scholarly journals STL-ATTLSTM: Vegetable Price Forecasting Using STL and Attention Mechanism-Based LSTM

Agriculture ◽  
2020 ◽  
Vol 10 (12) ◽  
pp. 612
Author(s):  
Helin Yin ◽  
Dong Jin ◽  
Yeong Hyeon Gu ◽  
Chang Jin Park ◽  
Sang Keun Han ◽  
...  
Keyword(s):  
Time Series ◽  
Prediction Accuracy ◽  
Crop Production ◽  
Time Series Data ◽  
Short Term Memory ◽  
Attention Mechanism ◽  
Series Data ◽  
Percentage Error ◽  
Hot Pepper ◽  
Types Of Information

It is difficult to forecast vegetable prices because they are affected by numerous factors, such as weather and crop production, and the time-series data have strong non-linear and non-stationary characteristics. To address these issues, we propose the STL-ATTLSTM (STL-Attention-based LSTM) model, which integrates the seasonal trend decomposition using the Loess (STL) preprocessing method and attention mechanism based on long short-term memory (LSTM). The proposed STL-ATTLSTM forecasts monthly vegetable prices using various types of information, such as vegetable prices, weather information of the main production areas, and market trading volumes. The STL method decomposes time-series vegetable price data into trend, seasonality, and remainder components. It uses the remainder component by removing the trend and seasonality components. In the model training process, attention weights are assigned to all input variables; thus, the model’s prediction performance is improved by focusing on the variables that affect the prediction results. The proposed STL-ATTLSTM was applied to five crops, namely cabbage, radish, onion, hot pepper, and garlic, and its performance was compared to three benchmark models (i.e., LSTM, attention LSTM, and STL-LSTM). The performance results show that the LSTM model combined with the STL method (STL-LSTM) achieved a 12% higher prediction accuracy than the attention LSTM model that did not use the STL method and solved the prediction lag arising from high seasonality. The attention LSTM model improved the prediction accuracy by approximately 4% to 5% compared to the LSTM model. The STL-ATTLSTM model achieved the best performance, with an average root mean square error (RMSE) of 380, and an average mean absolute percentage error (MAPE) of 7%.

scholarly journals Time-Series Classification Based on Fusion Features of Sequence and Visualization

2020 ◽  
Vol 10 (12) ◽  
pp. 4124
Author(s):  
Baoquan Wang ◽  
Tonghai Jiang ◽  
Xi Zhou ◽  
Bo Ma ◽  
Fan Zhao ◽  
...  
Keyword(s):  
Time Series ◽  
Human Brain ◽  
Time Series Data ◽  
Short Term Memory ◽  
Computational Cost ◽  
Open Data ◽  
Attention Mechanism ◽  
Series Data ◽  
Sequence Features ◽  
Fusion Features

For the task of time-series data classification (TSC), some methods directly classify raw time-series (TS) data. However, certain sequence features are not evident in the time domain and the human brain can extract visual features based on visualization to classify data. Therefore, some researchers have converted TS data to image data and used image processing methods for TSC. While human perceptionconsists of a combination of human senses from different aspects, existing methods only use sequence features or visualization features. Therefore, this paper proposes a framework for TSC based on fusion features (TSC-FF) of sequence features extracted from raw TS and visualization features extracted from Area Graphs converted from TS. Deep learning methods have been proven to be useful tools for automatically learning features from data; therefore, we use long short-term memory with an attention mechanism (LSTM-A) to learn sequence features and a convolutional neural network with an attention mechanism (CNN-A) for visualization features, in order to imitate the human brain. In addition, we use the simplest visualization method of Area Graph for visualization features extraction, avoiding loss of information and additional computational cost. This article aims to prove that using deep neural networks to learn features from different aspects and fusing them can replace complex, artificially constructed features, as well as remove the bias due to manually designed features, in order to avoid the limitations of domain knowledge. Experiments on several open data sets show that the framework achieves promising results, compared with other methods.

scholarly journals Implementation of Long Short-Term Memory and Gated Recurrent Units on Grouped Time-Series Data to Predict Stock Prices Accurately

2021 ◽  
Author(s):  
Armin Lawi ◽  
Hendra Mesra ◽  
Supri Amir
Keyword(s):  
Time Series ◽  
Stock Prices ◽  
Stock Price ◽  
Time Series Data ◽  
Short Term Memory ◽  
Series Data ◽  
Percentage Error ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory

Abstract Stocks are an attractive investment option since they can generate large profits compared to other businesses. The movement of stock price patterns on the stock market is very dynamic; thus it requires accurate data modeling to forecast stock prices with a low error rate. Forecasting models using Deep Learning are believed to be able to accurately predict stock price movements using time-series data, especially the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) algorithms. However, several previous implementation studies have not been able to obtain convincing accuracy results. This paper proposes the implementation of the forecasting method by classifying the movement of time-series data on company stock prices into three groups using LSTM and GRU. The accuracy of the built model is evaluated using loss functions of Rooted Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The results showed that the performance evaluation of both architectures is accurate in which GRU is always superior to LSTM. The highest validation for GRU was 98.73% (RMSE) and 98.54% (MAPE), while the LSTM validation was 98.26% (RMSE) and 97.71% (MAPE).

scholarly journals A Data-Driven Long Time-Series Electrical Line Trip Fault Prediction Method Using an Improved Stacked-Informer Network

Sensors ◽  
2021 ◽  
Vol 21 (13) ◽  
pp. 4466
Author(s):  
Li Guo ◽  
Runze Li ◽  
Bin Jiang
Keyword(s):  
Time Series ◽  
Prediction Accuracy ◽  
Time Series Data ◽  
Short Term Memory ◽  
Fault Prediction ◽  
Data Driven ◽  
Series Data ◽  
Sequence Prediction ◽  
Long Time Series ◽  
Long Time

The monitoring of electrical equipment and power grid systems is very essential and important for power transmission and distribution. It has great significances for predicting faults based on monitoring a long sequence in advance, so as to ensure the safe operation of the power system. Many studies such as recurrent neural network (RNN) and long short-term memory (LSTM) network have shown an outstanding ability in increasing the prediction accuracy. However, there still exist some limitations preventing those methods from predicting long time-series sequences in real-world applications. To address these issues, a data-driven method using an improved stacked-Informer network is proposed, and it is used for electrical line trip faults sequence prediction in this paper. This method constructs a stacked-Informer network to extract underlying features of long sequence time-series data well, and combines the gradient centralized (GC) technology with the optimizer to replace the previously used Adam optimizer in the original Informer network. It has a superior generalization ability and faster training efficiency. Data sequences used for the experimental validation are collected from the wind and solar hybrid substation located in Zhangjiakou city, China. The experimental results and concrete analysis prove that the presented method can improve fault sequence prediction accuracy and achieve fast training in real scenarios.

scholarly journals Multivariate Temporal Convolutional Network: A Deep Neural Networks Approach for Multivariate Time Series Forecasting

Electronics ◽  
2019 ◽  
Vol 8 (8) ◽  
pp. 876 ◽  
Author(s):  
Renzhuo Wan ◽  
Shuping Mei ◽  
Jun Wang ◽  
Min Liu ◽  
Fan Yang
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Prediction Accuracy ◽  
Time Series Data ◽  
Short Term Memory ◽  
Time Series Prediction ◽  
Multivariate Time Series ◽  
Series Data ◽  
Asymmetric Structure ◽  
Convolutional Network

Multivariable time series prediction has been widely studied in power energy, aerology, meteorology, finance, transportation, etc. Traditional modeling methods have complex patterns and are inefficient to capture long-term multivariate dependencies of data for desired forecasting accuracy. To address such concerns, various deep learning models based on Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) methods are proposed. To improve the prediction accuracy and minimize the multivariate time series data dependence for aperiodic data, in this article, Beijing PM2.5 and ISO-NE Dataset are analyzed by a novel Multivariate Temporal Convolution Network (M-TCN) model. In this model, multi-variable time series prediction is constructed as a sequence-to-sequence scenario for non-periodic datasets. The multichannel residual blocks in parallel with asymmetric structure based on deep convolution neural network is proposed. The results are compared with rich competitive algorithms of long short term memory (LSTM), convolutional LSTM (ConvLSTM), Temporal Convolution Network (TCN) and Multivariate Attention LSTM-FCN (MALSTM-FCN), which indicate significant improvement of prediction accuracy, robust and generalization of our model.

scholarly journals Improved Prediction of COVID-19 Transmission and Mortality Using Google Search Trends for Symptoms in the United States

