Neural Network Based Prediction Method Research Articles

A physics merged deep neural network-based prediction method for jet turbulent mixing noise

Turbulent mixing noise is a vital component of jet noise, and its rapid, accurate prediction has always been persistently pursued. Recent advancement in machine learning has been applied to jet noise prediction. However, these applications are pure curve fitting and lack physical constraints. In this study, a physics-merged deep neural network (PMNN)-based prediction method was developed for turbulent mixing jet noise by merging the physics of the jet noise. This deep neural network (DNN)-based method employed recent advancements in jet turbulent mixing noise containing large- and fine-scale turbulence structures. Two simple rational functions for large- and fine-scale turbulent noise similarity spectra were proposed to replace the original complex similarity spectra functions and incorporated into the DNN-based prediction method. For comparison, we present two data-driven DNN-based prediction methods (DDNN). The first DDNN method used the sound pressure level (SPL) as the output of neural networks, directly establishing the nonlinear relationship between the input features and SPL. In the second DDNN method, the dominant modes of the jet noise spectra extracted using the proper orthogonal decomposition method were merged with DNN. These DNN-based methods were then trained using a set of experimental data over a wide range of jet operating conditions. Their performance was evaluated and compared. It was evident that all these DNN-based methods were capable of predicting turbulent mixing noise reasonably well. In contrast to the DDNN methods, the PMNN method could provide insights into the jet turbulent mixing noise components. It demonstrates that the turbulent mixing jet noise spectra at the mid polar angle is generated by the large-scale noise component at low-frequency range and by the fine-scale noise component at high-frequency range.

Remaining useful life prediction of vehicle-oriented PEMFC systems based on IGWO-BP neural network under real-world traffic conditions

Accurately predicting the useful life can serve as a pivotal reference for effectively extending the lifespan of proton exchange membrane fuel cells (PEMFCs). Herein, this paper proposes a novel method that combines an improved grey wolf optimizer (IGWO) algorithm and a backpropagation (BP) neural network to predict the remaining useful life (RUL) of vehicle-oriented PEMFC systems using relative power loss rate (RPLR) as a health indicator under real-world traffic conditions. First, The Pearson correlation coefficient is used to analyze the correlation of various monitoring parameters and to verify the effectiveness of RPLR as a dynamic health indicator. Then, the IGWO-BP neural network-based prediction method is described and used to predict the RUL of PEMFC systems. Finally, the accuracy and reliability of the proposed method are validated against two separate datasets of PEMFC city buses operating under different traffic conditions. Compared with other methods, the proposed method has a relative error of less than 5 % and predicts a shorter RUL than the actual RUL. These findings illustrate that the proposed method has a high prediction accuracy and offers an early warning function, which is beneficial for practical applications.

Fusing sensitive degradation features with uncertainty analysis for RUL prediction of rotating machines

Recently, deep learning technology-based neural networks have been adopted for remaining useful life (RUL) prediction of rotating machines. However, there are still some shortcomings: (1) an individual degradation feature cannot sufficiently represent the degradation process, which has an adverse impact on the accuracy of prediction results; (2) most recurrent neural network-based prediction methods have difficulty in quantifying the uncertainty of the forecast results. In this paper, a fusing sensitive degradation features with uncertainty analysis for RUL prediction of rotating machines is proposed. Firstly, the statistical features contained in the vibration signal used to monitor the degradation of rotating equipment are extracted to construct the original feature set. Then, the weight coefficients of the monotonicity, correlation and robustness criteria are determined by the self-adjusting analytic hierarchy process. The sensitive features that describe the degradation process are selected from among the statistical features. Furthermore, the sensitive features are fed into residual networks and gated recurrent unit, and the spatial and temporal correlation of the features are considered to establish the health index (HI). Finally, the fitted HI is input into a Gaussian process regression model, and the prediction results with confidence intervals are obtained. To verify the effectiveness and superiority of the proposed method, two public bearing datasets and three model methods are used for comparative experiments.

