(2000-12-19)
High lights: Background, Current Work, Future Aim
Diagram and short version (in Chinese)
2 Self-organize System and Information.
2.1 Information Theory.
2.2 Ecology in Computational Intelligence.Synergetic Neural Network.
Hierarchy Neural Network.Agent-based Distributed Control System..
Behavior-Evolution-based Genetic Algorithm..My research work in computer science can be divided into two categories: image processing and self-organize system. Image processing and pattern recognition are the foundation for my research in self-organize system, which keys on computational intelligence. Also, many of algorithm proposed by me is experimented with or applied into the field of image processing. My master thesis, ¡°Research on Synergetic Associative Memory and its Application in Content Based Image Database Retrieval¡±, is a good example for this relationship.
I think my most important work is Hierarchy Neural Network and Behavior-Evolution-based Genetic Algorithm (BEGA). They are embodiments of my basic academic ideology. I am also pride of being one of the founders of DACS.
1 Image Processing
1.1 Image Database
PACS & Medical Image Processing
I was the designer and main developer of the project ¡°medical X-ray Digital Image Processing System¡± in 1999. I experienced the whole process of this project, from it¡¯s investigating and survey to its end. I fulfilled the task of client demand analysis, task planning, software design and coding and most of document work. I proposed ¡°Component PACS System¡± based on this software. My innovation is accepted by both medical and engineering experts.
There are some selected paper abstracts of mine in this field.
Bao Jie, Gao Jun, Zhang Xudong. Digital image self-adaptive acquisition in medical x-ray imaging. International Conference on Intelligent Information Processing, IFIP World Computer Conference 2000 (WCC200), Aug 2000,Beijing, China.
Abstract In this paper, we analysis the construction of video digital acquisition system in medical x-ray imaging and the main issues that have influence on the quality of acquired signal, and present a optimized self-adaptive digital image acquisition method against background nonuniformity and improperly setting of working point. The characteristic of medical image is used to extract valid region of image by integrating region growing, edge detecting and Hough transform, and to remove background by digital subtraction. After those preprocessing, self-adaptive parameters setting is executed based on dynamic analysis of histogram.
Wang Yixian, Gao Jun, Bao Jie. Valid region recognition in digital image self-adaptive acquisition of medical x-ray imagin
g. SPIE International Symposium on Optics and Opteoletronic Inspection and Control: Techniques, Applications and Inspection ,Nov,2000. Beijing, China.
Abstract: Setting of acquisition parameters during medical x-ray image acquisition has great influence on image quality. We have proposed a self-adaptive acquisition method based on image dynamic range analysis, which can adjust acquisition parameters automatically by software for best performance. One of the main problems in this method is that X-ray image contains many invalid regions that must be removed for analysis. Taking advantage of the characteristic of medical X-ray images (valid region close to a circle), We extract valid region of image by integrating region growing, edge detecting and general Hough transform. The result of edge-detection is irregular, with many surrounding glitches that will be detected and removed by curvature-based algorithm. The value of radius is estimated by the equation, and based on this value we can set a radius range to reduce Hough space volume. The parameters of the circle can be obtained by taking average of the range values.
Bao Jie, Gao Jun etal. Medical X-ray Digital Image Processing System Based on NSP. The fifth national conference on computer application, 1999,11(in Chinese)
Abstract In this paper, we discussed the realization of a medical x-ray digital image processing system based on NSP (Native Signal Process). In comparison with the traditional x-ray imaging technique, this system owns the features of high SNR (Signal Noise Ratio) and real-time video signal processing and data storing. It also provides the functions such as computer-aided diagnose, image sorting and management.
Bao Jie, Gao Jun etal. Component PACS System Design Orienting to Internet. Computer Engineering, 2000, 26(7): 9-10. (in Chinese)
Abstract This paper summarizes the development of PACS in China in recent years, analyzes the developing direction of PACS, proposes that the development of PACS should follow the principles of modularized design? department optimizing, and implement by stages. We present the concept of "Component PACS System" based on COM technique and DNA structure for large PACS. This System emphasizes the cooperation of departments, orients to Internet, resolves such problems as system inclusion, department optimization, telecommunication and flexible implement very well.
