CONSIDERATIONS FOR RENEWABLE ENERGY MINI-GRID SYSTEMS FOR ISOLATED AREAS IN UGANDA R. Okou, E. Niwagaba, O. Kyahingwa, M. Edimu, A.B Sebitosi* Department of Electrical and Computer Engineering, Makerere University, P.O.BOX 7062 Kampala, Uganda *Department of Mechanical Engineering, Stellenbosch, University, Private bag X1,7602 Maiteland, South Africa Corresponding author:
[email protected]
Abstract-This paper presents a model for a renewable energy mini-grid system for isolated areas in Uganda designed to provide an alternative to, the costly and emission ridden, thermal generator electricity supply that is currently used in a number isolated and rural areas of the country. It considers a detailed assessment of renewable energy resources including solar and wind energy. The area of study is an island in Uganda, Kalangala that receives an average solar flux of 300W per day on a horizontal surface. Preliminary studies reveal an estimated wind speed potential of 6m/s at 10m. The model is built on the basis of a detailed resource assessment, achieved using an Artificial Neural Networks (ANN) model to accurately predict wind speed and solar irradiance profiles of the island. The model results are benchmarked against ground based Meteorological data, Meteonorm data and ANN software forecasts. The model is used to perform a short term forecast of solar and wind energy for adequate planning, dispatch of power and financial viability of the system. This paper analyzes the energy cost as well as the cost comparison of the thermal generator system to that of the hybrid system. The technical viability is achieved using DIgSILENT power factory, where the existing island grid and PV plant are modelled. Various operational scenarios were created within the software to determine the impact of different photovoltaic (PV) penetration levels. The results show the availability of solar and wind resource, the possibility of solar integration and a realization of cost saving from the implementation of the designed renewable energy mini-grid Keywords: Isolated Mini grids, Artificial Neural Networks, Renewable Energy Integration, PV penetration levels. I. INTRODUCTION Energy is a vital force in the economic growth and development of any society. With the revitalisation of the Ugandan economy, coupled with a rapid increase in population growth, additional demands for energy have been observed. The rate of growth of demand for electricity outstrips the current generation, transmission and distribution capacities, thereby posing a major challenge to industrial growth and economic development. Electrification coverage in rural Uganda is still very poor with less than 6% of the rural population having access to the national grid, whereas the rural people are still an important majority in Uganda with about 88% of the population residing in rural/isolated areas of the country [1].
A mini-grid is a village-scale electrical distribution system served by an isolated generator of up to a few hundred kW in capacity. Power on these grids is often provided by diesel generators, but can be supplied by local, renewable energy sources (RES) such as micro-hydro, solar, biomass or wind. Mini-grids offer an intermediate solution between stand-alone individual home power systems and main-grid connection, and often prove to be more cost-effective and beneficial to the community than either of those alternatives. Renewable energy based mini-grids offer a significant opportunity to increase access to reliable electricity services for rural populations throughout the developing world Currently, there are a few mini grid projects in Uganda that include: Nyagak Power Station (3.5MW), Ngoma (65kW thermal generator), Kalangala (250kW thermal generator), Kisiizi(300kW), Arua, Yumbe, Moroto, Koboko, Kikagati, Buseruka, Isasha and Mpanga that are mostly mini hydros and thermal generator driven networks [4], yet Uganda receives on average, 3200 hours of sunshine per year and a mean solar radiation of 5.1 kWh /m2 per day on a horizontal surface. The average solar flux in some parts of the country based on 24 hours can be as high as 300W/m2 or more [5]. This resource could be used to electrify isolated areas. In fact, a preliminary review on an isolated Kalangala revealed 6.5m/s at 10 meters, resource that could be used to produce electricity for the island [6]. The use