Algorithm-based Back Propagation Neural Network Research Articles

Using a genetic backpropagation neural network model for credit risk assessment in the micro, small and medium-sized enterprises

In China, with the "Double Carbon" goal within reach, Micro, Small and Medium-sized Enterprises (MSMEs) emerge as pivotal contributors to economic advancement. However, they are now confronted with the imperative of transitioning towards green and low-carbon practices. To facilitate the attainment of peak carbon dioxide emissions and carbon neutrality, a refined approach is imperative. This entails precise capital allocation, enhanced financial services, streamlined management, and robust risk mitigation strategies. Consequently, conducting thorough credit risk assessments for MSMEs becomes a crucial endeavor. However, obtaining substantial loans for them proves challenging due to their elusive credit ratings and potential defaults. To address this issue, this study leverages machine learning and intelligent optimization algorithms to construct a classification model for default and credit ratings of MSMEs, utilizing their daily invoice data. Specifically, twelve indicators pertaining to default and credit ratings are extracted. Subsequently, Principal Component Analysis is employed to reduce dimensionality and synthesize all pertinent information. Following this, the Genetic Algorithm-based Back Propagation Neural Network (GA-BPNN) is utilized to delineate the relationship between indicators and default, as well as credit rating, respectively. The results indicate a prediction accuracy of 0.92 for default risk and 0.86 for credit rating. This underscores the efficacy of GA-BPNN in effectively classifying the underlying default risk and credit ratings of MSMEs, offering a promising approach for decision-making.

Adaptive STCPF for UCA pose estimation with improved GA-BPNN and multiple fading factors

In response to the challenge of pose estimation accuracy for unmanned carrier-based aircraft in complex maritime landing environments, we propose an adaptive strong tracking cubature particle filtering method based on an improved Genetic Algorithm-Based Backpropagation Neural Network (GA-BPNN) and multiple fading factors. Despite the effectiveness of strong tracking cubature Kalman filtering (STCKF) in adjusting the covariance of prediction estimation errors using fading factors, it still has limitations in terms of model correction and parameter estimation. Addressing this issue, our research delves deeper into the role of fading factors in adjusting process noise covariance and introduces two methods to maintain covariance matrix symmetry: direct matrix solving and trace matrix solving. This novel approach directly implements real-time dynamic feedback correction of system model parameters within the filtering framework, effectively resolving filtering divergence issues while maintaining a high level of conceptual clarity and interpretability. To further enhance the algorithm’s performance, we introduce an improved neural network algorithm to compensate for linear approximation errors in fading factor calculations, thereby improving filtering accuracy. Moreover, considering the limitations of cubature Kalman filtering when dealing with non-Gaussian systems, we integrate the adaptive STCKF based on the improved GA-BPNN into particle filtering as an importance function, aiming to enhance non-Gaussian processing performance and alleviate particle degeneracy issues. In conclusion, both simulation and real-world data tests confirm the superiority of our research approach.

A novel whale-based algorithm for optimizing the ANN approach: application to temperature compensation in pressure scanner calibration systems

To resolve the nonlinear measured values of pressure scanners with temperature change during pressure measurements, resulting in low accuracy, this paper proposes and develops a new calibration system based on neural networks for pressure scanners. Specifically, we first calibrate the pressure scanner production process, then design the sealing device applicable to the working environment of pressure scanners, and finally install the calibration system of the pressure scanner for calibration experiments to obtain accurate measurement results. In addition, at the algorithmic level, this paper forwards a whale optimization algorithm-based backpropagation neural network method instead of the traditional least squares method to complete the temperature compensation. Both the offline compensation results and the compensated online experimental results exhibit the high measurement accuracy of this temperature compensation method. The full-scale (FS) error is 0.07%, the coefficient of determination (R 2) = 99.27%, with similar results for all channels of the pressure sensor in the temperature environment of −40 °C to 60 °C and absolute pressures ranging from 0 to 1.1 MPa. This result presents a significant reduction in the FS error of 0.25%, R 2 = 91.32% based on the conventional algorithm compensation, and also applies to pressure scanners with wide temperature region, wide range, and high accuracy temperature compensation, which is crucial for their future research of low-cost overall calibration and high accuracy algorithms.

