• 沒有找到結果。

具有適應與創造能力之智慧型控制系統(3/3)

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(3/3)"

Copied!
7
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 成果報告

具有適應與創造能力之智慧型控制系統(3/3)

計畫類別: 個別型計畫 計畫編號: NSC94-2416-H-006-001- 執行期間: 94 年 08 月 01 日至 95 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: 完整報告 處理方式: 本計畫可公開查詢

中 華 民 國 95 年 10 月 12 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 (3/3)

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

計畫編號:NSC 94-2416-H-006-001

執行期限:94 年 8 月 1 日至 95 年 7 月 31 日

主持人:陳梁軒

國立成功大學工業管理科學系

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

一、中文摘要 本研究為第三 年之總結計畫,主要目 的在於建立智慧型控制系統的創造模組以 及驗 證系 統的 效率 。創造模組 有四個單 元,包括創造目標、創造程序、創造力衡 量與知識產生 器。創造程序能產 生有創意 的構想,由知識產生器將 這些構想轉換 成 系統的知識 。本研究以人工動 物之群體行 為模式,作為驗證系統之實例。 關鍵詞:智慧型控制系統 、創造程序、人 工動物 1. Abstract

The project of the third year aims to model the creation module of the proposed intelligent control system, and to demonstrate the system’s efficiency. The creation module consists of four elements: creative objective, creative process, creativity measurement and knowledge generator. The creative process is able to produce the original behavioral rules, and the knowledge generator is then to transform the novel rules to the system knowledge. The artificial animal (animat) colony is modeled and simulated to demon-strate the efficiency of proposed intelligent control system.

Keywords: Intelligent Control System,

Creative Process, Artificial Animal

2. Background and Objectives

Human intelligence has been studied and applied in many fields, e.g. philosophy, psychology, physiology, management, engi-neering, computer science and others. In common with all of these, intelligence re-quires the ability to sense the environment, to make decisions, and to generate actions. An intelligent control system (ICS), which en-dows intelligence within a control system, includes the three basic functions, i.e. per-ception, decision-making and action [1].

Intelligence as the ability to learn, rea-son and adapt has been widely investigated in the field of ICS; however, the creativity abil-ity of intelligence has been little studied. Creativity, as one of the greatest forms of in-telligence, has always been a challenge to AI. The arts, including painting, music, architec-ture, sculpture and the humanities, as well as social science such as management [2], edu-cation [3] and psychology [4], are not the only areas in which human creativity has been exhibited.

There are many fewer studies on crea-tivity for artificial systems or machines. In this study, we propose a computational model of creativity embedded in ICS, from the con-siderations of philosophy, psychology, and computer science. The proposed ICS consists of four modules: control module, perception module, adaptation module and creation module. The first modules have been pre-sented in the first two years in this project. This study, the third year of the project, will present the creation module and integrate the

(3)

four modules. Moreover, the animat colony model is presented to demonstrate the pro-posed ICS.

3. Methodology

The creation module plays the leading rule in the proposed ICS as briefed in below. 3.1. Creative Objectives

The creative objective has a great influ-ence on the anticipated results Among the various creative objectives, originality is im-portant. In this study, the creative objectives include originality, equilibrium, and adapta-bility.

With regard to originality, the creating work must be novel and unconventional. To achieve equilibrium, the creative process must be conformable with the balancing principles of Yin-Yang and Five Elements. For adaptability, the system should be able to adapt the environmental variations and oper-ate with acceptable efficiency.

3.2. Creative Process

Two main sub-procedures, i.e. the storm and association approaches, are embedded in the creation process, as described below. 3.2.1. Storm Approach

Storm approach has two operations as shown in Figs. 1 (a) and (b), i.e. the move operator and XOR operator, respectively. The move operator generates an initial action-trail, and XOR operator performs the exclusive OR computations to produce the new actions using the above action-trail. The move op-erator is as follows.

1) Construct a square matrix with c c size, where c is the number of all control ac-tions.

2) Randomly assign one of eight directional signs to each cell, as shown in Fig. 1 (a) with a 66 matrix.

3) Randomly assign an action sign coded by a number to each cell.

4) Randomly select an initial cell, e.g. cell (5, 1) in Fig. 1 (a).

5) Form an action-trail, starting at the initial cell and stopping when one of the four boundaries is reached, e.g. 6-2-1-4-5-2 in

Fig. 1 (a).