A fatigue crack growth prediction method on small datasets based on optimized deep neural network and Delaunay data augmentation

The neural network-based prediction method has shown great potential in the field of fatigue crack growth prediction due to its powerful nonlinear processing and generalization capabilities. However, traditional deep neural network-based methods often encounter problems such as overfitting, underfitting, and instability when predicting FCG rates on small datasets. To address these issues, this study proposes a SA-DNN prediction method for metal fatigue crack growth by introducing Delaunay data augmentation and simulated annealing algorithm into the basic DNN framework. The proposed SA-DNN model is then used to predict the fatigue crack growth process of AM60B magnesium alloy and 7055 Al alloy, and the intrinsic and extrinsic generalization abilities of the model are evaluated. In both the evaluation of intrinsic and extrinsic generalization abilities, the mean squared errors of the training, validation, and testing sets rapidly converge to the target error in various epochs. The lowest mean squared errors achieved are 1.07 e-5 and 1.241 e-5, respectively. Moreover, the results also indicate that the proposed model can accurately predict the fatigue crack growth process of metal materials with varying degrees of nonlinearity.

Cross-Domain Remaining Useful Life Prediction Based on Adversarial Training

Remaining useful life prediction can assess the time to failure of degradation systems. Currently, numerous neural network-based prediction methods have been proposed by researchers. However, most of the work contains an implicit prerequisite: the network training and testing data have the same operating conditions. To solve this problem, an adversarial discriminative domain adaption prediction method based on adversarial training is proposed to improve the accuracy of cross-domain prediction under different working conditions. First, an LSTM feature extraction network is constructed to mine the source domain data and the target domain data for deep feature representation. Subsequently, the parameters of the target domain feature extraction network are adjusted based on the idea of adversarial training to achieve domain invariant feature mining. The proposed scheme is experimented on a publicly available dataset and achieves state-of-the-art prediction performance compared to recent unsupervised domain adaptation prediction methods.

Machine learning-based model development for battery state of charge–open circuit voltage relationship using regression techniques

The development of the Battery Management System mainly depends on an accurate equivalent circuit of the battery model which can be used for the estimation of State of Charge (SoC) and State of Health (SoH). The accuracy of the predicted battery model mainly depends on the exactness of the SoC–Open Circuit Voltage (OCV) relationship which is obtained through the Static Test. Numerous research works reported on various higher-order SoC–OCV battery mathematical models which are unsuitable for real-time applications. Also, there is no unique model for the estimation of battery parameters. This work proposes a Machine Learning (ML) based battery static SoC–OCV prediction method using the concept of supervised learning. The performance of the proposed ML based techniques is compared with the conventional curve fitting techniques. The real-time data of Lithium Ion battery at different temperature profiles are used to analyze the performance of the proposed techniques. The Root Mean Square Error (RMSE) and R-Square (R2) indices are considered in the study to validate the performance of the proposed algorithm. From the study, it is revealed that the Neural network based prediction method is found to be the best method with high R2 and low RMSE indices. The proposed method has the advantages such as simplicity and high accuracy. This predicted battery model can be used in electric vehicle applications.

Generalized regression neural network-based prediction methods for overlying strata failure zone height

Generalized regression neural network-based prediction methods for overlying strata failure zone height

Hybrid deep neural network based prediction method for unsteady flows with moving boundary

Hybrid deep neural network based prediction method for unsteady flows with moving boundary

A Deep Neural Network-Based Method for Prediction of Dementia Using Big Data.

The rise in dementia among the aging Korean population will quickly create a financial burden on society, but timely recognition of early warning for dementia and proper responses to the occurrence of dementia can enhance medical treatment. Health behavior and medical service usage data are relatively more accessible than clinical data, and a prescreening tool with easily accessible data could be a good solution for dementia-related problems. In this paper, we apply a deep neural network (DNN) to prediction of dementia using health behavior and medical service usage data, using data from 7031 subjects aged over 65 collected from the Korea National Health and Nutrition Examination Survey (KNHANES) in 2001 and 2005. In the proposed model, principal component analysis (PCA) featuring and min/max scaling are used to preprocess and extract relevant background features. We compared our proposed methodology, a DNN/scaled PCA, with five well-known machine learning algorithms. The proposed methodology shows 85.5% of the area under the curve (AUC), a better result than that using other algorithms. The proposed early prescreening method for possible dementia can be used by both patients and doctors.

Shale gas well flowback rate prediction for Weiyuan field based on a deep learning algorithm

The flowback rate of a shale gas well is controlled by many factors, such as geological and engineering factors, which have a certain complexity. For the aforementioned reason, the knowledge-driven method often has difficulty effectively capturing the correlation rules among multiple datasets. As a data-driven algorithm, deep learning has strong advantages in data correlation analysis, nonlinear fitting and other applications. In this paper, an algorithm is set up to forecast the shale gas well flowback rate by using a deep learning algorithm based on the flowback characteristic factors and data from 286 shale gas wells in the Weiyuan field. An algorithm and prediction model of the flowback parameters of the Weiyuan shale gas wells is established to effectively capture their flowback characteristics. First, the general situation of the study area is briefly introduced, and the data used to predict the flowback rate in this area are analyzed. Second, based on the deep learning algorithm, two kinds of deep feedforward neural network structures and a neural network-based prediction method of shale gas well flowback rate are designed. Finally, according to the regional data, the prediction and analysis of the flowback rate of Weiyuan shale gas wells are carried out. The results show that the correlations between 17 factors, such as fracturing section length and shale gas well flowback rate, are not obvious, but that the deep learning method proposed in this paper can effectively and accurately capture the correlations between the data and predict the flowback rate. The results show high accuracy.

