Layer Back Propagation Neural Network Research Articles

Power Transformer Fault Diagnosis Based on Random Forest and Improved Particle Swarm Optimization–Backpropagation–AdaBoost

This paper proposes a novel fault diagnosis methodology for oil-immersed transformers to improve the diagnostic accuracy influenced by gas components in power transformer oil. Firstly, the Random Forest (RF) algorithm is utilized to evaluate and filter the raw data features, solving the problem of determining significant features in the dataset. Secondly, a multi-strategy Improved Particle Swarm Optimization (IPSO) is applied to optimize a double-hidden layer backpropagation neural network (BPNN), which overcomes the challenge of determining hyperparameters in the model. Four enhancement strategies, including SPM chaos mapping based on opposition-based learning, adaptive weight, spiral flight search, and crisscross strategies, are introduced based on traditional Particle Swarm Optimization (PSO) to enhance the model’s optimization capabilities. Lastly, AdaBoost is integrated to fortify the resilience of the IPSO-BP network. Ablation experiments demonstrate an enhanced convergence rate and model accuracy of IPSO. Case analysis using Dissolved Gas Analysis (DGA) samples compares the proposed IPSO–BP–AdaBoost model with other swarm intelligence optimization algorithms integrated with BPNN. The experimental findings highlight the superior diagnostic accuracy and classification performance of the IPSO–BP–AdaBoost model.

Transistor modeling based on LM‐BPNN and CG‐BPNN for the GaAs pHEMT

AbstractIn order to address the challenges of complex process and low precision in traditional device modeling, double hidden layer back propagation neural network (BPNN) are trained using the conjugate gradient (CG) algorithm and the Levenberg–Marquardt (LM) algorithm, the CG‐BPNN and LM‐BPNN models of small signal for gallium arsenide (GaAs) pseudomorphic high electron mobility transistor (pHEMT) are obtained and analyzed here. At first, the scattering parameters (S‐parameters) of GaAs pHEMT are divided into training set and test set randomly. Experimental results show that the CG‐BPNN model is better than another S‐parameters when predicting ImS12 with mean square error (MSE) of 7.6632e‐06, while LM‐BPNN model predicts ImS12 with MSE of 2.4672e‐06. Meanwhile, the MSE of CG‐BPNN model is higher than LM‐BPNN model when predicting all the S‐parameters. In addition, it shows a smaller fluctuation range for the error curve of LM‐BPNN model, which is more stable than the CG‐BPNN model. Therefore, the double hidden layer LM‐BPNN model is the better choice to characterize the small signal of GaAs pHEMT.

Application of back propagation neural network in the analysis of isothermal elastohydrodynamic lubrication

Developing stable and high precision solution methods has always been the most important research topic in the field of elastohydrodynamic lubrication (EHL). The emergence of artificial neural network(ANN) provide new ideas for the research of EHL solving methods. In this article, a model based on back propagation neural network (BPNN) is proposed to obtain the film thickness and pressure distribution of EHL. The model is established to predict the EHL characteristics with the data obtained from numerical calculation, and more numerical data are used to validate the results predicted by the ANN model. The orthogonal experimental method is utilized to tune model’s parameters. In addition, the method of increasing hidden layers is used to enhance the predictive ability of model. Finally, a double layer BPNN is completed that could predict oil pressure and film thickness accurately. And the exploratory research presented in this paper will provide a simple method for predicting lubrication behavior in industrial applications.

