Input Nodes Research Articles

Music emotion recognition algorithm based on BP neural network

Music plays a very important role in animation production. Because it could better express the emotion of the character, this paper uses BP neural network to identify the music emotion. This paper first introduced the structure of BP neural network. Then, the parameters and structure of the network were designed according to the category of music emotion. Finally, a three-layer BP neural network with 5 input nodes, 13 hidden layer nodes and 4 output nodes was constructed and applied to music emotion recognition. The recognition accuracy was 85.02%, which basically met the requirements of music emotion recognition and achieves the expected effect.

Designing a Pipeline for Predicting Power Grid Stability with Artificial Neural Network (ANN)

Renewable energy sources are becoming more popular, providing a much-needed alternative to traditional, limited, and climate-unfriendly energy sources. Wireless sensors, cloud computing, cyber security, and wide-area monitoring are basic communication and control technologies for smart grid applications. Design of communication and control architectures for the adoption of smart energy grids for rural loads and distributed energy, including energy storage solutions. In this work, a Machine Learning module called scikit-learn is used for pre-processing of labeled input data by using StandardScaler, KFold for cross-validation, and Confusion matrix for measuring performance. Also, the ML technique uses the binary classification method to divide the ‘stabf’ data into two parts as stable and unstable. Here deep learning-based Artificial Neural Network (ANN) has been used to evaluate the result and to predict new grid data to enhance stability. ANN takes 12 input nodes in the input layer and three hidden layers out of which two hidden layer takes 24 nodes and another one takes 12 nodes and an output layer consisting of a single node. Adam optimizer has been used for model compilation and loss function calculation ‘binary_crossentropy’ is used. Finally, after successful completion of the evaluation process, this model gives a test accuracy result of 98.33%.

Discrete-Time Approach to Operational Scheduling of Treelike Pipelines with Multiple Input and Output Nodes

Discrete-Time Approach to Operational Scheduling of Treelike Pipelines with Multiple Input and Output Nodes

EMI Suppression of a DC–DC Converter Using Predictive Pulsed Compensation

Rising voltages and wide bandgap semiconductors represent the main cause for rising electomagnetic interference (EMI) emissions of dc–dc converters in electric vehicles. Conventional passive filters are the limiting factor for the improvement of the power density and conventional active EMI filters are quite complex. This article presents a method in order to compensate the EMI generated by a dc–dc converter using a simple half-bridge gate-driver IC. The compensation signal is synchronized to the converter's switching events and injected through two capacitors at the converter's input nodes. Unlike analog active EMI filters, this method is able to compensate time delays of the filter's hardware components. This article presents an approach to describe such a compensation system analytically. Therefore, bandwidth and attenuation are determined due to the synchronization error and pulse amplitude. The presented prototype system and the analytical prediction are validated by measurements at the stationary operation of the converter in a CISPR 25 test setup. Finally, the performance of the method during dynamic operation of the converter is evaluated to demonstrate the high EMI suppression effectiveness of pulsed compensation.

Training of sparse and dense deep neural networks: Fewer parameters, same performance.

Deep neural networks can be trained in reciprocal space by acting on the eigenvalues and eigenvectors of suitable transfer operators in direct space. Adjusting the eigenvalues while freezing the eigenvectors yields a substantial compression of the parameter space. This latter scales by definition with the number of computing neurons. The classification scores as measured by the displayed accuracy are, however, inferior to those attained when the learning is carried in direct space for an identical architecture and by employing the full set of trainable parameters (with a quadratic dependence on the size of neighbor layers). In this paper, we propose a variant of the spectral learning method as in Giambagli etal. [Nat. Commun. 12, 1330 (2021)2041-172310.1038/s41467-021-21481-0], which leverages on two sets of eigenvalues for each mapping between adjacent layers. The eigenvalues act as veritable knobs which can be freely tuned so as to (1) enhance, or alternatively silence, the contribution of the input nodes and (2) modulate the excitability of the receiving nodes with a mechanism which we interpret as the artificial analog of the homeostatic plasticity. The number of trainable parameters is still a linear function of the network size, but the performance of the trained device gets much closer to those obtained via conventional algorithms, these latter requiring, however, a considerably heavier computational cost. The residual gap between conventional and spectral trainings can be eventually filled by employing a suitable decomposition for the nontrivial block of the eigenvectors matrix. Each spectral parameter reflects back on the whole set of internode weights, an attribute which we effectively exploit to yield sparse networks with stunning classification abilities as compared to their homologs trained with conventional means.

