This action might not be possible to undo. Are you sure you want to continue?

**System Identification and Control Using DNA Computing Algorithms
**

Ching-Huei Huang, Chun-Liang Lin and Horn-Yong Jan

**Abstract—A DNA computing algorithm (DNACA) with an
**

electron-ion interaction potential (EIIP) decoding scheme is proposed to identify a class of transfer functions. The DNACA includes crossover, mutation, enzyme and virus operators providing a highly modular, flexible, and accurate self-organizing structure. Simulation study based on the De Jong’s test functions show its superior performance when compared with the improved and standard genetic algorithms (GAs). The algorithm is also applied to control design with the simplest controller through special frameshift mutation such as enzyme and virus.

system identification (ID) and control design. System ID is essential for control designs based on the process model, which go from PID controller [11, 12] to more sophisticated methods such as H 2 control [13] and is applicable to the robust control design [14-16]. With regard to system ID, DNA strings are created here to represent transfer function models in which codons are used to represent coefficients of the denominator and numerator polynomials. Using frameshift operators (enzyme and virus), the order and numeric values of the transfer function model can be refined to fit the objective of finding the simplest ID model. The proposed approach is verified by testing a group of the De Jong’s functions given in [17]. Verification of the simulated results shows that the proposed method performed well, even for the system models with wide dynamic responses. Application to robust control design is also investigated and simulated result is presented. II. BIOLOGICAL COMPUTATION ALGORITHMS A DNA strand comprises of two complimentary strings in which four nucleotide base: adenine (A), cytosine (C), guanine (G), and thymine (T)—are arranged in various combinations. Nucleotides are paired along with two strings. One DNA string can be separated from the other through chemical process. As the other string is a perfect compliment, for the purpose of computation simplicity only one string is needed to further operations in next generation. The biological computation algorithm calculates the fitness function and gets the best solution during evolution of generations in terms of DNA coding schemes. The property of electro-ion interaction potential (EIIP) decoding can be applied in biological computation process and designed to calculate the fitness values. The system computation processes are as follows. A. The DNA Computation Algorithms The design flow of the DNACA and decoding process with EIIP as shown in Fig.1 is described in the follows. Step 1: Randomly generated a group of DNA sequences. Step 2: Perform DNA coding scheme for each DNA sequence. Step 3: The DNA codons within the DNA sequences are translated to amino acids. Step 4: The amino acids are converted to the EIIP levels.

Keywords—DNA computing algorithm, Electron-ion interaction potential, Systems identification

I. INTRODUCTION ECENTLY developed DNA computing algorithms (DNACAs) have inspired new methods that can simultaneously solve the parameter and structure optimization problems. The DNACAs based on the concept of bimolecular evolution was first developed by Adleman in [1]. Maley further detailed this kind of algorithms in terms of chemical processes and computer programming [2]. The operational features inherent in DNACAs make the algorithms implementable in the future DNA computers which are over a billion times possible to implement more computationally efficient than the conventional computers. The massively parallel nature of DNA in those computers means that computation may be millions or billions of times beyond today’s supercomputers [3-7]. In this paper, accuracy of the proposed DNACA with or without frameshift operators is verified first by numerical tests with the electron-ion interaction potential (EIIP) [8-10] decoding scheme. Two of the competitive applications of DNACAs are on the

Manuscript received January 31, 2009. This research was sponsored by National Science Council, Taiwan, R.O.C. under the Grant NSC 96-2628-E-005-086-MY2. C. H. Huang is with the Department of Electrical Engineering, National Chung Hsing University, Taichung, 402 Taiwan, R.O.C. (e-mail: d9564201@mail.nchu.edu.tw). C. L. Lin is with the Department of Electrical Engineering, National Chung Hsing University, Taichung, 402 Taiwan, R.O.C. (phone: +886-4-22851549 ext 708; fax: +886-4-22851410; e-mail: chunlin@dragon.nchu.edu.tw). H. Y Jan was with the Institute of Electrical and Communications Engineering, Feng Chia University, Taichung, 407 Taiwan, R.O.C.

