一种应用模糊神经网络对完全电机定子进行故障保护的新方法

该文研究一种由模糊逻辑和人工神经网络(ANN)组成的，对发电机定子线圈进行故障保护的综合方法。该方法只采用发电机机端电压、电流信号，并对这些信号进行特征提取，然后用到FNN的故障诊断和定位上。该技术由两个阶段组成：基于人工神经网络(ANN)的故障类型分类和基于一个包括模糊逻辑以及人工神经网络的综合网络的进行定子绕组故障精确定位。     

改善系统暂态稳定性的HVDC模糊神经控制器

提出了一种HVDC在线模糊神经控制器以提高交直流系统的暂态稳定性。该控制器的特点是结合了模糊系统处理复杂和不确定性问题及神经网络具有自学习能力的优点,选取整流侧交流母线电压相位误差及其变化率作为模糊逻辑控制部分的输入,其输出结果作为神经网络的一个输入,采用改进BP算法进行在线训练神经网络,神经网络的输出用来修正整流器的触发角,并利用NETOMAC软件对控制器主要参数进行了离线优化。仿真结果表明该控制器能有效地抑制有功功率振荡,改善发电机的功角特性,提高系统的暂态稳定性。     

基于ANN和模糊控制相结合的电力负荷短期预测方法

提出了一种ANN与模糊控制(FC)相结合的电力负荷短期预测方法。通过优化训练样本、变步长和变动量因子来改进BP算法,并采用在线自调整因子的模糊控制对预测误差进行在线智能修正。算例计算表明ANN与模糊控制相结合提高了预测精度。     

Partial discharge pulse pattern recognition using an inductive inference algorithm

This paper presents a novel approach in the area of time dependent partial discharge (PD) pulse pattern recognition, to applications based on the inductive learning (decision tree) approach. Different attributes based on pulse shape analysis are used as representative feature vectors that can accurately capture the unique and salient characteristics of the PD pulse shape. In the training phase, a decision tree is developed to relate the pulse shape with the cavity size by using inductive machine learning. The C4.5 machine learning algorithm is deployed to realize the tree using the training data, since it has the capability of inferring the rules and to produce the tree in terms of continuous features. During testing, the cavity size is recognized by means of the rules extracted from the decision tree. The dependency between the features and the classes are examined using the mutual information approach. The proposed algorithm possesses the inherent advantage of explaining the result via the self-created rule base as demonstrated by the results obtained. Those self-created rules can be employed as the basis for applying a fuzzy expert system for the classification of void sizes in an easily interpreted fashion.     

基于 BP 网络的锅炉过热汽温系统动态特性的辨识

提出了基于ＢＰ网络的锅炉过热器温度控制系统动态特性的辨识方法。应用可变斜率的改进的ＢＰ算法，极大地加速神经网络的学习速度。在伪随机信号作用下，对实际系统与辨识模型的响应进行比较，结果令人满意。     

Power quality disturbance classification using a statistical and wavelet-based Hidden Markov Model with Dempster–Shafer algorithm

A novel approach for power quality disturbance classification using Hidden Markov Model (HMM) and Wavelet Transform (WT) is proposed in this paper. The energy distributions of the signals are obtained by wavelet transform at each decomposition level which are then used for training HMM. The statistical parameters of the extracted disturbance features are used to initialize the HMM training matrices which maximize the classification accuracy. Fifteen different types of power quality disturbances are considered for training and evaluating the proposed method. The Dempster–Shafer algorithm is also used for improving the accuracy of classification. In addition, the effect of the noise is studied and the performance of a denoising method is also investigated. Simulation results in a 34-bus distribution system verify the performance and reliability of the proposed approach. Also the results obtained for practical data prove the capability of the proposed method for implementing in experimental systems.     

神经网络短期负荷预报模糊化改进

提出了一种简洁实用的基于模糊集的神经网络电力系统短期负荷预报方法 ,计及了天气和日期特征量 ,具有较高的预测精度。采用两种学习算法 ,用实际数据对ANN进行了训练 ,通过比较得出了优化模型。计算实例表明用该方法是可行的、有效的     

基于RSVP的面向不良信息检测人机协作系统研究

针对复杂的环境背景下不良信息的快速准确检测问题,提出了基于快速序列视觉呈现( rapid serial visual presentation, RSVP)的面向不良信息检测人机协作系统。 首先利用快速佩戴便携式采集系统采集了 12 名受试者的脑电数据;然后采用 Mallat 算法提取较低维度的时频特征,使用人工神经网络(ANN)和支持向量机(SVM)两种模型分类对比;最后在训练集中引入 不同次数的叠加平均数据以改善模型的分类性能。 实验结果表明,在含有 3 个目标的 60 张图像中平均正确输出至少 2 张目 标,AUC 值达到了 0. 9。 该系统在小批量数据集、环境变化复杂的不良图像信息检测中有着良好的性能,相较于人工检测提高 了效率。     

