| 摘要 | ABSTRACT
1. Research Purpose
The international crude oil price was dramatically dropped from $100.26/bbl. at June 2014 to $44.74/bbl. at January 2015. The high volatility of the oil price like this can cause a significant impact on power generation mix since natural gas and steam coal prices can be influenced by the oil price. It is, however, hard to find a quantitative analysis for the impact of oil price on the power mix. In the light of this, we setup an artificial neural network model to forecast long-term oil price by scenario. Also, based on fuel cost forecast data by power generation source, we analyzed the effect on future electricity market.
2. Research Results
A. Establishing NARX Neural Network
To forecast the oil, natural gas, and coal price from 2016 to 2029, NARX(Nonlinear AutoRegressive network with eXogenous inputs) neural netwrok model was established. Scenarios were also setup by the review results of World Energy Outlook(IEA, 2000~2015), Annual Energy Outlook(EIA, 2016), BP Energy Outlook(BP, 2016), and Commodity Markets Outlook(World Bank, 2016). As results, we selected ��World GDP Growth Rate���� ��GDP Growth Rate of China and India��, ��Reserve/Production Ratio��, and ��Major Events on the Oil Market�� as an exogenous variables for the NARX model. High, base, and low demand scenarios were also considered.
B. Estimation Results
We found that the performance of the R/P ratio case was the best and the case of all three exogenous variables included was also good. Considering the estimation results and the economic implication, we chose ��World GDP+R/P Ratio+Events�� scenario as a central scenario of our model.
In addition, the change of power generation cost according to the change of fuel cost by power generation source was examined. It seems unlikely that the order of economic dispatch between bituminous coal and LNG will vary between the two power sources, but it is looking at how the price of the two fuels should change in the future in order for LNG to replace bituminous coal.
3. Conclusion
This is a pioneer study for a long-term forecasting of energy price using machine learning. Using the results of this analysis, we predicted changes in power generation cost by power source and examined the effect on electricity market in the future. |
