• 沒有找到結果。

具有適應與創造能力之智慧型控制系統(2/3)

N/A
N/A
Protected

Academic year: 2021

Share "具有適應與創造能力之智慧型控制系統(2/3)"

Copied!
6
0
0

加載中.... (立即查看全文)

全文

(1)

行政院國家科學委員會專題研究計畫 期中進度報告

具有適應與創造能力之智慧型控制系統(2/3)

計畫類別: 個別型計畫 計畫編號: NSC93-2416-H-006-004- 執行期間: 93 年 08 月 01 日至 94 年 07 月 31 日 執行單位: 國立成功大學工業與資訊管理學系(所) 計畫主持人: 陳梁軒 報告類型: 精簡報告 處理方式: 本計畫可公開查詢

中 華 民 國 94 年 5 月 20 日

(2)

行政院國家科學委員會專題研究計畫成果報告

具有適應與創造能力之智慧型控制系統 (2/3)

An Intelligent Control System with Adaptability and Creativity (2/3)

計畫編號:NSC 93-2416-H-006-004

執行期限:93 年 8 月 1 日至 94 年 7 月 31 日

主持人:陳梁軒 國立成功大學 工業與資訊管理學系

E-mail:lhchen@mail.ncku.edu.tw

一、中文摘要 本研究為第二年計畫,主要目的在於 建立控制系統的感知模組及適應模組。感 知模組用以偵測環境的變化,及評估系統 內部的平衡狀態(基於陰陽五行原理)。並 依據所蒐集的資訊,經過評估後產生適當 的反應。適應模組能能調整系統本身的行 為以適應環應的變化。 關鍵詞:智慧型控制系統、陰陽五行 1. Abstract

The project of second year is to con-struct the perception module and adaptation module for the intelligent control system. The perception module can sense the varia-tion of environment and evaluate the balance state of inner system based on the Yin-Yang and Five Elements principle. According to the gathered information, the perception module can react properly after evaluating the information. The adaptation module is to adapt the various environments by adjusting the system’s behavior.

Keywords: Intelligent Control System,

Yin-Yang, Five Elements

2. Objectives

In this year’s project, there are two ob-jectives we want to achieve: 1) construct the model that can sense the variation of outer

environment and balance status of inner sys-tem, i.e. the perception module, and 2) build the adaptation module for adapting the vari-ous environments.

1) Perception Module: Senses the

equilib-rium of the inner system and the variation of the outer environment. A Five Element Balancing Chart (FEBC) that employs the balancing principle of Five Elements originally from traditional Chinese phi-losophy, is developed to identify system equilibrium. In addition, four kinds of observed energy and image functions (OEI function) are presented to detect en-vironmental circumstances. Two OEI functions regarding resource utilization and environmental security are the trig-gers to activate the adaptation module. In addition, the creation module is activated when system imbalance is detected by FEBC, or unusual environmental varia-tion is found by the other two OEI func-tions.

2) Adaptation Module: Composed of the

ac-tion explorer [1], and the knowledge gen-erator. The action explorer explores new behavioral rules for the system by a self-exploration process, and the knowl-edge generator then transforms these rules to the corresponding knowledge repre-sentation.

3. Methodology 3.1. Perception Module

3.1.1. Measuring system equilibrium by Five Element Balancing Chart (FEBC)

(3)

The evolutionary scheme of FEBC is based on the aspects of cellular automata (CA). Von Neumann first developed CA in 1953, and then Burks [2] synthesized the previous work as the 29-state CA. CA cur-rently has been applied to several fields, e.g. random number generators [3], music pro-ducer [4], and robotic path planning [5]. A CA system is usually composed of a regular lattice of sites (cells or automatons), and each cell has some features that are updated in discrete time steps according to local evolu-tion rules, which are the funcevolu-tions of the states of the cell itself and of its neighbors. As a result, many simple components act to-gether according to the evolutionary rules in order to produce various patterns of behavior, and they often act the self-organization be-haviors.

The novel FEBC is formed by a number of cells (i.e. a regular lattice), and each cell is represented by an element that is one of the Five Elements. According to the evolutionary rules and the sensed behaviors of the system, the FEBC is able to constitute a specific pat-tern, which is used to determine if the system is in equilibrium. Before using the FEBC, all possible actions of the system should be classified into the Five Elements, Wood, Fire,

Earth, Metal, and Water, according to some

pre-classified items, as in [6]. The initial FEBC can be constructed according to the specified directions and locations of the five elements, as shown in Fig. 1. Once a control output is generated from the controller, it will be transformed to one of the five elements and filled in a blank cell of FEBC.