Fabrication and Characterization of Lead-Free BNT-6BT Ultrasonic Transducers Designed by an Intelligent Optimization Algorithm

Lead-free piezoelectric material-based ultrasonic transducers have been researched for several years, but the inefficient properties and design difficulties have troubled lead-free ultrasonic transducers for a long time. To improve the performance and design efficiency of lead-free ultrasonic transducers, in this work, an equivalent circuit model and intelligent optimization algorithm were combined for use in a transducer design. Firstly, 0.94(Bi0.5Na0.5)TiO3-0.06BaTiO3(BNT-6BT) lead-free piezoelectric ceramics were prepared and characterized. Then, BNT-6BT ceramics were used to fabricate the ultrasonic transducers. An equivalent circuit model-based software, PiezoCAD, and a genetic algorithm-based back-propagation neural network were used to optimize the design of the transducers. A 3.03 MHz center frequency and 60.3% −6 dB bandwidth of the optimized transducers were achieved, which were consistent with the neural networks optimization results. To verify the application potential of the lead-free transducers, tungsten rods phantom imaging and polystyrene spheres with 300 μm diameter manipulation were completed by the transducers, and the experiment results indicate that the BNT-6BT lead-free transducers have great potential in further biological and biomedical applications.

Backpropagation Neural Network optimization and software defect estimation modelling using a hybrid Salp Swarm optimizer-based Simulated Annealing Algorithm

Software Defect Estimation (SDE) is a fundamental problem solving mechanism in the field of software engineering (SE). SDE is a task that identifies software models that are likely to have defects. In addition, SDE plays a vital overall role in improving software quality, reducing software development costs and accelerating software development processes. The Backpropagation Neural Network (BPNN) is a popular machine learning (ML) estimator widely utilized in SE estimation problems. Unfortunately, its performance depends on the initial weight and bias values. Metaheuristic optimization algorithms, as an alternative method, have proven to have strengths in parameter optimizations. Additionally, population-based metaheuristic algorithms suffer from low exploitation capabilities. In this paper, a new hybrid metaheuristic algorithm-based BPNN (SSA–SA) is proposed by hybridizing the Salp Swarm Algorithm (SSA) with the Simulated Annealing (SA) algorithm. The main goal of the hybridization is to adjust the balance between exploration and exploitation in SSA. The proposed algorithm is also assembled with the BPNN estimator to optimize its parameters to reduce the overall estimation error, which boosts the estimation accuracy. Thus, the proposed algorithm addresses the SDE problem. Experimental results prove the superiority of the proposed hybrid algorithm in optimizing BPNN parameters in comparisons against other estimators and algorithms in most SDE datasets and evaluation criteria.

Intelligent state of charge estimation of lithium-ion batteries based on L-M optimized back-propagation neural network

The state of charge (SOC) of lithium-ion batteries (LIBs) is a critical parameter of the battery management system (BMS), which represents the remaining capacity of LIBs. Precise SOC estimation is vitally important to ensure the safe and reliable operation of electric vehicles (EVs). In this paper, a series of intelligent SOC estimation methods using Levenberg-Marquard (L-M) algorithm based back-propagation neural network (BPNN) are proposed and compared with the extended Kalman filter (EKF) method. Firstly, genetic algorithm (GA) and particle swarm optimization (PSO) algorithm are used to optimize the three-layer BPNN based on L-M training (LMBP) and the multi-hidden-layer BPNN based on L-M training (LMMBP), which improve the estimated accuracy and convergence speed. Besides, it is verified that the LMMBP has better estimation performance than LMBP, the average absolute error (AAE) and the root mean square error (RMSE) of LMMBP can be reduced to 0.4% and 0.5% respectively under the United Kingdom Bus Cycle (UKBC). Finally, four typical driving cycles are used to carry out comparative analysis experiments, combining with the robustness evaluation results including measurement noises test, untrained driving cycles test, different batteries and piecewise training tests, it is validated that the intelligent SOC estimation methods proposed in this paper have high accuracy and strong robustness.

