SOLAR POWER FORECASTING

USING ARTIFICIAL NEURAL

NETWORKS
 ➢ Objectives
➢ Abstract
➢ Literature Survey
➢ Social Relevance
➢ Problem Statement
➢ Existing System
Content ➢ Proposed System
➢ Hardware and Software
Requirement
➢ System Architecture
➢ Modules
➢ Implementation Details
➢ References
2
 Objectives
➢ Prediction of solar irradiation shows
some uncertainties depending on
atmospheric parameters such as
temperature, cloud amount, dust and
relative humidity.

➢ Propose an accurate prediction model

to overcome uncertainty conditions PV
power generation
Abstract
In recent years, the rapid boost of variable energy
generations particularly from wind and solar energy
resources in the power grid has led to these generations
becoming a noteworthy source of uncertainty with load
behavior still being the main source of variability. Generation
and load balance is required in the economic scheduling of
the generating units and in electricity market trades. Energy
forecasting can be used to mitigate some of the challenges
that arise from the uncertainty in the resource. Solar power
forecasting is witnessing a growing attention from the
research community. The work presents an artificial neural
network model to produce solar power forecasts. Sensitivity
analysis of several input variables for best selection, and
comparison of the model performance with multiple linear
regression and persistence models are also shown.
 Introduction

➢ Forecasting/unit commitment for day ahead system helps the generating

station engineer to properly manage the power demand and these by
maintaining a balance between the generation and demand.
➢ Again, due to involvement of lots of environmental parameters such as
temperature, cloud quantity, dust exact prediction of PV power output
become a difficult task.
➢ A number of forecasting method have been introduced by many researcher
in last decade. All these forecasting are for long term prediction of solar PV
system.

5
 Literature Survey
Literature Survey-1

Title of the Paper Solar power forecasting using a hybrid

EMD-ELM method
Authors and Publication Irani Majumder et al (IEEE)

Methodology A forecasting method has been

mentioned that is contingent on a
hybrid empirical mode decomposition
(EMD) and Extreme Learning Machine
(ELM)
Merits Minimum computation

Demerits Less accuracy

6
 Literature Survey
Literature Survey-2

Title of the Paper Short-Term PV Power Forecasting Using

Adaptive Neuro Fuzzy Inference System

Authors and Publication Yash Pal et al (IEEE)

Methodology The ANFIS model (Adaptive Neuro Fuzzy

Inference Systems) is used to predict
photovoltaic power.

Merits Higher accuracy

Demerits Algorithmic complexity

7
 Literature Survey
Literature Survey-3

Title of the Paper A Solar Power Prediction Using Support

Vector Machines Based on Multi-source
Data Fusion
Authors and Publication Wang Buwei et al (IEEE)

Methodology Support vector machine based on data

fusion (SVM-DF) is proposed to run as the
main regression model.
Merits continuous space by identifying structures
in the mapping of input to output data

Demerits Larger training needed

8
 Literature Survey
Literature Survey-4

Title of the Paper A Hadoop based weather prediction model

for classification of weather data
Authors and Publication A. K. Pandey et al (IEEE)

Methodology Fuzzy logic (FL) and artificial neural

network fuzzy interface system (ANFIS)
methods are investigated for accurate
prediction of weather data on the basis of
mean square error.

Merits Easy to compute

Demerits Higher MSE

9
 Literature Survey
Literature Survey-5

Title of the Paper Improved ANN-based algorithm for

detection and classification of faults on
transmission lines
Authors and Publication Farhana Fayaz et al (IEEE)

Methodology A feed forward ANN with six inputs and

eleven outputs has been developed for the
detection and classification of faults

Merits Easy to access

Demerits Higher MSE

10
 Literature Survey
Literature Survey-6

Title of the Paper A hybrid deep learning model for short-

term PV power forecasting
Authors and Publication P. Li et al (Applied Energy)

Methodology Presented a hybrid deep learning model

combining wavelet packet decomposition
(WPD) and LSTM networks for one-hour-
ahead PV power forecasting

Merits Accuracy

Demerits More time

11
 Literature Survey
Literature Survey-7

Title of the Paper A Novel Short-Term Photovoltaic Power

Forecasting Approach based on Deep
Convolutional Neural Network

Authors and Publication Korkmaz et al (IJGE)

Methodology used a deep CNN structure combining
with Empirical Mode Decomposition
(EMD) algorithm
Merits greatly improved the accuracy of
prediction
Demerits Computational complexity

12
 Literature Survey
Literature Survey-8

Title of the Paper A kernel extreme learning machine-

based neural network to forecast very
short-term power output of an on-grid
photovoltaic power plan
Authors and Publication Yildiz et al (Energy source and
Utilization)
Methodology applied an ANN-based system for very
short-term (2 to 4-h) PV power
forecasting
Merits Improved efficiency
Demerits Latency
13
 Literature Survey
Literature Survey-9

Title of the Paper Solar power prediction based on satellite

images and support vector machine

Authors and Publication H. S. Jang et al (IEEE)

Methodology Support Vector Machine (SVM) used

for PV forecasting using satellite
images

Merits Huge dataset can be accessed

Demerits Inaccurate prediction

14
 Literature Survey
Literature Survey-10
Title of the Paper Short-term power prediction for
photovoltaic power plants using a hybrid
improved Kmeans-GRA-Elman model based
on multivariate meteorological factors and
historical power datasets
Authors and Publication Lin et al (Energy Conversion and
Management)
Methodology developed a multivariate neural
network ensemble forecast framework,
which substantially improved the
forecast accuracy in short-term
forecasting horizons.
Merits Accuracy
Demerits Computational complexity
15
Social Relevance

➢ Allocate power demand.

