Optimized Back Propagation Neural Network Research Articles

Salp swarm optimizer for modeling the software fault prediction problem

This paper proposes the salp swarm algorithm (SSA) combined with a backpropagation neural network (BPNN) to solve the software fault prediction (SFP) problem. The SFP problem is one of the well-known software engineering problems that are concerned with anticipating the software defects that are likely to appear during a software project or thereafter. In order to find the optimal BPNN parameters, a combination of SSA optimizer and BPNN named (SSA-BPNN) is proposed, so as to enhance prediction accuracy. The proposed method is evaluated against several datasets for the SFP problem. These datasets vary in both size and complexity. The results obtained are evaluated using a variety of performance measures (i.e., the AUC, Confusion Matrix, Sensitivity, Specificity, Accuracy, and Error Rate). The results obtained by SSA-BPNN are better than those obtained by the conventional BPNN over all of the datasets. The proposed method also has the ability to outperform several state-of-the-art methods over the same datasets in respect of most of the aforementioned performance measures. Therefore, the hybridization of SSA with BPNN is a valuable addition to the software engineering issues and can be utilized to achieve higher prediction accuracy for a variety of prediction problems.

Construction and verification of color fundus image retinal vessels segmentation algorithm under BP neural network

For the purpose of analyzing the application of back propagation (BP) neural network model in retinal vascular segmentation of color fundus images, in this study, the traditional BP neural network was first improved by adopting the additional momentum method. Second, adaptive histogram, morphological background subtraction, Gauss preprocessing matching filter and Heisen matrix were used to enhance the image and extract the features. Third, the retinal vascular segmentation algorithm for color fundus images was constructed based on optimized BP neural network. Finally, the DRIVE and MESSIDOR data sets of color fundus images were introduced to compare the proposed algorithm with the convolutional neural network (CNN) and pulse-coupled neural network (PCNN) algorithms in terms of performance. In addition, the three algorithms were also compared in terms of the sensitivity (Se), specificity (Sp), accuracy (Ac), average operation time for each image and F1 value. The results show that the BP neural network algorithm proposed in this study shows obvious advantage over the other two algorithms in the segmentation of color fundus images. In the DRIVE data set, the Se (81.37%), Sp (90.55%) and Ac (95.82%) of BP algorithm are the highest among the three; in the MESSIDOR data set, the Se (85.22%), Sp (91.08%) and Ac (96.16%) of BP algorithm are also highest; in the DRIVE and MESSIDOR data sets, the operation time of BP algorithm is (28.46 ± 3.19 ms; 24.73 ± 4.81 ms, respectively), which are significantly less than the other two algorithms. Besides, the F1 value of the proposed algorithm is obviously higher than that the other two algorithms. As a result, it is concluded that compared with the CNN algorithm and the PCNN algorithm, the proposed algorithm is more effective in the retinal vascular segmentation of color fundus images; the capillaries have better connectivity, and the proposed algorithm can improve the segmentation Ac while reducing the operation time.

The Identification Nanoparticle Sensor Using Back Propagation Neural Network Optimized by Genetic Algorithm

This study draws attention towards the application of identification nanoparticle (NPs) sensor based on back propagation (BP) neural network optimized by genetic algorithm (GA) in the early diagnosis of cancer cells. In this study, the traditional and optimized BP neural networks are compared in terms of error between the actual value and the predictive value, and they are further applied to the NP sensor for early diagnosis of cancer cells. The results show that the root mean square (RMS) and mean absolute error (MAE) of the optimized BP neural network are comparatively much smaller than the traditional ones. The particle size of silicon‐coated fluorescent NPs is about 105 nm, and the relative fluorescence intensity of silicon‐coated fluorescent NPs decreases slightly, maintaining the accuracy value above 80%. In the fluorescence imaging, it is found that there is obvious green fluorescence on the surface of the cancer cells, and the cancer cells still emit bright green fluorescence under the dark‐field conditions. In this study, a phenolic resin polymer CMK‐2 with a large surface area is successfully combined with Au. NPs with good dielectric property and bioaffinity are selectively bonded to the modified electrode through a sulfur‐gold bond to prepare NP sensor. The sensor shows good stability, selectivity, and anti‐interference property, providing a new method for the detection of early cancer cells.

