Simulink 深度学习
R2026b通过使用 Deep Learning Toolbox™ 中包含的 Deep Neural Networks、Python Neural Networks 和 Deep Learning Layers 模块库中的模块，或使用 Computer Vision Toolbox™ 中包含的 Analysis & Enhancement 模块库中的 Deep Learning Object Detector 模块，在 Simulink® 模型中实现深度学习功能。
要生成使用 Deep Learning Layers 模块库来表示网络的 Simulink 模型，请使用 exportNetworkToSimulink 函数。Deep Learning Layers 模块库中的模块支持代码生成。
Simulink 中某些深度学习功能使用的 MATLAB Function 模块需要支持的编译器。对于大多数平台，会随 MATLAB® 安装提供一个默认的 C 编译器。使用 C++ 语言时，必须安装兼容的 C++ 编译器。要查看支持的编译器列表，请打开支持和兼容的编译器，点击与您的操作系统对应的选项卡，找到 Simulink Product Family 表，并转至 For Model Referencing, Accelerator mode, Rapid Accelerator mode, and MATLAB Function blocks 列。如果您的系统上安装了多个 MATLAB 支持的编译器，可以使用 mex -setup 命令更改默认编译器。请参阅更改默认编译器。
默认情况下，Deep Neural Networks 模块库中的模块会根据模型配置和代码生成支持，自动选择使用代码生成或解释执行。使用解释执行时，模块通过 MATLAB 执行引擎运行，无需编译器。有关详细信息，请参阅Improve Performance of Deep Learning Simulations in Simulink。
函数
exportNetworkToSimulink | Generate Simulink model that contains deep learning layer blocks and subsystems that correspond to deep learning layer objects (自 R2024b 起) |
模块
 | Classify | Classify data using a trained deep learning neural network |
 | Predict | 使用经过训练的深度学习神经网络预测响应 |
 | Stateful Classify | 使用经过训练的深度学习循环神经网络对数据进行分类 |
 | Stateful Predict | 使用经过训练的循环神经网络预测响应 |
 | Deep Learning Object Detector | 使用经过训练的深度学习目标检测器检测目标 |
 | TensorFlow Model Predict | Predict responses using pretrained Python TensorFlow model (自 R2024a 起) |
 | PyTorch Model Predict | Predict responses using pretrained Python PyTorch model (自 R2024a 起) |
 | ONNX Model Predict | Predict responses using pretrained Python ONNX model (自 R2024a 起) |
 | Custom Python Model Predict | Predict responses using pretrained custom Python model (自 R2024a 起) |
 | Clipped ReLU Layer | Clipped Rectified Linear Unit (ReLU) layer (自 R2024b 起) |
 | GELU Layer | Gaussian error linear unit (GELU) layer (自 R2024b 起) |
 | Leaky ReLU Layer | Leaky rectified linear unit (ReLU) layer (自 R2024b 起) |
 | PReLU Layer | Parameterized rectified linear unit (PReLU) layer (自 R2026a 起) |
 | ReLU Layer | Rectified linear unit (ReLU) layer (自 R2024b 起) |
 | Sigmoid Layer | Sigmoid layer (自 R2024b 起) |
 | Softmax Layer | Softmax layer (自 R2024b 起) |
 | Swish Layer | Swish layer (自 R2026a 起) |
 | Tanh Layer | Hyperbolic tangent (tanh) layer (自 R2024a 起) |
 | Addition Layer | Addition layer (自 R2024b 起) |
 | Concatenation Layer | Concatenation layer (自 R2024b 起) |
 | Depth Concatenation Layer | Depth concatenation layer (自 R2024b 起) |
 | Multiplication Layer | Multiplication layer (自 R2024b 起) |
 | Convolution 1D Layer | 1-D convolutional layer (自 R2024b 起) |
 | Convolution 2D Layer | 2-D convolutional layer (自 R2024b 起) |
 | Convolution 3D Layer | 3-D convolutional layer (自 R2024b 起) |
 | Fully Connected Layer | Fully connected layer (自 R2024b 起) |
 | Grouped Convolution 2D Layer | Grouped 2-D convolutional layer (自 R2026b 起) |
 | Rescale-Symmetric 1D | 1-D input layer with rescale-symmetric normalization (自 R2024b 起) |
 | Rescale-Symmetric 2D | 2-D input layer with rescale-symmetric normalization (自 R2024b 起) |
 | Rescale-Symmetric 3D | 3-D input layer with rescale-symmetric normalization (自 R2024b 起) |
 | Rescale-Zero-One 1D | 1-D input layer with rescale-zero-one normalization (自 R2024b 起) |
 | Rescale-Zero-One 2D | 2-D input layer with rescale-zero-one normalization (自 R2024b 起) |
 | Rescale-Zero-One 3D | 3-D input layer with rescale-zero-one normalization (自 R2024b 起) |
 | Zerocenter 1D | 1-D input layer with zerocenter normalization (自 R2024b 起) |
 | Zerocenter 2D | 2-D input layer with zerocenter normalization (自 R2024b 起) |
