2022 Third International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT)

PREDICTION OF SOLAR POWER

GENERATION BASED ON MACHINE
2022 Third International Conference on Intelligent Computing Instrumentation and Control Technologies (ICICICT) | 978-1-6654-1005-2/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICICICT54557.2022.9917846

LEARNING ALGORITHM
Rinshy Annie Varughese Dr. R. Karpagam
Department of Electrical and Department of Electrical and
Electronics Engineering Electronics Engineering
Easwari Engineering College Easwari Engineering College
Chennai, India Chennai, India
[email protected] [email protected]

Abstract— Energy demand is growing and by 2050 in the process of developing alternative energy that are safe
solar energy will account for 11% total electricity and environmentally friendly for human beings. The
production. It has emerged as one of the most potential renewable energy sources are the future of the earth as it does
sources of alternative energy Even though the usage of not cause any adverse side effects that can be harmful to the
solar energy in residential places has increased, yet they earth. From energy point of view solar panels stand out for
are regarded as unpredictable and irregular power their environmentally friendly characteristics. It is abundant,
sources because the generated power output depends on free, green, renewable, and sustainable. Input and output
the geographical region, atmospheric conditions, which parameters can be linked using the machine learning
can vary drastically. Depending upon the weather technique.
conditions solar panels will work differently. Since the This makes the model suitable for data extraction and
power generation mostly depends on weather conditions predictive classification through model recognition. The
it is necessary to consider weather conditions. Because of production of solar energy depends primarily on the weather
the unpredictability in photovoltaic generations, it is conditions in the area where the panels are installed.
crucial to examine the effects of environmental However, the output power has daily periodicity, seasonal
circumstances on solar power system using machine and random fluctuations under various environmental
learning based approach. The machine learning weather conditions. These uncertain weather conditions make
algorithm shows great results in anticipating the power it difficult to reliably produce solar energy. Because each
with weather conditions as input models. The approach panel must respond to changes in electricity demand over
uses different databases, input, and mathematical time, solar panels must be able to predict how much power
relationships to predict the solar power generated. they will produce in future and respond accordingly to
Various machine learning algorithm would be applied to changing electricity demands.
get the patterns and to obtain the results with maximum Due to the uncertainty with the production of solar power it
accuracy and efficiency. This study demonstrates how a has led to significant costs for the operation, planning, and
variety of machine learning techniques may be used to market for power systems therefore, accurate forecasting of
predict the amount of energy a solar panel provides. the amount of solar power produced is required. Numerous
Various models were applied to the database and the most studies have been done to predict as accurately as possible the
appropriate machine learning predictive model was amount of solar power produced. However, solar panel exhibits
identified through coefficient of determination analysis. ‘intermittent’ effect depending on weather conditions such as
The results obtained after comparing the data for cloudiness, humidity, and precipitation. Due to the
different years are furnished. Temperature, relative intermittent and poor predictability, PV systems can cause
humidity, pressure, and wind speed are the independent fluctuations in the power produced.
factors, with power generated as the dependent variable. Solar radiation monitoring technology is not extensively used
The proposed model has provided prediction results with due to its high cost. The sensor employed in weather stations,
good accuracy. as well as the costs of maintenance and calibration, cannot be
completely relied upon to detect solar radiation. As a result,
various other factors must be considered when predicting the
Keywords—Machine learning, Photovoltaic, Root amount of solar electricity generated. Several ML algorithms
mean square error, Random Forest have been widely utilized to forecast daily power produced
by a solar panel. Machine algorithms are popular because
they are more precise than empirical models at making
I. INTRODUCTION
predictions. Accurate forecasts are needed to reduce the
Increase in demand for energy will lead to extinction impact of power generation uncertainty and maximize the
of fossil fuel and the increase in usage can lead to economic benefits of PV systems.
environmental pollution and global warming. Researchers are

