Comments on “Backpropagation Algorithms for a Broad Class of Dynamic Networks”

In a recent paper, De JesÚs proposed a general framework for describing dynamic neural networks. Gradient and Jacobian calculations were discussed based on backpropagation-through-time (BPTT) algorithm and real-time recurrent learning (RTRL). Some errors in the paper of De JesÚs bring difficulties for other researchers who want to implement the algorithms. This comments paper shows the critical parts of the publication and gives errata to facilitate understanding and implementation.      

基于神经网络改进算法的飞控系统故障诊断

应用神经网络技术对复杂的飞行控制系统进行故障诊断对提高飞机的可靠性和容错能力具有重要意义。为了提高网络的学习效率和稳定性,该文提出一种改进的径向基神经网络学习算法,使用混合共轭梯度优化算法对网络参数进行调整。利用神经网络对某型飞机的飞行控制系统进行故障诊断,仿真结果表明该神经网络具有较强的故障识别能力。     

ANASA-a stochastic reinforcement algorithm for real-valued neuralcomputation

This paper introduces ANASA (adaptive neural algorithm of stochastic activation), a new, efficient, reinforcement learning algorithm for training neural units and networks with continuous output. The proposed method employs concepts, found in self-organizing neural networks theory and in reinforcement estimator learning algorithms, to extract and exploit information relative to previous input pattern presentations. In addition, it uses an adaptive learning rate function and a self-adjusting stochastic activation to accelerate the learning process. A form of optimal performance of the ANASA algorithm is proved (under a set of assumptions) via strong convergence theorems and concepts. Experimentally, the new algorithm yields results, which are superior compared to existing associative reinforcement learning methods in terms of accuracy and convergence rates. The rapid convergence rate of ANASA is demonstrated in a simple learning task, when it is used as a single neural unit, and in mathematical function modeling problems, when it is used to train various multilayered neural networks.     

安全强化学习算法及其在CPS智能控制中的应用

信息物理系统(cyber-physicalsystem,CPS)的安全控制器设计是一个热门研究方向,现有基于形式化方法的安全控制器设计存在过度依赖模型、可扩展性差等问题.基于深度强化学习的智能控制可处理高维非线性复杂系统和不确定性系统,正成为非常有前景的CPS控制技术,但是缺乏对安全性的保障.针对强化学习控制在安全性方面的不足,围绕一个工业油泵控制系统典型案例,开展安全强化学习算法和智能控制应用研究.首先,形式化了工业油泵控制的安全强化学习问题,搭建了工业油泵仿真环境;随后,通过设计输出层结构和激活函数,构造了神经网络形式的油泵控制器,使得油泵开关时间的线性不等式约束得到满足;最后,为了更好地权衡安全性和最优性控制目标,基于增广拉格朗日乘子法设计实现了新型安全强化学习算法.在工业油泵案例上的对比实验表明,该算法生成的控制器在安全性和最优性上均超越了现有同类算法.在进一步评估中,所生成神经网络控制器以90%的概率通过了严格形式化验证;同时,与理论最优控制器相比实现了低至2%的最优目标值损失.所提方法有望推广至更多应用场景,实例研究的方案有望为安全智能控制和形式化验证领域其他学者提供借鉴.     

An Improved Active Noise Control Algorithm Without Secondary Path Identification Based on the Frequency-Domain Subband Architecture

Common active noise control (ANC) algorithms need to identify the secondary path transfer functions between the output of the adaptive control filters and the error sensors, and then use the information to guide the direction of control filter coefficient updating. Recently, Zhou proposed an ANC algorithm without secondary path identification, and we improve their algorithm in this paper. For single-tone and narrowband noise control, the direction of control filter coefficient updating has four choices 180$^{circ}$, 0$^{circ}$, and $pm {hbox {90}}^circ$. We test the four update directions and select the one that works the best. If for all four update directions, the system converges slowly or diverges, we adjust the step size and test again with the new step size. The multitone and broadband noise control problems are converted into several single-tone and narrowband noise control problems by means of a frequency-domain delayless subband architecture. Compared to Zhou's algorithm, our proposed method yields good performance and converges quickly. Simulation results confirm the effectiveness of our proposed algorithm.      