A data-driven model to determine the infiltration characteristics of air curtains at building entrances

The air curtain reduces heat exchange between the two sides by creating a virtual partition and works as a solution for improving building sealing and energy efficiency. Currently, the analytical numerical coupling method has achieved some success in describing the low-order theoretical descriptions of air curtain flow, but its application scope is limited. This paper introduces a data-driven model (DDM) to predict the operation state of the air curtain and the volume flow rate through the entrance. A computational fluid dynamics model is built to generate the dataset, which is validated by comparing velocity and volume flow rate with the published data in the literature. Three of the widely used algorithms are tested: support vector machine, random forest, and backpropagation neural network (BPNN). The main conclusions are as follows: (1) The combination of pressure difference and air supply velocity can quickly determine the operation state of the air curtain in the scene (f1-score = 0.9). (2) A single hidden layer BPNN can achieve high-precision prediction of volume flow rate (R2 = 0.92). (3) Compared to theoretical methods, the DDM can retain three-dimensional characteristics of the jet and capture additional details. The approach proposed in this paper can be applied to practical environments to rapidly and accurately optimize the operating parameters of air curtains.

Road Performance and Ice‐Melting Characteristics of Steel Wool Asphalt Mixture

Elastic pavements, low‐freezing point pavements, and energy exchange pavements meet the function of road ice and snow melting but may damage the pavement performance. To solve the above problems, this study intends to develop a noncontact snow‐melting pavement by using induction heating technology. The steel wool asphalt mixture was prepared, and the effects of steel wool on road performances including the high‐temperature performance, low‐temperature performance, water stability, and fatigue performance were studied. The influence of the asphalt mixture’s factors, environmental factors, and human factors on the ice‐melting characteristics of steel wool asphalt mixture was analyzed. A prediction model of the ice‐melting rate of steel wool asphalt mixture based on a double‐hidden layer backpropagation (BP) neural network was established. The results show that the road performance of the asphalt mixture mixed with steel wool mostly meets the requirements of the specification. The increase in steel wool content and length can improve the road performance to a certain extent. In terms of ice‐melting performance, the content and length of steel wool have a beneficial effect on the ice‐melting rate. The lower the ambient temperature or the thicker the ice layer, the slower the ice‐melting rate. The greater the heating intensity, the faster the ice‐melting rate. The average absolute error of the melting ice prediction model is 0.016, so the melting ice effect can be well predicted by this model.

Approach for Extreme Learning Machine-Based Microwave Power Device Modeling

The nonlinear representation of active devices plays an important role in microwave circuit design. Whereas, it takes a long time to extract a large amount of large signal data, and the problem of memory resource and CPU occupancy becomes significant. In order to address the problems in traditional large-signal modeling methods, in this paper an X-parameter modeling method for microwave power devices based on extreme learning machine (ELM) is proposed. To demonstrate the effectiveness of this method, a double layer back propagation (BP) neural network model is established. Then, harmonic balance simulations are used to verify the accuracy of these two models. After comparisons, it is proved that the three harmonic errors of double layer BP neural network model are 9.525dBm, 1.309dBm and 14.593dBm, respectively, and the three harmonic errors of ELM model are 0.673 dBm, 0.314 dBm, 3.09 dBm, respectively. Furthermore, the three harmonic modulus errors of double layer BP neural network model are 0.031, 0.002, 7.665e-4, respectively, and the errors of ELM model are 0.005, 0.001, 8.38e-5, respectively. Finally, in order to verify the accuracy of the predicted model in circuit design, the predicted X-parameter is used in the design of power amplifier. Moreover, the errors of the double layer BP neural network prediction model at 2.5 GHz, 5 GHz and 7.5 GHz are 1.142 dBm, 1.436 dBm and 2.294 dBm, respectively. The output power error of the ELM model at 2.5 GHz, 5 GHz and 7.5 GHz are 0.089 dBm, 0.311 dBm and 0.309 dBm, respectively. These experimental results demonstrate that the established ELM model is an efficient and valid approach for modeling GaN high electron mobility transistor types of nonlinear microwave devices.

Optimization of Copper Electroforming Process Parameters Based on Double Hidden Layer BP Neural Network.

