Scaling Renewable Energy Based Microgrids in ... - Andrew.cmu.edu

2016 IEEE PES Power Africa Conference

Scaling Renewable Energy Based Microgrids in Underserved Communities: Latin America, South Asia, and Sub-Saharan Africa Andrew Harrison Hubble, Taha Selim Ustun School of Electrical and Computer Engineering Carnegie Mellon University, PA, USA Abstract - With 1.3 billion people lacking access to electricity, the demand for power is increasing drastically. Electricity can no longer be considered a luxury, and is necessary for maintaining a modern life expectancy. This paper explores the feasibility of energy production and storage across South America, South East Asia and sub-Saharan Africa with National Renewable Energy Lab’s Geospatial Toolkit and the integrated HOMER interface. Multiple scenarios are constructed based on different community sizes with relative demand and load profiles. Optimization techniques are used to value both traditional generation options (diesel generation), their renewable counterparts (wind and solar), as well as energy storage methods (batteries). Each analysis is based on location-specific irradiance, wind speeds, and fuel prices. Economics remain at the forefront, as a community’s ability to purchase the generated power is the centerpiece of this entire investigation. With all three regions in a similar state of development, parallels can be drawn among them. By scrutinizing data from all three continents, similarities can be observed between population density and distribution, their demand and usage of electricity, and purchasing power parity. These similarities can be used to fill gaps in our understanding of how a community may use generated electricity, or how that demand will change over time. This is instrumental in scaling renewable energy based microgrids globally. Keywords — Microgrids, Generation, Energy Storage

Grid

Extension,

Distributed

I. INTRODUCTION As both traditional fossil fuels and emerging renewable energy (RE) technologies compete for market share of generation capacity, important decisions on their relative inclusion for small-scale power markets become ever more critical. Additionally, as rural and remote parts of the world are electrified, the decision between tethering to the grid and standalone systems becomes an important economic decision. Large grids and their capital costs can be prohibitively expensive for governments of developing countries, and are not efficient for sparsely populated regions [1]–[3]. In many instances, the vast distances between cities and settlements is a major factor in grid extension costs. Furthermore, the lower the population density, the more difficult it becomes to access the remote areas due to lack of roads and infrastructure, and the lower rural usage further extends payback periods. Microgrids can offer the same electrification results, without the expensive network of long-distance transmission lines and heavy capital cost of a traditional power plant. Two key issues arise with distributed renewables: the selection of renewable

978-1-4673-9981-4/16/$31.00 ©2016 IEEE

sources and their intermittent generation. Solar and wind – in the forms used today – will never be as consistent as traditional power plants. At best, their reliability can be predicted, but never controlled. This unpredictability has left renewable penetration on the fringes; dependent on backups and baseloads ready to take the brunt of the responsibility. The data which exists on potential sites is as intermittent as the generation itself. Creating better models for understanding both generation and usage is the first step in electrifying the developing world. While microgrids with decentralized production minimize investment in long distance transmission lines, all aspects of power production and transmission from source to sink must be evaluated, including the decision to include sources from renewables versus fossil fuels and in what quantities. Microgrids can be community-based and allow a region to take advantage of geo-specific renewable resources, instead of relying solely on fossil fuels. Their design often requires lengthy study periods and intense research on both the availability of the renewable resources as well as the populations’ usage, growth, and economic status. Currently, there exists no uniform microgrid design which is applicable to all or even most potential microgrid sites. Each microgrid is tailored to a specific location. Custom designed microgrids – whether or not they include renewables – are slow and costly when attempting to electrify the developing world. This paper seeks to reduce the design considerations over a range of communities to a manageable number of parameters. The streamlining will in turn allow underserved areas to rapidly electrify based on available local information and comparisons to similar communities. Fifteen countries were selected across three continents to represent a diversified sample of the developed and developing world. Countries were selected based on factors such as their size, population, renewable potential, economic status, and current electrical generation and transmission infrastructure, but priority was given to nascent countries. Simulations were run to determine the optimal mixture of solar, wind, and diesel generation, as well as battery storage capacity. These solutions could then be compared to estimated costs of extending the current grid to the community. The remainder of this paper is organized as follows: Section II focuses on the design, setup, and usage of the microgrid; Section III analyzes the commonalities and differences between solutions; Section IV compares the viability of building the proposed microgrid to an extension of the current grid, whereas Section V contains the conclusions and discussions, with Section VI proposing future work.

