Solar Energy Output Forecasting from SolarGIS Data for Connected Grid Station

i
Solar Energy Output Forecasting from SolarGIS Data for Connected
Grid Station
Submitted by
Saradindu Sengupta
(Reg no: 91616012)
In partial fulfillment of the requirements for the award of
Master of Science in Computer Science with
Specialization in Machine Intelligence
COCHIN UNIVERSITY OF SCIENCE AND TECHNOLOGY, KOCHI
Conducted by
Indian Institute of Information Technology and Management-Kerala
Technopark Campus
Thiruvanathapuram-695 581
May 2018

ii
Declaration
I, Saradindu Sengupta, MSMI-1613, hereby declare that this report is
substantially the result of my own work, except, where explicitly indicated in the
text and has been carried out during the period December 2017-April 2018.
Place: Thiruvananthapuram, Kerala
Date: 17.05.2018
Saradindu Sengupta

iii
Acknowledgment
I would first like to thank my thesis advisor Dr. Asharaf S of the Indian
Institute of Information Technology & Management , Kerala . The door to Prof.
Asharaf S's office was always open whenever I ran into a trouble spot or had
a question about my research or writing. He consistently allowed this paper
to be my own work, but steered me in the right the direction whenever he
thought I needed it.
I would also like to thank the experts who were involved in the validation
survey for this research project. Without their passionate participation and
input, the validation survey could not have been successfully conducted.
Finally, I must express my very profound gratitude to my parents for
providing me with unfailing support and continuous encouragement
throughout my years of study and through the process of researching and
writing this thesis. This accomplishment would not have been possible
without them.
Thank you.
Place: Thiruvananthapuram, Kerala
Date:17.05.2018
Saradindu Sengupta

iv
Abstract
Using random forest regression method, daily mean solar output generation
can yield promising result rather than conventional NWP model for
forecasting. Using that in practice also the goal was to create a user-friendly
application , with easy access, to provide accurate forecasting regarding
saving and conservation.This goal has been accomplished in three stages:
● Building a machine learning model that predicts the annual energy
production of a prospective solar installation.
● Building a model that predicts installation cost.
● Implementing these models on a user-friendly web app that shows users
how much they should expect to save on their energy bill each year by
switching to solar.
The random forest model out-performs its other rivals and also conventional
models , thus providing a better suited model to run with for forecasting.

1
Table of Contents
Abstract--------------------------------------------------------------------------------------------iii
LIST OF TABLES--------------------------------------------------------------------------------------------vii
LIST OF FIGURES----------------------------------------------------------------------------------vii
CHAPTER I----------------------------------------------------------------------------------------------------- 1
INTRODUCTION----------------------------------------------------------------------------------------------1
1.1 The necessity to predict solar radiation or solar production----------------4
1.2 Related Work-------------------------------------------------------------------------------------8
1.3 Available Forecasting Methodologies--------------------------------------------------9
1.3.1 NWP Model------------------------------------------------------------------------------- 9
1.4 Proposed Methodology---------------------------------------------------------------------12
CHAPTER II---------------------------------------------------------------------------------------------------13
Literature Review-----------------------------------------------------------------------------------------13
CHAPTER III--------------------------------------------------------------------------------------------------17
Materials & Methodologies--------------------------------------------------------------------------- 17
1.1 Materials------------------------------------------------------------------------------------------17
1.1.1 Data Sources and Viability------------------------------------------------------ 17
1.1.2 Data Collection----------------------------------------------------------------------- 19
1.2 Methodologies--------------------------------------------------------------------------------- 20
1.2.1 Parameters----------------------------------------------------------------------------- 20
1.2.2 Exploring the Data------------------------------------------------------------------ 21
1.2.3 Partial Correlation-------------------------------------------------------------------23
1.2.4 Baseline Model------------------------------------------------------------------------24
1.2.5 Model Selection---------------------------------------------------------------------- 25
CHAPTER IV------------------------------------------------------------------------------------------------- 26
Findings-------------------------------------------------------------------------------------------------------26

2
1.1 Comparing Model Performance-------------------------------------------------------- 27
1.2 Model Output/Size----------------------------------------------------------------------------28
1.3 Creating Web App----------------------------------------------------------------------------29
1.3.1 Using the Web App----------------------------------------------------------------- 30
CHAPTER V-------------------------------------------------------------------------------------------------- 31
Conclusion---------------------------------------------------------------------------------------------------31
1.1 Problems----------------------------------------------------------------------------------------- 32
1.2 Future Implementation-------------------------------------------------------------------- 32
REFERENCES------------------------------------------------------------------------------------------------33

3
LIST OF TABLES
Table
1 Specification of various storage options
2 Characteristics of electricity production plants
3 Comparison Between different model

4
LIST OF FIGURES
Figure
1 Prediction scale for energy management in an electrical network
2 Solar azimuth and zenith angles
3 a) Forecasting error versus forecasting models (left) [6,14].
b) Relation between forecasting horizons, forecasting models and the
related activities
4 In this graph of the western U.S., size represents number of
installations per ZIP code, and color represents amount of direct
radiation. Dark red indicates higher radiation, and larger
5 A scatter plot of the size of solar panels and their annual
6 The partial correlations between radiation factors and installation
factors.
7 A distribution of the differences in predicted return from actual return
shows a very strong positive skew, due to large prediction errors for
utility scale solar panels.
8 The distribution of annual energy production is highly positively
skewed, with some utility-scale installations producing many hundreds
of times more energy per year than the median.
9 Feature importance of the random forest model
10 The coefficients on an elastic net regression model predicting output
per unit size.
11 PV solar system cost and forecasting

