APP下载

Improved quantum bacterial foraging algorithm for tuning parameters of fractional-order PID controller

2018-03-07LIULuSHANLiangDAIYueweiLIUChenglinandQIZhidong

LIU Lu,SHAN Liang,*,DAI Yuewei,LIU Chenglin,and QI Zhidong

1.School of Automation,Nanjing University of Science and Technology,Nanjing 210094,China;

2.Key Laboratory of Advanced Process Control for Light Industry(Ministry of Education),Institute of Automation,Jiangnan University,Wuxi 214122,China

1.Introduction

In the past decades,process control has been developed rapidly in the industrial field,and the proportion integration differentiation (PID) controller is the most typical one.Because of its simple structure and strong robustness,the traditional PID controller is widely used in the field of industrial process control.However,it is difficult to achieve the expected performance because of its long adjustment time,large overshoot,and so on.Three parametersKp,KiandKdare adjusted to meet the preset performance functions which are usually composed of adjusting time,overshoot,and steady-state error[1].

The parameter settings of the PID controller are the key of the PID controller design because they directly affect the control performance of the system.In the past few years,the heuristic algorithms are applied to adjust the PID controller.The first proposed method is the Z-N method.In many areas of industrial control,it is difficult to find the optimal value using the Z-N method.After that,many experts and scholars focus on the research of the intelligent algorithms,such as the neural network and its mutation algorithm[2,3],but these intelligent algorithms require high control process and control law.Thus,the biological optimization algorithms,such as particle swarm optimization(PSO)algorithm,genetic algorithm(GA),and so on,are used to self-tune the PID controller parameters[4,5].

Among the bionic algorithms, the bacterial foraging optimization (BFO) algorithmis a typical one which was proposed by Passino in 2002 [6]. Although the BFO algorithm has achieved some success in application, the algorithm has the disadvantages of slow convergence and easy to fall into a local optimum. After that, many improved algorithms have been developed [7 – 9]. Majhi et al. presented an adaptive bacterial foraging algorithm to predict various stock indices [7]. Dasgupta et al. analyzed the chemotaxis operation of the algorithm, and presented an adaptive computational chemotaxis [8]. In [9], a new BFO algorithm was proposed based on the improved elimination-dispersal operation. These algorithms still have the defect of falling into a local optimum. In order to solve this defect, Huang et al. introduced the quantum mechanism into BFO and proposed a bacterial foraging algorithm with quantumbehavior [10]. In the quantum BFO (QBFO) algorithm,multi-qubit was used to encode the chromosome, and the diversity of the population was increased by the superposition of the gene, which makes the algorithm have better global search ability than the classical BFO algorithm.Subsequently, the QBFO algorithmis used in the optimization of practical problems [11 – 14]. In [11], the traveling salesman problem (TSP) was solved successfully by QBFO. In order to solve the discrete optimization problem,a similar algorithm was proposed in [12]. Based on [12],the performance of QBFO was compared with that of PSO and GA in [13], and the QBFO algorithm was successfully applied to spectrum sensing. In [14], the convergence of QBFO was proved and the QBFO algorithm was applied to adjust the parameters of cognitive radio networks.

In the QBFO algorithm,the value of the quantum rotation angle is given through the look-up table method,and the rotation angle obtained in this way is discrete and cannot fully reflect the basic situation of the solution space.Besides,the rotation angle is fixed,which limits the increase in the diversity of bacterial population.In this paper,an improved QBFO(IQBFO)algorithm is proposed,which can make the rotation angle adaptively and continuously updated,increase the diversity of the population and improve the global search ability.The tests with benchmark functions verify the effectiveness of the proposed algorithm,and the tuning results of PID parameters show that the proposed algorithm has a good ability to tune the PID parameters in the fractional-order servo system.

The rest of the paper is organized as follows.The improved BFO algorithm is presented in Section 2.In Section 3,the convergence of the proposed algorithm is analyzed.In Section 4,the parameters of IQBFO are analyzed,and some benchmark functions are adopted to evaluate the performance of IQBFO,which verifies the correctness of the improved method.In Section 5,the proposed algorithm is used to optimize the PID parameters of a typical fractional-order servo control system.Simulation results show that the improved method has better convergence property and stability.This paper is summarized and concluded in Section 6.

