Low-Voltage Planning for Rural Electrification in Developing Countries: A Comparison of LVAC and LVDC Microgrids—A Case Study in Cambodia

Abstract

This paper aims to define the optimal microgrid topology for rural electrification based on the lowest total cost by comparing LVAC and LVDC microgrids across three different scenarios. An LVAC radial topology is first designed using mixed-integer linear programming for phase balancing and the shortest path for connections, then implemented with a genetic algorithm to allocate and size solar home systems, forming an LVAC microgrid. Next, an LVDC topology is then derived from the LVAC structure and integrated with solar home systems under three scenarios: (1) using the same solar home system sizes, locations, and quantities as the LVAC microgrid; (2) using a genetic algorithm to re-determine solar home system sizes and locations, forming an LVDC microgrid; and (3) clustering the LVDC topology into nano-grids, each defined by genetic algorithm for solar home system sizing and placement and connected to the main feeder via bi-directional converters. Finally, all LVAC and LVDC scenarios are simulated over a 30-year planning horizon for analysis. A non-electrified village located in Cambodia has been selected for a case study to validate the proposed methods. The results have been obtained and provide a comparison of performance indicators (i.e., costs, energy production, losses, CO₂ emissions, and autonomous energy) among the microgrids (LVAC and LVDC). The LVAC microgrid produced lower total energy losses than the LVDC microgrid in all scenarios. However, when considering environmental impact, LVDC Scenario 2 is preferable. Based on the total cost results, the LVAC microgrid is considered more economical than the LVDC microgrid in each scenario in this study.

1. Introduction

Low-voltage (LV) microgrids are becoming of interest due to their capability to ensure a reliable power supply and improve overall system efficiency [1,2]. An LV microgrid is defined as a small-scale local power distribution system capable of operating in either isolated mode or in connection with the main grid [3,4]. It also provides a flexible and reliable approach for integrating distributed energy resources (DERs) such as photovoltaic (PV) systems, hydropower, wind turbines, and battery energy storage systems (BESs) [5,6,7,8]. They deliver a reliable energy supply and a high level of sustainability, reducing environmental impacts compared with traditional energy sources such as fossil fuels or coal [4,5,9]. Traditionally, the LVAC microgrid has been the standard choice due to its compatibility with existing networks [10,11,12]. However, the increasing adoption of LVDC systems offers a compelling alternative, especially as more and more DERs generate, use and store power in DC form [9,13]. As the energy landscape evolves, comparing LVAC and LVDC microgrids becomes crucial to determine the most effective configurations for future power networks [10]. Each system has its own advantages and challenges: LVAC is widely understood and standardized [14], whereas LVDC offers potential efficiency gains by eliminating the need for power conversion in DC-dominated systems [10]. To date, in the context of LV microgrid planning, researchers and distribution system operators (DSOs) still face challenges in deciding whether to operate an LV microgrid network in AC or DC form to minimize TOTEX and enhance overall system efficiency [10]. Meanwhile, many existing studies have proposed methodologies for optimizing LV microgrid topology (including siting and sizing of DERs) in either AC or DC form, or by comparing both topologies based on various research objectives. Consequently, these works have been reviewed in the following paper.

Reference [15] proposes a method that focuses on the optimal design of an LV distribution system for rural electrification, aiming for the lowest total cost. Initially, a topology is defined based on the first-fit bin-packing (FFBP) and SP algorithms. Next, two iterative techniques and genetic algorithms (GAs) are applied to determine the sizes and locations of PVs and DeBESs (decentralized BESs). This method is implemented in a non-electrified village as a test case. The study concludes that an LV system integrated with PVs and batteries is less expensive over a 30-year period than a system without PV and battery integration (i.e., one that requires increased cable size). In reference [16], the authors suggested employing simulated annealing (SA) and a GA to optimize the selection of local energy sources for a remote village in India. The goal was to minimize the total cost while satisfying various constraints, such as determining the optimal capacity of each supply option for electrification. The findings demonstrate that the GA achieves a lower total cost than SA. In [17], the authors employed a modified gradient-based metaheuristic optimizer (MGbMO) to determine the optimal size and placement of PV systems. The objective is to minimize energy purchase costs while ensuring that voltage levels and power flows stay within the prescribed limits. The study concluded that the MGbMO offers an efficient and effective approach for solving the optimal PV placement and sizing problem. In [18], the MILP approach is employed to optimize the sizing of PV systems for an AC microgrid in a rural area of Uganda. The study’s findings indicate that the optimized system demonstrates enhanced reliability and cost-effectiveness in comparison to the baseline scenario outlined in [19]. The results indicate that integrating the PV systems improves the voltage profile and reduces power losses. In [20,21,22,23], the studies compare the technical and economic aspects of microgrids versus grid extension for rural electrification. The test systems are based in Brazil, China, and Taiwan, and the analyses are validated using HOMER software. The findings indicate that microgrids are more cost-effective than grid extension. In another study [24], a GA was used to optimize the placement and sizing of PV systems in the IEEE 33-bus network, aiming to minimize system losses and enhance voltage profiles. In [25,26], the authors conducted an economic analysis comparing the grid-connected option with a grid-connected system incorporating PV using HOMER software. The methodology was validated on 45-bus and 129-bus distribution networks representing urban and rural areas, respectively. The results indicated that the grid-connected system with PV is more cost-effective than the grid-only option. Reference [27] conducts a cost analysis of a hybrid microgrid, taking into account the estimated load demand and available energy resources for rural electrification. Simulations are performed in HOMER for three villages in Ethiopia, Uganda, and Brazil, and the cases are compared based on the levelized cost of energy (LCOE). The results indicate that hybrid microgrid systems provide a cost-effective and practical solution for rural electrification in areas where grid extension is economically unfeasible. References [28,29,30] present methods—including the water cycle algorithm (WCA) and GA—to determine the optimal placement and sizing of PV systems. The total costs are calculated and compared across different scenarios. The results show that PV system integration enhances voltage profiles and is more cost-effective than system reinforcement or scenarios without PV.

