Deep Learning and Time Series
This document shows a list of bibliographical references on DeepLearning and Time Series, organized by type and year. I add some additional notes on each reference.
Table of contents
Deef Belief Network with Restricted Boltzmann Machine
Journal
2017
-
Summary: The paper proposes deep neural network (DNN)-based load forecasting models and apply them to a demand side empirical load database. DNNs are trained in two different ways: a pre-training restricted Boltzmann machine and using ReLu without pre-training.
Notes:
- Model 1 train -> greedy layer-wise manner
- Model 1 Fine-tuning connection weights -> Back-propagation
- Model 2 train -> ReLu
- Model Sizes -> trial and error
-
Summary: In this paper a Deep Belief Network (DBN) including two restricted Boltzmann machines (RBMs) was used to model load demand series.
2016
-
Summary: This paper presents a hybrid prediction method using DBNs (deep Belief Network) and ARIMA.
2014
-
Summary: This papers proposes a method for time series prediction using deep belief nets (DBN) (with 3-layer of RBMs to capture the feature of input space of time series data).
Notes:
- Mode Train -> greedy layer-wise manner
- Fine-tuning connection weights -> Back-propagation
- Mode sizes and learning rates -> PSO
Conference
2017
-
Summary: The paper looks into the drought prediction problem using deep learning algorithms. They propose a Deep Belief Network consisting of two Restricted Boltzmann Machines. The study compares the efficiency of the proposed model to that of traditional approaches such as Multilayer Perceptron (MLP) and Support Vector Regression (SVR) for predicting the different time scale drought conditions.
Notes:
- Model train -> unsupervised learning
- Model Fine-tuning connection weights -> Back-propagation
2016
-
Summary: This paper introduces a reinforcement learning method named stochastic gradient ascent (SGA) to the DBN with RBMs instead conventional BackPropagation to predict a benchmark named CATS data.
-
Summary: The paper proposes a deep learning approach, which hybridizes a deep belief networks (DBNs) and a nonlinear kernel-based parallel evolutionary SVM (ESVM), to predict evolution states of complex systems in a classification manner.
Notes:
- Top layer -> SVM
- Fine-tuning connection weights -> Back-propagation
-
Summary: In this paper, a deep learning method, the Deep Belief Network (DBN) model, is proposed for short-term traffic speed information prediction.
Notes:
- Model train -> greedy layer-wise manner
- Fine-tuning connection weights -> Back-propagation
- Model Sizes -> several ccombinations
2015
-
Summary: This paper proposes an ensemble of deep learning belief networks (DBN) for regression and time series forecasting on electricity load demand datasets. Another contribution is to aggregate the outputs from various DBNs by a support vector regression (SVR) model.
Notes:
- Top layer -> support vector regression (SVR)
Long short-term memory
Journal
2018
-
Summary: In this paper, they proposed a nonlinear-learning ensemble of deep learning time series prediction based on LSTMs, SVRM and EO (extremal optimization algorithm) that is applied on two case studies data collected from two wind farm.
Notes:
- Top layer -> support vector regression (SVR)
2017
-
Summary: This paper pses a traffic forecast model based on long short-term memory (LSTM) network, that considers temporal-spatial correlation in traffic system via a two-dimensional network which is composed of many memory units.
Conference
2018
-
Summary: This work proposes a time series forecasting model using a specific type of recursive neural networks, LSTM, for operation scheduling and sequencing in a virtual shop floor environment.
-
Summary: This paper uses two approaches to model gas consumption, Generalized Additive Models (GAM) and Long Short-Term Memory (LSTM). They compare the performances of GAM and LSTM with other state-of-the-art forecasting approachesand they show that LSTM outperforms GAM and other existing approaches.
-
Summary: The paper proposes a Bayesian optimised recurrent neural network (RNN) and a Long Short Term Memory (LSTM) network to ascertain with what accuracy the direction of Bitcoin price in USD can be predicted. The paper also implemented an ARIMA model for time series forecasting as a comparison to the deep learning models. The LSTM achieves the highest classification accuracy.
-
Summary: The paper proposes a deep learning method (long short-term memory (LSTM)) to estimate the remaining useful life of aero-propulsion engines. The proposed method is compared with the following methods: multi-layer perceptron (MLP), support vector regression (SVR), relevance vector regression (RVR) and convolutional neural network (CNN).
