arXiv subjects

Alan Rezazadeh

Publications and source records attributed to Alan Rezazadeh.

4 recordsLinked to original sources

arXiv2022

Application of Neural Network in the Prediction of NOx Emissions from Degrading Gas Turbine

This paper is aiming to apply neural network algorithm for predicting the process response (NOx emissions) from degrading natural gas turbines. Nine different process variables, or predictors, are considered in the predictive modelling. It is found out that the model trained by neural network algorithm should use part of recent data in the training and validation sets accounting for the impact of the system degradation. R-Square values of the training and validation sets demonstrate the validity of the model. The residue plot, without any clear pattern, shows the model is appropriate. The ranking of the importance of the process variables are demonstrated and the prediction profile confirms the significance of the process variables. The model trained by using neural network algorithm manifests the optimal settings of the process variables to reach the minimum value of NOx emissions from the degrading gas turbine system.

cs.LG↗

arXiv2021

Wind Turbine Gearbox Condition Based Monitoring

The main objective of this paper is finding effective gearbox condition monitoring methods by using continuously recorded monitoring SCADA (Supervisory Control and Data Accusation) data points. Typically for wind turbine gearbox condition monitoring; temperature readings, high frequency sounds and vibrations in addition to lubricant condition monitoring have been used. However, collection of such data, require shutting down equipment for installation of costly sensors and measuring lubricant quality. Meanwhile, operational data usually collected every 10 minutes, comprised of wind speed, power generated, pitch angle and similar performance parameters can be used for monitoring health of wind turbine components such as blades, gearbox and generator. This paper uses gear rotational speed for monitoring health of gearbox teeth; since gearbox teeth deterioration can be measured by monitoring rotor to generator rotation ratios over extended period of time. As nature of wind is turbulent with rapid fluctuations, a wind turbine may operate in variety of modes within relatively short period of time. Monitoring rotational speed ratio over time, requires consistent operational conditions such as wind speed and torques within the gearbox. This paper also introduces the concept of clustering such as Normal Mixture algorithm for dividing operating datasets into consistent subgroups, which are used for long term monitoring.

eess.SP↗

arXiv2021

Environmental Pollution Prediction of NOx by Process Analysis and Predictive Modelling in Natural Gas Turbine Power Plants

The main objective of this paper is to propose K-Nearest-Neighbor (KNN) algorithm for predicting NOx emissions from natural gas electrical generation turbines. The process of producing electricity is dynamic and rapidly changing due to many factors such as weather and electrical grid requirements. Gas turbine equipment are also a dynamic part of the electricity generation since the equipment characteristics and thermodynamics behavior change as the turbines age. Regular maintenance of turbines are also another dynamic part of the electrical generation process, affecting the performance of equipment. This analysis discovered using KNN, trained on relatively small dataset produces the most accurate prediction rates. This statement can be logically explained as KNN finds the K nearest neighbor to the current input parameters and estimates a rated average of historically similar observations as prediction. This paper incorporates ambient weather conditions, electrical output as well as turbine performance factors to build a machine learning model to predict NOx emissions. The model can be used to optimize the operational processes for reduction in harmful emissions and increasing overall operational efficiency. Latent algorithms such as Principle Component Algorithms (PCA) have been used for monitoring the equipment performance behavior change which deeply influences process paraments and consequently determines NOx emissions. Typical statistical methods of machine learning performance evaluations such as multivariate analysis, clustering and residual analysis have been used throughout the paper.

eess.SP↗

arXiv2020

Toe-Heal-Air-Injection Thermal Recovery Production Prediction and Modelling Using Quadratic Poisson Polynomial Regression

This research paper explores application of multivariable regression models using only reservoir temperatures for predicting oil and gas production in a Toe-Heal-Air-Injection (THAI) enhanced oil recovery process. This paper discusses effects of statistical interaction between thermocouples by using second degree quadratic polynomials, which showed significant production forecast accuracy. Interactions among thermocouples statistically include temperature of larger reservoir areas, hence improving the predictive models. The interaction of two thermocouples can be interpreted as temperature gradient of combustion zone as moving forward during THAI operations life cycle. Second degree polynomial regression including interactions showed major prediction improvement for both oil and natural gas productions compare to simple regression models. Application of Poisson regression slightly improved prediction accuracy for oil production and was less effective on improving natural gas production predictions. Quadratic Poisson regression models showed realistic production prediction method for both oil and gas production values, due to the nature of Poisson probability distribution which is non-negative for rates and count values.

eess.SP↗