Optimization of Backpropagation Neural Network under the Adaptive Genetic Algorithm

This study is to explore the optimization of the adaptive genetic algorithm (AGA) in the backpropagation (BP) neural network (BPNN), so as to expand the application of the BPNN model in nonlinear issues. Traffic flow prediction is undertaken as a research case to analyse the performance of the optimized BPNN. Firstly, the advantages and disadvantages of the BPNN and genetic algorithm (GA) are analyzed based on their working principles, and the AGA is improved and optimized. Secondly, the optimized AGA is applied to optimize the standard BPNN, and the optimized algorithm is named as OAGA‐BPNN. Finally, three different cases are proposed based on the actual scenario of traffic flow prediction to analyse the optimized algorithm on the matrix laboratory (MATLAB) platform by simulation. The results show that the average error distribution of the GA‐BPNN algorithm is about 1% with small fluctuation range, better calculation accuracy, and generalization performance in contrast to the BPNN. The average output error of the AGA‐BPNN fluctuates around 0 and remains in a relatively stable range as a whole in contrast to that of GA‐BPNN; the maximum fitness level keeps increasing during the evolution process but approaches the average value in later process, so the population diversity is hard to be guaranteed. The output error of the OAGA‐BPNN fluctuates little compared with that of AGA‐BPNN, and its maximum fitness continues to increase in the evolution process with guaranteed population diversity. In short, the OAGA‐BPNN algorithm can achieve the best performance in terms of calculation accuracy, generalization performance, and population evolution.

Optimization of Neural Network using Nelder Mead in Breast Cancer Classification

Classification is one of the data mining techniques which considered as supervised learning. Classification technique such as Backpropagation Neural Network (BPNN) has been utilized in several fields to increase human productivity. BPNN can give better results (more natural) compared with other statistical techniques. However, the learning process of BPNN could give an inefficient synapse weight of each hidden layer. This ineffective weight can affect the performance of the network. In this research, BPNN optimization using Nelder Mead to identifying the appearance of breast cancer is proposed. The datasets used are Breast Cancer Coimbra Dataset (BCCD), and Wisconsin Breast Cancer Dataset (WBCD). The testing result using accuracy and k-fold validation presents better performance compared with the original BPNN. Best average performance can be seen in the fifth fold of BCCD with 76.5217% of accuracy. Moreover, the highest average result of WBCD presented in the fourth fold with 91.1765% of average accuracy.

Mechanical behavior prediction of CF/PEEK-titanium hybrid laminates considering temperature effect by artificial neural network

This paper focuses on the mechanical behaviors of carbon fiber-reinforced PEEK-titanium hybrid laminates (TiGr) under different temperatures. The tensile strength, flexural strength and inter-laminar shear strength (ILSS) of TiGr laminates with different stacking structures were investigated at room temperature, 80, 150 and 220 °C, respectively. The results show that the tensile strength, flexural strength and ILSS gradually decrease as the temperature increases, and reduced by 12.3%, 45.4% and 29.1% for Type I, 12.9%, 48.8% and 30.1% for Type II, 16.1%, 45.1% and 39.4% for Type III at 220 °C. Moreover, a three-layers back-propagation (BP) neural network was developed to predict the corresponding mechanical behaviors of TiGr laminates, then the genetic algorithm (GA) was introduced to optimize the established BP neural network and the accuracies of the two prediction models were compared. It is found that both the trained BP and GA-BP neural networks could predict the mechanical performances of TiGr laminates well while the GA optimized BP neural network has better prediction results, with average absolute errors of 1.78%, 1.92% and 2.41% for tensile, flexural and inter-laminar shear strengths, respectively. Furthermore, the predictability of the trained models was validated with data from published literature, providing a new concept for the application of BP neural networks to fiber metal laminates.

