Solar irradiance forecasting and energy optimization ...

5 downloads 0 Views 2MB Size Report
increase in the need of solar irradiance forecasting for both solar PV applications and Passive Solar Architectural buildings. First, solar irradiance forecasting ...
Solar irradiance forecasting and energy optimization for achieving nearly net zero energy building A. Naveen Chakkaravarthy, M. S. P. Subathra, P. Jerin Pradeep, and Nallapaneni Manoj Kumar

Citation: Journal of Renewable and Sustainable Energy 10, 035103 (2018); doi: 10.1063/1.5034382 View online: https://doi.org/10.1063/1.5034382 View Table of Contents: http://aip.scitation.org/toc/rse/10/3 Published by the American Institute of Physics

Articles you may be interested in Carbon footprints for the supply of electricity to a heat pump: Solar energy vs. electric grid Journal of Renewable and Sustainable Energy 10, 023701 (2018); 10.1063/1.4997306 Japan, the new “El Dorado” of solar PV? Journal of Renewable and Sustainable Energy 10, 020401 (2018); 10.1063/1.5024431

JOURNAL OF RENEWABLE AND SUSTAINABLE ENERGY 10, 035103 (2018)

Solar irradiance forecasting and energy optimization for achieving nearly net zero energy building A. Naveen Chakkaravarthy,1 M. S. P. Subathra,1 P. Jerin Pradeep,1 and Nallapaneni Manoj Kumar2,a) 1

Department of Electrical Sciences, Karunya Institute of Technology and Sciences, Coimbatore 641114, Tamil Nadu, India 2 Faculty of Electrical and Electronics Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia (Received 11 April 2018; accepted 23 April 2018; published online 1 June 2018)

Solar energy and the concept of passive solar architecture are being increased in several areas to attain the net-zero energy concept. This paved the way for an increase in the need of solar irradiance forecasting for both solar PV applications and Passive Solar Architectural buildings. First, solar irradiance forecasting was done with 131 400 data sets (1-h data for 15 years) which was split into monthly mean for every year. This model was evaluated by forecasting the post-consecutive years one by one with the pre-consecutive years which includes the pre-forecasted years. This model was shown to have RMSE values of 11% to 24% for various seasonal forecasting using the Random Forest Algorithm in WEKA, which gave the annual irradiance results nearer to the PV Sol energy forecasting results. The R-value was in the range of 0.8 to 0.9 for various seasons which is good. Building Energy Optimization was carried out using BEopt 2.8 software designed by NREL. The chosen building was set to the standard parameters in India, and then, the optimization was done with various customized parameters and systems available in India to reduce the energy consumption from 192.2 MMBtu/yr to 109.1 MMBtu/ yr with a 7 kW Solar PV System to attain the net-zero energy concept. Published by AIP Publishing. https://doi.org/10.1063/1.5034382

I. INTRODUCTION

There has been a rapid increase in solar passive architectural buildings in the last few years to achieve the net-zero energy concept. Due to the dynamic change in solar radiation, reliable energy generation forecasting is necessary for grid operation in the case of solar energy generation and also for passive solar architectural building design for the optimal thermal performance of buildings. It is practically impossible to install radiation measuring instruments at every site due to the cost and inaccuracy of measurements on a large scale. Hence, the solar radiation data are made available only for few locations in the developing countries.1–4 So, it is necessary to arrive for different methodologies to estimate the solar irradiation by the commonly available weather data. However, using the data of parameters such as temperature, humidity, and wind speed to predict the solar radiation for an extended period is not so accurate as time plays the leading role in predicting the future radiation in the long term. Hence, time series data can give more accurate long-term results. In this paper, three algorithms were compared with the time series Global Horizontal Irradiation (GHI) data, and the best algorithm with the lowest error was used to predict the future values. The algorithms used are linear regression,5 SMO regression,6 and Random forest.7 This prediction is used to calculate the annualized energy production, cost, and CO2 emission for PV applications and also for the passive solar architectural design. To achieve the net-zero energy concept, the building materials and parameters should a)

Author to whom correspondence should be addressed: [email protected]

1941-7012/2018/10(3)/035103/12/$30.00

10, 035103-1

Published by AIP Publishing.

