Multilayer Feedforward Neural Network Research Articles

High-Tech Enterprise Financial Risk Early Warning System Based on Multi-Layer Feedforward Neural Network

High-Tech Enterprise Financial Risk Early Warning System Based on Multi-Layer Feedforward Neural Network

Adaptive control of a nonaffine nonlinear system using self-organising kernel extreme learning machine

We propose a self-organising kernel extreme learning machine (KELM) adaptive controller for nonaffine nonlinear systems. Literature survey reveals that neural network (NN) is extensively used to design adaptive controllers for nonlinear systems. When conventional NN, like multilayer feedforward NN and radial basis function NN (RBFNN), is used for controller design, the parameters of these networks converge slowly. Researchers have overcome this shortcoming by using extreme learning machine (ELM). The motivation to use KELM for controller design in our research is to utilise the advantages of ELM and radial basis function kernels. The structure of neural networks is seldom altered during training, resulting in unnecessarily small or large networks. The self-organising nature of our proposed controller caters to solving this problem. The structure of the self-organising KELM updates itself based on a threshold value set for the normalised change in the output weight. In our work, the control input meets three objectives: feedback linearisation, stabilisation of the linearised system and providing immunity to process and measurement noises. The update law for the hidden layer parameters of the KELM is obtained using the Lyapunov technique to ensure the overall stability of the system. A comparative analysis of different performance criteria is performed for trajectory tracking control, in the presence of process and measurement noises, for a numerical example and the Duffing–Holmes chaotic nonlinear system. The simulation results of these analyses demonstrate the superiority of the self-organising KELM compared to ELM and RBFNN based adaptive controllers. The experimental results with a rotary servo system validate the efficacy of the proposed controller in real-time systems. Furthermore, the robustness of the self-organising adaptive controller is verified with the results obtained for the servo system on varying the system parameter and operating condition.

SMART MINING: JOINT MODEL FOR PARAMETRIZATION OF COAL EXCAVATION PROCESS BASED ON ARTIFICIAL NEURAL NETWORKS

<p>In the present paper we propose a new artificial neural network model for the estimation of coal cutting resistance and excavator performance as a nonlinear relationship between the examined input (excavator movement angle in the left and right direction, slice height and thickness, coal unit weight, compressive and shear strength) and output factors (excavator effective capacity, maximum current/power/force/energy consumption, linear and areal cutting resistance). We analyze the dataset collected from three open-pit coal mines in Serbia: Field D, Tamnava Eastern Field and Tamnava Western Field (all part of the Kolubara coal basin). The model is developed using a multilayer feed-forward neural network, with a Levenberg-Marquardt learning algorithm. Results of the preformed analysis indicate satisfying statistical accuracyof the developed model (R>0.9). Additionally, we analyze the individual effects of input factors on the properties of coal cutting resistance and performance of the excavator, by invokling the multiple linear regression. As a result, we single out the statististically significant and physically possible interactions between the individual controlling factors</p>

Methodology for predicting dynamic response of multi-layer aluminum foam protected RC slabs under blast loading

Multi-layer aluminum foam with varying density distribution has been proven to exhibit significant blast alleviation capabilities under blast loads. Using the equivalent multiple degree of freedom (MDOF) method and rigid-perfectly-plastic-locking (RPPL) model, an analytical model for multi-layer aluminum foam protected RC (MF-RC) slabs is developed to predict displacement and energy properties. Studies on MF-RC slabs show that multi-layer aluminum foam with greater energy absorption causes more damage to RC slabs, revealing the limitations of the energy assessment method. RC slab displacement is used as a comprehensive assessment parameter for multi-layer aluminum foam using the pressure-impulse (P–I) diagram approach. Calculations and analyses of assessment indicators for different foam density distributions of triple-layer foam are carried out to compare their mitigation capacities under blast. Moreover, the multi-layer feedforward neural network (MFNN) is used as the surrogate model of the MDOF module, and an optimum Pareto solution between energy absorption and structural damage prevention is obtained using the non-dominated sorting genetic algorithm II (NSGA-II).