of RES as a driver of rural electrification has been demonstrated in other developing countries. Kenya is one of the countries in sub - Saharan Africa where rural electrification driven by RES is rapidly growing. Statistics from the Rural Electrification Authority (REA) in Kenya indicate that, currently, there are 127 rural electrification projects in the pipeline [7]. Another example is in South Africa in Lucingweni which is not connected to its national grid and has 50kW PV panels and 36kW wind generators serving 220 households [8]. The benefits of integrating RES on a mini-grid have been documented. Some of these include: Minimal contribution to global warming [9, 10] Improved public health and environmental quality [11] Inexhaustible energy resource [12] Other economic benefits [13, 14,] More stable energy market [15, 16] This paper considers the conditions in Uganda and investigates the impact of integrating RES, particularly solar and wind, on a mini-grid. The effect of different RES penetration levels on the grid is considered. The paper also
provides a cost analysis comparing the use of RES based supply and the current thermal plants. II. CASE STUDY ANALYZED The district of Kalangala in Uganda is located entirely within the boundaries of Lake Victoria, southwest of Entebbe, in Wakiso District. The coordinates of the town are: 00 18 32S, 32 13 30E (Latitude:-0.3084; Longitude: 32.2250). Kalangala has a total area of 9,066.8sq.km.The land area covers 454.8 sq.km representing only 5% of the total area of the district, while area under water is 8,612sq.km (95%).The perimeter of the District is 387 kilometres long. It encompasses 84 islands, 64 of which are inhabited. The population is estimated to be 5,900 people according to the Uganda census of 2002. Most of the population and commercial activities are focused in small, clustered fishing communities along the shores. The main and largest island houses Kalangala town, local government offices and the district’s secondary schools. The region is only accessible by ferry or boat and with no existence of bridges between the islands. The main commercial industries are fishing, tourism and oil production by BIDCO-Uganda. Subsistence agriculture is practiced by the inhabitants of the islands. The island has some uncommon and unspoiled sandy beaches. The electricity net-work of the island consists of nine transformers: One step-up 500 kVA 0.415/33 kV, five stepdown 50 kVA 33/0.415kVA and three step-down 25 kVA 33/0.415 kVA, 33 kV, distribution network with 9.4 km total length, two generators, 250 kVA and 350 kVA (one standby). The network also consists of two three phase synchronous generators (340 kVA and 250 kVA) generating at 415 V at 0.8 power factor. The generators are connected through a change over switch allowing generators to operate one at a time. Generator 1 operates from 9:00am to 2:00pm while generator operates from 2pm to midnight. The generators use approximately 500 litres of diesel each day costing the government about 36million shillings per month. The two installed generators have the capacity to generate up to 590 kVA at 100% loading but the current power that is being generated is 90 kW at 0.8 p.f because of the small load of the area. The generator parameters are as shown in the Table 1. Table 1: Generator parameters of the existing mini-grid
Parameters Power rating
Generator 1 3φ,340 kVA
Loading Voltage rating Current rating Power factor Frequency Connection type R.P.M Excitation voltage Excitation current
112 kVA 115/200/230/400 V 1707/981/853/491 0.8 50/60 Hz Delta/Wye. 1500 -
Generator 2 3φ, 250 kVA,200 kW 112 kVA 400/230 V 361A 0.8 50 Hz Wye. 1500 35v 3A
III. METHODOLOGY APPLIED In order to effectively carry out this study, site visits and participatory appraisal methodology research, data analysis i.e. resource assessment using neural networks, modelling and simulation of the existing mini grid network using DIgSILENT software and cost optimization using HOMER Energy software was performed. A. Load profile The daily load profile of the island was obtained by connecting a power analyzer on the network and averaged on five days. The results obtained were analysed and the obtained load profile is as shown in Figure 1.
Figure 1: Daily load profile based on the primary side of the StepuP transformer averaged for five days (Sunday-Thursday).