035103-2

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

be optimized like a passive solar architectural building. D’Agostino and Parker in their paper8 designed a framework for the cost-optimal design of nearly zero energy buildings (NZEBs), and Rhodes et al. in their paper9 did a survey on residential building energy auditing using BEopt software. Hence, using the optimal design parameters for the climate in Chennai and the PV system, the building is designed in such a way to reduce the energy consumption for achieving the nearly net-zero energy concept. II. METHODOLOGY A. Solar irradiance forecasting

The methods implemented here use the average monthly GHI for each year to predict the GHI value for the next consecutive years. WEKA stands for the Waikato Environment for Knowledge Analysis, which is developed by the University of Waikato. WEKA is a one-stop collection of machine learning algorithms for decoding many real-world data mining problems. It contains a lot of tools for data pre-processing, classification, regression, clustering, association rules and visualization. The input variable selection is the first step in developing the solar radiation models.10 For this methodology, hourly data for fifteen years were taken from the NREL database for Chennai (12.9814 N, 80.2432 E). A total of 131 400 data sets were sorted monthly, and an average value for each month of each year is tabulated in Table I. The prediction step is given as 1, and the year was presented at the time stamp for the forecasting. The data sets were then divided into four seasonal data sets and given as input to WEKA. The data sets were tested using three different time series forecasting algorithms, and the errors TABLE I. Input data for forecasting. Year

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

2000 2001

5.425 5.097

5.498 6.355

6.819 6.666

6.925 6.795

6.616 6.651

5.365 5.569

5.421 5.874

5.550 5.457

5.967 5.648

4.868 4.801

4.681 4.187

4.726 4.461

2002

4.951

5.963

6.931

7.320

6.490

5.801

5.750

5.447

6.141

4.681

4.360

4.850

2003 2004

5.403 5.478

6.142 6.173

6.826 6.876

7.032 6.896

5.850 5.877

6.141 6.206

4.965 5.495

5.390 5.705

5.734 5.454

5.040 4.862

4.691 4.809

4.624 4.955

2005

5.335

6.286

6.733

6.247

6.572

6.051

5.566

5.821

5.674

4.714

3.820

4.126

2006 2007

5.291 5.334

6.355 6.041

6.658 6.828

6.812 6.891

6.383 6.644

6.233 5.421

5.760 5.595

5.675 5.533

5.558 5.789

4.923 4.939

4.763 4.842

4.776 4.271

2008

5.217

6.115

6.048

6.956

6.720

5.862

5.870

5.183

5.807

6.133

5.678

5.396

2009 2010

5.485 5.145

6.342 6.136

6.479 6.823

6.789 6.942

6.296 6.270

6.535 5.475

5.758 5.303

5.931 5.238

6.007 5.551

5.614 4.873

4.202 4.525

4.196 3.789

2011

5.336

6.097

6.985

6.751

6.821

6.202

5.572

5.664

6.114

5.190

4.500

4.497

2012 2013

5.505 5.491

6.227 6.002

6.701 6.755

6.966 6.805

6.380 6.417

6.110 6.014

5.665 5.080

5.854 5.601

5.569 5.644

4.883 5.031

5.123 4.995

4.649 4.625

2014

5.350

5.913

6.876

7.023

6.519

6.129

5.608

5.606

5.955

5.005

4.307

4.197

TABLE II. Evaluation of training data. Spring

Summer

Fall

Winter

Algorithm

MAE

RMSE

MAE

RMSE

MAE

RMSE

MAE

RMSE

Linear regression SMO regression

0.1983 0.1115

0.2900 0.2135

0.2702 0.1781

0.3039 0.2607

0.1383 0.0527

0.1619 0.1277

0.3071 0.1739

0.3372 0.3354

Random forest

0.0925

0.1176

0.1483

0.1799

0.0598

0.0776

0.2001

0.2104

035103-3

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

TABLE III. Evaluation of testing data. Spring

Summer

Fall

Winter

Algorithm MAE

RMSE

MAE

RMSE

MAE

RMSE

MAE

RMSE

Linear regression

0.1650

0.2536

0.5503

0.6274

0.2150

0.2315

0.5753

0.8067

SMO regression

0.1065

0.2559

0.2032

0.3210

0.4273

0.4531

0.7351

0.8064

Random forest

0.0718

0.1170

0.1171

0.1249

0.2097

0.2412

0.2001

0.2104

TABLE IV. Forecasted values for Future 25 years (2016 to 2040) using the random forest algorithm. Year