Compressive strength of rubberized concrete: Regression and GA-BPNN approaches using ultrasonic pulse velocity

Rubberized concrete is one of the solutions to the disposal problem caused by large amounts of untreated waste rubber. To assess the performance of existing concrete structures, non-destructive-testing techniques offer a direct, quick, safe and reliable means of assessing the performance of concrete structures. Several researchers have proposed relationships, often of an exponential form, between the Ultrasonic Pulse Velocity (UPV) and compressive strength of rubberized concrete. This paper aims to propose a regression model and a genetic algorithm based backpropagation neural network (GA-BPNN) model that can be used to estimate compressive strength of rubberized concrete with UPV for different applications and accuracy requirements. Regression model in comparisons with GA-BPNN, firstly requires less computation work and will be easier for site measurement or environments without computers or certain softwares, secondly it has no barriers for users who are not familiar with machine learning models to approximately estimate the strength. The regression model comprises an Adjusted Regression Model which is a multi-variable non-linear model adjusted based on the ordinary exponential model incorporating other principal parameters, hence representing an improvement on the existing exponential model and two types of Stepwise Regression Model (pure linear and pure quadratic) will be employed. To achieve this, a database containing 158 pairs of data collected from previous literature is compiled. Results indicate that both three types of regression models and GA-BPNN are capable of effectively predicting the compressive strength of rubberized concrete with reasonable values of statistical indexes. More specifically, among three types of regression model, the pure quadratic stepwise regression model has relatively better performance with higher R and lower root-mean-square error values. Results also support that GA-BPNN has the highest accuracy compared to regression models and is proven to be reasonable for more precise estimations.

Non-Invasive Glucose Measurement Using Sub-Terahertz Sensor, Time Domain Processing, and Neural Network

This paper reports a non-invasive sub-Terahertz glucose concentration measurement system consisting of Sensor Unit (SU) and Processing Unit (PU). The SU. uses waveguide probe sensors to obtain the S parameters of glucose samples of different concentrations. These S parameters depend on the dielectric properties of glucose samples. The frequency-dependent permittivity values of various glucose samples are theoretically estimated using the double-Debye model for sensitivity and uncertainty investigations. The glucose sample concentration is used in the range 70-145 mg/dl to mimic healthy human bodies' blood glucose levels, ranging from 70 to 140 mg/dl. The S parameters obtained through SU are converted to the time domain to obtain real-valued impulse responses, which are normalized in PU, making the input data suitable for analysis using Levenberg-Marquardt (LM) algorithm-based Back Propagation Neural Network. The proposed SU. provides a sensitivity of 2 dB for 15 mg/dl change in glucose concentration, and PU exhibits an accuracy of ±5%, which falls within the clinical range specified for non-invasive based monitoring systems for diabetes. Overall, SU provides high sensitivity towards blood glucose measurement, and PU enhances the measurement system's readability by forming a non-linear relationship between S parameters and glucose concentration values.

Fault diagnosis of electric rudder system using PSOFOA-BP neural network

Electric rudder system exerts an immense influence on the movement of aircraft, missiles and ships. To solve the problem of low automation in traditional rudder system, we developed a novel intelligent electric rudder system test equipment, and proposed a new machine learning algorithm for fault diagnosis. This study mainly focuses on the multidimensional data set test and fault diagnosis of the rudder system. Additionally, so as to further the accuracy of the classification model, we optimized back propagation neural network (BPNN) with particle swarm optimization hybrid fruit fly algorithm (PSOFOA), which we called particle swarm optimization hybrid fruit fly algorithm-based back propagation neural network (PSOFOA-BP), and PSOFOA effectively resolved the weights and biases of BPNN. The experimental results indicate that the novel model proposed in this study has higher performance in fault diagnosis of rudder system, which can provides a new concept for rudder testing and production in the future.

Location of abnormal energy consumption and optimization of energy efficiency of hydraulic press considering uncertainty