Subsequently, the XOR operator is per-formed using the above action-trail. First, the decimal number for each action sign in the action-trail has to be converted into binary notation. Then, carry out XOR operation for the first two binary numbers, i.e. 110 and 010 in Fig. 1 (b), and perform XOR operation again for the produced binary number and the third binary number. Iteratively, the XOR operation is carried out for the produced bi-nary number in the previous XOR operation and the next binary number in the action-trail. Therefore, the new actions are found as {6, 4, 5, 1, 4, 6} in Fig. 1 (b).

3.2.2 Association Approach

Association approach can be treated as a kind of genetic operations of genetic algo-rithm. The eight operators, i.e. crossover, shift, swap, inverse, translation, insertion, deletion and mutation, carry out the opera-tions in this approach. Traditional GA has three basic genetic operators: selection, crossover, and mutation. Nearchou [5] pro-posed another three operators: swap, inser-tion and deleinser-tion. We [1] proposed the shift operator. The above seven operators are all employed in this study.

3.2.3 Algorithm

The algorithm of creation process is shown in Fig. 2. In the second line of Fig. 2, K denotes the number of total executions in the for-loop. In the while-loop, nS specifies the number of performing storm and associa-tion approaches, and nA is the number of generating solutions better than the previous one. NS/NA is the upper bound of nS/nA. T is the temperature regarding the accepted probability of a poor solution, and  is the reduction rate of T. h is a random number within [0, 1] relative to the possibility for accepting the poor solution.

In the main body of this algorithm, the number of executions of the for-loop is K. The primary procedure within the for-loop, the while-loop, has two conditions, i.e. nS < NS and nA < NA. Before performing the algorithm, initial solutions 0

and 0

(4)

are generated randomly, and then two new solutions B and A can be produced based on storm and association approaches, respectively. Afterwards, choose a better so-lution, i.e. new, from B and A based on the creativity measures. Compare new with the previous solution , and replace  with new if new is preferred (new ). However, if new is inferior to , then generates a random value h. Thenew re-places , if hexp

new /T

. As the while-loop is done, the temperature T will

decrease to T at each time step. While the

for-loop is finished, the final solution is

re-turned as the best outcome with respect to the creation objectives.

3.3. Creativity Measurement

This study develop a creativity measure combining the three indicators for evaluate creativity.

1) Behavior Abundance,a, evaluates the luxuriance of system’s behaviors. The be-havioral triangular relationship is proposed for this purpose. In Fig. 3, the neutral ac-tions is categorized to the second class, and put the two kinds of extreme actions in Classes one and three, respectively. The measure a is defined as ) 1 ( 2 1 1 1 ,    

   A N i i i a N A  (1)

where i,i1 specifies the characteristic distance of behavioral triangular relation-ship between the ith and (i + 1)th actions, and NA is the number of actions. The char-acteristic distance i,i1 is defined as

1 1

, 

ii i i (2) where i/i1 specify the class numbers, i.e. 1, 2, and 3, of actions i/(i + 1), respec-tively.

2) Behavior Equilibrium,e, measures the equilibrium of the system behaviors de-fined as

 

   Y Y Y Y e , max , min  (3)

where Y/Y is the number of actions classified as Yang/Yin. Yin and Yang are the two major attributes of the theory of Yin-Yang originated from traditional Chi-nese philosophy.

3) Behavior Adaptability,d, measures the adaptive ability. The perceptive output connecting to resource utilization u is con-sidered to construct the indicator as

      u u d u u u    if , if , 0 (4)

where u is the threshold of u.

Once the above three indicators are de-termined, the creativity measure C~ can be defined as

a n d

f

C~   ,, (5) 3.4 Knowledge Generator

The knowledge generator converts the creative solutions produced by creation proc-ess to system’s knowledge.

4. Application: An Animat Colony

4.1. Animat Colony Model

The behavior model of animat colony is to find food and communicating with other creatures. The model has four entities, i.e., creative animat with intelligence, random animat with common behavior, food and ob-stacle. Five features of the creative animat are Gender, Life, Emotion, Vigor and Direc-tion.

There are 10 sensors (their values are 0 or 1): S1 concerning the existence of any animat locates around the creative animat within 10 by 10 cells. S2/S3indicates whether the creative animat meets an animat with the same/opposite sex. S4/S5 specifies whether the creative animat meets a food/obstacle. If the animat with opposite sex agrees for courting, S6= 1. If at least one foods locate in north/south/ west/east to the creative animat within 20 cells, S7/S8/S9/S10 = 1. For the 7 ac-tions, the Ar/Ae/Ap/At/Ac/Am/Aw mean

(5)

per-forming the rest/eat/play/talk/court/mate/ walk. The other 5 actions Atc/Atf/Atb/Atl/Atr specify the creative animat has to turn clockwise/to north/to south/to west/to east. An example of the actual simulation run is shown in Fig. 4.

