Yixu Song

Short-term stock price prediction based on echo state networks

Neural network has been popular in time series prediction in financial areas because of their advantages in handling nonlinear systems. This paper presents a study of using a novel recurrent neural network-echo state network (ESN) to predict the next closing price in stock markets. The Hurst exponent is applied to adaptively determine initial transient and choose sub-series with greatest predictability during training. The experiment results on nearly all stocks of S&P 500 demonstrate that ESN outperforms other conventional neural networks in most cases. Experiments also indicate that if we include principle component analysis (PCA) to filter noise in data pretreatment and choose appropriate parameters, we can effectively prevent coarse prediction performance. But in most cases PCA improves the prediction accuracy only a little.

Automatic stock decision support system based on box theory and SVM algorithm

The stock market is considered as a high complex and dynamic system with noisy, non-stationary and chaotic data series. So it is widely acknowledged that stock price series modeling and forecasting is a challenging work. A significant amount of work has been done in this field, and in them, soft computing techniques have showed good performance. Generally most of these works can be divided into two categories. One is to predict the future trend or price; another is to construct decision support system which can give certain buy/sell signals. In this paper, we propose a new intelligent trading system based on oscillation box prediction by combining stock box theory and support vector machine algorithm. The box theory believes a successful stock buying/selling generally occurs when the price effectively breaks out the original oscillation box into another new box. In the system, two SVM estimators are first utilized to make forecasts of the upper bound and lower bound of the price oscillation box. Then a trading strategy based on the two bound forecasts is constructed to make trading decisions. In the experiment, we test the system on different stock movement patterns, i.e. bull, bear and fluctuant market, and investigate the training of the system and the choice of the time span of the price box. The experiments on 442 S&P500 components show a promising performance is achieved and the system dramatically outperforms buy-and-hold strategy.

Intelligent stock trading system based on improved technical analysis and Echo State Network

Stock trading system to assist decision-making is an emerging research area and has great commercial potentials. Successful trading operations should occur near the reversal points of price trends. Traditional technical analysis, which usually appears as various trading rules, does aim to look for peaks and bottoms of trends and is widely used in stock market. Unfortunately, it is not convenient to directly apply technical analysis since it depends on persons experience to select appropriate rules for individual share. In this paper, we enhance conventional technical analysis with Genetic Algorithms by learning trading rules from history for individual stock and then combine different rules together with Echo State Network to provide trading suggestions. Numerous experiments on S&P 500 components demonstrate that whether in bull or bear market, our system significantly outperforms buy-and-hold strategy. Especially in bear market where S&P 500 index declines a lot, our system still profits.

An adaptive modeling approach based on ESN for robotic belt grinding

Robotic belt grinding system has good prospect to release hand-grinder from their dirty and noisy working environment. However, as a kind of non-rigid processing system, it is a challenge to model its processes precisely for free-form surface because its performance is unstable due to a variety of factors, such as belt wear and belt replacement. In order to adapt to the variability, an adaptive modeling approach based on echo state network (ESN) is presented, whose major idea is to exhaust information from new data by using sliding window technique to select training samples. With machine learning paradigm this approach is more flexible than traditional ones which often base on formula and experimental curves. Experimental results of grinding turbine blades demonstrate this approach is workable and effective.

A Novel Control Method for Robotic Belt Grinding Based on SVM and PSO Algorithm

In this paper, a novel method for robotic belt grinding based on support vector machine and particle swarm optimization algorithm is presented. Firstly, the dynamic model of the robotic belt grinding process is built using support vector machine method. This is the basis of our work because the dynamic model shows the relation between the removal and control parameters (contact force and robot’s speed) of robot. Secondly, the method of reverse solution of the dynamic model is introduced. According to this method, control parameters of robot can be accurately calculated by the given value of removal. Finally, the PSO algorithm is introduced to get smooth and stable trajectories of the control parameters, because the trajectory jitter of the control parameters has a great influence on the grinding accuracy. The experiment results show that the novel method for robotic belt grinding performs well in the control of the robot parameters and the grinding accuracy is improved.