R

Issue 3, Volume 2, 2008

108

enzyme and virus operations.0000 【2】0. Each DNA string representing the transfer function is encoded as a vector in the following form: S = [S1 S2 Sk ] (1) Individual DNA strings S IN D . B. consider a transfer function consisted of a gain. IND = 1.0000 【0】0. IND=1. especially amino acids.0373 【4】0. Code of amino acid GCT GCC GCA GCG CGT CGC CGA CGG AGA AGG GAT GAC AAT AAC TGT TGC GAA GAG CAA CAG GGT GGC GGA GGG CAT CAC ATT ATC ATA TTA TTG CTT CTC CTA CTA AAA AAG ATG TTT TTC CCT CCC CCA CCG TCT TCC TCA TCG AGT AGC ACT ACC ACA ACG TGG TAT TAC GTT GTC GTA GTG Three-letter code ALA ARG ASP ASN CYS GLU GLN GLY HIS ILE LEU LYS MET PHE PRO SER THR TRP TYR VAL Ala（12） Arg（11） Asp（17） Asn（15） Cys（19） Glu（18） Gln（14） Gly（13） His（5） Ile（2） Leu（9） Lys（16） Met（3） Phe（1） Pro（7） Ser（10） Thr（8） Trp（20） Tyr（4） Val（6） One-letter code A R D N C E Q G H I L K M F P S T W Y V Amino acid Alanine Arginine Aspartic Asparagine Cysteine Glutamic Glutamine Glycine Histidine Isoleucine Leucine Lysine Mothionine Phenylalaine Proline Serine Threonine Tryptophan Tyrosine Valine Electron-ion interaction potential values 【2】0. Translation of DNA strands code.0761 【0】0. the representation possesses three merits in biology: (i) measure the chemical properties of bases. Basically. G}.0242 【0】0. Issue 3. Moreover. Electro-ion interaction potential (EIIP) applied here generates initial population and its subsequent offsprings. 2. DNA-Coding scheme Biologists have discovered 20 amino acids encoded by 64 codons. which are expressed in the versatile combinations of {A. each amino acid was represented by a specified number. k. and (iii) directly relate to DNA chromosomes. these numbers are essential to build a physical and mathematical model which interprets protein sequences information using signal analysis methods.0198 【3】0. 2008 109 .0057 Fig. First.0050 【1】0.… .0036 【3】0.…. a gene string for DNACAs starts with the codon ATG and ends with the codon TAA. SINF_G =[SSTA SINF_G(1) SSTP SSTP_INF]. Volume 2.0829 【4】0. (Sorting and saving the best solution decoded with EIIP in the DNACA). Table I. 2 Normalized EIIP wheel of amino acids. Fig. SSTP . TAG or TGA.0941 【2】0.k.0548 【2】0. T. The EIIP describes the average energy states of all valence electrons.0058 【3】0.0516 【1】0. are defined as SIND =[SINF SPARA]. which was referred to as the unique electron-ion interaction potential and it was irrespective of its position in a DNA string in [18]. Step 6: Calculate the fitness value (the fitness function may incorporate with a penalty term).0371 【3】0. SINF_D =[SSTA SINF_D(1) SINF_D(7) SSTP SSTP_INF]. As in biological DNA coding scheme. 1 Flowchart of DNACA for the solution search.1263 【0】0. Step 8: Repeat. its value for 20 amino acids and five types EIIP wheel nucleotides are summarized in Table I and Fig. C. a numerator polynomial and a denominator polynomial.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING Step 5: Perform DNA computing process with crossover. mutation. In the method presented.0946 【1】0.TGATAA} . SSTA ATG. (ii) preserve information about the properties of the bases.0823 【4】0.0829 【1】0. {TAG. SINF =[SINF_G SINF_D SINF_N].0959 【4】0. amino acid codon and EIIP value. Step 7: Store the results.

Accordingly. d d and the number of crossover points.ρ1 . is defined as Issue 3. SPARA_N =[SPARA_Nf SPARA_Ns ] =[SSTASCD1_1SSTPSSTASCD1_2 SSTASCD1_n SSTPSSTASCD2_1SSTPSSTASCD3_1 SCD3_m SSTPSSTP_D]..ρ . N (2) where the constants r. . as follows (i ) r∑DNAAA_ EIIP ⋅ 5i−1 − M i =1 L The two-point crossover operation is a process of exchanging DNA information.1. SSTP_N TAG where S IN D consists of the structure information S INF and the parameter information S PARA . If the population doesn’t contain enough encoded information to solve a particular problem. . ' that produces an intermediate population P ' t = ( a1' t . where Sk = ( x1x2 x3 ).. After crossover operation. a(i +1). i. are defined as the polynomials with the first or second-order polynomial. M and N control the range and (i ) resolution of the parameters. ai .ρ2 . aλ ) . { { Fig. excluding lead and end codon. the lead codon and the combination of end codons corresponding to each DNA string have to be excluded while single or multiple active points are randomly assigned for an operation. a(i +1). aλ ) in the t -th generation. The DNA sequence is divided into three sub-DNA sequences by the two points and crossovered each other by swapping the sequences between the first and third sub-DNA sequences..2 .. SPARA_D =[SPARA_Df SPARA_Ds ] =[SSTASC SSTPSSTASC N1_1 N1_2 SSTASC SSTPSSTASC SSTPSSTASC N1_nn N2_1 N3_1 SC N3_ m n SSTPSSTP_N]. a λt ) from t the original population P t = (a1t . is SSTP_G TGA. . ai .2 . . ai .L ⎤ ⎫ 1 1 2 2 ⎪⎣ ⎦⎪ =⎨ ⎬ ⎪ ⎡ai . This exploits current genetic potential. 3. and a(i +1) . Crossover Operation with DNA sequences The crossover operation causes recombination and information exchange of DNA sequences in a probabilistic way.ρ . Subscripts.( ρ +1) . the original DNA sequences between the two crossover points are exchanged as illustrated in Fig. a λt ) t from the original population Pt = (a1t .. Volume 2. Before executing the evolutionary process. the structure information S INF represents the coding information for the gain ( S PARA_G defined as ⎛ ⎧ at ⎫ ⎞ ⎧ ai' t ⎫ ⎪ ⎪ ⎪ ⎪ = ctp ⎜ ⎨ t i ⎬ ⎟ . a(i +1). none of the mixing strands can produce a satisfactory solution. G. f and s . ). 3 Example of two-point crossover operation. C. . In addition to the crossover and mutation operators. L is the length of the corresponding parameters in S PARA . N are used to identify lead and end codons.. which will be explained in more details subsequently. ∀i ∈{ 1. enzyme and virus. a(i +1) ∈ Sk . D. xi ∈{A.. Three types of crossover strategies have commonly been adopted [19].1. ai . .( ρ1 +1) .e. a DNA sequence can be drastically changed. The second crossover strategy is multi-point crossover.3. λ − 1 } ⎨ 't ⎬ ⎜ ⎪a(i +1) ⎪ ⎟ a(i +1) ⎪ ⎪ ⎩ ⎭ ⎭⎠ ⎝⎩ ⎧ ⎡a( i +1). SPARA_G =[SSTASCC SSTPSSTP_G]. ctp .. SSTP_D TAA. The mutation operator capable is spontaneously generating new strands which provide a mechanism to maintain the population’s diversity. T . a( i +1). SINF_N(7) SSTP SSTP_INF]. a(i +1). By this method. the decoded parameters are calculated. . a(i +1). there are two frameshift operators.( ρ2 +1) . .. SSTA and S ST P_ S with S s s INF. . DNAAA _ EIIP is the number of the i -th amino acid of S PARA . Two DNA sequences are chosen from the current population and swap partial genes with each other. and 1 ≤ ρ1 < ρ2 ≤ L . 2008 110 .. ai ∈ Sk . C}. ρ1. The first one is two-point crossover operation.( ρ +1) .2k + 1..L ⎤ ⎪ ⎦ ⎩ ⎣ ⎭ 't t 't t ai . denominator polynomial ( S PARA_D ) and numerator polynomial ( S PARA_N ) based on the combination of different numbers of amino acids. depending on the crossover points.. Normalization is performed based on the categorization of EIIP.. The variety of crossover algorithms work depending on the crossover rate ( Pc ) SPARA =[SPARA_G SPARA_D SPARA_N]. c m p . that ' produces an intermediate population P ' t = ( a1' t .INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING SINF_N =[SSTA SINF_N(1) SSTP_INF TGATGA. ai . ai .. ρ2 indicate the crossover poin t. . The DNACAs consist of several operators in which a crossover operation is used to generate a new strand that will retain beneficial features from the parent generation.