5
CHAPTER I
INTRODUCTION
An electrical operator should ensure a precise balance between the electricity
production and consumption at any moment. This is often very difficult to
maintain with conventional and controllable energy production system,
mainly in small or not interconnected (isolated) electrical grid (as found in
islands). Many countries nowadays consider using renewable energy sources
into their electricity grid. Photo voltaic systems are becoming important
sources of energy in electricity networks. However, electric utility companies
are required to guarantee electricity supply within certain ranges which is
difficult given the fluctuating nature of weather conditions.
With the dwindling fossil fuel resources, research in renewable energy has
gained significant impetus. The leading source for renewable energy is solar
power generation mainly through photo-voltaic cells. Advantages of using
solar energy include its immunity to imitative circumstances like the oil prices,
a clean source of energy, and reduction of imports and dependability on
external resources. Though photo-voltaic cells are considered as a major
source for future energy generation, their return on investment and upfront
cost is hindering their deployments. One of the reasons for this is lack of
predictable supply because of changing weather conditions. Since photo-
voltaic cells generate electricity by converting solar energy to electric current,
the amount of solar energy being provided in a day is very important to size
the photo-voltaic system. Therefore, the amount of electricity produced
depends upon solar irradiance in a particular day, which itself depends on
various parameters such as location, time, and weather patterns. Solar
irradiance is defined as the power per unit area received from the Sun in the
form of electromagnetic radiation in the wavelength range of the solar cell
being used. Unfavorable weather reduces the output of the solar plant to a
large extent. Therefore, in order to fulfill the energy requirements, a power
supply company needs to supplement the remaining amount by purchasing

6
power from different power generation companies running on costlier fossil
fuels. The power rates charged by these companies not only depend upon the
amount of the power required but also on the timeliness of the order. A
timely order placement with these companies not only helps to meet the
promised power supply goals, but also helps reduce the cost. Hence prior
knowledge of the power produced plays a crucial role in maintaining quality
of service and reducing cost. With solar power, its possible to predict the
production knowing current and the past information about the weather and
the irradiance. Various researchers have proposed forecasting mechanisms
with good results however a room for improvement still exists. There are two
orthogonal avenues of improvement in this domain, one is more efficient
algorithm design for forecasting and second is identification and
quantification of the effect of parameters on forecast. In this paper we
attempt to improve the state of the art in both the dimensions.
Capacity
Network
Optimization/Pro
duction/Consum
ption Network
System Safety
Fig: 1 Prediction scale for energy management in an electrical
network
Minutes/Seconds
(Power
management)
Years
(Global
management)
Months/Weeks
/Days
(Energy
management)

7
1.1 The necessity to predict solar radiation or solar production
One of the most important challenge for the near future global energy supply
will be the large integration of renewable energy sources (particularly non-
predictable ones as wind and solar) into existing or future energy supply
structure. An electrical operator should ensure a precise balance between the
electricity production and consumption at any moment. As a matter of fact,
the operator has often some difficulties to maintain this balance with
conventional and controllable energy production system, mainly in small or
not interconnected (isolated) electrical grid (as found in islands). The
reliability of the electrical system then become dependent on the ability of
the system to accommodate expected and unexpected changes (in
production and consumption) and disturbances, 5 while maintaining quality
and continuity of service to the customers. Then, the energy supplier must
manage the system with various temporal horizons. The integration of
renewable energy into an electrical network intensifies the complexity of the
grid 5 management and the continuity of the production/consumption
balance due to their intermittent and unpredictable nature [1, 2]. The
intermittent and the non-controllable characteristics of the solar production
bring a number of other problems such as voltage fluctuations, local power
quality and stability issues . Thus forecasting the output power of solar
systems is required for the effective operation of the power grid or for the
optimal management of the energy fluxes occurring into the solar system . It
is also necessary for estimating the reserves, for scheduling the power
system, for congestion management, for the optimal management of the
storage with the stochastic production and for trading the produced power in
the electricity market and finally to achieve a reduction of the costs of
electricity production. Due to the substantial increase of solar power
generation the prediction of solar yields becomes more and more important.
In order to avoid large variations in renewable electricity production it is
necessary to include also the complete prediction of system operation with
storage solutions. Various storage systems are being developed and they are
a viable solution for absorbing the excess power and energy produced by
such systems (and releasing it in peak consumption periods), for bringing

8
very short fluctuations and for maintaining the continuity of the power
quality.These storage options are usually classified into three categories:
● Bulk energy storage or energy management storage media is used to
decouple the timing of generation and consumption.
● Distributed generation or bridging power - this method is used for peaks
shaving - the storage is used for a few minutes to a few hours to assure
the continuity of service during the energy sources modification.
● The power quality storage with a time scale of about several seconds is
used only to assure the continuity of the end use power quality.
Category Discharge
power
(MW)
Dischar
ge Time
Stored
Energy
Representative
Application
Bulk energy 10 - 1000 1 - 8 H 10- 8000
MWh
Load leveling, generation
capacity
Distributed
generation
0.1 - 2 0.5 - 4 H 50 – 8000
kWh
Peak shaving, transmission
deferral
Power
quality
0.1 - 2 1 - 30 S 0.03 – 16.7
kWh
End-use power quality
/reliability
Table 1. Specification of various storage options

9
Table 1 shows these three categories and their technical specifications. As
shown, every type of storage is used in different cases to solve different
problems, with different time horizon and quantities of energy.Table 1 shows
that the electricity storage can be widely used in a lot of cases and
applications as a function of the time of use and the power needs of the final
user. Finally, it shows that the energy storage acts at various time levels and
their appropriate management requires the knowledge of the power or
energy produced by the solar system at various horizons: very short or short
for power quality category to hourly or daily for bulk energy storage. Similarly,
the electrical operator needs to know the future production (Figure 1) at
various time horizons from one to three days, for preparing the production
system and to some hours or minutes for planning the start-up of power
plants (Table 2). Starting a power plant needs between 5 min for a hydraulic
one to 40 hours for a nuclear one. 9 Moreover, the rise in power of the
electrical plants is sometimes low, thus for an effective balance between
production and consumption an increase of the power or a starting of a new
production needs to be anticipated sometimes well in advance.