In [31], the paper presents a method for determining the optimal electrification scheme, either a fully AC system or a hybrid AC/DC distribution with PVs and batteries. The PVs are sized and located using the bottom-up method. A case study was conducted to evaluate key performance indicators in terms of technical, economic and environmental aspects. The results show that the total cost of a full AC microgrid is lower than that of a hybrid AC/DC system. In [32,33], the authors propose an LVDC microgrid model for residential buildings. The model allows residents to share energy generated by a common rooftop PV system, which is distributed to individual user units through power converters. In papers [34,35], the authors proposed models for LVDC and LVAC microgrids and compared the system losses of the two configurations. The results showed that the LVDC system exhibits slightly lower losses than the LVAC system, resulting in improved overall system efficiency. References [36,37] present approaches for rural electrification in Madagascar using DC microgrids. The proposed DC microgrid is formed by interconnecting multiple nano-grids through a 72 VDC bus. The authors introduce different connection strategies based on the minimum spanning tree (MST) and sequential opening branches (SOBs) algorithms. Cost analyses of these topologies indicate that the LVDC microgrid configured using the MST approach is more economical than the one based on the SOB method. Another study, presented in [38], proposes a community-based, small-scale centralized PV system implemented as a nano-grid for rural areas in Bangladesh. In this approach, the PV system and battery installed in a single household act as a centralized generation source, supplying electricity to other households within a cluster of 15 to 20 houses.

Based on the reviewed papers, most studies focus on optimizing the siting (defining one or several optimal locations) and sizing of PV systems and battery systems, in a large scale, using PVs (up to 45 kW) and batteries for integration into the LV system [15]. In the context of LV planning for rural electrification in developing countries, most studies do not investigate the siting and sizing of SHSs (solar home systems) located at households. Specifically, in Cambodia, SHSs installed on rooftops are the most popular and effective method for household energy access in rural areas where there is no electricity from the grid [39]. This solution allows loads to be used 24 h a day, compared with loads that are fully supplied by LV diesel generators, which have limited operating hours and higher costs [9]. The size of an SHS could be up to 500 Wp [40]. It also includes a PV module, batteries, an inverter, and a charge controller. SHSs can operate independently or be integrated into either LVAC or LVDC microgrids, and selecting between LVAC and LVDC topologies remains a challenge in terms of lower-cost operation and higher efficiency. Therefore, to address these research gaps, this present work aims to propose a methodology to define an optimal microgrid topology for rural electrification based on the lowest TOTEX over 30 years, by comparing LVAC and LVDC microgrids across three different scenarios, including microgrids and nano-grids in terms of costs and technical aspects (i.e., energy production, losses, CO₂ emissions, and autonomous energy). The proposed methods introduce a novel development planning framework for rural electrification that enables consistent techno-economic comparisons between LVAC and multiple LVDC microgrid configurations over long-term planning.

The research contributions of this paper are as follows: (1) Developing a functional framework for microgrid design that identifies the optimal LV microgrid distribution mode, i.e., AC or DC, as well as the associated topology in terms of TOTEX and technical aspects. (2) This framework provides a comparison between four structures in order to assist DSOs in their LV investment decision-making process during the planning phase in the context of rural electrification. A set of key performance indicators are computed as a radar graph to help the decision-making process. (3) Validating the proposed methodology through a real case study of a non-electrified village, with potential application in other developing countries, not just Cambodia.

This paper is organized into several sections. Section 2 presents the research methodology for LVAC and LVDC microgrids. Next, Section 3 describes the case study and research hypotheses. The LVAC and LVDC microgrid architectures are detailed in Section 4. Section 5 provides the cost computations for economic analysis. The simulation results and discussion are presented in Section 6, and finally, Section 7 draws the conclusion and suggests future research.