Air Combat Maneuver Trajectory Prediction Model of Target Based on Chaotic Theory and IGA-VNN

Target maneuver trajectory prediction is an important prerequisite for air combat situation awareness and threat assessment. Aiming at the problem of low prediction accuracy in traditional trajectory prediction methods, combined with the chaotic characteristics of the target maneuver trajectory time series, a target maneuver trajectory prediction model based on chaotic theory and improved genetic algorithm-Volterra neural network (IGA-VNN) model is proposed, mathematically deducing and analyzing the consistency between Volterra functional model and back propagation (BP) neural network in structure. Firstly, the C-C method is used to reconstruct the phase space of the target trajectory time series, and the maximum Lyapunov exponent of the time series of the target maneuver trajectory is calculated. It is proved that the time series of the target maneuver trajectory has chaotic characteristics, so the chaotic method can be used to predict the target trajectory time series. Then, the practicable Volterra functional model and BP neural network are combined together, learning the advantages of both and overcoming the difficulty in obtaining the high-order kernel function of the Volterra functional model. At the same time, an adaptive crossover mutation operator and a combination mutation operator based on the difference degree of gene segments are proposed to improve the traditional genetic algorithm; the improved genetic algorithm is used to optimize BP neural network, and the optimal initial weights and thresholds are obtained. Finally, the IGA-VNN model of chaotic time series is applied to the prediction of target maneuver trajectory time series, and the experimental results show that its estimated performance is obviously superior to other prediction algorithms.

Rapid identification of cervical adenocarcinoma and cervical squamous cell carcinoma tissue based on Raman spectroscopy combined with multiple machine learning algorithms.

Cervical cancer has a long latency, and early screening greatly reduces mortality. In this study, cervical adenocarcinoma and cervical squamous cell carcinoma tissue data were collected by Raman spectroscopy, and then, the adaptive iteratively reweighted penalized least squares (airPLS) algorithm and Vancouver Raman algorithm (VRA) were used to subtract the background of the collected data. The following five feature extraction algorithms were applied: partial least squares (PLS), principal component analysis (PCA), kernel principal component analysis (KPCA), isometric feature mapping (isomap) and locally linear embedding (LLE). The k-nearest neighbour (KNN), extreme learning machine (ELM), decision tree (DT), backpropagation neural network (BP), genetic optimization backpropagation neural network (GA-BP) and linear discriminant analysis (LDA) classification models were then established through the features extracted by different feature extraction algorithms. In total, 30 types of classification models were established in this experiment. This research includes eight good models, airPLS-PLS-KNN, airPLS-PLS-ELM, airPLS-PLS-GA-BP, airPLS-PLS-BP, airPLS-PLS-LDA, airPLS-PCA-KNN, airPLS-PCA-LDA, and VRA-PLS-KNN, whose diagnostic accuracy was 96.3 %, 95.56 %, 95.06 %, 94.07 %, 92.59 %, 85.19 %, 85.19 % and 85.19 %, respectively. The experimental results showed that the model established in this article is simple to operate and highly accurate and has a good reference value for the rapid screening of cervical cancer.

Numerical Calculation and Uncertain Optimization of Energy Conversion in Interior Ballistics Stage

Gun firing is a process that converts propellant chemical energy to projectile kinetic energy and other kinds of energies. In order to explore the energy conversion process, firstly, the interior ballistics mathematical model and the barrel-projectile finite element model are built and solved. Then, the related variable values and energy values are obtained and discussed. Finally, for improving energy efficiency, the interval uncertainty optimization problem is modeled, and then solved using the two-layer nested optimization strategy and back-propagation (BP) neural network surrogate model. Calculation results show that, after optimization, the heat efficiency raises from 31.13% to 33.05% and the max rifling stress decreases from 893.68 to 859.76 Mpa, which would improve the firing performance and prolong the lifetime of the gun barrel.

Port Container Throughput Forecast Based on ABC Optimized BP Neural Network

In order to improve the accuracy of port container throughput prediction, an improved ABC-BP prediction model based on Artificial Bee Colony (ABC) was proposed. With the advantage of global search ability and difficulty in local optimization, the weights and thresholds of Back Propagation (BP) neural network are optimized, and the optimal weights and thresholds in the network are finally determined. Referring to the existing research results, Qingdao 2014-2019 (24 quarters) GDP, port throughput, total foreign trade import and export volume were selected as network input, and Qingdao port 2014-2019 port container throughput statistics were used as network output, a total of 24 sets of data were constructed, so as to build a BP neural network prediction model. The first 20 groups of data were used as the training set, and the last 4 groups of data were used as the test set for instance verification. The results show that, compared with the traditional BP neural network and the BP neural network prediction model optimized by Genetic Algorithm (GA), the ABC-BP model can significantly improve the prediction accuracy of port container throughput.