and a(' ti +1) . . .e.1 . and 1 ≤ p < L and p indicates the mutation point length L is generated. if mak = C(G) ⎟ ∨ ⎜ ⎢ . rj . . and a(' ti +1) ..3.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING ⎛ ⎧ ait ⎫ ⎞ ⎧ ai' t ⎫ ⎪ ⎪ ⎪ ⎪ ⎨ ' t ⎬ = cmp ⎜ ⎨ t ⎬ ⎟ . To realize the effect. 0 ≤ rj ≤ 1.k ⎤ ⎞⎪ ⎪⎛ ⎡ ai . λ } {ai't+1} = {ai.C ≡ G. that produces an ' intermediate population P ' t = ( a1' t . Virus Operation with DNA sequences From the biological viewpoint. p + 1.L } ai't . L) is generated randomly. ai. 4 Example of multi-point mutation operation.( j ) ∈ Sk where cvp = {ait.. The mask ma is defined as m a = ∪ Lj =1 { { Fig.. The crossover operation is performed in bit level. . distinctly from each other.( ρ1 ) . A mask.(ρ1 ) . aλ ) {([A (T )]. The Watson-Crick in the t -th generation and is defined as {ai't } = cvp{ait } ∀i ∈{ 1. 2. ai.. λ} .k ⎤ = ∪ L =1 ⎨⎜ ⎢ k ⎥ . i. an inserted DNA string makes an incursion into its parent’s string to form a new one.( j )} ={ai..k ⎤ ⎞⎫ ⎪ ⎪ = ∪ L =1 ⎨⎜ ⎢ . . a(ti +1) ∈ Sk where Sk = ( x1 x2 x3 ). Otherwise. Mutation injects new information into the generation that may be important but not contained in the initial population or lost in the selection process. The third crossover strategy is uniform crossover. . a binary array with ⎧ ai . 5. ai. . a λt ) from the original t population Pt = (a1t . more than one crossover point is selected in a pair of chromosomes. it is set to be C(G). each pair of DNA sequence is crossovered at different points because no predefined mask was used. 1 ≤ ρ k . if ρ k ≥ ρ mp ⎟ ⎬ ⎟ ⎜ ⎟ ⎜ ⎪⎝ ⎣ a( i +1). ait ∈ Sk . 3. In the uniform crossover strategy. k = ⎨ ⎩ ai . for the multi-point crossover strategy. The process of crossover positions in DNA sequence is selected randomly. ai. . ait ∈ Sk . 2. and ait ∈ S k .. 2008 111 . . C}. ai. k .T.2 . is defined as ⎧ ⎫ ⎞ ⎛ ⎡ a( i +1). xi ∈{A.(ρ j ) . T . . aλ ) . if r j ≥ ρ ma ) ∨ ([C(G )]. ai' .. . D. let the codons inducing virus influence be cvp that produces an intermediate population ' t P ' t = ( a1' t . the crossover positions are predefined in a mask..3. 2k + 1. T . it is helpful for escaping local optimal. G. the process of uniform crossover is performed bit by bit in a pair of DNA sequences. C} Similar to the multi-point crossover strategy. k . The operation of mutation m p . p − 1. . A=T.( ρ j ) .k ⎦ ⎩ ⎭ ai' t . ai . a λt ) from the original population P t = (a1t . 4. a('ti+1) ∈ Sk and ait. λ − 1} ⎧⎛ ⎡ ai ..k ⎥ ⎟ ⎜ ai . k ∈ S k where Sk = ( x1 x2 x3 ). Volume 2. . if ρ k < ρ mp ⎟ ∨ ⎜ ⎢ a ⎥ . .. and Sk = (x1x2 x3 ).2.. G.. λ − 1} complementarity is commonly considered in mutation operation.k ⎤ ⎞ ⎛ ⎡ a( i +1). ai. C} ∪ S t m (PARA_i (IND)) Insertvp Xk t S m (PARA_i (IND)) ∪ Insertvp Issue 3.. . ai. 2k + 1. for k ∈ {1. for k = p 't ai ∈ S k .( ρ2 ) .T=A. ⎜ ⎪ a( i +1) ⎪ ⎟ a( i +1) ⎪ ⎪ ⎩ ⎭ ⎭⎠ ⎝⎩ ∀i ∈ {1. .(ρ2 )}. ai' t = m p ( ait ) ∀ i ∈ {1. the major cause of disease is usually that a virus has intruded into creatures.. aλ ) . ρ mp ≤ L. If the j -th random number.. a λt ) from t the population P t = (a1t ... C}. 1 ≤ ρ1 < ρ j < ρ2 ≤ L. 2. . xi ∈{A.k ⎦ ⎠⎪ ⎠ ⎝ ⎣ i . All DNA sequences in a population are crossovered at the same positions. L} ρ k indicates the position of multi-point crossover ρ mp denotes the threshold In this crossover strategy. T . ( j = 1.. E. if r j < ρ m a ) } The uniform crossover is defined as ⎛ ⎧ ait ⎫ ⎞ ⎧ ai' t ⎫ ⎪ ⎪ ⎪ ⎪ ⎨ ' t ⎬ = cun ⎜ ⎨ t ⎬ ⎟ ⎜ ⎪a( i +1) ⎪ ⎟ ⎪a( i +1) ⎪ ⎩ ⎭ ⎭⎠ ⎝⎩ ∀i ∈ {1. modifying the original codons of DNA strings and reproducing the infected strings.. Through this mechanism. In other words. Mutation Operation with DNA sequence The mutation operation is to change DNA sequences in a probabilistic way which is determined by the mutation rate Pm .. a(ti +1) ∈ Sk where Sk = ( x1 x2 x3 ). if mak = A(T) ⎟ ⎬ k ⎟ ⎜ a( i +1).. xi ∈{ A. ait ∈ Sk . cun that ' produces an intermediate population P ' t = ( a1' t . G. G. xi ∈ { A. Mutation operation capable of spontaneously generating new strand provides a mechanism to maintain the population diversity as shown in Fig.k ⎥ ⎦ ⎦ ⎪⎝ ⎣ ⎠⎪ ⎠ ⎝⎣ ⎩ ⎭ ai' t . The direct consequence of this operation is that the resulting DNA string gains extra information in the next generation as displayed in Fig. is the j -th element in the binary array set as A(T).G ≡ C .