978-1-6654-1005-2/22/$31.00
Authorized licensed ©2022 IEEE
use limited to: Rajshahi University 396Downloaded on May 28,2024 at 11:37:46 UTC from IEEE Xplore. Restrictions apply.
Of Engineering and Technology.
 2022 Third International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT)

II. LITERATURE REVIEW literature include Support Vector Regression (SV), RF etc. are
Four tree-based approaches were examined by the widely used ML algorithm. The paper is organized in the
author [1] for solar intensity estimate (bagging, gradient manner outlined below. Ist section introductory, IInd section
boosting, random forest, and decision tree) utilizing Multi- reviews literature, and IIIrd section contains system model.
layer Perceptron (MLP) and Support Vector Regression Prediction approaches Section IV, performance evaluation
(SVR). The tree-based approach is straightforward and Section V, simulation and findings Section VI, and
accurate in its calculation of solar intensity. Yang et.al [2] two conclusion Section VII.
years data were used on 68 machine learning models for
hourly solar forecast and the result showed that tree- based III. SYSTEM MODEL
methods performed well. Support vector machine (svm) has
improved the existing ML model. Zeng and Qiao et al [3] Determine the solar power generated based on
estimated a short-term model based on the atmospheric various meteorological factors like relative humidity,
permeability using the method of least square SVM model. temperature, pressure, and wind speed. These historical
Photovoltaic one day ahead power output prediction using weather data (temperature, relative humidity, wind speed,
time series method were proposed by Yang and Dong et.al pressure) from 2017 to 2021 was collected from NASA site
[4]. The accuracy of ensemble-based forecast method can be where the longitude and latitude of the area was provided, and
improved by numerical weather prediction method. Two the data of power generated from a 20-watt panel was also
factors help to improve the prediction model which are solar included. In addition to weather data, day, month, and year
irradiance and weather parameters. Traditional time series was also provided. To validate the algorithm, the data from
prediction algorithms can only be used if the data is small. 2017 to 2020 is used as training set used and 2021 is used as
The author [5] proposed a SVM based technique for next-day the test set. The actual and predicted power plot value are
solar forecasting that considers a variety of factors such as plotted on a graph for the year 2021.The amount of power to
intensity and duration. For selecting the ideal parameters for be generated in 2022 is also predicted by using last five years
prediction, Jiang et al. [6] suggested a Support Vector dataset. Figure 1 shows the process flow of solar power
Machine technique with a radial basis function and a swarm prediction
optimization algorithm. D.P. Larson et al. looked at weather-
based forecasting algorithms that used historical weather data IV. PREDICTION TECHNIQUES
and earlier measurements of solar panel outputs in
meteorological data. S.G. Kim et.al [9] considered Five machine learning-based models are deployed
environmental features like weather and air pollution are namely:1. Decision tree algorithm; 2. Gradient boosting
considered for predicting the performance of solar panels. regressor.3. Random forests; 4. Linear regression; 5. Support
M.H Bergin [10] used weather features to anticipate the vector regression. Grouping data by time significantly
amount of solar electricity generated. Support vector increases the accuracy of the prediction results.
regression was proposed by Nageem et al. [8] to determine
the output of grid-connected solar panels. Various • Decision tree regressor: Regression is built in the
environmental parameters were all taken into consideration. tree structure form. The dataset is split into smaller
In M. M. Howlader [11] et.al proposed a Support vector subgroup and the decision tree is build.[21]
regression-based method based on the parameters like solar
radiation, temperature, coverage area of cloud, wind speed, • Gradient boosting [17]: An ensemble model that
atmospheric pressure, and tilt angle of the solar panel. builds a stage wise model using decision tree as
Hossain et.al,[14] proposed a forecasting method to forecast weak learner and by optimizing the loss function.
the solar power for the coming days with hourly resolution. The best possible next model, when combined with
An ensemble method has been proposed to use several previous model, minimizes prediction error.
meteorological and astronomical data to forecast solar energy
generation on an hourly basis for the coming day. Apart from • Random forest regressor: Random Forest [12] is a
the direct method, X. G. Agoua et al. [15] have found that method based on ensembles. Random forest is built
machine learning can be used to determine parameters for on the idea thar when a group of weak learners is
autonomous feature selection. Various statistical techniques united, they grow stronger.
such as the time series methods, regression methods,
regression trees were used by Y. K. Semero et.al.[16]. In • Linear regression is a supervised learning algorithm
statistical models, the output power is forecasted by depends on two continuous variables. This
analyzing the input variables which includes historical data regression method recognizes a linear relationship
and meteorological parameters. The statistical methods, between x (input) and y (output).
which are simple to use and play a significant part in solar
power forecasting. Supervised ML algorithms like random • Support vector regression [16] is a supervised
forest (RF) and support vector regression (SVR), as well as learning algorithm. Discrete values are predicted
the gradient boosting regressor, are becoming increasingly using this algorithm. Follows the same principle as
popular for building models due to their high accuracy. Support Vector Machine. SVR is used to determine
Machine learning in recent years, it has become increasingly the suitable fit line.
popular to predict solar power generation, achieving
acceptable performance levels. The widely used methods in