2.The improved BFO algorithm

QBFO is a new hybrid optimization algorithm, which combines the quantum evolutionary algorithm with the bacterial foraging algorithm.The features of QBFO are as follows:

(i)The chromosome encoded by binary represents bacterial population.

(ii)Gene is used to represent the location information of bacteria.

(iii)The chromosome is in superposition state or entangled state to enrich the diversity of bacterial population.

(iv)The quantum rotation gate is used for chemotaxis to update the state of chromosome.

2.1 Quantum bit and quantum rotation gate

Quantum bit is the smallest unit in quantum theory,which can be represented by a column matrix(α,β)Tor the following form

where α and β represent the probability amplitude of quantum states|0?and|1?,respectively.And α and β satisfy the normalization condition|α|2+|β|2=1.Equation(1)shows that the quantum bit is not only in state|0?and|1?,but also in superposition between the two. If a chromosome consists of p genes,the chromosome can describe 2pstates at the same time.For example,the probability amplitude of a three-qubit quantum is as follows:

The corresponding state of the chromosome can be described as

If the quantum chromosome is applied to represent the state in(3),it will require only one chromosome.However,the traditional chromosome needs at least four chromosomes,which further indicates that the superposition of quantum states in the chromosomecan enrich the diversity of the population.

In the QBFO algorithm,the state of the chromosome is updated with a single quantum rotation gate,which is expressed as follows:

where θ represents the radian rotated by U from state|0?to state|1?.The quantum rotation gate must meet the condition U+U=I,that is,preserving unitarity.

2.2 Design for QBFO algorithm

The main components of QBFO include parameter initialization,chemotaxis,reproduction and elimination-dispersal operations.In the chemotaxis operation,the core of QBFO is the updated process which is operated by the quantum gate.The aim of chemotaxis operation is to change the probability amplitude of each qubit, thereby the location of the bacteria can be updated. The specific implementation process is as follows.

(i)Initialization operation:initialize the following parameters:Siz,Nc,Nre,Ned,Ped,Q(t)and θ0.The initial population Q(t)is measured,and results of Q(t)are evaluated to obtain the initial local optimum and global optimum of the bacterial population.The system parameters are defined as follows:Siz denotes the population size;Ncdenotes the number of chemotaxis;Nredenotes the number of reproduction;Neddenotes the number of elimination-dispersal;Peddenotes the probability of elimination-dispersal;Q(t)denotes the initial population;θ0denotes the initial rotation angle.

(ii)Chemotaxis operation:the population Q(t)is updated by the quantum rotation gate,and the new population is evaluated to obtain the local optimum and global optimum.

(iii)Reproduction operation:the fitness values are in de-scending order,and then we reproduce the fitness values of the better half and remove that of the poorer half.

(iv)Elimination-dispersal operation:generate a random probability Prand,and compare it with the probability of elimination-dispersal.If Prand<Ped,the chemotaxis ope-ration will be executed;otherwise,it will keep the original information of the chromosome unchanged.

(v)When the end condition of the algorithm is satisfied,the optimal fitness value and the corresponding individual position information are output,and the algorithm is finished;otherwise,(iii)is executed.

2.3 The proposed IQBFO algorithm

In the classical QBFO algorithm,the quantum rotation angle is obtained through the look-up table,which cannot reflect the basic situation of solution space and restrict the increase of population diversity.Therefore,this paper proposes an improved quantum rotation angle which can make the quantum rotation angle adaptively and continuously updated.The modified operator is introduced to avoid the occurrence of invariant solutions in the process of updating the quantum gate.The modified operator can not only enrich the diversity of population,but also enhance the global search ability of the algorithm.

2.3.1 The modified probability of optimal rotation angle

Suppose that the population size,the chromosome length and the population are expressed by Siz,N,and Q(t),respectively,then the coding form of the k th individual in the population Q(t)is shown below:

In the QBFO algorithm,the initial values of the 2SizN probability amplitudes are set aswhich can not only ensure that all possible superposition states achieve the same probability,but also guarantee the unitary of the quantum rotation gate.Due to the same initialization value of each probability amplitude,the angles θiand θbare the same,which leads to the current optimal rotation angle being the same as the rotation angle of each qubit.θidenotes the angle of a quantum bit for the current bacterium and θbdenotes the angle of a quantum bit for the optimal bacterium.

