# Matlab-AE_MVTS
Generic Deep Autoencoder for Time-Series
This toolbox enables the simple implementation of different deep autoencoder. The primary focus is on multi-channel time-series analysis.
Each autoencoder consists of two, possibly deep, neural networks - the encoder and the decoder.
The following layers can be combined and stacked to form the neural networks which form the encoder and decoder:
- LSTM (Long-short term memory layers),
- Bi-LSTM (Bi-directional long-short term memory layers),
- FC with ReLU (Fully connected layers followed by a rectified linear unit).
There are two types of autoencoders available:
- AE (autoencoder)
- VAE (variational autoencoder)
The autoencoders can be easily parametrized using hyperparameters.
This toolbox is also mirrored to [MatlabFileExchange](https://de.mathworks.com/matlabcentral/fileexchange/111110-generic-deep-autoencoder-for-time-series)
-------------------------------------------------------------------------------------------
# Publishing information
This code can also be considered as supplemental Material to the Paper:
"Hybrid Machine Learning for Anomaly Detection in Industrial Time-Series
Measurement Data"
by: Anika Terbuch, Paul O'Leary and Peter Auer
Mai 2022
Link to paper: https://ieeexplore.ieee.org/document/9806663
# Example Plot
This plot shows the input and output of a multivariate time-series to a variational autoencoder built with this framework
![AutoencoderDeep](https://github.com/anikaTerbuch/Matlab-AE_MVTS/assets/58983404/ca537240-d2e0-4c59-b1a0-7abe78e1ce5b)
-------------------------------------------------------------------------------------------
# Class descriptions
The toolbox includes the following two classes:
1) AutoencoderDeep: generic framework for creating autoencoders suitable for multivariate time-series data
2) HyperparametersAED: the class used to parametrize objects of the class AutoencoderDeep.
-------------------------------------------------------------------------------------------
For easier and more convinient use the classes are wrapped into functions:
The following functions are provided:
-trainAutoencoderDeep: trains an autoencoder using the data provided.
-predictAutoencoderDeep: returns an prediction of the autoencoder passed to the function on the data provided to the function.
## AutoencoderDeep
ad 1)
The class AutoencoderDeep contains the following functions:
- decodingAED: decodes the latent representation back into the original domain.
- ELBOloss: calculates the Evidence Lower Bound (ELBO) of given data, given
latent encoding and the weighting factor of the two terms of the loss function.
- encodingAED: performs an encoding into the latent space of a trained
AutoencoderDeep on the samples provided to the function.
- gradientsRecErr: loss function of the AE - minimize the reconstruction loss
evaluates the encoder and decoder on the passed data and calculates
the gradients of the learnable parameters of the network.
- gradientsRecErrAndKL: loss function of the VAE - maximize the evidence lower bound
evaluates the encoder and decoder on the passed data and calculates
the gradients of the learnables of the network.
- layerArray2dlnetwork: converts a layer array to a dl-network.
- reconstructionAED: performs the reoncustriction (encoding followed by decoding) on the
given data with a trained AutoencodeDeep.
- reconstructionErrorPerSampleAEDvariableLength: given the reconstructed signal and the
real signal the reconstruction error (1-
norm) normalized by the varying length
of the time series is calculated.
- samplingVAE: performs the encoding into the latent space followed by the
reparametrization trick used in the calculation of latent
representations when using VAEs.
-setUpAndTrainAED: creates the encoder and decoder networks and trains them accoding
to the properties specified in the hyperparameter struct.
-setUpDecoderAED: sets up a neural network that forms the decoder of the network with the
layer-types and number of neurons specified in the hyperparameter
struct.
-setUpEncoderAED: sets up a neural network that forms the encoder of the network with
the layer-types and number of neurons specified in the
hyperparameter struct.
-setUpEncoderDecoderAED: function calls to functions used for setting up the encoder and
decoder networks.
-squaredReconstructionErrorPerSampleAEDvariableLength: given the reconstructed signal and
the real signal the reconstruction
error (2-norm) normalized by the
varying length of the time series is
calculated.
-trainingLoopAED: training loop for training the encoder and decoder simultaneously
based on ADAM, training can be performed using CPU-units or GPU-
units.