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

2016 2017

5.3654 5.4002

6.0354 6.0401

6.781 6.7724

6.8613 6.878

6.4777 6.4819

5.9969 6.0149

5.4966 5.5251

5.602 5.5903

5.8193 5.8178

5.0865 5.0626

4.6839 4.6914

4.4356 4.4523

2018

5.3957

6.0401

6.7364

6.8734

6.4717

6.0178

5.5242

5.6033

5.8157

5.0726

4.6875

4.4831

2019 2020

5.3934 5.3963

6.0337 6.0329

6.7538 6.754

6.8837 6.8852

6.4674 6.4609

6.004 6.007

5.5205 5.5202

5.5845 5.5952

5.8637 5.8622

5.0738 5.071

4.7008 4.684

4.4828 4.4766

2021

5.3919

6.0357

6.7536

6.8817

6.4613

6.0038

5.5224

5.5935

5.8575

5.069

4.6844

4.477

2022 2023

5.4162 5.4275

6.0306 6.0457

6.7456 6.4925

6.904 6.9271

6.4426 6.4203

5.9508 5.9328

5.5247 5.5368

5.5975 5.598

5.8778 5.8864

5.0753 5.3298

4.7063 4.958

4.4976 4.7018

2024

5.4538

6.0303

6.3388

6.9401

6.4068

5.8717

5.5556

5.6407

5.9326

5.2287

5.1157

4.627

2025 2026

5.4148 5.4307

6.0297 6.0255

6.334 6.1121

6.9231 6.9779

6.4282 6.4063

5.7417 5.6401

5.5757 5.6129

5.6176 5.6033

6.1991 6.4132

5.1721 5.3717

5.0784 5.3088

4.5164 4.6994

2027

5.4813

6.0328

5.7591

7.0338

6.3339

5.5063

5.6143

5.6472

6.4482

5.5177

5.8158

4.9129

2028 2029

5.5433 5.6101

5.9151 5.875

5.5213 5.1159

7.1006 7.1565

6.2585 6.1563

5.2869 5.0491

5.596 5.6219

5.6836 5.7415

6.7215 6.8335

5.6709 5.8611

6.2405 6.6937

5.0381 5.1828

2030

5.5523

5.9267

5.4112

7.0456

6.231

5.2183

5.5931

5.7221

6.8427

5.7072

6.2522

5.0148

2031 2032

5.4723 5.5941

5.9274 5.8945

5.4647 5.4021

6.9677 7.0029

6.3498 6.3622

5.2947 5.3144

5.2847 5.0445

5.5684 5.6311

5.6384 4.924

6.6199 7.5551

6.0507 6.1932

4.9709 5.2581

2033

5.8412

5.7704

5.4997

6.9785

6.3624

5.211

4.9088

5.6595

4.914

7.11

6.1524

5.2581

2034 2035

5.6762 5.4153

5.8528 5.856

5.4968 5.6252

6.9849 7.1979

6.3232 6.3484

5.2808 5.277

5.0603 5.1856

5.6419 5.6552

5.5553 6.0207

6.9175 6.9517

6.0936 6.0202

5.2151 5.1022

2036

5.5579

5.8549

5.7897

7.112

6.5397

5.2745

5.1698

5.6359

5.6925

6.9177

6.0219

4.9743

2037 2038

5.5718 5.4217

5.8474 5.9131

5.9224 6.121

7.0435 6.9707

6.4193 6.2306

5.2268 5.2606

5.1498 5.2793

5.6471 5.655

5.704 5.7646

6.8276 6.8268

6.0604 6.0596

5.0288 5.0308

2039

5.2814

6.0124

6.1314

6.9802

6.3913

5.2481

5.3187

5.8102

5.81

6.5261

6.1231

5.1027

2040

5.4266

5.9424

5.9233

7.0148

6.3881

5.6295

5.2039

5.5181

5.752

6.7469

6.081

5.0381

are tabulated in Tables II and III. The RMSE and MAE errors are comparatively less when the random forest algorithm was used for all seasons as seen in the tables. Hence, the random forest algorithm was used to predict the future GHI data for 25 years from 2016 to 2040. To increase the accuracy and reduce the RMSE error, the prediction was done for every month of every year giving the previous values as input for the consecutive iterations. The results of the predicted GHI values are tabulated in Table IV. 1. Correlation and sensitivity analysis

