太空生存太空生存太空生存 (40).pdf

Acta Astronautica 63(2008) Hua,Hong Liub,Chenliang Yangb,Enzhu HuaaDepartment of Bioengineering,Beijing University of Aeronautics and Astronautics,Beijing 100083,ChinabDepartment of Environmental Engineering,Beijing University of Aeronautics and Astronautics,Beijing 100083,ChinaReceived 21 May 2007;received in revised form 7 January 2008;accepted 20 February 2008Available online 7 April 2008AbstractAs a subsystem of the bioregenerative life support system(BLSS),light-algae bioreactor(LABR)has properties of highreaction rate,efficiently synthesizing microalgal biomass,absorbing CO2and releasing O2,so it is significant for BLSS toprovide food and maintain gas balance.In order to manipulate the LABR properly,it has been designed as a closed-loop controlsystem,and technology of Artificial Neural NetworkModel Predictive Control(ANN-MPC)is applied to design the controllerfor LABR in which green microalgae,Spirulina platensis is cultivated continuously.The conclusion is drawn by computersimulation that ANN-MPC controller can intelligently learn the complicated dynamic performances of LABR,and automatically,robustly and self-adaptively regulate the light intensity illuminating on the LABR,hence make the growth of microalgae inthe LABR be changed in line with the references,meanwhile provide appropriate damping to improve markedly the transientresponse performance of LABR.2008 Elsevier Ltd.All rights reserved.Keywords:BLSS;LABR;Control system;ANN-MPC1.IntroductionIn bioregenerative life support system(BLSS),thecore issue is of food supply and gas balance.BLSS ismainly made of higher plants and animals.If the systemis under emergencies;however,with their own restric-tions on the rate of metabolism,it is very difficult forrelying on them to restore the system to stability andequilibrium 1,2.For example,with respect to gases(mainly refer to CO2and O2)balance in BLSS,be-causeoftheabovereasons,higherplantsoranimalscan-not instantaneously influence upon the concentrationsCorresponding author.Fax:+861082339837.E-mail address:(H.Liu).0094-5765/$-see front matter 2008 Elsevier Ltd.All rights reserved.doi:10.1016/j.actaastro.2008.02.008of gases when some accidents happened to the system.As is well known,the system will undergo catastrophicdamage if the gases are in the state of imbalance fora long period.So it is necessary to choose the specieswhose reactions rate are faster and easier to controlas controllable and renewable tools for BLSS,mi-croalgae are the most promising candidate species bycontrast with high plants and animals 14 becausethey have characteristics of high growth rate,rapidlydecomposing the waste,absorbing CO2,releasing O2and effectively regulating gas composition.Here greenmicroalgae Spirulina platensis,has been selected to becultivated and used for BLSS,since besides precedingbenefits,S.platensis is a high nutrition food for crew,the protein,fat,carbohydrate and vitamin contents in1068D.Hu et al./Acta Astronautica 63(2008)10671075S.platensis are approximately 65%,4%,19%and6%,respectively,and can be digested and absorbedby humans more easily,furthermore its metabolism isextremely plastic and controllable 3,4,so finding amethodology for cultivating S.platensis will safeguardthe balance of material exchange as well as the equi-librium of the system 15.In the following context,the word microalgae refers in particular to Spirulinaplatensis.LABR is a complex system,due to it having non-linear and time-varying characteristics,and there areusually no fixed operating points in real application,soit is very difficult to employ conventional methods toachieve effective control for such system.Technologyof artificial intelligenceArtificial Neural Network-Model Predictive Control(ANN-MPC)is applied here,and combined with mathematical model and com-puter simulation to realize potent and robust controlfor LABR,and the result proved that ideal controllingeffect has been acquired.2.Materials and methods2.1.Light-algae bioreactorMicroalgae are photoautotrophs.Light undoubtedlyis the most important limiting factor for the photoau-totrophic growth of microalgae,and has a very great in-fluence upon its propagation,so light-algae bioreactor(LABR)has been designed as the form of microalgaecultured 5.LABR is a transparent device,in whichmicroalgae has been continuously cultivated by absorb-ing and decomposing various excretions or wastes,suchas urine and CO2in BLSS,and in the course of reac-tions microalgae will synthesize their own biomass,andsimultaneously release O2.Because the full benefits of microalgae for BLSShave been realized,the technological process had torender microalgal cultivation sufficiently controllable,so it is necessary to design the configuration of LABRas a closed-loop control system.When in the presenceof internal or external disturbances,LABR can elimi-nate their influences imposed upon BLSS and keep gasbalance 6.The configuration of LABR is shown inFig.1.Suppose that initially the concentration