Behavior Analysis of Sex based Cohorts Using the Toolset of Artificial Intelligence Based Insulin Sensitivity Prediction Methods

Abstract Tight glycaemic control (TGC) is a treatment in the intensive care in order to avoid stress-induced hyperglycaemia. The insulin sensitivity (SI) prediction is an essential step of the best performing, clinically applied so-called STAR (Stochastic-TARgeted) TGC protocol. Previous results showed performance improvement of the SI prediction using artificial intelligence methods. This study analyses the clinical performance of distinct artificial intelligence based SI prediction methods (2 different neural network based prediction methods: Classification Deep Network and Mixture Density Network with 3 different parametrizations and 2 variants: sex-specific and non sex-specific for each). In-silico validation was used for evaluation simulating the treatment of 171 virtual patients. Based on the results the number of input parameters involved into the prediction can effectively increase the reliability of the SI prediction. Improvements in the performance are also experienced in several cases by using sex-specific models.

A Deep Convolutional Neural Network to Improve the Prediction of Protein Secondary Structure

Background: Protein secondary structure prediction (PSSP) is a fundamental task in bioinformatics that is helpful for understanding the three-dimensional structure and biological function of proteins. Many neural network-based prediction methods have been developed for protein secondary structures. Deep learning and multiple features are two obvious means to improve prediction accuracy. Objective: To promote the development of PSSP, a deep convolutional neural network-based method is proposed to predict both the eight-state and three-state of protein secondary structure. Methods: In this model, sequence and evolutionary information of proteins are combined as multiple input features after preprocessing. A deep convolutional neural network with no pooling layer and connection layer is then constructed to predict the secondary structure of proteins. L2 regularization, batch normalization, and dropout techniques are employed to avoid over-fitting and obtain better prediction performance, and an improved cross-entropy is used as the loss function. Results: Our proposed model can obtain Q3 prediction results of 86.2%, 84.5%, 87.8%, and 84.7%, respectively, on CullPDB, CB513, CASP10 and CASP11 datasets, with corresponding Q8 prediction results of 74.1%, 70.5%, 74.9%, and 71.3%. Conclusion: We have proposed the DCNN-SS deep convolutional-network-based PSSP method, and experimental results show that DCNN-SS performs competitively with other methods.

Neural network–based prediction methods for height of water-flowing fractured zone caused by underground coal mining

The accurate calculation of the height of water-flowing fractured zone (WFFZ) in coal mine is a critical factor in ensuring mine safety and protecting surficial eco-environment. In view of inapplicability of traditional empirical formula for predicting the height of WFFZ, the correlation of the height of WFFZ and influence factors was analyzed firstly based on 82 collected groups of coalfield measured data in China. Results show that the mining thickness and mining depth have a significant effect on the height of WFFZ. Subsequently, the measured data were divided into two parts: 80% for training models and the remaining 20% for validation. Two prediction models, i.e., the multiple regression (MR) model and BP neural network (BPNN) model, were established and trained. A new merging model, multiple regression-BP neural network (MR-BPNN) model, was proposed by combining the multiple regression and BP neural network model. The prediction accuracy and generalization ability of the three models were verified by the recorded testing samples. Results of comparison suggest that three models all had better applicability for predicting the height of WFFZ in the coalmine, compared with the existing empirical prediction methods. More importantly, the MR-BPNN merging model combined the nonlinear mapping ability of neural network and empiric of multiple regression model, which could provide high-accurate, strong-generalized, and practical application for predicting the height of WFFZ of coalfield. In addition, the reason for the inapplicability of traditional empirical formula and the practicability of the proposed neural network-based prediction models were discussed.

