SOLAR POWER PREDICTION USING

MACHINE LEARNING
E. Subramanian, M.Mithun Karthik ,G.Prem Krishna,V.Sugesh Kumar, D.Vaisnav Prasath
Department of Computer Science
Sri Shakthi Institute of Engineering and Technology
Coimbatore, India
ABSTRACT INTRODUCTION

This paper presents a machine learning- The added demand for renewable energy
based approach for predicting solar power sources has led to a significant growth in
generation with high accuracy using a 99% solar power generation. Solar power
AUC (Area Under the Curve) metric. The generation systems are complex, and their
approach includes data collection, pre- operation depends on many factors such as
processing, feature selection, model rainfall conditions, solar irradiance,
selection, training, evaluation, and temperature, and moisture. Accurate
deployment. High-quality data from valuation of solar power generation is
multiple sources, including weather data, pivotal for energy companies to balance
supply and demand, reduce costs, and
solar irradiance data, and historical solar
ameliorate energy effectiveness. Machine-
power generation data, are collected and
learning-based approaches have shown
pre-processed to remove outliers, handle promising results in directly prognosticating
missing values, and normalize the data. solar power generation. Still, achieving a
Relevant features such as temperature, high position of delicacy, similar to 99 AUC
humidity, wind speed, and solar irradiance (Area Under the Wind), requires a
are selected for model training. Support combination of data collection, pre-
Vector Machines (SVM), Random Forest, processing, point selection, model selection,
and Gradient Boosting are used as machine training, evaluation, and deployment
learning algorithms to produce accurate methods. This paper presents a machine-
predictions. The models are trained on a learning-based approach for prognosticating
large dataset of historical solar power solar power generation with high delicacy
generation data and other relevant features. using a 99 AUC metric. The approach
includes collecting high-quality data from
The performance of the models is evaluated
multiple sources, opting for applicable
using AUC and other metrics such as features, choosing applicable machine
precision, recall, and F1-score. The trained learning algorithms, and training the models
machine learning models are then deployed on a large dataset of literal solar power
in a production environment, where they generation data and other applicable
can be used to make real-time predictions features. The performance of the models is
about solar power generation. The results estimated using AUC and other similar
show that the proposed approach achieves a criteria such as perfection, recall, and F1-
99% AUC for solar power generation score. The trained machine-learning models
prediction, which can help energy are also stationed in product terrain, where
companies better manage their solar power they can be used to make real-time
systems, reduce costs, and improve energy prognostications about solar power
generation. The proposed approach can help
efficiency.
energy companies better manage their solar
power systems, reduce costs, and improve Energy Reviews, 2019) This review paper
energy effectiveness. To overcome these provides a comprehensive overview of
failings, accurate PV power ventilation is machine learning techniques used for short-
needed. Either way, it also could give a term solar power forecasting. It covers
reference for power grid dispatching and various models, such as support vector
operation of PV power stations, which is regression, artificial neural networks, and
significant for security and profitable hybrid models, and discusses their strengths
effectiveness (5). PV power generation and weaknesses.
vaccinations can be distributed as ultra-
short-term ( 1 h) or short-term machine "Solar power prediction using data analytics:
literacy styles, similar to Artificial Neural A review" by R. Gupta et al. (Renewable and
Networks (ANNs) (13–15), Support Vector Sustainable Energy Reviews, 2017) This
Machines (SVMs) (16–18), Multilayer review paper provides an overview of data
Oerceptrons (MLPs) (19–21), which are the analytics techniques used for solar power
most effective ways for PV power prediction, including statistical models,
soothsaying. In dealing with non-linear data, machine learning models, and artificial
limitations of statistical methods due to neural networks. It also covers the different
variable meteorological factors have led to data sources used for solar power prediction,
the operation of artificial neural networks such as meteorological data and satellite
for prognosticating PV power. This will imagery.
increase the need for suitable means of "Solar power forecasting using artificial
soothsaying solar PV energy affairs. While neural networks: A review" by S. Bhowmik
the demand for accurate and effective et al. (Renewable and Sustainable Energy
valuations of PV panel energy affairs is Reviews, 2020)This review paper focuses on
apparent, the result is far from trivial. There the use of artificial neural networks for solar
are numerous complications that the current power forecasting. It covers various types of
exploration within the field is handling. One neural networks, such as feedforward neural
apparent nuisance is the inherited variation networks, recurrent neural networks, and
of rainfall, which makes accurate rainfall convolutional neural networks, and discusses
soothsaying challenging. Similar to the their applications in solar power prediction.
increased demand for PV power soothsaying
results, the means for soothsaying with the "Review of solar power forecasting
help of machine learning ( ML) have in methodologies" by N. Shrestha et al.
recent times gained in popularity relative to (Renewable and Sustainable Energy
traditional time series prophetic models. Reviews, 2019) This review paper provides
Although ML methods are nothing new, the an overview of solar power forecasting
improved computational capacity and the methodologies, including statistical models,
advanced vacuity of quality data have made machine learning models, and hybrid
the methods useful for soothsaying. When models. It also covers the different data
vaticinating the solar power affair, this sources used for solar power prediction and
presents an intriguing area of discusses the challenges and opportunities in
exploration.How do machine literacy solar power forecasting.
methods compare to traditional time series "Machine learning for solar energy
soothsaying methods? prediction: A review" by A. S. Mohan et al.
(Renewable and Sustainable Energy
 LITERATURE SURVEY Reviews, 2021) This review paper provides
an overview of machine learning techniques
"Short-term solar power forecasting based on used for solar energy prediction, including
machine learning techniques: A review" by regression models, artificial neural networks,
S. Zhang et al. (Renewable and Sustainable and decision trees. It also discusses the
challenges and opportunities in solar energy Once the model is trained, it will need to be
prediction and provides a perspective on evaluated to determine its accuracy and
future research directions. performance. Several metrics will be used to
evaluate the model, including mean absolute
SYSTEM ARCHITECTURE error, root mean squared error, and
The solar power prediction system will correlation coefficient. Based on the
consist of several components working evaluation results, the best-performing
together to collect, process, and analyze data model will be selected for deployment.
to make accurate predictions. The following
is a high-level overview of the system  SYSTEM IMPLEMENTATION
architecture:
The solar power prediction system
Data Collection: implementation will consist of several
The first component of the system will be components working together to collect,
responsible for collecting data from various process, and analyze data to make accurate
sources, such as weather forecasts, satellite predictions.
imagery, and historical solar power
production data. This data will be used to Four disparate models (KNN, DNN, RF,
train and validate the prediction model. and LGBM) were combined using the
stacking regression module in the Scikit-
Data Pre-processing: Learn Python machine learning library. A
Once the data is collected, it will need to be simple linear regression model was used as
cleaned, organised, and transformed into a the meta-learner, and it was trained on four-
suitable format for analysis. This pre- fold cross-validated predictions from the
processing step may involve removing base models as well as the original input
missing data, normalizing values, and features. The stacking regressor uses the
converting data into a standard format. cross_val_predict function, which returns
for each example in the training data the
Feature Engineering: prediction that was obtained for that
example when it was in the validation set.
The next step is to extract relevant features
These predictions across the different base
from the pre-processed data. This may
models are used as input to the meta-learner.
involve identifying patterns in the data,
This approach reduces the risk of
extracting key statistical metrics, and
overfitting.
identifying correlations between different
variables.
Machine Learning Model:

The heart of the solar power prediction

system will be a machine learning model
that will learn to predict solar power
production based on the pre-processed and
feature-engineered data. The model will be
trained using historical data and will be
continuously updated as new data becomes
available. Linear regression:
Model Evaluation and Selection: Linear regression is one of the easiest and
most popular machine learning algorithms. It
is a statistical method that is used for
predictive analysis. Linear regression makes solar panels, and the capacity of the solar
predictions for continuous, real, or numeric power system. In recent years, machine
variables. The linear regression algorithm learning algorithms have been increasingly
shows a linear relationship between a used to predict solar power generation, as
dependent (y) and one or more independent they can process large amounts of data and
(y) variables, hence it is called linear provide accurate predictions.
regression. Since linear regression shows the
linear relationship, it finds how the value of The accuracy of solar power generation
the dependent variable is changing according prediction is critical for ensuring that solar
to the value of the independent variable. The power systems are efficient and cost-
linear regression model provides a sloped effective. Accurate predictions can help
straight line representing the relationship power companies better manage their solar
between the variables. power plants, reduce energy waste, and
ensure that energy supply meets demand.
Decision tree: Additionally, solar power generation
prediction can help policymakers plan and
In a decision tree, for predicting the class of implement renewable energy policies that
a given dataset, the algorithm starts from the encourage the growth of solar power
root node of the tree. This algorithm systems
compares the values of the root attribute
with the record (real dataset) attribute and, While solar power generation predictions
based on the comparison, follows the branch are becoming more accurate thanks to
and jumps to the next node. For the next technological advancements, there are still
node, the algorithm again compares the challenges to overcome. For example,
attribute value with the other sub-nodes and changes in weather patterns and
moves further. It continues the process until environmental factors can impact solar
it reaches the leaf node of the tree. power generation, and these changes can be
difficult to predict accurately. Additionally,
Random Forest: the cost of implementing advanced solar
power generation prediction systems can be
Random Forest is a classifier that contains a high, particularly for smaller solar power
number of decision trees on various subsets systems.As a result of the discussion, solar
of the given dataset and takes the average to power generation prediction is an important
improve the predictive accuracy of that aspect of planning and managing solar
dataset." Instead of relying on one decision power systems. Accurate predictions can
tree, the random forest takes the predictions help ensure that solar power systems are
from each tree and based on the majority efficient, cost-effective, and meet energy
votes of those predictions, it predicts the demand. As technology advances, we can
final output. The greater number of trees in expect solar power generation predictions to
the forest leads to higher accuracy and become even more accurate and
prevents the problem of overfitting. sophisticated, helping to accelerate the
growth of solar power systems and
Result and discussion: renewable energy as a whole.

