Improving market clearing price prediction by using a committee machine of neural networks

Predicting market clearing prices is an important but difficult task, and neural networks have been widely used. A single neural network, however, may misrepresent part of the input-output data mapping that could have been correctly represented by different networks. The use of a "committee machine" composed of multiple networks can in principle alleviate such a difficulty. A major challenge for using a committee machine is to properly combine predictions from multiple networks, since the performance of individual networks is input dependent due to mapping misrepresentation. This paper presents a new method in which weighting coefficients for combining network predictions are the probabilities that individual networks capture the true input-output relationship at that prediction instant. Testing of the New England market cleaning prices demonstrates that the new method performs better than individual networks, and better than committee machines using current ensemble-averaging methods.     

Electricity price forecasting using artificial neural networks

Electricity price forecasting in deregulated open power markets using neural networks is presented. Forecasting electricity price is a challenging task for on-line trading and e-commerce. Bidding competition is one of the main transaction approaches after deregulation. Forecasting the hourly market-clearing prices (MCP) in daily power markets is the most essential task and basis for any decision making in order to maximize the benefits. Artificial neural networks are found to be most suitable tool as they can map the complex interdependencies between electricity price, historical load and other factors. The neural network approach is used to predict the market behaviors based on the historical prices, quantities and other information to forecast the future prices and quantities. The basic idea is to use history and other estimated factors in the future to “fit” and “extrapolate” the prices and quantities. A neural network method to forecast the market-clearing prices (MCPs) for day-ahead energy markets is developed. The structure of the neural network is a three-layer back propagation (BP) network. The price forecasting results using the neural network model shows that the electricity price in the deregulated markets is dependent strongly on the trend in load demand and clearing price.     

Energy clearing price prediction and confidence interval estimation with cascaded neural networks

The energy market clearing prices (MCPs) in deregulated power markets are volatile. Good MCP prediction and its confidence interval estimation will help utilities and independent power producers submit effective bids with low risks. MCP prediction, however, is difficult since bidding strategies used by participants are complicated and various uncertainties interact in an intricate way. Furthermore, MCP predictors usually have a cascaded structure, as several key input factors need to be predicted first. Cascaded structures are widely used, however, they have not been adequately investigated. This paper analyzes the uncertainties involved in a cascaded neural-network (NN) structure for MCP prediction, and develops the prediction distribution under the Bayesian framework. A computationally efficient algorithm to evaluate the confidence intervals by using the memoryless Quasi-Newton method is also developed. Testing results on a classroom problem and on New England MCP prediction show that the method is computationally efficient and provides accurate prediction and confidence coverage. The scheme is generic, and can be applied to various networks, such as multilayer perceptrons and radial basis function networks.     

Day-ahead price forecasting in restructured power systems using artificial neural networks

Over the past 15 years most electricity supply companies around the world have been restructured from monopoly utilities to deregulated competitive electricity markets. Market participants in the restructured electricity markets find short-term electricity price forecasting (STPF) crucial in formulating their risk management strategies. They need to know future electricity prices as their profitability depends on them. This research project classifies and compares different techniques of electricity price forecasting in the literature and selects artificial neural networks (ANN) as a suitable method for price forecasting. To perform this task, market knowledge should be used to optimize the selection of input data for an electricity price forecasting tool. Then sensitivity analysis is used in this research to aid in the selection of the optimum inputs of the ANN and fuzzy c-mean (FCM) algorithm is used for daily load pattern clustering. Finally, ANN with a modified Levenberg–Marquardt (LM) learning algorithm are implemented for forecasting prices in Pennsylvania–New Jersey–Maryland (PJM) market. The forecasting results were compared with the previous works and showed that the results are reasonable and accurate.     

Nonstationary speech analysis using neural prediction

Discusses extracting dynamic features of individual speakers from short speech segments for robust recognition. As a method of homomorphic signal processing, cepstrum analysis can separate the excitation and impulse response of the vocal channels when applied to speech signals. For short-time spectrum analysis, overlapping windows are used to divided speech into many frames. For each window, one cepstrum vector is obtained. It is assumed that for each frame (about 30 msec), the speech signal is stationary. However, the speech is basically nonstationary for long time intervals. Therefore, one must consider the dynamic changes between frames. Conventional methods often use only the static features of the short-time cepstrum. A neural network can be seen as a nonlinear dynamic system, which may express both the static and dynamic features of the signal at hand. For this purpose, a neural prediction network was designed to extract the inter- and intraframe correlations of cepstrum vectors, so as to obtain the robust features of individual speakers from very short speech epochs     

