Previous (Schematics (Advanced level))   |   Start   |   Next (How results are stored)

What happens when the simulation runs?

The IDA solver

Speeding up computation

IDA ICE builds a single large simultaneous system of equations for all processes in the building. This system of equations contains several ordinary differential equations and has therefore different “built in” time constants. The room air will for example react quickly on a convective heat load, while the ground layer below the building can have a time constant several orders of magnitude larger. This system of equations is solved with numerical methods that adapt the timestep to the frequency content of the solution. Short time constants in the model in combination with high frequency content in driving functions (many starts and stops) can lead to long execution times.

Internally generated starts and stops (events) will also lead to short timesteps. Try for example to use a thermostat for radiator control.

The shear number of equations in the system is naturally significant for the execution time, which increases roughly linearly with the problem size. Therefore, it is vital not to model any unnecessary detail, and not for example model a large number of identical objects such as windows or cooling beams, when they could be replaced by a single larger one. Each added zone will contribute about two thousand variables to the system of equations.

The single most important factor for speeding up calculations is to have a reasonable amount of detail in time schedules (driving functions). Describing many sharp transitions will usually only result in increased computation time and not have a significant impact on results. From version 4, automatic smoothing is applied to key schedules to minimize these problems. (Can be turned off under System parameters).

For a model with reasonably simple time schedules, it can often prove effective to loosen the numerical tolerance to make the solver take longer and thereby fewer timesteps. Two important solver parameters are Tolerance and Maximum timestep. These are accessed from the Advanced tab of the Simulation data dialog. Often the tolerance can be relaxed to, say, 0.1-0.3 from the standard value 0.02. For problems with equations that are difficult to solve, it can sometimes be beneficial to instead decrease the tolerance. (The solver spends less time on failed attempts to take long steps.)

Looser tolerance will normally lead to acceptable loss of accuracy for accumulated quantities such as monthly energy consumption. For computing design (extreme values) of quantities such as heating or cooling load, one should be more careful with using loose tolerances and a large max timestep. Too loose a tolerance will on the other hand lead to decreased robustness, forcing the solver to often have to back up and retry with a smaller timestep, ultimately leading again to even longer execution time (if indeed the simulation is successful).

To learn about the statistics of a simulation in terms of number of timesteps, variables, restarts etc. view the file screen.txt in the IDA temporary directory idamod48. At the end of this file some statistics are given.

Time-split Parallelization

The clock time of the simulation can be substantially decreased on a computer with more than one core, able to run concurrent processes by splitting the whole simulation period into smaller periods running on different cores. This is implemented in IDA ICE to been done automatically if the box “Time-split parallelization” in the Advanced Simulation Data tab is ticked. The number of processes is set under the Simulation Preferences menu.

There is no gain in simulation time if the number of processes is higher than the total available logical processors and this parallelization technique has some overhead which makes it beneficial only for long simulation periods. The Speed-up is roughly 2-3 using 4 processes and roughly 4 using 12 processes.

Numerical instabilities

With a tool like IDA ICE it is easy to build large non-linear systems of equations and solve them for thousands of time points. However, it is impossible to, even theoretically, guarantee the success of the solution procedure. Any non-linear system of equations may have more than a single solution or none at all. Numerical computer programs are in this respect different from most other types of software, where it may be, at least theoretically, possible to create a bug-free code. IDA ICE also has a more difficult task than most other building simulation software, where less freedom is given to create mathematically complex models.

A major part of the IDA development work is devoted to improving the solver performance on difficult cases. However, this work is altogether dependent on close interaction with users. It is vital that users, which have built reasonable and meaningful models that are difficult to solve, take the trouble of sending the model to the support office. This is most easily done by using the Mail support function on the Help menu.

Some physical processes more often lead to problematic models. Generally speaking, models with significant natural ventilation flows through openings or leaks are the most difficult to solve, especially when the effect of wind pressure is included. If, in addition, vertical temperature gradients are to be simultaneously computed, one obtains a severely stiff and non-linear system of equations.

Avoid to use vertical or horizontal openings unless the bidirectional flow is essential for the study (use large leaks instead). Absolutely do not use large openings for the purpose of recreating the exact geometry of a real building. Normally, precise geometry has a very small impact on results.

A common type of error is when the user has graphically defined a piecewise linear controller and unintentionally entered several points near each other, giving the curve (in micro scale) several sharp corners and jumps. The solver will invariably have problems to negotiate the sharp turns of such a graph. Check the table view of such a curve for unintentional points.

A general way of dealing with difficult cases is to decrease the tolerance parameter that was discussed in the previous section. This forces the solver to be more careful and take smaller steps, which in most cases improves robustness. A tolerance of 0.001 or even smaller can sometimes be used.

A frequent situation is that any change in input data makes a previously failed run go through. This is not as strange as it sounds, since each change will lead to a different sequence of timesteps and in this way the exact combination of values that led to the failed timestep is avoided.

Another often effective trick is to replace sharp steps in schedules with steep ramps, especially for the fan control schedule. This will enable the solver to gradually over a short time period approach the new solution and thereby reach it more securely. If, in addition, the start and endpoints of the ramp are marked as input events (double entries of the same point in the table view), the solver will be even more cautious. Input events are marked by repeating the same time point twice in the profile, something best done in the Data tab view.

Solver diagnostics

From version 4.7, a new useful diagnostic tool has been added. When the option ‘Show slowest model’ (Options menu, Preferences, Developer) has been selected, the component with the slowest convergence in each timestep will blink red during simulation, when viewed on the Schematic tab. For a case that runs well, the role of being the slowest model will change among several components during the simulation. On the other hand, for models with some particularly problematic equations, the components that are involved in this difficulty will blink more than others and this will frequently lead the user to understand the root of the problem.

Another useful diagnostic tool is the “real-time” update of the variables in the components at the advanced level. By opening an outline tab and choosing the variables to be visible in the schematic view of a component will show the actually computed values of the variables in each timestep during the simulation run as indicated in the following figure.

In ICE 5.0, we've introduced a feature for real-time visualization feedback, designed to enhance your simulation experience by providing instant visual insights into your work. Here's what you can do with this feature:

1. Monitor Multiple Variables: Use the Monitor dialog to track up to three variables, making it easier to track and compare data during your simulations (already been there).
2. Animate 3D Variables: Access the 3D tab to animate variables like temperatures, heat flows, shading control signals, PMV, PPD, air quality indicators, and ventilation air flows. You can do this in real-time while your simulation runs.
3. Real-Time Variable Updates: Stay updated with the real-time changes in your simulation through the outline tab. Right-click on a variable and select "monitor" to open a new monitor dialog.
4. Colored Temperature Representation: Visualize temperature variations in both the plant and air flow network with a color-coded schematic view

Previous (Schematics (Advanced level))   |   Start   |   Next (How results are stored)