Research on Hybrid Energy Storage Configuration in Grid Wind Power Scheduling Tracking Under Statistics and Frequency Decomposition

AbstractThe low accuracy of wind power scheduling influences the grid dispatch adversely, increasing the demand for spinning to reserve capacity and obstructing the grid frequency regulation. Considering the throughput characteristics of energy storage system, which can be used to compensate for wind farm power scheduling deviations, and smooth the grid power fluctuations, the hybrid energy storage (HES) is employed to enhance the dispatch ability of wind power generation. As one of the key techniques, desirable energy storage capacity configuration (ESCC) and control methods would accelerate the application of energy storage in the field of new resource. Combined with statistics and frequency decomposition of scheduling power deviation, HES capacity configuration and online dynamic power allocation method are proposed. First, by analysis of grid assessment indexes of wind power, scheduled wind power data are produced by improved adaptive error factor correction particle swarm optimization back-propagation neural network (AEFC-PSO-BPNN) prediction followed by wavelet packet smooth (WPS). After comparing with actual power, scheduling deviation statistics and frequency decomposition are applied in capacity and power configuration of energy storage, as well as dynamic power distribution control. With wind/storage simulation platform, then, feasibility of energy storage embedded in grid wind power scheduling deviation, regulation is verified under several combined methods, and the proposed ESCC methods are tested in application case by grid wind power indexes of root-mean-square error rate (RMSE), average volatility (AV), maximum throughout power and current (MTP, MTC), actual supercapacitor (SC), battery consumption capacity, and the number of crossings of state of charge (SOC) of HES. Finally, analyses and comparison of energy storage capacity requirements are carried out on different scheduling deviation control methods so as to explore the significant factors influencing capacity allocation. Applying these methods can improve the scheduling accuracy of grid wind power, reduce power fluctuations at the power common connected (PCC) point, and minimize the impact of accessed wind power to the grid as much as possible.

Temperature compensation for MEMS resonant accelerometer based on genetic algorithm optimized backpropagation neural network

Temperature compensation with high accuracy is crucial for improving the performance of MEMS resonant accelerometers. In this paper, we propose an effective temperature compensation method based on the backpropagation neural network (BP-NN). First, we analyzed the relationship among the input acceleration, the environmental temperature, the output frequencies, and the scale factor of a MEMS resonant accelerometer through the traditional polynomial fitting method. After that, we introduced the BP-NN improved by genetic algorithm (GA). Numerous experiments were performed to train the BP-NN model and establish the relationships between the input layer and the output layer. Comparison between single-beam working mode and symmetrical double-beam working mode of the MEMS resonant accelerometer proved that the latter had a better temperature compensation effect due to its minimized error caused by temperature measurement. Experimental results show that the maximum error of our approach is 0.017 % over the whole temperature range from -10°C to 80°C, which is 173-times better than the traditional polynomial fitting method.

Estimating waterway freight demand at Three Gorges ship lock on Yangtze River by backpropagation neural network modeling

The waterway freight volume through the Three Gorges ship lock (TGL) has increased significantly since it started operation in 2003 and exceeded the designed capacity of 100 million tons in 2011, 19 years earlier than expected. This made the TGL a bottleneck for the Yangtze River waterway transport development. Based on the historical freight volumes and types through the TGL, a combination of qualitative and quantitative analyses are conducted here to identify the key factors affecting freight demand. A genetic algorithm and backpropagation (GA–BP) forecasting model (optimized backpropagation neural network model using a genetic algorithm) was developed to estimate freight demand at the TGL over the period 2020–2040. We predict that freight demand will continue to increase, reaching 260.2 million tons in the basic scenario by 2040 (or 224.7 million tons in the conservative scenario and 276.7 million tons in the optimistic scenario). However, the growth rate will gradually decline. The freight composition will tend to be more stable and homogeneous, with over 64% of the freight related to investment and construction. From the perspective of freight volume evolution, the necessity of the Three Gorges New Locks Project is justified and necessary.