Jie BAO, Jun GAO. PACS Development in China and the Component PACS System. SPIE Medical Imaging 2001. San Diego, CA, USA, Feb 18-22,2001. (Accepted)
Abstract: In this paper, we summarize the development of PACS in China, present an analysis and classification of PACS developer in China. Some feature in the construction of PACS in China is discussed, such as the competition of international PACS developer, investment shortage and demand of high effectiveness, special management mode and traditional workflow and custom. Some important problems in the design of PACS are also discussed. 1. Should PACS include 3D image processing, DICOM interface and analog video acquisition? ; 2. Interface to electronic patient record and HIS Department optimization, 3. Optimize software according to the character of each imaging department under the assurance of interconnectivity between subsystems and data sharing; 4. Implement client-side low costly by Internet; 5. Develop system cooperatively, implemented by stages, customized by user. Component PACS structure is proposed to solve to these problems.
DACS
Inspire by the success of PACS and cooperated with the department of mechanic engineering and civil engineering, I propose the "drawing archiving and communication system"(DACS) to manage and process drawing database. It can not only archive large number of drawings and documents produced in engineering design, but also support remote or local transmission of graph and text. It changes the core of engineering design from single drawing or CAD file to the network database, and provide a better platform to the intelligent drawing processing. Then I organized a group to design and practice the DACS in automobile covering panel mold and bridge design.
There are some selected paper abstracts of mine in this field.
Abstract In this paper we propose the "drawing archiving and communication system"(DACS). DACS can archive, process and transmit drawings of different formats and their associated documents with uniform format. DACS also changes the focus of drawing management from single drawing or CAD file to network-based database. We also discuss following key issues in DACS and propose a practical system structure: design of database construction, the uniform encapsulation of raster and vector drawing, the linkage between drawings and documents, the quality of drawing input.
Gao Juan, Bao Jie, Huang Ling. Some Key Issues In Drawing Archiving And Communication System. The Third international conference on computer-aided industrial design, Dec 2000 ,Hang Kong, China
Abstract In this paper, we proposed the ¡°drawing archiving and communication system¡±(DACS) and its some key issues.1.Position of DACS .We should reserve information interface between DACS and the whole system of CIMS or CSCW, and let design departments are not ¡°isolated islands¡± any more. 2. Construction Policy of DACS. DACS is not simple CAD system but the requirement of collaborative design. DACS is also a system need for vast investment. A practical developing policy is component system and cooperative developing. 3. Data sources and processing methods. Raster drawings and vector drawings can be uniformly encapsulated in storage, but are different in processing methods. 4.Storage and transfer. We should choose to build a centralized archiving system according to work demand. It should be took mass storage and burst high bandwidth transmission demands into account when choosing transmission the rate of network and the storage devices.
1.2 Pattern Recognition
Pattern recognition is closely related to image process and artificial intelligence. It¡¯s the bridge between my research in image processing and self-organize system (specially, neural network)
Shape Recognition
My main work keys on Hough Transform and invariance feature recognition. Work in this part is combined with study on neural network and medical image processing; please refer to Neural Network section (Page 6) for detail.
Content-based Retrieval
With the base of image database and image recognition, I participate in several projects of content-based retrieval such as international cooperation project ¡°Associative memory and Content-based Image and Video Retrieval¡±(supported by NSF (National Science Foundation) of China and NSERC (National Science and Engineering Research Council) of Canada) and ¡°multi-resolution image analysis, understand and visualization¡± supported by HFUT. My main work in this field is: proposing that retrieve video by its caption feature and the processing flow of projection algorithm; using synergetic neural network as associative memory to retrieve image database; participate in a lab project to develop a video processing software.
There are some selected paper abstracts of mine in this field.
Yang Youqing, Gao Jun, Bao Jie. Caption Retrieval and Extraction Based on Video. Computer Application, 2000 ,10.(in Chinese)
Abstract Content-Based video and image database retrieval is one of the research hotspots in computer vision, image database and knowledge discovery, and is also the key technology in the next-generation multimedia database. There are captions in many videos such as news program and VCD, and those captions contain lots of image information. If we can extract them from video, it will be very useful to quick search for required video section. In this paper, we propose a method to retrieve and extract captions from video according to the particular characters of caption. The experimental result is satisfactory. Moreover, proposed method is also useful to retrieve and extract captions of other languages such as Japanese, Korean.
Bao Jie, Gao Jun, Yang Youqing etal. Caption-based Chinese Character Extraction in Streaming Video. Journal of LIIP-HUT(Technology Memo), Dec 1,1999.
