Showing papers in "Caai Transactions on Intelligent Systems in 2013"

Bat algorithm with the characteristics of Lévy flights

Abstract: The basic bat algorithm(BA) in the past research studies reveal deficiencies as apt to be premature and low precision of convergenceThis paper first analyzed the optimization mechanism and deficiency of bat algorithm(BA),and then considering the Levy flight behaviors of bats can simulate predatory more realistically,the study proposed substituting for the speed and location updating pattern of former algorithmThe proposed algorithm fully explored the trait of uneven random walks,so that clusters of short steps were connected by rare long steps,to avoid being trapped in local optimal solutionSimulation results for benchmark functions show that the proposed algorithm improved the global optimization ability remarkably and outperformed the basic BA and particle swarm optimization(PSO) in accuracy and convergence propertyTherefore,the proposed algorithm is an effective tool for solving the optimization of complex functions

A survey of the classification of data streams with concept drift

Abstract: Because the current machine learning algorithms all are essentially an optimization procedure that aims to ensure the generalization ability based on static learning environment, the classification data streams with concept drift has brought severe challenges to machine learning. In order to address these concerns, a survey was developed consisting of four aspects: the introduction to data streams and concept drift, the development process and future trends, the main research fields, and the new developments in the study field of the classification data streams with concept drift. The existing problems relating to classification data streams with concept drift were discussed at last.

Kinect-based object grasping by manipulator

Abstract: Kinect from Microsoft is often used in the robot system for target and real-time detection in unstructured environments. Background subtraction and frame difference algorithms based on the depth information collected by Kinect are used to obtain the information of the grasp point. A Rapidly-exploring Random Tree (RRT) algorithm based on the working space was used to carry out path plan of the end of manipulator, and gradient projection algorithm was used to do inverse kinematics trajectory optimization. The manipulator was able to accomplish object grasping tasks according to the joint angles. An experiment system was developed to verify the effectiveness of the proposed methods.

A kind of integrated navigation and control system design for substation robot based on the Kalman filtering and fuzzy algorithm

Abstract: This paper aims at examining the substation's industrial environment which has a number of uncertainty factors and strong external disturbance.This paper proposes a novel design method for the integrated navigation and control system which is used in a substation intelligent robot based on the Kalman filtering and fuzzy control theory,for the purpose of fulfilling an autonomous navigation,positioning and behavior control,and finishing the unattended inspection task.In particular,a navigation subsystem based on high-precision laser radar has the functions of positioning,and that of electronic compass(for navigation angle measurement),which effectively overcomes the failure of compass in substation with strong electromagnetic interference.The field test shows that the robot's actual track is close to the assigned track,achieving the independent inspection mission in a strong electromagnetic interference environment.In addition,the design also has the advantages of lower cost,higher precision(sub-meter),and more easy implementation for industrial use.

The formation control of multi-robot fish based on leader-follower and fuzzy feedback mechanism

Abstract: In the formation control of multi-robot fish under the leader-follower framework,the fuzzy feedback controller was introduced to traditional leader-follower formation control algorithm in efforts to cope with the relative independence of the leader and followers,as well as the desertion of the laggard robot fishes.The goal was to ensure that the pilot robotic fish could check the position of followers regularly.In order to enhance the stability of the formation control of multi-robot fish and complete the task of collaborative operation,the formation control problem of multi-robot fish system was transformed into the problem of followers' tracking the position and orientation of the leader with the addition of the feedback regulation.Simulations verify that the proposed method can better realize the formation control of the robotic fish system.

UCAV situation assessment based on fuzzy rules and dynamic ant colony-Bayesian network

Abstract: In order to solve the challenging problem of unmanned combat aerial vehicles(UCAV) situation assessment in complex environments, based on the introduction of ant colony optimization, Bayesian network and mathematical model, a hybrid fuzzy rules and dynamic ant colony-Bayesian network was proposed in efforts to examine the situation assessment of UCAVs The incomplete data was converted into a complete data packet by using a dynamic ant colony-Bayesian network, which can greatly simplify the complexity of learning, and ensure that the algorithm evolves into good structure The dynamic ant colony-Bayesian network algorithm was improved by using fuzzy logic The expert's experience was adopted in the form of fuzzy language and rules The single value assessment results were combined with the probability vector to evaluate the capacity level of UCAVs at different times, increase the intelligence of situation assessment, and practicality of engineering application A series of experiments verified the feasibility and effectiveness of the proposed hybrid method for situation assessment of UCAVs in the complicated combat environment

Application of mutation particle swarm optimization based BP neural network in the intrusion detection system

Abstract: A aiming at the properties of real-time performance and self-learning of the intrusion detection system( IDS),an improved particle swarm optimization( PSO) based on the mutation operator was proposed,which was used to optimize BP neural network,so as to accelerate convergence speed of BP neural network,thus,the MPSO_BP hybrid optimization algorithm is presented. In order to increase detection rate and lower false alarm rate of the intrusion detection system,a new intrusion detection model( MPBIDS) was put forward. Iris data set was applied to the three BP neural networks for simulation. Experiment results show that the optimized BP neural network had better convergence speed and accuracy. Based on this finding,the improved BP network was applied to intrusion detection,taking KDDCUP99 as the test data set. The simulation result proves that the IDS with improved BP network can improve the detection rate and reduce the false alarm rate.