In order to optimize the pulse electroforming copper process, a double hidden layer BP (back propagation) neural network was constructed. Through sample training, the mapping relationship between electroforming copper process conditions and target properties was accurately established, and the prediction of microhardness and tensile strength of the electroforming layer in the pulse electroforming copper process was realized. The predicted results were verified by electrodeposition copper test in copper pyrophosphate solution system with pulse power supply. The results show that the microhardness and tensile strength of copper layer predicted by “3-4-3-2” structure double hidden layer neural network are very close to the experimental values, and the relative error is less than 2.82%. In the parameter range, the microhardness of copper layer is between 100.3~205.6 MPa and the tensile strength is between 165~485 MPa. When the microhardness and tensile strength are optimal, the corresponding range of optimal parameters are as follows: current density is 2–3 A·dm−2, pulse frequency is 1.5–2 kHz and pulse duty cycle is 10–20%.

Drilling Efficiency Improvement and Rate of Penetration Optimization by Machine Learning and Data Analytics

Rate of Penetration (ROP) is one of the important factors influencing the drilling efficiency. Since cost recovery is an important bottom line in the drilling industry, optimizing ROP is essential to minimize the drilling operational cost and capital cost. Traditional the empirical models are not adaptive to new lithology changes and hence the predictive accuracy is low and subjective. With advancement in big data technologies, real- time data storage cost is lowered, and the availability of real-time data is enhanced. In this study, it is shown that optimization methods together with data models has immense potential in predicting ROP based on real time measurements on the rig. A machine learning based data model is developed by utilizing the offset vertical wells’ real time operational parameters while drilling. Data pre-processing methods and feature engineering methods modify the raw data into a processed data so that the model learns effectively from the inputs. A multi – layer back propagation neural network is developed, cross-validated and compared with field measurements and empirical models.

Modeling Height–Diameter Relationship for Poplar Plantations Using Combined-Optimization Multiple Hidden Layer Back Propagation Neural Network

Relationship of total height and diameter at breast height (hereafter diameter) of the trees is generally nonlinear, and therefore has complex characteristics, which can be accurately described by the height-diameter model developed using the back propagation (BP) neural network approach. The multiple hidden layered-BP neural network has several hidden layers and neurons, and is therefore considered more appropriate modeling approach compared to the single hidden layered-BP neural network approach. However, the former approach is not widely applied for tree height prediction due to absence of the effective optimization method, but it can be done using the BP neural network modeling approach. The poplar (Populus spp. L.) plantation data acquired from Guangdong province of China were used for evaluating the BP neural network modeling approach and compared its results with those obtained from the traditional regression modeling and mixed-effects modeling approaches. We determined the best BP neural network structure with two hidden layers and five neurons in each layer, and logistic sigmoid transfer functions. Relative to the Mitscherlich height-diameter model that had the highest fitting precision among the six traditional height-diameter models evaluated, coefficient of determination (R2) of the neural network height-diameter model increased by 10.3%, root mean squares error (RMSE) and mean absolute error (MAE) decreased by 12% and 13.5%, respectively. The BP neural network height-diameter model also appeared more accurate than the mixed-effects height-diameter model. Our study proposes the method of determining the optimal numbers of hidden layers, neurons of each layer, and transfer functions in the BP neural network structure. This method can be useful for other modeling studies of similar or different types, such as tree crown modeling, height, and diameter increments modeling, and so on.

RETRACTED:An assessment of software defined networking approach in surveillance using sparse optimization algorithm

Abstract Unmanned aerial vehicles (UAV) can be used as basic elements of the sensor network or an upgrade of existing network that are built with static wireless sensor nodes. Wireless Networks (WN) are utilized across in surveillance in all domains like natural disasters, agriculture, water, forest, military, buildings, health monitoring, disaster relief & emergency management, area and industrial surveillance, due to its wider applicability. Software Defined Networking (SDN) provide a hopeful resolution in bendy supervision WSNs by allowing the separation of the control logic from the sensor nodes/actuators. The advantage with this SDN-based supervision in structure of WSNs is that it enables centralized control of the entire WSN making it simpler to deploy network-wide management protocols and applications on demand. Synthetic Aperture Radar (SAR) images are difficult to analyze due to the presence of speckle noise. Speckle noise must be filtered out before applying to other image processing applications. Three Layered Feed Forward Back Propagation Neural Network (TLFFBPNN) has been proposed to suppress the speckle noise. GLCM properties have been extracted and Back propagation training algorithm is used to train the neural network.