Abstract According to Chinese Characters in streaming video, in this paper we extract video Chinese characters based on video captions, eliminate redundant information with feedback step by step by the feature of Chinese character, detect and separate Chinese captions from streaming video. First, according to changes of captions we extract key frames by method of spatial and temporal streams, eliminate temporal redundancy; Second, according to Chinese Characters, we eliminate spatial redundancy by method of mask FFT and so on; Third, we feed original key frames back to eliminate color redundancy; Finally, we analyze spatial locations according to Chinese character and spatial structure. Consequently, we achieve video Chinese characters extraction based on Chinese captions.
Visualization
In addition, I have some work in Visualization. One of my paper, Projection method with Coordinate rotate and interpolation of MRI Scanner had been awarded in Mathematical Contest in Modeling 1998(USA). My undergraduate thesis is Data Visualization and Processing. In graduate stage, I have done following work: analyzing and using Visualization Toolkit (VTK); designing the system structure of Paradise, a visualization platform; proposing a complex surface reconstruction algorithm in sparse data field.
2 Self-organize System and Information
An artificial intelligence system should be a self-organize system. The main points of design such systems are: increasing orderliness; strong information interchange; hierarchy structure and development; progressive centralization; progressive mechanization, etc. They are described in some of my Tech Memos and applied in the fields of neural network, agent system and genetic algorithm.
Information theory is part of my research because it is important for understanding the concept ¡° order¡± in self-organize system. It is also needed to explain communication, hierarchy and other phenomena in intelligent system.
2.1 Information Theory
There are some selected paper abstracts of mine in this field.
BAO Jie, GAO Jun. Integrate Degree Of System Hierarchy. The 11 th Annual Symposium of System Engineering Society of China, Yichang, China, Nov 2000. (In Chinese)
Abstract In this paper, we analysis the validity of ¡°integrate degree decreasing principle¡± in information systems and non-information systems, and propose that this principle is invalid in information systems because of different way of hierarchy formation. There is a ¡°integrate degree increasing principle¡± in information systems. This paper also discusses about the practical measurement of integrate degree in information systems, and the relationship between negentropy and integrate degree.
BAO Jie, GAO Jun. Base Information and Ergodic Hypothesis in Communication. Journal of Communication. (Under consideration) (In Chinese)
Abstract Shannon information is the change of the map of information destination states to information source states in communication. It¡¯s depended on the base information owned by information destination before communication. The base information is determined by the Ergodic hypothesis in communication, which is equivalent to supposing ¡°the unknown part of information source is in maximal entropy state¡± in isolated system, or more precisely, microcanonical ensemble distribution. But this hypothesis of equiprobable distribution will not be true in non-isolated system and be replaced by canonical ensemble distribution or grand canonical ensemble distribution?
Bao Jie, Gao Jun, Pan Mengxian. Uncertainty Relation in Measurement of Continuous Information Source. Tech Memo of LIIP-HUT 2000-4-9(in Chinese)
Abstract It¡¯s impossible to measure information of multi-dimension continuous information source with infinite precision because of the uncertainty relation. We propose a new proof of this relation in this paper, and reveal that the information source space should be discretized during information measure according to the uncertainty relation. There is minimum unit in information measurement, that is, information quantum.
2.2 Ecology in Computational Intelligence
The future of artificial intelligence is computational intelligence which including neural network, multi-agent system (MAS), evolution computation and artificial immune system. The basic idea of computational intelligence is ¡°social computation¡±, that¡¯s, complex intelligence can be obtained self-organizedly by simple intelligence individuals under some simple social rules (including competition, cooperation and so on). Such a ¡°social computation¡± system can be regarded as an artificial ecology system, which has similar property and development to nature ecology system. So basic laws of computational intelligence can be regarded as ¡°general ecology¡±. Some general laws in computational intelligence, such as order-increasing; information interchange; hierarchy structure and development; progressive centralization; progressive mechanization, are in fact general properties of a kind of self-organize systems. Therefore, the development of computational intelligence, especially the MAS, is closely related to the development of life sciences and social sciences. Neural Network (top-down) and MAS (bottom-up) are integrated methods to carry out the research in practice.
A notes about this field.
