Neural State-Space Model

Simulate neural state-space model in Simulink

Since R2022b

• 

×

Libraries:
System Identification Toolbox / Models

Description

Simulates neural state-space (idNeuralStateSpace) models.

Ports

Input

u — Input of the neural state-space model

scalar | vector

Input of the neural state-space model, specified as a vector signal with as many elements as the number of inputs you specify when creating the corresponding idNeuralStateSpace object. If the idNeuralStateSpace object specified in the block dialog does not have inputs, this port does not appear.

x0 — Initial state of the neural state-space model

scalar | vector

Initial state of the neural state-space model, specified as a vector signal with as many elements as the number of states you specify when creating the corresponding idNeuralStateSpace object.

Dependencies

To enable this input port, select the initial state source parameter as External. Then, connect the port to an upstream signal supplying the value of the initial state.

Output

y — Output of the neural state-space model

scalar | vector

Output of the neural state-space model, returned as a vector signal with as many elements as the number of outputs you specify when creating the corresponding idNeuralStateSpace object.

x — State of the neural state-space model

scalar | vector

State of the neural state-space model, output as a vector signal with as many elements as the number of states n_x that you specify when creating the corresponding idNeuralStateSpace object. These values are identical to the first n_x values in the y output.

Dependencies

To enable this output port, select the Output state parameter.

dx — State derivative/difference of the neural state-space model

scalar | vector

State derivative or state difference of the neural state-space model, output as a vector signal with as many elements as the number of states you specify when creating the corresponding idNeuralStateSpace object.

For continuous-time neural state space models, this value is the derivative of the state with respect to time. For discrete-time neural state space models, this value is the difference between the next state and the current one.

Dependencies

To enable this output port, select the Output state derivative/difference parameter.

z — Latent state of the neural state-space model

scalar | vector

Internal (latent) state of the neural state-space model, output as a vector signal with as many elements as the value of the LatentDim property that you specify when creating the corresponding idNeuralStateSpace object.

Dependencies

To enable this output port, specify the value of the LatentDim property of the idNeuralStateSpace object as a finite positive scalar and select the Output state parameter.

dz — Latent state derivative/difference of the neural state-space model

scalar | vector

Internal (latent) state derivative or latent state difference of the neural state-space model, output as a vector signal with as many elements as the value of the LatentDim property that you specify when creating the corresponding idNeuralStateSpace object.

For continuous-time neural state space models, this value is the derivative of the latent state with respect to time. For discrete-time neural state space models, this value is the difference between the next latent state and the current one.

Dependencies

To enable this output port, specify the value of the LatentDim property of the idNeuralStateSpace object as a finite positive scalar and select the Output state derivative/difference parameter.

Parameters

Model — Neural state space object

idNeuralStateSpace(1,'NumInputs',1,'NumOutputs',2) (default) | idNeuralStateSpace object

Specify an idNeuralStateSpace object that defines a neural state-space system by entering the name of an idNeuralStateSpace object from the MATLAB^® workspace.

Note

If you change the input/output configuration of Model in the workspace, click the ... button next to the Model edit field in the block dialog to refresh the Neural State-Space Model block.

Programmatic Use

Block Parameter: sys

Type: string, character vector

Default: "idNeuralStateSpace(1,'NumInputs',1,'NumOutputs',2)"

Initial state source — Initial state source

internal (default) | external

Specify whether to enter the initial condition directly on the dialog box or to receive the initial condition from an input port.

• If you set Initial state source to internal, enter the initial condition in the edit field under Initial state.

• If you set Initial state source to external, the input port x0 is added to the block. You must then connect an upstream signal supplying the value of the initial state to the x0 input port.

Programmatic Use

Block Parameter: x0type

Type: character vector

Values: 'internal' | 'external'

Default: 'internal'

Initial state — Initial state

0 (default) | scalar | vector

Specify the initial state of the neural state-space object as a vector of n_x entries, where n_x is the number of states in Model.

Use this parameter make sure that the system states reflect the true plant environment at the start of your simulation, to the best of your knowledge.

Programmatic Use

Block Parameter: ic

Type: string, character vector

Default: "0"

Output state — Add external output state port

off (default) | on

Select this parameter to add the x output port to the block. During simulation this port outputs the internal state of the idNeuralStateSpace object specified in Model.

Programmatic Use

Block Parameter: returnx

Type: string, character vector

Values: "off", "on"

Default: "off"

Output state derivative/difference — Add external output state derivative port

off (default) | on

Select this parameter to add the dx output port to the block. During simulation this port outputs the time derivative of the internal state of the idNeuralStateSpace object specified in Model.

Programmatic Use

Block Parameter: returndx

Type: string, character vector

Values: "off", "on"

Default: "off"

Extended Capabilities

C/C++ Code Generation
Generate C and C++ code using Simulink® Coder™.

Version History

Introduced in R2022b

R2024b: Code generation does not create temporary .m files

Starting in 2024b, temporary .m files will no longer be created in the current folder when updating the Neural State-Space Model Block for prediction and code generation.

However, if the structure of the dlnetwork object assigned to the state or output networks of the neural state-space object is not a multi-layer perceptron (MLP) network, then the implementation of the Neural State-Space Model Block uses the Predict (Deep Learning Toolbox) Block and .mat files will still be generated in the current folder. In all other cases, .mat files will also not be created.