We provide two datasets gengine1 and gengine2. Each dataset consists of N measurements of the gasoline engine D = {d₁,...,d_N}. Each measurement d = {(u₁, y₁),...,(u_T, y_T)} contains a trajectory of length T where the inputs are denoted as u_t and the outputs as y_t. The length of the trajectory may vary between measurements. The data points are recorded every 10 Hz. Each measurement is saved as a separate csv-file. The data is standardized to mean-zero and unit-variance on the training data set.

The dataset gengine1 contains in total 33 measurements. We used the measurements [10,11,12,13,14,15,16,17,18,19,30,31,32,33,34,35,36,37,38,39] for training and the measurements [53,54,55,56,57,58,59,60,61,62,63,64,65] for testing.

The first 5 columns correspond to the inputs u_t:

• Speed
• Load
• Lambda
• Ignition Angle
• Fuel Cutoff

The last 8 columns correspond to the outputs y_t:

• Particle Numbers
• CO
• CO2
• HC
• Nox
• O2
• Temperatue (Exhaus Manifold)
• Temperate (Catalyst 1)

The dataset gengine2 contains in total 22 measurements. We used the measurements [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15] for training and the measurements [16,17,18,19,20,21] for testing.

The first 4 columns correspond to the inputs u_t:

• Speed
• Load
• Lambda
• Ignition Angle

The last 5 columns correspond to the outputs y_t:

• Particle Numbers
• HC
• Nox
• Temperatue (Exhaus Manifold)
• Temperature (Catalyst 1)