An Efficient Improved Adaptive Genetic Algorithm for Training Layered Feedforward Neural Networks

Layered feedforward neural network training algorithm based on traditional BP algorithm may lead to entrapment in local optimum, and has the defects such as slow convergent speed and unsatis-fied dynamic character which reduce the study ability of the network. This paper presents an improved adaptive genetic algorithm (IAGA) for training the neural network efficiently that uses a forward adaptive technique and takes the advantages of the network architecture. The experimental results show that our al-gorithm outperforms BP algorithm, BGA algorithm and AGA algorithm, and the dynamic character,training accuracy and efficiency proved greatly.

Solar Radiation Estimation Based on a New Combined Approach of Artificial Neural Networks (ANN) and Genetic Algorithms (GA) in South Algeria

When designing solar systems and assessing the effectiveness of their many uses,estimating sun irradiance is a crucial first step.This study examined three approaches(ANN,GA-ANN,and ANFIS)for estimating daily global solar radiation(GSR)in the south of Algeria:Adrar,Ouargla,and Bechar.The proposed hybrid GA-ANN model,based on genetic algorithm-based optimization,was developed to improve the ANN model.The GA-ANN and ANFIS models performed better than the standalone ANN-based model,with GA-ANN being better suited for forecasting in all sites,and it performed the best with the best values in the testing phase of Coefficient of Determination(R=0.9005),Mean Absolute Percentage Error(MAPE=8.40%),and Relative Root Mean Square Error(rRMSE=12.56%).Nevertheless,the ANFIS model outperformed the GA-ANN model in forecasting daily GSR,with the best values of indicators when testing the model being R=0.9374,MAPE=7.78%,and rRMSE=10.54%.Generally,we may conclude that the initial ANN stand-alone model performance when forecasting solar radiation has been improved,and the results obtained after injecting the genetic algorithm into the ANN to optimize its weights were satisfactory.The model can be used to forecast daily GSR in dry climates and other climates and may also be helpful in selecting solar energy system installations and sizes.

Development of an electrode intelligent design system based on adaptive fuzzy neural network and genetic algorithm

The coating on the electrodes contains many kinds of raw materials which affect significantly on the mechanical properties of deposited metals. It is still a problem how to predict and control the mechanical properties of deposited metals directly according to the components of coating on the electrodes. In this paper an electrode intelligent design system is developed by means of fuzzy neural network technology and genetic algorithm,, dynamic link library, object linking and embedding and multithreading. The front-end application and customer interface of the system is realized by using visual C ＋＋ program language and taking SQL Server 2000 as background database. It realizes series functions including automatic design of electrode formula, intelligent prediction of electrode properties, inquiry of electrode information, output of process report based on normalized template and electronic storage and search of relative files.

Research on Financial Distress Prediction with Adaptive Genetic Fuzzy Neural Networks on Listed Corporations of China

To design a multi-population adaptive genetic BP algorithm, crossover probability and mutation probability are self-adjusted according to the standard deviation of population fitness in this paper. Then a hybrid model combining Fuzzy Neural Network and multi-population adaptive genetic BP algorithm—Adaptive Genetic Fuzzy Neural Network (AGFNN) is proposed to overcome Neural Network’s drawbacks. Furthermore, the new model has been applied to financial distress prediction and the effectiveness of the proposed model is performed on the data collected from a set of Chinese listed corporations using cross validation approach. A comparative result indicates that the performance of AGFNN model is much better than the ones of other neural network models.

Using fuzzy neural networks for RMB/USD real exchange rate forecasting

In order to aim at improving the forecasting performance of the RMB/USD exchange rate, this paper proposes a new architecture of fuzzy neural networks based on fuzzy logic, and the method of point differential, which guarantees not only the direction of weight correction, but also the needed precision for the BP algorithm. In applying genetic algorithms for optimal performance, this approach, in the forecasting of the RMB/USD real exchange rate from 1994 to 2000, obviously outperforms typical BP Neural Networks and exhibits a higher capacity in regard to nonlinear, time-variablility, and illegibility of the exchange rate.