-varSeqLen2dlarray: creates mini-batches of dl-arrays with minimum resampling.
## HyperparametersAED
ad 2)
The class HyperparametersAED contains the following functions:
-setDefaultHyperparametersAED: initializes the parameters to default values at instantiation of an object
-setHyperparametersAED: can be used to change the hyperparameters of the AutoencoderDeep by
passing the hyperaramter names to change and their values
('name value pairs') to this function.
-------------------------------------------------------------------------------------------
This derived class is tailored to handle data shown in
https://de.mathworks.com/help/deeplearning/ug/time-series-anomaly-detection-using-deep-learning.html
https://de.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html and
https://de.mathworks.com/help/deeplearning/ug/sequence-to-one-regression-using-deep-learning.html?searchHighlight=WaveformData&s_tid=srchtitle_WaveformData_3
The following impelementation was created using Matlab 2022a and 2022b.
没有合适的资源?快使用搜索试试~ 我知道了~
时间序列的通用深度自动编码器matlab代码.zip
共33个文件
m:22个
mlx:4个
txt:2个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
5星 · 超过95%的资源 1 下载量 18 浏览量
2024-04-26
23:04:32
上传
评论
收藏 1.54MB ZIP 举报
温馨提示
1.版本:matlab2014/2019a/2021a 2.附赠案例数据可直接运行matlab程序。 3.代码特点:参数化编程、参数可方便更改、代码编程思路清晰、注释明细。 4.适用对象:计算机,电子信息工程、数学等专业的大学生课程设计、期末大作业和毕业设计。
资源推荐
资源详情
资源评论
收起资源包目录
时间序列的通用深度自动编码器matlab代码.zip (33个子文件)
时间序列的通用深度自动编码器matlab代码
mcode
@HyperparametersAED
setDefaultHyperparametersAED.m 4KB
HyperparametersAED.m 1KB
setHyperparametersAED.m 10KB
wrapper_functions
trainAutoencoderDeep.m 5KB
predictAutoencoderDeep.m 5KB
@AutoencoderDeep
varSeqLen2dlarray.m 4KB
decodingAED.m 2KB
setUpEncoderAED.m 4KB
squaredReconstructionErrorPerSampleAEDvariableLength.m 3KB
encodingAED.m 2KB
setUpEncoderDecoderAED.m 781B
AutoencoderDeep.m 6KB
gradientsRecErr.m 3KB
layerArray2dlnetwork.m 1KB
samplingVAE.m 3KB
reconstructionAED.m 3KB
setUpAndTrainAED.m 2KB
gradientsRecErrAndKL.m 2KB
ELBOloss.m 2KB
trainingLoopAED.m 5KB
reconstructionErrorPerSampleAEDvariableLength.m 3KB
setUpDecoderAED.m 4KB
AutoencoderDeep.png 726KB
Introduction.mlx 88KB
Preamble.mlx 5KB
Example
Example1.pdf 492KB
Example1.mlx 104KB
Example2customHyperparameterSetting.pdf 161KB
Example2customHyperparameterSetting.mlx 7KB
CITATION.cff 1KB
mathworks_text.txt 8KB
license.txt 1KB
README.md 7KB
共 33 条
- 1
资源评论
- weixin_745586342024-05-16资源内容总结的很到位,内容详实,很受用,学到了~
Matlab科研辅导帮
- 粉丝: 1w+
- 资源: 7491
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 基于matlab的考虑设备动作损耗的配电网分布式电压无功优化源码.zip
- 互联网网络科技类网站模板整站打包下载.zip
- 电脑操作系统软件下载网站模板打包整站下载.zip
- 创维e910 mt8685刷机包
- 网站整站模板竞技游戏类网站模板打包下载.zip
- 14 Day of the Dragon King - Mary Pope Osborne.png
- 【Redis开发】资源Redis-v6.0.20.zip
- 《STM32单片机+DHT11温湿度++光敏电阻+蜂鸣器+OLED屏幕》源代码
- (完整版)银行计算机储蓄系统需求分析.doc
- 图像超分辨率FSRCNN的最优SSIM和PSNR的模型权重文件(x2、x3、x4)
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功