Add example models of full 5GDHC systems

AntoineGautier commented 4 years ago

This issue is related to:

URBANopt #60
URBANopt #46
And Buildings #1613

Example models of 5th generation DHC systems are generated in Buildings.Experimental.DHC.Examples to illustrate:

the connectivity of URBANopt components,
the scalability with the number of connected buildings,
the modularity which allows switching from one type of building model to another.
- This last point was validated with RC and Spawn models (see https://github.com/AntoineGautier/modelica-buildings/tree/issue1769_examples5GDHC_withSpawn). The feature branch will be merged with only time series load models to avoid resizing the ETS and DHC systems.

The development branch is issue1769_examples5GDHC.

The examples are still under development. TO DO:

[x] For models in DHC.Networks.BaseClasses:
- [x] Check that replaceable types cannot be propagated and document the redeclare logic.
- [x] Package the computation of Q_flow from temperature measurements in a reusable block
- Buildings.Experimental.DHC.Networks.BaseClasses.DifferenceEnthalpyFlowRate
[x] Add Dialog(tab="Assumptions") to allowFlowReversalBui and allowFlowReversalDis.
[x] Validate temperature sensor with no state with PI controller cf. direct feed through due to P term and algebraic loop.
[x] Devise how different building model types can be used within a single array instance.
- As discussed on 11/30/20 this is not supported by the array structure in Modelica, GMT will handle the multiple instances needed to address this use case.
[x] Remove Buildings.Controls.OBC.UnitConversions.From_degC in favor of attribute displayUnit
Sizing building models: the examples currently use the sizing parameters of the case SeriesConstantFlowTimeSeriesB3.
- Update spawn buildings parameters + design data for buildings & DHC: The feature branch will be merged with only time series load models to avoid resizing the ETS and DHC systems.
[x] Investigate the reason why flow reversal occurs.
- DHC.Examples.Combined.Generation5.Unidirectional.EnergyTransferStations.Validation.SwitchBox
[x] Select the best model for the switching box.
[x] Validate on/off control of heat pump.
[x] Add domestic hot water production to be consistent with the RN analysis already published.
[x] Add simulation scripts and validate yearly simulation results against the RN analysis already published.
[x] Develop a partial ETS/dedicated plant model to connect with the partial building model.
[x] Document the example models.
[x] Relocate models of generic use in DHC package.
Include an example model with DHC.EnergyTransferStations.Combined.Generation5.ChillerBorefield: this will allow testing the scalability of the "advanced" ETS model, benchmarking the computational performance with the simple ETS model, and compare the energy use between a free-cooling & direct connection system, and a heat recovery chiller & indirect connection system.
- This will be addressed separately.
[x] Refactor PartialBuilding with have_heaWat instead of have_watHea and have_chiWat instead of have_watCoo: addressed through https://github.com/lbl-srg/modelica-buildings/pull/2310.
[x] Scale the loads.

AntoineGautier commented 4 years ago

@amyeallen1 @khinkelman @nllong @YanfeiNREL You can checkout this branch to see whole DHC system example models. As discussed this is still under development.

AntoineGautier commented 3 years ago

EDIT (2/16/21): TODO

[x] Assess impact on simulation time of FC not in steady-state, tau=5*60 for junctions, solver tolerance (check that 1e-6 with DASSL does not yield nonrecoverable error).
- See https://github.com/lbl-srg/modelica-buildings/issues/1769#issuecomment-780937436.
[x] Update results below: error in setting COP nominal was corrected in 27b18a261cf3dffcd6872f058d19b499dea1b7ff.

Comparison of annual electricity use (in e+12 Wh) between

Buildings.Examples.DistrictReservoirNetworks.Reservoir1Constant from branch issue1613_rnbnComparison (after de-normalizing with totEneRN3 = 2.72961e+12)
Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesConstantFlow from current development branch issue1769_examples5GDHC e370360

Branch	`Etot`	`EHeaPum`	`EPum`	CPU-time (CVode 1e-6)
`issue1613_rnbnComparison`	4.0	2.5	1.5	21
`issue1769_examples5GDHC`	3.5	2.2	1.3	182

As mentioned in https://github.com/lbl-srg/modelica-buildings/issues/1769#issuecomment-780937436 the increase in simulation time is mostly due to the building mechanical system model. An alternative approach for coupling time series will be developed in #2302.

Comparison of annual electricity use (in e+12 Wh) between

Buildings.Examples.DistrictReservoirNetworks.Reservoir3Variable from branch issue1613_rnbnComparison (after de-normalizing with totEneRN3 = 2.72961e+12)
Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow from current development branch issue1769_examples5GDHC e370360
- [x] TODO: Double check the pump controller.
- The original models in issue1613_rnbnComparison use a time constant tau=3*60 for the control volume of each fluid junction (before and after each agent connection) whereas the refactored model uses a default tau=10.

