Multilayer Feedforward Network Research Articles

Presentation of artificial neural network models based on optimum theories for predicting accident severity on rural roads in Iran

This study focuses on predicting the number and severity of rural accidents using nine accident-related variables by the multi-layer perceptron (MLP) approach as a type of artificial neural network modeling method. The models were developed using the input parameters of shoulder width, roadway width, roadside hazard, access density, passing zone ratio, speed limit, pavement condition index, shoulder rumble strips, and centerline rumble strips. MLP models were created based on accident data that occurred from 2019 to 2020 on the Tehran–Qom and Tehran-Saveh rural roads in Iran. In order to achieve the highest accuracy, twenty MLP models have been built with various structures. The MLP model, consisting of a multilayer feedforward network with hidden sigmoid and softmax output, has been used in this study. In this regard, an attempt was made to present an optimum theory to select the best model. The most accurate model was chosen based on the R-value and root mean square error (RMSE), mean absolute error (MAE), f, SD, and R2. Results indicated that the R-value obtained from the optimum model was 0.912, representing the accurate performance of the selected model. In addition, access density, roadside hazard, and roadway width were identified as the most significant variables.

A novel model for efficient cluster head selection in mobile WSNs using residual energy and neural networks

Wireless Sensor Networks (WSNs) are essential for monitoring operational environments. Efficient selection of cluster heads is crucial in minimizing energy consumption. This study introduces an innovative model for determining the positions of cluster heads in a WSN, using the residual energy of each sensor node as a selection metric. The sensor incorporates a neural network model that has been trained using different significant models to determine the best location for the cluster head. Our model is compared with the Low Energy Adaptive Clustering Hierarchy (LEACH) algorithm to showcase its effectiveness, particularly in mobile node scenarios. We employ a feed-forward multilayer network, specifically Multilayer Perceptrons (MLPs), as the neural network technique. The proposed method comprises two stages: the training stage, where the neural network is trained with appropriate models, and the execution stage, where the trained network suggests the cluster head's location based on the sensor's conditions. The results demonstrate that the proposed method can accurately identify suitable locations for cluster heads in a sensor network and is capable of adapting to new models despite environmental changes and variations in input configurations.

Scheduling for the Flexible Job-Shop Problem with a Dynamic Number of Machines Using Deep Reinforcement Learning

The dynamic flexible job-shop problem (DFJSP) is a realistic and challenging problem that many production plants face. As the product line becomes more complex, the machines may suddenly break down or resume service, so we need a dynamic scheduling framework to cope with the changing number of machines over time. This issue has been rarely addressed in the literature. In this paper, we propose an improved learning-to-dispatch (L2D) model to generate a reasonable and good schedule to minimize the makespan. We formulate a DFJSP as a disjunctive graph and use graph neural networks (GINs) to embed the disjunctive graph into states for the agent to learn. The use of GINs enables the model to handle the dynamic number of machines and to effectively generalize to large-scale instances. The learning agent is a multi-layer feedforward network trained with a reinforcement learning algorithm, called proximal policy optimization. We trained the model on small-sized problems and tested it on various-sized problems. The experimental results show that our model outperforms the existing best priority dispatching rule algorithms, such as shortest processing time, most work remaining, flow due date per most work remaining, and most operations remaining. The results verify that the model has a good generalization capability and, thus, demonstrate its effectiveness.

Predicting the Power Requirement of Agricultural Machinery Using ANN and Regression Models and the Optimization of Parameters Using an ANN–PSO Technique

Optimizing the design and operational parameters for tillage tools is crucial for improved performance. Recently, artificial intelligence approaches, like ANN with learning capabilities, have gained attention for cost-effective and timely problem solving. Soil-bin experiments were conducted and data were used to develop ANN and regression models using gang angle, velocity ratio, soil CI, and depth as input parameters, while tractor equivalent PTO (PTOeq) power was used as an output. Both models were trained with a randomly selected 90% of the data, reserving 10% for testing purposes. In regression, models were iteratively fitted using nonlinear least-squares optimization. The ANN model utilized a multilayer feed-forward network with a backpropagation algorithm. The comparative performance of both models was evaluated in terms of R2 and mean square error (MSE). The ANN model outperformed the regression model in the training, testing, and validation phases. A well-trained ANN model was integrated with the particle-swarm optimization (PSO) technique for optimization of the operational parameters. The optimized configuration featured a 36.6° gang angle, 0.50 MPa CI, 100 mm depth, and 3.90 velocity ratio for a predicted tractor PTOeq power of 3.36 kW against an actual value of 3.45 kW. ANN–PSO predicted the optimal parameters with a variation between the predicted and the actual tractor PTOeq power within ±6.85%.

