The Short-Term Prediction of Length of Day Using 1D Convolutional Neural Networks (1D CNN)

Guessoum, Sonia; Belda, Santiago; Ferrándiz, José M.; Modiri, Sadegh; Raut, Shrishail; Dhar, Sujata; Heinkelmann, Robert; Schuh, Harald

doi:10.3390/s22239517

Cited by 23 publications

(3 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…When training a model on reflectivity curves with fixed discretization, a 1D CNN embedding network is parameterefficient and makes better use of the sequential characteristics of the data. While less popular than their 2D counterparts, 1D CNNs have been successfully used in a variety of tasks involving 1D signals, such as automatic speech recognition (Collobert et al, 2016) and time-series prediction (Guessoum et al, 2022). The 1D CNN, as shown in Fig.…”

Section: Embedding Networkmentioning

confidence: 99%

Neural network analysis of neutron and X-ray reflectivity data incorporating prior knowledge

Munteanu,

Starostin,

Greco

et al. 2024

J Appl Crystallogr

View full text Add to dashboard Cite

Due to the ambiguity related to the lack of phase information, determining the physical parameters of multilayer thin films from measured neutron and X-ray reflectivity curves is, on a fundamental level, an underdetermined inverse problem. This ambiguity poses limitations on standard neural networks, constraining the range and number of considered parameters in previous machine learning solutions. To overcome this challenge, a novel training procedure has been designed which incorporates dynamic prior boundaries for each physical parameter as additional inputs to the neural network. In this manner, the neural network can be trained simultaneously on all well-posed subintervals of a larger parameter space in which the inverse problem is underdetermined. During inference, users can flexibly input their own prior knowledge about the physical system to constrain the neural network prediction to distinct target subintervals in the parameter space. The effectiveness of the method is demonstrated in various scenarios, including multilayer structures with a box model parameterization and a physics-inspired special parameterization of the scattering length density profile for a multilayer structure. In contrast to previous methods, this approach scales favourably when increasing the complexity of the inverse problem, working properly even for a five-layer multilayer model and a periodic multilayer model with up to 17 open parameters.

show abstract

Section: Embedding Networkmentioning

confidence: 99%

Neural network analysis of neutron and X-ray reflectivity data incorporating prior knowledge

Munteanu,

Starostin,

Greco

et al. 2024

J Appl Crystallogr

View full text Add to dashboard Cite

show abstract

“…CNN model CNN models have superior capabilities in extracting local features (18). They can be employed in computer vision and time series prediction and analysis (19). CNN models consist of convolutional layers, pooling layers, activation functions, and dense layers (Figure 3).…”

Section: Deep Learning Model I Data Pre-processingmentioning

confidence: 99%

Forecasting and Analyzing Influenza Activity in Hebei Province, China, Using a CNN-LSTM Hybrid Model

Li,

Han

et al. 2024

Preprint

View full text Add to dashboard Cite

Background Influenza, an acute infectious respiratory disease, presents a significant global health challenge. Accurate prediction of influenza activity is crucial for reducing its impact. Therefore, this study seeks to develop a hybrid Convolution Neural Network - Long Short Term Memory neural network (CNN-LSTM) model to forecast the percentage of influenza-like-illness (ILI) rate in Hebei Province, China. The aim is to provide more precise guidance for influenza prevention and control measures. Methods Using ILI% data from 28 national sentinel hospitals in the Hebei Province, spanning from 2010 to 2022, we employed the Python deep learning framework PyTorch to develop four distinct forecasting prediction models. We assessed each model’s prediction performance using mean absolute error (MAE) and root mean square error (RMSE). Results The Seasonal Auto-Regressive Indagate Moving Average (SARIMA) model had the highest error among the four forecasting models, with a MAE value of 0.8913 and an RMSE value of 1.2098. The CNN-LSTM model had the lowest error, with MAE and RMSE values of 0.0.3987 and 0.5448, respectively. The CNN-LSTM model thus had a significantly better prediction performance compared to the SARIMA model, with a 55.26% decrease in MAE and a 54.97% decrease in RMSE. When compared to the standalone Convolution Neural Network (CNN) and Long Short Term Memory neural network (LSTM) models, the CNN-LSTM model showed performance enhancements of 32.86% for MAE and 28.60% for RMSE over CNN, and of 11.05% for MAE and 13.07% for RMSE over LSTM. Conclusion The hybrid CNN-LSTM model had better prediction performances than the SARIMA, CNN, and LSTM models. This hybrid model could provide more accurate influenza activity projections in the Hebei Province.

show abstract

“…A number of techniques and data combinations have been used and developed to improve the accuracy of EOP prediction, e.g., neural networks (Guessoum et al, 2022;Liao et al, 2012;Schuh et al, 2002), machine learning (Kiani Shahvandi et al, 2022;Lei et al, 2017), kriging Ligas, 2021, 2022), kalman filter (Gross et al, 1998;Nastula et al, 2020;Xu et al, 2012), singular spectrum analysis (Okhotnikov and Golyandina, 2019), or autoregressive models (Dill et al, 2018;Niedzielski and Kosek, 2011).…”

Section: Introductionmentioning

confidence: 99%

Short-term prediction of UT1-UTC and LOD via Dynamic Mode Decomposition and combination of least-squares and vector autoregressive model

Michalczak,

Ligas

2024

Reports on Geodesy and Geoinformatics

View full text Add to dashboard Cite

This study presents a short-term forecast of UT1-UTC and LOD using two methods, i.e. Dynamic Mode Decomposition (DMD) and combination of Least-Squares and Vector Autoregression (LS+VAR). The prediction experiments were performed separately for yearly time spans, 2018-2022. The prediction procedure started on January 1 and ended on December 31, with 7-day shifts between subsequent 30-day forecasts. Atmospheric Angular Momentum data (AAM) were used as an auxiliary time series to potentially improve the prediction accuracy of UT1-UTC and LOD in LS+VAR procedure. An experiment was also conducted with and without elimination of effect of zonal tides from UT1-UTC and LOD time series. Two approaches to using the best steering parameters for the methods were applied:. First, an adaptive approach, which observes the rule that before every single forecast, a preliminary one must be performed on the pre-selected sets of parameters, and the one with the smallest prediction error is then used for the final prediction; and second, an averaged approach, whereby several forecasts are made with different sets of parameters (the same parameters as in adaptive approach) and the final values are calculated as the averages of these predictions. Depending on the method and data combination mean absolute prediction errors (MAPE) for UT1-UTC vary from 0.63 ms to 1.43ms for the 10th day and from 3.07 ms to 8.05ms for the 30th day of the forecast. Corresponding values for LOD vary from 0.110 ms to 0.245 ms for the 10th day and from 0.148 ms to 0.325 ms for the 30th day.

show abstract

The Short-Term Prediction of Length of Day Using 1D Convolutional Neural Networks (1D CNN)

Cited by 23 publications

References 39 publications

Neural network analysis of neutron and X-ray reflectivity data incorporating prior knowledge

Neural network analysis of neutron and X-ray reflectivity data incorporating prior knowledge

Forecasting and Analyzing Influenza Activity in Hebei Province, China, Using a CNN-LSTM Hybrid Model

Short-term prediction of UT1-UTC and LOD via Dynamic Mode Decomposition and combination of least-squares and vector autoregressive model

Contact Info

Product

Resources

About