134


II. MICROGRID DESIGN AND LOAD-DEMAND OPTIMIZATION In order to investigate the commonalities between countries on different continents and with different renewable energy potentials, fifteen countries – five from each of the aforementioned continents – were selected to run the microgrid simulations, and are depicted in Figure 1.

the profiles have high evening peak demands, which is more than double the mid-day average. To optimize these load profiles, HOMER was utilized to consider varying system sizes of photovoltaic, wind turbines, diesel generators, batteries, and converters. Figure 3 illustrates a simple diagram of the microgrid setup.

Fig. 1. Selected Countries

The selected countries are: Argentina, Brazil, Ecuador, Paraguay, Uruguay, the Democratic Republic of the Congo, Ghana, Lesotho, Rwanda, Tanzania, Afghanistan, India, Indonesia, Nepal, and Singapore. These countries were chosen for their diverse population sizes and densities, current electrification rates, cost of diesel, port access and many other factors. The goal is to find commonalities between these vastly different countries. Utilizing NREL’s Geospatial Toolkit and HOMER, simulated microgrids were organized for 15 candidate sites in 15 different countries. While parameters such as irradiance, wind speed, and diesel fuel prices varied according to location, for the purposes of this paper, item costs such as photovoltaic panels, wind turbines, and even grid extension costs were held constant. At each of the 15 different sites, three unique load profiles were used to represent different community sizes and standalone systems, seen in Figure 2.

Fig. 2. Load Profiles for Various Scenarios

The community and standalone load profiles are products of NREL Geospatial Toolkit and depict typical energy consumption for varying communities as well as standalone systems described as a small home or clubhouse [4]. For Community I, the demand averages 5kWh/d spread over about ten small residences. Community II represents a community demanding an order of magnitude more of electricity. At 50kWh/d, this community represents small, medium, and large houses, in addition to a small store. The Standalone I system represents a single relatively large building or residence. All of

Fig. 3. HOMER Microgrid Diagram

The sizes and relative contribution of each generation type vary depending on the load profile and renewable availability. To standardize the optimization parameters, Table I illustrates the system sizes HOMER was allowed to select from. It’s worthwhile to note that batteries are measured in quantity – i.e. the batteries considered for Community I and Standalone I were Trojan T-105 batteries (1.35 kWh each), whereas Community II utilized L16P batteries (2.16 kWh each). TABLE I. OPTIMIZATION COMPONENTS Considered Components

Community I

Community II

Standalone I

Generator (kW)

0, 1, 5

0, 5, 7, 10, 18

0, 1, 5

Photovoltaic (kW)

0, 0.25, 0.5, 0.75, 1, 1.25, 1.5, 2, 2.5, 3

0, 2, 4, 6, 8

0, 0.25, 0.5, 0.75, 1, 1.25, 1.5, 2, 2.5, 3

Wind (kW)

0, 1, 2, 3, 4, 5

0, 1, 2, 3, 4, 5, 6

0, 1, 2, 3, 4, 5

Batteries (Each)

0, 2, 4, 6, 8, 10, 12, 16, 20

0, 5, 10, 15, 20, 25, 30

0, 2, 4, 6, 8, 10, 12, 16, 20

Converter (kW)

0, 0.75, 1, 2, 3

0, 2, 4, 6

0, 0.75, 1, 2, 3

HOMER optimizes the microgrid solution in addition to taking into account the break-even cost for grid extension. It can be difficult to find precise and current data for all candidate countries on the cost of transmission line extension; this, coupled with the fact that grid extension can service multiple sites – or at least provide a closer tie-in point, it is difficult to fully understand the true cost – and benefits – of grid extension. For the purposes of this paper, a grid extension cost of $20,000 per km was selected as a reasonable figure for developing and difficult to access areas [5]. An additional factor in the microgrid expense is the cost and availability of diesel fuel. While the cost of some equipment remains relatively constant from country to country (photovoltaics and wind turbines) neglecting import duties and other taxes, diesel fuel prices can vary wildly. Fuel prices vary even inside a country, and tend to increase in price in more

135


remote areas. For the purposes of this paper, average countrylevel prices were used and are outlined in in Table II [6]. TABLE II. DIESEL PRICES Country