10
Furthermore, the relevant horizons of forecast can and must range from 5
minutes to several days as it 9 was confirmed by Diagne et al. [6]. Elliston
and MacGill [10] outlined the reasons to predict solar radiation for various
solar systems (PV, thermal, concentrating solar thermal plant, etc.) insisting
on the forecasting horizon. It therefore seems apparent that the time-step of
Type of
electrical
generator
Power size
MW
Minimum
power capacity
percentage of
peak power
Rise speed in
power per min
percentage of
peak power
Starting time
Hour
Nuclear Power
Plant
400–1300 per
reactor
20% 1% 40 h (cold)-18
h
Steam thermal
plant
200–800 per
turbine
50% 0.5%-5% 11-20 h (cold)-
5 h
Fossil-fired
power plants
1–200 50-80% 10% 10 min-1 h
Combined-
cycle plant
100–400 50% 7% 1-4 h
Hydro power
plant
50–1300 30% 80%-100% 5 min
Hydro power
plant
25 30% 30% 15-20 min
Internal
combustion
engine
20 65% 20% 45-60 min
Table 2. Characteristics of electricity production plants

11
the predicted data may vary depending on the objectives and the forecasting
horizon. All these reasons show the importance of forecasting, whether in
production or in consumption of energy. The need for forecasting lead to the
necessity to use effective forecasting models. In the next section the various
available forecasting methodologies are presented.
1.2 Related Work
The current eras’ accelerating advances in all fields rely on the constant
supply of the electric power. The fossil fuels are key source of the reliable and
consistent energy provision, but they have high cost associated with them,
emit dangerous gases, and are subject to uncertainty of the international oil
prices. On the other hand, solar power is cheap, clean source of energy that
can be produced by every country, but its dependence on weather conditions
make it less reliable. To cope with this unreliability the prediction of the solar
power contributes towards a consistent supply. The prediction of solar power
is a multidisciplinary research that needs contribution from meteorology,
solar cell engineering, electrical engineering, and machine learning
computation. used the radial basis function (RBF) to model the solar radiation
based on sunshine duration and air temperature data. They use Multi-Layer
Perceptron (MLP) to model the hourly forecast of the solar radiation using
present values of air temperature and mean daily solar irradiance. [2]
compare different models including Fuzzy Logic, Neural Networks and
Autoregressive to predict the half daily values of solar radiation. In [3], 3-day
forecast of the solar irradiance has been modeled using the forecast data
provided by European Center for Medium-Range Weather Forecasts (ECMWF).
In [4], different weather parameters such as solar hours, latitude, longitude,
elevation, maximum and minimum air temperature, humidity, and rainfall has
been used to predict the solar power in Indonesia. It also used ANN to
generate a model that could predict the solar power of a particular region.A
recent addition is the use of computer algorithm based on fuzzy logic control
(FLC) to estimate the wind and solar energies in a hybrid renewable energy
system from natural factors [5]. The solar power was estimated using the
temperature and the lighting as input parameters. A detail introduction to the

12
current research on forecasting solar irradiance is presented in [6]. It is a in-
depth review to facilitate selection of the appropriate forecast method
according to needs. They also comment on the statistical approaches and
techniques based on images from satellite imagery. They also discuss
numerical weather prediction (NWP) and hybrid models. Online forecasting of
power production from PV systems is discussed in [7], to predict hourly
values of solar power for horizons of up to 36 h. A two-stage method of using
adaptive linear time series models for clear sky model and autoregressive
(AR) models NWPs is elaborated. [8] compare multiple regression techniques
for generating prediction models for solar power, including linear least
squares, and support vector machines using multiple kernel functions.
1.3 Available Forecasting Methodologies
1.3.1 NWP Model
The solar power forecasting can be performed by several methods; the two
big categories are the cloud imagery combined with physical models, and the
machine learning models. The choice for the method to be used depends
mainly on the prediction horizon; actually all the models have not the same
accuracy in terms of the horizon used. Various approaches exist to forecast
solar irradiance depending on the target forecasting time. The literature
classifies these methods in two classes of techniques:
Fig:2 Solar azimuth and zenith angles

13
● Extrapolation and statistical processes using satellite images or
measurements on the ground level and sky images are generally suitable
for short-term forecasts up to six hours. This class can be divided in two
sub-classes, in the very short time domain called “Now- casting’’ (0–3 h),
the forecast has to be based on extrapolations of real-time
measurements [5]; in the Short-Term Forecasting (3–6 h), Numerical
Weather Prediction (NWP) models are coupled with post-processing
modules in combination with real-time measurements or satellite data [5,
11].
● NWP models able to forecast up to two days ahead or beyond [12, 13] (up
to 6 days ahead [13]). These NWP models are sometimes combined with
post-processing modules and satellite information are often used [2].

14
The NWP models predict the probability of local cloud formation and then
predict indirectly the transmitted radiation using a dynamic atmosphere
model. The extrapolation or statistical models analyze historical time series of
global irradiation, from satellite remote sensing [15] or ground
measurements [16] by estimating the motion of clouds and project their
impact in the future [6, 13, 17]. Hybrid methods can improve some aspects of
all of these methods [6, 14]. The statistical approach allows to forecast hourly
solar irradiation (or at a lower time step) and NWP models use explanatory
variables (mainly cloud motion and direction derived from atmosphere) to
predict global irradiation N-steps ahead [15]. Very good overviews of the
forecasting methods, with their limitations and accuracy can be found in [1, 5,
6, 10, 12, 14, 18]. Bench-marking studies were performed to assess
comparisons and the accuracy of the results produced, as shown in this paper,
must be carefully evaluated in selecting the right method to use. As part of
COST Action ES1002. (European Cooperation in Science and Technology) [22]
on Weather Intelligence for Renewable Energies (WIRE) a literature review on
the forecasting accuracy applied to renewable energy systems mainly solar
and wind is carried out. In this paper an overview on the various
methodologies available for solar radiation prediction based on machine
Fig: 3 a) Forecasting error versus forecasting models (left) [6,14]. b) Relation
between forecasting horizons, forecasting models and the related activities
(right) [6, 14]

