Hammerstein-Wiener models describe dynamic systems using one or two static nonlinear blocks in series with a linear block. The linear block is a discrete transfer function and represents the dynamic component of the model. For more information about the structure of these models, see What are Hammerstein-Wiener Models?
You can estimate Hammerstein-Wiener models in the System Identification app or at the command line using the
nlhw command. To estimate a Hammerstein-Wiener model, you first prepare
the estimation data. You then configure the model structure and estimation algorithm,
and then perform estimation. After estimation, you can validate the estimated model as
described in Validating Hammerstein-Wiener Models.
You can use only uniformly sampled time-domain input-output data for estimating Hammerstein-Wiener models. Your data can have one or more input and output channels. You cannot use time series data (output only) or frequency-domain data for estimation. Use nonlinear ARX or nonlinear grey-box models for time series data.
To prepare the data for model estimation, import your data into the MATLAB® workspace, and do one of the following:
After importing the data, you can analyze data quality and preprocess data by interpolating missing values, filtering to emphasize a specific frequency range, or resampling using a different sample time. For more information, see Ways to Prepare Data for System Identification. For most applications, you do not need to remove offsets and linear trends from the data before nonlinear modeling. However, data detrending can be useful in some cases, such as before modeling the relationship between the change in input and output about an operating point.
After preparing your estimation data, you can configure your model structure, loss function, and estimation algorithm, and then estimate the model using the estimation data.
The Hammerstein-Wiener model structure consists of input and output nonlinear blocks in series with a linear block. The linear block is a discrete transfer function and represents the dynamic component of the model.
To configure the structure of a Hammerstein-Wiener model:
Configure the linear transfer function block.
Perform one of the following:
Specify model order and input delay for the linear transfer function as:
nb — Number of zeros plus one. nb is the length of the numerator (B) polynomial.
nf — Number of poles. nf is the order of the transfer function denominator (F polynomial).
nk — Delay from input to the output in terms of the number of samples.
For MIMO systems with Ny outputs and Nu inputs, nb, nf, and nk are Ny-by-Nu matrices.
Initialize the linear block using a discrete-time linear model — You can initialize using linear models at the command line only. The initialization sets the transfer function of the linear block to that of the specified linear model. For more information, see Initialize Hammerstein-Wiener Estimation Using Linear Model.
Configure the input and output nonlinearities, f and h respectively.
The default input and output nonlinearity estimators are piecewise linear
functions. See the
pwlinear reference page for
more information. To configure the input and output nonlinearity
Choose the type of input and output nonlinearity estimators, and configure their properties.
For a list of available nonlinearity estimators, see Available Nonlinearity Estimators for Hammerstein-Wiener Models.
Exclude the input or output nonlinear block.
You do not have to include both the input and the output nonlinearity in the model structure. When a model contains only the input nonlinearity f, it is called a Hammerstein model. Similarly, when the model contains only the output nonlinearity h, it is called a Wiener model.
For information about how to configure the model structure at the command line and in the app, see Estimate Hammerstein-Wiener Models at the Command Line and Estimate Hammerstein-Wiener Models in the App.
To configure the model estimation, specify the loss function to be minimized, and choose the estimation algorithm and other estimation options to perform the minimization.
The loss function or cost function is a function of the error between the model output and the measured output. For more information about loss functions, see Loss Function and Model Quality Metrics.
At the command line, use the
nlhw option set,
nlhwOptions to configure your loss
function. You can specify the following options:
OutputWeight — Specify a weighting of the
error in multi-output estimations.
Regularization — Modify the loss function to
add a penalty on the variance of the estimated parameters. For more
information, see Regularized Estimates of Model Parameters.
For details about how to specify these options in the app, see Estimate Hammerstein-Wiener Models in the App.
To estimate a Hammerstein-Wiener model, the software uses iterative search
algorithms to minimize the loss function. At the command line, use
nlhwOptions to specify the search algorithm and other
estimation options. Some of the options you can specify are:
SearchMethod — Search method for
minimization of prediction or simulation errors, such as Gauss-Newton
and Levenberg-Marquardt line search, and Trust-region reflective Newton
SearchOptions — Option set for the search
algorithm, with fields that depend on the value of
SearchMethod, such as:
MaxIterations — Maximum number
of iterations to perform.
Tolerance — Condition for
terminating iterative search when the expected improvement
of the parameter values is less than a specified
InitialCondition — By default, the software
treats the initial states of the model as zero and does not estimate the
states. You can choose to estimate initial states, which sometimes can
improve parameter estimates.
After preprocessing the estimation data and configuring the model structure,
loss function, and estimation options, you can estimate the model in the System Identification app, or using
nlhw. The resulting model is an
idnlhw object that stores all
model data, including model parameters and nonlinearity estimator. For more
information about these model objects, see Nonlinear Model Structures. You can
validate the estimated model as described in Validating Hammerstein-Wiener Models.
At the command line, you can use one of the following linear models to initialize the linear block of a Hammerstein-Wiener model:
Polynomial model of Output-Error (OE) structure
State-space model with no disturbance component (
model with K = 0)
Transfer function (
Typically, you use the
tfest commands to obtain the linear
model. You can provide the linear model when constructing or estimating a
Hammerstein-Wiener model. For example, use the following syntax to estimate a
Hammerstein-Wiener model using estimation data and a linear model
m = nlhw(data,LinModel)
Assigning the linear model orders as initial values of nonlinear model
nf properties of the
idnlhw) and delays
Setting the B and F polynomials of the linear transfer function in the Hammerstein-Wiener model structure.
During estimation, the estimation algorithm uses these values to adjust the
nonlinear model to the data. By default, both the input and output nonlinearity
estimators are piecewise linear functions (see
You can also specify different input and output nonlinearity estimators. For example, a sigmoid network input nonlinearity estimator and a dead-zone output nonlinearity estimator.
m = nlhw(data,LinModel,'sigmoidnet','deadzone')
For an example, see Estimate Hammerstein-Wiener Models Initialized Using Linear OE Models.