A Model Transfer Learning Framework With Back-Propagation Neural Network for Wine and Chinese Liquor Detection by Electronic Nose

Electronic nose, as a non-destructive instrument, is widely used in the field of gas analysis. In this work, E-nose was employed to distinguish wines and Chinese liquors by means of a machine learning technique. First, a multi-hidden layers Back-Propagation Neural Network (BPNN) was designed to build an identification model for the classification of different wines. Then, a BPNN-based transfer-learning framework was developed with minimal changes to the architecture of the BPNN-based model which was trained on the wine sample dataset. Experimental results revealed that the BPNN-based model performed with a 98.27% accuracy in identifying different wines, and the BPNN-based transfer-learning framework performed with a 93.4% accuracy in identifying Chinese liquors by only re-training the output layer. This reduced the model training costs compared with the complete retraining of a new classification model. Results demonstrated the effectiveness of the proposed BPNN-based transfer-learning model, which was capable of identifying different kinds of wines based on their own properties and could be easily applied to the classification of Chinese liquors. The model-based transfer learning framework offered promising potential for different classification tasks of various beverage.

Research on the Prediction Method of Centrifugal Pump Performance Based on a Double Hidden Layer BP Neural Network

With the aim of improving the shortcomings of the traditional single hidden layer back propagation (BP) neural network structure and learning algorithm, this paper proposes a centrifugal pump performance prediction method based on the combination of the Levenberg–Marquardt (LM) training algorithm and double hidden layer BP neural network. MATLAB was used to establish a double hidden layer BP neural network prediction model to predict the head and efficiency of a centrifugal pump. The average relative error of the head between the experimental and prediction obtained by the double hidden layer BP neural network model was 4.35%, the average relative error of the model prediction efficiency and the experimental efficiency was 2.94%, and the convergence time was 1/42 of that of the single hidden layer. The double hidden layer BP neural network model effectively solves the problems of low learning efficiency and easy convergence into local minima—issues that were common in the traditional single hidden layer BP neural network training. Furthermore, the proposed model realizes hydraulic performance prediction during the design process of a centrifugal pump.

Attentional Bias in Human Category Learning: The Case of Deep Learning.

Category learning performance is influenced by both the nature of the category's structure and the way category features are processed during learning. Shepard (1964, 1987) showed that stimuli can have structures with features that are statistically uncorrelated (separable) or statistically correlated (integral) within categories. Humans find it much easier to learn categories having separable features, especially when attention to only a subset of relevant features is required, and harder to learn categories having integral features, which require consideration of all of the available features and integration of all the relevant category features satisfying the category rule (Garner, 1974). In contrast to humans, a single hidden layer backpropagation (BP) neural network has been shown to learn both separable and integral categories equally easily, independent of the category rule (Kruschke, 1993). This “failure” to replicate human category performance appeared to be strong evidence that connectionist networks were incapable of modeling human attentional bias. We tested the presumed limitations of attentional bias in networks in two ways: (1) by having networks learn categories with exemplars that have high feature complexity in contrast to the low dimensional stimuli previously used, and (2) by investigating whether a Deep Learning (DL) network, which has demonstrated humanlike performance in many different kinds of tasks (language translation, autonomous driving, etc.), would display human-like attentional bias during category learning. We were able to show a number of interesting results. First, we replicated the failure of BP to differentially process integral and separable category structures when low dimensional stimuli are used (Garner, 1974; Kruschke, 1993). Second, we show that using the same low dimensional stimuli, Deep Learning (DL), unlike BP but similar to humans, learns separable category structures more quickly than integral category structures. Third, we show that even BP can exhibit human like learning differences between integral and separable category structures when high dimensional stimuli (face exemplars) are used. We conclude, after visualizing the hidden unit representations, that DL appears to extend initial learning due to feature development thereby reducing destructive feature competition by incrementally refining feature detectors throughout later layers until a tipping point (in terms of error) is reached resulting in rapid asymptotic learning.