2. Methodology

This section outlines the research methodology used to achieve the research objectives. The microgrid planning problem simultaneously involves discrete load allocation decisions for phase balancing, network routing optimization to construct the system structure, and the determination of the size and location (with nonlinear constraints related to PV curves and battery storage behavior) of the SHS. A unified optimization model could provide a global solution but at the cost of considerable computing time due to the large-scale mixed-integer nonlinear programming problem that would require searching across many variables simultaneously. In this context, a sequential MILP–SP–GA framework has been adopted, knowing that it was not guaranteed to provide a global optimum. To facilitate the understanding of the proposed methods for LVAC and LVDC microgrids, the proposed methods are explained separately in the following subsections.

2.1. LVAC Microgrid

For the long-term planning (i.e., over 30 years) of an LVAC microgrid in a non-electrified village or specific area, it is first essential to define an optimal LVAC radial topology, including topology building and phase balancing. Second, it is necessary to determine the optimal siting and sizing of the DERs, specifically SHSs, forming the LVAC microgrid.

Reference [41] provides a detailed description of the methodology used to define an optimal LVAC radial topology for a non-electrified village. Therefore, the same method is proposed for building an LVAC radial topology in this work. Each household/load is allocated to one of three phases (A, B, or C) of the main feeders using MILP to achieve phase balancing. Then, each household is connected directly to the main feeder at the nearest electric pole via a secondary feeder based on the shortest path (SP) algorithm, establishing an LVAC radial topology. Afterward, the GA is applied to determine the locations and sizes of the SHSs located at households within the LVAC radial topology. The objective function and constraint equations for the GA are specified in [2]. This reference provides a detailed procedural methodology for sizing and siting SHSs (PVs and DeBESs) within an LVAC radial topology, ultimately forming an LVAC microgrid. Consequently, these proposed methods (in references [2,41]) are combined in this work to propose an LVAC microgrid and validated with a new test case study. The architecture for the LVAC microgrid is illustrated in Section 4.1.

2.2. LVDC Microgrids

Figure 1 presents the flowchart of the proposed methodology for LVDC microgrids across three different scenarios, described in the following steps:

• Step 1: Input data for simulation. Detailed information about these data is provided in Section 3.
• Step 2: Define the LVDC topology. In this step, the LVDC topology is established based on the LVAC structure. Section 4.2 offers a simplified explanation about how the LVDC topology is derived from the LVAC structure.
• Step 3: Site and size SHSs. After obtaining the LVDC topology, this step involves siting and sizing the SHSs. Three different scenarios are proposed for integrating the LVDC topology with SHSs to form LVDC microgrids.

  Figure 1. Flowchart of the proposed methodology for LVDC microgrids.

  Figure 1. Flowchart of the proposed methodology for LVDC microgrids.

  ➢

  Scenario 1: In this scenario, the sizes, locations, and number of SHSs defined in the LVAC microgrid remain unchanged. These SHSs are used to connect with the LVDC topology to form an LVDC microgrid.

  ➢

  Scenario 2: Here, the LVDC topology is implemented using the GA to re-determine the sizes and locations of the SHSs. These SHSs are then integrated into the topology to form an LVDC microgrid. The objective function and constraints of the GA for this scenario are provided in Section 4.2, which also describes the architecture of the LVDC microgrids for both Scenario 1 and 2.

  ➢

  Scenario 3: In this scenario, the LVDC topology is first divided into clusters. Each cluster consists of one or more households or loads connected to a bi-directional DC/DC converter installed on an electric pole. Subsequently, the GA is applied within each cluster to re-determine the size and placement of SHSs, thereby forming a nano-grid. The nano-grids of all clusters are then interconnected via bi-directional DC/DC converters connected to the main feeders of the LVDC system, resulting in an LVDC microgrid. The architecture of the LVDC microgrid for this scenario is presented in Section 4.3.

• Step 4: At this stage, each LVDC microgrid scenario is simulated over a 30-year horizon using DC load flow analysis [42], incorporating an annual load growth rate of 3% [15]. The simulation results are then obtained and analyzed in the subsequent step.
• Step 5: TOTEX are calculated for each scenario and include capital expenditure (CAPEX), operational expenditure (OPEX_network) and income (OPEX_income). A comparison is then conducted between the LVAC microgrid and the LVDC microgrid for each scenario, with emphasis on energy production and consumption, environmental impacts, and cost analysis. Finally, the optimal microgrid topology is selected based on the TOTEX.

3. Case Study and Input Data

3.1. Site Locations Description

Located in Tboung Khmum province, Cambodia, a non-electrified village has been selected for a case study to validate the proposed methods.