Prediction of Regional Economic Trend with f Improved BP Neural Network Algorithm

Traditional BP neural network-based prediction method cannot accurately predictregional economic trend, and the operation efficiency is low. To this end, a method based on improved BP neural network algorithm is developed to predict regional economic trend. The Cobb-Douglas production function is used to select the most influential number of working people and economic investment in fixed assets, fiscal expenditure, gross foreign export value, retail sales of social consumer goods index as the sample data and normalized. The error of sample data is calculated by TangXC_BPModel model, and the the weights are modified uniformly so that and the regional economic trend is predicted accurately.

Hetero-Dimensional Multitask Neuroevolution for Chaotic Time Series Prediction

Chaotic time series prediction has important research and application value, and neural network-based prediction methods have problems such as low accuracy and difficulty in determining the number of nodes in the hidden layer. In recent years, as a representative of evolutionary multitask optimization algorithms(EMTO), multifactorial evolutionary algorithms (MFEA) have achieved excellent results in Multitask Optimization (MTO) problems, but it does not work well in solving hetero-dimensional multitask problem. We analyzed this phenomenon and proposed the concept of harmful transfer. Then to solve these problems, in this paper, we propose the hetero-dimensional multitask neuroevolution algorithms (HD-MFEA neuroevolution) for chaotic time series prediction. First, we propose hetero-dimensional assortative mating algorithm and self-adaption elite replacement algorithm, so that the hetero-dimensional multitask evolutionary algorithm (HD-MFEA) can overcome the problems caused by harmful transfer on hetero-dimensional multitask problem. Second, in order to adapt to the hierarchical characteristics of neural networks, we propose the neural level segmental coding strategy. Third, we consider the optimization of neural networks with different hidden layer nodes as hetero-dimensional multitask problem and use the HD-MFEA neuroevolution to solve this problem. We verify the effectiveness of the HD-MFEA on six hetero-dimensional multitask benchmark problems. The Mackey-Glass, Lorenz and Sunspot time series are used to demonstrate the performance of the HD-MFEA neuroevolution. Experimental results have showed that the HD-MFEA neuroevolution performs well in predicting the chaotic time series problems.

Machine learning-based prediction of component self-alignment in vapour phase and infrared soldering

Purpose This paper aims to investigate the self-alignment of 0603 size (1.5 × 0.75 mm) chip resistors, which were soldered by infrared or vapour phase soldering. The results were used for establishing an artificial neural network for predicting the component movement during the soldering. Design/methodology/approach The components were soldered onto an FR4 testboard, which was designed to facilitate the measuring of the position of the components both prior to and after the soldering. A semi-automatic placement machine misplaced the components intentionally, and the self-alignment ability was determined for soldering techniques of both infrared and vapour phase soldering. An artificial neural network-based prediction method was established, which is able to predict the position of chip resistors after soldering as a function of component misplacement prior to soldering. Findings The results showed that the component can self-align from farer distances by using vapour phase method, even from relative misplacement of 50 per cent parallel to the shorter side of the component. Components can self-align from a relative misplacement only of 30 per cent by using infrared soldering method. The established artificial neural network can predict the component self-alignment with an approximately 10-20 per cent mean absolute error. Originality/value It was proven that the vapour phase soldering method is more stable from the component’s self-alignment point of view. Furthermore, machine learning-based predictors can be applied in the field of reflow soldering technology, and artificial neural networks can predict the component self-alignment with an appropriately low error.

An Uncertainty-Incorporated Approach to Predict the Winner in StarCraft II Using Neural Processes

Real-time strategy (RTS) game is a kind of strategy game, in which the players compete for resources on 2D terrain by establishing the economy, training army, and guiding them into battle in real time. The winner prediction of the RTS games often involves studying a highly uncertain problem in an adversarial environment. In addition, the limit of the number of samples restricts on the application and performance of the prediction models. To obtain better winner prediction accuracy and maintain the prediction uncertainty under an adversarial environment, this paper proposes a neural network-based prediction method incorporated probability inference dealing with a small set of samples. This paper uses a dataset released based on SC2LE, a reinforcement learning environment released jointly by Blizzard Entertainment and DeepMind, and then employed the proposed neural processes model to build a winner prediction model. To verify, this paper implemented different features types' grouping and different game length grouping experiments for demonstrating better adaptability to such problems. Furthermore, this paper also implemented the SVM model experiments and compared the proposed method with the SVM model. Finally, when making predictions on a 1000 size testing data, the results show that the proposed prediction model achieves an accuracy of 0.811 at 200 and 0.821 at 1000 sizes of training sets, which is better than the SVM model with small training datasets.

Predicting Social Anxiety From Global Positioning System Traces of College Students: Feasibility Study.