USD/L

Country

USD/L

Country

USD/L

Afghanistan

1.19

Argentina

1.33

DRC

1.67

India

0.91

Brazil

1.02

Ghana

1.03

Indonesia

0.62

Ecuador

0.29

Lesotho

1.07

Nepal

1.04

Paraguay

1.16

Rwanda

1.41

Singapore

1.16

Uruguay

1.72

Tanzania

1.20

The last set of variables considered were irradiance and wind speeds. This data was taken from SWERA [7] based on community GPS coordinates. With these input parameters varying, and all other factors held constant, HOMER simulations for all 75 cases were found to yield robust solutions. III. OPTIMIZATION RESULTS AND EXTRACTED MICROGRID PARAMETERS

generation capacity. More importantly, it is optimal to use batteries on every scale. This is obvious when examining the Standalone I load profile. Each solution is dependent only on renewables, and therefore runs the risk of failing to meet demand since generation is intermittent. The counterpoint is that batteries are still optimal for solutions with generators as well. This makes sense because the generator provides a steady amount of power, whereas the load profile continues to change. The batteries allow the generator to shutdown, decreasing the amount of unused capacity generated. Additionally, generators appear in every community scenario. For instance, the Afghanistan case has the third best wind speeds, fourth best irradiance, and third highest diesel prices of the locations sampled. This scenario should lend itself to a high renewable, low-diesel solution, however, it does not. Despite the high renewables and high diesel prices, Afghanistan requires the use of a generator to meet peak demand. Figure 4 demonstrates the generator output for Afghanistan’s community II.

The optimal solutions to the above setup return a variety of generation and storage components, and it is the causes of the commonalities and differences that provide the most insight to creating a streamlined microgrid design approach. Table III below illustrates the components, but not sizing, of each generation and storage type. Fig. 4. Afghan Community II Generator Output

TABLE III. OPTIMAL GENERATION MIX Country

Community I

Community II

Standalone I

Afghanistan

P, W, G, B

P, W, G, B

W, B

India

P, W, G, B

P, W, G, B

P, B

Indonesia

P, G, B

P, G, B

P, B

Nepal

P, W, G, B

P, W, G, B

P, W, B

Singapore

P, G, B

P, G, B

P, B

Argentina

W, G, B

P, W, G, B

W, B

Brazil

P, G, B

P, G, B

P, B

Ecuador

P, G, B

G, B

P, B

Paraguay

W, G, B

W, G, B

W, B

Uruguay

P, W, G, B

P, W, G, B

P, W, B

DRC

P, G, B

P, G, B

P, B

Ghana

P, G, B

P, G, B

P, B

Lesotho

P, W, G, B

P, W, G, B

P, W, B

Rwanda

P, G, B

P, G, B

P, B

Tanzania

P, W, G, B

P, W, G, B

P, W, B

.P

= Photovoltaic W = Wind Turbines G = Diesel Generator B = Battery Storage

Several important occurrences are immediately noticeable. First, batteries occur in every scenario. Regardless of location, irradiance, wind speeds, or fuel costs, batteries are required to meet electrical demands. This presents itself in two ways. Since batteries are storage and not generation, they are not physically required to meet demand – these can be replaced by additional

The red band at the top of the figure shows the 5kW generator operating at full capacity from approximately 18:00 to 22:00, with little use during other times of the day. This is understandable, due to the high evening peak demand, but also generates an enormous amount of wasted generation potential. Over the course of the year, the generator operates at a capacity factor of only 19.2%. Of all the Community II simulations, the highest capacity factor for a generator was 47.6% in Ecuador, which has the lowest diesel price of 29 cents per liter. Despite Afghanistan’s high renewable availability and relatively high diesel prices, the generator is required because of the mismatch between peak demand and peak renewable generation. The oversizing of wind and solar generation, coupled with the batteries required to store the energy until it met peak demand would return a costly solution. It is important to note that the load profiles modeled by NREL are not the most accurate models for rural electricity usage. Often, there exists very little peak usage in the evenings owing to few high power devices in the community. Deeper investigations have found rural demand to consist of high midday usage and relatively little evening usage [2]. An issue that arises with smaller microgrids is the resolution of system sizes. Solar panels and batteries come in a wide variety of sizes, allowing for a finer resolution to truly optimize the problem. In contrast, wind turbines and generators do not have a fine resolution and developing models to calculate optimal generator sizes to fractions of a kW are not suitable for