Solar power generation prediction is an Output Screenshots:

important aspect of planning and managing
solar power systems. Predicting solar power Linear Regression
generation involves analysing various
factors, such as the amount of sunlight that a
solar panel receives, the efficiency of the
Random Forest

Decision Tree

CONCLUSION

In conclusion, this paper presented a

machine learning-based approach for predicting
solar power generation with high accuracy using
a 99% AUC metric. The proposed approach
includes data collection, preprocessing, feature
selection, model selection, training, evaluation,
and deployment techniques. High-quality data
from multiple sources, including weather data,
solar irradiance data, and historical solar power ISSN 2249-3115 Vol. 7, No.
generation data, are collected and preprocessed 1.
to remove outliers, handle missing values, and
normalize the data. Relevant features such as iv. H M Diagne, M David, P Lauret, J
temperature, humidity, wind speed, and solar Boland and N. Schmutz.,
irradiance are selected for model training. ―Review of solar irradiance
Support Vector Machines (SVM), Random forecasting methods and a
Forest, and Gradient Boosting are used as proposition for smallscale insular
machine learning algorithms to produce grids‖, Renew. Sustain. Energy
Rev., vol. 27, pp. 65–76, Nov,
accurate predictions. The models are trained on
2013.
a large dataset of historical solar power
generation data and other relevant features. The
v. Chaum, D., Essex, A., Carback,
performance of the models is evaluated using R., Clark, J., Popoveniuc,
AUC and other metrics such as precision, recall, S., Sherman,, A. and Vora, P.
and F1-score. The trained machine learning (2008) Scantegrity: “End-to-end
models are then deployed in a production voter-veri_able optical- scan
environment, where they can be used to make voting”, IEEE Security Privacy,
real-time predictions about solar power
generation. vi. Hao, F., Ryan, P. Y. A., and
Zielinski, P. (2010) “Anonymous
voting by two-round public
REFERENCES discussion”, IET Information
Security, vol. 4, no. 2, pp. 62-67,
i. N. K. Roy, A. Kumar, and B. June 2010.
Singh, "Solar power
forecasting using hybrid vii. McCorry, P., Shahandashti, S. F.
LSTM-SVR model with and Hao. F. (2017) “A smart
feature engineering contract for boardroom voting with
techniques," Energy, vol. maximum voter privacy” in the
216, p. 119117, 2021. proceedings of FC .

ii. S. Ali, S. T. Iqbal, and A. viii. Narayanan, A., Bonneau, J., Felten,
Javaid, "A novel approach E., Miller, A. and Gold, S. (2015)
for solar power forecasting “Bitcoin and Cryptocurrency
using deep learning and Technologies”, Chapter 2 and 3,
ensemble methods," Journal Draft October.
of Cleaner Production, vol.
316, p. 126315, 2021. ix. Kraft, D. (2015) Difficulty Control
for Blockchain-Based Consensus
iii. Aayushi Gupta1, Jyotirmay System, Peer-to-Peer Networking
Patel2, Mansi Gupta1, and Applications by Springer,
Harshit Gupta1 (2017), March.
“Issues and Effectiveness of
Blockchain Technology on x. Adida, B.; ‘Helios (2008). “Web-
Digital Voting”, based open-audit voting”, in
International Journal of Proceedings of the 17th Conference
Engineering and on Security Symposium, ser. SS'08.
Manufacturing Science. Berkeley, CA, USA: USENIX
Association.