( ρ j ) . ai .. might not possess the simplest so that there won’t be excessive control commands. ∈ S k and a ∈ Sk respectively.. i.. The newly formed codon breaks away from the doped position and the reacted remainder of codons are joined together to form a new one. Here. a λt ) from t the population Pt = (a1t . the weighting function Wrc ( s) works ω ∏ (s + C D1_ i )∏ ( s 2 + C D 2 _ i s + C D 3 _ i ) i =1 i =1 md (3) where n n = n n + 2 m n and n d = n d + 2 m d with nd ≥ nn assure the transfer function to be proper. .( ρ ) . . ait. The resulting transfer function..INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING Fig.ait. especially at higher frequencies.( ρ ) that 1 structure. Extension to Control Design Consider the control system shown in Fig. C}. ai . G. whereas J2 and J3 calculate. L } P0 ( s ) denotes the system to be identified. the one is not necessarily with the minimum order. The following S m ( P A R A _ i ( IN D )) t t Deleteep : S m ( P A R A _ i ( IN D )) ∅ D elete ep Fig. F. ai .e. ∀i (8) where λ i are the roots of the characteristic polynomial Λ ( s ) of the nominal closed-loop system. the start code. Volume 2.( ρ2 ) } ∀j ∈{ 2.ait. the magnitude and phase of the frequency where cep = {ait. the least mean square error scheme is adopted to construct a transfer function so that it ultimately mimics the dynamic behavior of the identified object. f t is the total sampling number over the testing frequency interval. A( jω ) ∞ = sup A( jω ) . 7 with its controller CE ( s ) described in the form of (3). a . The net effect is to help the reacting molecules go through chemical changes more rapidly. Configurable ID model There have been traditional or advanced approaches widely applied to deal with the system ID problem [20-22]. .( j ) phase(⋅) indicate.. however.( ρ1 ) .( ρ 2+1 ) . the overall errors of the magnitude and phase over the whole spectrum.ait. as follows Issue 3. . {ai' t+1} = {ai . In traditional system ID methodologies. xi ∈{A. 6. ANALYSIS METHODS A. The condition for internal stability is ensured by incorporating the following constraint: g1 Wrc ( jω )CE ( jω ) 1 + CE ( jω )Gn ( jω ) <1 ∞ (9) ∏ (s + C i =1 nd i =1 nn N 1_ i )∏ ( s + C N 2 _ i s + C N 3 _ i ) 2 mn where Gn (s) is the nominal plant model which could be obtained by using the previous ID technique. 6 Illustrations of enzyme operation. ait.( ρ1−1 ) .. respectively. a 't i t i 't i +1 . 2. aλ ) in the t -th generation and J3 (SIND ) : PAe = ∑ phase(PE ( jωi )) − phase( P0 ( jωi )) i =1 ft {ai' t } = cep {ait } ∀ i ∈ { 1. B. To realize the effect. 5 Illustrations of virus operation.2 . λ −1} and 1 ≤ ρ1 < ρ j < ρ2 ≤ L Sk = ( x1x2 x3 ). T . a ∈ Sk . Enzyme Operation with DNA sequence The enzyme mutation works to separate and connect two DNA substrings while removing an intermediate fragment. ait.λ } . and Mag (⋅) and response. The direct consequence of this operation is that the resulting DNA string loses some information in the next generation. 2008 112 .( ρ1+1 ) .( ρ2−1 ) . 3. Consider a class of transfer functions modeled as follows PE ( s ) = CC condition is a fundamental requirement for the nominal closed-loop stability: Re λi ( Λ( s) ) < 0. t i . let the codons eliminated by enzyme effect be cep . The conditions for robust stability and sensitivity reduction have been well known [5-7] and are given. the stop code. ai . We consider here a novel application of DNACA to tackle the problem. J1 denotes the magnitude error at the lowest frequency. III.3 .( j ) } ={ait..1 . as displayed in Fig. ω Lf represents the lowest frequency. respectively. The objective function for the current problem is defined as J = ∑ wi J i i =1 3 (4) where wi is the weighting factor for the corresponding term: J1 ( S IND ) : GeLow = ft Mag ( PE ( jω Lf )) 10 − exp(log( ω Lf )) − Mag ( P0 ( jω Lf )) 10 − exp(log( ω Lf )) (5) J2 (SIND ): Me = ∑ ln(Mag(PE ( jωi ))) − ln(Mag(P ( jωi ))) 0 i =1 (6) (7) 2 ' produces an intermediate population P ' t = ( a1' t .