136


Since every community I and II scenario contains a 1 and 5kW generator respectively, and standalone I contains no generators, the appreciable variation in solutions is contained solely in the renewable and fuel consumption aspects. Since the capital cost of the generator is comparatively small, the capital cost becomes a non-factor. Because every community has a generator, the cost optimization becomes fuel expenditures compared to capital cost of renewables (assuming all maintenance is equal). The comparison between high RE and low RE yields interesting simulation results, seen in Table IV. TABLE IV. RENEWABLE SIZING Country Paraguay

Photovoltaic (kW)

Wind (kW)

CIa

CII

SI

CI

CII

SI

Installed Capacity

0

0

0

1.0

6.0

0.5

7.50

India

0.25

4.0

0.5

1.0

2.0

0

7.75

Ecuador

0.25

0

0.5

1.0

0

0

1.75

DRC

1.50

8.0

0.5

0

0

0

10.0

aCI

(community I), CII (community II) and SI (standalone I)

These four countries represent the extreme cases of the study. Paraguay has moderate solar and excellent wind. India has excellent solar and moderate wind. Ecuador and the DRC have poor renewable capacity in general, and low and high diesel prices respectively. In this model, the capital cost of 1kW of wind generation ($4000) is substantially less expensive when compared to solar ($6000). Because Paraguay has the highest wind speeds of the countries selected, it returns no photovoltaics, concentrating wholly on wind generation. India uses a mix of the two, concentrating more heavily on wind for smaller grid sizes, and solar for larger ones. Over the three communities, the two countries recommend nearly the same RE capacity.

in this study were purposefully selected to be remote and far removed from existing transmission lines, such that it is far more economical to build the proposed microgrid than to tie into the existing national grid. In order to compare the microgrid and national grid solutions, a cost of grid extension must be well understood. Figure 5 shows breakeven distances with the previously mentioned $20,000/km grid extension cost. 10 Break-Even Grid Extension (km)

real world application. This can trap the model into requiring over-sized components, driving up the levelized cost of energy.

1

0.1

Community I

Community II

Standalone I

Fig. 5. Grid Breakeven Distances

While first glance indicated that these figures are tightly grouped, an interesting trend emerges when seen graphically. Dips in the breakeven distance emerge across all sizes for countries such as Argentina and Paraguay. This indicates highperforming microgrids, where a community would have to be closer to the grid to make a grid connection economically viable. As a whole, these breakeven distances favor the building of microgrids heavily. The highest breakeven distance seen here is 6.32km, which, when a corridor is drawn along the transmission lines, represents a tiny fraction of land in favor of tying into

The relationship between the two low-RE countries is deceptive at first glance. The Ecuadorian solutions utilize very little solar and wind, and focus very heavily on diesel power. This is because Ecuador has the lowest diesel prices ($0.29/L) and because every other solution purchases a generator as well, the levelized cost of energy for Ecuador is among the lowest of all the microgrids. The DRC relies heavily on renewables, in stark contrast to Ecuador. This is not, however, due to any abundance of irradiance or wind speeds, it is driven by the high cost of diesel. In effect, the DRC has no good solution and is forced to purchase large quantities of RE in order to meet demand. Unsurprisingly, the DRC has the highest levelized cost of energy. While these represent the extremes, this trend can be observed over the spectrum as diesel prices and available renewables rise and fall. IV. GRID EXTENSION/MICROGRID BREAKEVEN DISTANCE Despite the relationship between renewables and nonrenewables in the microgrid solution, the distance from the existing grid to the proposed microgrid ultimately provides the first line of feasibility. The locations of the communities used

Fig. 6. Brazilian Transmission Lines with 20km wide Corridor

137


existing grids. Figure 6 represents Brazil’s national grid, with a 20km wide corridor surrounding the transmission lines. For smaller systems, such as Standalone 1, the breakeven distances can be as low as 138 meters from the grid. These virtually negligible distances would require a proposed microgrid be nearly underneath existing transmission lines. The orange and yellow lines depict the existing grid infrastructure plus a 20km wide corridor (10km distance to the grid). While the coastal areas are well covered, a disproportionate amount of the interior remains too far to benefit. As the proposed size of the microgrid grows, so does the width of this breakeven corridor, but the corridor growth lags behind drastically. As the microgrid increases by an order of magnitude, the breakeven distance tends to triple. Determining the standard deviation of the breakeven distances gives a high level of certainty for the distance cutoff. The standard deviations are given in Table V. TABLE V. BREAKEVEN STANDARD DEVIATIONS Community I

