MultiSim is a simulation tool for energy systems consisting of multiple different parts like pipes, storages, valves, heat exchangers, etc. F.i. technical appliances such as space or water heating appliances can be designed, simulated and optimized.
MultiSim was mainly designed to solve the convection-diffusion-reaction-equation but can also be utilized to solve other differential equations. It features an adaptive step size solver, which is capable of yielding a stable solution even when loads fluctuate vigorously. Furthermore the solver can handle steps in flow variables due to interference of controllers (f.i. PID).
- Developer/maintainer: https://www.linkedin.com/in/johannes-elfner
The short documentation consists of:
- Installation instructions
- Description of available Components
- Getting startet
- Basic examples
- Summary of the validation of the simulation software
- Known limitations and To-do
To install the newest snapshot of MultiSim:
- Make a local clone of this repository or download the release to the installation path
cd
the console to the download/clone folderpip install -e .
to install MultiSim in editable mode orpip install .
if you want a fixed installation.
The newest snapshot of the branch master
is always fully operational, but may include code which will be deprecated in the next release.
If you want to stick to fixed releases, you can stick to the version published on PyPi and install MultiSim with:
pip install MultiSim
Fixed releases may be outdated by several months, so I recommend cloning this repository.
For slightly more detailed building, distribution and installation instructions, see INSTALL.rst.
MultiSim supports different types of components, most notably:
- Parts which require differential equations to be solved (called "basic parts" hereafter)
- Connectors and Actuators which can affect flow variables, either controlled by controllers, stationary or time series based
- Controllers which control actuators
- Boundary Conditions like in- and outflows to ambience, both stationary and time series based
- Compound parts consisting of multiple parts, actuators, controllers and boundary conditions
- Meters to track process variables of specific important parts/cells and perform basic calculations on-the-fly
- Utility tools for pre- and postprocessing, plotting, loading and storing data, ...
The following basic parts are currently available:
- Pipe
- Thermal storage (TES)
- Heat exchanger
- Numeric (by solution of differential equations)
- Non-numeric condensing flue gas heat exchanger based on fitting an ElasticNet regression to measurement data (very specific to the type of HEX)
Parts derived by class inheritance of the basic parts:
- Heated pipe
- Branched pipe, pipe with valve, pipe with pump (these are also compound parts)
The following actuators and connectors can be installed:
- Three way connector
- Mixing valve/splitting valve/three way valve
- Branch
- Pump
- Connector to ambient conditions/boundary conditions
All actuators/connectors can be controlled by controllers, set to a static value or follow a predefined behaviour by defining a time series.
parts/controllers defines the following controllers:
- PID controller
- Bang–bang controller (also 2 step or on–off controller) with a hysteresis
- Two sensor controller (switch on when sensor 1 is > or < than setpoint 1, switch off when sensor 2 is > or < than setpoint 2)
- Model predictive controller (CHP plant specific to optimally follow a predicted electric profile)
All controllers support setting:
- Fixed setpoints and setting the setpoint to a process variable of another part
- Control variable saturation limits
- Part specific control variable post processing like conversion to a specific value range
- Setting slopes to control variable changes
- Linking controllers to make controller action depend on another controller to construct control chains
PID controllers additionally support semi-automatic tuning by Ziegler-Nichols method.
Thus preferred tuning method for PID controllers is Ziegler-Nichols, since the parameters Kp_crit
and T_crit
can be passed directly to the controller while specifying the aggressiveness of the PID controller with rules like classic
or pessen-int
(Pessen integral rule).
Compound parts consisting of multiple other parts and controllers can be found in parts/part_modules. Part dimensions, such as pipe diameters, and controller coefficients have been fit to a wide range of flow speeds and temperatures, but may be adjusted if controls show instabilities or if the solver requires too many retries to find a stable solution. The following compound parts can be used:
- Gas boiler
- Chp plant, also with flue gas heat exchanger (based on fitting a model to manufacturer specific measurement data)
- Consumer appliances
- Space heating
- State-of-the-art water heating
- Low exergy water heating
New parts and controllers can be added either by defining completely new classes or by inheriting from existing parts.
There is also a list of sensors/meters which can be "installed" at any (numeric) cell of each part to track the state of this cell or perform calculations like energy flows, cumulated mass and energy flows etc. on the fly, such as:
- Temperature sensor
- Mass flow sensor
- Heat meter (power, mass flow, volume flow, temperature of hot and cold part, cumulated values)
The file utility_functions provides methods for pre- and post-processing of input/output data and basic plotting/statistic analysis. Also methods to open the *.hdf5
files, which are used to store the results on disk, are provided.
But: utility_functions.py
requires heavy refactoring!! This is scheduled for the next release.
