Natural gas, the electricity generation fuel of choice, would play an even more important role in the world’s energy sector in upcoming years. Volatility of natural gas markets could result in a high level of risk for both consumers and producers. A simple and precise forecast approach would enable hedging the associated risks. This thesis suggests a novel approach toward Natural-Gas price prediction modeling which provide best-in-class results. Our proposed model, has tried to be as simple as possible unlike much more unnecessary complicated models with numerous factors and parameters, while still be on par with more complex models. Complex models sometime have a lot of parameters which potentially introduce errors. These complicated models should use numerous simplifications and estimates to get to an answer. These assumptions would practically throw the model far away from the reality, because often these models must solve hundreds of equations and models simultaneously, which would exponentially increase the errors. Sometimes even getting the model to converge numerically is challenging. The suggested model is then compared to other modeling methods in performance assessment of in and out-of-sample forecasts, especially relative to the utility of the non-constant estimate of the volatility provided by ARCH, GARCH and TCM methods. The consistency and robustness of the forecast results for the selected model of each individual market (U.S., Germany and Japan) are compared to the other advanced financial corporate predictions. In this thesis, the optimum combination and configuration of various models are presented to have the best results for natural gas in each market. These models capture the volatility and leverage effect on the gas market, and their forecasting performances seemed to be better when comparing to the other candidates. Various data-sets have gone through dozens of analysis and studies with a considerable amount of different methodologies and codes. Various methods have been tested and evaluated for our purpose, like artificial neural networks (ANN), principal component analysis (PCA), temporal causal models (TCM), traditional modeling approaches, outliers’ detection methodologies, spectral analysis, Autocorrelations and Partial autocorrelations. Not all these methods and analysis made it into the final reports, for various respective reasons. Various data-set from diverse sources in distinctive time horizons have gone through different analysis and preparation methods before being used in the actual methodology, to ensure the input data, have sufficient accuracy, to avoid introduction of any drastic errors. Different software package and codes have been tested and used, including, statistical analysis packages, data analysis, analytics, big data analysis and data miners alongside more conventional software. As for the application of the method, a relatively efficient combined-cycle natural gas-fired power plant is simulated through the computer code, and the gas consumption of it is calculated. Then the economic evaluations have been done on the results in an in-house platform for entire lifespan of the power plant, considering dynamic natural gas price (based on our best model) and dynamic electricity price model. At the end, the greater goal of all these assessments and predictions, is to take the best set of decision to maximize our main objective. This objective, principally, in chemical or any industrial plant is to maximize the monetary revenue, bearing in mind the available limitations and resource. The concluding goal is to offer the best possible scenarios for decision makers, and assist them in taking optimum decision.
-
A novel approach towards natural gas price prediction, modeling and its application
MIRZAEI, FARSHAD;MIRZAEI, FARZAD
2016/2017
Abstract
Natural gas, the electricity generation fuel of choice, would play an even more important role in the world’s energy sector in upcoming years. Volatility of natural gas markets could result in a high level of risk for both consumers and producers. A simple and precise forecast approach would enable hedging the associated risks. This thesis suggests a novel approach toward Natural-Gas price prediction modeling which provide best-in-class results. Our proposed model, has tried to be as simple as possible unlike much more unnecessary complicated models with numerous factors and parameters, while still be on par with more complex models. Complex models sometime have a lot of parameters which potentially introduce errors. These complicated models should use numerous simplifications and estimates to get to an answer. These assumptions would practically throw the model far away from the reality, because often these models must solve hundreds of equations and models simultaneously, which would exponentially increase the errors. Sometimes even getting the model to converge numerically is challenging. The suggested model is then compared to other modeling methods in performance assessment of in and out-of-sample forecasts, especially relative to the utility of the non-constant estimate of the volatility provided by ARCH, GARCH and TCM methods. The consistency and robustness of the forecast results for the selected model of each individual market (U.S., Germany and Japan) are compared to the other advanced financial corporate predictions. In this thesis, the optimum combination and configuration of various models are presented to have the best results for natural gas in each market. These models capture the volatility and leverage effect on the gas market, and their forecasting performances seemed to be better when comparing to the other candidates. Various data-sets have gone through dozens of analysis and studies with a considerable amount of different methodologies and codes. Various methods have been tested and evaluated for our purpose, like artificial neural networks (ANN), principal component analysis (PCA), temporal causal models (TCM), traditional modeling approaches, outliers’ detection methodologies, spectral analysis, Autocorrelations and Partial autocorrelations. Not all these methods and analysis made it into the final reports, for various respective reasons. Various data-set from diverse sources in distinctive time horizons have gone through different analysis and preparation methods before being used in the actual methodology, to ensure the input data, have sufficient accuracy, to avoid introduction of any drastic errors. Different software package and codes have been tested and used, including, statistical analysis packages, data analysis, analytics, big data analysis and data miners alongside more conventional software. As for the application of the method, a relatively efficient combined-cycle natural gas-fired power plant is simulated through the computer code, and the gas consumption of it is calculated. Then the economic evaluations have been done on the results in an in-house platform for entire lifespan of the power plant, considering dynamic natural gas price (based on our best model) and dynamic electricity price model. At the end, the greater goal of all these assessments and predictions, is to take the best set of decision to maximize our main objective. This objective, principally, in chemical or any industrial plant is to maximize the monetary revenue, bearing in mind the available limitations and resource. The concluding goal is to offer the best possible scenarios for decision makers, and assist them in taking optimum decision.File | Dimensione | Formato | |
---|---|---|---|
2017_04_Mirzaei_Mirzaei.pdf
non accessibile
Descrizione: Thesis
Dimensione
4.74 MB
Formato
Adobe PDF
|
4.74 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/133930