2. objective functions considered should contain not only system performance indices but also structure information of the controller. Volume 2. α1 + α2 = 1 dr ( t ) + − (16) (17) e (t ) CE ( s ) uH ( t ) Gn ( s ) ( I + ΔGn ( s )) + + y (t ) J s ( S IND ) = 1≤ IND ≤ μ IND nd IND max nd Fig . the weight shifts gradually to the parameters selection emphasizing on the system performance improvement. ∀ω ≥ 0. M o with where δ ∈ (0. is used to control relative importance between stability and performance. nn ) ⎜ 3 i =1 max {ε p . Objective function The problems of control design can be viewed as requiring the discovery of a controller or a control strategy that takes the output variables or state variables of a problem as its inputs and produces the values of the control variable(s) as its outputs. the penalty term p ( ⋅) is defined as ⎛ 1 3 ⎞ Δbi ( ⋅) p ( ⋅) = ⎜ 1 − ∑ ⎟υ ( nd . Thus the systems may have poor relative stability. Balance between the two factors with the generation of evolution could be modified by appropriately adjusting the constant ς . i = 1. β represents the desired emphasis on controller’s complexity. Ess with 0 ≤ Ess ≤1 being the normalized steady-state error. Therefore. The cost formulation covering the error energy over the whole time horizon of interest has been used extensively for both deterministic inputs and statistical inputs. the following F′ .7 Typical closed-loop control system with uncertainties.1). nn ) = ⎨ and Δbi (⋅) = max {0. ∀ω is the weighting function used to 0 ≤ Mo ≤1 being the normalized maximum overshoot. Control design simultaneously satisfying above constraints ensures Wrs ( jω ) CE ( jω ) Gn ( jω ) 1 + CE ( jω ) Gn ( jω ) + Wrs ( jω ) < 1. DNACA-Based control design Although a complicated controller usually enables superior performance. a simpler structure with acceptable performance is more practically desirable. The term Wrs ( s ) satisfying ΔGn ( jω ) ≤ Wrs ( jω ) .5 ≤ β ≤ 1. Issue 3.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING g2 Wrs ( jω ) CE ( jω ) Gn ( jω ) 1 + CE ( jω ) Gn ( jω ) ∞ < δ (ω ) (10) objective function in time domain is defined to directly reflect the transient performance: J 2 = ε1 (1 − M o ) + ε 2 (1 − Tr ) + ε 3 (1 − Ess ) 2 2 2 (13) g3 Wrs ( jω ) 1 + CE ( jω )Gn ( jω ) < 1 − δ (ω ) ∞ (11) where the weighting factors ε i ≥ 0 with ∑ε i =1 3 i = 1 .3 with b1 = 1 . The DNA programming is well suited to resolve control design problems where no exact solution is known and where an exact solution is not required [23]. b2 = δ and b3 = 1 − δ . p (t ) where F ′(⋅) is the unconstrained fitness term converted from the following objective function: J all = (1 − β e−ς g ) J p + β e−ς g J s . Minimization of the following quadratic performance index is introduced here for the requirement of tracking accuracy: J1 = 1 T and g is the generation number. system designs by this criterion tend to display a rapid decrease in the large initial error. The multiplicative uncertainty is used here for its simplification and direct relationship with the complementary sensitivity function. 2008 113 . Tr with 0 ≤ Tr ≤ 1 being the normalized rise time. To complement the weakness. if nd < nn (18) υ ( nd . ς > 0 where (15) J p ( SIND ) = α1 J1 ( SIND ) + α2 J 2 ( SIND ). The linear ranking approach can be used to convert J all to ∫ T 0 eT ( t ) e ( t ) dt (12) where T is chosen sufficiently large so that e(t ) for T < t is negligible. if nd ≥ nn ⎩0. which possesses a sufficiently high gain in low frequencies to get a disturbance suppression property and eliminate the steady state error. gi (⋅) − bi }. However. The fitness function for the current problem is defined as bound the multiplicative uncertainty. Δbi ( ⋅)} ⎟ ⎝ ⎠ where ⎧1. 0. ε p is a small positive constant. The formulation of (15) places a heavier weight on the control structure selection during the early generations. ∀ω ≥ 0 1 + CE ( jω ) Gn ( jω ) f (⋅) = F ′(⋅)p (⋅) (14) which is a sufficient condition assuring robust performance of the control system simultaneously subject to plant uncertainties and external disturbances.