Vehicle Plate Number Localization Using Memetic Algorithms and Convolutional Neural Networks

This paper introduces the third enhanced version of a genetic algorithm-based technique to allow fast and accurate detection of vehicle plate numbers(VPLN)in challenging image datasets.Since binarization of the input image is the most important and difficult step in the detection of VPLN,a hybrid technique is introduced that fuses the outputs of three fast techniques into a pool of connected components objects(CCO)and hence enriches the solution space with more solution candidates.Due to the combination of the outputs of the three binarization techniques,many CCOs are produced into the output pool from which one or more sequences are to be selected as candidate solutions.The pool is filtered and submitted to a new memetic algorithm to select the best fit sequence of CCOs based on an objective distance between the tested sequence and the defined geometrical relationship matrix that represents the layout of the VPLN symbols inside the concerned plate prototype.Using any of the previous versions will give moderate results but with very low speed.Hence,a new local search is added as a memetic operator to increase the fitness of the best chromosomes based on the linear arrangement of the license plate symbols.The memetic operator speeds up the convergence to the best solution and hence compensates for the overhead of the used hybrid binarization techniques and allows for real-time detection especially after using GPUs in implementing most of the used techniques.Also,a deep convolutional network is used to detect false positives to prevent fake detection of non-plate text or similar patterns.Various image samples with a wide range of scale,orientation,and illumination conditions have been experimented with to verify the effect of the new improvements.Encouraging results with 97.55%detection precision have been reported using the recent challenging public Chinese City Parking Dataset(CCPD)outperforming the author of the dataset by 3.05%and the state-of-the-art technique by 1.45%.

A New Neuro-Fuzzy Adaptive Genetic Algorithm

Novel neuro-fuzzy techniques are used to dynamically control parameter settings ofgenetic algorithms (GAs).The benchmark routine is an adaptive genetic algorithm (AGA) that uses afuzzy knowledge-based system to control GA parameters.The self-learning ability of the cerebellar modelariculation controller (CMAC) neural network makes it possible for on-line learning the knowledge onGAs throughout the run.Automatically designing and tuning the fuzzy knowledge-base system,neuro-fuzzy techniques based on CMAC can find the optimized fuzzy system for AGA by the renhanced learningmethod.The Results from initial experiments show a Dynamic Parametric AGA system designed by theproposed automatic method and indicate the general applicability of the neuro-fuzzy AGA to a widerange of combinatorial optimization.

Study on Power Transformers Fault Diagnosis Based on Wavelet Neural Network and D-S Evidence Theory

>Transformer faults are quite complicated phenomena and can occur due to a variety of reasons.There have been several methods for transformer fault synthetic diagnosis,but each of them has its own limitations in real fault diagnosis applications.In order to overcome those shortcomings in the existing methods,a new transformer fault diagnosis method based on a wavelet neural network optimized by adaptive genetic algorithm(AGA)and an improved D-S evidence theory fusion technique is proposed in this paper.The proposed method combines the oil chromatogram data and the off-line electrical test data of transformers to carry out fault diagnosis.Based on the fusion mechanism of D-S evidence theory,the comprehensive reliability of evidence is constructed by considering the evidence importance,the outputs of the neural network and the expert experience.The new method increases the objectivity of the basic probability assignment(BPA)and reduces the basic probability assigned for uncertain and unimportant information.The case study results of using the proposed method show that it has a good performance of fault diagnosis for transformers.

Neural network fault diagnosis method optimization with rough set and genetic algorithms

Aiming at the disadvantages of BP model in artificial neural networks applied to intelligent fault diagnosis, neural network fault diagnosis optimization method with rough sets and genetic algorithms are presented. The neural network nodes of the input layer can be calculated and simplified through rough sets theory; The neural network nodes of the middle layer are designed through genetic algorithms training; the neural network bottom-up weights and bias are obtained finally through the combination of genetic algorithms and BP algorithms. The analysis in this paper illustrates that the optimization method can improve the performance of the neural network fault diagnosis method greatly.

