Forecasting India's Gasoline Consumption to Predict Future Ethanol Demand for Meeting Ethanol Blending Targets: A Comparison of Arima, GM (1, 1) and Lstm Models

Bishal Dey,Bidesh Roy,Subir Datta
DOI: https://doi.org/10.2139/ssrn.4363456
2023-01-01
SSRN Electronic Journal
Abstract:Ethanol is recognised as a fuel in India due to its potential to help the nation's ecology and energy security. In order to reap the rewards of ethanol, the Indian Government has started an Ethanol Blending Programme (EBP). Currently, India intends to begin using gasoline with 20% ethanol by 2025 and to complete the switch by 2030. India has, however, continually fallen short of the blending goal, which is because of the lack of infrastructure in the ethanol value chain. Moreover, this goal is still unattainable due to a lack of ethanol supply. So, improvement of infrastructure of the ethanol value chain requires a precise fuel demand projection. Besides, as the demand varies from year to year, so identifying a forecasting model is necessary which can produce accurate demand projection consistently with the addition of new data. This study attempts to identify the most suitable model among Autoregressive Integrated Moving Average (ARIMA), Grey model and Long-short term memory (LSTM), for this purpose based on the historical time series of two different datasets consist of different number of data points. The study reveals that LSTM model produces forecast more accurately with more consistency. Thus, this model is used for projecting the ethanol demand till 2030. The demand for ethanol to achieve a 20% mix will be 10474.81 million litres through 2030, according to an LSTM model estimate. Therefore, India is likely to face continuous ethanol shortages in order to fulfil future blending requirements, if the demand supply imbalance is not mitigated.
What problem does this paper attempt to address?