Branch	`Etot`	`EHeaPum`	`EPum`	CPU-time (CVode 1e-6)
`issue1613_rnbnComparison`	2.7	2.5	0.2	54
`issue1769_examples5GDHC`	2.5	2.3	0.3	248

[EDIT] (4/12/21): Impact of corrected borehole characteristics

See https://github.com/lbl-srg/modelica-buildings/pull/2373#discussion_r607348601

On Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesConstantFlow commit 97561a13c87fe3e862c61ee372783835bca631b4

Branch	`Etot`	`EHeaPum`	`EPum`	CPU-time (CVode 1e-6)
`issue1613_rnbnComparison`	4.0	2.5	1.5	21
`issue1769_examples5GDHC`	3.5	2.2	1.3	220

On Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow commit 97561a13c87fe3e862c61ee372783835bca631b4

Branch	`Etot`	`EHeaPum`	`EPum`	CPU-time (CVode 1e-6)
`issue1613_rnbnComparison`	2.7	2.5	0.2	54
`issue1769_examples5GDHC`	2.5	2.3	0.3	307

AntoineGautier commented 3 years ago

Parameters influencing simulation time

Summary of the main modifications with potential impact

The refactoring of the original models from issue1613_rnbnComparison includes the following modifications that may impact the simulation time.

Modeling of the in-building systems (distribution and terminal units)
Fluid junction: originally configured with a time constant of 5*60 s (note that there are around 15 junction instances, in addition to the two control volumes used in each ETS instance), changed to 10 s in the refactoring, and changed back to 5*60 s to enable the simulation of the variable flow control (otherwise too slow and stalling)
New model for the switch box: originally using pumps, refactored using three-port two-position directional valves
Temporization (hold of 15 min) used to command on/off the heat pump systems

Fan dynamics in terminal unit model

This is related to the fan component in Buildings.Experimental.DHC.Loads.Validation.BaseClasses.FanCoil2PipeCooling Buildings.Experimental.DHC.Loads.Validation.BaseClasses.FanCoil2PipeHeating

The model simulated has 3 buildings * 2 FCU / building = 6 instances.

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesConstantFlow",
    method="cvode",
    tolerance=1e-6,
    startTime=0,
    stopTime=6048000);

With energy and mass dynamics (fixed initial)

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 19.895 seconds
   CPU-time for initialization               : 0.731633 seconds
   Number of result points                   : 4317
   Number of grid points                     : 501
   Number of accepted steps                  : 91400
   Number of rejected steps                  : 5262
   Number of f-evaluations (dynamics)        : 143274
   Number of non-linear iteration            : 135901
   Number of non-linear convergence failures : 245
   Number of Jacobian-evaluations            : 3702
   Number of crossing function evaluations   : 93369
   Number of model time events               : 1759
   Number of state events                    : 159
   Number of step events                     : 0

In steady-state

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 16.8291 seconds
   CPU-time for initialization               : 0.724413 seconds
   Number of result points                   : 4317
   Number of grid points                     : 501
   Number of accepted steps                  : 71511
   Number of rejected steps                  : 3664
   Number of f-evaluations (dynamics)        : 105427
   Number of non-linear iteration            : 97862
   Number of non-linear convergence failures : 32
   Number of Jacobian-evaluations            : 2647
   Number of crossing function evaluations   : 73463
   Number of model time events               : 1759
   Number of state events                    : 159
   Number of step events                     : 0

So a gain of nearly 30% in the number of non-linear iterations, and about 15% in CPU-time for integration. However, the steady-state model yields this warning in Dymola.

Some variables are iteration variables of the initialization problem:
but they are not given any explicit start values. Zero will be used.

Iteration variables:
bui[1].bui.terUniCoo.fan.heaDis.WHyd
...

And heaDis is a protected component that prevents modifying the start attribute of WHyd. So currently the fan is modeled with a state.

Simulating the model with DASSL shows that most of the states that dominate the error correspond to the terminal unit model. The feature enhancement in #2302 should address this.

Time constant of control volume in junctions

This is related to the control volume in the fluid junction components, especially before and after each connection to an agent (ETS, plant or storage). The effect is highly tangible with closed loop control of the distribution system (as opposed to constant flow rate).

