You are on page 1of 11

Forecasting Selling Price of Petrol in

India Post Covid-19 Using Radial Basis


Function Technique

A AJITH KUMAR1, E LOKESH2, K KASIVISWANATH3


NATIONAL INSTITUTE OF TECHNOLOGY, TIRUCHIRAPPALLI-TAMILNADU
INDIA
1
114118001@nitt.edu, 2114118053@nitt.edu, 3114118045@nit.edu
*The Corresponding members

Keywords: RBF neural network; forecasting; Petroleum Products


ABSTRACT

Recently, the petroleum sector in INDIA has been liberalized which has a significant effect for
petroleum product distributors. India reported its first yearly increase in petroleum
consumption since February, proving the fuel-guzzling nation’s demand is back despite rising
coronavirus cases. Total petroleum demand rose 2.7% year-on-year in October, reaching 17.8
million tons, according to provisional data published by the oil ministry’s Petroleum Planning
& Analysis Cell. Road fuels reached pre-pandemic levels, but it was demand for naphtha that
really soared. Consumption of the petrochemical industry feedstock surged 15%. Diesel
demand soared 7.4% year-on-year to 6.5 million tonnes while petrol sales were up 4.5% at 2.54
million tonnes. The growth in diesel consumption is the highest in a year. Fuel demand had
snipped by 49% in April after a nationwide lockdown, imposed to curb the spread of
coronavirus, shut industries and took most vehicles off road. The 69-day nationwide lockdown
was followed by local and state-level restrictions. Restrictions have eased only slowly and in
phases, and localised restrictions in containment zones remain. The onset of the festive
season has fuelled a rise in consumption but public transport is not back to normal levels yet
as schools and educational institutions continue to remain shut in most parts of the country.
Demand for naphtha, which is used as industrial fuel for generating electricity as well as
producing petrochemicals, surged 15% to 1.3 million tonnes in October. This together with a
rise in other industrial fuels such as fuel oil indicates the return of economic activity. As all
fuel products are imported, we will be interested in the evolution by making forecasts of the
price of fuels in the Indian market. In this context, our paper aims mainly to study the selling
price of diesel and gasoline in order to provide precise forecasts to the company and to respect
the permissible error margin of 3%. To this end, we worked with a widely used approach for
price forecasting: artificial neural networks technique (radial basis function). Recently, it is
suggested to work with artificial neural networks in forecasting field as an alternative to the
traditional linear methods. We developed a radial basis function network to come up with
conclusions in terms of the superiority in forecasting performance. The error is further
minimized by applying radial basis function technique.
INTRODUCTION

India is the third largest energy consumer in the world after China and USA. It is also the fastest
growing energy consumer, consuming around 809.2 million tonnes of oil equivalent (MTOE)
in 2018. With a s share of 5.8% of the Worlds primary energy consumption, India’s primary
energy demand is expected to grow a CAGR of 4.2% during 2017-2040, much faster than any
major economy in the world. Oil and gas sector within the energy mix play a predominant role
as over one third of the energy required is met by hydrocarbons. Growing economy and
population growth are main drivers for oil and gas demand increasing every year.
There are many uncertain and unforeseen factors that affect the crude oil market, researchers
are paying more attention on the crude oil price forecasting which has widely been considered
as the most important and challenging issue. Nowadays, oil prices have gained more attention
and have been the sensitive subject for the financial Indian press. Since, the free play of supply
and demand is now governing the fuel prices. Consumers can choose any station they want and
use it freely. Thanks to this liberalization that is seen as an opportunity for the Indian economy,
all economic actors will be encouraged to streamline their attitude. However, the analysis of
the government’s approach reveals some gaps stopping the consumers benefits from this
liberalization. There are several risks that revolve around this measure that must be controlled.
In this sense, stock uncertainty threatens supply and presents a risk of rising prices.
The demands for petroleum products are demands that are derived from other primary activities
or human wants and needs. Therefore, the driving demand determinants for petroleum products
are the demands for these primary outputs, such as the transportation of goods or people, the
provision of heat for processing or cooking, etc. Hence demand forecasts have to be based on
forecasts of the levels of these petroleum using activities, subject to the possible use of
substitutes now and in the future. Given the rapid changes in the technology of energy using
systems and appliances that have been stimulated by the drastic changes in petroleum prices,
special attention must be placed on the potential feasibility and time-path of introducing such
new or improved technologies. India is heavily dependent on crude oil and LNG imports with
82.8% import dependence for crude oil and 45.3% for natural gas/LNG. The net foreign
exchange outgo is 63.305 billion US$ in the financial year 2017–18 on account of crude oil
imports. India generated 35.2 million tons of petroleum products from indigenous crude oil
production whereas the consumption of petroleum products is 204.9 million tons. Similarly
India generated 31.7 bcm natural gas locally against the consumption of 58.1 bcm. LNG price
is linked to the prevailing crude oil price in global markets. India is the second biggest importer
of crude oil and its products after China. In the year 2019, US is going to become net exporter
of LNG, LPG, crude oil and its products from its shale oil production boom. Shale oil
production cost would be the lower ceiling price for the crude oil in international trade as its
substantial production is consumed internally in US.
Our paper aims to ameliorate a forecast process for fuel prices by increasing the forecast
accuracy using radial basis function (RBF) technique To achieve this goal, we contribute in
our paper by forecasting the fuel named “HPCL petrol” prices for the year 2020 based on a
large and consistent historical data of the four previous months through RBF techniques.
(PRODUCTION AND CONSUMPTION OF PETORLEUM PRODUCTS IN INDIA)

