How do you handle 24h Simulink simulations with Simscape/Specialized Power Systems?

During my PhD work, I started wondering if there’s an efficient way to run long-duration simulations in Simulink, especially with a stop time of 86400 s (a full day).

I’m currently working on a PV system and studying the behavior of the photovoltaic array under varying irradiance conditions for a forecasting project. The issue is that using both Simscape and Specialized Power Systems libraries makes the simulation extremely heavy and slow.

I’ve read about several optimization techniques to reduce simulation time, such as:

• parallel/simultaneous simulation,
• saving and restoring states,
• using different solver modes,
• reducing switching frequencies,
• averaging techniques, etc.

While these methods definitely help, they still don’t reduce a full-day simulation to something practical like a few hours.

One solution I found was replacing some components with mathematical equations or simplified models. This speeds things up a lot, but obviously comes at the cost of accuracy. Another option is converting certain components into MATLAB functions instead of detailed Simulink blocks.

Still, I feel like this is a common problem, especially for people working on EMS, forecasting, or long-term energy simulations. I was hoping others here may have faced the same issue and found better approaches or best practices.

Any advice or recommendations would be greatly appreciated.

Answer 1

[Walter Roberson]

You use a discrete step time of 5e-6. You have a final time of 24 hours, and want to run it in "a few hours", which for the sake of discussion we will interpret as "6 hours".

In order to run 24 hours in 6, it follows that we need to be able to process 5e-6 seconds worth of computation in (6/24) as much time -- so in 1.25e-6 seconds.

You cannot use parallel simulations because you are iterating through time with a single configuration; parallel simulations are for the case of sweeping a parameter space.

If we assume 5 million instructions per second (projection from the 412,090 mips rating of Intel Core i9-9900K in 2018), that gives a budget of only 6 instructions per 1.25e-6 iteration.

Conclusion: it is not possible to condense 5e-6 step size for 24 hours, into 6 hours computation.