15
learning is presented. A lot of review papers are available, but it is very rare
to find a paper which is totally dedicated to the machine learning methods
and that some recent prediction models like random forest, boosting or
regression tree be integrated. In the next section the different methodologies
used in the literature to predict global radiation and the parameters used for
estimating the model performances are presented.
1.4 Proposed Methodology
Instead of using NWP model for weather simulation and solar power,
irradiance forecasting, the proposed method uses Random Forrest Regression
method as a bench-mark for daily mean power forecasting.
The final model used for production was a random forest regression model,
which has several benefits in this context. It is a non-parametric model, which
means it can predict a variable that is non-normally distributed. Because
there are a wide range of solar panels in the OpenPV dataset, with some
utility-scale installations producing thousands of times more energy per year
than small, residential panels, the data is very positively skewed. This means
that in order to use a model like linear regression without having biased
results, it would be necessary to log transform the data, or use a generalized
linear model like a Poisson regression. Random forest models also do well
with categorical features, and in this case there were a few such features,
including technology type and tracking type.

16
CHAPTER II
Literature Review
1. Direct Normal Irradiance (DNI):
Direct Normal Irradiance (DNI) is the amount of solar radiation received per
unit area by a surface that is always held perpendicular (or normal) to the
rays that come in a straight line from the direction of the sun at its current
position in the sky. Typically, you can maximize the amount of irradiance
annually received by a surface by keeping it normal to incoming radiation.
This quantity is of particular interest to concentrating solar thermal
installations and installations that track the position of the sun.
2. Diffuse Horizontal Irradiance (DHI):
Diffuse Horizontal Irradiance (DHI) is the amount of radiation received per
unit area by a surface (not subject to any shade or shadow) that does not
arrive on a direct path from the sun, but has been scattered by molecules
and particles in the atmosphere and comes equally from all directions.
3. Direct Radiation:
Direct radiation is also sometimes called "beam radiation" or "direct beam
radiation". It is used to describe solar radiation traveling on a straight line
from the sun down to the surface of the earth.
4. Diffuse radiation:
Diffuse radiation, on the other hand, describes the sunlight that has been
scattered by molecules and particles in the atmosphere but that has still
made it down to the surface of the earth.However, since diffuse radiation is
generally pretty equally distributed throughout the sky, the most diffuse
radiation is gathered when your solar panels are laying down horizontally.
The steeper your solar panels are tilted, the less of the sky they are facing
and the more of the sky's diffuse radiation they miss out on. If, for example,

17
your solar collectors are tilted at a 45° angle, they are facing away from
about a quarter of the sky and would only collect about three-fourths of the
diffuse radiation in the sky.
Still, because direct radiation is much more intense than diffuse radiation, the
amount of radiation missed by tilted solar panels is generally more than
compensated for by the extra radiation gained by tracking the sun.
5. Global Horizontal Irradiance (GHI):
Global Horizontal Irradiance is the total amount of shortwave radiation
received from above by a surface horizontal to the ground. This value is of
particular interest to photovoltaic installations and includes both Direct
Normal Irradiance (DNI) and Diffuse Horizontal Irradiance (DHI).
Global Horizontal (GHI) = Direct Normal (DNI) X cos(θ) + Diffuse Horizontal
(DHI)
6. Random Forest:
The Random Forest is one of the most effective machine learning models for
predictive analytic, making it an industrial workhorse for machine learning.
Background:
The random forest model is a type of additive model that makes predictions
by combining decisions from a sequence of base models. More formally we
can write this class of models as:
g(x)=f0(x)+f1(x)+f2(x)+...
where the final model g is the sum of simple base models fi. Here, each base
classifier is a simple decision tree. This broad technique of using multiple
models to obtain better predictive performance is called model ensembling.
In random forests, all the base models are constructed independently using a
different sub-sample of the data.

18
7. NWP Model:
Numerical Weather Prediction (NWP) data are the form of weather model
data we are most familiar with on a day-to-day basis. NWP focuses on taking
current observations of weather and processing these data with computer
models to forecast the future state of weather. Knowing the current state of
the weather is just as important as the numerical computer models
processing the data. Current weather observations serve as input to the
numerical computer models through a process known as data assimilation to
produce outputs of temperature, precipitation, and hundreds of other
meteorological elements from the oceans to the top of the atmosphere.
8. Azimuth Angle:
The azimuth angle is the compass direction from which the sunlight is coming.
At solar noon, the sun is always directly south in the northern hemisphere
and directly north in the southern hemisphere. The azimuth angle varies
throughout the day as shown in the animation below. At the equinoxes, the
sun rises directly east and sets directly west regardless of the latitude, thus
making the azimuth angles 90° at sunrise and 270° at sunset. In general
however, the azimuth angle varies with the latitude and time of year.
9. Elastic Net Regression:
Elastic net regularization method includes both LASSO (L1) and Ridge (L2)
regularization methods.
Overfitting : The core idea behind machine learning algorithms is to build
models that can find the generalised trends within the data. However, if no
measures are taken, sometimes the models tend to rote learn the data
instead of learning the patterns. During such cases, although the model fits
well to the training data ( model yields accurate results when evaluated on
training data), however, it evaluated poorly on the test data. This is called
overfit.
Regularization is used to prevent overfitting the model to training data. This
is achieved by slightly perturbing ( adding noise ) the objective function of

19
the model before optimizing it ( optimising a model means to find the model
parameters w* such that the argmin /argmax of the objective function is
found- in other words, it is to find the global optima of the objective function) .
In L1 Regularisation, a noise of magnitude lambda .|w*| is added while in L2
Regularisation, noise of magnitude lambda.|w*|. |w*| is added. where |w*| is
the magnitude of the optimal parameter vector.
10. Pyranometer:
A pyranometer is a type of actinometer used for measuring solar irradiance
on a planar surface and it is designed to measure the solar radiation flux
density (W/m2) from the hemisphere above within a wavelength range 0.3
μm to 3 μm. A typical pyranometer does not require any power to operate.
However, recent technical development includes use of electronics in
pyranometers, which do require (low) external power.