Approximation and interpolation with neural network

In this paper we show that multilayer feedforward networks with one single hidden layer.and certain types of activation functions can approximate univariant continuous functions defined on a compact set. arbitrarily well. In particular, our results contain some usual activation functions such as sigmoidal functions, RELU functions and threshold functions. Besides, since interpolation problems are highly related to approximation problem, we demonstrate that a wide range of functions have the ability to interpolate and generalize our results to functions which are not polynomial on R. Compared to existing results by numerous work, our methods are more intuitive and less technical. Lastly, the paper discusses the possibility of combining interpolation property and approximating property together, and demonstrates that given any Riemann integrable functions on a compact set in R, with several points on its graph, the finite combination of monotone sigmoidal functions can pass through these points and approximate the given function arbitrarily well with respect to L^1 (dx) (in the sense of Riemann integral) when the number of points getting large.

FLY-CAPS- A Hybrid Firefly Feature Optimized Capsule Networks for Plant Disease Classification in Resource Constriant Internet of Things (IoT)

Recent advancements in artificial intelligence, automation, and the Internet of Things (IoT) enable farmers to better monitor and diagnose all agricultural procedures with super-intellectual accuracy. These technologies also contribute to boosting the productivity of agriculture, which increases the country’s economy. Though these technologies help farmers increase productivity, the detection of plant diseases still needs heightened scrutiny for prevention and cultivation. Plant disease categorization has expanded with the introduction of deep learning algorithms, but it still needs more innovation in terms of accuracy and computing burden. Thus, a novel deep learning model based on capsule networks with firefly optimization and potent multi-layered feedforward prediction networks is proposed in this research. The handcrafted features in this proposed system are optimized before being extracted using a capsule network, which reduces the complexity overhead and is suitable for IoT devices with limited resources. Finally fed to the feed forward layers for better classification. The extensive experimentation has been tested with the Plant Village databases, which contain more than 50,000 images of healthy and infected plants. Performance criteria including recall, specificity, recall, accuracy, and f1-score are used to assess the proposed algorithm's performance. Additionally, its efficiency and computational cost are contrasted with those of other recent models. The suggested model has greater performance (95%) with reduced computing overhead, according to experimental data, which is advantageous for the new prediction approach and the welfare of the farmer.

ARTIFICIAL INTELLIGENCE APPLICATION IN PHOTOACOUSTIC OF GASES

Photoacoustic spectroscopy is a powerful, non-destructive, ultrasensitive technique that covers a wide range of applications including atmospheric monitoring, industrial, environmental, and biomedical practice. In this paper, our attention is focused on the area of artificial intelligence implementation in photoacoustic spectroscopy of gases. Artificial intelligence has been proven as a very successful, effective, and promising method for the accurate and real-time determination of photoacoustic signal parameters, related to relaxation, thermal and other physical properties of various media (i.e., for solving the inverse photoacoustic problem). To improve the sensitivity and selectivity of the photoacoustic method feedforward multilayer perceptron network is applied for real-time simultaneous determination of photoacoustic signal parameters: vibrational-to-translational relaxation time, and radius of the laser beam. Also, to solve the problem of finding optimal values of these photoacoustic parameters, metaheuristic algorithms, genetic algorithms and simulated annealing are used. The performance of artificial intelligence methods has been tested on a set of experimental signals generated in the (SF6+Ar) gas. The potential advantages and disadvantages of those methods are discussed.

Can Neural Quantum States Learn Volume-Law Ground States?

We study whether neural quantum states based on multilayer feed-forward networks can find ground states which exhibit volume-law entanglement entropy. As a testbed, we employ the paradigmatic Sachdev-Ye-Kitaev model. We find that both shallow and deep feed-forward networks require an exponential number of parameters in order to represent the ground state of this model. This demonstrates that sufficiently complicated quantum states, although being physical solutions to relevant models and not pathological cases, can still be difficult to learn to the point of intractability at larger system sizes. Hence, the variational neural network approach offers no benefits over exact diagonalization methods in this case. This highlights the importance of further investigations into the physical properties of quantum states amenable to an efficient neural representation.

