Micro Gas Turbine Electrical Energy Prediction

Donated on 4/23/2024

This dataset consists of measurements of electrical power corresponding to an input control signal over time, collected from a 3-kilowatt commercial micro gas turbine.

Dataset Characteristics

Sequential, Multivariate, Time-Series

Subject Area

Engineering

Associated Tasks

Regression

Feature Type

Real

# Instances

71225

# Features

1

Dataset Information

For what purpose was the dataset created?

Its original purpose was to learn the gas turbine's input-output temporal behavior with machine learning.

Are there recommended data splits?

In the original experiments we used experiments 1, 9, 20, 21, 23 and 24 for training and experiments 4 and 22 for testing. We used RMSE as the performance metric. See our referenced paper for experimental details.

Additional Information

The dataset comprises eight time series that depict the gas turbine’s behavior under diverse conditions. Each time series represents a separate experiment where the input control voltage was varied over time, and the resulting output electrical power of the micro gas turbine was measured. The time series vary in duration from 6,495 to 11,820 data points with a resolution of approximately 1 second, corresponding to approximately 1.8 to 3.3 hours. Each level of the input control signal corresponds to a stationary, i.e., constant, level of the output power. Notably, a noticeable delay occurs in the output signal during transitions in response to changes in the input control signal. There are rectangular and continuous time series: Four rectangular time series represent scenarios in which the input control signal changes instantaneously but the output power follows with a visible delay. Accurate modeling of both transitions and stationary phases is crucial for precise gas turbine modeling in these scenarios. The four remaining continuous time series represent scenarios in which the input control signal changes gradually and there are no visible delays in the output power. For these time series, learning transitions is less important for modeling the overall behavior.

Has Missing Values?

No

Introductory Paper

Knowledge-Guided Learning of Temporal Dynamics and its Application to Gas Turbines

By Pawel Bielski, Aleksandr Eismont, Jakob Bach, Florian Leiser, Dustin Kottonau, and Klemens Böhm. 2024

Published in 15th ACM International Conference on Future Energy Systems (e-Energy '24), Singapore

Variables Table

Variable NameRoleTypeDescriptionUnitsMissing Values
timeIDContinuousTimesecondsno
input_voltageFeatureContinuousInput control voltagevoltsno
el_powerTargetContinuousElectrical output powerwattsno

0 to 3 of 3

Dataset Files

FileSize
train.zip623.2 KB
test.zip264.3 KB

Reviews

There are no reviews for this dataset yet.

Login to Write a Review
Download (887.5 KB)
1 citations
5648 views

Creators

Pawel Bielski

pp.bielski@gmail.com

Karlsruhe Institute of Technology

Dustin Kottonau

Karlsruhe Institute of Technology

License

By using the UCI Machine Learning Repository, you acknowledge and accept the cookies and privacy practices used by the UCI Machine Learning Repository.

Read Policy