The peak load is 87.2 kW at around 7:45 pm; Peak hours range from 7:00 to 10:15pm, the average load is 51kW for the 15hours of power supply. B. Resource assessment The key step performed included: Studying and interpreting solar and wind real data logged from Namulonge Research Centre. This was used to validate results from the prediction models for Kalangala island and those derived from simulations on weather data software Meteonorm. The analyzed data particularly solar irradiance & wind speed were applied to Neural Network software Neuro-Intelligence as well as to the built Java Neural Networks Prediction model, and used to do a short term prediction i.e. 30minutes-1hour, then benchmarked against the real data and compared to expected data values to attain the error percentages. i. Real Data Analysis Real solar and wind data obtained from the automatic meteorological station of National Agriculture Crop Research Institute (NACRI) in Namulonge, Wakiso District for a period of four years (2008 to 2012) was analysed. The average highest wind speed in the area averaged over the 4 years was 2.2m/s2 at 10m. The highest wind speed observed was between the months of February to April as well as June to July. The highest wind speed in these months’ ranged from 3.22 to 8 m/s2.The lowest wind speed recorded was between January to December ranging from 0.4m/s2 to 0.9m/s2. Wind speed was
highest from 9:00am to 6:00pm due to the prevalent south east (SE) wind and lowest at night/early morning from 12:00am to 8:00am due to the dominant west south west (WSW) wind. The average solar flux was 167.49W/m2 averaged over the four years. The solar irradiance is greatest between 11:00am to 3:00pm although the area receives radiation for 11hours (from 7am to 6pm) with an average solar flux of over 300W/m2 solar radiation per day. ii. Meteonorm Data Meteonorm software is a solar resource prediction tool and was used to predict diffuse, direct and global solar irradiance, temperature and wind speed for Kalangala Island for 2012. The average wind speed at 10m height was 5.88m/s2. Wind speed was highest in the months of January to March (6.1m/s2) and lowest in June to September (5.66m/s2). Wind speed was highest during the morning hours (10:00am-12:00pm) as well as in the afternoon (12:00-5:00pm). Wind speed was lowest at night and early morning. Hence wind speed was fairly steady in the morning but strong in the later part of the day. The average monthly irradiance of global radiation horizontal for Kalangala was 146.67kWh/m2 therefore the average solar irradiance is 4.89kWh/m2/day for 2012. The average air temperature was 21.250C. The solar radiation was strongest between 10:00am to 4:00pm and the island receives solar radiation from 7:00am to 6:00pm thus 11sunshine hours. iii. Solar Data The yearly solar irradiance on a fixed plane, irradiance on optical inclined plane, irradiance on a plane at 900 were obtained and the average values calculated were 4.66kWh/m2/day, 4.1kWh/m2/day, 1.95kWh/m2/day respectively. The daily solar irradiance data was obtained and the average daily global solar flux was 499.2W/m2. The average estimated solar irradiance for Kalangala is above 4.76kWh/m2/day on a horizontal surface, solar flux of above 300W/m2 and the estimated average wind speed at 10m height is approximately 6m/s2.
In the testing window, an Actual vs. Output table is displayed as well as the absolute error (AE), absolute relative error (ARE), probability and match information. Visualization of the information is available in form of graphs (actual vs. output, error graphs) and a confusion matrix for each of the various network sets. Tables 2 and 3 present the results (using data for Namulonge) for the wind speed and solar radiation predictions using the different training algorithms provided by Alyuda Neuro Intelligence. Table 2: Wind Speed (m/s2)
Real Data 0.40
iv. Short Term Forecasts using Alyuda NeuroIntelligence We used the software to achieve the time series prediction through the following steps: Data Set Preparation: Wind and solar data sets were arranged in text files and used as inputs to the NN software Wind speed data for a period of 5 days at 30minutes intervals as well as solar radiation values of a similar pattern, arranged in a text, saved and opened as a new file in the Alyuda Neuro-Intelligence framework. The neural network was then switched to time series mode where the period/look ahead was set and the target established. The look ahead was set to 1 and 2 to achieve a 30minutes and 1hour prediction respectively. Data Analysis and Pre-processing: Once the file was opened, the data was analyzed. Missing, outliers, wrong data or unsuitable data for the neural network were sorted and identified/marked out. Preprocessed data included a column encoded list, column details
and encoded data which must be saved. This was an automatic preprocessing by just clicking the Analyze and Preprocess buttons respectively. Design, Training, Testing the Neural Network: In the design window, we were able to select the design architecture (number of inputs, hidden layers, neurons in each layer), define properties (hidden ylayer activation function, output neuron, output error function, output layer activation function) and carry out an architecture search (exhaustive or heuristic). The best fit architecture was chosen by pressing Activate Best Network button. For both solar and wind, online back propagation, batch back propagation, conjugate gradient descent, quasi Newton method, quick propagation and Levenberg Marquardt algorithms were used to train the various networks in order to obtain the optimal training algorithm to use for short term solar and wind predictions. We set the training condition to stop for a maximum of 500 iterations and an error change of MSE = 0.0001
