使用 NARX 网络和时延网络进行建模和预测
使用动态神经网络(包括带反馈的网络)求解时间序列问题
App
神经网络时间序列 | 使用动态神经网络求解非线性时间序列问题 |
函数
timedelaynet | Time delay neural network |
narxnet | Nonlinear autoregressive neural network with external input |
narnet | Nonlinear autoregressive neural network |
layrecnet | Layer recurrent neural network |
distdelaynet | Distributed delay network |
train | Train shallow neural network |
gensim | Generate Simulink block for shallow neural network simulation |
adddelay | Add delay to neural network response |
removedelay | Remove delay to neural network’s response |
closeloop | Convert neural network open-loop feedback to closed loop |
openloop | Convert neural network closed-loop feedback to open loop |
ploterrhist | Plot error histogram |
plotinerrcorr | Plot input to error time-series cross-correlation |
plotregression | 绘制线性回归图 |
plotresponse | Plot dynamic network time series response |
ploterrcorr | Plot autocorrelation of error time series |
genFunction | Generate MATLAB function for simulating shallow neural network |
示例和操作指南
基本设计
- 浅层神经网络时间序列预测和建模
使用神经网络时间序列和命令行函数进行时间序列预测。 - Design Time Series Time-Delay Neural Networks
Learn to design focused time-delay neural network (FTDNN) for time-series prediction. - 多步神经网络预测
了解多步神经网络预测。 - 设计时间序列 NARX 反馈神经网络
创建和训练外因输入非线性自回归网络 (NARX)。 - Design Layer-Recurrent Neural Networks
Create and train a dynamic network that is a Layer-Recurrent Network (LRN). - Deploy Shallow Neural Network Functions
Simulate and deploy trained shallow neural networks using MATLAB® tools. - Deploy Training of Shallow Neural Networks
Learn how to deploy training of shallow neural networks. - 磁悬浮建模
此示例说明 NARX(具有外部输入的非线性自回归)神经网络如何对磁悬浮动态系统建模。
训练可扩展性和效率
- 使用并行和 GPU 计算的浅层神经网络
使用并行和分布式计算,可以加快神经网络训练和仿真以及处理大量数据的速度。 - Automatically Save Checkpoints During Neural Network Training
Save intermediate results to protect the value of long training runs. - 优化神经网络训练速度和内存
使神经网络训练更加高效。
最优解
- 选择神经网络输入输出处理函数
对输入和目标进行预处理,以提高训练效率。 - 配置浅层神经网络输入和输出
了解如何在训练前使用configure
函数手动配置网络。 - 划分数据以实现最优神经网络训练
使用函数将数据分为训练集、验证集和测试集。 - 选择多层神经网络训练函数
不同问题类型的训练算法比较。 - 提高浅层神经网络泛化能力,避免过拟合
了解提高泛化能力和防止过拟合的方法。 - Train Neural Networks with Error Weights
Learn how to use error weighting when training neural networks. - Normalize Errors of Multiple Outputs
Learn how to fit output elements with different ranges of values.
概念
- How Dynamic Neural Networks Work
Learn how feedforward and recurrent networks work.
- Multiple Sequences with Dynamic Neural Networks
Manage time-series data that is available in several short sequences.
- Neural Network Time-Series Utilities
Learn how to use utility functions to manipulate neural network data.
- 浅层神经网络的样本数据集
试验浅层神经网络时要使用的样本数据集列表。
- 神经网络对象属性
了解定义网络基本特征的属性。
- Neural Network Subobject Properties
Learn properties that define network details such as inputs, layers, outputs, targets, biases, and weights.