4 0.95 (b) This is a multi-modal function.97 0.985 Fitness value 0.2 0. i = 1.12 ≤ xi ≤ 5. NUMERICAL STUDY To examine applicability and efficiency of the proposed DNACA. FD = 1 + Di ( x ) (|1 − D4 | +1) (24) For each test function. respectively. respectively. D4 = 0. the efficiency of DNACA with/without frameshift operators is illustrated as in Table III.3 0 5 10 15 20 25 30 Iterations 35 40 45 50 This is a typical test function to deal with the optimization.5 0. 2 j = 1. The results of the solution are searched by using the proposed DNACA as shown in Fig. and 3-dimentional quadric which function with zero-mean Gaussian noise.6 0. the iteration and population numbers are set to be 50 and 10. Taking D4 and D5 as the example.9 0. D3 = ∑ ix + Gauss(0. 5 .99 0.1 0 0 5 10 15 20 25 30 Iterations 35 40 45 50 1 FD = 1 1 2 .3 0.6 0.048 ≤ xi ≤ 2.4 0. 1) is a minimum of zero and it is very difficult to perceive because a deep parabolic valley is along with the relation x2 = x12 . D2 = 100( x2 − x12 ) 2 + ( x1 − 1) 2 .356 ≤ xi ≤ 65.965 0.048 0.9 0. 8 and Table II.7 Fitness value 0.7 Fitness value 0.002 − ∑ i =1 25 1 j + ∑ ( xi − aij ) i =1 2 6 .8 and 0.98 0. Both of the probability rates of enzyme and virus are 0. (d) Issue 3. Volume 2.2.5 and FD2 for D4 are given.356 (22) 1 0.1 0 0 5 10 15 20 25 30 Interations 35 40 45 50 (21) This is a continuous. it simultaneously prevents the redundant candidate strings from being repeated in the evolutionary process. − 5. − 2.12 i =1 3 (23) This is a generalized rastrigin’s function. − 1.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING IV. − 65. 9.7 ∑x i =1 3 2 i .975 0. unimodal.3.4 (20) 0.6 0. It possesses multiple local optimums that is suitable to examine performance of the optimization algorithms. 1 The corresponding fitness functions FD for D1. The point (1. 3.8 Fitness value 0.4.5 0. it has 25 local minimums lying approximately at the points a ij .28 ≤ xi ≤ 1. .96 0.2.12 (19) 0. Convergence of the fitness values for D4 and D5 is displayed as in Fig.9 0.12 ≤ xi ≤ 5. These results demonstrate efficiency and performance of the presented algorithm. respectively. − 5. 25 .995 0.3 0.8 0. i = 1.8 0.955 0.1). 2008 114 . the following De Jong’s test functions are considered: D1 = 1 0.28 i =1 4 i 3 (a) 1 0. as 0 5 10 15 20 25 30 Interations 35 40 45 50 (c) 1 0. D5 = ∑ [ xi2 − 10 cos(2π xi ) + 10].2 0.5 This is a simple 3-dimensional parabola with a spherical constant-cost contour. It is seen that the DNACA produces not only accurate solutions. convex. 2. The probability rates of crossover and mutation are 0.

1 0 0 5 10 15 20 25 30 Iterations 35 40 45 50 ACGGCGCC G CATTTCTCA GCCCACGA GTTAAATA A DNA string and TGAAGTC(-3 (-31. Clearly. 2008 115 .1068 2.6. Table II.012 ⎠⎝ 0. Moving further to the subsequent stages of evolution with the result obtained in the first stage as the initial model.3 0. 10(a). 26] for the De Jong’s function tests.2. the parameters and structure of the system model identified at this stage need to be refined. Volume 2. D4 D5 D4 D5 The parameter settings for the DNACA are set as follows: the population size is 70. (b. see Figs. (a.9 0. The summarized details are shown in Table IV. For the first frequency interval. the ID error attenuates with the increasing frequency interval of interest. Condition Function Fitness value Required generations for convergence Length (codons) 1.7587 ⎝ 300 ⎠⎝ 0. 9 Convergence of the fitness values for D4 and D5.991 1. The result reaches its optima over the whole frequency range after the fourth stage of ID.0164) DNACA is then applied to identify the following system 2 2×1.0911 0. Fig. the final results are shown in Fig. the DNACA generates a transfer function approximating the low frequency behavior of P0 ( s ) .000 With frameshift operations D4 Without frameshift operations With frameshift operations D5 Without frameshift operations 0.000 0. 10(b)-10(d). the standard GA [24] and the improved GA [25. As it is displayed in Fig.168 ⎠ (25) (e) Fig.7587 ⎠ P (s) = 0 2 2× 2.2 6. the maximum generation is 800. 8 Convergence of fitness values through DNACA.5 0. For each section.6239) G (31.1068 ⎠⎝ 2. (e) D5 . [10-2. every interval performed 200 generations of evolution). there exhibits a slight discrimination in the Bode magnitude and the phase plots of the two systems.038⎜ 2 2 0. the parameter range is [-200. D1 ( x ) D2 ( x ) D3 ( x ) D4 ( x ) D5 ( x ) Table III.5653) 1. the DNACA is used to optimize the structures and parameters of the identified transfer function.c) with frameshift operation. Comparison of DNACA’s efficiency with/ without frameshift operators for D4 and D5 .10.d) without frameshift operation. four frequency intervals are considered individually including [10-2. the rates of enzyme and virus are 0. 10]. (d) D4 . (c) D3 .0911 ⎠⎝ 0. respectively. Results of the proposed DNACA.103] (rad/s). (b) D2 . Every section encompasses the result obtained in the previous section as the basis for the optimization process.4.712s ⎞⎛ s2 2×1.3 6.17s ⎞⎛ s2 2×1. The frequency band under consideration is [10-2.6 Issue 3.8 0.168 ⎝ 4.0413) TTA(0) ACCCAGAACT TTGGGA(0 ATTGGTGG ) (-0. After 800 generations of evolution (i. the resolution is 15 codons.4014) decimal values AGGGCCAA of parameters GCACAGTC A T GCGCAGAT (31.3896) TTTTTTAGACC TTATAAG (-0.7 Fitness value 0.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING 1 0. (a) D1 . The aim is to find out a simplified transfer function to faithfully mimic the system behavior in the frequency domain.103] rad/s.6465 20 50 45 45 8.e. 1]. 200]. 102] and [10-2.2 0.4 0.6 0.0282) TACTGG(0) TATTGGGAAG CTGCAGCA (-0. [10-2.6 1.129s ⎞ ⎛ s ⎞⎛ s +1⎟⎜ + +1⎟⎜ + +1⎟ 1.59s ⎞ ⎛ s ⎞⎛ s +1⎟⎜ + +1⎟⎜ + +1⎟ s⎜ 2 2 0.