Machine Learning and Artificial Intelligence Techniques for Detecting Driver Drowsiness

The number of vehicles on the road rises in tandem with the development of vehicle manufacture. With the rapid growth of automobiles, the number of road accidents appears to be steadily increasing. Accidents are a typical occurrence in our daily lives. Road accidents are the top cause of death from injuries worldwide in the top 10. It now forms an amazingly important part of the global burden of illness. An estimated 1.2 million people are killed in accidents every year, and an estimated 50 million are injured while receiving bone marrow transplants in leading countries. Drowsiness, along with fatigue of drivers, are some of the major causes of road accidents. Every year, we see an increase in death numbers worldwide. This study also looks into automatically detecting driver drowsiness with AI, visuals, and CNN. The Driver Drowsiness System works on the concept of Multi-Layer Feed Forward Network, especially on Convolutional Neural Networks (CNN). CNN was developed with the help of around 7000 images of eyes in both drowsiness and non-drowsiness states with different facial architectures. These images were split into training (80 percent of images) and testing (20 percent of images) datasets. The images under the training dataset are given as the input for the network for training purposes. Backpropagation algorithms and optimizers are used to reduce the loss as much as possible. We created an algorithm for detecting, tracking, and analysing motor and visual effects to measure ROI.

Neuronal morphology and network properties modulate signal propagation in multi-layer feedforward network

The propagation and detection of weak signals play a vital role in the central nervous system's information processing. In this paper, a biophysical two-compartment model is adopted to investigate how the neuronal morphology and network properties modulate signal propagation in a multi-layer feedforward network (FFN). The numerical simulation results show that neurons with larger dendrites have higher firing rates and better responses to weak signals. Similarly, the output layer of FFN constructed by larger-dendrite neurons also exhibits better responses. A suitable chaotic current is necessary for the propagation of weak signals. Excessively strong or weak chaotic current leads to propagation failure. Sparse connection and weak synaptic strength optimize the responses of the output layer, which is consistent with real biological networks observed in the brain. It is found that weak signal propagation in FFN is highly correlated with the regulation of firing rate. Our results may provide novel insights into the modeling of complex networks and network function implementation.

Comparison of Linear Regression and Artificial Neural Network Models for the Dimensional Control of the Welded Stamped Steel Arms

The production of parts by pressing and subsequent welding is commonly used in the automotive industry. The disadvantage of this method of production is that inaccuracies arising during pressing significantly affect the final dimension of the part. However, this can be corrected by the choice of the technological parameters of the following operation—welding. Suitably designed parameters make it possible to partially eliminate inaccuracies arising during pressing and thus increase the overall applicability of this technology. The paper is focused on the upper arm geometry of a car produced in this manner. There have been two neural networks proposed in which the optimal welding parameters are determined based on the stamped dimensions and the desired final dimensions. The Levenberg–Marquardt back-propagation algorithm and the Bayesian regularised back-propagation algorithm were used as the learning algorithm for ANNs in multi-layer feed-forward networks. The outputs obtained from the neural networks were compared with a linear prediction model based on a on the design of experiment methodology. The mean absolute percentage error of the linear regression model on the entire dataset was 3 × 10−3%. A neural network with Levenberg–Marquardt back-propagation learning algorithm had a mean absolute percentage error of 4 × 10−3. Similarly, a neural network with a Bayesian regularised back-propagation learning algorithm had a mean absolute percentage error of 3 × 10−3%.

Prediction of secondary components in sugarcane brandy by application of artificial neural network

This study aimed to use artificial neural networks to predict the secondary components of sugarcane brandy. We got data on the characteristics of sugarcane brandy from the literature. We divided this data into input data and output data. Secondary components in sugarcane brandy were the output data. The architecture used for artificial neural networking was the multilayer feed-forward network, which features a hidden layer. These hidden neurons have the role of intervening between the input and output layers of the network. We separated the data into 70% for training and 30% for tests. The artificial neural network transfer function was Relu, with Adam as the training algorithm for weight change with a constant learning rate. The number of neurons in the hidden layer was determined using the mean square error. Twenty neurons in the hidden layer were analyzed and considered appropriate, since in the artificial neural network with a greater number of neurons, we observed no significant variation in the reduction of error.

Artificial neural network model for strength predictions of CFST columns strengthened with CFRP

This paper presents an optimised Artificial Neural Network (ANN) model for predicting the ultimate axial strengths of concentrically loaded Concrete-Filled Steel Tubular (CFST) short and slender columns strengthened with Carbon Fibre-Reinforced Polymer (CFRP). Since experimental data on CFRP strengthened CFST columns is limited, an accurate Finite Element (FE) model is developed and used to provide additional numerical data. A multi-layered feed-forward back-propagation network is proposed, optimised, and trained using the results of 76 experimental tests and 450 generated FE models. The accuracy of the ANN model is assessed through comparing its computed results with experimental data. A reliability analysis is performed using Monte Carlo Simulation (MCS) to evaluate the safety of the solutions computed by the ANN model. ANN-based equations and Graphical User Interface (GUI) are developed based on the trained ANN model for the determination of the ultimate axial strengths of CFST columns. The results show that the developed ANN model is capable of accurately predicting the ultimate axial strengths of CFRP strengthened CFST columns with a high degree of accuracy.