Real Data 0.40
Real Data 89 Real Data 34
30 MINUTES PREDICTION Conjugate Back Batch Back Gradient Propagation Propagation Descent 0.41 0.65 0.40 1 HOUR PREDICTION Conjugate Back Batch Back Gradient Propagation Propagation Descent 0.43 0.45 0.40 Table 3: Solar Radiation (W/m2) 30 MINUTES PREDICTION Back Batch Back Conjugate Propagation Propagation Gradient Descent 193.65 195 129.69 1 HOUR PREDICTION Back Batch Back Conjugate Propagation Propagation Gradient Descent 168.28 561.825 142.843
Quick Propagation 0.49 Quick Propagation 0.39
Quick Propagation 201.69 Quick Propagation 188.24
Using Alyuda neural network software, the solar prediction was not good, so we generally analysed the wind prediction and the results obtained were graphed as shown in Figure 2. Real data vs NN Output for various Algotithms 2.5 wind speed m/s2
Real Data 2 back propagation
1.5 1
Conjugate Gradient Descent Quick Propagation
0.5
11:30 12:00 12:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 4:30 5:00 5:30
0
Batch Propagation
Figure 2: Graphical representational of Alyuda NN Actual vs. Output for different training algorithms.
Resource prediction for solar radiation was most inaccurate; wind was predicted within reasonable margins. It was then made clear that prediction depends on the Training Algorithm. From the various training sets used, Back Propagation Algorithm most convenient for training the wind and solar data we had. Prediction accuracy decreased with increase in look ahead/ forecast time.
v) Neuroph Java Neural Networks prediction Model The Neural Network model was built on top of Neuroph Java Neural Network Framework and developed in Net: Beans (IDE). Neuroph was chosen because it was free, well documented, light weight and could be used to develop common neural network architectures. Most of the neural network functions were accessed from its’ well designed, open source Java library which contains a number of basic classes (training, ) which correspond to basic NN concepts. The model can be used for time series prediction as described in the process flow below: a. Getting Inputs from the file The model allows input files of either excel or text file but not comma separated files (csv) although text files are preferred as they are easy to parse (getting content from the file) and csv files were difficult to parse and therefore that format cannot be read. The public class file handler in neuroph was a written function that was used to open and read each line of the input files. b. Normalizing the inputs The inputs are scaled to a range of [0, 1] in this particular model using Equation (1). )∗ = +( − ⎧ ⎫ (1) ( − ) ⎨ = −D ⎬ ⎩ ⎭ − Where Imin Was set to 0, I max was set to 1; Dmax is the highest value in the range of inputs; Dmin is the lowest value in the range of the input values.
c. Setting the Network Architecture The Multilayered Perception in neuroph was defined with the architecture parameters. Transfer Function (TANH), inputs (4) and outputs (4) and hidden layers determined using the pruning algorithm to determine the best fit value, 2N +1 where N is the number of inputs. Therefore the best value for the hidden layer neurons was 9, however this number was varied as follows; 2,3,15 and 31. The neuroph model was then programmed to consider 5 network topologies by looping through the hidden layer. [4-2-4], [4-3-4], [4-9-4], [4-15-4], [431-4]. d. Setting the inputs as Training inputs & set training parameters Parameters are set in such a way that the training set should not be less than 20 elements. The training inputs are then formatted (into rows and comma separated) such that neuroph can use them for training. The perception parameters set include: training algorithm; back propagation, learning rule, maximum error (0.0001), learning rate (0.01) and maximum iterations. e. Training and saving the neural Network In the Train java class, the training data set from the Training Data class (where it was set as a neuroph input to the training set) is trained using MLP Supervised learning where by the neural network learns from the target/ desired output done by correctly mapping the inputs to the output so that the model can learn the data pattern and relationship. 1, 2, 3, 4-> [5, 6, 7, 8] 2, 3, 4, 5-> [6, 7, 8, 9] 3, 4, 5, 6 -> [7, 8, 9, 10] 4, 5, 6, 7-> [8, 9, 10, 11] 5…….(n-4)->[(n-3)……n] The network is trained using back propagation algorithm and each of the NN topologies is trained and their output saved. f. Testing the Network Testing the NN involves a computation output error which is the difference between the expected output and the NN output. The NN output is loaded (the 5 preciously saved outputs obtained) and the error computed in each case. g. Displaying the NN Output The Neural Network’s 4 outputs are displayed in as predictions for the set inputs.4.6.9 User Interface for Neural Network Prediction Model. For simplicity and applicability of the model, a Swing Graphical User Interface (SGUI) was used to design a user interface for the model. The Swing GUI contains symbols, tools and panels necessary to design and create a GUI in Net Beans. The training pallet gives you an opportunity to select the data you want to use in training, to set the learning rate, number of iterations, the maximum error, number of inputs and number of outputs IV. MODELING AND SIMULATION OF KALANGALA MINI-GRID The existing mini-grid was constructed by Uganda Electricity Distribution Company Limited (UEDCL). Using the data parameters obtained from UEDCL-Kalangala branch, the minigrid was simulated using DIgSILENT software. The mini-grid simulated in DigSilent is presented in Figure 3.