(PRODUCTION AND CONSUMPTION INDIGENOUS SALE OF PETORLEUM PRODUCTS IN INDIA)


THEORY AND METHOD

A Radial Basis Function network is an artificial forward single hidden layer feed neural

network that uses in the field of mathematical modelling as activation functions. A Radial Basis

Function (RBF) is a function that is only defined by distances from a centre. Exact position does

not matter; only relative position matters. Primarily, RBFs are used because of one property: at

the centre, the output (influence) is highest; at each distance unit away from the centre (in any

direction) the influence decays.

The input vector is the n-dimensional input in which a classification or regression task (only one

output neuron) is being performed on. A copy of the input vector is sent to each of the following

radial basis neurons. Each RBF neuron stores a ‘central’ vector — this is simply one unique

vector from the training set. The input vector is compared to the central vector, and the

difference is plugged into an RBF function. For example, if the central and input vectors were

the same, the difference would be zero. The normal distribution at x = 0 is 1, so the output of

the neuron would be 1. Hence, the ‘central’ vector is the vector at the centre of RBF function,

since it is the input that yields the peak output.

νj=‖cj−x‖= √∑N i=1 (x −c )


i j,i 2
----- (1)

We can determine the output of the hidden layer nodes by a nonlinear function: “activation

function” of the network. A typical selection for the activation function is the Gaussian

function.
f(ν)=e^(−ν2/2σ2) ---- (2)
where the width 𝜎²σ² is a variance. Thus, the output of the hidden layer node j is

Zj=f(vj) ----(3)
Between the hidden and the output layers, we apply a set of synaptic weights wj, j = 1, 2, … , k.
The nodes in the output layer serve only as summation units, which produce the final output of
the network. Considering a one-output network, the overall output will be

y=∑j=1kwjzj ----(4)
Likewise, if the central and input vectors are different, the output of the neuron decays

exponentially towards zero. The RBF neuron, then, can be thought of as a nonlinear measure of

similarity between the input and central vectors. Because the neuron is radial — radius-based

— the difference vector’s magnitude, not direction, matters. Lastly, the learnings from the RBF

nodes are weighted and summed through a simple connection to the output layer. Output nodes

give large weight values to RBF neurons that have specific importance to a category, and smaller

weights for neurons whose outputs matter less. In order to select the locations of the centres,

we have three main strategies. The first one consists of a random selection of a set of samples

from training set. The positions of the centres are set according to these samples. This approach

can be valid and then produce rational results only if the distribution of training data is

representative. The second approach consists of performing a pre-clustering on the training set.

We use then the centres of the clusters as the centres of the RBF network. The selection of the

centres is considered sub-optimal regarding final result’s accuracy since the clustering is

performed without the knowledge of the weights of the output nodes.


Once we selected the centres, we can determine a uniform width by the following equation
𝜎=√dM----(5)
where 𝑑=max1≤𝑖,𝑗≤𝑀‖𝑐𝑖−𝑐𝑗‖ is the maximum distance between the centres we chose and M is
the number of centres.

STEP LEARNING

 Dataset preparation and centre’s choice.


 The Gaussian width’s set.
 Weight’s initialization.
 Matrix of distances’ creation.

[f(IIti-C1II)…f(IIti-CmII)] w1 = output

Wm
 Error calculation.
 Weight’s correction using the gradient descent method.
𝑊𝑖+1=𝑊𝑖−𝛼×𝐸𝑖×𝑑𝑡𝑖[ 𝑓(‖𝑡𝑖−𝑐1‖)⋯𝑓(‖𝑡𝑖−𝑐𝑀‖)]
 Return to step (4) until we finish learning data.
STEP TEST
In this step, we do the same treatment for learning data and we calculate the output using the
following equation

[f(IIti-C1II)…f(IIti-CmII)] w1 = output

Wm
Calculate the error measure for every example.