The above situation is easy to make the algorithm produce invariant solutions in the early evolution stage,and leads to local convergence or even failure.In view of this,a new concept called modified probability of optimal rotation angle is proposed,as shown below:

where μ and ε are random numbers in the interval(0,1).

The purpose of the proposed optimal rotation angle is to prevent the generation of invariant solutions,which means that there is no need for a substantial change of the optimal rotation angle.According to(6),we can see that Pθshould be larger than ε.Considering that ε is a random number between 0 and 1,the value of Pθis set as 0.75 as a matter of experience.

2.3.2 The modified operator of probability amplitude

In order to further increase the population diversity and improve the global search ability of the algorithm,the modified operator of probability amplitude is applied to improve the probability amplitude[15].The modified operator of probability amplitude is defined as follows:

where γ is a relatively small positive number,andis the probability amplitude updated by the quantum gate.

2.3.3 The improved quantum rotation angle

The definition of the improved quantum rotation angle is as follows:

where θirepresents the angle of a quantum bit for the current bacterium;xiis a binary solution measured by a quantum bit of the current bacterium;θ0is the initial value of rotation angle;gen is the number of iterations;Maxgen is the maximum number of iterations and sgn(·)is the sign function.

In(9),M1and M2are used to control the direction of the rotation angle,and η and θ0are used to control the size of the rotation angle.Due to the continuous rotation angle obtained from(9),the improved QBFO algorithm based on this rotation angle can search the solution space in a comprehensive way and find the global optimal solution quickly.

3.Convergence analysis of the IQBFO algorithm

Theorem 1The IQBFO algorithm is a finite and homogeneous Markov chain.

ProofIn the IQBFO algorithm,the set of Q(t)=is a finite set.For any two states τiand τjin state space S,the probability transferred from τito τjis only related to τiand has nothing to do with the time k.Thus,the IQBFO algorithm is a finite and homogeneous Markov chain.

In the state space S,the state probability is affected by the modified probability of the optimal rotation angle,the modified operator of probability amplitude,reproduction and elimination-dispersal operations.The corresponding probability matrix are R,A,C and E,thus the transfer probability matrix of Markov chain of IQBFO can be described as F=RACE.

Theorem 2The transfer probability matrix R,A and E are both random and strictly positive,and the matrix C is random.

ProofTaking the transfer probability matrix R of rotation angle modified operation as an example,the state τiis mapped to the state τjaccording to a certain probability pτ,and the transfer probability is defined as follows:

Similarly,we can prove that the matrix A and E are random and strictly positive,and the matrix C is random.

Lemma 1[16]If A and B are n×n random matrices,then AB is also an n×n random matrix.

Theorem 3The transfer probability matrix F in Markov chain of IQBFO is strictly positive.

Lemma 4Suppose that U is a transfer probability matrix of regular homogeneous Markov chain,then

(ii)There exists a unique probability vector that satisfies ρTU= ρT.

Theorem 4IQBFO is globally convergent.

ProofSuppose that the state of transfer probability ma-trix F satisfies the following properties:

(i)f1is the global optimal solution;

(ii)fi(i=2,3,...,N-1)is the global suboptimal solution;

(iii)fNis the global worst solution.

Then the matrix F can be expressed in the form of a lower triangular matrix,as follows:

By Lemma 4,the row vector ofis a probability vector,and each element is greater than 0.From(14),we draw a conclusion that the state can eventually reach f1wherever the state starts,which clearly implies Theorem4.

4.Parameter selection and performance evaluation

4.1 Selection of θ0

In the IQBFO algorithm,the initial rotation angle θ0has a direct impact on the convergence rate and search efficiency.If θ0is too small,the algorithm will have a slow convergence rate and waste a lot of time;if θ0is too large,the amplitude of searching solution space is too large,which is easy to cause the loss of population diversity.Therefore,the selection of θ0is studied in this paper,which avoids the quantum rotation angle being too large or too small.