978-1-6654-1005-2/22/$31.00
Authorized licensed ©2022 IEEE
use limited to: Rajshahi University 397Downloaded on May 28,2024 at 11:37:46 UTC from IEEE Xplore. Restrictions apply.
Of Engineering and Technology.
 2022 Third International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT)

SOURCE DATA

MEASURE TEMPERATURE, RELATIVE

HUMIDITY, PRESSURE, WIND SPEED
•

DATA PROCESSING AND

CLEANING

TRAINING TESTING
DATA DATA

Fig. 2. Actual and anticipated results plot of April month 2021

MACHINE LEARNING PERFORMANCE

ALGORITHM METRICS
• Decision tree
• Gradient boosting MAE, MSE, R2
•• Random forest
•
• Linear regression
•
• Support vector
• FINDING SOLAR
regression
• POWER GENERATED

Fig. 1. Process flow of the solar power prediction

V. PERFORMANCE EVALUATION
The proposed method is a tool for forecasting solar power Fig. 3. Actual and anticipated results plot of February month 2021
one hour ahead of time. Various algorithms are compared
based on their performance evaluation metrics. Mean
Absolute Error (MAE), Mean Squared Error (MSE), and R2
(coefficient of determination) are some of the metrics used in
the model to assess the accuracy of each prediction approach.
The MSE is a statistic that measures the difference between
the expected and measured value power. In other words, it
indicates the match between the actual power generated and
the expected power generated.

VI. RESULTS OF SIMULATION

The simulation was completed, and the results were

obtained. Five algorithms were applied on the dataset. The
Figure 2 and 3 shows the actual and anticipated results power
plot of April and February month respectively 2021. Green
color represents true and red color represents predicted
values. Figure 4 represents the power prediction for the Fig. 4. Predicted power plot of April 2022
month of April 2022.Hour vs Power plot for the month of
April 2022 is shown in Figure 5. Performance metrics is
shown in Table I and Table II.

978-1-6654-1005-2/22/$31.00
Authorized licensed ©2022 IEEE
use limited to: Rajshahi University 398Downloaded on May 28,2024 at 11:37:46 UTC from IEEE Xplore. Restrictions apply.
Of Engineering and Technology.
 2022 Third International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT)

VII. CONCLUSION
Five machine learning (ML) models which includes
Linear regression, decision tree, random forest, gradient
boosting regressor and Support vector regression has been
evaluated using five years datasets for modelling the solar
power generation prediction. One hour ahead solar power
prediction has been proposed. The performance of each
algorithm has been evaluated by the performance metrics
used. In this study, three indices (R2, MSE, MAE) are used
to evaluate how well machine learning algorithms work.
According to the findings of the investigation, the following
conclusions can be drawn. Following conclusions can be
made based on the results. In terms of R2, the algorithms
utilized have shown to be successful. The study's R2 values
ranged from 48 to 88 percent. It is found that gradient
boosting with a R2 value of 88% shows better result as
compared to other algorithms Support vector regression has
Fig. 5. Hour vs Power plot of April month the highest error when the model is compared to another
algorithm.