A three-phase LV generator (off-grid mode) is proposed to supply power to this village, since the village is far away from the existing MV feeders, approximately 9.3 km (estimated by Google Maps via Google Chrome version 112.0.5615.138) [43]. Figure 2 provides an associated topology in XY coordinates, showing the locations of the LV generator, electric poles and all loads (households). There are 24 electric poles with an average distance of 40 m between each other along the road [44]. The system comprises 68 AC loads with a total initial power demand of 33.6 kW, as determined from a survey. A power factor of 0.95 is assumed.

3.2. Load, PV, and Decentralized Battery (DeBES) Curve

Figure 3 illustrates the daily load curve, PV curve, and decentralized battery (DeBES) curve in pu. These curves are used as inputs for the simulation. The load curve is obtained as an average behavior from the surveyed households (i.e., times of use of load consumption). The PV curve (24 h) is selected from the highest annual PV curve [44], this is to ensure that the PV panels produce maximum power. The DeBES of the SHS is used to reduce the high-power demand from the generator during nighttime, especially during the peak hour at 20:00. In this work, the DeBES is proposed to charge energy during daylight and start discharging it to the loads at 18:00. This is because the PV panels are unable to produce any more energy at 18:00, based on the PV curve.

3.3. Hypotheses

To simplify this work, there are some assumptions:

• No new loads are added to the network during the planning period.
• The daily load curve shapes remain the same, but the annual load growth is 3% for the entire load curve shape [44]. The load consumption during the rainy season is assumed to be 3% lower than during the dry season.
• All SHS units in this study are identical. In LVDC system, all the loads are considered to be DC loads (24 VDC) with the same power ratings as the AC loads in the LVAC system. The 24 VDC voltage level was selected based on its common use in DC appliances in Cambodia. Although this would lead to the use of higher cable cross-sections, the increase in voltage level would require households to install DC/DC converters, which would increase costs and losses.

3.4. Input Data for Simulation

Table 1. Input data for simulation to obtain the results.

Items		Values
Discount rate [41]		6%
Minimum and maximum voltage [45]		0.9 pu or 1.1 pu
Cost of fuel [9]		0.495 $/kWh
Cost of selling energy to households [46]		0.152 $/kWh
PV cost [9]		600 $/kW
DC charge controller [47]		30 $/piece
Battery cost [9]		105 $/kWh
Single-phase bi-directional inverter or converter [48]		400 $/kW
Three-phase bi-directional inverter or converter [49]		820 $/kW
DC cable length used per SHS		10 m/SHS
Maintenance cost (PV + battery + inverter/converter + charge controller) [50]		11.5 $/kW/year
LV generator cost [9]		500 $/kW
Efficiency of charge controller, inverter/converter, and battery [9,51]		95%
Degradation of charge controller, inverter/converter, PV, and battery [2]		0.5%/year
Lifespan of battery [9]		5 years
Lifespan of charge controller, inverter/converter, and LV generator [9]		15 years
Lifespan of PV panels [2]		25 years
Cable costs (1 core) [52]	4 mm2	76 $/km
	70 mm2	1330 $/km
	120 mm2	2280 $/km

shows the input data for simulation to obtain results and economic analysis.

4. LVAC and LVDC Microgrid Architectures

For ease of understanding, the architecture of each microgrid (LVAC and LVDC) is described in the following subsections.

4.1. LVAC Microgrid Architecture

By implementing the proposed method in reference [41] into the non-electrified village given in Figure 2, an LVAC radial topology is established, as depicted in Figure 4. Each household/load is assigned to one of the three phases and is directly connected to the main feeders via the secondary feeders. Next, the SHSs are allocated within this topology using the GA, as described in reference [2].

Figure 5 illustrates the integration of SHSs—determined by the GA—with the LVAC topology, forming an LVAC microgrid. Each SHS consists of a PV panel, a charge controller, a decentralized battery energy storage system (DeBES), and an AC/DC bi-directional inverter [39]. In this study, the nominal DC voltage of the SHS is 24 VDC, which is commonly used for SHSs in Cambodia. During the day, energy generated by the PV panels is delivered to the charge controller, which automatically manages the charging and discharging of the DeBES and supplies power to the household loads. Any excess energy is fed into the LV grid and stored in the centralized battery energy storage system (CeBES) through an AC/DC bi-directional inverter. At night, the DeBES discharges to supply the household loads, as the PV panels no longer generate energy. If additional power is needed, the CeBES discharges some to support the LV grid. Once the CeBES is depleted and reaches its minimum state of charge (SoC), the LV generator supplies the LV grid. The generator’s output power (V_L-L = 400 V) is delivered directly to the main feeders. Households without SHSs receive power directly from the LV grid or the main feeders. This process is repeated for the subsequent days.

