3 kW, as elucidated in Figure 3.
The property area covered by this data set ranges from 63.2 m2 to 3035.1 m2, as elucidated in Figure 4. Among the 42 households of the PRECON data set, 19% have an area of 252.9 m2, 14% have an area of 505.8 m2, 10% have an area of 227.6 m2, 7% have an area of 303.5 m2, 5% have an area of 1011.7 m2, 5% have an area of 126.5 m2, and 5% have an area of 101.2 m2. Each of the remaining property areas is covered by 2% of the households. These ranges of the peak load, average load, and property area are suitable enough to represent the consumption patterns of different social classes residing in an urban area of a developing country.
3.2. Data Preparation
Some entries in the PRECON data set were missing [28]. This might have occurred due to a fault in the installed smart meters or the absence of residents. The data missing for the short period were kept as it is, but those missing for longer periods were completed by using basic imputation techniques. To complete the missing data, the peak month of each household was noted. The peak month load data were scaled to replace the missing data for each household. After completion, the data were converted to per-hour data by averaging the values on a 60-minute basis to reduce complexity and computation time. This was performed in MATLAB. In the end, the data for the 42 households, each with 8760 entries representing hours of the year, were prepared.
3.3. Data Categorization
The prepared data were categorized based on three variables which were the base load, peak load, and average load. If this categorization were based on a single variable, this would have been performed by mere visual inspection. Because of three variables, categorization was performed through K-means clustering in Excel. The steps included in the methodology for K-means clustering were as follows:(1)The data including the base loads, peak loads, and average loads of the 42 households of the PRECON data set were arranged in tabular form.(2)The household numbers (21, 42, and 15) were chosen stochastically to represent the central points of clusters 1, 2, and 3, respectively. Three clusters were chosen because the aim was to categorize the households into three categories representing three scales of the load.(3)The multidimensional Euclidean distances between the three initial clusters and other houses of the data were determined.(4)Each household of the data was placed in the cluster from which its distance was minimum. This resulted in three clusters each having different houses included.(5)The mean values of the three variables of all the houses included in the given cluster were determined separately to represent the values of the three variables of the new clusters.(6)The multidimensional Euclidean distances between the three new clusters and other houses of the data were determined.(7)The next iteration was performed by repeating all the steps from 4–6.(8)The iterations were continued till the values of the variables of the three clusters did not change in the next iteration.
A total of eleven iterations were performed to develop the final form of the three clusters on a kW usage basis. The final result of K-means clustering was the formation of three cluster centers, having defined values of their base load, peak load, and average load, and the number of houses in each cluster. This is given in Table 1. Cluster 2 is representing the small-scale load, cluster 1 is representing the medium-scale load, and cluster 3 is representing the large-scale load. These scales are decided based on the load variables of each cluster, as there is no standard value yet defined to scale the residential load into these categories to the best of our knowledge.
4. Feeder Synthesis
The three clusters representing three categories of households were expanded for the sake of feeder synthesis. The distribution that the expanded data follow was uniform. The data expansion was performed in a Monte Carlo simulation. The data were expanded such that a single feeder represented a community having a total of 500 houses, including the PRECON households, with per-hour electric power usage values. Each household had 8760 per-hour electric power values a year. A question can arise here: is not there a chance of unequal electric power usage by the households of a community? The answer to this is that, in structured communities, households of similar property areas are expected to have electricity usage patterns that are quite close. Most of the time, a structured community represents a specific social class. An example of that structured community is given in Figure 5, which is Gulberg, Lahore, Pakistan. The synthesized data had statistics (mean, standard deviation, etc.) similar to those of the original data, which proved the validity of this method of feeder synthesis.
The monthly energy consumption for each feeder, along with the peak electric load, is tabulated in Table 2. The monthly energy consumption for each feeder is also graphically represented in Figure 6. It is evident from this consumption data that, almost every month, both the energy consumption and peak load of feeder 2 are smallest and those of feeder 3 are highest. This elucidates that the feeders are rightly grouped based on their consumption. The annual base load, peak load, average load, and annual energy for each of the three feeders are tabulated in Table 3. Again, feeder 1 represents medium-scale usage, feeder 2 represents small-scale usage, and feeder 3 represents large-scale usage by residential consumers.