TABLE I. R2 (%) of algorithm used ACKNOWLEDGEMENT

S.NO ALGORITHM R2 (%) The authors wish to thank Department of Science &
Technology, Government of India for funding the Research
1. Random forest 87 Infrastructure under the Scheme entitled “Funds for the
Improvement of S&T Infrastructure (DST-FIST)” Ref. No.
SR/FST/ College – 110/2017
2. Gradient boosting 88

REFERENCES
3. Linear regression 53
[1] Hassan, Muhammed A.“Exploring the Potential of Tree-Based
Ensemble Methods in Solar Radiation Modeling.” Applied Energy, vol.
203, Oct. 2017, pp. 897–916,4 Nov.2019.
4. Support vector regression 48
[2] Yagli, Gokhan Mert,“Automatic Hourly Solar Forecasting Using
Machine Learning Models.” Renewable and Sustainable Energy
Reviews, vol. 105, May 2019.
5. Decision tree 80
[3] Zeng, Jianwu, and Wei Qiao. “Short-Term Solar Power Prediction
Using a Support Vector Machine.” Renewable Energy, vol. 52, Apr.
2013, pp. 118–127, 10.1016/j.renene.2012.10.009.
[4] Yang, Dazhi, and Zibo Dong. “Operational Photovoltaics Power
Forecasting Using Seasonal Time Series Ensemble.” Solar Energy, vol.
TABLE II. MAE and MSE 166, May 2018.
[5] Belaid, S., and A. Mellit. “Prediction of Daily and Mean Monthly
S.NO ALGORITHM Mean absolute Mean squared Global Solar Radiation Using Support Vector Machine in an Arid
error error Climate.” Energy Conversion and Management, vol. 118, June 2016.
[6] Jiang, He, and Yao Dong. “A Nonlinear Support Vector Machine
1 Random forest 0.0074 0.0002 Model with Hard Penalty Function Based on Glowworm Swarm
Optimization for Forecasting Daily Global Solar Radiation.” Energy
. Conversion and Management, vol. 126, Oct. 2016, pp. 991–1002.
2 Gradient 0.0080 0.0002 [7] Larson, David P.“Day-Ahead Forecasting of Solar Power Output from
. Photovoltaic Plants in the American Southwest.” Renewable Energy,
boosting vol. 91, June 2016, pp. 11–20, 2019.
3 Linear 0.0214 0.0008 [8] Nageem, Ruby, and Jayabarathi R. “Predicting the Power Output of a
. Grid-Connected Solar Panel Using Multi-Input Support Vector
regression Regression.” Procedia Computer Science, vol. 115, 2017, pp. 723–730,
4 0.0362 0.0017 2020
Support vector
. [9] Kim, Seul-Gi, “A Two-Step Approach to Solar Power Generation
regression
Prediction Based on Weather Data Using Machine Learning.”
5 Decision tree 0.0089 0.0003 Sustainability, vol. 11, no. 5, 12 Mar. 2019.
. [10] M.H. Bergin, C. Ghoroi, D. Dixit, J.J. Schauer and D.T. Shindell,
“Large Reductions in Solar Energy Production Due to Dust and
Particulate Air Pollution,” Environmental Science & Technology
Letters, Vol. 4, No. 8, pp. 339-344, 2017

978-1-6654-1005-2/22/$31.00
Authorized licensed ©2022 IEEE
use limited to: Rajshahi University 399Downloaded on May 28,2024 at 11:37:46 UTC from IEEE Xplore. Restrictions apply.
Of Engineering and Technology.
 2022 Third International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT)