A compounding-model comprising back propagation neural network and genetic algorithm for performance prediction of bio-based lubricant blending with functional additives

Purpose The amount, type and addition conditions of additives of lubricants should be continuously adjusted to obtain appealing performance. To obtain the optimal pretreatment parameters and reduce the cost of time-consuming experiments, the purpose of this paper is to establish an optimal back propagation neural network (BPNN) model combined with genetic algorithm (GA) in this work. Design/methodology/approach Using trimethylolpropane trioleate as the base oil and three types of phosphorus compounds as additives, 25 sets of lubricant formulas were designed regarding lubricant performances of average friction coefficient, average spot diameter, disk wear volume and extreme pressure. The data set was used for training and learning of BPNN and then combined with GA to optimize BPNN with continuously optimization by adjusting various parameters. Findings Comparing prediction data of BPNN with actual test data, correlation coefficients were above 90%, indicating that the model could accurately predict the performance of lubricants. When combined with GA, all performance errors were less than 5%, indicating that BPNN could be optimized by GA to obtain an accurate combined model for prediction of lubricant performance. The best additive formula with excellent performances was obtained from the BPNN–GA model. Originality/value This work developed a new method to study lubricant compounding. The combined model was expected to provide a theoretical basis and guidance for the compounding optimization of lubricant additives with high efficiency and low cost and to expand the scope to practical applications. Peer review The peer review history for this article is available at: https://publons.com/publon/10.1108/ILT-05-2020-0165/

Impact of cost–benefit analysis on financial benefit evaluation of investment projects under back propagation neural network

In order to realize the financial benefit evaluation of the investment project, the investment project of power transmission and transformation of a power grid enterprise in Sichuan province is taken as an example. First, the overall cost of the investment project is analyzed and introduced, and the sales revenue, running costs, taxes and surcharges in the four years from 2016 to 2019 are calculated, and the financial benefit evaluation index and calculation method are determined. Then, the Fuzzy optimal back propagation neural network (BPNN) is constructed based on BPNN and fuzzy optimization strategy for cost–benefit analysis. Then an improved Gravitational search algorithm (IGSA) algorithm based on GSA algorithm is proposed applied to the optimization of fuzzy optimal BPNN. Finally, the influence of some uncertainty factors on the financial benefit of investment project is measured through break-even analysis and sensitivity analysis. The results show that the operation index of the substation is stable and the electricity sold shows an increasing trend after its operation. The financial situation of the investment project is optimistic with high economic benefits. The internal rate of return of the capital can reach 17.13%, and the payback period of the investment is less than 7 years. The sensitivity of the internal rate of return of project investment from high to low is unit price of electricity sold, electricity sold and operation cost. The output value of the samples trained by the fuzzy optimal BPNN is close to the expected value, with a small error of about 0.3% at most, and the training speed is fast and the training effect is good. The actual output result of the financial evaluation index is basically consistent with the expected output result of the expert survey, with a small error. Moreover, the application of the model to other similar investment projects still has good financial benefit evaluation results and good generalization ability. In conclusion, BPNN can be used to calculate the cost and benefit and predict the financial benefit of investment projects. According to the forecast results, a feasible investment scheme is proposed, which can be used for reference in the financial benefit evaluation of similar investment projects.

Study on prediction model of diesel engine with regulated two-stage turbocharging system based on hybrid genetic algorithm-particle swarm optimization method at different altitudes

ABSTRACT Aiming at the problem of complicated mechanism modeling for regulated two-stage turbocharging (RTST) system of diesel engine, a hybrid genetic algorithm–particle swarm optimization (GA-PSO) method was applied to optimize back propagation neural network (BPNN) performance prediction model of diesel engine. First, using a high-altitude (low-pressure) simulation test bench for diesel engines, a full-load performance test of RTST diesel engine at different altitudes was carried out, and the regulation rules of diesel engine intake characteristics and two-stage turbocharger (TST) matching characteristics with altitude were obtained. Based on the test data, the BPNN is used to establish the boost pressure prediction model of the RTST system of the diesel engine. For the problem of slow convergence and low prediction accuracy of the BPNN, adaptive improvement measures such as momentum and learning rate are introduced, and the GA-PSO optimizes the weights and thresholds of BPNN. The performance comparison shows that hybrid GA-PSO is superior to particle swarm optimization (PSO) and genetic algorithm (GA) in terms of evolution speed and convergence accuracy. Genetic algorithm–particle swarm optimization–back propagation neural network (GA-PSO-BPNN) model has higher prediction accuracy and higher accuracy than particle swarm optimization–back propagation neural network (PSO-BPNN) model and genetic algorithm–back propagation neural network (GA-BPNN) model. Compared with the original model, the root-mean-square error of the model is reduced by 50%, the mean error percentage (MEP) is reduced by 38%, and the sum of the squared errors is reduced by 79%. The performance verification of the GA-PSO-BPNN model shows that the MEP of the model prediction results is 5.5%, and the GA-PSO-BPNN model has higher prediction accuracy.