BackgroundSocial anxiety is highly prevalent among college students. Current methodologies for detecting symptoms are based on client self-report in traditional clinical settings. Self-report is subject to recall bias, while visiting a clinic requires a high level of motivation. Assessment methods that use passively collected data hold promise for detecting social anxiety symptoms and supplementing self-report measures. Continuously collected location data may provide a fine-grained and ecologically valid way to assess social anxiety in situ.ObjectiveThe objective of our study was to examine the feasibility of leveraging noninvasive mobile sensing technology to passively assess college students’ social anxiety levels. Specifically, we explored the different relationships between mobility and social anxiety to build a predictive model that assessed social anxiety from passively generated Global Positioning System (GPS) data.MethodsWe recruited 228 undergraduate participants from a Southeast American university. Social anxiety symptoms were assessed using self-report instruments at a baseline laboratory session. An app installed on participants’ personal mobile phones passively sensed data from the GPS sensor for 2 weeks. The proposed framework supports longitudinal, dynamic tracking of college students to evaluate the relationship between their social anxiety and movement patterns in the college campus environment. We first extracted the following mobility features: (1) cumulative staying time at each different location, (2) the distribution of visits over time, (3) the entropy of locations, and (4) the frequency of transitions between locations. Next, we studied the correlation between these features and participants’ social anxiety scores to enhance the understanding of how students’ social anxiety levels are associated with their mobility. Finally, we used a neural network-based prediction method to predict social anxiety symptoms from the extracted daily mobility features.ResultsSeveral mobility features correlated with social anxiety levels. Location entropy was negatively associated with social anxiety (during weekdays, r=−0.67; and during weekends, r=−0.51). More (vs less) socially anxious students were found to avoid public areas and engage in less leisure activities during evenings and weekends, choosing instead to spend more time at home after school (4 pm-12 am). Our prediction method based on extracted mobility features from GPS trajectories successfully classified participants as high or low socially anxious with an accuracy of 85% and predicted their social anxiety score (on a scale of 0-80) with a root-mean-square error of 7.06.ConclusionsResults indicate that extracting and analyzing mobility features may help to reveal how social anxiety symptoms manifest in the daily lives of college students. Given the ubiquity of mobile phones in our society, understanding how to leverage passively sensed data has strong potential to address the growing needs for mental health monitoring and treatment.

Calculation of surface settlements caused by EPBM tunneling using artificial neural network, SVM, and Gaussian processes

Increasing demand on infrastructures increases attention to shallow soft ground tunneling methods in urbanized areas. Especially in metro tunnel excavations, due to their large diameters, it is important to control the surface settlements observed before and after excavation, which may cause damage to surface structures. In order to solve this problem, earth pressure balance machines (EPBM) and slurry balance machines have been widely used throughout the world. There are numerous empirical, analytical, and numerical analysis methods that can be used to predict surface settlements. But substantially fewer approaches have been developed for artificial neural network-based prediction methods especially in EPBM tunneling. In this study, 18 different parameters have been collected by municipal authorities from field studies pertaining to EPBM operation factors, tunnel geometric properties, and ground properties. The data source has a preprocess phase for the selection of the most effective parameters for surface settlement prediction. This paper focuses on surface settlement prediction using three different methods: artificial neural network (ANN), support vector machines (SVM), and Gaussian processes (GP). The success of the study has decreased the error rate to 13, 12.8, and 9, respectively, which is relatively better than contemporary research.

Neural network based prediction method for preventing condensation in chilled ceiling systems

Condensation is prone to occur at the startup moment in chilled ceiling systems, due to the infiltration and accumulation of moisture during system-off. To prevent condensation, an effective method is to operate the dedicated outdoor air system (DOAS) to dehumidify indoor air before operating chilled ceiling system. The pre-dehumidification time is critical. However, there is little experience in determining the pre-dehumidification time in both research and practice. In this study, neural network (NN) is used to predict condensation risk and the optimal pre-dehumidification time in chilled ceiling systems. Two NN models are developed to predict the temperature on the surface of chilled ceiling and indoor air dew-point temperature at the startup moment so as to evaluate the risk of condensation. The third NN model is developed to predict the optimal pre-humidification time for condensation prevention. Both training data and validation data are obtained from simulation tests in TRNSYS. The results show that 30 min pre-dehumidification is sufficient for the simulated building in Hong Kong. The influence of infiltration rate on the pre-dehumidification time is also investigated. This study also shows that NN-based method can be used for predictive control for condensation prevention in chilled ceiling systems.