The evolution rules of the FEBC are de-fined as follows:

1) Rule 1. According to the creation and

con-trol cycles [6], fill the element, trans-formed from the controller’s output, into the neighborhood of an element that can create the controller’s element.

2) Rule 2. If the new element, generated by

the controller, is located on one of the four neighbor positions, i.e. front, back, left, and right, of an element that is controlled by the new element, the element will be

eliminated.

3) Rule 3. After filling in an element

pro-duced from the controller, select a kind of element having the maximal amount and change one element chosen randomly as its “creative” element. If two or more kinds of elements have the same amounts, select one randomly.

4) Rule 4. If the new element cannot be filled

according to Rule 1, or if at least one kind of element disappeared in FEBC, it is called a status of imbalance.

3.1.2. Measuring environmental variations

Four indicators, namely resource utili-zation, environmental security, environ-mental variety, and environenviron-mental equilib-rium, are used to measure stability of the en-vironment. The four indicators described be-low are then considered in the four OEI func-tions to measure the environmental varia-tions.

1) Resource utilization: Measures whether or

not the resources are gathered efficiently. In order to gather the resources, the system has to consume energy internally The re-source utilization u is defined as

i o u h h = (1) where h and o h are the total amount i of resources gathered and the total con-sumption of energy in the system, respec-tively.

2) Environmental security: A change of

en-vironment could cause the sensing of the system to be inaccurate. Assuming that the probability of sensory inaccuracy is p, the initial reliability r in terms of p is defined as

p

r= 1− (2)

However, the environmental reliability is changeable in the control process so that a variable reliability r is presented as v

(

p

)

v p c

(4)

where υp ~ N

(

0,σp

)

is a normal

distri-bution with mean 0 and standard deviation p

σ , and c is a constant. When the vari-able reliability rv is decreased so it is much

smaller than the initial reliability r, the control system cannot cope with the envi-ronment efficiently. The security measure s is therefore defined by the ratio of r versus

rv as

r r

s= v (4)

3) Environmental variety: The environmental

variety of is defined by the ratio of the number of distinct object patterns Ν p that can be detected (or sensed) by the system versus the total number of objects

o

Ν appearing in the environment. The more distinct object patterns there are, the higher the variety is. The pattern may be explained, as the target point and obstacles in robotic path planning. The variety indi-cator v is defined as o p v Ν Ν = (5)

4) Environmental Equilibrium: According to

the Yin-Yang theory, we first categorize the environmental objects into the classes of Yin or Yang [6]. Next, calculate the changing rate w, i.e.

u c

T S

w= (6)

where Sc denotes the total number of

ob-jects belonging to Yin or Yang within the time interval Tu. Let w+ and w− specify the

changing rates of Yang and Yin, respec-tively. The equilibrium indicator, b, is de-fined by the relative changing rates of Yang and Yin as

− + = w w b (7)

The environment is more balanced, if the indicator b is closer to one.

The above four indicators are then ap-plied to build up the OEI functions. Consid-ering the system energy and one of the four indicators, an OEI function is determined to measure environmental variation. The system energy e(, as described in the resource

utili-zation, can be determined according to the applications. In addition, there are no regular forms of energy function, which would re-quire it to be developed for a particular ap-plication.

The OEI function O~ is then defined as

( )

e i f

O~ = (,( (8)

where i( indicates the image representing one of the four environmental indicators, and

O~ can be expressed as any function of

en-ergy and image. The four OEI functions re-garding to the four indicators⎯ u, s, v, and

b⎯ are Ou ~ , Ob ~ , Os ~ , and Ov ~ , respectively. The thresholds of four OEI functions, u

O~ , O~b, O~s and O~v are εu, εb, εs and

v

ε , respectively; and they are defined within [0, 1]. For applications, the adaptive mecha-nism is activated, once O~u < εu or O~s <

s

ε ; in addition, the creative mechanism is enabled while O~b < εb or O~v < εv.