Hysteresis modeling of antiferroelectric micro-cantilever with anticommutative multilayer perceptrons

The discrete Preisach model is popular to describe the hysteresis of smart materials, such as the piezoelectric and ferroelectric material. Re-explaining the discrete Preisach model of antiferroelectric materials after introducing an extended Preisach function, the special multilayer perceptrons with only two input nodes was designed intrinsically to satisfy the anticommutative property of hysteresis without superfluous hidden nodes (called aMLP), where its output equals to the difference between two predictive results that two signals in the normal and reverse sequence are imported to the aMLP model. Several training algorithms, including the normal/hybrid generic algorithm (GA) and the normal/hybrid particle swarm optimization (PSO), were used to adjust the parameters of the aMLP model, respectively. The comparing results indicate that the aMLP models designed with the winner operation function pair of [tanh, ReLU] and trained by both normal and hybrid PSO algorithms suggest the better performance when applying the working voltages more than 50 V according to the experimental data. It provides another accurate way for an aMLP model to estimate the anticommutative property of hysteresis of smart materials.

Determining the Factors Affecting the Boiling Heat Transfer Coefficient of Sintered Coated Porous Surfaces

The boiling heat transfer performance of porous surfaces greatly depends on the morphological parameters, liquid thermophysical properties, and pool boiling conditions. Hence, to develop a predictive model valid for diverse working fluids, it is necessary to incorporate the effects of the most influential parameters into the architecture of the model. In this regard, two Bayesian optimization algorithms including Gaussian process regression (GPR) and gradient boosting regression trees (GBRT) are used for tuning the hyper-parameters (number of input and dense nodes, number of dense layers, activation function, batch size, Adam decay, and learning rate) of the deep neural network. The optimized model is then employed to perform sensitivity analysis for finding the most influential parameters in the boiling heat transfer assessment of sintered coated porous surfaces on copper substrate subjected to a variety of high- and low-wetting working fluids, including water, dielectric fluids, and refrigerants, under saturated pool boiling conditions and different surface inclination angles of the heater surface. The model with all the surface morphological features, liquid thermophysical properties, and pool boiling testing parameters demonstrates the highest correlation coefficient, R2 = 0.985, for HTC prediction. The superheated wall is noted to have the maximum effect on the predictive accuracy of the boiling heat transfer coefficient. For example, if the wall superheat is dropped from the modeling parameters, the lowest prediction of R2 (0.893) is achieved. The surface morphological features show relatively less influence compared to the liquid thermophysical properties. The proposed methodology is effective in determining the highly influencing surface and liquid parameters for the boiling heat transfer assessment of porous surfaces.

Electricity Theft Detection in Power Consumption Data Based on Adaptive Tuning Recurrent Neural Network

Electricity theft behavior has serious influence on the normal operation of power grid and the economic benefits of power enterprises. Intelligent anti-power-theft algorithm is required for monitoring the power consumption data to recognize electricity power theft. In this paper, an adaptive time-series recurrent neural network (TSRNN) architecture was built up to detect the abnormal users (i.e., the electricity theft users) in time-series data of the power consumption. In fusion with the synthetic minority oversampling technique (SMOTE) algorithm, a batch of virtual abnormal observations were generated as the implementation for training the TSRNN model. The power consumption record was characterized with the sharp data (ARP), the peak data (PEA), and the shoulder data (SHO). In the TSRNN architectural framework, a basic network unit was formed with three input nodes linked to one hidden neuron for extracting data features from the three characteristic variables. For time-series analysis, the TSRNN structure was re-formed by circulating the basic unit. Each hidden node was designed receiving data from both the current input neurons and the time-former neuron, thus to form a combination of network linking weights for adaptive tuning. The optimization of the TSRNN model is to automatically search for the most suitable values of these linking weights driven by the collected and simulated data. The TSRNN model was trained and optimized with a high discriminant accuracy of 95.1%, and evaluated to have 89.3% accuracy. Finally, the optimized TSRNN model was used to predict the 47 real abnormal samples, resulting in having only three samples false predicted. These experimental results indicated that the proposed adaptive TSRNN architecture combined with SMOTE is feasible to identify the abnormal electricity theft behavior. It is prospective to be applied to online monitoring of distributed analysis of large-scale electricity power consumption data.

On the Classification of a Greenhouse Environment for a Rose Crop Based on AI-Based Surrogate Models

A precise microclimate control for dynamic climate changes in greenhouses allows the industry and researchers to develop a simple, robust, reliable, and intelligent model. Accordingly, the objective of this investigation was to develop a method that can accurately define the most suitable environment in the greenhouse for an optimal yield of roses. Herein, an optimal and highly accurate BO-DNN surrogate model was developed (based on 300 experimental data points) for a quick and reliable classification of the rose yield environment considering some of the most influential variables including soil humidity, temperature and humidity of air, CO2 concentration, and light intensity (lux) into its architecture. Initially, two BO techniques (GP and GBRT) are used for the tuning process of the hyper-parameters (such as learning rate, batch size, number of dense nodes, number of dense neurons, number of input nodes, activation function, etc.). After that, an optimal and simple combination of the hyper-parameters was selected to develop a DNN algorithm based on 300 data points, which was further used to classify the rose yield environment (the rose yield environments were classified into four classes such as soil without water, correct environment, too hot, and very cold environments). The very high accuracy of the proposed surrogate model (0.98) originated from the introduction of the most vital soil and meteorological parameters as the inputs of the model. The proposed method can help in identifying intelligent greenhouse environments for efficient crop yields.

Towards Grading Chest X-rays of COVID-19 Patients Using a Dynamic Radial Basis Function Network Classifier

The high volume of COVID-19 Chest X-rays and less number of radiologists to interpret those is a challenge for the highly populous developing nations. Moreover, correct grading of the COVID-19 stage by interpreting the Chest X-rays manually is time-taking and could be biased. It often delays the treatment. Given the scenario, the purpose of this study is to develop a deep learning classifier for multiple classifications (e.g., mild, moderate, and severe grade of involvement) of COVID-19 Chest X-rays for faster and accurate diagnosis. To accomplish the goal, the raw images are denoised with a Gaussian filter during pre-processing followed by the Regions of Interest, and Edge Features are identified using Canny's edge detector algorithm. Standardized Edge Features become the training inputs to a Dynamic Radial Basis Function Network classifier, developed from scratch. Results show that the developed classifier is 88% precise and 86% accurate in classifying the grade of illness with a much faster processing speed. The contribution lies in the dynamic allocation of the (i) number of Input and Hidden nodes as per the shape and size of the image, (ii) Learning rate, (iii) Centroid, (iv) Spread, and (v) Weight values during squared error minimization; (vi) image size reduction (37% on average) by standardization, instead of dimensionality reduction to prevent data loss; and (vii) reducing the time complexity of the classifier by 26% on average. Such a classifier could be a reliable assistive tool to human doctors in screening and grading COVID-19 patients and in turn, would help faster management of the patients as per the stages of COVID-19.

Systematic Approaches for Optimal Scheduling of Straight Pipelines with Batch-Centric Formulations

Scheduling of the transportation of liquids by multiproduct pipelines has been receiving considerable attention in the literature. The preferred methods for solving this problem rely on mathematical programming and use continuous-time formulations that move the virtual entities of batches instead of the real entities of products. Such batch-centric formulations require the user to convert the products initially in the pipeline into batches, which is trivial to do for linear or branched systems with a single input node but not for straight systems with multiple input nodes. While the dependence of solution quality on the initial batch sequence has been recognized early, no systematic procedure covering all possibilities has been presented yet. In this paper, we propose a novel scheduling algorithm that starts generating all possible initial batch sequences with a new planning formulation, before exploring all generated sequences with either a sequential or simultaneous approach, using the scheduling formulation of Liao, Q.; [ AIChE J. 2019, 65, e16712]. We then generalize this formulation to consider automatic batch selection, by including the logical constraints of the planning model.

Hybrid Extreme Learning Machine dan Firefly Algorithm untuk Meramalkan Nilai Tukar Rupiah terhadap Dolar

Every country has a currency as a medium of exchange and the movement of its exchange rate can affect the economy of the country. In Indonesia, since the freely floating exchange rates system has been applied in August 1997, the value of rupiah currency in the foreign exchange market can change at any time. Considering the massive impacts of exchange rate fluctuation on the economy, then forecasting the exchange rate of rupiah against the US dollar is important to help Indonesia’s economic growth. The aims of this thesis is to predict the estimated exchange rate of rupiah against the US dollar in the future by using hybrid artificial neural network extreme learning machine (ELM) method and firefly algorithm (FA). In the training process, ELM-FA hybrid has a role to obtain the best weight and bias. The weight and bias that obtained will be used for forecasting and to know the success rate of the training process, the validation test process is required. Based on the implementation of program and simulation for some parameter values on the exchange rate data from Jan 2015 until Jan 2018, with four input and hidden nodes, and one output node, obtained the smallest MSE of the training is 0.000480513 with MSE of the testing is 0.0000854107. The relatively small MSE value indicates that ELM-FA network is able to recognize the data pattern well and able to predict the test data well.

The Prediction of Chlorophyll Content in African Leaves (Vernonia amygdalina Del.) Using Flatbed Scanner and Optimised Artificial Neural Network

African leaves (Vernonia amygdalina Del.) is a nutrient-rich plant that has been widely used as a herbal plant. African leaves contain chlorophyll which identify compounds produced by a plant, such as flavonoids and phenols. Chlorophyll testing can be carried out non-destructively by using the SPAD 502 chlorophyll meter. However, it is quite expensive, so that another non-destructive method is developed, namely digital image analysis. Relationships between chlorophyll content and leaf image colour indices in the RGB, HSV, HSL, and Lab* space are examined. The objectives of this study are 1) to analyse the relationship between texture parameters of red, green, blue, grey, hue, saturation(HSL), lightness (HSL), saturation( HSV), value(HSV), L*, a*, and b* against the chlorophyll content in African leaves using a flatbed scanner (HP DeskJet 2130 Series); and 2) built a model to predict chlorophyll content in African leaves using optimised ANN through a feature selection process by using several filter methods. The best ANN topologies are 10-30-40-1 (10 input nodes, 40 nodes in hidden layer 1, 30 nodes in hidden layer 2, and 1 output node) with a trainlm on the learning function, tansig on the hidden layer, and purelin on the output layer. The selected topology produces MSE training of 0.0007 with R training 0.9981 and the lowest validation MSE of 0.012 with R validation of 0.967. With these results, it can be concluded that the ANN model can be potentially used as a model for predicting chlorophyll content in African leaves.

CNN‐combined graph residual network with multilevel feature fusion for hyperspectral image classification

The application of graph convolutional networks (GCN) in hyperspectral image (HSI) classification has become a promising method, thanks to its flexible convolution operation in any irregular image region. For the classification of HSI, GCN can extract more superpixel-level features with a topological structure, in comparison to the traditional convolutional neural networks (CNNs) using fixed square kernels distilling pixel-level features. To fully leverage the different levels of features, this study proposes a novel deep network referred to as a CNN-combined graph residual network ( C 2 GRN), which integrates the multilevel graph residual module and spectral-spatial features continuous learning module. During the extraction of topology information using the former module, HSI pixels are divided into superpixels and served as input nodes of the module to reduce the computational complexity and obtain the multilevel spatial relevance between adjacent superpixels. Besides, for the latter module, the spectral-spatial features are learnt continuously, which could obtain the finer pixel-level features. Finally, the captured spectral-spatial features of different levels are concatenated. This strategy could not only adequately utilize the correlation and difference of adjacent spatial but also obtain the finer and more valuable spectral-spatial information, which makes a significant boost in the HSI classification. Additionally, the experiment results demonstrate the superiority and availability of the C 2 GRN on three benchmark datasets of HSI, compared with the state-of-the-art methods for the classification of HSI.