As a high energy consumption machine, there is plenty of abnormal energy consumption in the operation of a hydraulic press, which leads to energy loss and reduces energy efficiency. The time series of energy consumption data are characterized by intense uncertainty, nonlinearity, and non–Gaussianity. The traditional method is used to locate it, which has the problems of a large error and uncertain locating result. Therefore, an abnormal energy consumption locating technique considering uncertainty is developed. To quantitatively evaluate the correlation between abnormal machine components and energy consumption, an uncertainty evaluation method combining entropy weight and fuzzy theory is proposed, which aims to provide a decision basis for the location of abnormal energy consumption. To accurately locate abnormal energy consumption, a novel localization technique is proposed, which is called genetic algorithm–based wavelet neural network. Firstly, the energy consumption data is processed by wavelet packet decomposition and reconstruction. The energy value is calculated, and then the dimension reduction is performed by principal component analysis. Finally, energy value is input into the genetic algorithm–based backpropagation neural network model for training and anomaly location. We integrate the proposed method into the energy management system and perform energy efficiency optimization analysis. The results show that the proposed method can improve the energy efficiency of a hydraulic press, reduce energy cost, and reduce environmental pollution. This provides a solution for solving the cleaner production problem of high energy consumption manufacturing industry and realizing sustainable intelligent manufacturing.

Complementary Ensemble Adaptive Local Iterative Filtering and Its Application to Rolling Bearing Fault Diagnosis

A fault diagnosis model for rolling bearing based on complementary ensemble adaptive local iterative filtering (CEALIF), Laplacian score (LS) feature selection, and genetic algorithm-based backpropagation neural network (GA-BPNN) is proposed in this article. When the rolling bearing fails, the field-measured waveforms usually shows strong nonlinearity and non-stationary characteristics. Adaptive Local Iterative Filtering (ALIF) is an alternative novel approach to empirical mode decomposition to decompose complex signals into multiple intrinsic mode functions (IMFs), but modal aliasing will occur in actual processing. Aiming at the phenomenon of modal aliasing, a noise-assisted analysis methodology, namely complementary ensemble adaptive local iterative filtering, which could overcome the modal aliasing problem of ALIF. This article applies it to the pre-processing of rolling bearing time series. Then, the time domain (TD) statistical features of the IMFs, their Fourier frequency domain (FD) features, and the time frequency domain (TFD) energy features are extracted to capture the fault information. Meanwhile, to avoid feature redundancy and enhance the diagnostic performance, the LS is adopted to rank the features to improve the fault characteristics. Subsequently, the optimized feature vectors are entered into the GA-BPNN to automatically achieve the fault type recognition. The experimental data analysis results of rolling bearings indicate that the model can effectively diagnose the degree and type of failure.

Crop Growth Stage GPP-Driven Spectral Model for Evaluation of Cultivated Land Quality Using GA-BPNN

Rapid and accurate evaluation of cultivated land quality (CLQ) using remotely sensed images plays an important role for national food security and social stability. Current approaches for evaluating CLQ do not consider spectral response relationships between CLQ and spectral indicators based on crop growth stages. This study aimed to propose an accurate spectral model to evaluate CLQ based on late rice phenology. In order to increase the accuracy of evaluation, the Empirical Bayes Kriging (EBK) interpolation was first performed to scale down gross primary production (GPP) products from a 500 m spatial resolution to 30 m. As an indicator, the ability of MODIS-GPPs from critical growth stages (tillering, jointing, heading, and maturity stages) was then investigated by combining Pearson correlation analysis and variance inflation factor (VIF) to select the phases of CLQ evaluation. Finally, a linear Partial Least Squares Regression (PLSR) and two nonlinear models, including Support Vector Regression (SVR) and Genetic Algorithm-Based Back Propagation Neural Network (GA-BPNN), were driven to develop an accurate spectral model of evaluating CLQ based on MODIS-GPPs. The models were tested and compared in the Conghua and Zengcheng districts of Guangzhou City, Guangdong, China. The results showed that based on field measured GPP data, the validation accuracy of 30 m spatial resolution MODIS GPP products with a root mean square error (RMSE) of 7.43 and normalized RMSE (NRMSE) of 1.59% was higher than that of the 500 m MODIS GPP products, indicating that the downscaled 30 m MODIS GPP products by EBK were more appropriate than the 500 m products. Compared with PLSR (R2 = 0.38 and RMSE = 87.97) and SVR (R2 = 0.64 and RMSE = 64.38), the GA-BPNN model (R2 = 0.69 and RMSE = 60.12) was more accurate to evaluate CLQ, implying a non-linear relationship of CLQ with the GPP spectral indicator. This is the first study to improve the accuracy of estimating CLQ using the rice growth stage GPP-driven spectral model by GA-BPNN and can thus advance the literature in this field.