➢ Future prediction.

➢ Estimate Power generation by solar.

 Problem Statement

➢ Forecasting/unit commitment for day ahead system helps the generating

station engineer to properly manage the power demand
➢ Again, due to involvement of lots of environmental parameters such as
temperature, cloud quantity, dust exact prediction of PV power output
become a difficult task.
➢ Maintaining a balance between the generation and demand is the main
concern

17
 ➢ Statistical Non-Learning
Approach Models
➢ Statistical Learning Approach
Models
➢ Intraday Forecasting
Existing System ➢ Short term forecasting
➢ Medium term Forecasting
➢ Long term forecasting
➢ Data Synthesis
➢ Data Analysis

18
 Proposed System

➢ In this work , for the classification of solar power , multilayer feed forward
network with back propagation algorithm is proposed

➢ The Artificial Neural Network using back propagation with momentum

learning and adaptive learning rate are often used for forecasting and
prediction applications.

➢ The main goal is to classify data in terms of higher accuracy.

19
 Hardware and Software Requirements
Hardware Specifications
➢ Processor INTEL® PENTIUN (R) CPU A1018
➢ Clock Speed 2.10 GHz
➢ RAM 2 GB
➢ Hard Disk 500 GB

Software Requirements
➢ Operating System Windows Xp/7.
➢ Platform Python
➢ Front End Pyspark
➢ Back end PySQL
20
 System Architecture

Data Set Test Image

Preprocessing Preprocessing

Feature Classification
ANN
Extraction Result

21 Training
 ➢ Dataset collection
➢ Pre-processing
Modules ➢ ANN classification
➢ Performance analysis

22
 Dataset Collection

➢ The datasets used in the experiments are collected from three PV plants
distributed in different areas in Southeastern China, 2019.
➢ The information of the collected datasets. Output power and
meteorological factors (i.e., temperature, cloud amount, dust and relative
humidity) are managed separately from a power plant.
➢ The power generation data are presented at five-minute intervals, whereas
the meteorological data are presented at one-hour intervals due to the
limitations of the weather collection system.

23
 Pre-Processing
➢ In order to avoid large differences in data scales that can impair the
effectiveness of the model, the data pre-processing is normalization, so
that the data are restricted in range 0 to 1. The normalization speeds up
the gradient descent to the optimal solution, increases the comparability
of the data, and improves the precision of the data. The formula is:

𝜘𝑖𝑛 − min 𝜘𝑖𝑛

𝑥𝑖𝑛 ′ =
max(𝜘𝑖𝑛 ) − min 𝜘𝑖𝑛

➢ where 𝑥𝑖𝑛 ′ is the normalized input data; 𝑥𝑖𝑛 is the origin input data(PV
output power and meteorological factors data); and max(∙) and min (∙) are
24 the maximum and minimum values of the origin input data, respectively.
 ANN Classification
➢ For classification, the Feedforward Neural Network is presented using back
propagation with momentum learning and adaptive learning rate.
➢ It has two main steps namely,
1. Feed-forward: Data from input layer is fed forward through each layer and
then output is generated in the final layer.
2. Back-propagation: This is the learning step. Once the output is generated,
the error is calculated w.r.t. the expected output and then this error is
propagated in backwards direction to adjust the weights to reduce the error.
This is the learning step.
➢ Consider the neural network shown in Figure . It has three neural units in
each layers and b1, b2, b3 are included as units (with input as 1)
for bias. Figure has one input layer, one output layer (layer L) and 2 hidden
25 layers (L-1 and L-2).
 Figure
Hidden Layers
Output Layer
Input Layer L–2=2 L–2=3 L=4

𝑤211 𝑤311 𝑤411 𝑎41

𝑥1 𝑥21 𝑥31 𝑥41
𝑤212 𝑤312 𝑤412
𝑤213 𝑤313 𝑤413

𝑤221 𝑤321 𝑤421

𝑤322 𝑤422 𝑎42
𝑥2 𝑤222 𝑥22 𝑥32 𝑥42
𝑤423
𝑤223 𝑤323
𝑤231 𝑤331 𝑤431
𝑤232 𝑤332 𝑤432
3
𝑥3 𝑤233 𝑥23 3 𝑤 33 𝑥33 𝑤433 𝑥43 𝑎43
𝑏21 𝑏 1 𝑏3 𝑏41 𝑏42
2
𝑏22
b1 = 1 b2 = 1 𝑏33 b3 = 1 𝑏43
𝑏23
26
 Performance Analysis
𝑇𝑃
➢ Sensitivity = 𝑇𝑃+𝐹𝑁

