Optimizing the LSTM Deep Learning Model for Arctic Sea Ice Melting Prediction

The National Oceanic and Atmospheric Administration reports a 95% decline in the oldest Arctic ice over the last 33 years [1], while the National Aeronautics and Space Administration states that summer Arctic Sea Ice Extent (SIE) is shrinking by 12.2% per decade since 1979 due to warmer temperatures [2]. Given the rapidly changing Arctic conditions, accurate prediction models are crucial. Deep learning models developed for Arctic forecasts primarily focus on exploring convolutional neural networks (CNN) and convolutional Long Short-Term Memory (LSTM) networks, while the exploration of the power of LSTM networks is limited. In this research, we focus on enhancing the performance of an LSTM network for predicting monthly Arctic SIE. We leverage five climate and atmospheric variables, validated for their correlation with SIE in prior studies [3]. We utilize the Spearman’s rank correlation and ExtraTrees regressor to enhance our understanding of the importance of the five variables in predicting SIE. We further enhance our predictor variables with seasonal information, lagged time steps, and a linear regression simulated SIE that accounts for the influence of past SIE on current SIE. Statistical methods guide our selection of data scalers and best evaluation metrics for our model. By experimenting with hyperparameter optimization and advanced deep learning training techniques, such as batch sizes, number of neurons, early stopping, and model checkpoint, our model achieved a Mean Absolute Error (MAE) of 0.191 and R2 of 0.996, underscoring its ability to account for nearly all the variance in our data and holds great promise for the prediction of SIE.

Multipath Selection Algorithm Based on Dynamic Flow Prediction

Traditional traffic management techniques appear to be incompetent in complex data center networks, so proposes a load balancing strategy based on Long Short-Term Memory (LSTM) and quantum annealing by Software Defined Network (SDN) to dynamically predict the traffic and comprehensively consider the current and predicted load of the network in order to select the optimal forwarding path and balance the network load. Experiments have demonstrated that the algorithm achieves significant improvement in both system throughput and average packet loss rate for the purpose of improving network quality of service.

Wind Speed Prediction Based on Improved VMD-BP-CNN-LSTM Model

Amid the randomness and volatility of wind speed, an improved VMD-BP-CNN-LSTM model for short-term wind speed prediction was proposed to assist in power system planning and operation in this paper. Firstly, the wind speed time series data was processed using Variational Mode Decomposition (VMD) to obtain multiple frequency components. Then, each individual frequency component was channeled into a combined prediction framework consisting of BP neural network (BPNN), Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) after the execution of differential and normalization operations. Thereafter, the predictive outputs for each component underwent integration through a fully-connected neural architecture for data fusion processing, resulting in the final prediction. The VMD decomposition technique was introduced in a generalized CNN-LSTM prediction model;a BPNN model was utilized to predict high-frequency components obtained from VMD, and incorporated a fully connected neural network for data fusion of individual component predictions. Experimental results demonstrated that the proposed improved VMD-BP-CNN-LSTM model outperformed other combined prediction models in terms of prediction accuracy, providing a solid foundation for optimizing the safe operation of wind farms.

深度学习在钢结构货架变形预测中的应用研究

随着工业化和物流行业的发展,钢结构货架在仓储和物流系统中越来越重要,因此准确预测其变形至关重要。文章介绍了一种基于双向长短时记忆网络(BiLSTM)和注意力机制的预测算法,该算法利用时间序列数据,通过深度学习模型进行训练,能够更细致地分析和预测钢结构货架的变形。结合一个典型应用验证了模型性能,证实了其高稳健性和出色的预测精度。实验结果表明,该模型能够准确地预测钢结构货架的变形情况,其平均误差仅为0.15%~3.33%。这些结果表明了该算法在钢结构货架自动化监测领域的潜在应用前景,为其结构变形预测提供了一种可行的解决方案。

PREDICTION OF MECHANICAL PROPERTY OF WHISKER REINFORCED METAL MATRIX COMPOSITE: PART-II. VERIFICATION & APPLICATION

The present paper continues the discussion in Part I. Model and Formulation. Based on the theory proposed in Part I, the formulae predicting stiffness moduli of the composites in some typical cases of whisker orientations and loading conditions are derived and compared with theoretical representatives in literatures, experimental measurement and commonly used empirical formulae. It seems that (1) with whisker reinforcing and matrix hardening considered, the present prediction is in well agreement with the experimental measurement; (2) the present theory can predict accurate moduli with the proper pre calculated parameters; (3) the upper bound and lower bound of the present theory are just the predictions of equal strain theory and equal stress theory; (4) the present theory provides a physical explanation and theoretical base for the present commonly used empirical formulae. Compared with the microscopic mechanical theories, the present theory is competent for modulus prediction of practical engineering composite in accuracy and simplicity. [WT5”HZ]