2021 ◽  
Author(s):  
Meshrif Alruily ◽  
Mohamed Ezz ◽  
Ayman Mohamed Mostafa ◽  
Nacim Yanes ◽  
Mostafa Abbas ◽  
...  
Keyword(s):  
Public Health ◽  
Time Series ◽  
Computational Models ◽  
Time Series Data ◽  
Short Term Memory ◽  
Emerging Infectious Diseases ◽  
The United States ◽  
Series Data ◽  
Percentage Error ◽  
Google Search

ABSTRACTAccurate forecasting of emerging infectious diseases can guide public health officials in making appropriate decisions related to the allocation of public health resources. Due to the exponential spread of the COVID-19 infection worldwide, several computational models for forecasting the transmission and mortality rates of COVID-19 have been proposed in the literature. To accelerate scientific and public health insights into the spread and impact of COVID-19, Google released the Google COVID-19 search trends symptoms open-access dataset. Our objective is to develop 7 and 14 -day-ahead forecasting models of COVID-19 transmission and mortality in the US using the Google search trends for COVID-19 related symptoms. Specifically, we propose a stacked long short-term memory (SLSTM) architecture for predicting COVID-19 confirmed and death cases using historical time series data combined with auxiliary time series data from the Google COVID-19 search trends symptoms dataset. Considering the SLSTM networks trained using historical data only as the base models, our base models for 7 and 14 -day-ahead forecasting of COVID cases had the mean absolute percentage error (MAPE) values of 6.6% and 8.8%, respectively. On the other side, our proposed models had improved MAPE values of 3.2% and 5.6%, respectively. For 7 and 14 -day-ahead forecasting of COVID-19 deaths, the MAPE values of the base models were 4.8% and 11.4%, while the improved MAPE values of our proposed models were 4.7% and 7.8%, respectively. We found that the Google search trends for “pneumonia,” “shortness of breath,” and “fever” are the most informative search trends for predicting COVID-19 transmission. We also found that the search trends for “hypoxia” and “fever” were the most informative trends for forecasting COVID-19 mortality.

scholarly journals Time–frequency time–space LSTM for robust classification of physiological signals

2021 ◽  
Vol 11 (1) ◽  
Author(s):  
Tuan D. Pham
Keyword(s):  
Time Series ◽  
Network Architecture ◽  
Time Series Data ◽  
Short Term Memory ◽  
Physiological Signals ◽  
Series Data ◽  
Time Frequency ◽  
Time Space ◽  
Deep Recurrent Neural Network

AbstractAutomated analysis of physiological time series is utilized for many clinical applications in medicine and life sciences. Long short-term memory (LSTM) is a deep recurrent neural network architecture used for classification of time-series data. Here time–frequency and time–space properties of time series are introduced as a robust tool for LSTM processing of long sequential data in physiology. Based on classification results obtained from two databases of sensor-induced physiological signals, the proposed approach has the potential for (1) achieving very high classification accuracy, (2) saving tremendous time for data learning, and (3) being cost-effective and user-comfortable for clinical trials by reducing multiple wearable sensors for data recording.

scholarly journals Comparison of Multi-Temporal PlanetScope Data with Landsat 8 and Sentinel-2 Data for Estimating Airborne LiDAR Derived Canopy Height in Temperate Forests

2020 ◽  
Vol 12 (11) ◽  
pp. 1876 ◽  
Author(s):  
Katsuto Shimizu ◽  
Tetsuji Ota ◽  
Nobuya Mizoue ◽  
Hideki Saito
Keyword(s):  
Time Series ◽  
Prediction Accuracy ◽  
Time Series Data ◽  
Airborne Lidar ◽  
Canopy Height ◽  
Forest Monitoring ◽  
Series Data ◽  
Landsat 8 ◽  
Multi Temporal ◽  
Sentinel 2

