Main Content


Evaluate output values of idnlarx or idnlhw mapping object array for given set of input values



Value = evaluate(MO,X) computes the value of a mapping object array MO that contains objects such as idWaveletNetwork objects for the inputs in X. For a list of available mapping objects, see Available Mapping Functions for Nonlinear ARX Models or Available Nonlinearity Estimators for Hammerstein-Wiener Models. Use this syntax when you want to independently evaluate the output of a mapping object array that you are using to estimate an idnlarx or idnlhw model.


Value = evaluate(MO,X,varnames) specifies the list of variable names varnames that orders the input data for the MO.Inputs property. Use this syntax when the columns of data in X are not in the order that the nonlinear model blocks require.


collapse all

Create a linear input signal u with slope m.

m = 0.1;
u = m*[1:100]';

Create an idDeadZone object with a dead zone between 3 and 5.

MO = idDeadZone;
MO.ZeroInterval = [3 5]
MO = 
Dead Zone
Inputs: u(t)
Output: y(t)
    ZeroInterval: [3 5]
            Free: [1 1]

Evaluate the output of MO using u as the input.

e = evaluate(MO,u);
title('Dead Zone Output')

The plot shows a zone of no response between 3 and 5.

Load the data, which contains input matrix u and output matrix y. u contains data for six inputs and y contains data for two outputs. Encapsulate u and y in an iddata object with a sample time of 0.02 sec.

load motorizedcamera
z = iddata(y,u,0.02,'Name','Motorized Camera','Timeunit','s');

Estimate idnlarx Model and Predict Model Output

Specify the order nn for the nlarx estimation.

na = [2 1;1 4];
nb = [1 4 2 2 1 3;3 3 3 1 2 4];
nc = [ones(2,3),zeros(2,3)];
nn = [na nb nc];

Estimate an idnlarx model after first turning off the normalization option. Specify two different mapping objects for the output function.

opt = nlarxOptions('Normalize',false);
sys = nlarx(z, nn, [idWaveletNetwork(3); idSupportVectorMachine('linear')],opt)
sys =

Nonlinear ARX model with 2 outputs and 6 inputs
  Inputs: u1, u2, u3, u4, u5, u6
  Outputs: y1, y2

  Linear regressors in variables y1, y2, u1, u2, u3, u4, u5, u6

Output functions:
  Output 1:  Wavelet network with 3 units
  Output 2:  Support Vector Machine function using a Linear kernel

Sample time: 0.02 seconds

Estimated using NLARX on time domain data "Motorized Camera".
Fit to estimation data: [98.87;91.79]% (prediction focus)    
FPE: 23.94, MSE: 22.44

Predict the model output using one-step-ahead prediction and store the result in yp.

yp = predict(sys,z,1,predictOptions('InitialCondition','zero'));
yp = yp.OutputData;

Use evaluate to Map Regressors to Mapping Object Output

Use getreg to compute and store the regressor data in D.

D = getreg(sys,z);
D = D{:,:};

The two mapping objects, y1 and y2, use different regressors. Because of this difference, evaluate requires the regressor names to order the regressors correctly for the object inputs. Use getreg to retrieve the regressor names and store them in regnames. View the first five regressor names.

regnames = getreg(sys);
ans = 5x1 cell

Evaluate the output of sys.OutputFcn using the regressor data and names.

yev = evaluate(sys.OutputFcn,D,regnames);

Compare Output Results

Plot the estimated and evaluated results for the two model outputs y1 and y2.

t = z.SamplingInstants;
title('One-Step prediction result comparison')
legend('Result using PREDICT','Result using GETREG, EVALUATE')
xlabel('Time (seconds)')

The predict and evaluate results are essentially the same.

Input Arguments

collapse all

Mapping object array, specified as a single mapping object or an array of length ny, where ny is the number of both mapping objects and outputs. For an idnlarx model sys, MO represents the sys.OutputFcn property. For an idnlhw model sys, MO represents the sys.InputNonlinearity or sys.OutputNonlinearity property.

Input values at which to evaluate MO, specified as a single numeric value or an nv-by-nx matrix, where nv is the number of points at which to evaluate MO and nx is the number of inputs to MO. For nonlinear ARX models, the inputs are the input regressor signals. For Hammerstein-Wiener models, the inputs are either the true input signals (input nonlinearity) or the outputs of the linear block (output nonlinearity).

Variable names associated with each column of X, specified as a string array of length nx, where nx is the number of input columns. varnames{i} corresponds to X(:,i). The software uses varnames to order the data columns in accordance with the property MO.Inputs. If you do not specify varnames, then the columns are ordered monotonically. Therefore, omit specifying varnames only when the columns of X are already in the same order as in MO.Inputs.

Output Arguments

collapse all

Evaluated values, returned as a numeric value or an nv-by-ny array of numeric values, where nv is the number of points to evaluate and ny is the number of outputs.

Version History

Introduced in R2007a