Gas Turbine Emission Prediction – MohsenMajidiPishkenari

Python can be a useful tool for analyzing by using machine learning techniques. Here’s a general outline of how I could approach it:

1. Data collection: Gather a dataset of gas turbines.

2. Data preprocessing: Clean and preprocess the dataset by handling missing values, normalizing or standardizing features, and splitting it into training and testing sets.

3. Feature selection: Identify the most relevant features. This step helps reduce complexity and improve model performance.

4. Model training: Choose a suitable machine learning algorithm, such as logistic regression, support vector machines, or random forests. Train the model using the training set and evaluate its performance.

5. Model evaluation: Assess the performance of the trained model using evaluation metrics like accuracy, precision, recall, and F1 score. Adjust hyperparameters if necessary to improve the model’s performance.

Recently, I worked with a dataset from a gas turbine located in Turkey’s northwestern region. This dataset was generated in 2015 by the Center for Machine Learning and Intelligent Systems (doi:10.3906/elk-1807-87). More specifically, it consists of 7384 instances from 11 sensors.

The purpose of the study was to analyze CO and NOx emissions, and the data range includes gas turbine parameters and ambient variables. I created three models for gas turbine emission prediction:

1. Linear regression
2. Random forest
3. XGBRegressor

How it works?

A) Data preprocessing

The histogram diagram for each variable to check normality is:

And to check the outliers, we have:

The scatter plots for CO variable are:

And the heatmap plots are:

I use these libraries for classification of a dataset of Gas turbine:
seaborn, pandas, TensorFlow, Numpy and matplotlib.

B) Modeling

To select NOx as target, we should drop CO feature and then normalize data.

For modeling step, I use LinearRegression , RandomForestRegressor and XGBRegressor methods from sklearn package.

For each model, we should calculate Mean Absolute Error (MAE).

The results show that MAE for Random forest is lower than other and the linear regression

has worst MAE.

C) Results

The results showed that the Mean Absolute Error (MAE) for the Random forest model (2.867) was better than the other models.

Abbreviation

AT: Ambient temperature
AP: Ambient pressure
AH: Ambient humidity
AFDP: Air filter difference pressure
GTEP: Gas turbine exhaust pressure
TIT: Turbine inlet temperature
TAT: Turbine after temperature
CDP: Compressor discharge pressure
TEY: Turbine energy yield
CO: Carbon monoxide
NOx: Nitrogen oxides

Posted in Python by Mohsen Majidi Pishkenari