Application of deep learning technique to the sea surface height prediction in the South China Sea

A deep-learning-based method,called ConvLSTMP3,is developed to predict the sea surface heights(SSHs).ConvLSTMP3 is data-driven by treating the SSH prediction problem as the one of extracting the spatial-temporal features of SSHs,in which the spatial features are“learned”by convolutional operations while the temporal features are tracked by long short term memory(LSTM).Trained by a reanalysis dataset of the South China Sea(SCS),ConvLSTMP3 is applied to the SSH prediction in a region of the SCS east off Vietnam coast featured with eddied and offshore currents in summer.Experimental results show that ConvLSTMP3 achieves a good prediction skill with a mean RMSE of 0.057 m and accuracy of 93.4%averaged over a 15-d prediction period.In particular,ConvLSTMP3 shows a better performance in predicting the temporal evolution of mesoscale eddies in the region than a full-dynamics ocean model.Given the much less computation in the prediction required by ConvLSTMP3,our study suggests that the deep learning technique is very useful and effective in the SSH prediction,and could be an alternative way in the operational prediction for ocean environments in the future.

A Spatial-Temporal Attention Model for Human Trajectory Prediction

Human trajectory prediction is essential and promising in many related applications. This is challenging due to the uncertainty of human behaviors, which can be influenced not only by himself, but also by the surrounding environment. Recent works based on long-short term memory(LSTM) models have brought tremendous improvements on the task of trajectory prediction. However, most of them focus on the spatial influence of humans but ignore the temporal influence. In this paper, we propose a novel spatial-temporal attention(ST-Attention) model,which studies spatial and temporal affinities jointly. Specifically,we introduce an attention mechanism to extract temporal affinity,learning the importance for historical trajectory information at different time instants. To explore spatial affinity, a deep neural network is employed to measure different importance of the neighbors. Experimental results show that our method achieves competitive performance compared with state-of-the-art methods on publicly available datasets.

Short-time prediction for traffic flow based on wavelet de-noising and LSTM model

Aiming at the problem that some existing traffic flow prediction models are only for a single road segment and the model input data are not pre-processed,a heuristic threshold algorithm is used to de-noise the original traffic flow data after wavelet decomposition.The correlation coefficients of road traffic flow data are calculated and the data compression matrix of road traffic flow is constructed.Data de-noising minimizes the interference of data to the model,while the correlation analysis of road network data realizes the prediction at the road network level.Utilizing the advantages of long short term memory(LSTM)network in time series data processing,the compression matrix is input into the constructed LSTM model for short-term traffic flow prediction.The LSTM-1 and LSTM-2 models were respectively trained by de-noising processed data and original data.Through simulation experiments,different prediction times were set,and the prediction results of the prediction model proposed in this paper were compared with those of other methods.It is found that the accuracy of the LSTM-2 model proposed in this paper increases by 10.278%on average compared with other prediction methods,and the prediction accuracy reaches 95.58%,which proves that the short-term traffic flow prediction method proposed in this paper is efficient.

Prison Term Prediction on Criminal Case Description with Deep Learning

The task of prison term prediction is to predict the term of penalty based on textual fact description for a certain type of criminal case.Recent advances in deep learning frameworks inspire us to propose a two-step method to address this problem.To obtain a better understanding and more specific representation of the legal texts,we summarize a judgment model according to relevant law articles and then apply it in the extraction of case feature from judgment documents.By formalizing prison term prediction as a regression problem,we adopt the linear regression model and the neural network model to train the prison term predictor.In experiments,we construct a real-world dataset of theft case judgment documents.Experimental results demonstrate that our method can effectively extract judgment-specific case features from textual fact descriptions.The best performance of the proposed predictor is obtained with a mean absolute error of 3.2087 months,and the accuracy of 72.54%and 90.01%at the error upper bounds of three and six months,respectively.

Ship motion extreme short time prediction of ship pitch based on diagonal recurrent neural network

A DRNN (diagonal recurrent neural network) and its RPE (recurrent prediction error) learning algorithm are proposed in this paper .Using of the simple structure of DRNN can reduce the capacity of calculation. The principle of RPE learning algorithm is to adjust weights along the direction of Gauss-Newton. Meanwhile, it is unnecessary to calculate the second local derivative and the inverse matrixes, whose unbiasedness is proved. With application to the extremely short time prediction of large ship pitch, satisfactory results are obtained. Prediction effect of this algorithm is compared with that of auto-regression and periodical diagram method, and comparison results show that the proposed algorithm is feasible.