基于随机森林遥相关因子选择的月径流预报北大核心CSCD

流域径流过程与大尺度气候因子之间存在遥相关关系,如何从众多的水文、气象、大气环流及洋流等因子中找出与径流密切关联的因子,是中长期径流预报的一个难题。将基于贝叶斯优化的随机森林模型应用于对水文、气象、气候因子构成的高维度因子集进行因子选择,根据变量重要性评分挑选对月径流影响较大的预报因子,构建广义回归神经网络、极限学习机、支持向量回归径流预报模型。将该方法应用到金沙江流域,相较于线性相关法,基于随机森林输入因子选择的方法提高了模型泛化性能;遥相关因子的引入既实现了流域月径流高精度预报,又从物理机制上提供了支撑。     

Myoelectric signal analysis using neural networks

It is shown that the capacity of a discrete Hopfield network for functional minimization allows it to extract the time-series parameters from a myoelectric signal (MES) at a faster rate than the previously used SLS algorithm. With a two-dimensional signal space consisting of one of the parameters and the signal power, a two-layer perceptron trained using back-propagation has been used to classify MES signals from different types of muscular contractions. The results suggest that neural networks may be suitable for MES analysis tasks and that further research in this direction is warranted.     

Chromosome classification using backpropagation neural networks

The feasibility of an artificial neural network as a chromosome classifier was examined in this study, using the relative length, the centromeric index, and the density distribution of G-banded chromosome as feature vectors. The two-layer neural network trained with the error backpropagation training algorithm showed good potential in classification of Giemsa-banded human chromosomes. The minimum classification error was obtained with the configuration that had 27 input nodes and 24 PEs in the hidden layer. However, this study also showed some problems. Only two experiments, which had 25 and 50 density distribution samples, respectively, were carried out, due to the long computation time of the backpropagation neural network. Also, the centromere finding algorithm used in this study could not apply to telocentric chromosomes (group D and group G) because of their very small short arms; their centromere locations were determined manually. The algorithm must be modified so that it can be applied to all types of chromosomes to reduce the preprocessing time. Better training algorithms to reduce training time are needed. The error backpropagation training algorithm requires very long training times. Next, finding the optimal number of input nodes that gives the minimum classification error requires a trial and error experiment. Finally, other chromosome features that reduce the classification error need to be examined     

A new method for system optimization by neural networks

This paper proposes a new method of optimizing the operation of a power transmission system, based on a structured neural network. Because the structured neural network is built by prewiring, the network construction time is very short. Simulations of transmission loss of reduction have been carried out. It is shown that the optimization performance of the proposed method is satisfactory in practice. © 1997 Scripta Technica, Inc. Electr Eng Jpn. 118 (4): 27–34, 1997     

Computing with neural networks

The resurgence of interest in neural networks is discussed. This interest is prompted by two facts. First, the nervous systems of simple animals can easily solve problems that are very difficult for conventional computers. Second, the ability to model biological nervous system functions using man-made machines increases understanding of that biological function. Two popular models of neural networks, the feedforward model and the feedback model, are described. Adaptation or learning, a major focus of neural network research, is detailed     

基于盲数和神经网络的电价预测模型

利用盲数理论结合经典电价预测方法,在有限的历史电价、负荷及其他相关数据的基础上对电价进行准确预测。首先提出了基于盲数和神经网络市场出清电价预测模型,利用BP神经网络对历史数据进行训练学习,在得到网络学习权重后,结合盲数理论,引用盲数代替实数进行价格预测。算例结果表明,模型消除不确定性因素对预测结果的影响,实际历史电价都很好地落在了预测结果最大置信度的置信区间中,较好地完成了预测任务,证实了设计的可行性和模型的可靠性。     

基于自组织映射神经网络的市场清算电价预测

市场清算电价预测是电力市场中交易决策的基础。人工神经网络是电价预测较为理想的方法,但依然存在一些问题,如样本训练有时需要很长时间,存在收敛问题,特别是当样本特征量不明显的时候,这种现象更为突出。针对这一问题,利用自组织映射的聚类特性将历史数据进行特征分类和筛选处理,处理后形成的新数据用于训练三层BP神经网络,仿真结果表明,经过这种数据处理后,网络的收敛速度得到了显著提高,且预测效果良好。     

Blind signal separation using cellular neural networks

In this paper a two-layer cellular neural network (CNN) is used to separate blind signals. the topological structures of the CNN and the inner parameters are presented. the first CNN layer functions as an adaptive filter which converges asymptotically to an equilibrium point in the mean. A stochastic stability model is used to find conditions under which cells in the first layer converge. Conditions leading to correct equilibrium solutions are also presented using this model. the second CNN layer functions as a signal separator. Simulations show that the CNN blind signal separator has strong robustness and works even better than the theory predicts.     