Hybrid chaos search genetic algorithm and meta-heuristics method for short-term load forecasting

A hybrid chaos search genetic algorithm (CGA) /fuzzy system (FS), simulated annealing (SA) and neural fuzzy network (NFN) method for load forecasting is presented in this paper. A fuzzy hyper-rectangular composite neural networks (FHRCNNs) was used for the initial load forecasting. Then, we used CGAFS and SA to find the optimal solution of the parameters of the FHRCNNs, instead of back-propagation (BP) (including parameters such as synaptic weights, biases, membership functions, sensitivity factor in membership functions and adjustable synaptic weights). First, the CGAFS generates a set of feasible solution parameters and then puts the solution into the SA. The CGAFS has good global optimal search capabilities, but poor local optimal search capabilities. The SA method on the other hand has good local optimal search capabilities. We combined both methods to try and obtain both advantages, and in doing so eliminate the drawback of the traditional artificial neural networks (ANN) training by BP (where the weights and biases are always trapped into a local optimum, which then leads the solution to sub-optimization). Finally, we used the CGAFS and SA combined with NFN (CGAFSSA–NFN) to see if we could improve the quality of the solution, and if we actually could reduce the error of load forecasting. The proposed CGAFSSA–NFN load forecasting scheme was tested using the data obtained from a sample study, including 1 year, 1 week and 24-h time periods. The proposed scheme was then compared with ANN, evolutionary programming combined with ANN (EP–ANN), genetic algorithm combined with ANN (GA–ANN), and CGAFSSA–NFN. The results demonstrated the accuracy of the proposed load-forecasting scheme.     

A novel fuzzy neural network based distance relaying scheme

This paper presents a new approach to distance relaying using fuzzy neural network (FNM). The FNN can be viewed either as a fuzzy system, a neural network or fuzzy neural network. The structure is seen as a neural network for training and a fuzzy viewpoint is utilized to gain insight into the system and to simplify the model. The number of rules is determined by the data itself and therefore a smaller number of rules is produced. The network is trained with the backpropagation algorithm. A pruning strategy is applied to eliminate the redundant rules and fuzzification neurons, consequently a compact structure is achieved. The classification and location tasks are accomplished by using different FNN's. Once the fault type is identified by the FNN classifier the selected fault locating FNN estimates the location of the fault accurately. Normalized peaks of fundamental voltage and current waveforms are considered as inputs to all the networks and an additional input derived from the DC component is fed to fault locating networks. The peaks and DC component are extracted from sampled signals by the EKF. Test results show that the new approach provides robust and accurate classification/location of faults for a variety of power system operating conditions even with resistance in the fault path     

基于模糊神经网络的永磁同步电动机矢量控制系统

该文提供了一种基于自适应模糊神经网络的永磁同步电动机（PMSM）矢量控制系统速度控制器的实施方案。模糊神经网络控制器（FNNC）包括神经网络控制器（NC）和模糊逻辑控制器（FC）两部分，它同时具有神经网络自学习能力和模糊逻辑在处理不确定信息方面的能力。人工神经网络（ANN）的初始权值和阈值通过离线训练的方式获得。在实际的运行过程中，利用模糊控制器的输出对神经网络的权值和阈值进行实时调整。仿真结果表明利用所提出的模糊神经网络来建立永磁同步电动机矢量控制系统的速度控制器，当电机参数改变或者受到外部扰动时，系统具有良好的动态特性。     

Implementation of an artificial-neural-network-based real-time adaptive controller for an interior permanent-magnet motor drive

This paper presents the implementation of an artificial-neural-network (ANN)-based real-time adaptive controller for accurate speed control of an interior permanent-magnet synchronous motor (IPMSM) under system uncertainties. A field-oriented IPMSM model is used to decouple the flux and torque components of the motor dynamics. The initial estimation of coefficients of the proposed ANN speed controller is obtained by offline training method. Online training has been carried out to update the ANN under continuous mode of operation. Dynamic backpropagation with the Levenburg-Marquardt algorithm is utilized for online training purposes. The controller is implemented in real time using a digital-signal-processor-based hardware environment to prove the feasibility of the proposed method. The simulation and experimental results reveal that the control architecture adapts and generalizes its learning to a wide range of operating conditions and provides promising results under parameter variations and load changes.     