Jie BAO, Jun GAO. Sociobiological Principles in Computational Intelligence. Tech Memo of LIIP-HUT 2000-5-11
Some basic principles or phenomena in sociobiology such as Kin Selection or ¡° selfish gene¡± hypothesis which can lead to system ordering (e.g. ¡°Evolutionarily Stable Strategy¡±), and methodologies of sociobiology such as game theory, can be found in one kind of artificial societies: the Computational intelligence system. As a branch of artificial intelligence, Computational intelligence includes neural network, genetic algorithms, evolution computing, artificial immune system, and multi-agent system. Its philosophy is ¡°social computing¡±, i.e., complex intelligence can ¡°emergence¡± from competition and cooperation between Computational individuals which have only simple intelligence, with simple social regulations. From the viewpoint of dynamics, and supported strongly by recent research, Computational intelligence system and bio-society system are similar. For example, sub-population evolution and symbiosis group selection recent proposed in evolution computing is the embodiment of kin selection; we can also let genetic individuals of different ranks have different search strategies, instead of using classical algorithm that all individuals have the same strategy, to build a food chain in artificial evolution computing system and got desired solution eventually from the evolutionarily stable state; Darwinism methods and game theory¨Cbased negotiation are widely used in multi-agent ecology system. Those all are not accidentally but determined by the basic common property of computational intelligence system and bio-society system of being order-increasing system, or ¡°information system¡±.
Neural Network
Synergetic Neural Network
I have participated in projects of ¡°Research of Associative Memory Based on Mean Field Theory¡± supported by NSF, ¡°Research of Parallel Optical-electronic Shape Recognition System¡± supported by National Science Foundation of Anhui province?¡°Research of Shape Recognition System based on Associative Memory¡± supported by Ministry of Education(MOE), ¡°Research of some key issues on synergetic associative memory¡± supported by MOE. I also applied and supervised a project on synergetic associative memory supported by HFUT. (Only 8 such projects this year)
My main work in this area is: the relationship between fault-tolerance and memory capability of associative memory; research of invariance based on Hough transform; use synergetic associative memory as the recognition layer of Shape Recognition System; optical-electronic implement of synergetic neural network; Experiment plan of Haken network based on Matlab; Experiment of Quick Haken Network with disequilibrium attention parameters; learning in design synergetic neural network with disequilibrium attention parameters; introduce and improve ESDN and diffusive Haken network; analysis of the memory capability of Haken network; compare Haken network with SOM; research on stereo matching algorithm based on synergetic neural network.
There are some selected paper abstracts of mine in this field.
Jun GAO, Jie BAO et al. Optical-electronic Shape Recognition System Based on Synergetic Associative Memory. SPIE Photonics West 2001 ,Electronic Imaging 2001 Applications of Artificial Neural Networks in Image Processing VI. Jan 2001. San Jose, USA.
Abstract: A shape recognition system should be a fault-tolerant and real-time system. Synergetic associative memory has excellent dynamic performance of no spurious states, low complexity and high recognition speed, furthermore, it can be implemented by optical approach and meets the real-time requirement. Haken mode is dek/dt=rkek-B?k¡¯?k(ek¡¯2ek)-C?(ek¡¯2) ek. There ek is order parameters, rk is attention parameter. A problem is realizing this model needs many iteration steps, which is against real-time requirement. It can be proved that, for balanced attention parameters, pattern corresponding to maximal initial order parameter will govern the whole system eventually; more generally, for unbalanced attention parameters, if rj-ri>(lnei(0)/ lnej(0))/to,then the order parameter corresponding to pattern i will vanish. So, we can determine the winner in the first competition without iteration. This model is named ¡°Quick Haken Model¡±. This network can be implemented by the scheme as: input vector ->(prototype vector matrix)-> order parameters, (with attention parameter vector)->[adjudge]-> (adjoint vector matrix) -> recognized pattern.
Gao Jun, Bao Jie. Research on Invariance in Optical-electronic Shape Recognition System based on synergetic theory. Tech Memo of LIIP-HUT, Nov 6, 2000.
Abstract: To a shape recognition system, the most important demand is that it possesses invariance to recognition object and the real time character. Taking into account both invariance and real time character, we create a system that is called Optical-electronic Shape Recognition System based on synergetic theory. We use parallel Optical-electronic Hough transform in pre-process layer and character pick-up layer, we adopt optic synergetic neural network to invariability recognition in recognition layer. Hough transform and character pick-up method can effetely pick up invariability character of unrecognized shape; Compared with electronic system and traditional neural network, parallel Optical-electronic implement and synergetic pattern recognition improves system¡¯s real time character in character pick-up and recognition phases.