A modified teaching-learning-based optimization algorithm and application in neural networks

Abstract: In order to improve the output accuracy of back propagation neural network,a modified teaching-learning-based optimization( MTLBO) algorithm is proposed to train the weight and threshold value of neural network.In the MTLBO method,the "Teaching"phase and "Learning"phase were modified on the basis of TLBO algorithm,and a new "Self-Learning"mechanism was proposed to intensify global searching ability. Finally,the function fitting experiment and the tractor gearbox diagnosis experiment were used to test the performance of the proposed algorithm. Simulations show that this algorithm has a better convergence,prediction accuracy and robustness compared to the genetic algorithm( GA) and the basic teaching-learning-based optimization( TLBO) algorithm.

Analysis of the present research status and key technology of the humanoid emotion-interactive countenance robot

Abstract: An increasing amount of research is being focused on the contemporary field of robotic intelligence. As a result,the humanoid emotion-interactive countenance robot is attracting wider attention. This paper summarizes some of the research achievements regarding the countenance robot in Japan,the USA,the European Union and China. Additionally this paper analyzes the theoretical framework and key technology of the research for the humanoid emotion-interactive countenance robot,and finally the authors discuss the future development trends and put forward some opinions on the future direction of the research.

A novel SIR model with infection delay and nonuniform transmission in complex networks

Abstract: In order to analyze and understand the spreading behavior of infectious diseases, the authors propose to examine susceptible-infected-removed (SIR) model. The researchers simultaneously introduce into the epidemic model the two factors: influencing disease spreading behavior, and infection delay and nonuniform transmission, utilizing the SIR model. Based on the mean-field approximation and large-scale numerical simulations, the analytical results of critical thresholds of disease spreading were derived, along with the infection delay and the nonuniform transmission having a distinct impact on the critical threshold. The infection delay can greatly decrease the critical threshold and facilitate the spread of epidemics, while the nonuniform transmission can augment the critical threshold and hinder the epidemic spreading in complex networks. Current results are conducive to further understand the epidemic spreading inside the complex real systems, as well as to fully consider the roles of infection delay, transmission factors and topological structure of population in the spreading of diseases. The results also provide a number of theoretical evidence to design more effective epidemic prevention and containment measures.

Conflict resolution of multiple airplanes in free flight based on the genetic algorithm

Abstract: In order to resolve the conflict among airplanes in free flight,the study proposed to examine a genetic algorithm to quickly solve the best routeThe genetic algorithm was considered to be a simplification,generalization and strong robustnessBy applying the genetic algorithm,the conflict relief among multi-planes can be resolved respectively by altering the heading,speed,and the effective flight mechanism when multiple airplanes are flying relatively at the same timeThe simulation results show that the algorithm can achieve the optimal conflict relief route quickly by utilizing both methods,and if there is a conflict at a point among multi-planes,using the method of changing the heading is more applicable

Immersion and invariance theory and its application in nonlinear adaptive control

Abstract: The actual engineering demands and the complexities of the nonlinear system theory have led it to become the most attractive and challenging research field in control subjects,therefore,a new nonlinear control law design method—immersion and invariance( II) theory has been introduced With this method,a( partially) asymptotically stable system with a dimension less than that of the controlled system is firstly selected as the target system,then the immersion mapping and control law are designed to make the controlled dynamics of the original system is an immersion image of the target system In addition,the control law can keep the image of the target system as an invariant and attractive manifold,and render the trajectory of the closed-loop system bounded Focusing on the unknown point mass model,a new nonlinear immersion and invariance adaptive control law is designed for realizing the precise tracking of a reference command Compared with the adaptive control law based on the certainty-equivalent principle,the simulation results show that the proposed II adaptive control law is quite effective for processing a system with unknown parameters

Qualitative modeling and simulation of complex system based on the combination of GM( 1,N) and QSIM

Abstract: Aiming at the characteristics of complex system simulations,such as a lack of information,and the possibility of QSIM modeling method employing differential equations,a qualitative modeling method combining QSIM and GM( 1,N) is proposed. First the realtive researches are reviewed. The variable space expression of uncertain information based on the cloud model is provided. Then the process and principles of combination of QSIM and GM( 1,N) qualitative simulation are given. Lastly,the modeling method was applied to the case of modeling in the modeling and simulation of the complex system to verify the feasibility of the method. The results show that this method has full use of fewer system information. Both quantitative and qualitative information can effectively be integrated into modeling and simulation of complex systems.