Monitoring maize growth on the North China Plain using a hybrid genetic algorithm-based back-propagation neural network model

Crop growth and early yield information is crucial for the establishment and adjustment of agricultural management plans. Timely, precise and regional assessments of crop growth conditions and production greatly benefit the national economy and agriculture. In this study, remotely sensed leaf area index (LAI) and vegetation temperature condition index (VTCI) data retrieved from the Global LAnd Surface Satellite (GLASS) and Moderate-resolution Imaging Spectroradiometer (MODIS) data were selected as the key indicators of maize growth and a hybrid genetic algorithm (GA)-based back-propagation neural network (BPNN) (GA-BPNN) model was applied to provide complementary information on maize growth at the main growth stage. GA-BPNN models with different architectures were established, and an architecture with 10, 9 and 1 nodes in the input, hidden and output layers, respectively, achieved the best training and testing performance. The root mean square error (RMSE) values of the training and testing samples were 588.2 kg/ha and 663.4 kg/ha, respectively. Thus, the hybrid model with the best architecture (10-9-1) was selected to calculate the values of the integrated growth monitoring index (G) at the regional scale with a 1 km spatial resolution in the study area from 2010 to 2018. The results showed that the monitored maize growth well reflected the actual situation and the correlations between G values and sites’ measured variables, such as the maize yield and soil relative humidity, were higher than that of a pure BPNN model. The linear relationship between the GA-BPNN-based G values and maize yields was analyzed to estimate the maize yield in the North China Plain. Most of the RMSE and mean absolute percentage error (MAPE) values between the estimated and actual maize yields were less than 700.0 kg/ha and 10.0%, respectively. Considering that the estimation errors of most statistical samples were small and there were no obvious differences between the estimated maize yields in the adjacent regions of the northern, central and southern plain, the GA-BPNN-based yield estimation model provided reliable and spatially continuous estimates of maize yield.

Prediction of critical properties of biodiesel fuels from FAMEs compositions using intelligent genetic algorithm-based back propagation neural network

ABSTRACT Biodiesels are considered as promising fuels to substitute diesel fuel which can fill the gap of energy shortage while maintaining the diesel engine’s efficiency. Variations in the properties of different biodiesel fuels are caused by the varied fatty acid methyl esters (FAMEs) compositions derived from their parent oils. Therefore, correlating key properties such as cetane number (CN), kinematic viscosity (KV), iodine value (IV) and cold filter plugging point (CFPP) with FAMEs compositions of each biodiesel fuel is significant to developing whatever new types of fuels applied on diesel engines. In this study, an intelligent genetic algorithm (GA)-based back propagation neural network (BPNN) model was proposed to predict the properties of biodiesel fuels according to FAMEs compositions. The hybrid BPNN-GA model has five inputs (methyl palmitate, methyl stearate, methyl oleate, methyl linoleate and methyl linolenate) corresponding to the FAMEs compositions and outputs with estimated fuel properties, with the GA assisting on training to find out local minimum deviation and updated weighting configurations. It was found that the intelligent learning-training method proposed hybrid BPNN-GA model enabled to map the non-linear relationships between the FAMEs compositions and key properties of biodiesel fuels with fairly good agreement. The predicted value of fuel properties agrees with measured ones with R-square up to 96%, along with lower value (less than 10%) over Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) than those of other empirical correlations. In addition, a sensitivity analysis was conducted to in-depth investigate the FAMEs compositions on key properties. It was concluded that saturated FAMEs have positive impacts on CN, KV and the CFPP, while IV is typically dependent on unsaturated FAMEs. Therefore, it is attainable to formulate new types of alternative fuels based on the required properties on diesel engine applications.

Thermal positioning error modeling of machine tools using a bat algorithm-based back propagation neural network

Thermal error of a machine tool is one of the main reasons affecting the machining accuracy. Heat production and heat transfer of a machine tool are too complicated to predict the generated thermal error accurately. According to the nonlinear and time-varying characteristics of thermal error, the back propagation (BP) neural network is perfectly suitable for thermal error modeling, which has been extensively used to map the nonlinear relationship. However, traditional BP neural network usually has poor prediction performance under different operating conditions. Therefore, a new swarm intelligent optimization algorithm, bat algorithm (BA), is introduced to optimize BP neural network and improve its performance. The focus of this paper is the application of the combined algorithm (bat algorithm-based back propagation neural network) to solve the problem of thermal error modeling. Thermal positioning error experiments were conducted on a three-axis experiment bench. The experimental results show that thermal positioning error model built by BA-BP neural network is more stable and has high prediction accuracy and strong robustness, which can provide a candidate method for thermal error modeling.