B. Solar penetration levels Solar penetration levels were calculated according to Equations (2) and (3) = (2)
10%
=
(3)
e.g. 10% Solar penetration = 10 kW 20% Solar penetration = 22.5 kW 50% Solar penetration = 90 kW 75% Solar penetration = 270 kW
Figure 3: Grid network for the existing mini-grid system
For each penetration level, an operation scenario at for LS was created and saved in the simulated grid network. An AC load flow was carried out and results were imported into excel and analysed. Figure 4 presents the results for the case one (load less than supply) and PV being integrated at a point near the greatest load in the network
The colour coded voltage levels shown in Figure 3 are described in Table 4. Table 4: Key showing voltage levels for each colour on the grid
COLOUR Pink Black
VOLTAGE LEVEL/kV 33 0.415
An AC load flow was then executed within the software basing on the Newton Raphson Iterations; the system report obtained showed that the voltage values for different bus terminals were the same as those obtained from KalangalaUEDCL branch. All the voltages ranged between 0.95 to 1.05pu which is the agreed voltage range for the Kalangala mini-grid network. This showed that the mini-grid network was simulated with minimal errors hence good enough to use to conduct the analysis. A. Creating different load scenarios Three different load scenarios were then created in DIgSILENT before the integration of solar PVs onto the grid. These scenarios were later used to clearly understand the impact of solar integration. The load scenarios considered are; Case one: Load less than supply; this is the current situation. The load is 0.0872MW and the generation is 0.09MW Case two: Load equal to supply; A 3.11% difference between the load and supply was applied to each load as a percentage increase so as to get to a situation of load equal to supply. Case three: Load greater than supply; Based on the developments taking place on the island and past increase in load, it was forecasted that within the next 5-8 years the load would have increased not more than 300%, therefore a 300% increase was applied to each load and new load values were updated in the data manager.
Figure 4: Voltage (p.u) at selected bus terminals for different PV penetration levels
Figure 4 indicates that the voltage increased at the bus terminals when solar was integrated onto the network. All the voltages at the bus terminals remained in the agreed voltage range of 0.95 to 1.05p.u V.
COST ANALYSIS
Cost analysis and optimization as well as sensitivity analysis 0f the system was done using Homer Energy analysis with the aim of closely matching the cheapest source of energy production. The following system scenarios were considered: Cost of Existing Grid Cost of Solar and Diesel Generator as back up Cost of Solar and Grid Connection A. Cost of existing grid The system was designed using Homer Energy analysis based on Kalangala Island’s load profile for a year. Figure 5 shows the schematic diagram of the existing grid modelled using HOMER.
(200kW, 89.6kW) and fuel consumption values (58.3L/hr, 26L/hr) respectively. iv. Schedule Forced on weekday and weekend schedule for this generator was set from 2:00pm to 12:00pm. v. Emission and Advanced Settings Default values were maintained.