20
CHAPTER III
Materials & Methodologies
In global horizontal irradiance forecasting the models can be used in three
different ways [24]:
● structural models which are based on other meteorological and
geographical parameters;
● time-series models which only consider the historically observed data of
solar irradiance as input features (endogenous forecasting);
● hybrid models which consider both, solar irradiance and other variables
as exogenous variables (exogenous forecasting). The proposed method
uses the last mentioned way,i.e., hybrid method.
1.1 Materials
1.1.1 Data Sources and Viability
Data for this project came from three sources, both managed by the NREL.
The first is The OpenPV Project, which contains data related to over one
million solar panel installations across the U.S. This dataset includes the
following:
1. Annual energy production
2. Installation cost
3. Size
4. Orientation
5. Tilt
6. Installer
7. Technology type
etc.

21
The second dataset comes from the National Solar Radiation Database
(NSRDB) API. This dataset includes hourly measures of:
8. Radiation
9. Temperature
10. Wind speed
11. Position of the sun
The third dataset comes from CleantechSolar Energy Corporation Pvt Ltd(R)
‘ s own rooftop solar plant from Singapore and US west coast. The data at
Cleantech Solar’s sites are recorded using proprietary data logger provided
by SERIS( Solar Energy Research Institute , Singapore, NUS) which contains
data in the following way
12. Global Horizontal Irradiance
13. Global Tilted Irradiance
14. Total Power pushed to the Grid
15. Humidity
16. Wind Speed
As a failsafe process to ensure unhindered availability of data, satellite data
also use here provided as commercial use by SolarGIS which shows as
follows:
17. Global Diffusion Irradiance
18. Humidity
19. Wind Speed
1.1.1.1 Data Source Description
The data provided by NREL is in plain CSV file format provided according to
ZIP code and contains over one million solar system data logged on bi-
monthly basis.
The data provided by NSRDB is in form of API and can be accessed on daily
basis at once. This data contains mainly sensor data rather than actual

22
system output, load generation and grid pumping data.
The third data provided by CleantechSolar Energy Corporation Pvt Ltd comes
from company owned cloud storage which contains time series data for
sensors and electric power generation as well.
1.1.2 Data Collection
The data for NREL’s OpenPv project came as a simple CSV download-able
from their website here [https://openpv.nrel.gov/search], accessible by ZIP
code in amount of millions of solar power station installation of either rooftop
or industrial.
The NSRDB API only allows one thousand daily queries, so in order to gather
local radiation data for the roughly fifteen thousand ZIP codes in the OpenPV
dataset, used a python script, set it to run every 24 hours, and deployed it
on a remote Amazon Web Services EC2 instance. The script pulled hourly
data from all of 2017 (the most recent data available), averaged it for the
year, and saved it to a local Neo4J database. Once pulled down radiation
data for all one million solar panel installations, merged the two datasets
according to ZIP code, the most granular location measure available.
The data from CleantechSolar Energy Corporation Pvt Ltd, came from Digital
Ocean hosted server where every station’s data comes on certain time of the
day.

23
1.2 Methodologies
1.2.1 Parameters
Based on exploratory analysis, the following variables have been chosen to
build a model for annual energy production.Direct Normal Irradiance (DNI) is
the amount of radiation that travels directly from the sun to the earth,
whereas Diffuse Horizontal Irradiance (DHI) is the amount of radiation that
reflects off particles in the air before hitting the surface of the earth. Diffuse
irradiance is therefore higher in places with more cloud cover or more dust in
the air to block the travel of sunlight to the earth.
In the northern hemisphere, the further north an installation is located, the
more it ought to be tilted. The optimal tilt difference is simply the difference
between the tilt of a panel and its latitude. Technology type refers to what
type of silicon is used in the panel. Some panels move to track the sun, and
tracking type refers to whether a panel is fixed or has some kind of tracking.

24
1.2.2 Exploring the Data
Though most solar installations are in places where direct irradiance (DNI) is
high, it appears the size of an installation actually plays the biggest role in
determining a panel’s energy output. Size is highly correlated with annual
energy production. When we look at a scatter plot of size and annual
production, the linear relationship between the two is clear. The larger an
installation, the higher its capacity for transforming solar radiation into
electricity, and thus the observed relationship.
Fig: 4 In this graph of the western U.S., size
represents number of installations per ZIP code,
and color represents amount of direct radiation.
Dark red indicates higher radiation, and larger

25
Fig: 5 A scatter plot of the size of solar panels and their annual

26
1.2.3 Partial Correlation
The above graph shows the partial correlations between installation factors
and radiation factors. The red box highlights the partial correlations between
these factors and annual energy production. Partial correlations are useful
because they show the correlation between two variables when the effects of
other variables are held constant. Dark red indicates high positive correlation,
dark blue indicates high negative correlation, and white indicates no
correlation.
Surprisingly, there appears to be no more than very slight correlations
between the radiation measures (DNI and DHI) and energy production. We
might expect—given that we’re talking about solar energy after all—that
these factors would play a large role in determining how much energy a solar
panel generates. In the U.S., this may not end up being the case because
Fig: 6 The partial correlations between radiation
factors and installation factors. The red box
highlights partial correlations with annual energy
production.