A hybrid health condition monitoring method in milling operations

Cutting chatter has been a very important issue in the milling operations due to its unexpected and uncontrollable characteristics. Developing an effective healthy condition monitoring method is critical to identify cutting chatter exactly. In this paper, a hybrid healthy condition monitoring (HHCM) method, that combines variational mode decomposition (VMD) with genetic algorithm-based back propagation neural network (BPNN) model, is developed for cutting chatter detection and state classification in complex and non-stationary milling operations. First, cutting chatter vibration signal is decomposed into multiple mode components by the VMD. Then, Shannon power spectral entropy is adopted to extract features from decomposed vibration signals. Furthermore, BPNN model is optimized by traditional genetic algorithm to identify and classify machine states in milling operations. Last, impeller milling experiments are conducted and results show that the proposed HHCM method can effectively realize cutting chatter detection and state classification during milling operations.

Thermal Error Compensation Method Based on Floyd Algorithm

In this paper, a new compensation method of thermal error is proposed based on Floyd algorithm and homogeneous transformation matrices (HTMs). With the topological structure and measured data, the thermal error modelling is established. Then, according to the improved Floyd minimum-distance method and ant colony algorithm-based back propagation neural network (ACO-BPN), the error compensation models were established respectively. A comparison between the Floyd algorithm compensation method and a typical method ACO-BPN is performed, and the results shows that the Floyd algorithm has advantages of higher compensation accuracy, more effective and shorter time than ACO-BPN. Under variable temperatures conditions, the Floyd compensation method also has higher compensation accuracy.

Credibility forecasting in short‐term load forecasting and its application

It is very helpful for power system operation to assess and forecast the uncertainty of the load forecasting. The improved credibility assessment index of the short-term load forecasting results is presented to assess the uncertainty in this study. The forecasting method for credibility of short-term load forecasting is also proposed by adopting genetic algorithm-based back propagation neural network. In the case study, it is proved that the improved credibility assessment can evaluate the short-term load forecasting results and the forecasting methods. Meanwhile, the credibility forecasting can contribute to optimise reserve capacity in power system scheduling.

Application of GRAM and AFSACA-BPN to thermal error optimization modeling of CNC machine tools

Thermal errors are major contributor to dimensional errors of a part during precision machining. Error compensation is an effective method to reduce thermal errors. Accurate modeling of thermal errors is a prerequisite for thermal error compensation. In this paper, five key temperature points of a computer numerical control (CNC) machine tool were selected based on grey relational analysis method (GRAM). One thermal error model based on the five key temperature points was proposed using artificial fish swarm and ant colony algorithm-based back-propagation neural network (AFSACA-BPN). AFS is applied to generate initial pheromone value of ACA, which improves the computational efficiency of BPNs and prediction accuracy of thermal error modeling. One thermal error real-time compensation system was developed based on the proposed model. An experiment was carried out to verify the performance of the compensation system. Experiment results show that the diameter error of the workpiece reduced from 23 to 10 μm after compensation.

Training a Feed-Forward Neural Network with Artificial Bee Colony based Backpropagation Method

Back-propagation algorithm is one of the most widely used and popular techniques to optimize the feed forward neural network training. Nature inspired meta-heuristic algorithms also provide derivative-free solution to optimize complex problem. Artificial bee colony algorithm is a nature inspired meta-heuristic algorithm, mimicking the foraging or food source searching behaviour of bees in a bee colony and this algorithm is implemented in several applications for an improved optimized outcome. The proposed method in this paper includes an improved artificial bee colony algorithm based back-propagation neural network training method for fast and improved convergence rate of the hybrid neural network learning method. The result is analysed with the genetic algorithm based back-propagation method, and it is another hybridized procedure of its kind. Analysis is performed over standard data sets, reflecting the light of efficiency of proposed method in terms of convergence speed and rate.