Evaluation of Feature Selection Methods in Estimation of Precipitation Based on Deep Learning Artificial Neural Networks

Precipitation is the most important element of the water cycle and an indispensable element of water resources management. This paper’s aim is to model the monthly precipitation in 8 precipitation observation stations in the province of Hamadan, Iran. The effects and role of different feature weights pre-processing methods (Weight by deviation, Weight by PCA, Weight by correlation and Weight by Support Vector Machine) on artificial intelligence modeling were investigated. Deep learning method based on a multi-layer feed-forward artificial neural network that is trained with Stochastic Gradient Descent using back-propagation (DL-SGD) and Convolutional Neural Networks (CNN) modelling were applied. The precipitation of each station is modeled using the precipitation values of the other stations. The best result, among all scenarios, at the Vasaj station according to the DL-SGD method (CC = 0.9845, NS = 0.9543 and RMSE = 10.4169 mm) and at the Varayineh station according to the CNN method (CC = 0.9679, NS = 0.9362 and RMSE = 16.0988 mm) were estimated.

Construction of rehabilitation training simulation model based on energy monitoring and transmission by wireless sensor networks

In today's world, with the advent of information technology, the speed of social development has also been greatly improved, and the national demand is also increasing. Among them, national health issues have always been the focus of countries. In addition to basic medical testing, the rehabilitation training for special groups in the medical industry is also gradually becoming more mature. In traditional wireless sensor networks, the energy consumption of nodes is often high, because each node needs to continuously send and receive data, leading to energy waste. In addition, many nodes remain active even when data transmission is not required, further increasing energy consumption. The coverage range of traditional wireless sensor networks is influenced by the number and distribution of nodes, often resulting in coverage vulnerabilities and blind spots. This makes it impossible for the network to fully cover the target area, reducing the reliability and accuracy of monitoring. Based on the above situation, this paper studies the construction of rehabilitation training simulation model by using wireless sensor network monitoring coverage and transmission technology on the basis of energy acquisition. Firstly, it discusses the development trend of rehabilitation training technology in various countries and the social status of rehabilitation training. Through the use of wireless sensor network monitors to monitor the target data collection and model construction. The target coverage and transmission speed are optimized by combining reinforcement learning algorithm SARSA and multilayer feedforward BP neural network. The sensor coverage and transmission of monitoring targets are also studied. The results show that the optimized wireless sensor system has a good effect in rehabilitation training.

Hidden classification layers: Enhancing linear separability between classes in neural networks layers

Many Deep Learning approaches are based on variations of standard multi-layer feed-forward neural networks. These are also referred to as deep networks. The basic idea is that each hidden neural layer accomplishes a data transformation which is expected to make the data representation “somewhat more linearly separable” than the previous one to obtain a final data representation which is as linearly separable as possible. However, determining the optimal network parameters for these transformations is a crucial challenge. In this study, we propose a Deep Neural Network architecture (Hidden Classification Layer, HCL) which induces an error function involving the output values of all the network layers. The proposed architecture leads toward solutions where the data representations in the hidden layers exhibit a higher degree of linear separability between classes compared to conventional methods. While similar approaches have been discussed in prior literature, this paper presents a new architecture with a novel error function and conducts an extensive experimental analysis. Furthermore, the architecture can be easily integrated into existing frameworks by simply adding densely connected layers and making a straightforward adjustment to the loss function to account for the output of the added layers. The experiments focus on image classification tasks using four well-established datasets, employing as baselines three widely recognized architectures in the literature. The findings reveal that the proposed approach consistently enhances accuracy on the test sets across all considered cases.