A sensitivity analysis is a technical methodology for analyzing and studying the behavior of a model and assessing the significance of every input parameter on the values of the output variable of the model. The primary objective of the correlation and sensitivity analysis is to find out how the output data are influenced by the change in the input data. With the hands on

035103-4

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 1. Correlation factor for overall prediction.

TABLE V. Correlation factor. R2 value

R value

Overall (average of all months yearly) Spring

0.8252 0.7055

0.9084 0.8399

Summer

0.7428

0.8618

Fall Winter

0.6919 0.6406

0.8318 0.8004

Season

this type of analysis, it is possible to evaluate which input parameter is more significant than others or the most significant or the least significant for global solar radiation prediction. In this part, to find out the R-value, a correlation analysis was performed, and regression plots were developed. The R-value for the overall actual and predicted values was 0.9084, and for the seasonal prediction, the correlation factor ranged from 0.8004 to 0.8618. The regression plots are shown in Fig. 1, and the correlation factor is tabulated in Table V. B. Energy and performance analysis of the PV system

The simulation of the chosen PV system in Chennai was done using PV Sol software. The simulation was done to compare the energy forecasting results done using WEKA’s random forest algorithm with the forecasted results of PV Sol software. PV Sol Premium 2018 was used to perform a real-time simulation of the PV systems with a layout model of the PV system installed or to be designed. It takes the inbuilt weather data for the selected place and runs the simulation based on the design parameters given. The selected PV system was a 7.5 kW grid-connected system geographically located at 12.9814 N, 80.2432 E. The energy generated from this system is only used for the VRF airconditioners and a water pump. The total number of panels used here is 60 which are polycrystalline panels from the manufacturer UPV of 125 Wp rating, and the layout is shown in Fig. 2. All the panels are oriented south with a tilt angle of 30 . The inverter used is the 8.5 kVA Fronius SYMO solar grid tied inverter, and the modules are mounted by the traditional roof-top rack mounting system.

035103-5

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 2. PV system roof-top layout.

Energy production forecast specific to each month is shown in Figure 3. The results obtained from the PV Sol simulation show that the annual irradiance onto the horizontal surface was 1941.8 kWh/m2/year as shown in Fig. 4. The specific annual yield of the system was 1476.96 kWh/kWp with the performance ratio of 78.9%. The grid feed-in in the first year including the module degradation was 11 020 kWh/year, and without the module, degradation is 11 077 kWh/year. With the radiation results obtained from the WEKA prediction methodology, the annual energy generated from the radiation over a year was calculated using

FIG. 3. Energy production forecasting.

035103-6

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 4. Annual Irradiance per module. TABLE VI. Technical parameters and result comparison. Technical parameters

PV Sol

BEopt

WEKA

Yearly radiation (kWh/m2) Annual average energy generation (kWh/year)

1941.8 11 077

… 12 177

2121 11 923

PV capacity (kWp)

7.5

7.5

7.5

Performance ratio (%) Yield factor (kWh/year/kWp)

78.9 1477

7.5 1624

75 1590

Capacity factor (YF/8760)

16.86

18.53

18.15

E ¼ A  r  I  PR; where E is the energy in kWh, A is the total solar panel area in m2, r is the solar panel yield in percentage, I is the annual average irradiance without shading, and PR is the performance ratio. The energy generated was 11 923.4 kWh/year with the default performance ratio of 75%.11,12 Then, the energy generated by the PV system simulated using building energy optimization software (BEopt) is 12 177 kWh with the performance ratio of 75%. The comparison of technical parameters and the results of three software programs are tabulated in Table VI. C. Economic and CO2 mitigation assessment

