Wind Power Forecasting - An Application of Machine learning in

Renewable EnergyDr. Gul Muhammad Khan

Abstract

The advancement in renewable energy sector being the focus of research these days, a novel neuro evolutionary technique is proposed for modeling wind power forecasters.

The word uses the robust technique of

Cartesian Genetic Programming to evolve ANNfor development of forecasting models.

These Models predicts power generation of a wind based power plant from a single hour up to a year - taking a big lead over other proposed models by reducing its MAPE to minimum values for a single day hourly prediction.

Results when compared with other models in the literature demonstrated that the proposed models are among the best estimators of wind based power generation plants proposed to date.

Introduction

The environmental hazards like pollution and global warming have compelled the power generation sectors to come up with

different green energy production techniques.

EU Wind power association Report

• 9.616GW wind based power plants installation in EUROPE in 2011

• 105.6GW capacity achieved by the end of 2012

• 7% of the total generated power in 2012 was based on wind energy

So

Considering these facts, one must adopt techniques that are feasible and provide optimum forecasting results regarding wind power generation.

Estimation of future production of wind power is one of them.

Techniques used for forecasting wind power generation includes Stochastic, Probabilistic and Machine learning techniques.

CGP (Cartesian Genetic Programming)

CGP utilizes a two dimensional programming architecture that is incorporated

by nodes or genes

CGP (cont.)

In CGP, a genotype is represented by a string of integers with thecorresponding phenotype a two dimensional nodal network.

The genotype is evolved by changing the connectivity andfunctions of nodes in the network, thus obtaining a range oftopologies.

CGP evolved ANN

Input to a particular node of a genotype g

Output of a particular node p

Wind Power Prediction (Case Study)

Sotavento Galicia wind farms

Online Available at:http://www.sotaventogalicia.com/en/real-time-data/instantaneous-wind-farm

Simulation Attributes For wind power prediction

• Number of inputs taken (24hrs data)

• Training Data Set (1 year)

• Accuracy and Error Calculation

• Fitness

Sliding Window Mechanism

Training Accuracy

Network initially with 200 instances of hourly wind power production as its input

exhibits 4.9% Mean Absolute %age Error

Testing Results

Network initialized with 150 instances of hourly wind power production as its

input has 95.29% Mean absolute Percentage Accuracy that descends to 5.79%

of MAPE for increasing the input range that is in the favor of model

Comparison with other Models for the sameInput data type and prediction horizon

CGP based ANN has its lead over other state of the are models for its

1.04% MAPE value, depicting its perfection and adoptability

Actual Verses Forecasted terrain

Graphical comparison of 30 days actual verses forecasted generated wind power

Conclusion• Based on CGP evolved ANN, three different forecasting models have been

proposed in the work whereas Each model is forecasting generating windpower for the next 1 hour.

• A MAPE value of as low as 4.71% has been achieved for a full one year.

• The Accuracy of the model goes up to 98.951% for single day prediction,evidencing the perfection of the proposed CGPANN Models for short termforecasting.

• The proposed solution can be further enhanced by consideringparameters such as wind-speed and its direction at the site, instantaneoushumidity, atmospheric pressure and air temperature.

Questions??