In order to select a suitable θ0,the classical function named Sphere is used to test the IQBFO algorithm,which is defined as

In the test process,the order of Sphere is set as 2,and the search range is set as[-100,100].The function Sphere takes the global minimum at x=(0,0).In the IQBFO algorithm,the parameters are set as follows:Siz=40,Nc=25,Nre=5,Ned=4,Ped=0.25.Run the algorithm 50 times and the performance of different values of θ0is listed in Table 1.

Table 1 Test results of θ0with different initial rotation angles

As can be seen from Table 1,the IQBFO algorithm can get the same optimal value for different θ0.When θ0is 0.3π,the IQBFO algorithm can get the minimum value regardless of the average value or the worst value,which indicates that the algorithm has strong search ability.Furthermore,the variance of test results is minimum,which shows that the algorithm has strong stability.The simulation results show that the IQBFO algorithm has better searching ability when θ0is 0.3π.In order to evaluate the performance of the proposed IQBFO algorithm,the performance IQBFO is compared with that of the QBFO algorithm and the quantum genetic algorithm(QGA)through seven benchmark functions in Table 2.Since these three algorithms are binary coded,the classical particle swarm optimization(CPSO)[18]algorithm is introduced to highlight the optimization ability of the IQBFO algorithm.

4.2 Performance evaluation

In Table2,f1-f4are multi modal functions,and f5-f7are unimodal functions.The seven benchmark functions are used to evaluate the performance of the proposed IQBFO algorithm.During the evaluation process,the initial rotation angle θ0of IQBFO is 0.3π.For the QBFO algorithm,the values of Nc,Nre,Nedand Pedare set as that in IQBFO.For the QGA algorithm,the parameters are set as:the maximum number of iterations MaxI=500,crossover probability Pcross=0.7 and mutation probability Pm=0.15.For the CPSO algorithm,the parameters are set as:the population size Siz=40,the maximum iteration number MaxI=500,the inertia factor w=0.8,the learning factor c1=1.5 and c2=1.5.

The test difficulty increases with the increase of the di-mension.In order to better reflect the effectiveness of the proposed algorithm,the dimension in the test is set as 30.Running 50 times,the performances of the algorithm are analyzed from two aspects:the average value of the fitness function and the number of iterations Iter.The test results are listed in Table 3.

Table 2 Basic information of benchmark functions

Table 3 Performance comparisons of IQBFO,QBFO,QGA and CPSO algorithms

As can be seen from Table 3,compared with QBFO and QGA algorithms,the proposed IQBFO algorithm can get the minimum value for each benchmark function,which shows that the algorithm has better convergence precision.In addition,for most benchmark functions,the IQBFO algorithm has the minimum number of iterations,which indicates that the proposed algorithm has a fast convergence rate.Compared with the CPSO algorithm,the convergence rate of IQBFO is relatively slow,but it has good convergence precision.The reason for this phenomenon can be given:the CPSO algorithm is easy to fall into a local optimum,which leads to the fast convergence of the CPSO algorithm.In conclusion,the IQBFO algorithm can converge quickly with a relatively small number of iterations whether they are unimodal functions or multi modal functions.

In order to present the performance of the proposed algorithm more directly,taking three multi modal functions f1-f3as an example,the curves of the adaptive function value with the number of iterations are drawn,as shown in Fig.1–Fig.3.

Fig.1 Change curve of adaptive value of Griewank with the number of iterations

Fig.2 Change curve of adaptive value of Ackley with the number of iterations

Fig.3 Change curve of adaptive value of Rastrigrin with the number of iterations

Fig.1 and Fig.2 show that for f2,the convergence rate of IQBFO is slower than that of CPSO at the early stage of iteration,but in the latter stage the IQBFO can converge to the global optimal solution rapidly.Fig.3 shows that for f1and f3,the proposed algorithm has the fastest speed in the whole optimization phase.

Although the proposed IQBFO algorithm increases the complexity,it can obtain better accuracy and robustness.If the algorithmis simplified,such as reducing the number of chemotaxis,the performance of the system will decrease due to the reduction of the iteration number.