Figure 5: Screen shot of Schematic Diagram of Existing grid using HOMER
The components are described below: a. 340 kVA generator i. Costing A 272 kW generator costs $45,000 initially, $38,000 to replace at the end of its life, and operation and maintenance costs $1/hr. Sensitivity cost variations considered the initial cost to go as low as $30,000 and as high as $50,000. The replacement cost sensitivity values were linked to the capital cost multiplier. ii. Generator Sizes Considered The generator sizes considered ranged from 0 to 272kW used to search for the optimal system. The generator was labelled 340kVA Generator, abbreviated as GEN1, type set to AC, life time set to 25,000hrs (as its slip is 1500rpm) and the minimum load ratio of not less than 30%. iii. Fuel The type was set to Diesel, Intercept Coefficient was 0.0026L/hr/kW rated, slope set to 0.2934L/hr/kW output as per the fuel curve calculator set with the output power values (272kW, 89.6kW) and fuel consumption values (80.512L/hr, 27L/hr) respectively. iv. Schedule Forced on Weekday and weekend schedule for this generator was set from 9:00am to 2:00pm. v. Emission and Advanced Settings Default values were maintained. b. 250 kVA generator i. Costing The 200 kW generator costs $30,000 initially, $26,000 to replace at the end of its life, and $1 per hour for operation and maintenance. Sensitivity cost variations considered the initial cost to go as low as $20,000 and as high as $45,000. The replacement cost sensitivity values were linked to the capital cost multiplier. ii. Generator sizes considered They ranged from 0 to 200kW used to search for the optimal system. The generator was labelled 250kva Generator, abbreviated as GEN2, type set to AC, life time set to 25,000hrs (as its slip is 1500rpm) and the minimum load ratio of not less than 30%. iii. Fuel The type was set to Diesel, Intercept Coefficient was 0.001072L/hr/kW rated, slope set to 0.2926L/hr/kW output as per the fuel curve calculator set with the output power values
c. Primary Load The daily load data we obtained using the power analyzer connected to the generators to obtain the day profile for the 15hours of electricity supply as shown in Table 5. Table 5: Hourly load data
HOURS OF THE DAY 9:00-10:00 10:00-11:00 11:00-12:00 12:00-13:00 13:00-14:00 14:00-15:00 15:00-16:00 16:00-17:00 17:00-18:00 18:00-19:00 19:00-20:00 20:00-21:00 21:00-22:00 22:00-23:00 23:00-24:00
DEMAND(kW) 23.525 34.925 34.7 40.25 41.325 45.275 44.675 44.15 45.35 54.225 63.4125 84.4 75.45 60.15 49.46
The annual baseline load profiles were therefore synthesized by HOMER. The primary load was labelled Kalangala Island, type was AC, the random variability day to day parameter was 15% and time step to time step was 20%, the scaled annual average was 734kWh/day with a peak of 151Kw used for simulation. d. Resource Inputs The economic inputs: Interest rate was varied for 12% and 15% as per the current Bank Of Uganda interest rate at 12%. Diesel inputs: A litre of diesel fuel costs about USD $1.31. The several design alternatives were considered by HOMER and the optimal simulated existing grid system parameters were as shown in Table 6. From the HOMER Simulations, the LCOE is about UgX.1700 per kWh - using the two diesel generators. This is close to the actual UgX.1200/- (unsubsidized) per kWh spent by UEDCL for this Island. The Fuel consumption is 89,693L/yr costing the utility $ 117,290.8 (about UgX.304.9m) annually and the Operating cost of the system is about $166,509 (UgX.432.9m)/yr.
Table 6: Existing Grid Specification and cost summary
Parameters Hours of Operation (Hr/yr) Mean Output (kW) Electrical Production (kWh/yr) Fuel Consumption (L/yr)
340kVA Diesel Generator 1,825
250kVA Diesel Generator 3650
81.6 148,920
66.1 241,128
19,923
69,770
Specific Fuel 0.335 0.279 Consumption (L/kWh) COST SUMMARY Net Present Cost (USD) $ 1,371,955 Leveled Cost of Energy $ 0.654/kWh (LCOE) Operating Cost (USD) $166,509/yr
B. Cost of hybrid system HOMER simulation software was used to design the system with energy sources solar and the two diesel gen sets. It also consisted of batteries, converter and the primary load as shown in Figure 6.