27
solar radiation is high enough that panels reach their capacity for energy
production, and are unable to produce more energy even with more sun
exposure. If we were to compare the performance of solar panels in the U.S.
to solar panels in the arctic, we might see that radiation is more highly
correlated with energy production than we see here. An alternative
explanation is that the level to which size correlates with output is biasing the
results of the partial correlations.
Like in the scatter plot, there’s again a very strong positive correlation
between size and energy production. It is interesting to note that there is a
slight negative partial correlation between temperature and energy
production. Though temperature obviously correlates highly with areas of
high sun exposure, when controlling for the effect of radiation, high
temperatures actually cause solar panels to produce energy less efficiently,
and degrade more quickly.
1.2.4 Baseline Model
The NREL OpenPV dataset includes two measures of annual energy
production, one that is a self reported value, and another predicted value
based on a user’s inputs. Using these two values, Established a baseline R²
score of .915. An R² score measures how much of the variance around the
mean a prediction captures. So, if predictions were as good as guessing the
average of the predicted variable every time, the R² score would be zero, and
if the predictions were perfect, the R² would be one. At an R² score of .915,
the baseline predictions are very accurate. Above is a scatter plot of the
predicted and reported values, showing how closely they correlate.

28
At an average return rate from a utility company of $0.1024 per kilowatt hour
of energy generated, the median error in the baseline is equivalent to +/-
approximately $60.00 in predicted savings on energy. The above chart shows
the distribution of the differences between predicted and actual output for
the baseline model, in dollars. The median is a more robust measure of
central tendency than the average in this situation, given the large positive
skew in the data. It’s also a better measure of the prediction error for small-
scale, residential solar customers. Over the course of the lifetime of a solar
panel, this error compounds, amounting to thousands of dollars difference
between the predicted and actual return.
1.2.5 Model Selection
The final model used for production was a random forest regression model,
which has several benefits in this context. It is a non-parametric model, which
means it can predict a variable that is non-normally distributed. Because
there are a wide range of solar panels in the OpenPV dataset, with some
utility-scale installations producing thousands of times more energy per year
than small, residential panels, the data is very positively skewed. This means
that in order to use a model like linear regression without having biased
Fig: 7 A distribution of the differences in predicted return from actual
return shows a very strong positive skew, due to large prediction errors for
utility scale solar panels.

29
results, it would be necessary to log transform the data, or use a generalized
linear model like a Poisson regression. Random forest models also do well
with categorical features, and in this case there were a few such features,
including technology type and tracking type.
The performance of this model was compared against three others, including:
1. Ordinary Least-Squares Regression
2. Elastic Net Regression
3. 2-Layer Feed Forward Neural Network
Built the regression models and the random forest model using Scikit-Learn,
and the neural network using Keras with a Tensorflow backend.
Fig: 8 The distribution of annual energy production is highly positively skewed,
with some utility-scale installations producing many hundreds of times more
energy per year than the median.

30
CHAPTER IV
Findings
Using the random forest model, achieved an R² value of .973 on validation
data. The validation data were the same values used to generate the baseline
score, so the comparison is completely like-to-like. In terms of annual savings,
the median error drops from the baseline of $60.00 to +/- approximately
$15.00 in predicted savings on energy.
Feature importance of a random forest model are a clear and easy way to
interpret how much different variables contribute to predictions. More
specifically, in Figure 9. they show the percent increase to mean squared
error were a variable to be excluded from the model. We see that size
contributes the most to predicting energy production, with an 87%
importance. Direct and diffuse irradiance also play a role. These findings
Fig: 9 Feature importance of the random forest model

31
indicate that while it’s most important to build as large an installation as
possible, building in places with high direct irradiance and low diffuse
irradiance will help to produce more energy. Wind speed also plays a role,
likely because high wind speeds correlate with areas of less shade, and high
winds keep debris from collecting on panels. The last feature here is optimal
tilt difference, which plays a comparatively small role in determining energy
output. The more off a solar panel is from its optimal tilt, the less energy it
will generate. Other factors in the model have a negligible importance.
1.1 Comparing Model Performance
While the random forest model ultimately performs best, it has several trade-
offs. It took the longest time to hone in on the optimal parameters, and the
interpret-ability of its results is lacking compared to linear regression and
elastic net. The latter two models give a clear relationship between changes
to factors like size and radiation and expected annual energy production,
whereas random forest only provides feature importance. The neural network
model suffers from both the drawback of time and a lack of interpret-ability.
Given the relative simplicity of predicting energy output, a neural network
appears to be an unnecessarily complex model for this situation. With more
training time and parameter tuning, the neural network model would likely
match the performance of the random forest model, but the marginal
benefits are clearly limited.
Table 3. Model Comparison

32
1.2 Model Output/Size
Because size dominates the performance of the models when used as a
predictive factor, I built a regression model predicting output per unit size to
gain a more granular understanding of the relationship between other factors
and energy production. Though this model performs more poorly than when
size is included as a feature, it helps demonstrate the effect of radiation and
installation factors on output. The model scores an R² of .683 on validation
data.
Fig: 10 The coefficients on an elastic net regression model predicting output
per unit size. These values show the positive or negative effect a change to
these factors has on the predicted variable.

33
Again Scikit-Learn has been used to build the elastic net regression for this
model. Elastic net uses two types of regularization, lasso and ridge, which
help to prevent over-fitting due to outliers and inconsequential features. Put
simply, over-fitting is when a model learns the intricacies of its training data
so specifically that it does not generalize well to other data.
The results of the model show how much positive or negative effect an
increase in each factor has on annual energy production per unit size of an
installation. Direct irradiance is the most important feature, with a large
positive impact. Both diffuse irradiance and optimal tilt difference have
substantial negative impacts on energy output, which is a fairly intuitive
result. We see that a solar panel with fixed tracking will produce less energy
as well, because a panel that does not track will have less exposure to
radiation. Presumably it requires energy to operate a tracking solar panel,
and this finding suggests that generally, the energy generated from tracking
outweighs the energy required to operate the tracking system. Finally, mono-
crystalline panels outperform poly-crystalline. We see this result because
mono panels contain purer silicon, which leads to increased efficiency.
1.3 Creating Web App
Using the random forest model, designed a web app that allows users to
input information about where they want to build solar panels, and learn how
much they would save on their energy bill. The web app also uses a second
model to predict installation cost based on the following factors:
1. Size
2. Installer
3. Tracking
4. Technology
Built the website using Django, a python-based web framework. then set it up

