© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.

org (ISSN-2349-5162)

SOLAR ENERGY FORECASTING USING

MACHINE LEARNING
YERAMOLU VENKATA DURGA DEVI. PADALA SREENIVASA REDDY

PG Scholar, Department of Computer Science, Associate Professor in Computer Science,

SVKP & Dr K S Raju Arts & Science SVKP & Dr K S Raju Arts & Science College,
College, Penugonda, W.G.Dt., A.P, India Penugonda, W.G.Dt., A.P, India.

Psreddy1036@gmail.com
devyaramolu@gmail.com

ABSTRACT: gradually decreasing costs of power generation.

Solar power, in particular, has the potential to
As a result of the climate and energy crises,
account for a larger share of growing energy needs
renewable energy generation, including solar
as it becomes more cost-effective. According to
generation, has grown significantly. In smart grid,
reports, photovoltaic (PV) module costs have
photovoltaic (PV) generation is becoming
dropped by roughly four-fifths, making residential
increasingly prevalent. Because the solar source at
solar PV systems up to two-thirds cheaper than in
ground level is highly dependent on cloud cover
2010 [1]. As the cost of installing PV modules
variability, atmospheric aerosol levels, and other
decreases, the cost of operations and maintenance
atmosphere parameters, solar power is intermittent
(O&M) gradually consumes a large portion of the
and variable. Large-scale solar generation's
cost of power generation. Maintaining PV
inherent variability poses significant challenges to
operations may be considered simpler and less
smart grid energy management. Accurate
expensive than other alternative energy sources,
forecasting of solar power/irradiance is critical to
such as wind power and natural gas; however,
ensuring the smart grid's economic operation. We
because power supply is largely dependent on
employ a number of variable selection techniques,
changing weather, determining power price and
which lead us to believe that air temperature,
managing production budgets for system operators
relative humidity, and dew point are the most
and power market participants become critical
informative weather parameters for predicting
issues to keep power plants commercially viable.
solar energy generation at the power plant. A
Solar PV system operators use long and short term
tensor flow-based sequential algorithm is used in
solar power forecasts to schedule generation,
this project to train and predict solar energy
estimate operating reserves, and ensure system
models.
stability through output fluctuation. Forecasts are
INTRODUCTION: also used by market participants to manage their
generation portfolios. Given that 38 GW of solar
1.1 Introduction
capacity was traded on the energy market in
Alternative power generation has received a lot of Germany in 2014 [3,] solar power forecasting has
attention over the last decade due to the rapidly a significant impact on market price and cost-
growing interest in renewable energy and the efficiency of power generation. As a result, solar
JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d437
© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.org (ISSN-2349-5162)

power forecasting is now an important part of PV Intensifying energy demand is paving the way for
system management. Solar power forecasting the integration of renewable solar energy with non
techniques have been extensively studied not only renewable energy resources. Solar energy, unlike
in the solar power industry but also in academic other non renewable energy sources, is
communities (See recent surveys for an overview continuous. Accurate solar power forecasting is
of operating reserves and ensuring system stability required for effective utilisation of spontaneously
through output fluctuation). Forecasts are also available energy. On a solar dataset, this study
used by market participants to manage their aims to predict solar power using deep neural
generation portfolios. Given that 38 GW of solar networks (DNNs) and various machine learning
capacity was traded on the energy market in (ML) techniques such as linear regression, support
Germany in 2014 [3,] solar power forecasting has vector regression, random forest, and so on. The
a significant impact on market price and cost- dataset is used to extract solar power energy every
efficiency of power generation. As a result, solar five minutes. Furthermore, a comparison study is
power forecasting is now an important part of PV conducted between DNNs and ML techniques,
system management. Solar power forecasting which aids in crafting appropriate decisions to
techniques have been extensively researched not select appropriate forecasting and prediction
only in the solar power industry but also in techniques.
academic circles ( An overview can be found in
This study proposed a combination of different
recent surveys.
ML and DL techniques that were used to
Traditional forecasting techniques have mostly determine the best combination in order to obtain
relied on physical models that calculate solar accurate results that could then be compared to the
power based on irradiation or a simple linear/non- results of other models. A hierarchical and layered
linear regression model. However, due to the non- approach was used in this study to determine the
linear dependence of solar power generation closest possible solar radiance value from the solar
efficiency on meteorological variables such as irradiance value of the previous few days. SVR
irradiation and temperature, as well as irregular and GB appear to perform the best on the given
errors in input data exhibited by inaccurate dataset, with roughly equivalent scores, of all the
sensors, they require extensive pre-processing to techniques discussed. Although GB outperforms
refine input data and suffer from poor accuracy SVR in terms of output stability, SVR has more
with incomplete inputs. Thus, deep learning, a variations in output compared to similar inputs,
cutting-edge machine learning technique based on making it less stable. Seasonal inputs do not
artificial neural networks that has achieved appear to work well with DL models. The study
significant performance improvements in a variety gave users the option of using one of several
of prediction problems, has recently been models to forecast solar irradiance based on
introduced for forecasting solar power. previous data and predict solar radiance. This
method assisted in determining a specific attribute
or feature independently of other attributes from
2. LITERATURE SURVEY
JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d438
© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.org (ISSN-2349-5162)

any other model if the value of that attribute had critical to ensuring the economic operation of the
previously been measured. As a result, predicting smart grid.
solar radiance value is independent of a single
3.2 Proposed system:
model. In short, the best algorithm for solar power
prediction can be determined using various ML • In the proposed system, a tensorflow sequential
and DL techniques. Hyper parameter optimization model algorithm is used to predict solar energy
and other techniques can also be used to carry out forecasting. Unlike statistical models and AI
additional modifications based on the best-suited techniques, physical models use solar and PV
algorithm in order to improve accuracy even models to predict solar irradiance/power.
further.
Advantages:

• The accuracy of predicting solar energy is high

3. OVERVIEW OF THE SYSTEM when compared to previous models.

3.1 Existing System • Applications of solar forecasting in energy

management using Trained model can improve
Existing system:
effective management of solar energy.
• A model based on least-square support vector 3.3 System Modules
machine (LS-SVM) is proposed for short-term
Dataset Collection:
solar power prediction. The LSSVM model
outperforms a reference AR model in terms of This module collects a dataset from Kaggle that
forecasted atmospheric transmissivity, which is includes weather conditions as features and power
converted to solar power. output as labels.