Some parts have already been refactored to multisim/_utility/:
- Meters, also see section Meters
- plotting provides basic plotting methods for validation and heatmap plots and also some formatting helpers. More will be added soon.
- Statistical error measures provides basic error measures useful for validation, such as the (adjusted) coefficient of determination, MSE, RMSE, CV(RMSE), NME, ...
Import MultiSim and create your simulation environment instance with:
import multisim as ms
my_sim = ms.SimEnv()
Now simply follow the detailed step-by-step instructions printed to the console.
We will cover three basic examples in this section, all covering the temperature control of a three-way-valve flowing into a thermal energy storage (TES):
- A stable PID controller (loop tuned with Ziegler-Nichols)
- An instable PID controller. Stable at first for small steps in the process variable, but instable with persisting oscillations for larger steps.
- A bang-bang controller to control the pump.
The appliance/setup to simulate is, in all three cases, the following:
With the temperature of the water flowing into port B of 'pipe_in' describing a step from 50.0 °C to 85.0 °C after 300 s.
This small example covers controlling the mixing temperature of the three-way-valve via a stable PID controller. The PID controller is tuned using the class Ziegler-Nichols rule.
The full executable example as a Python script can be found at doc/examples/basic_loop_w_stable_pid.py.
To set up the simulation environment, first start by loading the required modules and defining boundary conditions like the temperatures and temperature time series:
import matplotlib as mpl
import matplotlib.pyplot as plt
import pandas as pd
import multisim as ms
# define temperatures
sp_pid = 40.0 # setpoint for the PID in degree celsius
theta_low = 20.0 # degree celsius
theta_high = pd.Series(
data=50.0, index=pd.date_range('2021-01-01', periods=1000, freq='1s')
)
theta_high.iloc[300:] = 85.0
Now create an instance of the simulation environment, and set some basic options:
# create simulation environment
my_sim = ms.Models()
# set disksaving, simulatiion timeframe and solver
my_sim.set_disksaving(save=True, start_date='infer', sim_name='sim_a')
my_sim.set_timeframe(timeframe=900, adaptive_steps=True)
my_sim.set_solver(solver='heun', allow_implicit=False)
Next define specifications for pipes and the ports at each pipe. Dimensions can be set individually for each port. In this case all ports and parts share the same specifications.
# define pipe specifications for all pipes and ports
pipe_specs = {'all': {'pipe_type': 'EN10255_medium', 'DN': 'DN25'}}
# set general specifications for all parts
general_specs = dict(
insulation_thickness=1e-2, # insulation around pipe in meters
insulation_lambda=0.035, # insulation heat conductivity in W/(m*K)
T_init=theta_low, # initial temperature of all cells
T_amb=theta_low, # ambient temperature of all parts
material='carbon_steel',
pipe_specs=pipe_specs,
)
Now add the parts to the simulation environment:
my_sim.add_part( # add pipe_in with the valve to control
part=ms.ap.PipeWith3wValve,
name='pipe_in',
length=2.0, # in meters
grid_points=20, # number of numeric cells to calculate
valve_location=5, # location of the three-way-valve in grid_points
start_portA_opening=0.5, # initialize the valve
lower_limit=0.0, # lower limit for the valve, can be 0 <= x < 1
upper_limit=1.0, # upper limit for the valve, can be 0 < x <= 1
**general_specs,
)
my_sim.add_part(
part=ms.ap.Tes, # add a thermal energy storage
name='TES',
volume=0.5, # volume in m**3
grid_points=20,
outer_diameter=1.0, # outer diameter in meters
shell_thickness=5e-3, # shell/casing thickness in meters
new_ports=None, # add no additional ports/connectors
**general_specs,
)
my_sim.add_part(
part=ms.ap.PipeWithPump, # add a pipe with a pump
name='pipe_out',
length=1.0,
grid_points=10,
start_massflow=0.75, # initialize massflow in kg/s
ctrl_required=False, # set to constant or time series based
const_val=0.75, # constant massflow
**general_specs,
)
Add open ports (connections to ambient conditions, connections crossing the control volume of the simulation environment, other boundary conditions (BC)). Open ports can be either constant or time series based.
my_sim.add_open_port('BC_theta_low', constant=True, temperature=theta_low)
my_sim.add_open_port('BC_theta_high', constant=False, temperature=theta_high)
my_sim.add_open_port('BC_out', constant=True, temperature=theta_low)
Connect parts at ports and also boundary conditions to parts:
my_sim.connect_ports(
first_part='BoundaryCondition',
first_port='BC_theta_low',
scnd_part='pipe_in',
scnd_port='B',
)
my_sim.connect_ports(
first_part='BoundaryCondition',
first_port='BC_theta_high',
scnd_part='pipe_in',
scnd_port='A',
)
my_sim.connect_ports(
first_part='pipe_in', first_port='AB', scnd_part='TES', scnd_port='in',
)
my_sim.connect_ports(
first_part='TES', first_port='out', scnd_part='pipe_out', scnd_port='in',
)
my_sim.connect_ports(
first_part='pipe_out',
first_port='out',
scnd_part='BoundaryCondition',
scnd_port='BC_out',
)
Add and set PID control to control the 3-way-valve.