4.2. LVDC Microgrid Architecture for Scenarios 1 and 2

Figure 6 shows the LVDC system architecture for Scenarios 1 and 2, derived from the LVAC structure in Figure 4. In this LVDC topology, all connections between loads and electric poles remain identical to those in the LVAC system. However, the main feeders consist of only two lines, corresponding to the positive (P) and negative (N) conductors.

The DC voltage of the main feeders is set at 230 VDC, chosen to align with the voltage level used in the LVAC system. For both scenarios, a DC/DC converter is installed at each household to step down the voltage from 230 VDC (main feeder voltage) to 24 VDC (load voltage), as shown in Figure 6. Consequently, the secondary feeders continue to operate at 230 VDC, the same as the main feeders.

For the LVDC microgrid in Scenario 2, it is necessary to re-determine the sizes and locations of the SHSs within this LVDC topology. The GA is commonly used to solve nonlinear problems with constraints. In power distribution planning, this algorithm is widely and effectively applied to optimize various tasks such as determining the location and size of power generations, minimizing power losses in distribution systems, and optimizing investment costs [25,30]. Since this optimization involves multiple variables, a nonlinear objective function, and various constraints, the GA is selected as the most suitable method for this optimization. The objective function and constraint equations are presented as follows:

• Objective function:

$$\min \left({\displaystyle \sum _{t=1}^{24}{P}_{dc losses}\left(t\right)+{\displaystyle \sum _{t=1}^{24}{P}_{dc excess}\left(t\right)}}\right)$$

(1)

• Constraints:

$$V_{\min }\le V_{system}\le V_{\max }$$

(2)

$$I_{system}\le I_{\max }$$

(3)

where P_{dc losses} represents the DC power losses in the LVDC system (including both main and secondary feeders), and P_{dc excess} denotes the reversed DC power at the slack bus (generator electric pole). The CeBES and the LV generator continue to operate in the LVDC microgrid for Scenarios 1 and 2, supplying power to the loads in the same manner as described for the LVAC microgrid. An AC/DC converter is added to convert the AC power from the LV generator into DC power for the main feeders.

4.3. LVDC Microgrid Architecture for Scenario 3 (Nano-Grid)

Figure 7 shows the LVDC system architecture for Scenario 3. In this scenario, if an electric pole is connected to one or more households/loads, the group of these loads is identified as a cluster.

The index numbers in Figure 7 indicate the cluster indices within the LVDC topology. In this study, a total of 22 clusters is identified. Specifically, in the seventh cluster, a bi-directional DC/DC converter is installed at the electric pole to step down the voltage from 230 VDC (main feeder voltage) to 24 VDC. Each household or load in this cluster is connected to the secondary side of the converter via a secondary feeder, resulting in a secondary feeder voltage of 24 VDC.

Next, each cluster is sited and sized with SHSs, forming them into nano-grids. Figure 8 illustrates the process of siting and sizing SHSs for each cluster. Starting with the first cluster (n = 1), the size and location of its SHSs are determined using the GA, based on the objective function and constraint equations given in Equations (1)–(3). First, the GA searches for the optimal SHS size for each household within the cluster. Next, the household (equipped with an SHS) that provides the best value of the objective function is selected as the best location among the households in the cluster. Once the SHSs for the first cluster are defined, the same process is repeated for the remaining clusters until all are completed. In the end, the sizes and locations of SHSs for each cluster are established. Note: There is only one household/load in each cluster containing SHSs that supplies power to other households/loads within its cluster. In this work, the population size is set to 50. Since there is no specific instruction regarding the population size, this value is selected based on those commonly used in similar research [44]. The MATLAB ga function and its default parameters were used. To reduce the search space, different PV panel power ratings [430 W, 440 W, 450 W, 530 W, 540 W, and 550 W] were used as input options. Pⁿ_{dc losses} is the DC power loss within the nth cluster, and Pⁿ_{dc excess} is the DC power reversed from the nth cluster to the main feeder of the LVDC system via the DC/DC bi-directional converter.

Figure 9 shows the proposed diagram of all nano-grids integrated together. The nano-grids of each cluster are connected to the main LVDC feeder through the converters, forming an LVDC microgrid. This configuration allows each nano-grid to exchange power through the DC bus or the main feeders (230 VDC) of the LVDC system. When additional power is needed, it is supplied by the CeBES/LV generator.

4.4. CO₂ Emissions and Autonomous Energy

To calculate CO₂ emissions, all energy sources are considered in this study, including PVs, batteries, and the LV generator. The emission factors for PVs, batteries, and the LV generator are 38 g-CO₂/kWh, 33 g-CO₂/kWh, and 1270 g-CO₂/kWh, respectively [9,53,54].

