Forecasting of preprocessed daily solar radiation time series using neural networks

Presenter : Cheng-Han Tsai 　Authors : Christophe Paoli, Cyril Voyant, Marc Muselli, Marie-Laure Nivet

SOLAR ENERGY, 2010

1

Outlines

• Motivation• Objectives• Methodology• Experiments• Conclusions• Comments

2

Motivation

• A lot of methods’ performance be affected by disruptors such as diffuse, ground-reflected and seasonal climate.

3

Objectives

• This paper has used a MLP and pre-processing for the daily prediction of global solar radiation to deal with the above problems.

4

Methodology

5

Methodology

ARIMA Bayesian Markov chains KNN

6

Methodology

ARIMA Bayesian Markov chains KNN

7

Methodology

ARIMA Bayesian Markov chains KNN

8

Methodology

ARIMA Bayesian Markov chains KNN

9

Experiments

10

Experiments

11

Cleaning the measure errors

Ad-hoc time series preprocessing

Corrected time series

Forecasting methods & Predicted irradiation

Experiments

12

Ad-hoc time series preprocessing

Clearness index Clear sky index

Experiments

13

Experiments

14

Experiments

15

Conclusions

• This prediction model has been compared to other prediction methods

• These simulation tools have been successfully validated on the DC energy prediction

16

Comments

• Advantages–This paper considers seasonal factors

• Applications– Solar radiation prediction

17