Similar code detection algorithm based on sequence clustering

Abstract: In order to improve efficiency of similar detecting between the codes of source programs,similar code detection algorithm based on sequence clustering was proposed in this paper.First,the algorithm extracts the source code by partitioning it into multi-segments according to its structure.Secondly,parts of the codes in each segment were transformed and the sequences were then clustered taking the weighted edit distance as the similar measure standard.Next,similar code fragments were obtained,achieving the objective of detecting similar functions among multi-codes of the source program.The experimental results based on a series of real and synthetic programs reveal the validity and scalability of the algorithm.

Infant safety management system based on integrated RFID and ZigBee network

Abstract: In order to improve the informatization and automatization level of infant safety management in a kindergarten and reduce the occurrence of infant safety accidents,this paper describes the application of the method of integrating radio frequency identification and wireless communication. A set of wireless radio frequency identification system integrating the ZigBee network was designed and utilized,and it was discovered that the system may be applied to the infant safety management in the kindergarten. In this paper,the authors elaborate on the structure of the system,introduce the design of specific hardware in the read-write device as the main control module,RF transceiver module,display module and electronic label,and give the design structure of the system software and the flow process diagram on the design of the main program. It is shown by actual deployment and validation that,the ZigBee network of the system is stable and reliable,and the data transceiver is accurate. It may adapt to complex building environments and achieve long distance recognition without orientation configuration. The RFID system may greatly improve the efficiency of infant safety management in the kindergarten.

Path planning for autonomous underwater vehicles by using QPSO

Abstract: In efforts to address the submersible path planning problem in complex undersea environment,a path planning method for autonomous underwater vehicles was put forward,which is based on quantum-behaved particle swarm optimization(PSO).First,the bathymetric data was extracted from a nautical chart,and the intensive-specification depth data was acquired by dealing with the natural neighbor interpolation and the random midpoint displacement interpolation.Next,we were able to establish the undersea 3D model and determine a path security testing program,along with the principle to prevent collisions.The influence of the ocean current size,and direction on autonomous underwater vehicle navigation and the influence of the turning angle of the path points on navigation were transformed into corresponding path lengths.At last,the total length was used as the fitness function and the optimal path was obtained by iteration of quantum-behaved particle swarm optimization(QPSO).A safe and simple path was achieved as the result of the simulation,verifying effectiveness and feasibility of the method.

Support vector machine imbalanced data classification based on weighted clustering centroid

Abstract: Classification of imbalanced data has become a research hot topic in machine learning.Traditional classification algorithms assume that different classes have balanced distribution or equal misclassification cost,thus,making it hard to get ideal result of classifications.A support vector machine(SVM) classification method based on weighted clustering centroid was proposed in this paper.First,unsupervised clustering was applied to the positive and negative samples respectively to extract the clustering centroid of each clustering,which was represented the most in compactness of the clustering sample.Next,all clustering centroids formed a new set of balance training.In order to minimize the information loss during clustering,each clustering centroid was associated with a weight factor that was defined proportional to the number of samples of the class.Finally,all clustering centroids and weight factors participated in the training of the improved SVM model.Experimental results show that the proposed method can make the sample selected from model train sets more typical and improve the classification performance better than other sampling techniques for dealing with imbalanced data.

Botnet detection algorithm based on neural network

Abstract: The most current botnet detection algorithm are typically based on network traffic analyzing technologies that usually need packet payload. The botnet detection algorithm also relies on information obtained by external systems or malicious behaviors of bots that do not automatically store the features of botnet traffic and do not have the ability of associative memory. As a result, this paper proposes a botnet detection algorithm based on BP neural network which trains the BP neural network classifier through a lot of botnet and normal traffic samples and allows it to learn how to identify botnet traffic and automatically remember the features of botnet traffic and therefore, detect the infected hosts effectively. The neural network classifier takes the host-pairs as analysis objects, extracts the traffic features of communications between two hosts and takes the feature vectors of host-pairs as input, thus, effectively distinguishing the normal hosts and bots. The experimental results show that the detection rate of our algorithm can achieve to 99% and the false positive rate is below 1% and the algorithm has a good performance.