Autonomous energy refers to the proportion of total energy generated by the PVs of SHSs relative to the total energy supplied by all sources. The total energy from all sources includes the energy supplied by the generator at the slack bus as well as the PVs. Therefore, the autonomous energy can be expressed by the following equation:

$$[\%] Autonomous energy={\displaystyle \frac{Energy from PVs}{Energy from all sources}}\times 100\%$$

(4)

5. Economic Analysis

This section outlines the cost formulations used to calculate all expenses associated with each microgrid, including CAPEX, OPEX_network, OPEX_income, and TOTEX.

5.1. CAPEX and OPEX_network

CAPEX comprises the costs of LV cables (C_{LV cable}), the LV generator (C_generator), batteries (C_{CeBES + DeBES}), PV panels (C_PV), DC/DC converters, AC/DC converters, and charge controllers (C_{converter + charg.}), as well as replacement costs when components reach the end of their lifespan. The CAPEX formulation is expressed by the following equation:

$$\begin{array}{c}CAPEX=C_{generator}+C_{LV cable}+C_{PV}+C_{CeBES+DeBES} +C_{converter+ch\arg .}+\left({\displaystyle \sum _{k=1}^5\frac{C_{CeBES+DeBES}}{{\left(1+r\right)}^{k\times 5}}}\right)\\ +\left({\displaystyle \frac{C_{generator}+C_{converter+ch\arg .}}{{\left(1+r\right)}^{15}}}\right)+\left({\displaystyle \frac{C_{PV}}{{\left(1+r\right)}^{25}}}\right)\end{array}$$

(5)

where k is the index ranging from one to five used to determine battery replacement costs, and r is the discount rate [%]. Next, OPEX_network includes the operating costs of the LV generator as well as the maintenance costs for PV panels, batteries, inverters/converters, and charge controllers. Therefore, OPEX_network can be written as follows:

$$OPE{X}_{network}={\displaystyle \sum _{i=1}^{30}\left(\frac{\left(E_{generator}\left(i\right)\times C{F}_{k\$/kWh}\right)+Main.}{{\left(1+r\right)}^i}\right)}$$

(6)

where

$$E_{generator}\left(i\right)={\displaystyle \sum _{t=1}^{8760}{P}_{generator}\left(t,i\right)}$$

(7)

CF_$/kWh represents the cost of energy purchased from the LV generator [$/kWh] and i denotes the index year. E_generator and P_generator are the energy and power supplied by the LV generator in the i^th year, respectively. Next, t represents the time step, which is 1 h for each interval.

5.2. Income (OPEX_income)

OPEX_income includes revenues from selling energy to households/loads (OPEX_{income LV}) and the salvage value of equipment, specifically the PV panels. These are considered as income in this context. Therefore, the OPEX_income equation is expressed as follows [55]:

$$OPE{X}_{income}=OPE{X}_{income LV}+Salvage$$

(8)

where

$$OPE{X}_{income LV}={\displaystyle \sum _{i=1}^{30}\left(\frac{\left(E_{loads}\left(i\right)\times C{S}_{k\$/kWh}\right)}{{\left(1+r\right)}^i}\right)}$$

(9)

$$E_{loads}\left(i\right)={\displaystyle \sum _{t=1}^{8760}{P}_{loads}\left(i,t\right)}$$

(10)

CS_$/kWh represents the cost of selling energy to households [$/kWh]. E_loads and P_loads denote the total load energy and power in the ith year, respectively.

5.3. Total Expenditure or Total Cost (TOTEX)

TOTEX represents the total costs for each microgrid: it includes CAPEX, OPEX_network, and OPEX_income. Thus, the TOTEX formulation is as follows:

$$TOTEX=CAPEX+OPE{X}_{network}-OPE{X}_{income}$$

(11)

6. Simulation Results and Discussion

In this work, the proposed methods have been simulated in MATLAB R2021a. The simulation results and analyses are provided in the following subsections.

6.1. LVAC and LVDC Microgrid Topologies

Figure 10 demonstrates the LVAC microgrids and LVDC microgrids in Scenarios 1, 2, and 3. The cross-sections of the main and secondary feeders are 70 mm² and 4 mm², respectively, for the LVAC and LVDC microgrids in Scenarios 1 and 2. However, for the LVDC microgrid in Scenario 3, the cross-sections of both the main and secondary feeders are increased to 120 mm² each to maintain the voltages within the accepted values [45].

Table 2. Simulation results.

Items	LVAC	LVDC Sce. 1	LVDC Sce. 2	LVDC Sce. 3
Total number of SHSs	37	37	50	45
Total PV output power (12:00) [kW]	16.65	16.65	22.5	20.25
Total max. DeBES power (20:00) [kW]	23.9	23.9	32.3	29
Total size of DeBES [kWh]	144.3	144.3	195	175.5

provides the simulation results of the total number of SHSs for each microgrid, including the total PV output power, maximum DeBES discharge power, and DeBES size.