34
to run on an Amazon Web Services EC2 instance with a Gunicorn HTTP
server. When a user inputs their information, the app queries the NREL solar
radiation database to find their local radiation data, and calculates their
expected annual energy production and installation. A Scrapy web crawler
that finds their local average per-kilowatt return from a utility company, and
uses that to calculate their average savings per year, as well as how long it
would take to pay off the installation cost is also in use. The app can be
checked here[http://www.solarcalculator.xyz/].
1.3.1 Using the Web App
The web app can be accessed from here [ http://www.solarcalculator.xyz].
To access the system, initially it is without any restriction. After opening the
website,
Step 1. Input the desired address line 1, 2 and city along with Zip code,
for US and Singapore both.
Step 2. Input the total covered area in square meters preferably and
the tilt angle of the solar system
Step 3. Azimuth value determines the direction of the PV solar station.
The result will be shown in the right hand corner, in a way like showing
optimal tilt angle as well as total initial cost and annual return.

35
CHAPTER V
Conclusion
Over the past 10 years, installation costs for solar energy technology have
dropped an astonishing 60%. This form of renewable energy is more
accessible now than ever before. Yet over that same period, soft costs, such
as sales and marketing, have remained almost completely stagnant.
According to the Solar Energy Industries Association (SEIA), in Q4 2016, soft
costs accounted for 67% of installation costs for residential solar². This has
moved the impetus for growth in the solar industry from the development of
cheaper technologies to a focus on ways to attack soft costs by more
efficiently spreading information to potential solar customers.
If there is a choice between paying to install tracking, more tilt, etc., versus
installing additional panels, the additional panels will almost always be the
right move.
Fig 11: PV solar system cost and forecasting

36
Using machine learning, the model gives highly accurate predictions of the
expected return on energy generated by a prospective solar panel, and made
it easily accessible at solarcalculator.xyz. Tools such as this, which use the
machine learning techniques described above, will make information
regarding one’s ability to switch to solar more widely available, ultimately
bringing down soft costs of installation and accelerating the transition to
renewable energy. But there are some hiccups in the process mentioned
below.
1.1 Problems
The main problem arising before handling the data was the availability which
is one of the reason that the project now is mainly aimed at US and
Singapore. The unusual communication hiccups causes serious damage to
the data modeler.
1.2 Future Implementation
In future implementations, the learning model can be improved using better
sourced data and meaningful parameters. Also the amount of output
measurement treated here are very low thus increasing parameters involving
output generation is a positive step towards robust system.
This system not only work for solar power forecasting or return on investment
showing but also belongs to a suit of operations where the same model of
solar power forecasting and NWP modeling is used indicating cleanliness
drives of solar panels as solar panels need regular washing, specially in
desert areas. It will decrease human intervention in mundane jobs and
increase efficiency if the whole solar PV system.

37
REFERENCES
[1] B. Espinar, J.-L. Aznarte, R. Girard, A.M. Moussa, G. Kariniotakis,
Photovoltaic Forecasting: A state of the art, in: OTTI - Ostbayerisches
Technologie-Transfer-Institut, 2010: p. Pages 250-255- ISBN 978-3-941785-
15-1.https://hal-mines-paristech.archives-ouvertes.fr/hal-
00771465/document (accessed April 14, 2018).
[2] V. Lara-Fanego, J.A. Ruiz-Arias, D. Pozo-Vázquez, F.J. Santos-Alamillos, J.
Tovar-Pescador, Evaluation of the WRF model solar irradiance forecasts in
Andalusia (southern Spain), Sol. Energy. 86 (2012) 2200–2217.
doi:10.1016/j.solener.2011.02.014 (accessed April 14, 2018).
[3] A. Moreno-Munoz, J.J.G. De la Rosa, R. Posadillo, F. Bellido, Very short term
forecasting of solar radiation, in: 33rd IEEE Photovolt. Spec. Conf. 2008 PVSC
08, 2008: pp. 1–5. doi:10.1109/PVSC.2008.4922587 (accessed April 14, 2018).
[4] D. Anderson, M. Leach, Harvesting and redistributing renewable energy:
on the role of gas and electricity grids to overcome intermittency through the
generation and storage of hydrogen, Energy Policy. 32 (2004) 1603–1614.
doi:10.1016/S0301-4215(03)00131-9 (accessed April 14, 2018).
[5] M. Paulescu, E. Paulescu, P. Gravila, V. Badescu, Weather Modeling and
Forecasting of PV Systems Operation, Springer London, London, 2013.
http://link.springer.com/10.1007/978-1- 4471-4649-0 (accessed April 14,
2018).
[6] H.M. Diagne, P. Lauret, M. David, Solar irradiation forecasting: state-of-
the-art and proposition for future developments for small-scale insular grids,
in: n.d. https://hal.archives- ouvertes.fr/hal-00918150/document (accessed
April 14, 2018).