A digital direction of arrival estimator based on fast on-line sequential random vector functional link network

This article develops a fast online sequential random vector functional network. In this network, input nodes are connected to both hidden nodes and output nodes. This architecture results in high-speed learning and testing. The network estimates the direction of arrival from signals received by the sparse array in real-time. Neither noisy signal nor effect of mutual coupling among array elements affect the estimation appreciably using the network. The network shows competitive performance in terms of learning speed, accuracy, short event regression time, simplicity, and robustness in comparison to other state-of-the-art methods such as support vector regression, extreme learning machine, and online sequential extreme learning machine. Finally, validation of the network translation into a hardware platform using a fast digital high-speed reconfigurable Xilinx Virtex-5 field-programmable gate array embedded processor for single-source detection completes the work. Comparison with reported measurement shows results with good accuracy and low root mean square error both by the network and its hardware-implemented platform. The direction of arrival estimation is thus possible using the network either through a computer or dedicated hardware as per necessity.

An Enhanced Input Differential Pair for Low-Voltage Bulk-Driven Amplifiers

This article presents a low-voltage high-transconductance input differential pair for bulk-driven amplifiers. The proposed structure employs two bulk-driven flipped voltage follower (FVF) cells as nonlinear tail current sources to enhance the slewing behavior. This method also increases the transconductance of the proposed amplifier two times against the conventional one. The enhanced topology is merged with a conventional bulk-driven input differential pair using cross-coupled connections to significantly increase the transconductance. These circuitry ideas lead to an improvement in the amplifier's specifications, such as dc gain, slew rate (SR), and input noise without any degeneration in other parameters. Moreover, thanks to the use of the bulk terminals as the input nodes and also a simple common-source structure as the second stage, rail-to-rail input, and output swings are achieved, respectively. The proposed amplifier was fabricated in TSMC 0.18- μm CMOS technology. Under a supply voltage of 0.5 V, the measurement results show that the proposed amplifier achieves a dc gain of 78 dB, a gain bandwidth of 7.5 kHz, and an SR of 8.6 V/ms with just 91-nA current dissipation.

Development and Validation of Artificial Neural Networks for Survival Prediction Model for Patients with Spontaneous Hepatocellular Carcinoma Rupture After Transcatheter Arterial Embolization.

BackgroundSpontaneous rupture bleeding is a fatal hepatocellular carcinoma (HCC) complication and a significant determinant of survival outcomes. This study aimed to develop and validate a novel artificial neural network (ANN)-based survival prediction model for patients with spontaneous HCC rupture after transcatheter arterial embolization (TAE).MethodsPatients with spontaneous HCC rupture bleeding who underwent TAE at our hospital between January 2010 and December 2018 were included in our study. The least absolute shrinkage and selection operator (LASSO) Cox regression model was used to screen clinical variables related to prognosis. We incorporated the above clinical variables identified by LASSO Cox regression into the ANNs model. Multilayer perceptron ANNs were used to develop the 1-year overall survival (OS) prediction model for patients with spontaneous HCC ruptured bleeding in the training set. The area under the receiver operating characteristic curve and decision curve analysis were used to compare the predictive capability of the ANNs model with that of existing conventional prediction models.ResultsThe median survival time for the whole set was 11.8 months, and the 1-year OS rate was 47.5%. LASSO Cox regression revealed that sex, extrahepatic metastasis, macroscopic vascular invasion, tumor number, hepatitis B surface antigen, hepatitis B e antigen, tumor size, alpha-fetoprotein, fibrinogen, direct bilirubin, red blood cell, and γ-glutamyltransferase were risk factors for OS. An ANNs model with 12 input nodes, seven hidden nodes, and two corresponding prognostic outcomes was constructed. In the training set and the validation set, AUCs for the ability of the ANNs model to predict the 1-year OS of patients with spontaneous HCC rupture bleeding were 0.923 (95% CI, 0.890–0.956) and 0.930 (95% CI, 0.875–0.985), respectively, which were higher than that of the existing conventional models (all P < 0.0001).ConclusionThe ANNs model that we established has better survival prediction performance.