An efficient stock market prediction model using hybrid feature reduction method based on variational autoencoders and recursive feature elimination

In this study,the hourly directions of eight banking stocks in Borsa Istanbul were predicted using linear-based,deep-learning(LSTM)and ensemble learning(Light-GBM)models.These models were trained with four different feature sets and their performances were evaluated in terms of accuracy and F-measure metrics.While the first experiments directly used the own stock features as the model inputs,the second experiments utilized reduced stock features through Variational AutoEncoders(VAE).In the last experiments,in order to grasp the effects of the other banking stocks on individual stock performance,the features belonging to other stocks were also given as inputs to our models.While combining other stock features was done for both own(named as allstock_own)and VAE-reduced(named as allstock_VAE)stock features,the expanded dimensions of the feature sets were reduced by Recursive Feature Elimination.As the highest success rate increased up to 0.685 with allstock_own and LSTM with attention model,the combination of allstock_VAE and LSTM with the attention model obtained an accuracy rate of 0.675.Although the classification results achieved with both feature types was close,allstock_VAE achieved these results using nearly 16.67%less features compared to allstock_own.When all experimental results were examined,it was found out that the models trained with allstock_own and allstock_VAE achieved higher accuracy rates than those using individual stock features.It was also concluded that the results obtained with the VAE-reduced stock features were similar to those obtained by own stock features.

A Short-Term Climate Prediction Model Based on a Modular Fuzzy Neural Network

In terms of the modular fuzzy neural network (MFNN) combining fuzzy c-mean (FCM) cluster and single-layer neural network, a short-term climate prediction model is developed. It is found from modeling results that the MFNN model for short-term climate prediction has advantages of simple structure, no hidden layer and stable network parameters because of the assembling of sound functions of the self-adaptive learning, association and fuzzy information processing of fuzzy mathematics and neural network methods. The case computational results of Guangxi flood season (JJA) rainfall show that the mean absolute error (MAE) and mean relative error (MRE) of the prediction during 1998-2002 are 68.8 mm and 9.78%, and in comparison with the regression method, under the conditions of the same predictors and period they are 97.8 mm and 12.28% respectively. Furthermore, it is also found from the stability analysis of the modular model that the change of the prediction results of independent samples with training times in the stably convergent interval of the model is less than 1.3 mm. The obvious oscillation phenomenon of prediction results with training times, such as in the common back-propagation neural network (BPNN) model, does not occur, indicating a better practical application potential of the MFNN model.

PREDICTION OF MECHANICAL PROPERTY OF WHISKER REINFORCED METAL MATRIX COMPOSITE: PART-I. MODEL AND FORMULATION

Based on study of strain distribution in whisker reinforced metal matrix composites, an explicit precise stiffness tensor is derived. In the present theory, the effect of whisker orientation on the macro property of composites is considered, but the effect of random whisker position and the complicated strain field at whisker ends are averaged. The derived formula is able to predict the stiffness modulus of composites with arbitrary whisker orientation under any loading condition. Compared with the models of micro mechanics, the present theory is competent for modulus prediction of actual engineering composites. The verification and application of the present theory are given in a subsequent paper published in the same issue

An Approach for Improving Short-Term Prediction of Summer Rainfall over North China by Decomposing Interannual and Decadal Variability

A statistical downscaling approach was developed to improve seasonal-to-interannual prediction of summer rainfall over North China by considering the effect of decadal variability based on observational datasets and dynamical model outputs.Both predictands and predictors were first decomposed into interannual and decadal components.Two predictive equations were then built separately for the two distinct timescales by using multivariate linear regressions based on independent sample validation.For the interannual timescale,850-hPa meridional wind and 500-hPa geopotential heights from multiple dynamical models＇ hindcasts and SSTs from observational datasets were used to construct predictors.For the decadal timescale,two well-known basin-scale SST decadal oscillation （the Atlantic Multidecadal Oscillation and the Pacific Decadal Oscillation） indices were used as predictors.Then,the downscaled predictands were combined to represent the predicted/hindcasted total rainfall.The prediction was compared with the models＇ raw hindcasts and those from a similar approach but without timescale decomposition.In comparison to hindcasts from individual models or their multi-model ensemble mean,the skill of the present scheme was found to be significantly higher,with anomaly correlation coefficients increasing from nearly neutral to over 0.4 and with RMSE decreasing by up to 0.6 mm d-1.The improvements were also seen in the station-based temporal correlation of the predictions with observed rainfall,with the coefficients ranging from-0.1 to 0.87,obviously higher than the models＇ raw hindcasted rainfall results.Thus,the present approach exhibits a great advantage and may be appropriate for use in operational predictions.