Convergence Analysis and Application for Multi‐Layer Neural Network Based on Fractional‐Order Gradient Descent Learning

AbstractFractional order calculus, with its inheritance and infinite memory properties, is a promising research area in information processing and modeling of certain physical systems, system identification, and control. In this paper, a fractional‐order gradient descent method is proposed for backpropagation training of multilayer feedforward neural networks. In particular, the Caputo derivative is used to define the measurement function and consider the smooth regular term. In addition, the monotonicity of the error function and the strong (weak) convergence theorem of the algorithm are rigorously proved. Finally, numerical experiments prove the correctness and effectiveness of the algorithm theory.

Coarse-Grained Artificial Intelligence for Design of Brush Networks.

The ability to synthesize elastomeric materials with programmable mechanical properties is vital for advanced soft matter applications. Due to the inherent complexity of hierarchical structure-property correlations in brush-like polymer networks, the application of conventional theory-based, so-called Human Intelligence (HI) approaches becomes increasingly difficult. Herein we developed a design strategy based on synergistic combination of HI and AI tools which allows precise encoding of mechanical properties with three architectural parameters: degrees of polymerization (DP) of network strands, nx, side chains, nsc, backbone spacers between side chains, ng. Implementing a multilayer feedforward artificial neural network (ANN), we took advantage of model-predicted structure-property cross-correlations between coarse-grained system code including chemistry specific characteristics S = [l, v, b] defined by monomer projection length l and excluded volume v, Kuhn length b of bare backbone and side chains, and architecture A = [nsc, ng, nx] of polymer networks and their equilibrium mechanical properties P = [G, β] including the structural shear modulus G and firmness parameter β. The ANN was trained by minimizing the mean-square error with Bayesian regularization to avoid overfitting using a data set of experimental stress-deformation curves of networks with brush-like strands of poly(n-butyl acrylate), poly(isobutylene), and poly(dimethylsiloxane) having structural modulus G < 50 kPa and 0.01 ≤ β ≤ 0.3. The trained ANN predicts network mechanical properties with 95% confidence. The developed ANN was implemented for synthesis of model networks with identical mechanical properties but different chemistries of network strands.

Applying artificial neural network to approximate and predict the transient dynamic behavior of CO2 combined cooling and power cycle

The CO2 combined cooling and power cycle (CCP) is a promising alternative for waste heat recovery due to its environmental friendliness and excellent performance. However, the transient dynamic behavior analysis and control of CCP systems are challenged by the instability of waste heat sources. In transient dynamic modeling, artificial neural networks, with their nonlinear mapping capabilities and relatively low computational requirements, prove advantageous over dynamic simulation models. In this study, six commonly used artificial neural network architectures are employed for approximating and predicting the transient dynamic behavior of CCP systems and subjected to preliminary applications. Results show that the multilayer feedforward neural network is the most suitable among the six networks for predicting and approximating the CCP system's transient dynamic behavior. Based on this model, a trajectory optimization control strategy is designed, leading to a 5.3 % improvement in CCP net power. This research underscores the effectiveness of artificial neural networks in the field of CCP dynamic modeling, offering valuable guidance for its application.

Assembly of low-stiffness parts through admittance control with adaptive stiffness

Existing studies on robot assembly mostly focus on high-stiffness metal parts, which are unsuitable for assembling low-stiffness parts. Additionally, when assembly is achieved by the conventional admittance control method, the number of parameters to be adjusted is significant, and unstable contact force may occur. This study proposes two algorithms, namely admittance control with variable stiffness (ACVS) and admittance control with adaptive stiffness (ACAS), which are suitable for robotic assembly of low-stiffness parts. The proposed ACVS is achieved without a position trajectory, which reduces the number of parameters that need to be adjusted by the user. Furthermore, overshoot is suppressed by calculating the stiffness parameter in real time. The ACAS scheme adjusts the stiffness value in ACVS through a multilayer feedforward neural network; therefore, its overshoot is lower than that of ACVS, and the contact force stability is enhanced. The ACAS scheme thus characterized by a slight overshoot and stable contact force enables stable assembly of low-stiffness parts via the maximum force generated by the robot.