Observer-Based Speed Estimation Method for Sensorless Vector Control Using Artificial Neural Network

This paper presents a novel approach to sensorless vector control of induction motor drives. The method is based on an adaptive flux observer in the rotorspeed reference frame in which an artificial neural network (ANN) is employed to modify the estimated rotor flux to improve the performance of speed estimation. The adopted ANN is a feed-forward neural network identified off-line. It uses the backpropagation learning process to update their weights. The data for training are obtained from a computer simulation and experimental data file of a vector control system. Then, the estimated rotor flux is used in the speed estimation that will feedback to the vector control system. The proposed method has the advantages of better accuracy at low speed range and speed following under heavy loads. Experimental results show the effectiveness of the proposed method.     

Investigation of extended fuzzy reasoning and neural classification for load-shedding prediction to prevent voltage instability

This paper first illustrates an intelligent load-shedding approach, which makes use of knowledge base and extended fuzzy reasoning, for curative control to prevent a power system from moving towards a dynamic voltage-insecurity state following disturbances. Using the same training patterns, layered artificial neural networks are then employed to perform the same graded classification. From the investigation, it is found that there is difficulty for even well trained artificial neural networks to return consistently satisfactory results for previously untrained cases.     

An Integrated Approach to Economic Emission Load Dispatching Using Neural Network and Goal-Attainment Methods

This paper presents an integrated approach comprising artificial neural network (ANN) and goal-attainment (GA) methods to economic emission load dispatching (EELD) in power system operation and scheduling phases. The GA method, which is one of the most powerful tools for multiobjective optimization problems, is quantitatively performed to grasp trade-off relations between the two conflicting objectives (economy and emission impact) in the training set creation phase. Finally, a radial basis function ANN is trained by the orthogonal least squares learning algorithm to reach the optimal generations. The ANN models so developed have been tested to solve EELD problem on 6-bus and 71-bus power systems. The test results demonstrate that the proposed approach is capable of obtaining well-coordinated optimal solutions suitable both in accuracy and speed while allowing flexibility in the operation of the generators.     

Detection and classification of single and combined power quality disturbances using fuzzy systems oriented by particle swarm optimization algorithm

In this paper, a new approach for the detection and classification of single and combined power quality (PQ) disturbances is proposed using fuzzy logic and a particle swarm optimization (PSO) algorithm. In the proposed method, suitable features of the waveform of the PQ disturbance are first extracted. These features are extracted from parameters derived from the Fourier and wavelet transforms of the signal. Then, the proposed fuzzy system classifies the type of PQ disturbances based on these features. The PSO algorithm is used to accurately determine the membership function parameters for the fuzzy systems. To test the proposed approach, the waveforms of the PQ disturbances were assumed to be in the sampled form. The impulse, interruption, swell, sag, notch, transient, harmonic, and flicker are considered as single disturbances for the voltage signal. In addition, eight possible combinations of single disturbances are considered as the PQ combined types. The capability of the proposed approach to identify these PQ disturbances is also investigated, when white Gaussian noise, with various signal to noise ratio (SNR) values, is added to the waveforms. The simulation results show that the average rate of correct identification is about 96% for different single and combined PQ disturbances under noisy conditions.     

神经网络应用于电力变压器故障诊断

将电力变压器油气分析法作为检测数据来源，利用神经网络这一强有力的故障诊断工具，有效地诊断电力变压内部故障。仿真结果表明，用神经网络诊断变压器故障具有更加优秀的性能。文中，作者采用的ＢＰ网络模型及算法，并对网络训练过程中一些技巧问题进行了讨论。     

一种改进共轭梯度在线学习滑模控制算法

对于一类基于T-S模糊模型描述的非线性不确定系统,滑模控制算法的稳态误差和动态品质与T-S模糊算法描述模型的准确度相关,利用最小二乘支持向量机算法(LSSVM)学习T-S模糊模型可以很好的逼近实际系统.但是由于LSSVM算法对数据量有一定要求,而且学习速度比较慢,对要求动态响应较高或者内存较小的系统不适用.提出了一种基于改进共轭梯度在线学习算法学习T-S模型,可以实时逼近实际模型,配合滑模控制算法可以达到控制系统的渐进稳定.在不同误差条件下对该控制算法进行仿真实验,在随机误差幅值100以内,系统稳态误差为0.01,同时对时变误差表现出快速的稳定特性,显示了该控制算法较强的鲁棒性.最后,实验还对随机误差幅值为500的系统验证了T-S模型对于系统学习数据的随机误差具有去噪能力.     

乙醇-水溶液电化学浸渍体系的神经网络优化

乙醇-水溶液体系电化学浸渍具有很高的实用价值。采用人工神经网络对实验数据进行建模。采用基于学习参数模糊自适应调整的误差反向传播算法提高网络训练速度。适当拓宽实验各因素的水平范围,经过不同因素、不同水平间的组合模拟,利用建立的神经网络模型预测出二元电化学浸渍体系的适当工艺条件。