2017
-
Summary: The paper evalues a series of long short-term memory neural networks with deep neural layers (LSTM-DNN) using 16 settings of hyperparameters and investigates their performance on a 90-day travel time dataset. Then, the LSTM is tested along with linear models such as linear regression, Ridge and Lasso regression, ARIMA and DNN models under 10 sets of sliding windows and predicting horizons via the same dataset.
-
Summary: The paper proposes a LSTM deep learning methodology for predicting future price movements from large-scale high-frequency time-series data on Limit Order Books.
2016
-
Summary: The paper proposes a model incorporating a sequence-to-sequence model that consists two LSTMs, one encoder and one decoder. The encoder LSTM accepts input time series, extracts information and based on which the decoder LSTM constructs fixed length sequences that can be regarded as discriminatory features. The paper also introduces the attention mechanism.
-
Summary: This paper proposes an application of deep learning models, Paragraph Vector, and Long Short-Term Memory (LSTM), to financial time series forecasting.
-
Summary: This paper explores a deep learning model, the LSTM neural network model, for travel time prediction. By employing the travel time data provided by Highways England dataset, the paper construct 66 series prediction LSTM neural networks.
-
Summary: This paper presents an energy load forecasting methodology based on Deep Neural Networks (Long Short Term Memory (LSTM) algorithms). The presented work investigates two LSTM based architectures: 1) standard LSTM and 2) LSTM-based Sequence to Sequence (S2S) architecture. Both methods were implemented on a benchmark data set of electricity consumption data from one residential customer.
Notes:
- Model train -> Backpropagation
-
Summary: This paper explores the application of Long Short-Term Memory Networks (LSTMs) in short-term traffic flow prediction.
-
Summary: This paper compares conventional machine learning methods with modern neural network architectures to better forecast analgesic responses. The paper applies the LSTM to predict what the next measured pain score will be after administration of an analgesic drug, and compared the results with simpler techniques.
-
Summary: This paper uses a stacked long short-term memory model to learn and predict the patterns of traffic conditions (that are collected from online open web based map services).
Notes:
- Model sizes and learning rates -> several ccombinations
Auto-Encoders
Journal
2017
-
Summary: This paper proposes a stacked autoencoder Levenberg–Marquardt model to improve forecasting accuracy. It is applied to real-world data collected from the M6 freeway in the U.K.
Notes:
- Fine-tuning connection weights -> Levenberg-Marquadt
2016
-
Summary: This paper proposed a novel spatiotemporal deep learning (STDL)-based air quality prediction method that inherently considers spatial and temporal correlations. A stacked autoencoder (SAE) model is used to extract inherent air quality features.
Notes:
- Model Train -> greedy layer-wise manner
- Top layer -> logistic regression
- Fine-tuning connection weights -> Back-propagation
- Model sizes -> several ccombinations
Conference
2016
-
Summary: The paper introduces an architecture based on Deep Learning for the prediction of the accumulated daily precipitation for the next day. More specifically, it includes an autoencoder for reducing and capturing non-linear relationships between attributes, and a multilayer perceptron for the prediction task.
Notes:
- Top layer -> multilayer perceptron
- Model sizes and learning rates -> several combinations
2015
-
Summary: This paper compares a deep learning network (Stacked Denoising Auto-Encoders (SDAE)) against a standard neural network for predicting air temperature from historical pressure, humidity, and temperature data gathered from meteorological sensors in Northwestern Nevada. In addition, predicting air temperature from historical air temperature data alone can be improved by employing related weather variables like barometric pressure, humidity and wind speed data in the training process.
Notes:
- Top layer -> feed-forward neural network
2013
-
Summary: This paper presents a study of deep learning techniques (Stacked Denoising Auto-Encoders (SDAEs)) applied to time-series forecasting in a real indoor temperature forecasting task.
Combination of the above
Journal
2018
-
Summary: The paper proposes a framework for forecasting electricity prices. They use four different deep learning models for predicting electricity prices. In addition, they also consider that an extensive benchmark is still missing. To tackle that, they compare and analyze the accuracy of 27 common approaches for electricity price forecasting. Based on the benchmark results, they show how the proposed deep learning models outperform the state-of-the-art methods and obtain results that are statistically significant.
2017
-
Summary: The paper investigates the performance of deep belief network (DBN) and long short-term memory (LSTM) to conduct short-term traffic speed prediction with the consideration of rainfall impact as a non-traffic input. To validate the performance, the traffic detector data from an arterial in Beijing are utilised for model training and testing. The experiment results indicate that deep learning models have better prediction accuracy over other existing models. Furthermore, the LSTM can outperform the DBN to capture the time-series characteristics of traffic speed data.