Hierarchy Neural Network
My plan for hierarchy neural network began in 1998 and it is important in my whole academic system. But it is not until recently have I the time to fulfill it. My research on synergetic neural network provides me a good experiment platform.
Bao Jie, Gao Jun etal. Neural Network based on Hierarchy Architecture. Information Processing, as a part of SPIE's 46th Annual Symposium. San Jose, CA, 2001. (Accepted, but was given up because of reason)
Abstract: Bio-neural system has obviously hierarchical Architecture. About all, the brain is divided into two hemispheres; one of them is adept in logical reasoning and the other is good at association of conceptions. The pallium of brain is also divided into many functional areas, such as vision area and olfaction area. In each of those areas, cells are arranged in a very orderly way: they are organized in narrow columns, which work together for special stimulation or task. This type of hierarchical organization has many advantages compared with classical ANN. It has excellent robustness in that partial absence of areas will not lead to serious malfunction of whole system. Learning in this structure can be divided into global learning and local learning more clearly and can be much quicker.
In fact, the architecture of some classical neural networks has partly applied the hierarchy principle. For example, MP model - Perceptron ¨C Multi-layer Perceptron ¨CBP network is a typical hierarchy architecture. But they have limited performance in the aspects of robustness and learning with huge volume samples. Large volume of high dimension samples will lead to need for so many nodes that it will be very difficult in learning and working. So we propose Hierarchy Neural Network based on hierarchy feature of biological information processing. It uses simple neural networks as basic units to construct complex neural network and allot learning task at global and local levels. We experiment this algorithm in synergetic neural network with the introduce of multi-local competition to resolve the problem of weak-robustness and slow-convergence resulted form sheerly global competition.
Multi-Agent System
I lead two groups in MAS as a ¡°bottom-up¡± method to apply my basic principles. Apart from Agent-based Distributed Control System and BEGA, we have done much other work, too. We introduced the ecological computation and immune computation to our university and had developed several experimental software by Visual C++ and Java.
Agent-based Distributed Control System
In March 2000, I proposed a distributed control system could be regarded as an agent system. Then I proposed the basic principles in enacting the ¡®laws¡¯ of control agent society. My work was highly recognized by Prof. Jiang Jianguo, the Dean of School Computer and Information. I help him to apply for a Key Project of MOE and a project supported by NSF of Anhui province. Prof. Jiang sent his students to our laboratory to research on those projects under my lead.
There are some selected paper abstracts of mine in this field.
Bao Jie, Gao Jun, Jiang Jianguo. Development and Intelligentization of Distributed Control System. Tech Memo of LIIP-HUT 2000-3-26 (in Chinese)
Abstract: We discuss a develop history of distributed control system in this paper by stages of DCS, Field Bus System and Intelligent Control Maintenance and technical Management System (ICMMS). Two important trends in the development of distributed control system are the intelligentization of control matters and increasing synergism of subsystems. Its further development has close relationship with agent technology and CSCW. The intelligentization of control system is not realized by preparing a lot of behavior rules in advance, but by the self-organization of control matters from the view of agent. The key issue in the self-organization is how to enact law system that can enable global cooperation between agents with incomplete local information under the guidance of synergetic principles, such as order increasing, hierarchy structure, progressive centralization and progressive mechanization.
Han Xiaozheng, Bao Jie etal. Learning In Mutiagent Distributed Control Matters. The Third international conference on computer-aided industrial design, Dec 2000 ,Hang Kong, China
Abstract:The history of distributed control system has seen DCS, FCS and ICMMS etc. The megatrends of the development of distributed control system are the intelligentization of control matters and the collaboration of sub-systems. The control matters have increasing demand for autonomy for response to environmental change. The main difficulty of this control method is how to deal with the interference between complex environmental change and control matters robustly and quickly. An effective solution is multi-agent-based DCS. Agent can change its control behavior by learning and can find out optimal or sub-optimum control scheme. For example, we experiment on a sensor-actuator agents control system and prove that SOM has better performance for avoid local optimal than traditional hill climbing method.
Behavior-Evolution-based Genetic Algorithm
Inspired by the possibility of behavior evolution I proposed BEGA in Jun 2000. It can be regarded as GA or MAS because its individuals can adjust its behavior responsive to environment change.
I and Mr. Wang Yixian (a junior graduate student) had received a foundation support from HFUT on ¡°Behavior-Evolution-based Genetic Algorithm and its application in network management¡±. I lead a group composed by students from three departments on this project.