5.Tuning of PID parameters of fractional-order servo system based on IQBFO

In order to verify the effectiveness of the proposed IQBFO algorithm,we apply the proposed algorithm to tune the PID controller of a two-loop position servo control system.Compared with the integer-order PID controller,the fractional-order PID controller increases the complexity of the system with better flexibility and robustness,although it has added two additional parameters.Due to the addition of two parameters,it is difficult to identify the parameters of the target model by using the ordinary identification algorithm,and it is easy to fall into a local optimum.Therefore,the proposed IQBFO algorithm is adopted to identify the parameters of the fractional-order controller.The structure of the position servo system is shown in Fig.4.

Fig.4 Structure of position servo system

where

According to(18),the whole system can be simplified,as shown in Fig.5.

Fig.5 simplified structure of position servo system

5.1 Design of IQBFO-PID controller

The fractional-order PID controller was first proposed by professor Podlubny[19],and the transfer function is defined as follows:

where Kp,Kiand Kdrepresent the proportion,integral and differential gains of the fractional-order PID controller,respectively.λ and μ are integral and differential orders,respectively.The design of the IQBFO-PID controller is shown in Fig.6.R(s)represents the input signal,E(s)represents the error,Y(s)represents the output signal,and GC(s)represents the control object.

In this paper,the fitness function is shown as follows:

where σ,tr,tsand essdenote overshoot,rise time,setting time and steady state error,respectively.

Fig.6 IQBFO-PID controller

The typical input signals of the position servo system include step,constant and sinusoidal signals.There are no σ,trand tsin the constant velocity and sinusoidal output signals,thus the classical evaluation function called integral of time multiplied by the absolute value of error(ITAE)[20]is introduced to test the constant velocity and the sinusoidal signals,which is shown as follows:

5.2 Simulation experiment and result analysis

The parameters of the servo system are listed in Table 4.

Table 4 Parameters setting of servo system

In the test,each simulation time is set as 5 s,and the reasonable ranges of P,I,D,λ and μ are set as[0,2].Three typical input signals of the position servo system are as follows:

According to the tuning method of PID controller of the actual servo system,the test steps are listed as follows.

Firstly,the parameters of the step input signal are adjusted,and the optimal values of the five parameters of Kp,Ki,Kd,λ and μ are obtained.Then,these five parameters are used to adjust the constant velocity input signal,and the algorithm is applied to tune the feed-forward coefficient Kfe.Finally,these six parameters are used to adjust the sinusoidal input signal.

Running 30 times,the tuning results are shown in Table 5.The corresponding performance criterion of step response is shown in Table 6.The corresponding output curves of three typical signals are shown in Fig.7–Fig.9.

Table 5 Results of parameter tuning

Table 6 Performance criterion of step response

Fig.7 Curve of 70°step response

Fig.8 Error curve of 30°/s constant velocity signal

Fig.7–Fig.9 and Table 6 show that the proposed IQBFO algorithm can obtain a faster convergence rate and smaller convergence accuracy compared with other two kinds of quantum algorithms.At the same time,it can reflect the characteristics of the CPSO algorithm:it has a faster convergence rate in the early stage and it is easy to fall into a local optimum value in the later stage,which will affect the final convergence precision and steady-state time.In order to further highlight the performance of the proposed algorithm,a nonlinear time-delay process is introduced to the position servo system,which represents some uncertain nonlinear factors in the servo system.Parameter settings remain unchanged and the time delay is set as 100 ms.The proposed algorithm is tested by the step signal and the corresponding output curve is in Fig.10.

From Fig. 10, we can see that the proposed IQBFO algorithm still outperforms other algorithms.Besides,it is clear to see that all algorithms can converge to a corresponding value with the introduction of the time-delay process.The reason lies in that the fractional-order controller is not sensitive to its control parameter and the parameters of the controlled object in a certain range,which can enhance the anti-jamming performance of the system.

6.Conclusions

In this paper,an IQBFO algorithm is proposed,which can not only make the quantum rotation angle adaptively and continuously updated,but also enhance the global search ability of the algorithm.Furthermore,the parameter identification of the fractional-order controller is realized by using the proposed algorithm.In the initialization process,the modified probability of the optimal rotation angle is adopted to avoid the existence of invariant solutions.The modified operator of probability amplitude is introduced to enrich the diversity of population.The performance of the IQBFO algorithm is tested by several benchmark functions and compared with that of QBFO,QGA and CPSO algorithms.The test results show that the IQBFO algorithm can converge quickly with small convergence precision.The parameter tuning of the PID controller for the fractional order servo system shows that the proposed algorithm can effectively adjust the parameters of the PID controller.