Based on the GA, an SHS contains a PV panel with a capacity of 450 W and a maximum DeBES power of 646 W at 20:00. According to the DeBES curve in Figure 3, the DeBES size is found to be 3.9 kWh [51], including a 20% minimal SoC. Based on the results, there are 37 units of SHSs found for the LVAC microgrid in Figure 10a, providing 16.65 kW of total maximum PV output power at 12:00. At 20:00, the maximum total DeBES discharge power is 23.9 kW (including efficiency), and the total size is 144.3 kWh. As mentioned, the LVDC microgrid in Scenario 1 (Figure 10b) has the same sizes, locations, and number of SHSs as the LVAC microgrid; therefore, 37 units of SHSs are integrated into this microgrid. Additionally, the results of PV and DeBES for LVDC Scenarios 2 (Figure 10c) and 3 (Figure 10d) are also provided in Table 2. The SHSs installed per cluster (LVDC Scenario 3) are indicated.

6.2. Performance Indicators

Table 3. Performance indicators.

Items		LVAC	LVDC Scenario 1	LVDC Scenario 2	LVDC Scenario 3
Vmin at 30th year [pu]		0.97	0.90	0.91	0.91
EPV [MWh]		1090.74	1090.74	1473.87	1326.4
Egenerator [MWh]		2258.41	2536.28	2175.74	2330.6
Energy reversed at generator bus [MWh]		0.145	0.134	0.23	0.22
Eloads [MWh]		3120.5	3120.5	3120.5	3120.5
Losses [MWh]	SHSs	202.49	472.53	504.9	493.73
	Main and sec. feeder	26.27	34.14	24.37	37.14
	Total	228.76	506.67	529.27	530.87
CeBES [kWh]		3.9	3.9	3.9	3.9
CO2 emissions [tone]		2909.6	3262.5	2819.2	3003.6
Autonomous Energy [%]		32.5	30.0	40.3	36.2
Gradual electrification		No	No	No	Yes

provides the simulation results of each microgrid over 30 years. Based on the results, the LVAC microgrid exhibits the highest minimal voltage compared to LVDC in those three scenarios. This is because the LVDC systems consist of only two lines for the main feeders, and these feeders have long connections to loads. The LVDC microgrid in Scenario 2 generates the highest total energy from PV panels (1473.87 MWh), which is approximately 35% higher than that of the LVAC microgrid or the LVDC microgrid in Scenario 1. As a result, the required energy from the generator is reduced compared to the LVAC microgrids and the LVDC microgrids in Scenario 1 and 3. This also increases the autonomous energy in Scenario 2 to 40.3%, the highest among all scenarios. Next, the LVDC microgrid in Scenario 3 experiences highest total energy losses, including losses from SHSs as well as the main and secondary feeders, particularly on the main and secondary feeders, due to the nano-grids in each cluster operating at a low voltage of 24 VDC.

In contrast, the LVAC microgrid exhibits significantly lower total energy losses than the LVDC microgrids in all scenarios, with total losses approximately 2.32 times lower than those observed in LVDC Scenario 3. Next, the size of the CeBES used for each microgrid is 3.9 kWh: it is sized based on the power exchanges at the slack bus. This CeBES is replaced with one of the same size every 5 years due to its lifespan. Lastly, in terms of environmental impact, the LVDC microgrid in Scenario 2 has the best performance, reducing CO2 emissions by about 3% compared to the LVAC microgrid. In contrast, the LVDC microgrid in Scenario 1 produces the highest CO₂ emissions, increasing them by 3% compared to the LVAC structure, due to its greater reliance on energy supplied by the generator.

On the other hand, LVDC Scenario 2 has 13 more SHS units than LVDC Scenario 1, despite both scenarios using the same load curve. As illustrated in Table 3, this difference is attributed to the GA implemented in Scenario 2, which seeks to minimize losses in the main and secondary feeders by reducing the energy drawn from the generator compared to Scenario 1. As a result, the contribution from PV panels increases, leading to a higher number of SHS units in LVDC Scenario 2. Appendix A.1 illustrates the annual energy over 30 years, including energy losses in the main and secondary feeders, energy losses in SHSs, energy generated by PV panels, energy supplied by the generator, and total energy demand.

6.3. Cost Comparison

The 30-year cost calculations for each microgrid are presented in Figure 11. According to the results, the LVAC microgrid achieves the lowest CAPEX, approximately 59.1% lower than the highest CAPEX, which occurs in LVDC Scenario 3. It also delivers the lowest LCOE across all LVDC scenarios. In Scenario 1, the LVDC topology draws more energy from the slack bus, leading to a higher OPEX_network and, consequently, an increased TOTEX. In Scenario 2, although the LVDC shows the lowest OPEX_network, its TOTEX remains higher than that of the LVAC microgrid, mainly due to its higher CAPEX. Scenario 3, where the LVDC is configured into nano-grids, proves less cost-effective, producing a higher TOTEX than LVDC Scenario 2 without nano-grids. Finally, OPEX_income is fairly similar across all microgrids, as total load energy remains constant and differences in the salvage value are small.