The application of echo state network in stock data mining

Stock data, which is among the most complicated time series, is difficult to analyze and mine. Neural network has been a popular method for data mining in financial area since last decade. In this paper, we explore the use of Echo State Networks (ESNs) to perform time-series mining on stock markets. The Hurst exponent is applied to adaptively determine initial transient and choose sub-series with greatest predictability before training. With the capability of short-term memory provided by ESN, a stock prediction system is built to forecast the close price of the next trading day based on history prices and technical indicators. The experiment results on S&P 500 data set suggest that ESN outperforms other conventional neural networks in most cases and is a suitable and effective way for stock price mining.

Automatic Target Segmentation based on Texture for Microscopic Images of Natural Medical Herbal Powders

The authentication of natural medical herbs has crucial impact on the clinical curative effect. Much of these herbs have been ground into powders in the market, leading to the difficulty of identification. Microscopic images of these powders contain important evidences for identification. Currently identification based on these microscopic images are mostly conducted by manual observation. Identification aided by computer vision technique is an important research subject recently. These microscopic images usually contain variety of substance, and most of them are noises, thus the target segmentation is necessary for identification. An effective automatic target segmentation algorithm based on texture is proposed in this paper. Our method consists of two steps: “Preliminary Segmentation” and “Further Segmentation”. Firstly, gradient transform and image fusion are conducted for image preprocessing, then each pixel is encoded into a feature vector based on texture and clustered into two groups: background and foreground. Secondly, taking the continuity of edge and the locality of target into consideration, energy equations are established, and maximum flow-minimum cut algorithm is applied to solve them. Three groups of images are tested to evaluate our method, and the experimental results show that our method achieves a better segmentation compared with Grab-Cut, and additionally user inter-action is not required in our method.

Notice of Retraction An adaptive SVR modeling method based on VFS for robotic grinding

The performance of a model, which is trained with offline data, is highly relied on the conditions in which the system is working. When the working conditions change, the prediction accuracy of the model will be reduced significantly. To solve this problem, we propose an adaptive SVR modeling method based on vector-field-smoothed (VFS) algorithm. This method can adapt the model quickly to new working conditions by using only a few adaptive samples. Also, it can extend the feature subspace which the model covers so as to enhance the generalization ability of the model. The experimental results show that the model using this method can achieve a much better performance than the original model, as well as the model using other adaptive SVR modeling method.

Automatic target segmentation based on texture for microscopic images of Chinese herbal powders

The identification of Chinese herbal powders is usually based on physical or chemical detection, but that is far from enough to identity dozens of herbal species. Microscopic images of these powders contain variety of information, and important evidence for identification. These images usually contain variety of substance, and most of them are noises, which makes the target segmentation become a difficult job. An effective automatic target segmentation algorithm based on texture is proposed in this paper. Our method consists of two steps: “Preliminary Segmentation” and “Further Segmentation”. Firstly, feature vector of texture is extracted and clustered into two groups: background and foreground; secondly, taking the continuity of edge and the locality of target into consideration, energy equations are established, and Maximum flow-Minimum cut Algorithm is applied to solve them. Three groups of images are used to test our method: microscopic images of Chinese herbal powders, Brodaze Images, and natural texture images. And the experimental results show that our method achieves a better segmentation compared with Grab-Cut, and additionally user inter-action is not required in our method.

A Fast Events Relationship Extraction Method Based on Semi-CRFs

Event relationship extraction is a new research domain which has attracted more and more attentions. It is because that the relationships among different events can provide a lot of important information on special fields such as national defense, crime-solving or Anti-Terrorism. But there are innumerous event description web pages on the internet and the relationship among them is perplexing, so obviously it is impossible to extract them manually. In order to extract the relationship automatically, recognizing the key elements of the event is an indispensable preprocessing phase. At this step, semi-CRFs is a good Name Entity Recognition model, but it is too slow to fit the requirements of large scale processing. This paper has accelerated the semi-CRFs inference algorithm and presents a fast events elements extraction method. Based on it, the events relationship map is also extracted automatically. All event elements such as the agents, location, time or related organization are used as slots to connect different events so as to provide more plentiful information for the supervisor.