An adaptive fuzzy neural network for MIMO system modelapproximation in high-dimensional spaces

An adaptive fuzzy system implemented within the framework of neural network is proposed. The integration of the fuzzy system into a neural network enables the new fuzzy system to have learning and adaptive capabilities. The proposed fuzzy neural network can locate its rules and optimize its membership functions by competitive learning, Kalman filter algorithm and extended Kalman filter algorithms. A key feature of the new architecture is that a high dimensional fuzzy system can be implemented with fewer number of rules than the Takagi-Sugeno fuzzy systems. A number of simulations are presented to demonstrate the performance of the proposed system including modeling nonlinear function, operator's control of chemical plant, stock prices and bioreactor (multioutput dynamical system).     

Wavelet CMAC for MIMO uncertain nonlinear systems using a modified social ski-driver algorithm

This study uses a Mexican hat wavelet membership function for a cerebellar model articulation controller (CMAC) to develop a more efficient adaptive controller for multiple input multiple output (MIMO) uncertain nonlinear systems. The main controller is called the adaptive Mexican hat wavelet CMAC (MWCMAC), and an auxiliary controller is used to remove the residual error. For the MWCMAC, the online learning laws are derived from the gradient descent method. In addition, the learning rate values are very important and have a great impact on the performance of the control system; however, they are difficult to choose accurately. Therefore, a modified social ski driver (SSD) algorithm is proposed to find optimal learning rates for the control parameters. Finally, a magnetic ball levitation system and a nine-link biped robot are used to illustrate the effectiveness of the proposed SSD-based MWCMAC control system. The comparisons with other existing control algorithms have shown the superiority of the proposed control system.     

A learning automata-based algorithm for determination of the number of hidden units for three-layer neural networks

There is no method to determine the optimal topology for multi-layer neural networks for a given problem. Usually the designer selects a topology for the network and then trains it. Since determination of the optimal topology of neural networks belongs to class of NP-hard problems, most of the existing algorithms for determination of the topology are approximate. These algorithms could be classified into four main groups: pruning algorithms, constructive algorithms, hybrid algorithms and evolutionary algorithms. These algorithms can produce near optimal solutions. Most of these algorithms use hill-climbing method and may be stuck at local minima. In this article, we first introduce a learning automaton and study its behaviour and then present an algorithm based on the proposed learning automaton, called survival algorithm, for determination of the number of hidden units of three layers neural networks. The survival algorithm uses learning automata as a global search method to increase the probability of obtaining the optimal topology. The algorithm considers the problem of optimization of the topology of neural networks as object partitioning rather than searching or parameter optimization as in existing algorithms. In survival algorithm, the training begins with a large network, and then by adding and deleting hidden units, a near optimal topology will be obtained. The algorithm has been tested on a number of problems and shown through simulations that networks generated are near optimal.     

Nonlinear optimized adaptive trajectory control of helicopter

This paper attempts to develop an optimized adaptive trajectory control system for helicopters based on the dynamic inversion method. This control algorithm is implemented by three time-scale separation architectures. Pseudo control hedging (PCH) is used to protect the adaptive element from actuator saturation nonlinearities and also from the inner-outer-loop interaction. In addition, to augment the attitude control system, two online adaptive architectures that employ a neural network are used. By tuning the neural network based on the system model, a better and faster learning will be achieved, but this is a frustrating and time consuming process. Due to complexity in accurate tuning of neural network, this paper introduces a non-dominated sorting genetic algorithm II (NSGA-II) for off-line optimization of the neural network. Thus, in the proposed method, the neural network can compensate model inversion error caused by the deficiency of full knowledge of helicopter dynamics more accurately. The effectiveness of proposed method is demonstrated by numerical simulations.     

Adaptive sliding mode approach for learning in a feedforward neural network

An adaptive learning algorithm is proposed for a feedforward neural network. The design principle is based on the sliding mode concept. Unlike the existing algorithms, the adaptive learning algorithm developed does not require a prioriknowledge of upper bounds of bounded signals. The convergence of the algorithm is established and conditions given. Simulations are presented to show the effectiveness of the algorithm.     

Implementing online natural gradient learning: problems and solutions

The online natural gradient learning is an efficient algorithm to resolve the slow learning speed and poor performance of the standard gradient descent method. However, there are several problems to implement this algorithm. In this paper, we proposed a new algorithm to solve these problems and then compared the new algorithm with other known algorithms for online learning, including Almeida-Langlois-Amaral-Plakhov algorithm (ALAP), Vario-/spl eta/, local adaptive learning rate and learning with momentum etc., using sample data sets from Proben1 and normalized handwritten digits, automatically scanned from envelopes by the U.S. Postal Services. The strong and weak points of these algorithms were analyzed and tested empirically. We found out that using the online training error as the criterion to determine whether the learning rate should be changed or not is not appropriate and our new algorithm has better performance than other existing online algorithms.     

A MULTISWARM COMPETITIVE PARTICLE SWARM ALGORITHM FOR OPTIMIZATION CONTROL OF AN ETHYLENE CRACKING FURNACE

A fuzzy C-means (FCM) multiswarm competitive particle swarm optimization (FCMCPSO) algorithm is proposed, in which FCM clustering is used to divide swarms adaptively into different clusters. The large-scale swarms are according to the standard particle swarm optimization (PSO) algorithm, whereas the small-scale swarms search randomly in the neighborhood of the optimal solution to increase the probability of jumping out of the local optimization point. Within every cluster, the adaptive value gained by competitive learning is respectively found and arranged in order. Swarms of small adaptive value were integrated with the neighboring swarms of large adaptive value to search the optimal solution competitively by the swarms. The algorithm's validity was tested by benchmark functions and compared with other PSO algorithms. Furthermore, an integrated FCMCPSO-radial basis function neural network was studied for nonlinear system modeling and intelligent optimization control of cracking depth of an ethylene cracking furnace application in a chemical process.     

基于知识引导的自适应动态多模态差分进化算法

为充分利用问题求解过程知识,提升动态多模态优化算法的计算资源利用效率,提出一种基于知识引导的自适应动态多模态差分进化算法.首先,利用自组织映射神经网络实现种群自聚类,形成稳定的小生境;然后,通过对种群全局知识和个体邻域知识的综合学习,设计一种基于知识引导的自适应差分进化算法,在对种群进化状态进行实时监测和分析的基础上,逐层递进地引导不同种群个体自适应地选择最符合当前进化需求的变异方式,提升种群搜索效率,平衡种群多样性与收敛性;最后,针对问题动态特性,设计一种基于历史动态过程知识引导的自适应动态响应机制,通过对历史寻优经验的自适应学习,预测生成新环境下的潜在精英个体,引导种群实现精准快速的多峰定位.实验结果表明,所提出算法能够有效解决动态多模态优化问题,且在不同动态环境设置下其求解性能均优于对比算法.     

A fuzzy neural network based on fuzzy hierarchy error approach

This paper presents a novel fuzzy neural network which consists of an antecedent network and a consequent network. The antecedent network matches the premises of the fuzzy rules and the consequent network implements the consequences of the rules. In the network learning and training phase, a concise and effective algorithm based on the fuzzy hierarchy error approach is proposed to update the parameters of the network. This algorithm is simple to implement and it does not require as many calculations as some other classic neural network learning algorithms. A model reference adaptive control structure incorporating the proposed fuzzy neural network is studied. Simulation results of a cart-pole balancing system demonstrate the effectiveness of the proposed method     

Robust Control of an LUSM-Based $Xhbox{--}Yhbox{--}theta$ Motion Control Stage Using an Adaptive Interval Type-2 Fuzzy Neural Network

The robust control of a linear ultrasonic motor based $Xhbox{--}Yhbox{--}theta$ motion control stage to track various contours is achieved by using an adaptive interval type-2 fuzzy neural network (AIT2FNN) control system in this study. In the proposed AIT2FNN control system, an IT2FNN, which combines the merits of an interval type-2 fuzzy logic system and a neural network, is developed to approximate an unknown dynamic function. Moreover, adaptive learning algorithms are derived using the Lyapunov stability theorem to train the parameters of the IT2FNN online. Furthermore, a robust compensator is proposed to confront the uncertainties including the approximation error, optimal parameter vectors, and higher order terms in Taylor series. To relax the requirement for the value of lumped uncertainty in the robust compensator, an adaptive lumped uncertainty estimation law is also investigated. In addition, the circle and butterfly contours are planned using a nonuniform rational B-spline curve interpolator. The experimental results show that the contour tracking performance of the proposed AIT2FNN is significantly improved compared with the adaptive type-1 FNN. Additionally, the robustness to parameter variations, external disturbances, cross-coupled interference, and frictional force can also be obtained using the proposed AIT2FNN.      