The economics of this PV system was compared with the two proposed cases, one with the full initial cost and the other with the zero initial value (monthly payment for produced units), which is also known as the power purchased agreement. The cost of 60 poly-crystalline PV modules is `330 000, and the price for three 8.2 kVA Fronius SYMO are `429 000. The lightning protection rod and the earthing requirements used here are locally manufactured and the cost of which is `35 000. Hence, the total cost of the PV system including the installation cost of `35 000 will be `829 000. Case 1: Full initial cost invested in installation and recurring investments: The annual saving and the payback period were calculated using the following formulas13,14 where the feed-in tariff rate is `3.8 and a lifetime of the system is 25 years:

035103-7

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

Annual savings Payback period

¼ ¼

Annual energy Initial investment

 /

Feed-in tariff Rate Annual savings

Returns for

¼

Annual savings



25

The annual saving calculated on an average was `33 385 per year, and the payback period was 14.6 years. The economic saving return for 25 years is `834 625. Case 2: Zero initial cost (monthly payment for produced units): The total initial cost without the recurring investments of the inverters in the 9th and the 18th year is `543 000. The recurring expenditure of the inverters for two replacements within the lifetime of the PV system was `286 000 which accounts for a total cost of `829 000 for 25 years. The interest rate of 6.25% per year accounts for a maturity value of `2 580 030 for a fixed deposit. For the power purchase agreement (PPA) signed for 20 years at the rate of 8` per unit of solar power produced, the cost will be `1 907 680 for a total unit of `238 460 of 20 years. Hence, in the view of simple interest, Case 2 (i.e., power purchase agreement) is more economical than Case 1 for the installation of the solar PV system. The CO2 emission factor is 1.57 kg/kWh for electricity [taken from Department of Environment, Food and Rural Affairs (DERFA)], and the CO2 emission factor of PV is 0.105 kg/kWh.15 The CO2 mitigation and emissions are given by the following formulas:14,16 CO2 mitigation by PV plant CO2 emission from PV plant

¼ ¼

Annual energy generation Annual energy generation

 ¼

Emission factor CO2 per kWh

Net CO2 reduction

¼

CO2 mitigation by PV plant



CO2 emission from PV plant

where the emission factor is 1.57 kg/kWh, and CO2 per kWh emitted because of the PV plant is 0.105 kg/kWh. The CO2 mitigation by the PV plant will be 18 719.11 kg/year, and the CO2 emission from the PV plant will be 1251.91 kg/year. The net CO2 reduction because of the solar PV system will be 17 467.2 kg/year. D. Building energy optimization

BEopt is a simulation-based building optimization model which has been developed to design the most energy and cost-effective combination of renewable energy measures for a residential building prototype in Chennai. BEopt is a building energy optimization software program that uses a logical order sequence search technique to optimize the building design starting from a base configuration and parameters. It uses Energy Plus and TRNSYS to perform lively active simulations of the building. Energy Plus estimates hourly building heating, cooling, water heating, and appliance loads, while TRNSYS calculates the renewable energy production for solar PV and water heating to achieve the net-zero energy concept.17 Data include a vast library of energy efficient parametric options, related to the building envelope, parametric design, appliances, heating and cooling systems, and the PV system. This also includes technical features coupled with life operation, cost, maintenance, life expectancy, and replacement costs. The climatic variation which has a potential impact on the estimation of cooling loads has been added to the calculations. The climatic data of the location are given in the form of the ISHRAE (Indian Society of Heating, Refrigerating & Air-conditioning Engineers) weather data file. The simulation shared data include both the model input and output. The core data input such as the building setup, efficiency options, climatic conditions, energy, and economic parameters are designed and customized as per the Indian standards, local requirements, and the availability of the material. The unavailable elements are designed as per Indian standards for the input. Further data are also available on the building layout and geometry, materials, costs, appliances, lighting, envelope, and systems. Figure 5 shows how the sketch-up integration simplified

035103-8

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 5. Building overall layout (2496 ft2).

the creation of the building model. Figure 6 shows the difference in choosing energy and cost efficiency options related to the building orientation, walls, roofs, ceilings, foundation, windows, airflow, thermal mass, and space conditioning. Different economic and technical data have been defined and are available within the designed system file. The selected location, Chennai, is on the southeastern coast of India in the northeast part of Tamil Nadu on a flat coastal plain. Its average elevation is around 6.7 m (22 ft.), and its highest point is 60 m (200 ft.). It is a tropical dry climate and lies in the warm equatorial region. The hottest months late May to early June have maximum temperatures of 35  C–40  C, and the coldest part of the year in December and January has minimum temperatures around 19  C–25  C. Hence, the place is hot dry with a significant level of humidity, and the materials used in the building are designed with higher R-values to resist the heat flow and at the same time to allow little heat to pass through when it is cold. The insulation materials used are

FIG. 6. Parametric difference and cost of the building.