Community II

Standalone I

Mean Distance (km)

0.833

4.620

0.174

Standard Deviation

0.121

0.945

0.022

Since two standard deviations constitutes 95% of cases, it is safe to assume that if a proposed 5 kWh/day microgrid is at least 1.075 km away from the closest grid connection point, there is a high likelihood that an independent microgrid will be more economical. If that were extended to three standard deviations, there would be over 99% likelihood that a 5kWh/day community 1.196 km away would be more economical. The comparison to grid extension does not include the grid’s generation capacity either. The assumption that there is already sufficient capacity to absorb the proposed community becomes problematic for the utility once multiple extensions are considered. V. CONCLUSIONS The relationship between effectiveness of building a standalone microgrid and tying into an existing grid heavily favors microgrids in developing countries or countries whose existing grids are not far-reaching. With the size of the proposed microgrid and distance from the closest national grid connection point, a highly accurate prediction can be made about the two solutions’ relative economic viability. Breakeven distances of hundreds of meters can be economically feasible, with distances as low as 138 meters seen for three different 1 kWh/day systems as distant as South America to the Middle East. While the internals of the microgrid will vary with location, their costs remain constant enough for this prediction to remain accurate. Taken to only two standard deviations (>95% certainty), the breakeven distance only increases by less than two kilometers for the largest of the three cases. Furthermore, the diesel generation components remains constant throughout

the study because of the evening peak demand. The optimal microgrid could be completely renewable if demand were distributed more evenly. The solutions examined in this paper are optimal given their input parameters, but do not represent the most optimal forms mathematically. The availability of generation and storage components severely restricts the ability to reduce costs and unusable energy. System sizes are forced to be oversized to meet demand. VI. FUTURE WORK The simulations presented in this paper can be made more accurate by obtaining local pricing information. The cost per liter of diesel used were national averages and do not accurately reflect fuel prices seen in remote areas. Other factors, including government incentives, import duties, and plans for future expansion can increase or decrease the microgrid feasibility. The load profiles from Geospatial Toolkit are indicative of classic demand. The evening peak suggests a high number of electric devices being turned on simultaneously, but it is important to note that in the rural setting, many of these devices are absent. Adopting a more accurate load profile could impact these results heavily, hopefully removing the diesel generation requirement. Additionally, understanding the threshold of peak demand for generators appearing in the solution will add to the robustness. REFERENCES [1] J. P. Murenzi and T. S. Ustun, “The Case for Microgrids in Electrifying sub-Saharan Africa,” IEEE Renew. Energy Congr., pp. 1–6, 2015. [2] P. Buchana and T. S. Ustun, “The Role of Microgrids & Renewable Energy in Addressing sub-Saharan Africa’s Current and Future Energy Needs,” IEEE Renew. Energy Congr., pp. 1–6, 2015. [3] U. Deichmann, C. Meisner, S. Murray, and D. Wheeler, “The economics of renewable energy expansion in rural SubSaharan Africa,” Energy Policy, vol. 39, no. 1, pp. 215–227, 2011. [4] National Renewable Energy Laboratory, “NREL Geospatial Toolkit.” Golden, Colorado, 2005. [5] J. M. Lukuyu and J. B. Cardell, “Hybrid Power System Options for Off-Grid Rural Electrification in Northern Kenya,” Smart Grid Renew. Energy, vol. 5, pp. 89–106, 2014. [6] German Agency for International Cooperation, “Pump Price for Diesel Fuel,” World Bank, 2015. [Online]. Available: http://data.worldbank.org/indicator/EP.PMP.DESL.CD. [Accessed: 01-Dec-2015]. [7] “Solar and Wind Energy Resource Assessment,” United Nations Environment Programme. [Online]. Available: http://en.openei.org/apps/SWERA/. [Accessed: 15-Dec2015].

138