5. BESS Selection
To perform DSM, the electrical load profiles on a per-hour basis were imported in the SAM for each feeder. The electric load annual growth rate was kept at 0.5% to get the margin while implementing this strategy in the future. The project’s lifetime was 25 years. The type of battery selected for the study was the lead-acid flooded battery that is commonly and readily available at an affordable rate for an average residential consumer. The nominal cell voltage and capacity for the lead-acid flooded battery were modeled as 2 V and 30 Ah, respectively. The battery and inverter costs were also modeled in the project with 5% contingency costs. The initial state of charge (SoC) of the battery was kept at 100%, which represents a new battery. During battery operation, the minimum SoC was 70% and the maximum SoC was 95%. This was done to ensure that the battery degrades/ages slowly. Manual dispatch was opted in the dispatch storage options to allow the battery to discharge during peak hours only [35]. The battery bank replacement threshold was set at 70% of its capacity. The battery was scheduled on a 24-hour basis. The initial value of battery energy was calculated for each feeder using the devised relationship of equation (1), where represents kWh of the battery, th is the hour of the day, and represents the load for the given hour:
The performance index of the NPV was determined for each feeder to estimate the profitability of the proposed DSM strategy. The mathematical relationship used to determine the NPV is given in equation (2). The payback period (PP) of the entire project was also calculated from equation (3) for each feeder. The increment in the LF and the reduction in the utility bill, along with the distributor’s profit, were also checked to ensure that the proposed method was responding as expected. The formulae to determine the daily LF are given in equation (4):
The objective function was to minimize the battery size such that the NPV is maximum, the PP is minimum, and the feeder LF is an improved version of the old LF. The mathematical model of the objective function is given in equation (5). The gradient descent algorithm was used to perform optimization. The runtime to obtain the final solution for each of the three residential feeders was around 6.14 s.
6. Results and Discussion
6.1. BESS Rating
The optimal battery power and energy determined for each of the three residential feeders are tabulated in Table 4. The high efficiency (∼92%) of the BESS developed for each feeder is an indication that the battery is working at its maximum possible efficiency at each scale.
6.2. BESS State of Charge
The heat maps representing the SoC of the battery bank proposed for each of the three feeders throughout the day are presented in Figures 7–9. It can be seen in the heat maps that the battery has its maximum SoC of ∼95% from 12 p.m. to 4 p.m., as it charges to its maximum during this time. The SoC of the battery bank reduces gradually from 95% to a minimum SoC of ∼70% during the peak hours in the range of 6 p.m. to 11 p.m., as it discharges during that time. After the peak hours, the battery retains its 70% SoC until the next charging cycle occurs.
The lead-acid battery available in the market whose characteristics were modeled in the simulation is Volta/Osaka P-140s [40], and the inverter chosen was UFFULL FU9000D [41]. The electrical characteristics of this equipment are given in Table 5. The number of batteries to be connected in series or parallel in the battery bank depends on the required voltage and current. The required current depends on the load which decides the battery capacity. The values of these electrical parameters can be determined by considering the electrical power and energy of the battery banks recommended for each of the three feeders, as shown in Table 4.
6.3. Economic Analysis
The input parameters modeled for the financial modeling in the techno-economic feasibility analysis of the BESS installation at the feeder level are presented in Table 6. These parameters include the values of battery cost per capacity, battery cost per kW, contingency cost (to accommodate uncertainties), inflation rates [42], discount rates [43], electricity unit price defined by Lahore Electric Supply Company (LESCO) electricity tariff [35], and expected electricity bill escalation rates. All of these values were considered for the time duration during which the project simulations were carried out (2021). Three types of electricity rates were included in the energy charges: the fixed monthly charge, peak-time charge, and off-peak time charge [35]. The electricity rates were inclusive of taxes that keep on varying depending on present circumstances. For example, the fuel price adjustment (FPA) charge varies each billing month. The electricity bill escalation rate was kept at 0.5% to provide a margin. No incentives were included in the project as no standard incentivizing is currently active in Pakistan to promote DSM.
The economic aspects of this research work, which are the NPV performance index, PP, distributor’s profit, and consumer utility bill saving as the result of the virtual implementation of the DSM project, are presented in Table 7.
It is to be noted here that the distributor’s profit and the consumer utility bill savings are equal as the difference between the peak and off-peak electricity rates is partially divided. This means that the distributor will store the energy during the off-peak time and that it will sell this energy to residential consumers at the peak time, at a rate that is equal to the off-peak rate added to the part (1/2) of the difference between the peak and off-peak electricity rates. The highest profit of around US$25k was for residential consumers with the highest percentage of electricity consumption, i.e., feeder 3, as per Table 3.
The cash flow representing the economic trend of the DSM project proposed in this research work is in favor of its implementation, as elucidated in Figure 10. The only major investment in this project is the capital cost that is to be invested to design and install the recommended battery banks for each feeder. After this initial investment, the project is expected to provide a remarkable gain during its lifetime. The after-tax cash flow is positive for each feeder, but it is highest for the feeder representing the community’s large-scale consumption of electricity.