3.2. Adaptation Module

The adaptation module has two ele-ments: action explorer (AE) and knowledge generator (KG). The former is to search the new control actions and the later transforms these actions to the system knowledge. A three-stage self-exploration process is ap-plied to explore new actions for the AE. This process includes three critical points, i.e. failure, return, and lead points. On the basis of human adaptive behavior when faced with a dilemma, we first return to some past action in the return stage, i.e. the return point. For example, Fig. 2 indicates that the failure point is action 7, the return point is action 4, and the number of returned actions is C1 = 3. Second, in the lead stage, the number of new actions, C2, is determined, and the lead point

(5)

is the last action found. Figure 2 shows that the lead point is action e and C2 = 5. Third, in the exploration stage, C2 new actions are ex-plored, and denoted as C3, such as C3 = {Ea,

Eb, Ec, Ed, Ee} in Fig. 2.

The KG transforms the discovered be-haviors into symbolic rules for SN or fuzzy rules for FNN, as described in the following.

1) Production of Fuzzy Rules for FNN

Applying a simple algorithm proposed by Wang and Mendel [7] generates fuzzy rules. Using this algorithm, we can form the numerical input-output data pairs and gener-ate the fuzzy rules.

2) Production of Symbolic Rules for SN

The symbolic rules can be directly de-coded from the AE.

4. Self-Assessment of This Project

In the project of second year, we have successfully modeled the perception module and adaptation module. The perception mod-ule integrates the Chinese traditional phi-losophy, i.e. Yin-Yang and Five Elements theories, and the computer science to per-ceive the environment and system itself. It is important to the control system and it likes

human sensorium. The adaptation module can lead the system to adapt the different and unknown environments.

Reference

[1] L.-H. Chen, C.-H. Chiang, “New approach to in-telligent control systems with self-exploring process,” IEEE Trans. Syst., Man, Cybern. B, Vol. 33, No. 1, pp. 56−66, 2003.

[2] J. von Neumann, “Theory of Self-Reproducing Automata,” in: A.W. Burks, (Ed.), University of Illinois Press, Champaign, IL, 1966.

[3] M. Tomassini, M. Sipper, M. Perrenoud, “On the generation of high-quality random numbers by two-dimensional cellular automata,” IEEE Trans. Comput., Vol. 49, No. 10, pp. 1146−1161, 2000. [4] P. Dahlstedt, M.G. Nordahl, “Living Melodies:

coevolution of sonic communication,” Leonardo, Vol. 34, No. 3, pp. 243–248,2001.

[5] P.G. Tzionas, A. Thanailakis, P.G. Tsalides, “Colli-sion-free path planning for a diamond-shaped ro-bot using two-dimensional cellular automata,” IEEE Trans. Robot. Automat., Vol., 13, No. 2, pp. 237−250, 1997.

[6] D. Connelly, Traditional Acupuncture, Law of the Five Elements, 2nd ed., Traditional Acupuncture Institute, Columbia, MD, 1994.

[7] L.X. Wang, J. Mendel, “Generating fuzzy rules by learning from examples,” IEEE Trans. Syst., Man, Cybern.,Vol. 22, pp. 1414−1427, 1992.

(6)

Figure 1. An initial state of the Five Element Balancing Chart

Figure 2. An illustration of self-exploration process

W F E M A W: Wood F: Fire E: Earth M: Metal A: Water 1 2 4 5 6 7 3 a d c e Failure point Return point Lead point Return stage: C1 = 3 (7 → 4)

Lead stage: C2 = 5 (a →e)

Exploration stage: C3 = {Ea, Eb, Ec, Ed, Ee} Original action New action b

數據

Figure 1. An initial state of the Five Element Balancing Chart

參考文獻

相關文件

Let f being a Morse function on a smooth compact manifold M (In his paper, the result can be generalized to non-compact cases in certain ways, but we assume the compactness

2.1.1 The pre-primary educator must have specialised knowledge about the characteristics of child development before they can be responsive to the needs of children, set

 Promote project learning, mathematical modeling, and problem-based learning to strengthen the ability to integrate and apply knowledge and skills, and make. calculated

1.9 Chapters 3 to 7 cover the concerns and suggestions received and elaborate on our support measures covering the five proposed actions, including enhancing schools’

In order to facilitate school personnel of DSS schools in operating their schools smoothly and effectively and to provide new DSS schools a quick reference on the

Using this formalism we derive an exact differential equation for the partition function of two-dimensional gravity as a function of the string coupling constant that governs the

The principal chiral model has two conserved currents corresponding to the G × G symmetry of the action.. These currents are

Courtesy: Ned Wright’s Cosmology Page Burles, Nolette &amp; Turner, 1999?. Total Mass Density