Data mining in forecasting PVT correlations of crude oil systems based on Type1 fuzzy logic inference systems

Pressure–volume–temperature properties are very important in the reservoir engineering computations. There are many empirical approaches for predicting various PVT properties based on empirical correlations and statistical regression models. Last decade, researchers utilized neural networks to develop more accurate PVT correlations. These achievements of neural networks open the door to data mining techniques to play a major role in oil and gas industry. Unfortunately, the developed neural networks correlations are often limited, and global correlations are usually less accurate compared to local correlations. Recently, adaptive neuro-fuzzy inference systems have been proposed as a new intelligence framework for both prediction and classification based on fuzzy clustering optimization criterion and ranking. This paper proposes neuro-fuzzy inference systems for estimating PVT properties of crude oil systems. This new framework is an efficient hybrid intelligence machine learning scheme for modeling the kind of uncertainty associated with vagueness and imprecision. We briefly describe the learning steps and the use of the Takagi Sugeno and Kang model and Gustafson–Kessel clustering algorithm with K-detected clusters from the given database. It has featured in a wide range of medical, power control system, and business journals, often with promising results. A comparative study will be carried out to compare their performance of this new framework with the most popular modeling techniques, such as neural networks, nonlinear regression, and the empirical correlations algorithms. The results show that the performance of neuro-fuzzy systems is accurate, reliable, and outperform most of the existing forecasting techniques. Future work can be achieved by using neuro-fuzzy systems for clustering the 3D seismic data, identification of lithofacies types, and other reservoir characterization.     

周期时变系统的鲁棒自适应重复控制

针对周期时变系统,提出一种鲁棒自适应重复控制方法.该方法利用周期学习律估计周期时变参数,并结合鲁棒自适应方法处理非周期不确定性.与现有重复控制不同的是,在控制器设计中引入了新变量—周期数,利用周期系统的重复特性,使界的逼近误差随周期数的增加而逐渐减少,保证了系统的全局渐近稳定性.同时将该方法应用于一类非线性参数化系统,使系统在非参数化扰动的情形下,输出误差仍能收敛于0,倒立摆模型的仿真验证了此结果.该设计方法适用于消除神经网络逼近误差对重复控制系统的影响,理论证明了基于神经网络的鲁棒自适应重复控制系统中所有变量的有界性和输出误差的渐近收敛性,关于机械臂模型的仿真结果验证了受控系统具有良好的跟踪性能.     

An intelligent biological inspired evolutionary algorithm for the suppression of incubator interference in premature infants ECG

Electromagnetic interference produced by the incubator medical equipments may interrupt or degrade the premature infant’s electrocardiography (ECG) signal. The premature infant’s ECG is usually contaminated by an interference caused by the incubator devices. The interference cancellation system is designed using an adaptive learning ability of artificial neural network Levenberg–Marquardt (LM) algorithm. In this paper the swarm intelligent-LM algorithm is used for the electromagnetic interference cancellation in infant ECG signal. The swarm intelligent algorithm is used for the optimization by selecting the optimized number of neurons in the hidden layer, learning rate and momentum factor of the neural network. Also, this paper presents a comparison of residual mean square error (RMSE) values for neural network trained by LM algorithm, hybrid genetic-LM algorithm and hybrid swarm intelligent-LM algorithm. The LM algorithm is used for the weight updating and reducing the content of electromagnetic interference noise present in the signal. The performance analysis of the proposed noise cancellation approach is compared with gradient based and evolutionary based algorithms. The result analysis shows that the interferences in infant ECG signal is removed successfully using the proposed approach.     

The layer-wise method and the backpropagation hybrid approach tolearning a feedforward neural network

Feedforward neural networks (FNNs) have been proposed to solve complex problems in pattern recognition and classification and function approximation. Despite the general success of learning methods for FNNs, such as the backpropagation (BP) algorithm, second-order optimization algorithms and layer-wise learning algorithms, several drawbacks remain to be overcome. In particular, two major drawbacks are convergence to a local minima and long learning time. We propose an efficient learning method for a FNN that combines the BP strategy and optimization layer by layer. More precisely, we construct the layer-wise optimization method using the Taylor series expansion of nonlinear operators describing a FNN and propose to update weights of each layer by the BP-based Kaczmarz iterative procedure. The experimental results show that the new learning algorithm is stable, it reduces the learning time and demonstrates improvement of generalization results in comparison with other well-known methods.