Research on the expert system reasoning model based on Agent

Abstract: The traditional expert system reasoning model structure has poor adaptability in acquiring knowledge. From the viewpoint of system science, the complex adaptive system theory is used to improve the structure and operation mechanism of a traditional expert system. Firstly, an Agent was introduced to simulate neurons in the human brain and load the knowledge interacting in the expert system reasoning model. Then an expert system reasoning model of complex adaptation was constructed based on the Multi-Agent interaction. Consequently, the knowledge acquiring mechanism, knowledge base and reasoning engine were unified into the Agents interaction in the complex adaptive expert system. Finally, by designing the expert system reasoning model prototype in decision-making of international sporting events bidding, the effectiveness of the expert system reasoning model based on Agent was verified. The results of the prototype running show that the expert system reasoning structure based on Multi-Agent model can effectively improve the adaptability of expert system knowledge acquisition. That provides a new idea for studying the expert system closer to human intelligence.

Automatic recognition of multi-interference source signals in the common channel based on independent component analysis

Abstract: A new method was presented to solve the problem of automatic recognition of multi-interfering signals in the common channel.Based on the independent component analysis(ICA),the multi-interfering blind signal separation technology was adopted to separate the interfering signals,which are mixed at the same time.Also the algorithm selected the features of each signal and interference in time domain,frequency domain and high-order cumulant domain to complete the recognition of the interfering signals.The computer simulation utilizes the automatic recognition method to solve the problem of two types of signals and four interferences mixed in the same channel,and the results show that after five times of iteration the algorithm achieves convergence with a good performance index of 0.21.The results also indicate that when SNR is more than 10 dB,the accurate interference separation rate is above 95%.And when SNR is less than 10 dB,the recognition rate drops greatly while the separation rate is hardly influenced,which proves the correctness and validity of this method.

Research on selection of paradigm words in the judgment of emotional tendency

Abstract: In light of the weakness of standard words selection,which exists in previous research studies,a method for selecting standard words in the judgment of emotional tendency was proposed in this paper By considering three key aspects,which include degree of emotion,tendency of emotion and ambiguity of emotion,the most representative typical sentiment words were chosen to act as standard words based on the definition of standard words Firstly,initial standard words were screened from emotion words issued in HowNet,the degree of emotion of these candidate standard words was computed,then the emotional tendency degree of the positive emotion words and negative emotion words,which have high ranking of emotion degree,was calculated,respectively Finally,the larger emotional tendency words were used as final standard words The results of test shows the methods used in this paper can gain a high accuracy for judging the tendency of emotion

Application of adaptive probabilistic neural network in Chinese liquor E-Nose

Abstract: In order to explore the possibility of hard liquor quality recognition by an electronic nose, the Chinese liquor of Yanghe Haizhilan, Jinshiyuan Shengjiedai, Anhui Yingjiadaqu, and Niulanshan Chenniang were analyzed by using self-made new wireless electronic nose for recognition of hard liquor quality. Firstly, the steady-state response and slope values were extracted after smoothing the collected data. Secondly, principal component analysis PCA was used to reduce the dimension of the eigenvector, and the obtained first two principal components scores were then used as the input parameters of the probabilistic neural network recognition model. Next, the aim was to overcome defect of traditional probabilistic neural network smoothing factor which would cause classification error easily. The method of adaptive probabilistic neural network identification model was presented, utilizing differential evolution algorithm to optimize the set of parameters. The results show that differential evolution-probabilistic neural network obtained a high recognition accuracy and noise immunity compared to back propagation, particle swarm optimization-probabilistic neural network and support vector machine. The experiment also proved that the electronic nose can effectively detect different liquor brands in China.

An emotion-based method of topic detection from Chinese microblogs

Abstract: Previous research studies have laid the foundation in the area of traditional topic detection and shown there are some effective ways to detect topics.However,the traditional algorithms do not work well in special situations for Chinese microblogs.In order to raise the recall ratio,the focus of this paper proposes to examine methods for detecting topics.The key to topic detection method,examines how to handle the structure of microblog with emotional content weighting,which is based on the argument that the negative words tend to carry more information.The existing topic detection methods for short messages merge emotional incination into the topic detection by first raising the weight of short messages containing negative emotion in the topic detection,then clustering the topics by a clustering method based on self-inquiry.The experiment on a real microblog dataset show that the approach provided in this paper can cluster topics and detect topics effectively,and also increase the recall ratio.

Gait recognition based on dual-view and multiple feature information fusion

Abstract: In view of low recognition rate of single-view and complexity of multi-view algorithm,a research was conducted examining the gait recognition under dual-viewCurrent research on the contour characteristic of the human body in frontal view and the dynamic characteristics of human walking in side view was examined using the complementary features of the gait information under multi-viewAlso the gait sequences were obtained utilizing the two views respectively,and then preprocessed to obtain simply connected body silhouettesNext,the Procrustes mean shape was extracted from the front view,and the active energy images(AEI) was calculated by side viewHowever,each of the AEI was projected to a low-dimensional feature subspace via two-dimensional local preserving projections(2D-LPP)The final recognition result was obtained by fusing recognition results of two perspectivesThe experiments in CASIA dataset(Dataset B) obtained a high recognition rate and achieved the expected effect of recognition