Nomenclature: setpoint (SP), control variable (CV), process variable (PV).
For this, the critical coefficient Kp_crit
causing permanent oscillations and the period of the oscillations T_crit
have to be estimated before. This can be done by increasing Kp
of a PID controller set to loop_tuning='manual'
and terms='P'
until oscillations after a step start pertaining.
my_sim.add_control(
ms.ap.PID,
name='pid_valve',
actuator='pipe_in', # controlled actuator
process_CV_mode='part_specific', # allow post-processing of CV in part
CV_saturation=(0.0, 1.0), # clip CV
controlled_part='pipe_in', # part where the PV is found
controlled_port=-1, # port or cell where the PV is found in its part
reference_part='none', # use another part as source of the SP
setpoint=sp_pid, # use defined constant value
sub_controller=False, # controller action is not depending on another ctrl
off_state=0.0, # which value shows that the controller is off?
time_domain='discrete', # integral and derivative calculation type
deadtime=0.0, # in seconds
slope=(-0.1, 0.1), # in units/s
invert=False, # invert action to allow reversed operation
terms='PID', # which coefficients to use
loop_tuning='ziegler-nichols', # semi-automatic loop tuning or manual?
rule='classic', # loop tuning rule
Kp_crit=0.025, # critical Kp value
T_crit=5.0, # period of the oscillations in seconds
filter_derivative=False, # low pass filter of the derivative term
anti_windup=1.0, # anti windup for the integral term
)
Initialize simulation (set up parts and controllers, preallocate arrays, calculate topology...) and run it:
my_sim.initialize_sim()
my_sim.start_sim()
Add meters:
meters = ms.Meters(my_sim, start_time=theta_high.index[0])
meters.temperature(name='theta_mix', part='pipe_in', cell=-1)
meters.heat_meter(
name='hm',
warm_part='pipe_in',
warm_cell=-1,
cold_part='pipe_out', # massflows will be calculted on the cold cell
cold_cell=0,
)
meters.massflow(name='mflow_A', part='pipe_in', cell=0)
meters.massflow(name='mflow_AB', part='pipe_in', cell=-1)
return results as a dictionary of kind {part:{'res': temperatures, 'dm': massflows}}
:
results = my_sim.return_stored_data()
For plotting of the results, define the plot index:
plot_idx = results['TES']['res'].index
Now plot the temperatures and massflows of the valve respectively of part 'pipe_in'
:
fig_valve, (ax_valve_flow, ax_valve_theta) = plt.subplots(
1, 2, sharex=True, figsize=(16 / 2.54, 6 / 2.54)
)
# plot massflows
ax_valve_flow.plot(
plot_idx, results['meters']['mflow_A'], label=r'$\dot{m}$ port A'
)
ax_valve_flow.plot(
plot_idx,
results['meters']['mflow_AB'].sub(results['meters']['mflow_A'].values),
label=r'$\dot{m}$ port B',
)
ax_valve_flow.plot(
plot_idx, results['meters']['mflow_AB'], label=r'$\dot{m}$ total'
)
# plot temperatures
ax_valve_theta.hlines(
theta_low, plot_idx[0], plot_idx[-1], label=r'$\theta$ port A'
)
ax_valve_theta.plot(
plot_idx, theta_high.reindex(plot_idx), label=r'$\theta$ port B'
)
ax_valve_theta.plot(
plot_idx, results['meters']['theta_mix'], label=r'$\theta$ mix'
)
# legends, ax labels, formatting and layout
ax_valve_flow.legend()
ax_valve_theta.legend(loc='center right')
ax_valve_flow.set_ylabel(r'massflow $\dot{m}$ in kg/s')
ax_valve_theta.set_ylabel(r'temperataure $\theta$ in °C')
ax_valve_flow.set_xlabel('simulation time in min:s')
ax_valve_theta.set_xlabel('simulation time in min:s')
ax_valve_flow.xaxis.set_major_formatter(mpl.dates.DateFormatter('%M:%S'))
fig_valve.tight_layout(pad=0.1)
As you can see, the PID tuned with the Ziegler-Nichols method reaches the new SP quite fast and stable with only minor oscillations:
And finally plot a heatmap of the TES temperature:
# resample and select every second point in y-axis to reduce plot size:
tes_heatmap_rs = results['TES']['res'].resample('5s').mean().iloc[:, ::2]
fig_tes = plt.figure(figsize=(16.0 / 2.54, 6.0 / 2.54))
ax_tes = fig_tes.gca()
ms.plotting.heatmap_from_df(
tes_heatmap_rs,
ax=ax_tes,
ylabel=('TES height', 'm'),
cbar=True,
cbar_label=(r'Temperature\; $\theta$', '°C'),
vmin=20.0,
plt_kwds={'shading': 'gouraud'},
)
Which yields the temperature of the TES over the time:
Now we'll try what happens when we switch to manual loop tuning and increase the Kp
coefficient to be slighty Kp > Kp_crit
.