of CO2isequilibrium at the required level in BLSS atmosphere,if some disturbances suddenly occur,the concentrationsof CO2may deviate from its initial state,and simul-taneously LABR detects such change and executes toeliminate such deviation as following sequence.Firstly,the sensor1 measures the actual concentrationof CO2,the difference between current and desired con-centration of CO2can be considered a certain functionof microalgal biomass which can be deemed as a refer-ence input for LABR,Secondly,the sensor2 measuresthe microalgal biomass in the bioreactor as a feedbacksignal and compares it with the reference,the ANN-MPC controller will use the difference between themto adjust the light intensity,thus stimulate or inhibit thegrowth of microalgae to speed up or slow down in thebioreactor to produce an ideal output in agreement withthe reference,and such procedure corresponds to indi-rectly influence on concentrations of CO2and O2asneeded,because these two gases are tightly correlatedwith assimilatory quotient of microalgae,hence makesthe gases in BLSS rebalance at required level.2.2.Development of mathematical model for LABRLife systems generally have highly nonlinear,time-varying specifications and lots of non-specific interac-tions in life systems cannot be ascertained now,hencethey are essentially different from mechanical systems7,8.Herein kinetic equations just as Monod type orHenriMichaelisMenten type are utilized to describethe growth of microalgae.LABR was divided into threehierarchies,i.e.,gas phase,liquid phase and biologi-cal phase,the mathematical models for them have beendeveloped respectively,based on the theories of sys-tem dynamics,stoichiometry and experimental data aswell,and then multidimensionally coupled them intoa whole.In developing a reasonably simplified math-ematical model for LABR;however,some distributedparameters,nonlinearities and time-varying factors thatmaybepresentintheLABRhavebeenignoredbasedonapproximations and assumptions 9,10.Mathematicalmodel have been obtained in terms of first-order nonlin-ear ordinary differential equations to represent and de-pict the dynamics of the LABR 1116,the parametersin the mathematical model have been specified by ex-perimental data and quoting from the documents 11,and then the simulation model of LABR has been de-veloped by MatLab7.4/Simulink6.6 based on the math-ematical model to conduct the computer experiments.2.3.ANN-MPC for LABR as a closed-loop controlsystemThe dynamic features of biological systems and theircomponents are extremely complicated,in the course ofdevelopment of the mathematical model of system,weoften ignore these factors,but they act the actual role inthe real system.Furthermore,the state variables of theLABR will change over the large scope,i.e.,they willnot stay at a fixed operating point,so the linearizationprocedure is based on the expansion of the nonlinearD.Hu et al./Acta Astronautica 63(2008)106710751069Inflow-pipe ANN-MPCSensor1 CO2f(x)qivqov+-Sensor 2Microalgal Biomass Light Intensity Reference Vent-pipe Feedback Fig.1.Schematic diagram of LABR as a closed-loop control system.PlantNeural Network Model Learning AlgorithmymError+-Iy?Fig.2.Schematic diagram of LABR having ANN-MPC controller.model functions into a Taylor series about a operatingpoint and the retention of only the linear term cannot beapplied here,because higher order terms of the Taylorseries expansion cannot be neglected,therefore us-ing traditional controllers for mechanic systems(likeProportional plus Integral plus Derivative controller,PID controller)are often unable to achieve the desiredresults or entirely fail to control.At present,artificial neural networks have been ap-plied very successfully in the identification and controlof dynamic systems.The universal approximation capa-bilities of the multilayer perceptron make it be a popularchoice for modeling nonlinear systems and for imple-mentinggeneral-purposenonlinearcontrollers17.TheANN-MPC executing process mainly comprises twostages,which is represented in Fig.2.The first stage issystem identification,i.e.,ANN-MPC learns and trainsits neural network to represent the forward dynamics ofthe plantthe simulation model of LABR.The predic-tion error between the plant output and the neural net-work output is used as the neural network learning andtraining signals 18,19.The second stage is model predictive control,basedon the receding horizon technique.The neural networkmodel in ANN-MPC predicts the plant response overa specified time horizon.The predictions are used by anumerical optimization program to determine the con-trol signal that minimizes the following performancecriterion(Eq.(1)over the specified horizon:J=N2?N1y?