Fault Location Estimator Design for Power Distribution System Using Artificial Neural Network

Fault location in distribution system is critical issue to increase the availability of power supply by reducing the time of interruption for maintenance in electric utility companies. Fault location estimator for power distribution system using artificial neural network is developed for line to ground, line to line, line to line to ground and three phases to ground faults in distribution system. To develop this estimator is one of rural radial power distribution feeder in Ethiopia, south west reign, Aba substation Tarcha line feeder is used as a test feeder. This feeder is simulated using ETAP software to generate data for different fault condition, with different fault resistance and loading conditions, which is the fault phase voltage and current. The generated data is preprocessed and put as an input for neural network to be trained. MATLAB R2016a neural network toolbox to train ANN and programming toolbox is used to develop graphic user interface for fault estimator. The feed forward multi-layer network topologies of neural network with improved back propagation, Liebenberg Marquardt learning algorithm is used to train the network. After the network is trained the mean square error performance, regression plot and error histogram analysis was made and found to have an excellent performance with regression coefficient 0.99929 , validation performance of 0.000102 and error histogram range 0.015 to 0.019. In this thesis for practical implementation the fault records at the test feeder is handled by intelligent electronic device (IED) installed at the substation feeders. The fault record of IED can be read by PCM600 tool using laptop or manually using IEDs human machine interface, this fault recorded data feed to the graphic user interface to estimate the fault location as well as the fault type. Finally it is found that artificial neural networks are one of the alternate options in fault estimator design for distribution system where sufficient distribution network data are available with narrow fault location distance range from the substation. This has benefits in assisting for maintenance plan, saving efforts in fault location finding and economic benefits by reducing interruption time. Keywords : Power distribution, artificial intelligence, neural network, feed forward network DOI: 10.7176/JETP/12-5-01 Publication date: November 30 th 2022

Improving Returns on Strategy Decisions through Integration of Neural Networks for the Valuation of Asset Pricing: The Case of Taiwanese Stock

Most of the growth forecasts in analysts’ evaluation reports rely on human judgment, which leads to the occurrence of bias. A back-propagation neural network (BPNN) is a financial technique that learns a multi-layer feedforward network. This study aims to integrate BPNN and asset pricing models to avoid artificial forecasting errors. In terms of evaluation, financial statements and investor attention were used in this case study, demonstrating that modern analysts should incorporate the evaluation advantages of big data to provide more reasonable and rational investment reports. We found that assessments of revenue, index returns, and investor attention suggest that stock prices are prone to undervaluation The levels of risk-taking behaviors were used in the classification of robustness analysis. This study showed that when betas range from 1% to 5%, both risk-taking levels of investors can hold buying strategies for the long term. However, for lower risk-taking preferences, only when the change exceeds 10 percent, the stock price is prone to overvaluation, indicating that investors can sell or adopt a more cautious investment strategy.

Temperature-controlled propagation of spikes in neuronal networks

Temperature plays a vital role in the functioning of biological organisms and there often exists an optimal temperature for their best performance. In this work, we investigate the role of temperature on spike propagation in scale-free and small-world neuronal networks, where a single neuron is chosen randomly for receiving a stimulus current. Upon exploiting the dominant phase-advanced driving (DPAD) method, the complex neuronal network is seen as a regular feed-forward multilayer neuronal network. The propagation route is then clearly identified, and many traveling-like waves are formed along the propagation route. Interestingly, we find that temperature not only controls the shortest path of propagation but also regulates the response time of a single neuron. The propagation speed is also maximized for an optimal choice of temperature at which the spike rapidly propagates through the entire neuronal network. Our findings extend the current understanding of the neuronal networks functioning and provide new insights into the existence of an optimal temperature as seen in our experiments on several living biological systems.

Deep multimodal predictome for studying mental disorders.