38
[7] A. Hammer, D. Heinemann, E. Lorenz, B. Lückehe, Short-term forecasting
of solar radiation: a statistical approach using satellite data, Sol. Energy. 67
(1999) 139–150. doi:10.1016/S0038- 092X(00)00038- 4 (accessed April 14,
2018).
[8] E. Lorenz, J. Remund, S.C. Müller, W. Traunmüller, G. Steinmaurer, D. Pozo,
J.A. Ruiz-Arias, V.L. Fanego, L. Ramirez, M.G. Romeo, others, Benchmarking of
different approaches to forecast solar irradiance, in: 24th Eur. Photovolt. Sol.
Energy Conf. Hambg. Ger., 2009: p. 25. http://task3.iea-
shc.org/data/sites/1/publications/24th_EU_PVSEC_5BV.2.50_lorenz_final.pdf
(accessed April 14, 2018).
[9] M. Saguan, Y. Perez, J.-M. Glachant, L’architecture de marchés électriques :
l’indispensable marché du temps réel d’électricité, Rev. Déconomie Ind.
(2009) 69–88. doi:10.4000/rei.4053 (accessed April 14, 2018).
[10] B. Elliston, I. MacGill, The potential role of forecasting for integrating
solar generation into the Australian national electricity market, in: Sol. 2010
Proc. Annu. Conf. Aust. Sol. Energy Soc., 2010.
http://solar.org.au/papers/10papers/10_117_ELLISTON.B.pdf (accessed April
14, 2018).
[11] E. Lorenz, J. Kühnert, D. Heinemann, Short term forecasting of solar
irradiance by combining satellite data and numerical weather predictions, in:
Proc. 27th Eur. Photovolt. Sol. Energy Conf. Valencia Spain, 2012: pp. 4401–
440.
http://meetingorganizer.copernicus.org/EMS2012/EMS2012-359.pdf (accessed
April 14, 2018).
[12] D. Heinemann, E. Lorenz, M. Girodo, Forecasting of solar radiation, Sol.
Energy Resour. Manag. Electr. Gener. Local Level Glob. Scale Nova Sci. Publ.
N. Y. (2006). https://www.uni-
oldenburg.de/fileadmin/user_upload/physik/ag/ehf/enmet/publications/solar/c
onference/200 5/Forcasting_of_Solar_Radiation.pdf (accessed April 14, 2018)).

39
[13] R. Perez, S. Kivalov, J. Schlemmer, K. Hemker Jr., D. Renné, T.E. Hoff,
Validation of short and medium term operational solar radiation forecasts in
the US, Sol. Energy. 84 (2010) 2161–2172. doi:10.1016/j.solener.2010.08.014
[14] M. Diagne, M. David, P. Lauret, J. Boland, N. Schmutz, Review of solar
irradiance forecasting methods and a proposition for small-scale insular grids,
Renew. Sustain. Energy Rev. 27 (2013) 65–76. doi:10.1016/j.rser.2013.06.042
[15] E. Lorenz, A. Hammer, D. Heinemann, Short term forecasting of solar
radiation based on satellite data, in: EUROSUN2004 ISES Eur. Sol. Congr.,
2004: pp. 841–848. https://www.uni-
oldenburg.de/fileadmin/user_upload/physik/ag/ehf/enmet/publications/solar/c
onference/200
4/eurosun/short_term_forecasting_of_solar_radiation_based_on_satellite_data.
pdf (accessed April 15, 2018).
[16] G. Reikard, Predicting solar radiation at high resolutions: A comparison of
time series forecasts, Sol. Energy. 83 (2009) 342–349.
doi:10.1016/j.solener.2008.08.007 (accessed April 15, 2018).
[17] Mohammed, Azhar Ahmed,Aung, Zeyar, Ensemble learning approach for
probabilistic forecasting of solar power generation (accessed April 15, 2018).
[18] Cyril Voyant, Gilles Notton, Soteris Kalogirou, Marie-Laure Nivet,
Christophe Paoli, Fabrice Motte, Alexis Fouilloy, Machine Learning methods
for solar radiation forecasting: a review (accessed April 15, 2018).
[19] Jawaid, Faizan,NazirJunejo, Khurum, Predicting daily mean solar power
using machine learning regression techniques (accessed April 15, 2018).
[20] Haupt, Sue Ellen,Kosovic, Branko, Big Data and Machine Learning for
Applied Weather Forecasts: Forecasting Solar Power for Utility Operations.

40
[21] J. Remund, R. Perez, E. Lorenz, Comparison of solar radiation forecasts
for the USA, in: Proc 23rd Eur. PV Conf., 2008: pp. 1–9. http://ww.w.iea-
shc.org/data/sites/1/publications/Comparison_of_USA_radiation_forecasts.pdf
[22] COST | About COST, (n.d.). http://www.cost.eu/about_cost (accessed April
15, 2018).
[23] R.H. Inman, H.T.C. Pedro, C.F.M. Coimbra, Solar forecasting methods for
renewable energy integration, Prog. Energy Combust. Sci. 39 (2013) 535–576.
doi:10.1016/j.pecs.2013.06.002 (accessed April 15, 2018).
[24] S.K. Aggarwal, L.M. Saini, Solar energy prediction using linear and non-
linear regularization models: A study on AMS (American Meteorological
Society) 2013–14 Solar Energy Prediction Contest, Energy. 78 (2014) 247–256.
doi:10.1016/j.energy.2014.10.012 (accessed April 15, 2018).
[25] Z. Şen, Ş.M. Cebeci, Solar irradiation estimation by monthly principal
component analysis, Energy Convers. Manag. 49 (2008) 3129–3134.
doi:10.1016/j.enconman.2008.06.006. [26] M. Zarzo, P. Martí, Modeling the
variability of solar radiation data among weather stations by means of
principal components analysis, Appl. Energy. 88 (2011) 2775–2784.
doi:10.1016/j.apenergy.2011.01.070 (accessed April 15, 2018).
[27] C. Paoli, C. Voyant, M. Muselli, M.-L. Nivet, Forecasting of preprocessed
daily solar radiation time series using neural networks, Sol. Energy. 84 (2010)
2146–2160. doi:10.1016/j.solener.2010.08.011 (accessed April 15, 2018).
[28] D.J. Hand, Data Mining: Statistics and More?, Am. Stat. 52 (1998) 112–
118. doi:10.1080/00031305.1998.10480549 (accessed April 15, 2018).

Solar Energy Output Forecasting from SolarGIS Data for Connected Grid Station

More Related Content

Similar to Solar Energy Output Forecasting from SolarGIS Data for Connected Grid Station (20)

More from SARADINDU SENGUPTA (7)

Recently uploaded (20)

Solar Energy Output Forecasting from SolarGIS Data for Connected Grid Station