Двошарова нейронна мережа зворотного поширення на основі обмежених машин Больцмана для прогнозування добової концентрації ультрадисперсних частинок РМ2.5

Particulate matter 2.5 (PM2.5) pollution is an actual problem in the modern world and forecasting of the daily concentration of PM25 is a challenging task for researchers. In this study, a novel neural network model that effec­tively forecasts daily PM2.5 in Hangzhou city was developed in the form of a restricted Boltzmann machines double layer back propagation neural net­work model (RBM-DL-BPNN). Air quality index, the air pollutants, e.g., particulate matter 10 (PM10), PM25, SO2, CO, NO2, O3, and meteorological parameters (temperature, dew point, humidity, pressure, wind speed, and precipitation) of Hangzhou city were used in this study to train and test three models: RBM-DL-BPNN, double layer back propagation neural network (DL-BPNN), and back propagation neural network (BPNN). The results of experiments and analyses performed indicate that RBM-DL-BPNN has a smaller mean absolute percent error (MAPE), smaller overall daily absolute percentage errors, and more results in terms of absolute percentage error within the range 0-50 % than DL-BPNN and BPNN.

2족 보행로봇의 실시간 작업동작 생성을 위한 지능제어에 관한 연구

In this paper, we propose a new learning control scheme for various walk motion control of biped robot with same learning-base by neural network. We show that learning control algorithm based on the neural network is significantly more attractive intelligent controller design than previous traditional forms of control systems. A multi layer back propagation neural network identification is simulated to obtain a dynamic model of biped robot. Once the neural network has learned, the other neural network control is designed for various trajectory tracking control with same learning-base.The biped robots have been received increased attention due to several properties such as its human like mobility and the high-order dynamic equation. These properties enable the biped robots to perform the dangerous works instead of human beings. Thus, the stable walking control of the biped robots is a fundamentally hot issue and has been studied by many researchers. However, legged locomotion, it is difficult to control the biped robots. Besides, unlike the robot manipulator, the biped robot has an uncontrollable degree of freedom playing a dominant role for the stability of their locomotion in the biped robot dynamics. From the simulation and experiments the reliability of iterative learning control was illustrated.Keywords : Learning control, Biped Robot, Neural Network, Real-Time

Tourism sustainability in Tibet – Forward planning using a systems approach

The purpose of this study is to employ a nonlinear dynamic evaluation method to assess the tourism sustainability of Tibet Autonomous Region (TAR), China, a new emerging tourism destination. The methodology draws on system dynamics and Back Propagation (BP) neural network. According to 7 setting principles, this study identifies 13 tourism sustainability indicators including conventional tourism income, tourism resources stock, pollution stock, etc., as well as specific residents’ tourism cognition, seasonal difference, accessibility, etc. Then a system dynamics model including the 13 indicators (variables) and other relevant auxiliary variables is established. Based on the numerical simulation, using a three layers BP neural network optimized by genetic algorithm and particle swarm algorithm, this study evaluates the future sustainability dynamically and compares the sustainability evolution from 2014 to 2050 under different development strategies. The research results not only provide information useful for the dynamic control and scientific management of the future sustainable tourism development, but also provide a systems approach to evaluate regional tourism sustainability.

Activity prediction and molecular mechanism of bovine blood derived angiotensin I-converting enzyme inhibitory peptides.

Development of angiotensin I-converting enzyme (ACE, EC 3.4.15.1) inhibitory peptides from food protein is under extensive research as alternative for the prevention of hypertension. However, it is difficult to identify peptides released from food sources. To accelerate the progress of peptide identification, a three layer back propagation neural network model was established to predict the ACE-inhibitory activity of pentapeptides derived from bovine hemoglobin by simulated enzyme digestion. The pentapeptide WTQRF has the best predicted value with experimental IC50 23.93 μM. The potential molecular mechanism of the WTQRF / ACE interaction was investigated by flexible docking.