Traffic flow prediction of urban road network based on LSTM-RF model

Traffic flow prediction,as the basis of signal coordination and travel time prediction,has become a research point in the field of transportation.For traffic flow prediction,researchers have proposed a variety of methods,but most of these methods only use the time domain information of traffic flow data to predict the traffic flow,ignoring the impact of spatial correlation on the prediction of target road segment flow,which leads to poor prediction accuracy.In this paper,a traffic flow prediction model called as long short time memory and random forest(LSTM-RF)was proposed based on the combination model.In the process of traffic flow prediction,the long short time memory(LSTM)model was used to extract the time sequence features of the predicted target road segment.Then,the predicted value of LSTM and the collected information of adjacent upstream and downstream sections were simultaneously used as the input features of the random forest model to analyze the spatial-temporal correlation of traffic flow,so as to obtain the final prediction results.The traffic flow data of 132 urban road sections collected by the license plate recognition system in Guiyang City were tested and verified.The results show that the method is better than the single model in prediction accuracy,and the prediction error is obviously reduced compared with the single model.

The long-term prediction of the oil-contaminated water from the Sanchi collision in the East China Sea

The condensate and bunker oil leaked from the Sanchi collision would cause a persistent impact on marine ecosystems in the surrounding areas. The long-term prediction for the distribution of the oil-polluted water and the information for the most affected regions would provide valuable information for the oceanic environment protection and pollution assessment. Based on the operational forecast system developed by the First Institute of Oceanography, State Oceanic Administration, we precisely predicted the drifting path of the oil tanker Sanchi after its collision. Trajectories of virtual oil particles show that the oil leaked from the Sanchi after it sank is mainly transported to the northeastern part of the sink location, and quickly goes to the open ocean along with the Kuroshio. Risk probability analysis based on the outcomes from the operational forecast system for years 2009 to2017 shows that the most affected area is at the northeast of the sink location.

Cross-Band Spectrum Prediction Based on Deep Transfer Learning

Spectrum prediction is a promising technology to infer future spectrum state by exploiting inherent patterns of historical spectrum data.In practice,for a given spectrum band of interest,when facing relatively scarce historical data,spectrum prediction based on traditional learning methods does not work well.Thus,this paper proposes a cross-band spectrum prediction model based on transfer learning.Firstly,by analysing service activities and computing the distances between various frequency points based on Dynamic Time Warping,the similarity between spectrum bands has been verified.Next,the features,which mainly affect the performance of transfer learning in the crossband spectrum prediction,are explored by leveraging transfer component analysis.Then,the effectiveness of transfer learning for the cross-band spectrum prediction has been demonstrated.Further,experimental results with real-world spectrum data demonstrate that the performance of the proposed model is better than the state-of-theart models when the historical spectrum data is limited.

Verification of Short-Term Predictions of Solar Soft X-ray Bursts for the Maximum Phase (2000-2001) of Solar Cycle 23

We present a verification of the short-term predictions of solar X-ray bursts for the maximum phase (2000–2001) of Solar Cycle 23, issued by two prediction centers. The results are that the rate of correct predictions is about equal for RWC-China and WWA; the rate of too high predictions is greater for RWC-China than for WWA, while the rate of too low predictions is smaller for RWC-China than for WWA.

Prediction Model of Weekly Retail Price for Eggs Based on Chaotic Neural Network

This paper establishes a short-term prediction model of weekly retail prices for eggs based on chaotic neural network with the weekly retail prices of eggs from January 2008 to December 2012 in China.In the process of determining the structure of the chaotic neural network,the number of input layer nodes of the network is calculated by reconstructing phase space and computing its saturated embedding dimension,and then the number of hidden layer nodes is estimated by trial and error.Finally,this model is applied to predict the retail prices of eggs and compared with ARIMA.The result shows that the chaotic neural network has better nonlinear fitting ability and higher precision in the prediction of weekly retail price of eggs.The empirical result also shows that the chaotic neural network can be widely used in the field of short-term prediction of agricultural prices.

Improved Prediction of Metamaterial Antenna Bandwidth Using Adaptive Optimization of LSTM

The design of an antenna requires a careful selection of its parameters to retain the desired performance.However,this task is time-consuming when the traditional approaches are employed,which represents a significant challenge.On the other hand,machine learning presents an effective solution to this challenge through a set of regression models that can robustly assist antenna designers to find out the best set of design parameters to achieve the intended performance.In this paper,we propose a novel approach for accurately predicting the bandwidth of metamaterial antenna.The proposed approach is based on employing the recently emerged guided whale optimization algorithm using adaptive particle swarm optimization to optimize the parameters of the long-short-term memory(LSTM)deep network.This optimized network is used to retrieve the metamaterial bandwidth given a set of features.In addition,the superiority of the proposed approach is examined in terms of a comparison with the traditional multilayer perceptron(ML),Knearest neighbors(K-NN),and the basic LSTM in terms of several evaluation criteria such as root mean square error(RMSE),mean absolute error(MAE),and mean bias error(MBE).Experimental results show that the proposed approach could achieve RMSE of(0.003018),MAE of(0.001871),and MBE of(0.000205).These values are better than those of the other competing models.