Investigation of Strength and Fatigue Life of Rubber Asphalt Mixture.

Strength and fatigue life are essential parameters of pavement structure design. To accurately determine the pavement structure resistance of rubber asphalt mixture, the strength tests at various temperatures, loading rate, and fatigue tests at different stress levels were conducted in this research. Based on the proposed experiments, the change law of rubber asphalt mixture strength with different temperatures and loading rates was revealed. The phenomenological fatigue equation of rubber asphalt mixture was established. The genetic algorithm optimized backpropagation neural network (GA-BPNN) is highly reliable for optimizing production processes in civil engineering, and it has a remarkable application effect. A GA-BPNN strength and fatigue life prediction model was created in this study. The reliability of the prediction model was verified through experiments. The results showed that the rubber asphalt mixture strength decreases and increases with the increase of temperature and loading rate, respectively. The goodness of fit of the rubber asphalt mixture strength and fatigue life prediction model based on the GA-BPNN could reach 0.989 and 0.998, respectively. The indicators of the fatigue life prediction model are superior to the conventional phenomenological fatigue equation model. The GA-BPNN provides an effective method for predicting the rubber asphalt mixture strength and fatigue life, which significantly improves the accuracy of the resistance design of the rubber asphalt pavement structure.

Hysteresis modeling of piezoelectric actuator using particle swarm optimization-based neural network

Piezoelectric actuators have been received much attention for the advantages of high precision, no wear and rapid response, etc. However, the intrinsic hysteresis behavior of the piezoelectric materials seriously degraded the output performance of piezoelectric actuators. In this paper, to decrease such nonlinear effects and further improve the output performances of piezoelectric actuators, a modified nonlinear autoregressive moving average with exogenous inputs model, which could describe the rate-dependent hysteresis features of piezoelectric actuators was investigated. In the experiment, the different topologies of the proposed back propagation neural network algorithm were compared and the optimal topology was selected considering both the tracking precision and the structure complexity. The experimental results validated that the modified nonlinear autoregressive moving average with exogenous inputs model featured the hysteresis characteristics description ability with high precision, and the predicted motion matched well with the real trajectory. Then, the initial parameters of the back propagation neural network algorithm were further optimized by particle swarm optimization algorithm. The experimental results also verified that the proposed model based on particle swarm optimization–back propagation neural network algorithm was more accurate than that identified through the conventional back propagation neural network algorithm, and has a better predicting performance.

Statistically Optimized Back-Propagation Neural-Network Model and Its Application for Deformation Monitoring and Prediction of Concrete-Face Rockfill Dams

AbstractConcrete-face rockfill dams (CFRDs) are widely used in hydropower engineering. Deformation monitoring and safety operation of CFRDs is of great significance in ensuring the safety of human ...

Prediction of PM2.5 Mass Concentration Based on the Back Propagation (BP) Neural Network Optimized by t-Distribution Controlled Genetic Algorithm

In the past few years, human-health has been severely impacted from PM2.5 and has thus been a very popular topic of study. Furthermore, monitoring and control of PM2.5 are becoming one of the major environmental problems. In view of this, the present work targets at the establishment of an optimized BP neural network model based on t-distributed control genetic algorithm (BPM-TCG). Subsequently, in order to verify the performance of the proposed BPM-TCG, comparison analyses were performed among the prediction results generated from BPM-TCG, BP neural network model and BP-GA according to hourly data of PM2.5 mass concentration, analysis of corresponding meteorological factors, and gas pollutant concentrations from October 2017 to August 2018 at Qiqihar University monitoring point. The experimental results showed that BPM-TCG had the highest prediction accuracy and the best generalization ability, excellent applicability and commonality. Additionally, it may provide a basis for predicting the mass concentration of PM2.5, and thereby control and prevent the air pollution.