Clean energy sources such as photovoltaic (PV) panels
are widely employed. However, their performance is affected by
the surroundings. A hybrid optimization technique that
comprised an ant lion optimizer (ALO) and artificial neural
network (ANN) is presented in this study, to forecast the PV cell
temperature and output power. The optimizer's major purpose
was to create and improve an ANN approach that was based on
training and forecasting. The ALO was used as MVO and GA to
obtain the optimal hidden layers neurons number, weights, and
biases, of the proposed ANNs. The accuracy of the multilayer
feed forward neural networks (MFFNN) was evaluated using the
data from the MFFNN-MVO, MFFNN-GA and MFFNN-ALO
models. The panel output power and temperature were regulated
by three variables: solar irradiation, ambient temperature, and
wind speed. The Saudi Arabia, Shaqra City PV station with 4-
kW output power is the source of the two years testing and
training. For the MFFNN-GA, MFFNN-MVO, and MFFNNALO models, the NRMSE for DC power predicting compared to
2019 observ
1. PV Solar Power Forecasting
Based on Hybrid MFFNN-ALO
Adel Alblawi1, Taghreed Said2, M.
Talaat3,4,*, M.H. Elkholy4,
1Mechanical Engineering Department, College of Engineering, Shaqra University, Ar Riyadh, Saudi Arabia
2Electrical Engineering Department, Higher Technological Institute, 10th of Ramadan City, Egypt
3Electrical Engineering Department, College of Engineering, Shaqra University, Ar Riyadh, Saudi Arabia
4Electrical Power & Machines Department, Faculty of Engineering, Zagazig University, 44519 Zagazig, Egypt.
*m_mtalaat@eng.zu.edu.eg
4. The grid management for the uncertain and variable
power generation.
Introduction
5. Introduction
The ANN is preferred over traditional methods in terms
of solar energy prediction:
The high ability to learn from previous
measurements.
Build accurate relationships between all
variables.
Predict what is new in the future.
The main problems of utilizing ANNs alone is the
stagnation in a local minimum.
7. The accuracy of the suggested MFFNNs is assessed using:
𝑁𝑅𝑀𝑆𝐸 =
1
𝑁
.
𝑖=1
𝑁
𝐸𝑖
2 /𝑣𝑎𝑟 𝑃𝐷𝐶
𝐸𝑖 = 𝑃𝐷𝐶 − 𝑃𝑖
The training's objective function (OF):
OF= min ( 𝑁𝑅𝑀𝑆𝐸)= min
1
𝑁
.
𝑖=1
𝑁
𝐸𝑖
2 /𝑣𝑎𝑟 𝑃𝐷𝐶
The minimization of the NRMSE problem is subjected to:
Weight limits. Bias limits.
Problem Formulation
8. Forecasting Model
1. Data preparation:
For the proposed Neural networks:
Solar radiation, ambient temperature, and wind
speed as the inputs to the forecasting model.
During 2018, 3566 sets of environmental
measurements were used for the three input
parameters.
In addition to the PV power or cell temperature
as the output parameter
10. Results and discussion
The ALO algorithm is relatively recent and was chosen
based on its benchmark performance.
In terms of optimization, GA is regarded as standard
because of the enormous number of real-world optimization
studies by GA.
The accuracy of the solution, convergence rate and the time
taken for solving the optimization problem are the factors
that are equally significant to the optimization technique.
11. Results and discussion
1. The predicting models training:
The process of training and testing were developed using
exact measured data from the photovoltaic module of 4-
kW output in 2018 and 2019, which has:
Incidence irradiance, ambient temperature, and wind
speed as an inputs
DC output power/cell temperature as an output.
Each hidden layer's biases were determined via training
with NRMSE as a fitness function.
12. Results and discussion
1. The predicting models training:
The best training performance of MFFNN-ALO for DC
output power as Network output.
13. Results and discussion
1. The predicting models training:
The best training performance of MFFNN-ALO for cell temperature as
Network output.
14. Results and discussion
2. The DC output power Prediction:
The MFFNN-ALO model forecasted and measured DC power output.
15. Results and discussion
2. DC output power Prediction
The comparison between MFFNN-ALO and other models
presented in [1] for DC power prediction is
The MFFNN-ALO algorithm has the most prevalent contribution.
The ALO optimizer has the advantage of minimizing NMSE by
21.7% when compared to GA and by 3% when compared to
MVO.
ANN model Input parameters
DC power output
NRMSE Time
MFFNN-ALO ambient temp., solar
irradiance, and wind
speed
6.08 E-04 3861
MFFNN-MVO 7.11E-04 4269
MFFNN-GA 2.78E-03 6487
16. Results and discussion
3. The Cell Temperature Prediction:
The MFFNN-ALO projected values vs. experimental cell temperature.
17. Results and discussion
2. DC output power Prediction
The comparison between MFFNN-ALO and other models
presented in [1] for Cell Temperature prediction.
The MFFNN-ALO algorithm has the most prevalent contribution.
The ALO optimizer has the advantage of minimizing NMSE by
78% when compared to GA and by 14.5% when compared to
MVO.
Besides efficient in equality solution MFFNN has the best
computing efficiency.
The balance between exploitation and exploration.
ANN model Input parameters
Cell temperature
NRMSE Time
MFFNN-ALO ambient temp., solar
irradiance, and wind
speed
5.77 E-03 3127
MFFNN-MVO 5.95E-03 3257
MFFNN-GA 7.37E-03 3229
18. Conclusions:
The MFFNNs-ALO proposed method is robust and effective
with high precision and less computational time.
The proposed MFFNN-ALO model was compared against
MFFNN-MVO, MFFNN-GA, and MFFNN networks.
For the MFFNN-ALO network, the NRMSE for DC power
predicting compared to 2019 observed data was 6.08.
For the MFFNN-ALO network, the NRMSE for Temperature
of PV cells predicting compared to 2019 observed data was
5.77 E-03.
19. Acknowledgement
The authors are grateful to the Saudi Arabian Ministry of
Education's Deputyship for Research and Innovation for
financing this research through project number (IFP2021-077).
20. References:
1. M. Talaat, T. Said, M. A. Essa, A. J. I. J. o. E. P. Hatata, and E. Systems,
"Integrated MFFNN-MVO approach for PV solar power forecasting
considering thermal effects and environmental conditions," vol. 135, p.
107570, 2022.