Time-series data in SSP

[robha67]

The possibility to store simulation stimuli (time-series data) in the SSP resources such that any SSP supporting integration tool can interpret the resource and invoke it for executing the simulation. In this context, the time-series data could represent some founding design requirements.

[JochenKoehler]

Web meeting 2022-09-16

• Robert will check if a first definition of requirement could be written as start for a discussion
• Dag references to a paper with similar topic - will put link here: Modelica conference Munich 2012 (Leo Gall, Martin Otter)

[DagBruck]

The paper is "Proposal for a Standard Time Series File Format in HDF5" at https://ep.liu.se/ecp/076/050/ecp12076050.pdf.

[chrbertsch]

Do you know the "Signal Tables" by @MartinOtter :
https://www.mdpi.com/2079-9292/11/18/2811

[DagBruck]

Do you know the "Signal Tables" by @MartinOtter :
https://www.mdpi.com/2079-9292/11/18/2811

Yes, that is a very useful paper. However, I did not see any explicit comparison of the performance and size requirements of the serialized SignalTable. As mentioned in the paper, the "trick" DSRES applies to signals of the type a=b or a=-b reduces the file size by up to 75%, which is significant.

[MartinOtter]

As described in the paper and used in the SignalTables.jl prototype, "a=b" is utilized, but not "a=-b" (similar to FMI 3.0). The main reason was to simplify the description and the implementation. I did not yet made serious comparisons between SignalTables.jl and DSRES format, because it is some effort to have exactly the same (realistic) model in Modelica and in Julia/Modia/SignalTables.jl and store exactly the same variables. It would be possible to utilize "a=-b" as well by some additions to the current description and implementation.

[trulede]

Could you do this using a Model (FMU) to represent the time series in the simulation?

Define your time series data as a Parameter (binary with MimeType according to the format), and then connect that to a Component (Model/FMU) which supports that MimeType format, for the purpose of introducing that data to the simulation.

As I read over the Binary Parameter specification, it does not seem possible to reference a resource file directly. It might be necessary to extend that to include a source attribute, similar to Component source (anyURI). However, the annotation mechanism is available on a Parameter and would also suffice.

Edit: With binary types increasing representing complex data sources (signals, buses etc.) it might be interesting to represent them as Components. With that, all the existing mechanisms of the SSP become available for creating the structural elements around that data source. There might be some overlap with the new FMI3 clocks, where the source of those clocks might be component of the Simulation environment, or some other (external) system, rather than an FMU.

[DagBruck]

Another possibility regarding storage file formats is recon, as described in https://ep.liu.se/ecp/096/113/ecp14096113.pdf.

[pmai]

F2F 2024-06-13: To be handled via the common CSV format and other approaches to be defined in FMI-LS-Ref and harmonized across the MA.