Group) Interest rates: 10% and 15% and BOU of 12%. Several design alternatives were considered and the optimal system specifications were as follows: a. Optimal system configuration results The most cost effective system, i.e. the system with the lowest net present cost, is the solar-generator set-batteryconverter set-up with the solar operating under a cycle charging (CC) strategy (a dispatch strategy whereby the solar operates at full output power to serve the primary load: surplus electrical production goes towards lower-priority objectives). A solar system of 5MW, 272kW generator, Battery of 16,000 Surrette 6CS25P, Inverter and rectify of 100kW. For this setup, the total net present cost (NPC) is $745,910, the cost of energy (COE) is 0.301$/kWh. However due to the small existing load, slow economic growth rate and low electricity load growth of the Island, 5MW solar system was not considered. Recommended system configuration results obtained are as shown in Tables 7 and 8. Table 7: Hybrid System Architecture
PV Array 340kva Diesel Generator Battery Inverter Rectifier Dispatch Strategy
1500kW 272kW 8000 Surrette 6CS25P 1000kW 1000kW Cycle Charging
Table 8: Recommended Hybrid System specifications and cost summary
PV Specifications Rated Capacity- 1500kW Mean Output- 227kW Mean Output per day – 5448kWh/d Maximum output- 1262kW Figure 6: Screen shot of HOMER diagram for the hybrid PV gen-battery-converter set-up
a. Solar PV system The cost of the PV system is 3$/Watt and Operation & Maintenance cost is $0/yr. b. Diesel inputs Diesel fuel was costed at the present value of USD $1.31/L. Fluctuation in fuel prices was catered for. The values considered included $ 1.3, $ 1.31 and $ 1.4. c. Battery Bank Surrette 6CS25P battery was chosen with Capital costs varied between $1200 -$1375 and we set a minimum battery life of 4years. d. Converter The capital and replacement costs for the converter were taken as USD1200/kW and lifespan as 15 years. C. Economic Inputs A sensitivity analysis was performed on the interest rates. According to the Central Bank of Uganda, the real interest rate to Uganda was 12% in 2013. Therefore a sensitivity analysis was applied on the interest rates as follows. (The World Bank
340kVA Diesel Generator Rated Capacity- 272kW Mean Output- 81.8kW Maximum Output- 203kW
Hours of Operation1,825hr/yr Fuel Consumption50,042L/yr COST SUMMARY Net Present Cost $585,839 Levelized Cost of Energy $0.279/kWh (LCOE) Operating Cost $68,317/yr
The total net present cost (NPC) for this setup is $585,839 the cost of energy (COE) is 0.279 $/kWh. D. Cost of Solar and Grid Connection The grid connection is shown in Figure 7. The component descriptions follow: a. Grid Inputs The cost structure of the grid inputs was defined by setting the cost of buying power from the grid to 0.462$/kWh (UgX.1200), and the demand rate set to 1.292$/kW/month. The other system inputs included: PV-1500kW, Grid1000kW..
compared to the actual USD $ 0.654 (UGX.1200/kWh) spent using the generator system and this is further reduced to $0.003/kWh when the PV system is connected to the grid. The GOU (REA) will be able to save $0.1825/kWh, about $98,192/yr = UgX.255.3m/yr on operating cost with the implementation of the hybrid solar/diesel system. Further, the GOU (REA) will be able to save over USD $74,000 (UgX190m/yr) on (fuel and 0 & M/yr) with the implementation of the hybrid solar/diesel System. Figure 7: Screen shot for Schematic Diagram of the Hybrid system connected to the grid
Without considering the cost of grid connection, the cost summary for the solar and grid system is presented in Table 9. Table 9: Cost summary for PV-Grid Connected system
Net Present Cost (USD) Levelized Cost of Energy (USD) Operating Cost
$5,659 $0.003/kWh $81/yr
Once the grid is extended to the Island with the PV system in place, the Levelled COE will significantly fall to 0.003$/kWh. Table 10 shows the impact of grid extension on different cost components. Table 10: Influence of grid extension of costs
Capital Cost($/km)
650,000
Price of submarine cable extension per km
Grid power price($/kWh)
0.18
Tariff from generation(12cents) to transmission(3cents) and losses(3cents)
O&M Cost($/yr/km)
160
The LCOE obtained is $0.003/kWh at a break even grid extension distance of -0.642km. b. Cost comparison between the existing grid, hybrid system, solar and grid connection The cost of installing and running the hybrid system (Diesel and Solar) as well as grid extension was compared to that of the simulated existing grid system using generators and the grid extension and presented in Table 11. Table 11: Cost Comparison of the 3 systems
1. 2. 3.