There are some selected paper abstracts of mine in this field.
Bao Jie, Gao Jun, Wang Yixian. Behavior Information in Inheritance and Behavior Evolution Model. Information Processing, as a part of SPIE's 46th Annual Symposium. San Jose, CA, 2001. (Accepted, but was given up because of reason)
Behavior information in inherit: According to classical central dogma of molecular genetics, coded DNA information is transferred and translated by RNA, then control the compose of DNA. But, a lot of DNA segment is not corresponding to any protein. Reproduce and express gene is not only a data transfer and reproduce process but also a ¡° dynamic process¡±. Gene contain not only phenotype¡¯s segment (about 4% of DNA), but also segment that control organism behavior. A lot of experiments show that behavior had its inherit foundation, and it can be divided into inborn behavior and acquired behavior.
Model of Behavior¡¯s evolution: Traditional GA is based on classical central dogma. That¡¯s, inherit code determines phenotype, then determines fitness with no inheritance of behavior. Based on the hypothesis of behavior evolution, we propose that if not only translate inherit code into phenotype but also behavior, the inherit individual changed into a kind of agent. This kind of behavior character can be described as apperceive-effect-representation collection.
Behavior-evolution-based GA (BEGA) can also be divided into inborn algorithm and acquired learning algorithm. Main idea of inborn GA is: individual choose some certain behavior and strategy though competition until some strategy is in predominance or come into an Evolutionarily Stable Strategy (ESS). Acquired learning GA keys on the automated strategy generating.
Applications: algorithm is tested in several areas including intelligent e-mail management system, optimize design of feedforward network and pile foundation stress analysis.
Wang Yixian, Gao Jun, Bao Jie. Behavior Evolution-based Genetic Algorithm. Information Processing, as a part of SPIE's 46th Annual Symposium. San Jose, CA, 2001.
PURPOSE: GA is an effective way to solve complex problem. But it has obvious limitation. It not considers the life¡¯s behavior character. Heredity and evolution of life is regarded as only the evolution of structure without consideration of the storage and evolution of behavior information under the interaction between life and outside. Based on biology and current work of GA, we propose a new algorithm - behavior evolution based GA (BEGA).
METHOD :Algorithm keys on the concept of behavior evolution. In instinct model, we construct an instinct behavior set; each behavior in this set has a fitness function that can be used to evaluate its responsiveness. The set use receiver to receive information and respond differently to different information. With this information, fitness function of each behavior is been adjusted. Behavior strategy responsive to special outside information is determined by competition. In the learning model, we construct a learning behavior set in the same way as instinct behavior set, but behavior in this set can evolute into new behavior by operators of crossover, mutation etc. Behavior with low fitness will be eliminated, new behaviors are more advanced and complex than old behaviors. By repeating those operators, we can get a hierarchy behavior learning set.
RESULT:Using this algorithm as core method, we design an e-mail management system. This system can receive letters on-line, filtering information that we don¡¯t want to receive, and can also do something by the user¡¯s favorite (for example, classify the letters, rank the web sites). More important, the system¡¯s behaviors can evolute through learning.
Wang Chun, Bao Jie, Huang Xi. Application of Neural Network based on Behavior Evolutionary Genetic Algorithm in pile foundation. The International Deep Foundations Congress (ASCE) February 14-16, 2002 Orlando, Florida (under consideration)
Abstract: In pile foundation ,influencing elements of pile capacity are very complex ,which are non-linear and even non-differentiable and can't be expressed with a function directly .It seems BP neural network could solve this problem in that it can simulate any complex non-linear mapping theoretically .However ,it encounters low convergence rate and local minimum. We proposed a neural network based on behavior evolutionary genetic algorithm (BEGA) which BEGA is applied in learning algorithm.The BEGA retains the ability of stochastic global searching of traditional genetic algorithm, and endow the genetic individuals of behavior perception and evolution ability based on the fact that a large part of germ plasm in the organism do not be translated into proteins and correspond to behavior control. The BEGA has better global convergence and very strong self-adaptive ability with environment. We use BEGA to code network's weights and thresholds and form chromosomes and then decode. During pile construction, the interaction of pile and soil influences the pile capacity as well as too many uncertain elements that also influence the pile capacity but not always. By analyzing all the affecting elements and characteristics of BEGA neural network we propose a new conception¡ªposition parameter and consider construction process as behavior environment.
[Return to Jie Bao's Homepage]