Transient stability assessment using artificial neural networks

This paper presents a study of the feasibility of using artificial neural networks (ANNs) in transient stability assessment for power systems. In the study ANNs have been developed to synthesize the complex mapping that carries the power system operating variables and fault locations into the Critical Clearing Times. The training of the ANNs was achieved through the method of backpropagation. The critical fault clearing time values were obtained by the Extended Equal Area Criterion method and used for training. In this work, an attempt was made to avoid the restrictions on load and topology variations. The parameters of the ANNs consist of the generation and loading levels. None of these inputs require any computation. This feature is desirable for on-line transient stability assessment purposes.Training of the ANNs was achieved using a combined production learning phase. The training patterns were not limited to a given collection of samples. This scheme eliminates the problem that an ANN may be influenced by the regions of attraction of a specific category.     

融合边缘检测和递归神经网络的视频表情识别

为有效解决传统视频人脸表情识别通常只关注单张视频帧的空间特征,而忽略了相邻帧之间隐藏的时间特征的问题,提出一种结合边缘检测和递归神经网络的视频表情识别方法,利用梯度边缘检测准确地提取输入图像的纹理信息,同时提出一种分片交叉LSTM结构,提取出图像序列中隐藏的时空特征。实验在CK+和MMI视频库上进行,在OCNN-RNN网络中分别取得88.4%和69.7%的识别率,在GCNN-RNN网络中分别取得89.8%和73.6%的识别率,最终使用提出的加权随机搜索方法融合GCNN-RNN和OCNN-RNN两个网络之后,分别取得了94.6%和79.9%的识别率,均优于单流网络算法,证明了所提算法的有效性。     

Composite learning adaptive backstepping control using neural networks with compact supports

The ability to learn is crucial for neural network (NN) control as it is able to enhance the overall stability and robustness of control systems. In this study, a composite learning control strategy is proposed for a class of strict‐feedback nonlinear systems with mismatched uncertainties, where raised‐cosine radial basis function NNs with compact supports are applied to approximate system uncertainties. Both online historical data and instantaneous data are utilized to update NN weights. Practical exponential stability of the closed‐loop system is established under a weak excitation condition termed interval excitation. The proposed approach ensures fast parameter convergence, implying an exact estimation of plant uncertainties, without the trajectory of NN inputs being recurrent and the time derivation of plant states. The raised‐cosine radial basis function NNs applied not only reduces computational cost but also facilitates the exact determination of a subregressor activated along any trajectory of NN inputs so that the interval excitation condition is verifiable. Numerical results have verified validity and superiority of the proposed approach.     

基于混沌因子及相空间重构后的神经网络短期电价预测的研究

影响电价因素众多,但在现实中不可能获得所有信息的资料,在这种信息不完全的情况下,为了更好地提高电价预测精度,通过分析电价和负荷时间序列的混沌性,用C-C方法分别重构其相空间,揭示出其本身蕴涵的规律,并采用数据挖掘技术中的相似搜索技术,挖掘出与预测日变化规律最相似的时间序列作为样本,利用BP神经网络这一具有高度自学习自适应能力的网络,拟合电价序列的重构函数.利用美国PJM电力市场的实际数据进行了实例预测,结果显示出良好的预测精度,并比传统BP网络能更好地预测休息日电价.     

Position sensorless driving of BLDCM using neural networks

A sensorless driving method of brushless DC motors (BLDCM) using neural networks has been studied in this paper. Considering the nonlinear characteristics and the parameter error of the modeling, neural networks are introduced to estimate the electromotive force (EMF). The results of simulation and experiment using offline trained neural networks show that the proposed method is useful. In addition, the robustness about the parameters is discussed. © 2007 Wiley Periodicals, Inc. Electr Eng Jpn, 162(4): 64–71, 2008; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20240     

基于人工神经网络输电线的速断保护

介绍了用人工神经网络实现输电线速断保护区内区外动作情况的判断和定位的功能,并通过对单相接地故障和三相故障的EMTP仿真测试发现,用神经网络可以在一个周期以内准确地显示和定位故障,证明了该方法的切实可行性。     

Intelligent control using multiple models and neural networks

Most adaptive control algorithms for nonlinear discrete time systems become invalid when the controlled systems have non‐minimum phase properties and large uncertainties. In this paper, an intelligent control method using multiple models and neural networks (NN) is developed to deal with those problems. The proposed control method includes a set of fixed controllers, a re‐initialized neural network (NN) adaptive controller and a free‐running NN adaptive controller. The bounded‐input‐bounded‐output (BIBO) stability and performance convergence of the system are guaranteed by the free‐running adaptive controller, while the multiple fixed controllers and the re‐initialized adaptive controller are used to improve the transient response. Simulation results are presented to demonstrate the effectiveness of the proposed method. Copyright © 2007 John Wiley & Sons, Ltd.