nnidenthelp_main.m nncontrol 工具箱 matlab 源码程序 - matlab工具箱源码

www.gusucode.com > nncontrol 工具箱 matlab 源码程序 > nncontrol/private/nnidenthelp_main.m

% OVERVIEW
%
% The Plant Identification GUI is an interactive environment for developing
% a Neural Network capable of modeling a given plant.
%
% Flip through the remaining Topics for a detailed description of how
% to use these and other Plant Identification GUI features.
%
% MENUS
%
% The menus provide additional options for setting up and configuring
% the controller. The menus available are as follows.
%
% 1) File:
% a) Import Network: Import neural network plant model weights.
% b) Export Network: Export neural network plant model weights.
% c) Save: Load all parameters into the Simulink controller block.
% d) Save and Exit: Load all parameters into the Simulink controller block and close this menu.
% e) Exit Without Saving: Close the Plant Identification GUI and all related windows.
%
% 2) Window:
% Show and switch between all the open windows.
%
% 3) Help:
% a) Main Help: Open the general Indirect Adaptive Control GUI help text.
% b) All other Help menus: Open tool specific help text.
%
% NEURAL NETWORK PLANT STRUCTURE
%
% The two-layer neural network plant has an input layer with a tansig transfer
% function. There are two sets of inputs to the plant model: delayed values of
% the plant output and delayed values of the controller output. The output
% layer has a purelin transfer function. You can set the size of the hidden
% layer.
% For the NARMA-L2 controller, the plant model has a more complex structure.
% The inputs to the network are the same, but the network has four layers
% instead of two. See the User's Guide for a complete description.
%
% SIMULINK PLANT MODEL
%
% You enter the name of a simulink file that has the plant model to be
% used in the identification process.
%
% The Simulink model must have one inport block and one outport block.
% The Simulink model will be used to generate data for the plant
% identification. Random inputs will be applied to the model to
% generate the training data.
%
% NEURAL NETWORK INPUTS
%
% The neural network plant model has two inputs available:
%
% 1)Delayed Controller Outputs.
% 2)Delayed Plant Outputs.
%
% For each input you must specify the number of delays to be used.
% The delays are based on the sample time defined in the Sampling Interval
% field. For each plant input, you can select any nonzero value for
% the number of delays.
%
% The sampling time is given in seconds.
%
% TRAINING FUNCTION
%
% The Plant Identification algorithm has the following algorithms available
% for training:
%
% 1) trainbfg: BFGS quasi-Newton backpropagation
% 2) trainbr: Bayesian regularization backpropagation
% 3) traincgb: Conjugate gradient backpropagation with Powell-Beale
% restarts.
% 4) traincgf: Conjugate gradient backpropagation with Fletcher-Reeves
% updates.
% 5) traincgp: Conjugate gradient backpropagation with Polak-Ribiere
% updates.
% 6) traingd: Gradient descent backpropagation.
% 7) traingdm: Gradient descent with momentum backpropagation.
% 8) traingda: Gradient descent with adaptive learning rate backpropagation.
% 9) traingdx: Gradient descent with momentum & adaptive learning rate
% backpropagation.
% 10) trainlm: Levenberg-Marquardt backpropagation.
% 11) trainoss: One step secant backpropagation.
% 12) trainrp: Resilient backpropagation algorithm (RPROP).
% 13) trainscg: Scaled conjugate gradient backpropagation.
%
% TRAINING DATA
%
% You have two options for obtaining the data used to train the neural
% network plant model:
%
% 1) Import Training Data: Here you have a file with the data
% used for training. The data is retrieved from a .mat file whose name
% you enter in the appropriate field. The data file can contain a structure
% with fields named U and Y for the input and output of the plant, respectively.
% It can also obtain two individual arrays.
%
% 2) Generate Training Data: You allow the GUI to generate the random
% training data to be used in the identification process. You must
% define the minimum and maximum values of the random control signal.
% The simulink file with the plant model is used to generate the targets.
% If the user selects Limit Output Data, the GUI will stop the target
% generation process each time a limit is violated. The simulation
% process will then continue with new initial conditions. The number of
% training samples will define how many random inputs will be applied
% to the simulink plant to generate the targets.
%
% The data will be normalized to a range 0-1 if you select the Normalize
% Training Data option. This option is preferred when trainbr is used as the
% training function.
%
% TRAINING EPOCHS
%
% Defines the number of iterations that will be applied to train the neural
% network plant model.
%
% USE VALIDATION/TESTING DATA
%
% The Validation option is used to stop training early if the network
% performance on the validation data fails to improve or remains the same
% for 5 epochs in a row.
%
% The Testing option is used to test the generalization capability of the
% trained network. The error on a test data set is monitored and displayed
% during training.
%
% If any of these options are selected, 25 % of the data is used for each
% (validation or testing) option, allowing a minmum of 50 % for training
% if both options are selected. After training, graphs are created to present
% the training data (and the validation and testing data if selected). You can
% then continue training or repeat the training with new random initial weights.