 | Zerocenter 3D | 3-D input layer with zerocenter normalization (自 R2024b 起) |
 | Zscore 1D | 1-D input layer with zscore normalization (自 R2024b 起) |
 | Zscore 2D | 2-D input layer with zscore normalization (自 R2024b 起) |
 | Zscore 3D | 3-D input layer with zscore normalization (自 R2024b 起) |
 | Batch Normalization Layer | Batch normalization layer (自 R2024b 起) |
 | Instance Normalization Layer | Instance normalization layer (自 R2026a 起) |
 | Layer Normalization Layer | Layer normalization layer (自 R2024b 起) |
 | Inverse Zerocenter | Inverse zero-center normalization (自 R2026a 起) |
 | Inverse Zscore | Inverse Z-score normalization (自 R2026a 起) |
 | Inverse Rescale-Zero-One | Inverse rescale-zero-one normalization (自 R2026b 起) |
 | Inverse Rescale-Symmetric | Inverse rescale-symmetric normalization (自 R2026b 起) |
 | Average Pooling 1D Layer | 1-D average pooling layer (自 R2024b 起) |
 | Average Pooling 2D Layer | 2-D average pooling layer (自 R2024b 起) |
 | Average Pooling 3D Layer | 3-D average pooling layer (自 R2024b 起) |
 | Global Average Pooling 1D Layer | 1-D global average pooling layer (自 R2024b 起) |
 | Global Average Pooling 2D Layer | 2-D global average pooling layer (自 R2024b 起) |
 | Global Average Pooling 3D Layer | 3-D global average pooling layer (自 R2024b 起) |
 | Global Max Pooling 1D Layer | 1-D global max pooling layer (自 R2024b 起) |
 | Global Max Pooling 2D Layer | 2-D global max pooling layer (自 R2024b 起) |
 | Global Max Pooling 3D Layer | 3-D global max pooling layer (自 R2024b 起) |
 | Max Pooling 1D Layer | 1-D max pooling layer (自 R2024b 起) |
 | Max Pooling 2D Layer | 2-D max pooling layer (自 R2024b 起) |
 | Max Pooling 3D Layer | 3-D max pooling layer (自 R2024b 起) |
 | Flatten Layer | Flatten layer (自 R2024b 起) |
 | GRU Layer | Gated recurrent unit (GRU) layer for recurrent neural network (RNN) (自 R2025a 起) |
 | GRU Projected Layer | Gated recurrent unit (GRU) projected layer for recurrent neural network (RNN) (自 R2025a 起) |
 | LSTM Layer | Long short-term memory (LSTM) layer for recurrent neural network (RNN) (自 R2024b 起) |
 | LSTM Projected Layer | Long short-term memory (LSTM) projected layer for recurrent neural network (RNN) (自 R2024b 起) |
 | Dropout Layer | Dropout layer (自 R2024b 起) |
 | Identity Layer | Identity layer (自 R2026a 起) |
 | Permute Layer | Permute layer (自 R2026b 起) |
 | Reshape Layer | Reshape layer (自 R2026b 起) |
 | Scaling Layer | Scaling layer (自 R2026a 起) |
 | Spatial Dropout Layer | Spatial dropout layer (自 R2026a 起) |
主题
深度学习层模块
- List of Deep Learning Layer Blocks and Subsystems
Discover all the deep learning layer blocks and subsystems in Simulink. - Implement Unsupported Deep Learning Layer Blocks
Learn how to implement unsupported deep learning layer blocks.
图像
- 使用 SqueezeNet 对 Simulink 中的图像进行分类
此示例说明如何使用 Classify 模块对 Simulink® 中的图像进行分类。 - Acceleration for Simulink Deep Learning Models
Improve simulation speed with accelerator and rapid accelerator modes. - Lane and Vehicle Detection in Simulink Using Deep Learning
This example shows how to use deep convolutional neural networks inside a Simulink® model to perform lane and vehicle detection. - Classify ECG Signals in Simulink Using Deep Learning
This example shows how to use wavelet transforms and a deep learning network within a Simulink (R) model to classify ECG signals. - Classify Images in Simulink with Imported TensorFlow Network
Import a pretrained TensorFlow™ network usingimportTensorFlowNetwork, and then use the Predict block for image classification in Simulink.
