Example on generating time series data for machine learning

[ognjenkundacina]

Hi,

I am exploring the possibility of using tespy for simulating thermal power plants and generating time series datasets to apply machine learning algorithms. I would appreciate a basic example or guidance on how to simulate a system (for starters, the choice of the particular element is not important, let it be heat pump, gas turbine, or others). Specifically, I'm interested in varying input variables like load and ambient temperature over time, solving the system, and collecting the outputs for each timestep into a dataset.

Additionally, I am considering simulating a fault in one of the components in future simulations. The fault should ideally introduce a gradual performance deterioration rather than an immediate breakdown. Any suggestions on how to approach this or examples of similar setups would be extremely helpful.

Thank you for any help or pointers you can provide!

[fwitte]

Hi @ognjenkundacina,

thank you for reaching out. I have been quite busy lately, I will give you a response later!

[fwitte]

I am exploring the possibility of using tespy for simulating thermal power plants and generating time series datasets to apply machine learning algorithms. I would appreciate a basic example or guidance on how to simulate a system (for starters, the choice of the particular element is not important, let it be heat pump, gas turbine, or others). Specifically, I'm interested in varying input variables like load and ambient temperature over time, solving the system, and collecting the outputs for each timestep into a dataset.

In principle, you could apply a simulation of a heat pump following variable load from a typical district heating system, e.g. https://zenodo.org/records/2562658. It has heat demand, and feed flow and return flow temperatures. If you want to work with an air source heat pump, you could try to pull some air temperature data for the same timestamps.

There is a but: Can you elaborate a little bit more on what you are trying to achieve with the machine learning algorithms? Predict the performance or something similar? In that case it seems a little bit overkill to me, because linear interpolation on a sufficiently fine grid could do the trick as well.

Additionally, I am considering simulating a fault in one of the components in future simulations. The fault should ideally introduce a gradual performance deterioration rather than an immediate breakdown. Any suggestions on how to approach this or examples of similar setups would be extremely helpful.

In principle, you can do this. Let's say, there is some kind of fault in the compressor which leads to a decrease of efficiency over time. You could simply apply that changed efficiency value for every time step and get the results (assuming steady-state operation is still valid). The challenge here would be, how to model the effect of the fault. If you have equations for that linking specific component parameters to other things like the mass flow or pressure it will be rather easy actually. Here it would also be helpful to get a little bit more insights.

Thank you

[ognjenkundacina]

Thank you for the detailed answer!

My goal is to use machine learning to detect subtle changes in the data that could indicate future faults or signal that it's time for maintenance due to decreased efficiency. I assume most measurement values will vary significantly due to varying system load or external parameters, making these subtle faults hard to detect—a good task for ML. I would like to use tespy to generate these scenarios. Although I am not a domain expert, some examples that come to mind are:

• Gradually reducing fan speed by altering the airflow rate, affecting cooling efficiency and condenser pressure.
• Reducing boiler feed pump efficiency by increasing power consumption for the same flow rate or reducing the flow rate for a given power input.
• Lowering condenser water flow rate to simulate blockages or pump issues.
• Reducing boiler combustion air flow to simulate restricted air intake.
• Increasing oxygen concentration in the flue gas to indicate inefficient fuel burning.

However, I am unsure if these can be modeled in tespy and plan to go through the documentation in detail. If you have any suggestions on top of your mind, that would be appreciated.

Thank you,

Ognjen