𝑇𝑁
➢ Specificity = 𝑇𝑁+𝐹𝑃

𝑇𝑃+𝐹𝑁
➢ Accuracy = 𝑇𝑃+𝐹𝑃+𝑇𝑁+𝐹𝑁

2𝑇𝑃
➢ F1Score = 2𝑇𝑃+𝐹𝑃+𝐹𝑁

𝑇𝑃
➢ Precision = 𝑇𝑃+𝐹𝑃

𝑠𝑢𝑝 sup
➢ HD = max { 𝑑𝑚 𝑠, 𝑟 , 𝑑𝑚 𝑠, 𝑟 , }
𝑟𝑅 𝑠𝑆

27
 Data Set
S.No A B C D
1 DATE_TIME DC_POWER MODULE DATA
2 15-5-202 34.875 22.35346 0
3 15-5-202 278 22.89328 0
4 15-5-202 614.875 24.44244 0
5 15-5-202 1166.857 27.18565 0
6 15-5-202 0661.5 28.88848 0
7 15-5-202 1856.375 29.60564 0
8 15-5-202 1842.286 29.54711 0
9 15-5-202 1877.875 31.41254 0
10 15-5-202 3246 35.52871 0
11 15-5-202 3917.5 40.31806 0
12 15-5-202 4322 39.08195 1
13 15-5-202 4257.125 45.00923 1
14 15-5-202 5706.714 46.61771 1
28 15 15-5-202 4015.5 39.13633 0
 Data Set

S.No A B C D
DATE_TIME DC_POWER MODULE DATA
16 15-5-202 3219.286 40.930558 0
17 15-5-202 6454.429 52.54774 1
18 15-5-202 5286.5 47.63374 1
19 15-5-202 6008.375 49.24972 1
20 15-5-202 6637.286 44.04833 1
21 15-5-202 4399.75 47.68545 1
22 15-5-202 5388.375 50.00699 1
23 15-5-202 6829.5 49.84136 1
24 15-5-202 8617.5 47.83638 1
25 15-5-202 6226.125 49.18358 1

29
 Data Processing

30
 Plot of generated power
Histogram of DC_POWER
7

0 2000 4000 6000 8000 10000

31
 Temperature Plot
Histogram of Temperature
10

25 30 35 40 45 50 55

32
 Comparison

S.No Method MAE Specificity

1 MLR 0.214 0.867

2 Persistent 0.162 0.912

3 Proposed 0.099 0.978

33
 Conclusion
➢ The artificial neural networks model outperforms the multiple linear
regression analysis MLR model and the persistence model.
➢ The performance of the ANN depends on how well it is trained and on the
quality of the data that is used.
➢ The feed-forward ANN with 14 weather variables and with hourly step size
for forecasts performed better than the recursive neural networks.
➢ The normalized input data doesn’t improve the performance, but removing
the night hours slightly improves the model performance.
➢ Plotting the data, investigating the correlation and sensitivity analysis
between the variables, as well as data cleansing of outliers are essential
34 data preparation steps before building the forecasting model.
 Reference
➢ [1] D. Giorgi, M. P. Congedo, and M. Malvoni. "Photovoltaic power forecasting using statistical methods: impact of
weather data." IET Science, Measurement & Technology, vol.8, pp. 90-97, 2014.

➢ [2] J. Antonanzas, N. Osorio, R. Escobar, "Review of photovoltaic power forecasting", Solar Energy, vol.136, pp.78-111,
2016.

➢ [3] S. A. Karim, B. S. Singh, R. Razali and N. Yahya, "Data Compression Technique for Modeling of Global Solar
Radiation," 2011 IEEE International Conference on Control System, Computing and Engineering, 2011, pp.348-352

➢ [4] R Xiao, J Wang and F Zhang, "An Approach to Incremental SVM Learning Algorithm," 12th IEEE International
Conference on Tools with Artificial Intelligence, 2000, pp. 268- 273.

➢ [5] N. A. Gounden, S. A. Peter, H. Nallandula, "Fuzzy logic controller with MPPT using line-commutated inverter for
three-phase grid-connected photovoltaic systems," Renewable Energy,vol.34, pp.909-915,2009.

➢ [6] B. Wolff, J. Kühnert, E. Lorenz, O. Kramer, and D. Heinemann, "Comparing support vector regression for PV
power forecasting to a physical modeling approach using measurement, numerical weather prediction, and cloud
motion data," Solar Energy, vol.135, pp.197-208, 2016.

➢ [7] B. Wolff, E. Lorenz, and O. Kramer, "Statistical learning for short-term photovoltaic power predictions,"
35 Computational Sustainability,vol.19, pp.31- 45, 2016
Thankyou