035103-9

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 7. Energy saving Vs. annualized energy cost.

designed to have lower U-values to resist the heat. The R-value is defined as the resistivity of a material to the heat flow, whereas the U-value is the effectivity of the material as an insulator. Data outputs can be visualized in different forms: energy consumption, energy savings, selected efficiency measures, costs, and renewable production. Provided data allow the identification of the NZEB target within the cost-optimal curve which reports global costs per year ($/yr) and energy saving (%) as shown in Fig. 7. Incremental and cumulative costs of the location can be visualized as well. The reduction of electricity consumption towards NZEBs was derived as shown in Fig. 8. Data comparison before and after the building optimization about the parameters such as energy, utility bill, CO2 emission, and heating/cooling loads is shown in Figs. 8–11, respectively. III. RESULTS AND DISCUSSION

The comparative results obtained from the three methods are much similar with a variation within 1000 kWh. The annual energy generation using PV Sol was 11 077 kWh/year, and when using BEopt Software, it was 12 177 kWh/year. The energy generation calculated using the forecasted irradiance value from WEKA was 11 923 kWh/year. Hence, the results of PV Sol and BEopt justify that the random forest algorithm in WEKA is optimal. The correlation factor of the forecasted values supports these findings with the overall high R-value of 0.9084. This

FIG. 8. Energy usage in MMBtu/yr.

035103-10

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

FIG. 9. Annual utility bill for $/year.

FIG. 10. Co2 emissions in metric tons/year.

FIG. 11. HVAC capacities in kBtu/h.

035103-11

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

model was shown to have RMSE values of 11% to 24% for various seasonal forecasting using the random forest algorithm in WEKA. This gave the annual irradiance results nearer to the PV Sol energy forecasting results and BEopt results. In the economic analysis of the solar PV system, the total cost of Case 1 was calculated to be 829 000` with an annual saving of 33 385 `/yr and a payback period of approximately 14.6 years. The economic saving for 25 years is 834 625`. When the initial investment is not done as in Case 2 and when the plant is commissioned by the power purchase agreement, the cost of the power produced for the agreement period of 20 years will be 1 907 680` after which the power is free. The initial amount may be deposited for a cumulative interest rate of 6.25% to get a maturity value of 2 580 030`. Hence, with the above results, it is significant that for a large PV system, commissioning the system by power purchase agreement is economical than the full initial payment as in Case 1. The results obtained from the building energy optimization show that the net-zero energy concept can be achieved for the selected building with the 7 kW Solar PV System as shown in Fig. 8. The energy consumption of the system was reduced from 192.2 MMBtu/yr to the minimum of 109.1 MMBtu/yr with negative upgradation cost, i.e., the cost of the building before optimization was 66 488 US$, and after optimization, it was 66 202 US$, which shows that the optimized building is comparatively less than the traditional one. The optimized building with a 7 kW PV system can achieve the net-zero energy concept with an initial investment of 87 407 US$, which is 21 205 US$ more than the cost and energy optimized building. The HVAC capacity has been lowered more than half for both cooling and heating systems as shown in Fig. 11. From Fig. 10, the CO2 emission has been dipped to 7.1 tons/year from 12.3 tons/year for the energy optimized building, and the CO2 emission was drawn to a negative value of 0.9 tons/year for the net-zero energy building. IV. CONCLUSION

In this paper, solar irradiance forecasting and building energy optimization were carried out to achieve the net-zero energy concept. At first, the solar irradiance forecasting was done until 2040 by considering the hourly data for fifteen years. The forecasted model was shown to have RMSE values of 11% to 24% for various seasonal forecasting using the random forest algorithm. The building energy optimization was performed for the chosen building with standard parameters used widely in India. During the optimization, various customized parameters and systems available in India were considered. It was concluded that the energy consumption was reduced from 192.2 MMBtu/yr to 109.1 MMBtu/yr by building energy optimization, and with a support of the 7 kW Solar PV System, the net-zero energy concept can be achieved. 1

