Assignment#3: DR - Azhar UL Haq

Assignment#3
Forecasting Techniques for Power Generation from Wind and

Solar Energy System
Subject: Selected topics of power system

Submitted to:
Dr.Azhar UL Haq
Submitted By:
Muhammad Muzammal Islam
(CMS ID: 275348)
Muhammad Javed
(CMS ID: 273422)
NUST College of Electrical and Mechanical Engineering

Rawalpindi
Abstract
Power generation from solar and wind resources has increased in recent years and main
challenges associated with integration of solar and wind energy generation with traditional power
network are Variability and Ramp events. These challenges have their effect on network
balancing, power management, power generation scheduling and commitments of power
generating units. These challenges lead towards developing different forecasting techniques like
physical variables, statistical variables, artificial intelligence and hybrid techniques.
Introduction
The primary purpose of predicting intermittent solar and wind energy generation is to find out as
efficiently as possible the output power of power plants in a close term (15-30 minutes, or hour
based and complete day time periods based). Reliable forecasting techniques reduce the cost and
power consumption as well as improving the power factor.
The recent study has shown that the reliability of integrated wind and solar irradiance system is
twice greater than either technique used separately. Characteristics of solar power generation are
very different from the wind power generation characteristics like variation in self correlation
and enough power production during the mid time of the day.
During the end duration of 2018, the global accumulated wind power production reached an
amount of 600 GW and solar power generation capacity approached to 480 GW. The persistent
forecasting technique assumes that the energy generation by solar irradiance and wind speed is
same at present time step as previous time. Two broad classifications of forecasting techniques
are physical and statistical.
Recently the researches have main focus on building the forecasting tools for short term
predictions ranging from few minutes to a few days-ahead in order to assure reliable grid
operation. Complete day forecasting act as priory data for network operators to perform different
tasks like load on/off, congestion management, scheduling, load flow prediction and reserves
power generation allocation etc. Grid operation is largely effected by the effect of cloud motion
on solar intensity and greater deviation of wind speed, that results in ramp events and variability.
Reasons for Forecasting:

Important goal of forecasting in renewable energy production is balancing of network operators
(TSOs and DSOs), economic dispatch, Load flow analysis and congestion management.
Forecasting is necessary because there is a crucial relation between wind speed and the wind
power generation, it means a little change at wind speed can alter wind power considerably.
Forecasting Techniques
Persistence method:
For short time forecasting the persistence method is used as a reference model where we assume
that the solar irradiance at time “t” will absolutely the same as past value of time “t-1”.
Persistence model is more reliable for a short forecast of 0-6 hours. If we increase forecast
horizon from 6 hours then this model’s reliability reduces.
𝐼𝑡 = 𝐼𝑡−1
Where 𝐼𝑡 is solar irradiance at present time t.
𝐼𝑡−1 is solar irradiance at past time t-1.
Fig 1. Persistence Method Forecast
Physical Techniques:
 Numerical Weather Prediction (NWP)

 Satellite based (Cloud Imagery)
 Numerical Weather Prediction (NWP):
In NWPs techniques we develop some physical variables and then form relationship between
them. The physical variables are wind pressure, wind speed, temperature, humidity, topography,
surface roughness etc. NWPs models consists of further sub-models which represent grid points.
The output of sub-models shows the wind speed, shape and features of site. As far as solar
irradiance forecasting is concerned there are fourteen (14) global NWPs models for this.
Fig2. Schematic diagram of Numerical Weather Prediction

https://www.researchgate.net/figure/General-structure-of-a-numerical-weather-prediction-NWP-
b-conventional-chemical_fig1_280236082
 Satellite based (Cloud Imagery):

The solar irradiance is highly effected by cloud optical depth and cloud cover. Therefore cloud
assessment plays an important role in forecasting. The application of satellite and sky images is
to evaluate the cloud structure in past time step and then showing the clouds motion and
magnitude in future time step.
Fig 3. Schematic diagram of forecasting based on sky images
Linear Statistical Approaches:

Linear statistic based models are members of class of time data series forecasting techniques.
These statistical models are mainly based on historical time series data and this time series data
is used to establish a relationship between different variables to predict their future values. Some
particular statistical methods are Autoregressive (AR) based, Moving average based and
Autoregressive-moving average ARMA.
We used AR methods for the prediction of wind speed. The AR method used along with second
order blind identification (SOBI). Because the results which obtained from AR-SOBI model has
greater forecasting efficiency than that of persistence and separate autoregressive method.
In case of solar irradiance the most of the statistic based methods that used for wind are also
valid for solar irradiance. But in case of solar irradiance the input does not stationary and it
varies according to cloud cover and cloud depth conditions. So, that’s why the author established
by multiplicative ARMA methods to develop the global series radiation because ARMA models
require stationary input. The other statistical technique that is used for solar irradiance is coupled
AR and dynamical system (CARDS) model. This model is to be developed for time series solar
radiation prediction for hourly and sub-hourly time scales. Hence solar data based on seasonal
predictions so it is de-seasoned by using power spectrum analysis and Fourier series.
Fig4. Schematic diagram of ARMA identification Model
www.researchgate.net/figure/Block-diagram-of-the-auto-regressive-moving-average-ARMA-
identification-model_fig12_226537209/download
Artificial Neural Networks (ANN) Approaches:
The main inspiration behind ANNs is the cognitive approach adapted by human brain to solve
problems. The design of ANNs is such that they have self-learning ability from experience and
drawing a relationship between input and output variables. ANNs are classifies into two
categories, FFNN and RNN. In feed-forward NNs the computation flow is only from input to
output (forward direction) but in Recurrent NNs there are feedback loops too which helps in
modeling of several dynamic systems.
FFNN was implemented with wind speed and wind direction asinput and wind power as an
output and a very good performance was demonstrated. It was also demonstrated that the effect
of wind direction on wind power is less than wind speed. A comparison of SARIMA and
ADALINE NNs for forecasting wind speed in Mexico depicted that SARIMA model is better
than ADALINE. In comparison between RBFNN (Radial basis function) and ERNN (Elman
recurrent) predicted that RBFNN gives less error than ERNN. A comprehensive analysis of
RBFNN, FFBP and ADALINE revealed that under different model architecture and parameters
different accuracies were obtained for same dataset. To improve the learning process of NNs
various techniques are used including GNN, PSO and multi-agent BPNN. A comparison between
SVM and MLP demonstrated that SVMs provided better prediction as compared to MLP.
For forecasting solar irradiance, based on MLP structure an ANN model was developed
considering day-ahead predictions. In this model the solar irradiance and air temperature were
taken as an input and day-ahead solar forecast as an output. As a result of this model, the
correlation
coefficient of performance prediction for clear sky days was 98% and for poor weather days it
was 94%. Feed forward wavelet networks (combined effect of wavelet theory and NNs) were
also used to solar irradiance forecasting. Adaptive wavelet-networks were used for solar
irradiance forecasting in Algeria and they gave considerably good performance as compared to
other NNs.
In short, the main purpose of ANNs and SVMs is to learn and train the historic data trends to
provide future forecast without any prior mathematical model.
Fig5. FFNN
Fig6. RNN
https://en.wikipedia.org/wiki/Artificial_neural_network
Fuzzy Logic Models:

Fuzzy logic models based on generalization of classical logic models. These models usually
based on non-linear designing of input to output variables and they consists of continuous range
[0 1] of functions. The main application of fuzzy logic models is that we used them in such a
place where exact models of systems are not known or there is an ambiguity present in
formulation of systems.
Fuzzy logic models based approaches are used for wind speed forecasting that consists of
probabilistic data system. This probabilistic data system is used for modeling of random and
deterministic sets in order to introduce the third probability dimension in model. We can predict
the wind power from raw data that is based on fuzzy modeling. The algorithm used for
calculation of the number of models based on fuzzy C-means and in order to develop the
relationship between functions the BP-Algorithm is used. The efficiency of these fuzzy models
can increase by wavelet transform (WT). Then Author used the combined technique of
ARTMAP and wavelet transform (WT) for wind power prediction that is based on wind speed
and wind direction information.
Fuzzy-based approaches are also used for Solar irradiance forecasting .The author basically used
the temperature and solar intensity forecast to design the fuzzy data models of solar irradiance
system. In order to calculate the ambiguity in solar irradiance the interval type-2 fuzzy data
based models are used. The main theme of using this interval type-2 fuzzy model is that it not
only tells us about uncertainty range of solar irradiance but it also tells us about forecasting value
that is useful in order to develop predictions intervals (PIs). In short fuzzy based approaches are
helps us to map the non-linear data of wind and solar intensity without requiring historical data
series.
Fig7. Schematic diagram of Fuzzy Logic model

Assignment#3: DR - Azhar UL Haq

Uploaded by

Copyright:

Available Formats

Assignment#3: DR - Azhar UL Haq

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Assignment#3: DR - Azhar UL Haq

Uploaded by

Copyright:

Available Formats

Assignment#3

Forecasting Techniques for Power Generation from Wind and

Subject: Selected topics of power system

NUST College of Electrical and Mechanical Engineering

Reasons for Forecasting:

Fig 1. Persistence Method Forecast

 Numerical Weather Prediction (NWP)

Fig2. Schematic diagram of Numerical Weather Prediction

 Satellite based (Cloud Imagery):

Linear Statistical Approaches:

Fig4. Schematic diagram of ARMA identification Model

Fuzzy Logic Models:

Fig7. Schematic diagram of Fuzzy Logic model

You might also like