Particle Swarm Optimization Algorithm vs Genetic Algorithm to Develop Integrated Scheme for Obtaining Optimal Mechanical Structure and Adaptive Controller of a Robot

The performances of Particle Swarm Optimization and Genetic Algorithm have been compared to develop a methodology for concurrent and integrated design of mechanical structure and controller of a 2-dof robotic manipulator solving tracking problems. The proposed design scheme optimizes various parameters belonging to different domains (that is, link geometry, mass distribution, moment of inertia, control gains) concurrently to design manipulator, which can track some given paths accurately with a minimum power consumption. The main strength of this study lies with the design of an integrated scheme to solve the above problem. Both real-coded Genetic Algorithm and Particle Swarm Optimization are used to solve this complex optimization problem. Four approaches have been developed and their performances are compared. Particle Swarm Optimization is found to perform better than the Genetic Algorithm, as the former carries out both global and local searches simultaneously, whereas the latter concentrates mainly on the global search. Controllers with adaptive gain values have shown better performance compared to the conventional ones, as expected.

The Prediction of Dynamic Coefficients for Tilting-Pad Journal Bearings Based on an AGA-BP Neural Network

To provide real-time dynamic coefficients of tilting-pad journal bearings( TPJBs) for the dynamic analysis of a rotor-bearing system accurately,an improved error back propagation( BP) neural network model is built in this paper.First,the samples are gained by solving the Reynolds equation with the finite differential method based on hydrodynamic lubrication theory.Secondly,the adaptive genetic algorithm( AGA) is applied to optimize the initial weights and thresholds of the BP neural network before training.Then,with a number of trial calculations,the optimum parameters for the neural network are obtained.Finally,an application case of the neural network is given as well as the results analysis.The results show that the AGA can efficiently prevent the training of the neural network from falling into a local minimum,and the AGA-BP neural network of dynamic coefficients for TPJBs built in this paper can meet the demand of engineering.

Finding roots of arbitrary high order polynomials based on neural network recursive partitioning method

This paper proposes a novel recursive partitioning method based on constrained learning neural networks to find an arbitrary number (less than the order of the polynomial) of (real or complex) roots of arbitrary polynomials. Moreover, this paper also gives a BP network constrained learning algorithm (CLA) used in root-finders based on the constrained relations between the roots and the coefficients of polynomials. At the same time, an adaptive selection method for the parameter d P with the CLA is also given. The experimental results demonstrate that this method can more rapidly and effectively obtain the roots of arbitrary high order polynomials with higher precision than traditional root-finding approaches.

Adaptive and intelligent prediction of deformation time series of high rock excavation slope

Deformation of high rock excavation slope has nonlinear evolution characters. It is very difficult to build mechanical model to describe this nonlinear evoution. A genetic-neural network model has been initially proposed for adaptive and intelligent prediction of deformation of slopes, which used artificial neural network to represent nonlinear evoution of sloPe deformation. Number 0f history points of displacement inputted to the model, topologies of neural network, and learning process of model were adaptive and automatically determined using genetic algorithm. The obtained model was thus optimal at global range, and gave predictions of horizontal displacement at succedent three months for the three measurement points with average relative error of 1. 4 % compared with the measured values. Results from one step prediction and multi-step prediction were combined with the measurements.

基于改进实数编码遗传算法的神经网络超参数优化

针对神经网络超参数优化效果差、容易陷入次优解和优化效率低的问题,提出一种基于改进实数编码遗传算法(IRCGA)的深度神经网络超参数优化算法——IRCGA-DNN(IRCGA for Deep Neural Network)。首先,采用实数编码方式表示超参数的取值,使超参数的搜索空间更灵活;然后,引入分层比例选择算子增加解集多样性;最后,分别设计了改进的单点交叉和变异算子,以更全面地探索超参数空间,提高优化算法的效率和质量。基于两个仿真数据集,验证IRCGA-DNN的毁伤效果预测性能和收敛效率。实验结果表明,在两个数据集上,与GA-DNN(Genetic Algorithm for Deep Neural Network)相比,所提算法的收敛迭代次数分别减少了8.7%和13.6%,均方误差(MSE)相差不大;与IGA-DNN(Improved GA-DNN)相比,IRCGA-DNN的收敛迭代次数分别减少了22.2%和13.6%。实验结果表明,所提算法收敛速度和预测性能均更优,能有效处理神经网络超参数优化问题。