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow",
    method="cvode",
    tolerance=1e-6,
    startTime=0,
    stopTime=6048000);

tau=5*60 (setting from original models in issue1613_rnbnComparison)

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 26.3316 seconds
   CPU-time for initialization               : 0.807683 seconds
   Number of result points                   : 4319
   Number of grid points                     : 501
   Number of accepted steps                  : 118454
   Number of rejected steps                  : 6985
   Number of f-evaluations (dynamics)        : 186083
   Number of non-linear iteration            : 178706
   Number of non-linear convergence failures : 299
   Number of Jacobian-evaluations            : 4409
   Number of crossing function evaluations   : 120433
   Number of model time events               : 1759
   Number of state events                    : 160
   Number of step events                     : 0

tau=10 (default setting for the base classes in Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Networks)

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 47.0535 seconds
   CPU-time for initialization               : 0.75922 seconds
   Number of result points                   : 4319
   Number of grid points                     : 501
   Number of accepted steps                  : 252732
   Number of rejected steps                  : 18448
   Number of f-evaluations (dynamics)        : 368372
   Number of non-linear iteration            : 360288
   Number of non-linear convergence failures : 291
   Number of Jacobian-evaluations            : 7193
   Number of crossing function evaluations   : 254736
   Number of model time events               : 1759
   Number of state events                    : 160
   Number of step events                     : 0

CPU-time for integration is nearly doubled. The whole-year simulation will eventually stall. However, 5' transport delay at each junction seems also unrealistic. A plug flow pipe model will probably be better suited to include temperature dynamics.

Solver tolerance

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow",
    method="cvode",
    tolerance=1e-6,    // Default setting.
    startTime=0,
    stopTime=6048000);

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 26.3316 seconds
   CPU-time for initialization               : 0.807683 seconds
   Number of result points                   : 4319
   Number of grid points                     : 501
   Number of accepted steps                  : 118454
   Number of rejected steps                  : 6985
   Number of f-evaluations (dynamics)        : 186083
   Number of non-linear iteration            : 178706
   Number of non-linear convergence failures : 299
   Number of Jacobian-evaluations            : 4409
   Number of crossing function evaluations   : 120433
   Number of model time events               : 1759
   Number of state events                    : 160
   Number of step events                     : 0

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow",
    method="cvode",
    tolerance=1e-4,     // Change in tolerance.
    startTime=0,
    stopTime=6048000);

Integration terminated successfully at T = 6.048e+06
   CPU-time for integration                  : 20.6832 seconds
   CPU-time for initialization               : 0.771738 seconds
   Number of result points                   : 4339
   Number of grid points                     : 501
   Number of accepted steps                  : 79861
   Number of rejected steps                  : 1004
   Number of f-evaluations (dynamics)        : 119938
   Number of non-linear iteration            : 112354
   Number of non-linear convergence failures : 1319
   Number of Jacobian-evaluations            : 5526
   Number of crossing function evaluations   : 82065
   Number of model time events               : 1759
   Number of state events                    : 170
   Number of step events                     : 0

So a gain of nearly 20% in CPU-time for integration. However, this comes with a high number of

cvode: CVODE CVode At t = 165552, mxstep steps taken before reaching tout.

The conclusion differs with DASSL solver, with a gain closer to 50%. The log also shows that the integration order may effectively be reduced from 5 to 4 with the relaxed tolerance.

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow",
    method="dassl",
    tolerance=1e-6,    // Default setting.
    startTime=0,
    stopTime=6048000);

Integration terminated successfully at T = 6048000
   CPU-time for integration                : 41.2 seconds
   CPU-time for one grid interval          : 82.4 milliseconds
   CPU-time for initialization             : 0.81 seconds
   Number of result points                 : 4317
   Number of grid points                   : 501
   Number of accepted steps                : 83111
   Number of f-evaluations (dynamics)      : 175368
   Number of crossing function evaluations : 86702
   Number of Jacobian-evaluations          : 26127
   Number of model time events             : 1759
   Number of input time events             : 0
   Number of state events                  : 159
   Number of step events                   : 0
   Minimum integration stepsize            : 1.48e-10
   Maximum integration stepsize            : 796
   Maximum integration order               : 5

simulateModel("Buildings.Experimental.DHC.Examples.Combined.Generation5.Unidirectional.Examples.SeriesVariableFlow",
    method="dassl",
    tolerance=1e-4,    // Default setting.
    startTime=0,
    stopTime=6048000);

Integration terminated successfully at T = 6048000
   CPU-time for integration                : 22.7 seconds
   CPU-time for one grid interval          : 45.4 milliseconds
   CPU-time for initialization             : 0.776 seconds
   Number of result points                 : 4317
   Number of grid points                   : 501
   Number of accepted steps                : 37070
   Number of f-evaluations (dynamics)      : 77966
   Number of crossing function evaluations : 40845
   Number of Jacobian-evaluations          : 16863
   Number of model time events             : 1759
   Number of input time events             : 0
   Number of state events                  : 159
   Number of step events                   : 0
   Minimum integration stepsize            : 1.48e-08
   Maximum integration stepsize            : 1.77e+03
   Maximum integration order               : 4

lbl-srg / modelica-buildings