Characterizing neuropsychiatric disorders is challenging due to heterogeneity in the population. We propose combining structural and functional neuroimaging and genomic data in a multimodal classification framework to leverage their complementary information. Our objectives are two-fold (i) to improve the classification of disorders and (ii) to introspect the concepts learned to explore underlying neural and biological mechanisms linked to mental disorders. Previous multimodal studies have focused on naïve neural networks, mostly perceptron, to learn modality-wise features and often assume equal contribution from each modality. Our focus is on the development of neural networks for feature learning and implementing an adaptive control unit for the fusion phase. Our mid fusion with attention model includes a multilayer feed-forward network, an autoencoder, a bi-directional long short-term memory unit with attention as the features extractor, and a linear attention module for controlling modality-specific influence. The proposed model acquired 92% (p< .0001) accuracy in schizophrenia prediction, outperforming several other state-of-the-art models applied to unimodal or multimodal data. Post hoc feature analyses uncovered critical neural features and genes/biological pathways associated with schizophrenia. The proposed model effectively combines multimodal neuroimaging and genomics data for predicting mental disorders. Interpreting salient features identified by the model may advance our understanding of their underlying etiological mechanisms.

Minimum Functional Length Analysis of K-Mer Based on BPNN.

BP neural network (BPNN), as a multilayer feed-forward network, can realize the deep cognition to target data and high accuracy to output results. However, there were still no related research of k-mer based on BPNN yet. In present study, BPNN was used to train and test binary classification data of each classification mode respectively. All k-mer were divided into two categories according to the X + Y content or completely random mode. Results showed that 1) For classification mode of X + Y content, the accuracy of k-mers classification was 100 percent, no matter k ≤ 6 or k ≥ 7; 2) For completely random classification mode, the accuracy of classification is 100 percent for k-mers of k ≤ 6; But for k-mers of k ≥ 7, the accuracy is less than 100 percent, and with the increase of k value, the accuracy of classification gradually decreases (gradually approaches 50 percent). The k-mers of k ≥ 7 should be the basic functional fragment of nucleic acid, and perform basic nucleic acid function in the DNA sequence. The k-mers of k ≤ 6 should be the basic component fragment of nucleic acid, and no longer perform basic nucleic acid function.

Modelling of urea hydrolysis kinetics using genetic algorithm coupled artificial neural networks in urease immobilized magnetite nanoparticles.

The presence of urea in runoff from fertilized soil could be contributing to the growth of dangerous blooms. Enzymatic urea hydrolysis is a well-known outstanding process that, when integrated with nanotechnology, would be much more efficient. This research provides a novel perspective on magnetic nanobiocatalysts that reduce diffusion barriers in effective urea hydrolysis. Surprisingly, the model developed with the use of a Genetic Algorithm (GA) and an Artificial Neural Network (ANN) demonstrated that the system's diffusion restrictions were reduced. In order to forecast accurate outputs using artificial intelligence (AI), a neural network with one hidden layer and 20 neurons was built utilizing multilayer feed-forward network and showed highest output (diffusion co-efficient) with least mean square error (MSE). The diffusion coefficients of free urease, urease immobilized onto porous MNs (U-aMNs), and nanobiocatalyst, i.e. urease immobilized onto surface modified MNs (U-MNβ), were 1.9×10-17, 12.62×10-16, and 15.48×10-16cm2/min, respectively. These results revealed that the addition of Chitosan to the surface of MNs had a considerable impact on enzyme dispersion. The decrease in Damkohler number (Da) from 2.37±0.26 for U-aMNs to 2.19±0.11 for U-MNβ suggested a beneficial effect in overcoming diffusion constraints. Pseudo-first order and pseudo-second order models were used to analyze urea uptake kinetics, with the former model offering the best fit to the system, with R2 values that were much closer to unity.

Artificial Neural Networking (ANN) Model for Drag Coefficient Optimization for Various Obstacles

For various obstacles in the path of a flowing liquid stream, an artificial neural networking (ANN) model is constructed to study the hydrodynamic force depending on the object. The multilayer perceptron (MLP), back propagation (BP), and feed-forward (FF) network models were employed to create the ANN model, which has a high prediction accuracy and a strong structure. To be more specific, circular-, octagon-, hexagon-, square-, and triangular-shaped cylinders are installed in a rectangular channel. The fluid is flowing from the left wall of the channel by following two velocity profiles explicitly linear velocity and parabolic velocity. The no-slip condition is maintained on the channel upper and bottom walls. The Neumann condition is applied to the outlet. The entire physical design is mathematically regulated using flow equations. The result is presented using the finite element approach, with the LBB-stable finite element pair and a hybrid meshing scheme. The drag coefficient values are calculated by doing line integration around installed obstructions for both linear and parabolic profiles. The values of the drag coefficient are predicted with high accuracy by developing an ANN model toward various obstacles.