Developing accurate methods for estimating forest structures is essential for efficient forest management. The high spatial and temporal resolution data acquired by CubeSat satellites have desirable characteristics for mapping large-scale forest structural attributes. However, most studies have used a median composite or single image for analyses. The multi-temporal use of CubeSat data may improve prediction accuracy. This study evaluates the capabilities of PlanetScope CubeSat data to estimate canopy height derived from airborne Light Detection and Ranging (LiDAR) by comparing estimates using Sentinel-2 and Landsat 8 data. Random forest (RF) models using a single composite, multi-seasonal composites, and time-series data were investigated at different spatial resolutions of 3, 10, 20, and 30 m. The highest prediction accuracy was obtained by the PlanetScope multi-seasonal composites at 3 m (relative root mean squared error: 51.3%) and Sentinel-2 multi-seasonal composites at the other spatial resolutions (40.5%, 35.2%, and 34.2% for 10, 20, and 30 m, respectively). The results show that RF models using multi-seasonal composites are 1.4% more accurate than those using harmonic metrics from time-series data in the median. PlanetScope is recommended for canopy height mapping at finer spatial resolutions. However, the unique characteristics of PlanetScope data in a spatial and temporal context should be further investigated for operational forest monitoring.

scholarly journals Distance variable improvement of time-series big data stream evaluation

2020 ◽  
Vol 7 (1) ◽  
Author(s):  
Ari Wibisono ◽  
Petrus Mursanto ◽  
Jihan Adibah ◽  
Wendy D. W. T. Bayu ◽  
May Iffah Rizki ◽  
...  
Keyword(s):  
Time Series ◽  
Standard Deviation ◽  
Standard Method ◽  
Time Series Data ◽  
Series Data ◽  
Percentage Error ◽  
Traffic Demand ◽  
Incremental Model ◽  
Chernoff Bound ◽  
The Mean

Abstract Real-time information mining of a big dataset consisting of time series data is a very challenging task. For this purpose, we propose using the mean distance and the standard deviation to enhance the accuracy of the existing fast incremental model tree with the drift detection (FIMT-DD) algorithm. The standard FIMT-DD algorithm uses the Hoeffding bound as its splitting criterion. We propose the further use of the mean distance and standard deviation, which are used to split a tree more accurately than the standard method. We verify our proposed method using the large Traffic Demand Dataset, which consists of 4,000,000 instances; Tennet’s big wind power plant dataset, which consists of 435,268 instances; and a road weather dataset, which consists of 30,000,000 instances. The results show that our proposed FIMT-DD algorithm improves the accuracy compared to the standard method and Chernoff bound approach. The measured errors demonstrate that our approach results in a lower Mean Absolute Percentage Error (MAPE) in every stage of learning by approximately 2.49% compared with the Chernoff Bound method and 19.65% compared with the standard method.

scholarly journals An External Agribusiness Risk Analysis Using KBDI: A Case of Veldfires in the Northern Territory of Australia

2020 ◽  
Vol 3 (1) ◽  
pp. 37
Author(s):  
Toyi Maniki Diphagwe ◽  
Bernard Moeketsi Hlalele ◽  
Dibuseng Priscilla Mpakathi
Keyword(s):  
Time Series ◽  
Probability Distribution ◽  
Crop Production ◽  
Time Series Data ◽  
Northern Territory ◽  
Return Level ◽  
Series Data ◽  
Burned Area ◽  
Trend Test ◽  
P Value

The 2019/20 Australian bushfires burned over 46 million acres of land, killed 34 people and left 3500 individuals homeless. Majority of deaths and buildings destroyed were in New South Wales, while the Northern Territory accounted for approximately 1/3 of the burned area. Many of the buildings that were lost were farm buildings, adding to the challenge of agricultural recovery that is already complex because of ash-covered farmland accompanied by historic levels of drought. The current research therefore aimed at characterising veldfire risk in the study area using Keetch-Byram Drought Index (KBDI). A 39-year-long time series data was obtained from an online NASA database. Both homogeneity and stationarity tests were deployed using a non-parametric Pettitt’s and Dicky-Fuller tests respectively for data quality checks. Major results revealed a non-significant two-tailed Mann Kendall trend test with a p-value = 0.789 > 0.05 significance level. A suitable probability distribution was fitted to the annual KBDI time series where both Kolmogorov-Smirnov and Chi-square tests revealed Gamma (1) as a suitably fitted probability distribution. Return level computation from the Gamma (1) distribution using XLSTAT computer software resulted in a cumulative 40-year return period of moderate to high fire risk potential. With this low probability and 40-year-long return level, the study found the area less prone to fire risks detrimental to animal and crop production. More agribusiness investments can safely be executed in the Northern Territory without high risk aversion.

Sign in / Sign up

Export Citation Format

Share Document