In order to further verify the effectiveness of the proposed algorithm in the actual system,in the future work,the algorithm will be verified in the actual system which takes the industrial personal computer as the controller.

[1]HUANG G,WU D,YANG W,et al.Self-tuning of PID parameters based on the modified particle swarm optimization.Proc.of the 8th IEEE World Congress on Intelligent Control and Automation,2010:5311–5314.

[2]KANG J,MENG W,ABRAHAM A,et al.An adaptive PID neural network for complex nonlinear system control.Neuro-computing,2014,135(8):79–85.

[3]ATTARAN S M,YUSOF R,SELAMAT H.A novel optimization algorithm based on epsilon constraint-RBF neural network for tuning PID controller in decoupled HVAC system.Applied Thermal Engineering,2016,99:613–624.

[4]LIU X.Optimization design on fractional order PID controller based on adaptive particle swarm optimization algorithm.Nonlinear Dynamics,2015,84(1):379–386.

[5]NEATH M J,SWAIN A K,MADAWALA U K,et al.An optimal PID controller for a bidirectional inductive power transfer system using multi objective genetic algorithm.IEEE Trans.on Power Electronics,2014,29(3):1523–1531.

[6]PASSINO K M.Biomimicry of bacterial foraging for distributed optimization and control.IEEE Control Systems,2002,22(3):52–67.

[7]MAJHI R,PANDA G,MAJHI B,et al.efficient prediction of stock market in-dices using adaptive bacterial foraging optimization(ABFO)and BFO based techniques.Expert Systems with Applications,2009,36(6):1–8.

[8]DASGUPTA S,DAS S,ABRAHAM A,et al.Adaptive computational chemotaxis in bacterial foraging optimization:an analysis.IEEE Trans.on Evolutionary Computation,2009,13(4):919–941.

[9]DEVI S,GEETHANJALI M.Application of modified bacterial foraging optimization algorithm for optimal placement and sizing of distributed generation.Expert Systems with Applications,2014,41(6):2772–2781.

[10]HUANG S,ZHAO G L.A comparison between quantum inspired bacterial foraging algorithm and GA-like algorithm for global optimization.International Journal of Computational Intelligence&Applications,2012,11(3):1969–2013.

[11]LI F,ZHANG Y T,WU J L,et al.Quantum bacterial foraging optimization algorithm.Proc.of the IEEE Congress on Evolutionary Computation,2014:1265–1272.

[12]CAO J L,GAO H Y.A quantum-inspired bacterial swarming optimization algorithm for discrete optimization problems.Proc.of International Conference on Advances in Swarm Intelligence,2012:29–36.

[13]DIAMANTOPOULOS A,LEREUN C,RASUL F,et al.A quantum bacterial foraging optimization algorithm and its application in spectrum sensing.International Journal of Modelling identification&Control,2013,18(3):234–242.

[14]GAO H Y,LI C W.Quantum-inspired bacterial foraging algorithm for parameter adjustment in green cognitive radio.Journal of Systems Engineering&Electronics,2015,26(5):897–907.

[15]HAN K H,KIM J H.Quantum-inspired evolutionary algorithms with a new termination criterion,Hεgate,and two phase scheme.IEEE Trans.on Evolutionary Computation,2004,8(2):156–169.

[16]FANG Z B,LIAO B Q.Random process.Hefei:University of Science&Technology China Press,1994.

[17]RUDOLPH G.Convergence analysis of canonical genetic algorithms.IEEE Trans.on Neural Networks,1994,5(1):96–101.

[18]KENNEDY J,EBERHART R.Particle swarm optimization.Proc.of the IEEE International Conference on Neural Networks,1995:1942–1948.

[19]PODLUBNY I.Fractional-order systems andPIλDμcontrollers.IEEE Trans.on Automatic Control,1999,44(1):208–214.

[20]ABD-ELAZIM S M,ALI E S.A hybrid particle swarm optimization and bacterial foraging for power system stability enhancement.Complexity,2015,21(2):245–255.


登录APP查看全文