6.4. Sensitivity Analysis
The sensitivity analysis was performed for the DSM systems proposed for each feeder. The results of the sensitivity analysis are given in Tables 8–13. The purpose of the sensitivity analysis was to observe the impact of changing specific inputs (inflation rates, discount rates, and electricity bill escalation rate per year) on a specific output (NPV). It also accounted for the uncertainties associated with the financial model parameters. The financial parameters including the inflation rate, discount rate, and electricity bill escalation rate per year are expected to vary in uncertain ways in developing countries having fluctuating economic conditions. Keeping this fact in view, the following two cases were considered in this study to observe the impact on the NPV:(1)Inflation rate against the electricity unit price escalation rate(2)Discount rate against the electricity unit price escalation rate
The tolerance that was considered to account for the variation in the financial parameters is ±25%. The highlighted NPV values in each of these tables are the values obtained at 0% sensitivity. A trend can be seen in the variation of the NPV with the financial model parameters. While moving from left to right in each row, the NPV is decreasing. While moving from top to bottom in each column, the NPV is increasing. For the first case, a maximum NPV of US$25.9 k, US$45.3 k, and US$48.9 k could be achieved if inflation were 6.8% and the electricity bill escalation rate per year were 0.625%. For the second case, a maximum NPV of US$38.8 k, US$68.3 k, and US$74.4 k could be achieved if inflation were 6.0% and the electricity bill escalation rate per year were 0.625%. The results of the sensitivity analysis elucidate the fact that even if the input parameters of inflation and discount rates change shortly, our approach will not introduce any losses. This is evident from the positive NPV values in these tables for each value of the input variables.
6.5. Load Factors
It is evident from the monthly and annual profiles of Figures 11–13 that the BESS for each feeder is scheduled such that the total electricity demand remains the same after DSM implementation. These electric power (kW) versus time (h) plots represent the total electricity load, total electricity to the load from the battery bank, total electricity to the battery from the grid, and total electricity to the load from the grid. It is visible in these profiles that, as expected, the battery charges from the grid during the off-peak time (4 hours) and discharges during the peak time (4 hours).
The improvement in another technical aspect, i.e., LFs of the three feeders representing different scales of electricity usage, as a result of the suggested DSM strategy, is included in Table 14. The LF of feeder 1 improved from 0.83 to 0.88, which is an improvement of 6%. The LF of feeder 2 improved from 0.81 to 0.82, which is an improvement of 1%. The LF of feeder 3 improved from 0.86 to 0.92, which is an improvement of 7%. For the feeder representing small-scale energy usage, the LF improvement is minimum. The reason is that morning peaks are more prominent in electricity consumption patterns than evening peaks, as elucidated in Figure 12. One of the goals of this research work was to reduce consumer electricity billing. For this reason, even though the morning peaks are higher in the feeder representing small-scale energy usage, the evening peaks are still shaved.
7. Conclusion
In this research work, a DSM strategy is proposed to benefit three parties at a time, which are utility, distributors, and residential consumers. The PRECON data set used was categorized based on electricity usage in three clusters through K-means clustering and expanded using uniform distribution in a Monte Carlo simulation to synthesize three feeders. DSM is suggested by installing the battery banks at the feeder level. The optimally sized battery banks selected for feeders numbered 1, 2, and 3 had a power of 50 kW, 90 kW, and 100 kW and battery energies of 670 kWh, 1207 kWh, and 1360 kWh, respectively. By implementing the proposed DSM method, the LF of the small-scale, medium-scale, and large-scale consumption feeders improved by 1% to 0.82, 6% to 0.88, and 7% to 0.92, respectively. The profit to the distributor and utility bill savings for residential consumers, for the feeders representing small-scale, medium-scale, and large-scale consumption, was US$23 k, US$12 k, and US$25 k, respectively.
The analyses performed in this research work are on the higher level of abstraction of power engineering, limiting the scope to load analyses and power system economics. The DSM strategy suggested in this research work is expected to provide the national benefit by managing the peak load without compromising electricity user desires. The increase in the carbon footprint caused by peakers is also expected to reduce by using a BESS at peak times. The deterioration of the transmission infrastructure to provide users with the peak load is also expected to be limited. This research work can be extended in the future by investigating the suggested DSM using other types of storage systems instead of lead-acid flooded batteries. Researchers can incorporate electricity generation options other than utilities, along with storage systems, and compare the efficiency of different solutions.
Abbreviation
| DSM: | Demand-side management |
| PRECON: | Pakistan Residential Energy Consumption |
| BESS: | Battery energy storage system |
| SAM: | System advisor model (tool) |
| NPV: | Net present value |
| LF: | Load factor |
| DPC: | Distribution Planning Code |
| NEPRA: | National Electric Power Regulatory Authority |
| SoC: | State of Charge |
| EB: | Battery energy |
| th: | Time in hours |
| EL: | Load for the given hour |
| PP: | Payback period |
| Ct: | Net cash flow at time t |
| i: | Discount rate |
| LFold: | Load factor before DSM |
| LFnew: | Load factor after DSM |
| LESCO: | Lahore Electric Supply Company (distribution company) |
| FPA: | Fuel price adjustment. |
Data Availability
Date sharing is not applicable.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This research work was performed at the National University of Sciences and Technology (NUST), Islamabad, Pakistan, and funded by Taif University Researchers supporting project number (TURSP-2020/121), Taif University, Taif, Saudi Arabia.