Graph optimization neural network with spatio-temporal correlation learning for multi-node offshore wind speed forecasting

Wind speed accurate forecasting plays an important role in preserving the stability of offshore wind power. Most of current predictions are based on a single wind node. It is difficult for this methods to capture wind high dimensional features and latent spatio-temporal dependencies. In this paper, we propose a general graph optimization neural network specifically for multi-node offshore wind speed prediction named spatio-temporal correlation graph neural network. The proposed model firstly employs graph convolution, which performs Laplace transforms instead of the traditional convolution, to capture the potential spatial dependencies from the nodes' relationship and historical time series better. The channel-wise attention makes the original concentrated weights within a region constructed by adjacency matrix disperse to all input nodes, which distinguishes the nodes’ contributions and generates high dimensional spatial features. Long short-term memory is applied to extract temporal correlation from the high dimensional spatial features. The model has ability to excavate the full potential of spatial-temporal dependencies from multiple wind nodes to the utmost. Experiments select 120 wind nodes in the China sea for prediction. The results show that the proposed model can be very competitive with state-of-the-art methods and holds great performance on multi-node and multi-step wind speed forecasting.

Investigation of pedestrian jaywalking behaviour at mid-block locations using artificial neural networks

This research presents a framework based on artificial neural networks (ANNs) to predict jaywalker’s trajectory while crossing the road. In this process, different causal and conditional variables related to jaywalking, such as, gender, direction of crossing, walking or running, cell phone use, roadway lane number, etc. are taken as input variables. The dataset comprises of 2504 samples which is collected under non-lane based heterogeneous traffic conditions. Through testing predictive performance of several ANN architectures based on correlation co-efficient and mean square error, the best ANN architecture to predict jaywalker’s movement is determined. The optimal ANN architecture comprises of 9 input nodes, 10 hidden nodes and 1 output node. This study also determines the microscopic variables, i.e., speed, flow and density, associated with jaywalking. The results indicate that the average speed of male jaywalkers is higher than female jaywalkers. The density is found to be higher in the last lanes of the crossing paths. In addition, this research develops jaywalker trajectories in order to understand their position shifting strategies. Findings suggest that ‘median to sidewalk’ movement trajectories tend to move closer to the foot-over bridge, whereas, the ‘sidewalk to median’ movement trajectories tend to move further away from foot-over bridge. The outcome of this research is expected to assist driver assistance technology as well as Connected and Autonomous Vehicle (CAV) technologies by allowing vehicles to safely navigate through both clustered and individual jaywalkers.

A Multiple Regression Convolutional Neural Network for Estimating Multi-Parameters Based on Overall Data in the Inverse Heat Transfer Problem

Abstract The inverse heat transfer problem (IHTP) is a central task for estimating parameters in heat transfer. It is ill-posedness that is characterized by instability and non-uniqueness of the solution. Recently, novel algorithms using deep learning (DL) and neural networks have been applied to various sparse data in the IHTP. In order to overcome the optimization problem of input nodes under sparse data, we try to use the overall data of the target as the basis for inversion. In this work, we used a multiple regression convolutional neural network (MRCNN) to estimate multi-parameters in the IHTP. Computational fluid dynamics and DL are fused to provide datasets for training of the proposed model. The proposed model was verified by experiments with a cubic cavity. Additionally, the MRCNN model was used to predict the different parameters of the more complex armored vehicle model. The results showed that the model has good prediction accuracy for estimating multi-parameters on different datasets. These attempts of introducing convolutional neural network to the IHTP in the present study were successful and it was significant for the study of the IHTP of estimating multi-parameters.