Development of surrogate-optimization models for a novel transcritical power cycle integrated with a small modular reactor

In recent years, various types of surrogate optimization models have been proposed to reduce the computational time and to improve the emulation accuracy. In this study, by leveraging an ANN surrogate model developed earlier, a comprehensive and efficient optimization algorithm is conceived for the global optimal design of an integrated regenerative methanol transcritical cycle. It combines a unique converging/diverging classifier model into the surrogate model to form a surrogate-based model, which significantly improves the prediction accuracy of the objective function. Six binary classifiers are explored and the multi-layer feed-forward (MLF) neural network classifier is selected. In addition, within the five global optimizers being explored, the basin-hopping (BH) and dual-annealing (DA) are selected. The optimal surrogate-based model and global optimizers are then combined to form a unique surrogate-optimizer model. The surrogate-optimizer model is slightly outperformed by the physics-based model in terms of the optimization results, the time consumption of the surrogate-optimizer model during the optimization searching process is 99% less than that of the physics-based model. As the results, the surrogate-optimizer model is slightly outperformed by the physics-based model in terms of the optimization results, where the Levelized Cost of Energy (LCOE) of the Surrogate-DA and Surrogate-BH models are 77.912 and 78.876 $/MWh, respectively, compared to the 77.190 $/MWh of the Baseline model with fairly close penalties between them. In the meantime, the time consumption of the surrogate-optimizer model during the optimization searching process is 99% less than that of the physics-based model.

Research on music signal feature recognition and reproduction technology based on multilayer feedforward neural network

Abstract In this paper, a multi-layer feed-forward neural network is used to construct a Meier spectrogram recognition system. By analyzing the algorithmic role of recurrent neural, the backpropagation algorithm is applied to update the weights in the neural network to obtain the mapping relationship between audio input and output. Combined with the algorithmic formula of the spectrum, the short-time Fourier transform is used to analyze the audio information. By architecting a multilayer feedforward recurrent neural network, the music signals are fused and classified. The cross-entropy loss function is applied to calculate the accuracy of micro and macro averages to improve the accuracy of music signal feature recognition. The results show that the feedforward recurrent neural network has the lowest error rate in different note recognition, and the error rate for “do” is 4%.

3D model of water saturation, effective porosity and volume of shale in upper Assam Shelf, India using multi-attribute regression and cascade-probabilistic neural network

Quantitative interpretation of geophysical data for reservoir identification to study the vertical and horizontal variation of lithofacies, involves integration of multiple disparate datasets. We use a 3D seismic data and two wells (Km and Te) to predict a petrophysical model within the Paleocene to Oligocene formations, and the Precambrian basement in the upper Assam Shelf. The conventional well data were examined in each well to identify reservoirs within the formations. This is followed by a model based seismic inversion of 3D post-stack seismic data resulting in a 3D impedance model. Next, the cascade-probabilistic neural network (CPNN) was trained using the well log derived attributes (target label data) at the collocated seismic observations (observed features). We were able to successfully train the network to predict water saturation, effective porosity, and volume of shale from the inverted seismic traces with the cross-correlation of range 0.96, 0.97, 0.97, respectively. The optimum set of multi-attributes at well-seismic tie locations are extracted from the seismic data and the selected optimum set of attributes are used at all locations in the application of CPNN. CPNN is found to train the target data better than probabilistic neural network and multi-layer feedforward neural network. Hydrocarbon reservoirs are found in shaly sandstones within the Kopili and Sylhet formations of Eocene and Paleocene ages, and Precambrian basement. The hydrocarbon reservoir shows the variation in water saturation 20% to 50%, with the volume of shale of range 40% to 60%, and the effective porosity of range 8% to 30% respectively.