K. D. V. Siva Krishna Rao, M. Premalatha, and C. Naveen, “Method and strategy for predicting daily global solar radiation using one and two input variables for Indian stations,” J. Renewable Sustainable Energy 10(1), 013701 (2018). H. Khalid and M. Zakaria, “Estimation of solar radiation in southern areas of Pakistan using radiation models,” J. Renewable Sustainable Energy 8(4), 043701 (2016). 3 A. Pacurar, N. Stefu, O. Mares, E. Paulescu, D. Calinoiu, N. Pop, R. Boata, P. Gravila, and M. Paulescu, “Forecasting hourly global solar irradiation using simple non-seasonal models,” J. Renewable Sustainable Energy 5(6), 063140 (2013). 4 A. K. Yadav, H. Malik, and S. S. Chandel, “Selection of most relevant input parameters using WEKA for artificial neural network based solar radiation prediction models,” Renewable Sustainable Energy Rev. 31, 509–519 (2014). 5 S. Gupta, “A regression modeling technique on data mining,” Int. J. Comput. Appl. 116(9), 27–29 (2015). 6 S. K. Shevade, S. S. Keerthi, C. Bhattacharyya, and K. R. K. Murthy, “Improvements to the SMO algorithm for SVM regression,” IEEE Trans. Neural Networks 11(5), 1188–1193 (2000). 7 L. Breiman, “Random forests,” Mach. Learn. 45(1), 5–32 (2001). 8 D. D’Agostino and D. Parker, “A framework for the cost-optimal design of nearly zero energy buildings (NZEBs) in representative climates across Europe,” Energy 149, 814–829 (2018). 9 J. D. Rhodes, W. H. Gorman, C. R. Upshaw, and M. E. Webber, “Using BEopt (EnergyPlus) with energy audits and surveys to predict actual residential energy usage,” Energy Build. 86, 808–816 (2015). 10 R. Meenal and A. I. Selvakumar, “Assessment of SVM, empirical, and ANN based solar radiation prediction models with most influencing input parameters,” Renewable Energy 121, 324–343 (2018). 11 W. Sprenger, H. R. Wilson, and T. E. Kuhn, “Electricity yield simulation for the building-integrated photovoltaic system installed in the main building roof of the Fraunhofer Institute for solar energy systems ISE,” Sol. Energy 135, 633–643 (2016). 12 A. Chel, G. N. Tiwari, and A. Chandra, “A simplified method of sizing and life cycle cost assessment of building integrated photovoltaic system,” Energy Build. 41(11), 1172–1180 (2009). 2

035103-12 13

Naveen Chakkaravarthy et al.

J. Renewable Sustainable Energy 10, 035103 (2018)

W. Wang, Y. Liu, X. Wu, Y. Xu, W. Yu, C. Zhao, and Y. Zhong, “Environmental assessments and economic performance of BAPV and BIPV systems in Shanghai,” Energy Build. 130, 98–106 (2016). P. Sandwell, N. L. A. Chan, S. Foster, D. Nagpal, C. J. Emmott, C. Candelise, S. J. Buckle, N. Ekins-Daukes, A. Gambhir, and J. Nelson, “Off-grid solar photovoltaic systems for rural electrification and emissions mitigation in India,” Sol. Energy Mater. Sol. Cells 156, 147–156 (2016). 15 Nallapaneni, Manoj Kumar, K. Sudhakar, and M. Samykano, “Techno-economic analysis of 1 MWp grid connected solar PV plant in Malaysia,” Int. J. Ambient Energy (published online, 2017). 16 Z. S. M. Nadoushani and A. Akbarnezhad, “Effects of the structural system on the life cycle carbon footprint of buildings,” Energy Build. 102, 337–346 (2015). 17 D. D’Agostino and D. Parker, “Data on cost-optimal nearly zero energy buildings (NZEBs) across Europe,” Data Brief 17, 1168–1174 (2018). 14