[11] M. M. Howlader, M. M. Howlader, M. S. A. Khan, “Gis-based solar

irradiation forecasting using support vector regression and
investigations of technical constraints for pv deployment in Bangladesh
in 2017 4th (ICAEE) ,Sep. 2017, pp. 675-680.
[12] Lahouar, A., and J. Ben Hadj Slama. “Hour-Ahead Wind Power
Forecast Based on Random Forests.” Renewable Energy, vol. 109,
Aug. 2017.
[13] Voyant, Cyril “Machine Learning Methods for Solar Radiation
Forecasting: A Review.” Renewable Energy, vol. 105, May 2017, pp.
569–582, 10.1016/j.renene.2016.12.095.
[14] Hossain, Monowar,“Application of Extreme Learning Machine for
Short Term Output Power Forecasting of Three Grid-Connected PV
Systems.” Journal of Cleaner Production, vol. 167, Nov. 2017, pp.
395–405, 10.1016/j.jclepro.2017.08.081.
[15] Zhang, Xinmin “A Solar Time-Based Analog Ensemble Method for
Regional Solar Power Forecasting.” IEEE Transactions on Sustainable
Energy, vol. 10, no. 1, Jan. 2019.
[16] Agoua, Xwegnon Ghislain, “Short-Term Spatio-Temporal Forecasting
of Photovoltaic Power Production.” IEEE Transactions on Sustainable
Energy, vol. 9, no. 2, Apr. 2018, pp. 538–546,
10.1109/tste.2017.2747765.
[17] R. Al-Hajj, A. Assi and M. M. Fouad, "Stacking-Based Ensemble of
Support Vector Regressors for One-Day Ahead Solar Irradiance
Prediction," 2019 8th International Conference on Renewable Energy
Research and Applications (ICRERA), 2019, pp. 428-433, doi:
10.1109/ICRERA47325.2019.8996629.
[18] Tiwari, Soumya “Short Term Solar Irradiance Forecast Using
Numerical Weather Prediction (NWP) with Gradient Boost
Regression.” 2018 9th IEEE International Symposium on Power
Electronics for Distributed Generation Systems (PEDG), June 2018.
[19] Sangrody H, Zhou N and Zhang Z (2020) “Similarity based models for
day ahead solar PV generation Forecasting” IEEE ACCESS 1-1 Digital
Object Identifier 10.1109/ACCESS.2017
[20] Gautam, M “Household Energy Management Model to Maximize
Solar Power Utilization Using Machine Learning.” Procedia Computer
Science, vol. 165, 2019.
[21] Hephzibah Jose Queen, Jayakumar J, Deepika T J, K. Victor Sam
Moses Babu “Machine learning based predictive techno economic
analysis of power system” 10.1109/ACCESS.2021.3110774, IEEE
Access
[22] I. Arora, J. Gambhir and T. Kaur, "Solar Irradiance Forecasting using
Decision Tree and Ensemble Models," 2020 Second International
.
Conference on Inventive Research in Computing Applications
(ICIRCA),,pp.675-681,doi: 10.1109/ICIRCA48905.2020.9182876.
[23] Gigoni, Lorenzo,“Day-Ahead Hourly Forecasting of Power Generation
from Photovoltaic Plants.” IEEE Transactions on Sustainable Energy,
vol. 9, no. 2, Apr. 2018.
[24] Obando, Edgar Dario,” Solar Radiation Prediction Using Machine
Learning Techniques: A Review.” IEEE transactions, vol.17, no. 04, Apr.
2019, pp. 684-697, 10.1109/tla.2019.8891934.Accessed 19 Apr.2022

978-1-6654-1005-2/22/$31.00
Authorized licensed ©2022 IEEE
use limited to: Rajshahi University 400Downloaded on May 28,2024 at 11:37:46 UTC from IEEE Xplore. Restrictions apply.
Of Engineering and Technology.