Furthermore we add small integral and derivative terms Ki
and Kd
.
You can find the full executable example as a Python script at doc/examples/basic_loop_w_instable_pid.py.
The only change compared to the previous stable PID example is within the controller construction in the following lines:
my_sim.add_control(
ms.ap.PID,
name='pid_valve',
actuator='pipe_in', # controlled actuator
process_CV_mode='part_specific', # allow post-processing of CV in part
CV_saturation=(0.0, 1.0), # clip CV
controlled_part='pipe_in', # part where the PV is found
controlled_port=-1, # port or cell where the PV is found in its part
reference_part='none', # use another part as source of the SP
setpoint=sp_pid, # use defined constant value
sub_controller=False, # controller action is not depending on another ctrl
off_state=0.0, # which value shows that the controller is off?
time_domain='discrete', # integral and derivative calculation type
deadtime=0.0, # in seconds
slope=(-0.1, 0.1), # in units/s
invert=False, # invert action to allow reversed operation
terms='PID', # which coefficients to use
loop_tuning='manual', # no loop tuning set
Kp=0.03, # Kp coefficient
Ki=0.02, # Ki coefficient == 1/Ti
Kd=0.01, # Kd coefficient == Td
filter_derivative=False,
anti_windup=1.,
)
Now the PID is stable at first and reaches equilibrium even faster than the stable PID. But this comes at the cost of persistent instability once a larger step in the PV is introduced, such as after 5 minutes:
Since the amplitude of the oscillations is still small and the product of the period with the massflow is small compared to the TES volume, the effect on the temperature in the TES is negligible:
Assuming in more complex systems other controllers are relying on a stable output of this controller, even oscillations this small my cause a chain reaction leading to an undesired behavior of the complete system. Furthermore instabilities may cause the solver to require much smaller steps to solve the differential equations.
coming soon
MultiSim has been fully validated following standard ANSI/BPI-2400-S-2015. A stricter set of statistic measures than provided in the standard has been used.
Since MultiSim was part of a PhD thesis, validation results will be added as soon as the thesis has been published (approximately December 2021).
One of the systems simulated to validate MultiSim, was the following heating and PWH (potable water hot) appliance:
The 2D-KDE prediction-realization plot, created with multisim.plotting.prediction_realization_2d_kde shows that almost all points lie close to the halving diagonal:
This is also confirmed by the statistical error measures in the plots. The coefficient of determination is, in both cases, >0.9
. ANSI defines validation bounds of CV(RMSE) <= 0.3
and |NME| <= 0.05
. Both bounds are satisfied.
Finally, the heatmap plot of the TES temperature, plotted with multisim.plotting.heatmap_from_df, shows only minor differences:
Further validation results of other parts and with more plots will be added to the full documentation as soon as everything is officially published and can be referenced.
Even though MultiSim is fully operational, many things have to be refactored,
replaced, improved or deprecated. Especially the core class SimEnv
and the
most basic parts like pipes and TES require a general overhaul. Thus
enhancements should start here.
Furthermore current tests used for TDD are based on proprietary measurement data. Thus these tests cannot be published. Hence tests included in this public GitHub repo are merely truncated stumps. Using free data to integrate extensive tests will be an important step.
The documentation is currently missing but will be added step by step.
Other enhancements could be:
-
Extend the tests using
pytest
with non-proprietary data. -
Implementing the implicit differential equation solver in numba to speed things up considerably. Implicit solving is currently slowing down the simulation progress.
-
Move the outer explicit solver loop to numba. This should also improve the performance by several percent points.
-
Check if local Nusselt number calculation is implemented in each relevant part.
-
Fully implement parts with compound structures.
-
Refactor type checks during part-adding as much as possible, using
@property
may help. Move away from using to manykwargs
-based arguments. Include more specific type hints and default args handling. -
Move to Python 3.8, using type hints and assignment expressions.
-
There is a lot of chaos in
utility_functions
. This needs some heavy refactoring and tidying. -
Write a documentation.
MultiSim depends mainly on numpy
, numba
, pandas
, scipy
, and
matplotlib
. For some parts scikit-learn
is a dependency.