Prediction of compressive strength and feature importance analysis of solid waste alkali-activated cementitious materials based on machine learning

This study examines the analysis and treatment of diverse types of solid waste alkaline-activated cementitious materials utilizing machine learning techniques. To account for the variability of solid waste alkaline-activated cementitious materials, three crucial chemical components (aluminum trioxide, calcium oxide, and silicon dioxide) were chosen as representative features. A total of 274 pertinent datasets were collected and split into training and testing sets with an 80% to 20% ratio for analysis. Six machine learning models, including multilayer feedforward neural network, genetic algorithm neural network, support vector machine, random forest, radial basis function neural network, and long short-term memory network, were employed to construct predictive models. The optimal hyperparameters were identified via grid search. The performance of the models was assessed using the training and testing sets, revealing that all models exhibited strong predictive and generalization capabilities on both sets. Among the models, the support vector machine model achieved the highest performance, yielding an R2 value of 0.9054, MAE of 4.1460, and NRMSE of 0.0997. Through the utilization of SHAP (Shapley Additive Explanations) analysis, the interplay between factors and the significance of features were examined. Calcium oxide, water-to-binder ratio, silicon dioxide, modulus of water glass, and aluminum trioxide were recognized as the primary factors that influence the compressive strength. The research findings offer guidance for optimizing the performance of alkaline-activated cementitious materials made from solid waste.

Applying feature selection and machine learning techniques to estimate the biomass higher heating value

The biomass higher heating value (HHV) is an important thermal property that determines the amount of recoverable energy from agriculture byproducts. Precise laboratory measurement or accurate prediction of the HHV is essential for designing biomass conversion equipment. The current study combines feature selection scenarios and machine learning tools to establish a general model for estimating biomass HHV. Multiple linear regression and Pearson’s correlation coefficients justified that volatile matter, nitrogen, and oxygen content of biomass samples have a slight effect on the HHV and it is better to ignore them during the HHV modeling. Then, the prediction performance of random forest, multilayer and cascade feedforward neural networks, group method of data handling, and least-squares support vector regressor are compared to determine the intelligent estimator with the highest accuracy toward biomass HHV prediction. The ranking test shows that the multilayer perceptron neural network better predicts the HHV of 532 biomass samples than the other intelligent models. This model presents the outstanding absolute average relative error of 2.75% and 3.12% and regression coefficients of 0.9500 and 0.9418 in the learning and testing stages. The model performance is also superior to a recurrent neural network which was recently developed in the literature using the same databank.

Development of Artificial Neural Network-Based Prediction Model for Evaluation of Maxillary Arch Growth in Children with Complete Unilateral Cleft Lip and Palate.

Cleft lip and palate (CLP) are the most common congenital craniofacial deformities that can cause a variety of dental abnormalities in children. The purpose of this study was to predict the maxillary arch growth and to develop a neural network logistic regression model for both UCLP and non-UCLP individuals. This study utilizes a novel method incorporating many approaches, such as the bootstrap method, a multi-layer feed-forward neural network, and ordinal logistic regression. A dataset was created based on the following factors: socio-demographic characteristics such as age and gender, as well as cleft type and category of malocclusion associated with the cleft. Training data were used to create a model, whereas testing data were used to validate it. The study is separated into two phases: phase one involves the use of a multilayer neural network and phase two involves the use of an ordinal logistic regression model to analyze the underlying association between cleft and the factors chosen. The findings of the hybrid technique using ordinal logistic regression are discussed, where category acts as both a dependent variable and as the study's output. The ordinal logistic regression was used to classify the dependent variables into three categories. The suggested technique performs exceptionally well, as evidenced by a Predicted Mean Square Error (PMSE) of 2.03%. The outcome of the study suggests that there is a strong association between gender, age, and cleft. The difference in width and length of the maxillary arch in UCLP is mainly related to the severity of the cleft and facial growth pattern.

