nwidPEM function

Identify a full directed network or a single target module in a directed network using prediction error minimization (PEM), based on the implementation in the System Identification toolbox. The network model is transformed into a state space model to support optimization with pem.

Syntax

• model = nwidPEM(data,model,orders,options)  performs identification based on the provided data, initialized network model, model orders and option set. The identified predictor model is returned.

Input arguments

• data

  nwdata object

  Network data object. The input excitation, input node and output node signals (as referred to by their labels) in the Rset, Dset and Yset of the predictor model must be present in the data.

• model

  nwpredmodel object	nwmodel object

  Initialized network/predictor model object. Use an nwmodel object to identify the full network, and an nwpredmodel object to identify a single target module. The properties of the predictor mappings G, T and H are used in the algorithm, with some restrictions:

  • The algorithm currently only supports Variable = z^-1, hence only discrete-time identification is possible.
  • H must specify a square matrix (number of noise signals is equal to the number of output nodes) and is enforced to be monic.

• orders

  structure array

  Model orders. Can be used to override properties of the initialized network model object. Use a structure array with optional fields G, T and H, which should each be structure arrays themselves with the optional fields npnum, npden and d (to set the analogous PredictorMap properties). Use an empty structure array to use the initialized network model object without modifications.

• options

  nwidPEMOptions object

  Option set for nwidPEM function. Can be constructed using options = nwidPEMOptions for the default option set or options = nwidPEMOptions(Name=Value,...) to set specified options.

Output arguments

• model

  nwpredmodel object	nwmodel object

  Estimated network/predictor model object. This is an nwmodel object after full network identification, and an nwpredmodel object after single module identification. The pnum and pden properties of the G, T and H fields are populated with the estimated parameter values. The e properties of G, T and H contain the fixed and estimated transfer functions of the predictor model connections.