Nonlinear ARX models extend the linear ARX model to the nonlinear case. For information about the structure of nonlinear ARX models, see What are Nonlinear ARX Models?
You can estimate nonlinear ARX models in the System Identification app or at
the command line using the
To estimate a nonlinear ARX model, you first prepare the estimation
data. You then configure the model structure and estimation algorithm,
and then perform the estimation. After estimation, you can validate
the estimated model as described in Validating Nonlinear ARX Models.
You can use uniformly sampled time-domain input-output data or time-series data (no inputs) for estimating nonlinear ARX models. Your data can have zero or more input channels and one or more output channels. You cannot use frequency-domain data for estimation.
To prepare the data for model estimation, import your data into the MATLAB® workspace, and do one of the following:
In the System Identification app — Import data into the app, as described in Represent Data.
At the command line —
Represent your data as an
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.
A nonlinear ARX model consists of standard and custom regressors, and a nonlinearity estimator. The block diagram represents the structure of a nonlinear ARX model in a simulation scenario.
To configure the structure of a nonlinear ARX model:
Configure the model regressors.
Choose the standard and custom regressors based on your knowledge of the physical system you are trying to model.
Specify standard regressors in one of the following ways:
Specify model orders and delay to create the set of standard regressors. For more information, see Nonlinear ARX Model Orders and Delay.
Initialize using a linear ARX model. You can perform this operation at the command line only. The initialization configures the nonlinear ARX model to use standard regressors of the linear model. For more information, see Initialize Nonlinear ARX Estimation Using Linear Model.
Specify custom regressors. Custom regressors are arbitrary functions of past inputs and outputs, such as products, powers, and other MATLAB expressions of input and output variables. Specify custom regressors in addition to, or instead of, standard regressors for greater flexibility in modeling your data.
Include only a subset of standard and custom regressors as inputs to the nonlinear function of the nonlinear estimator block. Including only a subset of regressors can help reduce model complexity and keep the estimation well-conditioned.
The choice of the subset to use can require multiple trials. You can examine a nonlinear ARX plot to help you gain insight into which regressors have the strongest effect on the model output. Understanding the relative importance of the regressors on the output can then help you decide which regressors to include in the nonlinear function.
Configure the nonlinearity estimator block.
Specify and configure the nonlinearity estimator, F(x).
Here, x is a vector of the regressors, and r is the mean of the regressors x. is the output of the linear function block and is affine when d ≠ 0. d is a scalar offset. represents the output of the nonlinear function block. Q is a projection matrix that makes the calculations well conditioned. The exact form of F(x) depends on your choice of the nonlinearity estimator. The default nonlinearity is a wavelet network. For information about the available nonlinearity estimators, see Available Nonlinearity Estimators for Nonlinear ARX Models.
You can also perform one of the following tasks:
Exclude the nonlinear function from the nonlinearity estimator such that F(x) = .
Exclude the linear function from the nonlinearity estimator such that F(x) = .
Note: You cannot exclude the linear function from tree partitions and neural networks.
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
nlarxOptions to configure
your loss function. You can specify the following options:
Focus — Specifies whether
the simulation or prediction error is minimized during parameter estimation.
By default, the software minimizes one-step prediction errors, which
correspond to a
Focus value of
If you want a model that is optimized for reproducing simulation behavior,
Minimization of simulation error requires differentiable nonlinear
functions and takes more time than one-step-ahead prediction error
minimization. Thus, you cannot use
when minimizing the simulation error because these nonlinearity estimators
are not differentiable.
OutputWeight — Specifies
a weighting of the error in multi-output estimations.
Regularization — Modifies
the loss function to add a penalty on the variance of the estimated
parameters. For more information, see Regularized Estimates of Model Parameters.
To estimate a nonlinear ARX model, the software uses iterative
search algorithms to minimize the error between the simulated or predicted
model output and the measured output. At the command line, use
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
SearchOption — Option set
for the search algorithm, with fields that depend on the value of
MaxIter — Maximum number
Tolerance — Condition for
terminating iterative search when the expected improvement of the
parameter values is less than a specified value.
After preprocessing the estimation data and configuring the
model structure and estimation options, you can estimate the model
in the System Identification app,
nlarx at the command
line. The resulting model is an
that stores all model data, including model regressors and parameters
of the nonlinearity estimator. For more information about these model
objects, see Nonlinear Model Structures. You can validate the estimated
model as described in Validating Nonlinear ARX Models.
At the command line, you can use an ARX structure polynomial
idpoly with only A and B as
active polynomials) for nonlinear ARX estimation. To learn more about
when to use linear models, see When to Fit Nonlinear Models.
Typically, you create a linear ARX model using the
arx command. You can provide the linear
model when constructing or estimating a nonlinear ARX model. For example,
use the following syntax to estimate a nonlinear ARX model using estimation
data and a linear ARX model
m = nlarx(data,LinARXModel)
Assigning the linear ARX model orders and delays as
initial values of the nonlinear ARX model orders (
idnlarx object) and delays (
The software uses these orders and delays to compute standard regressors
in the nonlinear
ARX model structure.
Using the A and B polynomials
of the linear model to compute the linear function of the nonlinearity
LinearCoef parameter of the nonlinearity
estimator object), except if the nonlinearity estimator is a neural
During estimation, the estimation algorithm uses these values to adjust the nonlinear model to the data.
Note: When you use the same data for estimation, a nonlinear ARX model initialized using a linear ARX model produces a better fit to measured output than the linear ARX model itself.
By default, the nonlinearity estimator is the wavelet network
wavenet object). You can also
specify different input and output nonlinearity estimators. For example,
you can specify a sigmoid network nonlinearity estimator.
m = nlarx(data,LinARXModel,'sigmoid')
For an example, see Estimate Nonlinear ARX Models Initialized Using Linear ARX Models.