8) −2 3 s(s+6. V. Molecular computation of solutions to combinational problems.8536 db (at 0. 0.8030 db 1.49s3 +21.4) (s+ 109. The performance index of the objective function J 2 is ignored in the following case.40)(s+ 18. A controller is obtained after 20 generations of evolution as Issue 3. CONCLUSIONS A DNA computing algorithm with new coding method based on EIIP of DNA codons is introduced.1 0. For the proposed DNACA.9802 ( s + 6.09 0. Capability for seeking the solutions of the multiple variable functions is confirmed by considering a class of De Jong’s functions as the testing object.964) 10 ~10 5 0. C.06 0. 10] .01 rad/sec) 1.32)(s+52.08 0.8. 1021-1024. 10 and 11. the weighting factor ε i is 0.38s2 +2. practice. the controller with lower order.4. Results of four step system ID.79) 10 ~1 (at 0. and prospects.199)(s+7. Adleman. Science. vol. the dashed line shows the identified system response within the following frequency intervals: (a) [10-2. 1]. 11. both the enzyme and virus rates are 0.2401 0. ID error (a) Order of num/den exp–((a)x0. 266.103](rad/s).6)(s+0.2. The EIIP is used instead of the traditional binary coding system.06 rad/sec) 0.81) (s+54. 201-229.7315 db (at 260. Fig.942×10−6 s −7.234s4 +10. Although this value is slightly worse than the one obtained through using Matlab. (c) [10-2.199)(s+7. Using Matlab to solve for the robust control problem defined by (10) gives CHopt = 4. the real line indicates the original system response.67)(s+9.05 0.1355(s+0.1) −2 2 s(s+2.5593)(s+192. Convergence of the H ∞ norm. pp.04 0. C.03 .02 .45)(s+62. Numerical experiments confirm its excellence for these problems.5593)(s+192. the population size is set as 50.768s +0. (d) [10-2.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING CE ( s ) = 3.01 rad/sec) 5 0. 2008 116 . M. 1998. The proposed DNACA has also been successfully extended to deal with the system ID and control design problem.1801 and the resulting g 2 = 0. The algorithm’s multiple mutation mechanism allows it to simplify the model structure simultaneously while generating the optimal parameter set during the evolution process.8536 db (at 0.1)(s+ 193.23 The final result for the robust stability index is g 2 = 0.973×10-7s4 +1.071×10−6 s2 +9. 10. Changes of the summation of denominator and numerator order. pp.2332(s+0.53) ⋅ −2 s(s+ 1.8)(s+194. mutation rate is 0. Maley.03 0. while sacrificing performance a bit.414)(s+189. REFERENCES [1] [2] L.2401 H∞ 0. Next. Table IV.1(b)) polynomials (b) Fig.97(s+5.29)(s+69. Results of ID in four frequency intervals Performance indices Estimation Model Max.2.07 2. the maximum and minimum bounds of the information fragments are [1. Convergent behavior of the controller’s structure and g 2 are shown in Figs.503) s + 17. is of more practical. crossover rate is 0. 6.1) 10 ~10 7 0.044×10−6 s3 +5.2975 0 2 4 6 8 10 12 14 16 18 20 Iterations Fig.236)(s+3.2332(s+0. Volume 2.1) −2 s(s+2. (b) [10-2. control design based on DNACA is examined.0346 2. 10].964) 10 ~10 12 0. DNA computation theory. 10. 1994. vol. 102]. Evolutionary Computation.081)(s+103.683×10−9 s5 +2.01 rad/sec) 2.