RESULT AND DISCUSSION

In this part, we will model the data the price of “HPCL Petrol” to make prediction to determine
future selling price using radial basis function technique in forecasting.
The below figure shows the price of petrol over the years in India

The radial basis ANN model comprising two layers is trained for implementing the back-
propagation algorithm to minimize the mean squared error with one parameter, month, as the
input and the desired output as petrol selling price. Feature selection is one of the core concepts
in machine learning which hugely impacts the performance of our model. Irrelevant or partially
relevant features can negatively impact model performance. Feature selection and data cleaning
should be the first and most important step of our model designing. Subsequently, the dataset
was randomly divided into two disjoint subsets of training set (60% of total dataset) which help
us train our dataset to find the adequate model and testing set (40% of total dataset) to validate
the model found. The training set is applied in order to develop the network. After the training
phase, the reliability and accuracy of the network were perused with the test data. In order to
optimize the prediction error of the network, a compromise must be made between the
parameters of the network, which includes the speed, the goal, the number of neurons, and the
number of neurons to add in the hidden layer; to have this compromise, we have to do several
tests for the different cases. In order to optimize the prediction error of the network, a
compromise must be made between the parameters of the network, which includes the speed,
the goal, the number of neurons, and the number of neurons to add in the hidden layer; to have
this compromise, we have to do several tests for the different cases. The percentage of total
dataset used in the training step (60%). In fact, this percentage will not enable us to predict
40% of total dataset. We will need to increase the percentage of training. In the next step, we
will consider 80% of total dataset for the training stage, and 20% for testing the model.

MATLAB CODE:

The Matlab code is presented as follows:

 net=newrb(Input,Output,0.1,1.0,10,20);
 net.trainFcn='trainlm';
 net.performFcn='mse';
 net=train(net,Input,Output);
 X;
 Y=sim(net,X);
 perf=msee(net,t,Y,0.1);

ERROR COMPARISON FOR SEVERAL COMBINATIONS OF PARAMETERS


GOAL SPREAD MN DF RELATIVE
ERROR (%)
0.1 1 20 30 8.37
0.1 0.25 10 25 6.71
0.1 0.5 15 20 4.67
0.1 0.75 10 25 3.86
0.1 1 20 30 1.95

Goal: mean goal of the quadratic error.

Spread: propagation of basic radial functions.


MN: maximum number of neurons.

DF: number of neurons to add between displays significance of bold.

From the above data the combination that allows us to minimize the error in this case is as follows
 GOAL=0.1
 SPREAD=1
 MN=20
 DF=2.0
As a conclusion, learning with 80% of the database is better than the first case because we have
low error for the case of 80% compared to the case of 60%. The output is then calculated and
given below

FORECAST RESULTS OF RBF MODEL


MONTH REAL VALUE PREDICTED VALUE ERROR (&)
AUGUST Q1 (1) 80.43 80.312 1.05068624
AUGUST Q2 (2) 82.09 82.168 -1.84055658
SEPTEMBER Q1 (3) 82.14 81.475 0.68486282
SEPTEMBER Q2 (4) 81.06 81.475 -0.35897542
OCTOBER Q1 (5) 81.12 81.475 -0.21568755
OCTOBER Q2 (6) 82.06 81.682 2.65852658
NOVEMBER Q1 (7) 83.23 81.826 3.72856175

The above data shows the calculated output and presented. The graph for the corresponding
data is given below:
we can see that the RBF model can be selected and used for modelling and forecasting the
future sales in petroleum manufacturing since the error does not exceed the margin permissible
by the company. Furthermore, the error is much minimized when comparing it is compared to
other forecasting techniques and methods.
In this context, we saw the selling price of ‘HPCL Petrol’ selling price through Radial Basis
Function technique and developed RBF neural network based on the data to come up with
conclusions in forecasting performance. In this method we have seen the error is minimized
and found to be 2.0% and showed us RBF technique is much accurate for forecasting compared
to other techniques and methods.

ACKNOWLEDGEMENT
We would like to express our thanks of gratitude to our Professor “Dr R Murugesan”, who gave
us the opportunity to this project and for their valuable guidance has been the one that helped
to patch this paper and his support and instructions has served as the major contributor towards
completing our project.

We would also like to thank our parents and friends who have helped us with their suggestions
and helped in a various phase of the completion of the project.

Last but not the least we would like to thank our classmates who have helped us a lot.

DATE:23/11/2020 A Ajith Kumar - 114118001


E Lokesh - 114118053
K Kasiviswanath - 114118045
Reference

1. https://www.bloombergquint.com/business/indian-oil-demand-posts-first-annual-
growth-since-feburary
2. http://petroleum.nic.in/
3. petroleum.nic.in/sites/default/files/AR_2019-20E.pdf
4. Radial Basis Functions, RBF Kernels, & RBF Networks Explained Simply | by Andre
Ye | Analytics Vidhya | Sep, 2020 | Medium
5. Radial Basis Function Network - an overview | ScienceDirect Topics
6. https://eneken.ieej.or.jp/en/data/pdf/349.pdf
7. India petroleum demand: Covid-19 impact: India’s petroleum consumption falls to 10-year
low, Energy News, ET EnergyWorld (indiatimes.com)
8. Zhang, G, Patuwo, EB, Hu, MY. Forecasting with artificial neural networks: the state
of the art. Int J Forecasting 1998; 14: 35–62.
9. RBF Neural Networks Research Papers - Academia.edu
10. https://www.atlantis-press.com/article/25879353.pdf

You might also like