Modelling and multi-response optimization of hole sinking electrical discharge micromachining of titanium alloy thin sheet

Thin sheets of titanium alloys are widely used in aerospace and automotive industries for specific applications. The creation of micro holes with requisite hole quality in thin sheets of these alloys using energy of electric discharge is a challenging task for manufacturing engineers. Hole sinking electrical discharge micromachining (HS-EDMM) is one of the most promising micromachining processes to create symmetrical and non-symmetrical micro holes. The present paper is related to selection of optimum parameter settings for obtaining maximum material removal, minimum tool wear and minimum hole taper in HS-EDMM. In this paper an attempt has been made to develop an integrated model (ANN-GRA-PCA) of single hidden layer back propagation neural network (BPNN) for prediction and grey relational analysis (GRA) coupled with principal component analysis (PCA) hybrid optimization strategy with multiple responses of HSEDMM of Ti-6Al-4V. Experiments have been conducted to generate dataset for training and testing of the network where input parameters consist of gap voltage, capacitance of capacitor and the resulting performance parameters are represented by material removal rate (MRR), tool wear rate (TWR), and hole taper (Ta). The results indicate that the integrated model is capable to predict and optimize process performance with reasonable accuracy under varied operating conditions of HS-EDMM. The proposed approach would be extendable to other configurations of EDMM processes for different materials.

The use of artificial neural network for modeling in vitro rumen methane production using the CNCPS carbohydrate fractions as dietary variables

The objective of this trial was to investigate the suitability and accuracy of modeling the rumen methane production of mixed rations for cattle using artificial neural network (ANN). The three layer back propagation neural network (BP) which included the input, the hidden and the output layers, was used for modeling. Two datasets used in the trial were from Dong and Zhao (2013). The first dataset which contained the CH4, CO2 and total gas production and the Cornell Net Carbohydrate and Protein System (CNCPS) carbohydrate fractions of forty-five rations was for training the BP model and the second dataset which contained ten rations was for testing the BP model. The predicting performances of the BP models with different number of neurons in the hidden layer and different number of variables in the output layer were compared, and the effective BP models were established. Paired t-test showed that no difference was found between the observed and the predicted CH4, CO2 and total gas production based on the BP models (p>0.05). Model performance analysis based on the test data showed the root mean square prediction errors (RMSPE%) were 3.89%, 2.95% and 4.23%, and the determination coefficients (r2) between the observed and the predicted values were 0.95, 0.97 and 0.92 for CH4, CO2 and total gas, respectively. Testing of the BP models indicated that the in vitro CH4, CO2 and total gas production of mixed rations for cattle could be reliably and accurately predicted based on the CNCPS carbohydrate fractions using BP models. The BP models showed similar accuracy with the multiple regression model for predicting the CH4 production and better accuracy for predicting the CO2 and the total gas production than the multiple regression models.

Firearm Classification using Neural Networks on Ring of Firing Pin Impression Images

This paper implements two layer neural networks with different feedforward backpropagation algorithms for better performance of firearm classification us-ing numerical features from the ring image. A total of 747 ring images which are extracted from centre of the firing pin impression have been captured from five different pistols of the Parabellum Vector SPI 9mm model. Then, based on finding from the previous studies, the six best geometric moments numerical fea-tures were extracted from those ring images. The elements of the dataset were further randomly divided into the training set (523 elements), testing set (112 el-ements) and validation set (112 elements) in accordance with the requirement of the supervised learning nature of the backpropagation neural network (BPNN). Empirical results show that a two layer BPNN with a 6-7-5 configura-tion and tansig/tansig transfer functions with ‘trainscg’ training algorithm has produced the best classification result of 98%. The classification result is an improvement compared to the previous studies as well as confirming that the ring image region contains useful information for firearm classification.