SHRISTI SHRESTHA 073BEL342
A CASE STUDY ON
Artificial Neural Network Approach to
Photovoltaic System Power Output Forecasting
Tatjana Konjić
Faculty of Electrical Engineering
University of Tuzla
Amra Jahić
Graduate School
University of Applied Sciences Offenburg
Jože Pihler
Faculty of Electrical Engineering and Computer Science
University of Maribor
I
Scope
The amount of global cumulative solar
Photovoltaic (PV) capacity amounts to hundreds of
gigawatts.
Mitigating these technical challenges for successful
integration of these PV power plants into the
system is possible when the power output of these
plants can be forecasted more accurately.
The findings of this review article is an
encouraging stance on the possibility of utilizing
modern ANN techniques for power forecasting of
Nepalese PV power plants.
II
Objective
Find Best ANN Model for Power Forecasting of a PV Power Plant Every Hour (6 AM
to 8 PM) in August 2013 Using Solar Irradiation, Ambient Temperature and Model
Temperature Data of the Plant Through July 2012 - July 2013
ANN
Model
Solar Radiation
Model Temperature
Ambient Temperature
Total Power Output
III
IV
Model Used
Multi-layer Perceptron (MLP)
V
Data Used
Dataset 1 had summer data of 2012 and 2013 (6-month Data
Each)
Dataset 2 had year-round data July 2012 - July 2013
Hourly data of three inputs (solar radiation, ambient temperature
and module temperature) and one output (power output of PV
system) from 6 am to 8 pm for years 2012 and 2013
VI
Learning Algorithm
Levenberg-Marquardt backpropagation (trainlm)
Scaled conjugate gradient backpropagation (trainscg)
Gradient descent backpropagation (traingd)
Polak-Ribier conjugate gradient backpropagation (traincgp)
Fletcher-Powell conjugate gradient backpropagation (traincgf)
BFGS Newton backpropagation (trainbfg)
VII
Findings
VIII
Model Accuracy
Confidence Interval (0.5 to 2 Times Standard Deviation)
88% to 96% for models based on 6 months database
82% to 94% for models based on 12 months database
IX
Reference Paper
Konjic, Tatjana & Jahic, Amra & Pihler, Jože. (2015). Artificial Neural Network
Approach to Photovoltaic System Power Output Forecasting.
Thank You
X

ANN Case Study.pdf

  • 1.
    SHRISTI SHRESTHA 073BEL342 ACASE STUDY ON Artificial Neural Network Approach to Photovoltaic System Power Output Forecasting Tatjana Konjić Faculty of Electrical Engineering University of Tuzla Amra Jahić Graduate School University of Applied Sciences Offenburg Jože Pihler Faculty of Electrical Engineering and Computer Science University of Maribor I
  • 2.
    Scope The amount ofglobal cumulative solar Photovoltaic (PV) capacity amounts to hundreds of gigawatts. Mitigating these technical challenges for successful integration of these PV power plants into the system is possible when the power output of these plants can be forecasted more accurately. The findings of this review article is an encouraging stance on the possibility of utilizing modern ANN techniques for power forecasting of Nepalese PV power plants. II
  • 3.
    Objective Find Best ANNModel for Power Forecasting of a PV Power Plant Every Hour (6 AM to 8 PM) in August 2013 Using Solar Irradiation, Ambient Temperature and Model Temperature Data of the Plant Through July 2012 - July 2013 ANN Model Solar Radiation Model Temperature Ambient Temperature Total Power Output III
  • 4.
  • 5.
    V Data Used Dataset 1had summer data of 2012 and 2013 (6-month Data Each) Dataset 2 had year-round data July 2012 - July 2013 Hourly data of three inputs (solar radiation, ambient temperature and module temperature) and one output (power output of PV system) from 6 am to 8 pm for years 2012 and 2013
  • 6.
    VI Learning Algorithm Levenberg-Marquardt backpropagation(trainlm) Scaled conjugate gradient backpropagation (trainscg) Gradient descent backpropagation (traingd) Polak-Ribier conjugate gradient backpropagation (traincgp) Fletcher-Powell conjugate gradient backpropagation (traincgf) BFGS Newton backpropagation (trainbfg)
  • 7.
  • 8.
    VIII Model Accuracy Confidence Interval(0.5 to 2 Times Standard Deviation) 88% to 96% for models based on 6 months database 82% to 94% for models based on 12 months database
  • 9.
    IX Reference Paper Konjic, Tatjana& Jahic, Amra & Pihler, Jože. (2015). Artificial Neural Network Approach to Photovoltaic System Power Output Forecasting.
  • 10.