Prediction of Petrophysical Properties using Post Stack Seismic Inversion and Geostatistical Techniques over F-3 Block, Netherlands - A Comparative Study

To estimate petrophysical parameters, seismic inversion techniques have been frequently used for estimating attributes like P-impedance, elastic impedance, S-impedance, density, Vp / Vs ratio, and gamma-ray logs from seismic and well-log data. These characteristics enable us to comprehend subsurface lithology for geo-seismic analysis, including its extent and shape. Four different post-stack inversion techniques, including bandlimited inversion (BLI), colored inversion (CI), maximum likelihood sparse spike inversion (MLSSI), and model-based inversion (MBI), have been applied to the post-stack seismic data from the F3 block in the Netherlands in this study. The objective is to compare the efficacy of these inversion methods over F3 block seismic data. For post-stack inversion, the data was inverted into a very high-resolution P-impedance volume. The analysis depicts that all four inversion methods provide mutually consistent subsurface information with marginally better MBI results. Furthermore, geostatistical techniques have been used intensively for further testifying the results obtained from post-stack inversion methods. The geostatistical techniques use seismic data-derived attributes and inverted impedance-derived attributes as input to estimate P-wave velocity, porosity, and density away from the boreholes. Two geostatistical methods namely probabilistic neural network (PNN) and multilayer feed-forward neural network (MLFN) are used for the analysis. Porosity, density, and P-wave velocity have been predicted using both techniques which highlighted different characteristics of the subsurface with very detailed information. The derived results show that reservoir properties have been better estimated with the combination of MBI and PNN techniques for F3 block data over the other methods used in this study.

Deep learning analysis of UPLC-MS/MS-based metabolomics data to predict Alzheimer's disease

ObjectiveMetabolic biomarkers can potentially inform disease progression in Alzheimer's disease (AD). The purpose of this study is to identify and describe a new set of diagnostic biomarkers for developing deep learning (DL) tools to predict AD using Ultra Performance Liquid Chromatography Mass Spectrometry (UPLC-MS/MS)-based metabolomics data. MethodsA total of 177 individuals, including 78 with AD and 99 with cognitive normal (CN), were selected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort along with 150 metabolomic biomarkers. We performed feature selection using the Least Absolute Shrinkage and Selection Operator (LASSO). The H2O DL function was used to build multilayer feedforward neural networks to predict AD. ResultsThe LASSO selected 21 metabolic biomarkers. To develop DL models, the 21 biomarkers identified by LASSO were imported into the H2O package. The data was split into 70% for training and 30% for validation. The best DL model with two layers and 18 neurons achieved an accuracy of 0.881, F1-score of 0.892, and AUC of 0.873. Several metabolomic biomarkers involved in glucose and lipid metabolism, in particular bile acid metabolites, were associated with APOE-ε4 allele and clinical biomarkers (Aβ42, tTau, pTau), cognitive assessments [the Alzheimer's Disease Assessment Scale-cognitive subscale 13 (ADAS13), the Mini-Mental State Examination (MMSE)], and hippocampus volume. ConclusionsThis study identified a new set of diagnostic metabolomic biomarkers for developing DL tools to predict AD. These biomarkers may help with early diagnosis, prognostic risk stratification, and/or early treatment interventions for patients at risk for AD.

Detection of Epilepsy patients using coot optimization based feed forward multilayer neural network

ABSTRACT A familiar nervous system disorder characterised by seizures is called as Epilepsy. It is indeed hard to control the suitable type as an outcome of insufficient EEG information. In order to overcome these issues, a Multilayer Neural Network (MLNN)-based classifier is proposed to recognise if the patients are affected by epileptic disease or not. EEG signal is a contribution, and the input signal is preprocessed using antialiasing filter, finite impulse response, and band pass filter to eradicate unwanted noise present in the signal. After preprocessing, the features extracting process is done, and four extraction techniques are proposed in order to calculate the feature coefficient. The feature extraction outcome is fed into the MLNN classifier to predict the disease. MLNN performs with Coot-Optimization to reduce error and increase prediction accuracy. The future ideal applied in Matlab-software carried out numerous act metrics, and these parameters attained better performance such as accuracy of 96.5%, error of 0.03, precision of 98%, specificity is 97%, sensitivity is 95%, and so on. This displays the effectiveness of the future ideal than existing approaches such as ANN, SVM, KNN and NB. Based on this proposed classification, the epileptic disease prediction can be improved on this technique and can provide a living standard for patients.