System
Initial Cost ($)
LCOE ($/kWh)
Total NPC ($)
Base case (Existing generator grid) Solar + 272kW Gen Solar + Grid Connection
66,000
0.654
1,371,955
50,023 5,023
0.279 0.003
585,839 5659
The COE is significantly lowered using the Hybrid solar diesel system to USD $ 0.279/kWh (UgX.725.4/kWh)
VI. CONCLUSION A. Power quality From the study carried out on simulation of the existing mini-grid in DIgSILENT, integration of solar can affect the power quality in terms of voltage increase of the network. This impact becomes more significant with higher penetration levels. At high solar penetration levels the voltage increases, however for the Kalangala mini-grid network, the increase is not that significant in that even at higher penetration levels, the voltage remains in the required range of 0.9 to1.05 p.u. B. Resource prediction From the neural network study that we carried out, back propagation offers a better resource prediction for short term (30 minutes to 1 hour). C. Cost The Government of Uganda would save if it considers a hybrid for Kalangala Island and this could be applied to all other mini-grid systems in the country. Wind was not considered as a lot of considerations like locating wind turbines must be considered while locating a wind turbine, the “Java neural network prediction model” that we developed could not predict wind accurately and hence we resorted to designing considering solar alone. As DIgSILENT was not able to provide real time load flow analysis with real time load and real time solar radiation, hence it was not possible to do a real time voltage analysis, it is therefore necessary to use such software in order to study real time voltages changes due to solar PV integration onto the mini-grid. REFERENCES [1] www.ugandainvest.com, “overview of Uganda’s energy sector” September, 11th, 2012. [2] ERA-Uganda “Developments and Investment Opportunities in Renewable Energy Resources in Uganda”, May 2009. [3] ERA-Uganda “independent grid projects” September, 11th, 2012. [4] 7m-magazine “Progress of rural electrification in Uganda” September, 11th, 2012. [5] Muhammad Ziad “Minimum Network Element 3, Alternate Energy Solutions for Minne3 SOLAR, WIND, HYBRID” (2010, January). [6] Ben Sebitosi, “Mekonorn Weather Data” simulated on June, 28th, 2012, unpublished. [7] Camco UNDP Climate Parliament, “Mini-grid Toolkit field study report for Kenya, Mozambique and Zambia” [8] http://www.energy.gov.za/files/esources/renewables/r_hy brid.html reviewed on September, 13th, 2012.
[9]
Working Group III of the Intergovernmental Panel on Climate Change [O. Edenhofer, R. Pichs-Madruga, Y. Sokona, K. Seyboth, P. Matschoss, S. Kadner, T. Zwickel, P. Eickemeier, G. Hansen, S. Schlömer, C. von Stechow (Eds)].Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, Intergovernmental Panel on Climate Change (IPCC) “IPCC Special Report on Renewable Energy Sources and Climate Change Mitigation” 2011,1075 pp. (Chapter 9). [10] National Renewable Energy Laboratory (NREL). “Renewable Electricity Futures Study.2012” Volume 1, pg. 210. [11] Machol, Rizk “Economic value of U.S. fossil fuel electricity health impacts. Environment International” 2013, 5.
[12] “Renewable Electricity Futures Study”2012. [13] The Solar Foundation, Solar Energy Industries Association (SEIA); “Solar Industry Data, National Solar Jobs Census 2011” 2011. [14] Navigant Consulting “Job Creation Opportunities in Hydropower.” 2009 [15] Geothermal Energy Association “Green Jobs through Geothermal Energy.” 2010. [16] SEIA “Solar Market Insight Report” 2012. [17] AWEA “Federal Production Tax Credit for Wind Energy” 2012 [18] Unger, David J. “Wind-The Christian Science” 2012, November 19.
-