C. I. He received the M. D. John Wiley ＆ Sons. Grimble. Ghaffari. 39-49.S. Bumble. Hybr. 2007. Taiwan. 2008. pp. Optimization and Machine Learning. vol. Veljkovic.79-88. Manuel. Taiwan. vol. He is currently a Distinguished Professor in the Department of Electrical Engineering at National Chung Hsing University. Ching-Huei Huang was born in Zhanghua. 82-102. IEEE International Conference on Systems. Int. C. ROC. and prospects. Taiwan. Cosic. 2008. pp. Cambridge. and J.. Hashiyama. vol. 643-648. Postlethwaite. pp. Wood. PID Controllers. Hagglund. F. Z. 1998. Taoyuan. Botta. 1999. Astrom and B. Pavlovic. IEEE Trans. Mazinan and N. Issue 3. Ghorbani. D degree in the Department of Electrical Engineering at National Chung Hsing University. Chem. 3. 3. ROC. WSEAS Trans. Taichung. H. S. in 1978. He received his Ph. F.. Esté. and K.1-12. vol.INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] N. Veljkovic.261-265. Neural Networks.. Calderon. K.179-191. Hayes. Med. K. and robust control. J. Contr. Princeton. H. WSEAS International Conference on Automatic Control. He received the B. Robust Industrial Control: Optimal Design Approach for Polynomial Systems. Evolutionary programming made faster. from 1995 to 2003. T. Inf. Yao. Genetic Algorithms in Search. C. V. Manetti. Boston. Ann Arbor. Proc. 1994. in 1958. A. ROC. Astrom and T. in 2000. USA. 2007. Contr. Okuma. 2. Taichung. His research interests are in evolutionary algorithms. ROC. S. V. Syst. Proc. in 2003.and M. Taichung. Prentice Hall. pp.D degree from Institute of Electrical and Communications Engineering. Perovic. vol. Feng Chia University. in 2000 and 2003. 1996. 261-266. The Orchestral Analog of Molecular Biology. pp. Massa. Machine Learning Neural Networks. and O. Hardware/Software environment for process ID. Zhou. evolutionary algorithms. The relationship of the resonant recognition model to effects of low-intensity light on cell growth. S. Syst. John Wiley ＆ Sons. 319-323. pp. J. Lorenzo. Evo. Dietrich. Hung. pp. 2003. R. 2008 117 . Goldberg. DNA computing implementing genetic algorithms. Y. Contr. L. ROC. 503-508. Wittenmark. Chen. and G. Kluwer Academic Publishers. N. 1989. He received the Ph. E. W. Prentice Hall. 14. Vercammen. Tainan. Taiwan. ROC. ROC. ROC. in 1978. He was an Associate Professor and Professor in the Department of Automatic Control Engineering at Feng Chia University. K. 1995.441-53. Prague. Neuro-controller design using genetic optimized pole placement method. J. Addison-Wesley. degree in Department of Mechanical Engineering of Chung Cheng Institute of Technology. in 2007. J. M. WSEAS International Conference on Telecommunications and Informatics. Y. 1995. V. Glover. Luis. no. pp. ROC. DNA Computation theory. Doyle and K. Lam. Chem. Antipov. MIT Press. Debyser. Novel virtual screening protocol based on the combined use of molecular modeling and electron-ion interaction potential techniques to design HIV-1 integrase inhibitors. WSEAS Trans. pp. B. L. Intell. Tuning of the structure and parameters of a neural network using an improved genetic algorithm. Constrained model predictive control implementation for a heavy-duty gas turbine power plant. Altamiranda. Instrument Society of America. Z. MI. C. A. S. degree in automatic control engineering from Feng Chia University. degree in 2006. and U. An evolutionary adaptation strategy for dynamic systems ID. pp.. E. 2005. C. H. Biol. Taiwan. 14. Tintori. K. in 1991. 2008. and motor control. Imitation of Life: How Biology is Inspiring Computing. Rahnama. Taichung. Istanbul. 249-258. A. G. and M. vol. Peter. H.. 1995. vol. ROC. vol. D. Proc. practice. Inc. Rentea. X. 1999. V. Skogestad and I. H. 1999. Application of the EIIP/ISM bioinformatics concept in development of new drugs. and D. Chun-Liang Lin was born in Tainan. Model. Evolutionary Computation. vol. A. and robust control. Accelerated evolutionary computation using fitness estimation. Veljkovic. Adeli and S. MA. A. Int. 2005.S. Taiwan. robust controller design and hard real time implementation. 1975. intelligent control. Dr. 6. C. Maley. H. and he is currently working toward the Ph. Y. 3. 56. An Analysis of the Behavior of a Class of Genetic Adaptive Systems. J. Ono. 201-229. 2. Hanaki. Taiwan. E. 5. optimal control. J. Philadelphia College. Taiwan. 4. Comp. 2004. Genetic Algorithms and Fuzzy System. S. J. University of Michigan. Frank. Liu. Ling. Proc. M. pp. ROC. and E. H. His research interests are in bioinformatics.S.. Lemieux. Ibrahim. Volume 2. J. Veljkovic. Morles. Radiat. Vojisavljevic. DIMACS Workshop on Evolution as Computation. K. Hüther. Multivariable Feedback Control. S. Cedeno. L. WSEAS Trans. DNA computing approach to knowledge representation. He received the M. Jong. Multiple modeling and fuzzy predictive control of a tubular heat exchanger system. Sadati. J. Cur. His research interests include guidance and control..D. 1989. Taichung. pp. degree in aeronautical and astronautical engineering from the National Cheng Kung University. T. Tsuboi. and S.. Modeling and Simulation. Lin received the Distinguished Research Award from National Science Council of Taiwan. IEEE Tran. Syst. 507-516. Forbes. pp. Syst. Lin. 2003. and M. 2007. Robust and Optimal Control. pp. Taiwan. Horn-Yong Jan was born in Taoyuan. N. 2006. vol. Taiwan. J. Adaptive Control. Inc. J. Witvrouw.

- Dna Computing
- Race for the Double Helix
- paulAMS57-60-2008_DNA
- BCH3020 Questions and Answers
- dna computing
- Biomol Part 2 Dr. Dani
- Error-Prone Replication for Better or Worse
- Interesting Paper About Malondialdehyde
- Mecanica Estadistica de Desnaturalizacion de ADN
- 1 Quim Exam Model DNA the Molecule of Heredity TestBank_Basics.bio10a_chapter11
- tmpBED1.tmp
- The Science of DNA Fingerprinting
- lec_7
- A brief understanding of DNA synthesis.doc
- pnas00005-0140
- Biofunctionalized AlGaN/GaN high electron mobility transistor for DNA hybridization detection
- Lecture 6 April 5th 2009
- Text
- Amgen Inc. v. F. Hoffmann-LaRoche LTD et al - Document No. 852
- Article
- dnafingerprinting-110118120746-phpapp02
- 0046351a6fc3eb0074000000
- 3.3 & 7.1 DNA Structure HL Caitlin WORD
- Dna
- Chapt11 Lecture
- L1. Introduction and Historical Background Part I
- Stoneking MtDNA Review
- Type II Topoisomerase - Wikipedia, The Free Encyclopedia
- Advanced Chemistryprize2015
- William M. Shih, Joel D. Quispe and Gerald F. Joyce- A 1.7-kilobase single-stranded DNA that folds into a nanoscale octahedron
- bio-25

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd