The identification of sites for establishing power plants operated by renewable sources of energy is a pressing need in the 21st century as nonrenewable sources are no longer fit for our energy needs. This is especially important in a country like India, where supply to demand ratio is very low.
In this paper, a machine learning technique named Decision Tree (CART) is proposed as a novel method to predict whether the site is liable of establishing solar power plant.
The decision tree method is able to do feature selection implicitly and performs very well even when the dataset is huge.
Experiments were performed with the dataset gathered from the model proposed by Dev Gaurav et al. [2].
No other energy source matches to the energy potential of sunshine. Coal,
Uranium, Petroleum, and Natural Gas are TOTAL recoverable reserves, whereas the solar energy has a giant potential per year.
Non renewable energy, such as coal and petroleum, require costly explorations and potentially dangerous mining and drilling, and they will become more expensive as supplies shrink and demand surges.
Renewable energy yields only minute levels of carbon emissions and thus helps combat climate change triggered by fossil fuel usage.
Zamo, M., Mestre, O., Arbogast, P., and Pannekoucke, O.: A benchmark of statistical regression methods for short-term forecasting of photovoltaic electricity production. Part II: Probabilistic forecast of daily production. In
Solar Energy, vol. 105, pp. 792-803. (2014)
Gaurav, D., Mittal, D., Vaidya, B. and Mathew, J.: A GSM based low cost weather monitoring system for solar and wind energy generation. In
Applications of Digital Information and Web Technologies (ICADIWT),
IEEE, 2014 Fifth International Conference, pp. 1-7. (2014)
Jung, J., and Broadwater, R. P.: Current status and future advances for wind speed and power forecasting. In Renewable and Sustainable Energy
Reviews, 31, pp. 762-777. (2014)
Pedro, H. T., and Coimbra, C. F.: Assessment of forecasting techniques for solar power production with no exogenous inputs. in Solar Energy, vol.
86(7), pp. 2017-2028. (2012)
Diagne, M., David, M., Lauret, P., Boland, J., and Schmutz, N.: Review of solar irradiance forecasting methods and a proposition for small-scale insular grids. in Renewable and Sustainable Energy Reviews, vol. 27, pp. 65-76.
(2013)
Sharma, N., Sharma, P., Irwin, D., and Shenoy, P.: Predicting solar generation from weather forecasts using machine learning. In Smart Grid
Communications (SmartGridComm), 2011 IEEE International Conference on IEEE, pp. 528-533, October. (2011)
Decision tree approach is a technique under supervised learning, the learning proceeds through a set of decision rules by using the attributes available in the dataset. A tree is constructed having the information related to decision rules.
For each data located at particular node, the algorithm splits it into feature and threshold value.
The function that minimizes the impurity function H() is given by
G ( Q ,
)
nleft
Nm
H ( Qleft (
))
nright
H ( Qright (
))
Nm
The function that classify samples according to class labels is
P mk
1
Nm
I ( y i
k )
This technique is simple and efficient, does not require data normalization and the cost is logarithmic in the number of samples used for training. It is able to do feature selection implicitly.
Research data utilized for classification was gathered from the model proposed by Dev Gaurav et al. [2].
The total number of instances for this experimentation was 8000 recorded rows over a period of eight months, from September 2012 to April 2013.
The features taken into consideration are comprised of daily weather conditions including
Light intensity
Temperature
Relative humidity
The class label for each instance was either “1” signifying that the conditions are favorable in context of setting up power plant or “0” signifying that the weather conditions are not favorable to set up power plant.
A comparison study for analyzing the model which we have proposed, with the Nearest centroid model and stochastic gradient descent (SGD) was incorporated. Decision Tree does not entail any parameter values in opposition to stochastic gradient descent.
The performance of proposed and comparison models was measured by computing accuracy utilizing confusion matrix criteria.
Confusion Matrix
Actual
Class
Class=1
Class=0
Predicted Class
Class=1 Class=0
F
11
F
01
F
F
10
00
The accuracy is then given by
Accuracy
( F
11
( F
11
F
10
F
00
)
F
01
F
00
)
Nearest Centroid
Stochastic Gradient descent
Decision tree
Method
Stochastic
Gradient descent
Nearest Centroid
Decision Tree
Training
Accuracy
92.605
90.301
100
Testing Accuracy
92.866
89.820
99.874
The results indicate that the proposed model outperforms Nearest Centroid and Stochastic Gradient Descent models.
This technique was a novel and peculiar approach towards the problem of determining the feasibility of a location for setting up solar power plant.
It encourages the use of renewable energy sources and ensure the utilization of non-renewable resources in a sustainable way.