Conference
2017
-
Summary: The paper proposes a system to predict monthly energy consumption using deep learning techniques. Three deep learning models were studied: Deep Fully Connected, Convolutional and Long Short-Term Memory Neural Networks.
2016
-
Summary: This work aims to investigate the use of some of deep learning architectures (deep belief networks and aunto-encoders) in predicting the hourly average speed of winds in the Northeastern region of Brazil.
Notes:
- Model Train -> greedy layer-wise manner
- Fine-tuning connection weights -> Levenberg-Marquadt
- Model sizes -> several combinations
-
Summary: This paper introduces different Deep Learning and Artificial Neural Network algorithms, such as Deep Belief Networks, AutoEncoder, and LSTM in the field of renewable energy power forecasting of 21 solar power plants.
Others
2018
-
Summary: This paper presents a novel deep learning architecture called multivariate convolutional neural network for time series classification, in which the proposed architecture considers multivariate and lag-features characteristics.
Type: Convolutional neural network
2017
-
Summary: This paper proposes a convolutional neural network (CNN) framework for time series classification. Two groups of experiments are conducted on simulated data sets and eight groups of experiments are conducted on real-world data sets from different application domains.
Type: Convolutional neural network
-
Summary: This paper exploits the applicability of and compares the performance of the Feed-forward Deep Neural Network (FF-DNN) and Recurrent Deep Neural Network (R-DNN) models on the basis of accuracy and computational performance in the context of time-wise short term forecast of electricity load. The herein proposed method is evaluated over real datasets gathered in a period of 4 years and provides forecasts on the basis of days and weeks ahead.
Type: Recurrent neural network
-
Summary: The study proposes a deep denoising recurrent temporal restricted Boltzmann machine network for long-term prediction of time series.
Notes:
- Model train -> layer by layer
- Model Fine-tuning connection weights -> Back-propagation
Type: Recurrent restricted Boltzmann machine
2016
-
Summary: The paper presents a deep convolutional factor analyser (DCFA) for multivariate time series modeling. The network is constructed in a way that bottom layer nodes are independent. Through a process of up-sampling and convolution, higher layer nodes gain more temporal dependency.
Type: Convolutional neural network
-
Summary: The present study uses three years' worth of point-of-sale (POS) data from a retail store to construct a sales prediction model that, given the sales of a particular day, predicts the changes in sales on the following day.
Type: Not specified
-
Summary: This work reports on employing the deep learning artificial intelligence techniques to predict the energy consumption and power generation together with the weather forecasting numerical simulation. An optimization tool platform using Boltzmann machine algorithm for NMIP problem is also proposed for better computing scalable decentralized renewable energy system.
Type: a novel optimization tool platform using Boltzmann machine algorithm for NMIP
2015
-
Summary: This study investigates deep learning techniques for weather forecasting. In particular, this study will compare prediction performance of Recurrence Neural Network (RNN), Conditional Restricted Boltzmann Machine (CRBM), and Convolutional Network (CN) models. Those models are tested using weather dataset which are collected from a number of weather stations.
Type: Recurrent neural network, convolutional neural network
-
Summary: The aim of this paper is to present deep neural network architectures and algorithms and explore their use in time series prediction. Shallow and deep neural networks coupled with two input variable selection algorithms are compared on a ultra-short-term wind prediction task.
Type: MultiLayer Perceptron.
2014
-
Summary: This paper explores the feature learning techniques to improve the performance of traditional feature-based approaches. Specifically, the paper proposes a deep learning framework for multivariate time series classification in two groups of experiments on real-world data sets from different application domains.
Type: Multi-Channels Deep Convolution Neural Networks
-
Dmitry Vengertsev (2014). Deep Learning Architecture for Univariate Time Series Forecasting
Summary: This paper overviews the particular challenges present in applying Conditional Restricted Boltzmann Machines (CRBM) to univariate time-series forecasting and provides a comparison to common algorithms used for time-series prediction. As a benchmark dataset for testing and comparison of forecasting algorithms, the paper selected M3 competition dataset.
Type: Conditional Restricted Boltzmann Machines (CRBM)
Reviews
2017
-
John Cristian Borges Gamboa 2017. Deep Learning for Time-Series Analysis
Summary: This paper presents review of the main Deep Learning techniques, and some applications on Time-Series analysis are summaried.
2014
-
Summary: This paper gives a review of the recent developments in deep learning and unsupervised feature learning for time-series problems.