(t+j)ym(t+j)2+?Nu?j=1u(t+j 1)u(t+j 2)2,(1)where N1,N2and Nudefine the horizons over whichthe tracking error and the control increments are evalu-ated.The u variable is the tentative control signal,y?isthe desired response and ymis the neural network modelresponse.The?value determines the contribution thatthe sum of the squares of the control increments has onthe performance index.The optimization determines thevalues of u that minimize J,and then the optimal I isinput to the plant.The ANN-MPC controller uses a neu-ral network model to predict future plant responses topotential control signals,an optimization algorithm thencomputes the control signals that optimize future plantperformance.This sequence completes the execution ofANN-MPC associated with any reference inputs.Fur-thermore as designing an ideal ANN-MPC controllerfor LABR,it is preferable that the controller provideappropriate damping,and transient response of LABRbe sufficiently fast as well as reasonably damped 6.1070D.Hu et al./Acta Astronautica 63(2008)10671075Fig.3.Overall mathematical model of LABR.Table 1Values of main parameters of the modelParametersValuesUnitSignificationV3.15LVolume of liquid phaseVG0.1LVolume of gas phaseQ60L/minTotal flow rate of gas in gas phaseF3.08L/minThe inflow and outflow rateCO2D_o0.01mg/LThe concentration of CO2outfluxO2D_o4.5mg/LThe concentration of O2outfluxXo0.26mg/LThe concentration of microalgal biomass outfluxNu_i345mg/LThe concentration of CO(NH2)2influxNu_o0.2mg/LThe concentration of CO(NH2)2outfluxNh_o0.15mg/LThe concentration of NH3.N outfluxNo_o0.22mg/LThe concentration of NO3.N outflux3.Results3.1.Mathematical model of LABR and computersimulation for its validationThe whole mathematical model of LABR is shownin Fig.3,and Table 1 lists the main parameter valuesthat appear in the model.The validity was verified byboth computer simulations and real experiments.Although the dynamic characteristics of LABR canbe studied by simulations under varied parameters,con-ditions,and their combinations,here we only select twotypical parameters,light intensity and temperature,asa demo to illustrate the system dynamic performances,D.Hu et al./Acta Astronautica 63(2008)106710751071020406080100 120 140 160 180 2000.10.20.30.40.50.60.70.80.911.1R2=0.94std=0.01R2=0.99std=0.07R2=0.99std=0.07R2=0.96std=0.01Time(h)Microalgal Biomass(g/L)Experimental DataSimulation Curve0501001502002503000.20.40.60.811.21.41.6Time(h)Microalgal Biomass(g/L)Experimental DataSimulation Curve0501001502002503000.20.40.60.811.21.41.6Time(h)Microalgal Biomass(g/L)Light Intensity:202mol/(m2*s)Experimental DataSimulation Curve020406080100 120 140 160 180 2000.20.30.40.50.60.70.80.9Time(h)Microalgal Biomass(g/L)Light Intensity:141 mol/(m2*s)Experimental DataSimulation CurveLight Intensity:101mol/(m2*s)Light Intensity:182mol/(m2*s)Fig.4.A comparison of microalgal biomass outputs under different light intensity from simulation and experiment.and carry out validation test.Comparisons of outputs ofmicroalgal biomass from simulations and experimentsare illustrated in Figs.4 and 5,where correlation coef-ficients and standard deviations are shown,too.Fromthe figures,we can see results of simulations seem tobe exactly agreeable to that of experiments,so modelis valid for its real counterpart.3.2.Simulation for closed-loop system of LABRhaving ANN-MPCThe growth of microalgae is distinguished by the factthat it can be limited by light as well as by substancesdissolved in the medium of substrate,so manipulationof either factor should be able to control the growth ofmicroalgae,nevertheless it is easy to maintain a con-stant and abundant supply of these components,just likeurine,CO2,minerals and other substances in BLSS,sothe cultural substrate is not regarded here as a limitingfactor.The compositions of substrate(Table 2),inflowand outflow rate are unchangeable in our experiments15.Sincecultureilluminationconsumesgreatamountsoflight,and ensures that the light factor is maximally ex-ploited,it is characteristic that any consideration of cul-tivation productivity for microalgae must be based firstand foremost on microalgae dependence on light inten-sity as a main limiting factor,and therefore LABR canbe regarded as a single-input and single-output(SISO)system.Experiments show that the utilization of red lightsof 620nm wave length,light intensity within the rangeof 50.300?mol m2 s1can significantly effectthe growth of microalgae,so red light-emitting diode(LED)used as the light source in our experiments,andANN-MPC controls the intensity of light by increasingor reducing their numbers.In order to select the best one among candidate ANN-MPCs having different structures and parameters,theANN-MPC controllers are trained and simulated offline1072D.Hu et al./Acta Astronautica 63(2008)106710750204060801001201400.20.30.40.50.60.70.80.91R2=0.95std=0.02R2=0.92std=0.05R2=0.92std=0.02R2=0.93std=0.02Time(h)Microalgal Biomass(g/L)Temperature:28Experimental DataSimulation Curve0204060801001201400.20.40.60.811.21.41.6Time(h)Microalgal Biomass(g/L)Temperature:30Experimental DataSimulation Curve0204060801001201400.20.30.40.50.60.70.80.91Time(h)Microalgal Biomass(g/L)Temperature:32Experimental DataSimulation Curve0204060801001201400.20.30.40.50.60.70.80.91Time(h