序列
- 在 Simulink 中预测和更新网络状态
此示例说明如何在 Simulink® 中使用Stateful Predict模块预测经过训练的循环神经网络的响应。此示例使用预训练的长短期记忆 (LSTM) 网络。 - 在 Simulink 中分类和更新网络状态
此示例说明如何在 Simulink® 中使用Stateful Classify模块对经过训练的循环神经网络进行数据分类。此示例使用预训练的长短期记忆 (LSTM) 网络。 - Speech Command Recognition in Simulink
Detect the presence of speech commands in audio using a Simulink model. - Time Series Prediction in Simulink Using Deep Learning Network
This example shows how to use an LSTM deep learning network inside a Simulink® model to predict the remaining useful life (RUL) of an engine. - Simulate Calorie Burn Using Neural Network in Simulink
This example shows how to include a simple fully connected neural network in a Simulink® model that predicts calorie burn when given five time steps of sensor readings from a smart watch. (自 R2024b 起) - 使用深度学习进行电池荷电状态估计
定义需求、准备数据、训练深度学习网络、验证稳健性、将网络集成到 Simulink 中以及部署模型。 (自 R2024b 起) - Physical System Modeling Using LSTM Network in Simulink
This example shows how to create a reduced order model (ROM) that acts as a virtual sensor in a Simulink® model using a long short-term memory (LSTM) neural network. - Classify Motor Faults Using Deep Learning
This example shows how to train a deep learning model to classify faults in a permanent magnet synchronous motor (PMSM) using simulated data across various revolutions per minute (RPM). (自 R2025a 起) - Improve Performance of Deep Learning Simulations in Simulink
This example shows how to use code generation to improve the performance of deep learning simulations in Simulink®.
强化学习
- Control Water Level in a Tank Using a DDPG Agent (Reinforcement Learning Toolbox)
Use DDPG agent to control water level in a tank modeled in Simulink. - Train DDPG Agent for Adaptive Cruise Control (Reinforcement Learning Toolbox)
Train a DDPG agent for an adaptive cruise control application. - Train DQN Agent for Lane Keeping Assist Using Parallel Computing (Reinforcement Learning Toolbox)
Train a DQN agent for an automated driving application using parallel computing. - Train DDPG Agent for Path-Following Control (Reinforcement Learning Toolbox)
Train a DDPG agent for lane following control.
Python 协同执行
- Classify Images Using TensorFlow Model Predict Block
Classify images using TensorFlow Model Predict block. - Classify Images Using ONNX Model Predict Block
Classify images using ONNX Model Predict block. - Classify Images Using PyTorch Model Predict Block
Classify images using PyTorch Model Predict block. - Predict Responses Using TensorFlow Model Predict Block
Predict Responses Using TensorFlow Model Predict block. - Predict Responses Using ONNX Model Predict Block
Predict Responses Using ONNX Model Predict block. - Predict Responses Using PyTorch Model Predict Block
Predict Responses Using PyTorch Model Predict block. - Predict Responses Using Custom Python Model in Simulink (Statistics and Machine Learning Toolbox)
This example shows how to use the Custom Python Model Predict (Statistics and Machine Learning Toolbox) block for prediction in Simulink®.
代码生成
- 从 Simulink 应用生成深度学习代码
生成用于在桌面或嵌入式目标上部署的 C/C++ 和 GPU 代码 - Export Network to FMU
This example shows how to export a trained network as a Functional Mock-up Unit (FMU). (自 R2023b 起)
精选示例
Lane and Vehicle Detection in Simulink Using Deep Learning
Use deep convolutional neural networks inside a Simulink® model to perform lane and vehicle detection. This example takes the frames from a traffic video as an input, outputs two lane boundaries that correspond to the left and right lanes of the ego vehicle, and detects vehicles in the frame.
Classify ECG Signals in Simulink Using Deep Learning
Use wavelet transforms and a deep learning network within a Simulink (R) model to classify ECG signals. This example uses the pretrained convolutional neural network from the Classify Time Series Using Wavelet Analysis and Deep Learning example of the Wavelet Toolbox™ to classify ECG signals based on images from the CWT of the time series data. For information on training, see Classify Time Series Using Wavelet Analysis and Deep Learning (Wavelet Toolbox).
Time Series Prediction in Simulink Using Deep Learning Network
Use an LSTM deep learning network inside a Simulink® model to predict the remaining useful life (RUL) of an engine. You include the network inside the Simulink model by using a Stateful Predict block, which predicts the RUL at every simulation step.
使用前馈神经网络在 Simulink 中估计电池荷电状态
此示例说明如何在 Simulink® 模型内使用前馈深度学习网络来预测电池荷电状态 (SOC)。使用 Predict 模块将该网络包含在 Simulink 模型中，该模块在每个仿真步骤预测 SOC。
Profile Deep Learning Network for Deployment
Profile a deep learning network in Simulink. Profiling your network before training is a good way of checking that the network meets your requirements, before spending time on training. This example shows how to check:
Generate AUTOSAR C Code for Deep Learning Model
Integrate a trained deep neural network in a Simulink model mapped to an AUTOSAR Classic software component.