Nevertheless, it has to be mentioned that this analysis concerns the electrification of a village all at once. In the case of staged electrification, the nano-grid-based architecture corresponding to Scenario 3 would be the best option, since it enables a bottom-up approach (gradual electrification). Indeed, with the LVDC Scenario 3 structure, the operator would have to spend 205.56 k$ at once, whereas in the nano-grid-based topology, the mean cost per nano-grid is around 9.34 k$. Appendix A.2 depicts the cumulative TOTEX evolution of LVAC and LVDC for all scenarios over 30 years.

Figure 12 shows bar charts of total investment costs (positive values) and incomes (negative values) over 30 years for each microgrid, along with a breakdown of costs by item. In LVDC Scenario 3, cable investment is significantly higher—about 4.39 times that of LVDC Scenarios 1 and 2—due to larger cable sections in both the main and secondary feeders to maintain acceptable system voltage levels. Additionally, in all LVDC scenarios, each household requires a DC/DC converter to step down the voltage from the 230 VDC grid to the 24 VDC load level. This increases the investment cost for converters, as they become approximately 2.78 times higher than that of the LVAC microgrid. Finally, LVDC Scenario 2 has the highest salvage value because of the greater number of SHSs, resulting in higher incomes compared to the other microgrids.

To conclude, based on the lowest TOTEX, the LVAC microgrid is more economical than the LVDC microgrid across all scenarios considered in this study.

7. Conclusions and Future Works

This paper presents methods for identifying an optimal LV microgrid by comparing LVAC and LVDC configurations across three scenarios in terms of TOTEX for rural electrification, energy autonomy and CO₂ emissions. An LVAC radial topology is designed using a GA to determine the locations and sizes of SHSs, integrating each SHS into the LVAC network. Based on this structure, an LVDC topology is developed and integrated with SHSs, forming LVDC microgrids across the three scenarios. The proposed methods are validated through a case study of a non-electrified village in Cambodia. The results indicate that the LVAC microgrid consistently achieves lower total energy losses than the LVDC microgrid across all scenarios. Conversely, the LVDC nano-grid in Scenario 3 exhibits the highest total energy losses, especially along the main and secondary feeders, because the nano-grids in each cluster operate at a low voltage level of 24 VDC. When environmental impact is taken into account, the LVDC microgrid—particularly in Scenario 2—is more favorable, as this configuration includes the highest number of SHSs compared with the other systems. The LVAC microgrid results in the lowest CAPEX due to having fewer components (i.e., DC/DC converters), whereas LVDC Scenario 2 achieves the lowest OPEX because it includes a higher number of SHSs. The LVDC system organized into clusters forming nano-grids appears to be less cost-effective, as it leads to a higher TOTEX value than the LVDC microgrid without nano-grids (i.e., LVDC scenario 2). To conclude, in terms of TOTEX, the LVAC microgrids are more economical than the LVDC microgrids in all scenarios.

To enhance the comprehensiveness and realism of future research, several improvements should be considered. First, forecasts for new loads connected to the system should be developed, and load profiles could be updated daily based on measurements or requests from local DSOs, where feasible. The uncertainties in fuel costs, component prices, electricity selling prices, and the discount rate should be considered in the long term. Next, it is also advisable to periodically upgrade SHS sizes to accommodate annual load growth. Also, protection and stability studies could be performed to confirm the superiority of the LVAC structure. A comprehensive reliability assessment incorporating component failure rates and restoration times has to be considered in future work to evaluate how SAIDI and SAIFI are impacted. Additional case studies are necessary to further validate the proposed methods, and the results should be benchmarked against those obtained from existing microgrid software, such as HOMER. In this research, the LVAC microgrid has a lower CAPEX, while the LVDC microgrid provides a lower OPEX. This provides a perspective that rather than choosing a fully LVAC or LVDC system, future systems may combine both approaches. It should be noted that deriving the LVDC structure from the existing LVAC layout may partially constrain the design from achieving an optimal LVDC topology. This is due to differences in main feeder routing (e.g., nano-grids), converter placement, and varying distribution voltage levels. Accounting for these factors may lead to more meaningful comparisons and more reliable conclusions.

Although the present work focuses primarily on a TOTEX-based comparison, LVDC architectures may provide additional long-term advantages in terms of operational flexibility and compatibility with the increasing penetration of DC-native loads (e.g., electronics, LED lighting, battery systems, and electric mobility). These benefits were not explicitly quantified or monetized in the current study and therefore represent an important direction for future research. Incorporating DC load growth scenarios and converter reduction effects could lead to a more comprehensive assessment of LVDC microgrid scalability and long-term operational value.