基于小波变换与IAGA-BP神经网络的短期风电功率预测

为提高风功率预测精度,减轻输出风能波动性对风电并网不利影响,提出了基于WT-IAGA-BP神经网络的短期风电功率预测方法。利用风速分区、3σ准则及拉格朗日插值法清洗风电场历史数据;其次,依据小波重构误差,选择db4小波分别提取风速、风向、历史风功率的不同频率特征信号,并引入改进自适应遗传算法(IAGA)对各序列BP神经网络的初始权值与阈值寻优,使用Sigmiod函数通过适应度值自适应改变交叉概率与变异概率;构建各序列的WT-IAGA-BP模型对短期风功率组合预测。通过仿真分析,并与ELM、IAGA-BP、WT-ELM及WT-LSSVM方法对比,验证该方法具有更高的预测精度和更好的预测性能。

基于深度神经网络的永磁直线电机仿真与优化

针对永磁直线同步电机(permanent magnet linear synchronous machine,PMLSM)有限元仿真模型的计算时间长,不能直观地显示结构参数与输出推力的关系,无法指导电机结构参数优化等问题,提出基于子域解析法和深度神经网络算法的PMLSM改进仿真模型,根据麦克斯韦方程组计算得到电机的磁通密度、空载反电势等性能数据,结合深度神经网络算法拟合出电机结构参数与输出推力的非线性关系。基于此模型,使用自适应遗传算法对PMLSM的推力密度进行优化,并与有限元仿真结果对比。结果表明:PMLSM改进仿真模型的计算速度是有限元模型的87.1倍,推力计算结果与有限元结果的平均误差为2.87%,优化后的电机推力密度提高了5.7%。

基于AGA-RBF神经网络模型的叶绿素a质量浓度预测研究

叶绿素a质量浓度是预测湖泊水华形成的重要影响因子,但常用的径向基(radial basis function,RBF)神经网络存在容易陷入局部极值,导致预测精度欠佳.针对这一问题,采用自适应遗传算法(adaptive genetic algorithm,AGA)对RBF神经网络进行优化,构建基于AGA-RBF神经网络预测模型,以莆田东圳水库为应用案例,对叶绿素a质量浓度进行预测,通过采集到的数据对预测模型进行仿真,对比均方根误差(RMSE)、相对误差(RE)以及平均相对误差(MRE),验证改进后的AGA-RBF模型具有更好的预测精度,以期对叶绿素a质量浓度进行长期预测.

基于量测数据处理的中低压配电网线损分析方法研究

为了提升电力网络中低压配电网网络线损的预测精度,文中对神经网络的结构、训练方法等基本理论进行了研究。针对传统神经网络在复杂结构下训练时梯度消失和陷入局部最优的现象,引入了一种自适应遗传算法(AGA)。该算法通过网络的基本结构来确定染色体上的基因位数,使用一种可变交叉、变异概率策略,从而有效提升了训练时的稳定性及效率。基于配电网的关键指标体系,对某10 kV配电网络完成数据采集,并使用量测数据进行了算法的性能仿真实验。实验结果表明,在相同的迭代条件下,改进后的算法相比传统神经网络算法对330条配电线路的平均预测精度提高了2.20%。此外,算法在迭代过程中的稳定性更强,即使在更低的目标精度下,也不会出现过拟合现象。

基于自适应遗传优化神经网络的航空装备故障诊断

针对改进反向传播神经网络在航空装备故障诊断中存在的缺陷和不足,将自适应遗传算法与改进反向传播算法相结合构成混合算法用以训练人工神经网络。以改进反向传播神经网络的初始权值空间为切入点,利用改进遗传操作对其开展多点自适应遗传优化,然后运用改进反向传播算法开展局部精确搜索,最终实现全局最优。以某型飞机电气控制盒和某型飞机自动驾驶仪飞行控制盒的故障诊断为例对所提算法进行仿真研究,结果表明自适应遗传算法与改进反向传播算法相结合的方法收敛速度快、诊断精度高,对于具有复杂输入输出关系的工程样本具有较好的诊断结果。