4.2. Simulation Results

The simulation stops at the time step 157. The initial total amount of animats is 18. At time step 57, the animats amount in-creases to 24 due to the mating event. From time step 104, the animats amount decreases due to the death of random animats. The creator is activated at the time step 62, since the Op

~

< 0.15 (see Fig. 5). This message alerts the system to enhance the ability of gathering resources. The measure O~p is promoted to 0.2035 after 10 time steps. The behavior rule is produced as follows.

IF S1= 1or S3= 1 THEN do {Ar, Ap, At} (19)

This rule indicates that the creative animat will do the actions {Ar, Ap, At}, if at least one random animat is around the creative animat within 1010 cells, or the creative animat neighbors a random animat with the same sex.

5. Self-Assessment of This Project

In the project of third year, we have successfully modeled the creation module. Moreover, an animat colony model is simu-lated to demonstrate the proposed intelligent control system. The simulation results showed that the proposed system emerges its adaptability and creativity.

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] D.K. Beattie, “Creativity in art: the feasibility of assessing current conceptions in the school con-text,” Assessment Educ., Vol. 7 ,No. 2, pp. 175192, 2000.

[3] I. Marušič, “Some observations regarding the edu-cation of landscape architects for the 21st century,” Landscape Urban Planning, Vol. 60, pp. 95103, 2002.

[4] R. Conti, M.A. Collins, M.L. Picariello, “The impact of competition on intrinsic motivation and creativity: considering gender, gender segregation and gender role orientation,” Personality Individ-ual Differences, Vol. 30, pp. 12731289, 2001. [5] C. Nearchou, “Adaptive navigation of autonomous

vehicles using evolutionary algorithms,” Artificial In-telligence Eng., Vol. 13, No. 3, pp. 159173, 1999.

(6)

Figure 1. An illustration of storm approach

Procedure CREATION PROCESS

Define i, K, n ,S n ,A N ,S N , T,A, h  For i = 1 to K Let nS = 0 and nA = 0 While nS < NS and nA < NA Create initial 0 

; run storm approachB Create initial 0; run association approachA

A B newBest  

 ,

If new, new and nAnA1 If newand hexp

new /T

,

new    and nAnA1 End_of_If 1   S S n n End_of_While T T End_of_For Return ()

Figure 2. Algorithm of the creation process

Figure 3. Diagram of behavioral triangular relationship

3 2 5 5 3 3 5 1 2 4 4 3 1 2 4 2 4 2 4 6 6 2 1 3 5 2 4 1 2 6 2 6 4 2 4 2 6 2 4 1 5 2 4 5 1 4 6 XOR Operations 6 4 5 1 4 2 1 3 Close Relationship Far Relationship Close Relationship

(7)

Figure 4. Screen shot of an actual simulation run: ‘@’/‘O’/‘&’/‘X’represents creative ani-mat/random animat/food/obstacle

Figure 5. Sensory information of animat colony simulation 0 20 40 60 80 100 120 140 160 0 0.5 1 1.5 p O~  a p ~ -s p ~ . m p ~  0.15

數據

Figure 2. Algorithm of the creation process
Figure 4. Screen shot of an actual simulation run: ‘@’/‘O’/‘&amp;’/‘X’represents creative ani- ani-mat/random animat/food/obstacle

參考文獻

相關文件

Therefore, in this research, we propose an influent learning model to improve learning efficiency of learners in virtual classroom.. In this model, teacher prepares

- - A module (about 20 lessons) co- designed by English and Science teachers with EDB support.. - a water project (published

There is no general formula for counting the number of transitive binary relations on A... The poset A in the above example is not

Since Dolby AC-3(abbreviated as AC-3) is the main technology of the surrounding sound format, in this thesis, we proposes a data model for mining the relationship between

IPA’s hypothesis conditions had a conflict with Kano’s two-dimension quality theory; in this regard, the main purpose of this study is propose an analysis model that can

For obtaining the real information what the benefits of a KMS provides, this study evaluated the benefits of the Proposal Preparation Assistant (PPA) system in a KMS from a case

This study chose a qualitative research method to explore more in-depth information access strategy for the establishment of many commodities, institute of emphasis from

The purpose of this study is that in the future planning of new or converted semiconductor plant, the plant facilities to be demand for the plant systems