• For forecasting mean hourly global solar Cloud Coverage ; Visibility ; Temperature ; Dew
radiation, several AI techniques, including linear, Point ; Relative Humidity ; Wind Speed ; Station
feed-forward, recurrent Elman, and Radial Basis Pressure ; Altimeter
Function NNs, as well as the adaptive neuro-fuzzy
Preprocessing:
inference scheme, are proposed.
Unwanted values are removed in this step. The
dataset is removed, and the features and labels are

Disadvantages: processed in order to save them in various

variables.
• Because of the inherent variability of large-scale
Splitting Data:
solar generation, smart grid energy management
faces significant challenges. To test the efficiency of the algorithm, the data set
is divided into testing and training sets, which are
• Because current system forecasting is inaccurate,
used for fitting into the algorithm and predicting
effective forecasting of solar power/irradiance is
based on the test set and calculating accuracy.

JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d439
© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.org (ISSN-2349-5162)

Algorithm Initialization: Test set:

Tensor flow sequential model is used in this step

to train the model by fitting testing set features
and labels into the algorithm using the fit function,
and the model is saved in.h5 format.

Prediction:

Prediction is performed based on user-supplied

input values, and the result is displayed on a web
page.

Input page:

4. RESULTS

Data set:

Enter values:

Result:

5. Conclusion

The use of the Neural Network technique in solar

radiation modelling and prediction is studied using
datasets. According to the findings, the proposed
ANN model has a high level of accuracy. The

JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d440
© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.org (ISSN-2349-5162)

model takes into account input parameters such as Energy, vol. 67, no. 1-3, 1999, pp. 139–
average temperature, maximum temperature, 150., doi:10.1016/s0038-092x(00)00038-4.
minimum temperature, and altitude and produces 3. http://s35695.mini.alsoenergy.com/Dashbo
solar radiation as an output. A comparison of the ard/2
proposed and measured data demonstrates that the a5669735065572f4a42454b772b714d3d
model can predict with new data. This model will 4. https://machinelearningmastery.com/use-
be suitable for predicting solar power based on keras-deep-learning-models-scikit-learn-
various weather factors, and the prediction can be python/.
checked using the flask web application. 5. Kingma, Diederik P., and Jimmy Ba.
”Adam: A Method for Stochastic
Future Scope:
Optimization.” 3rd International
Conference for Learning Representations
(2015). Print.
The use of the Artificial Neural Network
6. Martin, R., et al. “Machine learning
technique in solar radiation modelling and
techniques for daily solar energy
prediction is investigated for four cities in India.
prediction and interpolation using
According to the findings, the proposed ANN
numerical weather models.” Concurrency
model's accuracy can be improved. The model
and Computation: Practice and Experience,
takes into account input parameters such as
vol. 28, no. 4, 2015, pp. 1261–1274.,
average temperature, maximum temperature,
doi:10.1002/cpe.3631.
minimum temperature, and altitude and produces
7. Mellit, Adel. “Artificial Intelligence
solar radiation as an output. A comparison of the
technique for modelling and forecasting of
proposed data with the measured data reveals that
solar radiation data: a review.”
the model agrees well with the IMD measured
International Journal of Artificial
data. This model is appropriate for predicting solar
Intelligence and Soft Computing, vol. 1,
power predictions for locations in India and can be
no. 1, 2008, p. 52.,
used for solar energy applications.
doi:10.1504/ijaisc.2008.021264.
References: 8. NOAA website:

1. Gensler- Janosch, A., et al. “Deep https://www.ncdc.noaa.gov/cdo-

Learning for solar power forecasting — web/datatools/l cd

An approach using Auto Encoder and

LSTM Neural Networks.” 2016 IEEE ABOUT AUTHORS:

International Conference on Systems,

Man, and Cybernetics (SMC), 2016 YERAMOLU VENKATA DURGA DEVI. is

2. Hammer, A., et al. “Short-Term currently pursuing MCA in SVKP & Dr K

forecasting of solar radiation: a statistical S Raju Arts & Science College, affiliated,

approach using satellite data.” Solar to Adikavi Nannaya University,

Rajamahendravaram. His research interests
JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d441
© 2021 JETIR August 2021, Volume 8, Issue 8 www.jetir.org (ISSN-2349-5162)

include Data Structures Web

Technologies, Operating Systems and
Artificial Intelligent.

PADALA SREENIVASA REDDY is

working as an Associate in the Department
of Computer Science in SVKP & Dr K S
Raju Arts &Science College, Penugonda,
A.P. He received Master’s Degree in
Computer Applications from Andhra
University A.P. He attended and presented
papers in conferences and seminars. His
areas of interests include Operational
research , Probability and Statistics,Design
and Analysis of Algorithm, Big Data
Analytics

JETIR2108426 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org d442