Research On The Separation Model For HPLC-DAD Data Set And Its Solution By Optimization Algorithm

The task for separating High Performance Liquid Chromatography-Diode Array Detector (HPLC-DAD) data set is to extract spectra, which can be used to identify what the compound is, and the chromatogram peaks, which can be used to calculate the amount of the compound, for all compounds contained in the data set from instrument. This work is the foundation and prerequisite for related researches. However, there is big limitation existing in current separation methods, especially when the HPLC-DAD data set is complex with severe overlap among different compounds. In order to improve the accuracy and effectiveness for HPLC-DAD data set separation, to reduce the requirement of the instrument, to enhance the flexibility and adaptability of the separation methods, this paper proposes a new separation method based on the reference curve parameter assessment and optimization, which is successfully applied to the separation of HPLC-DAD data set.Firstly, this paper constructs a new framework for HPLC-DAD data set separation. This framework begins with the reference curve for chromatogram peaks. Several kinds of reference curves are analyzed, selected and constructed to approximate chromatogram peaks in the data set. Then, a large group of parameters are initialized for further calculation. Following, a parameter assessment (PA) model is buided to calculate the quality of these parameters. In order to solve the PA model builed, some algorithms based Evolutionary Algorithm and Swarm Intelligent method are proposed, which can compute all the optimal parameters simultaneously. Finally, an estimator is proposed to calculate the spectra based on the calculated chromatogram peaks and the HPLC-DAD data set.Secondly, the constructure of the PA model is analyzed and discussed based on the framework builded above. The function of the PA model is to give a value for the inputted paramters, which indicates the error between the reference curve constructed by the inputted parameter and a certain chromatogram peak contained in the HPLC-DAD data set. The major work for the PA model is to design the curve generated (CA) model contained in the PA model, whose function is to generate a curve to compare with the inputted curve. This paper designs two CA models including Independent Component Analysis constrained by Reference curve (ICARC) model and the Reference Curve Measurment (RCM) model. ICARC model introduces the shape constraint for the chromatogram peak into the objective function of the Independent Component Analysis (ICA) model. Through eight groups of performance tests, the ICARC model fulfills the requirements of the design. Further research finds that it is not necessary for the chromatogram peaks contained in the HPLC-DAD data set to be independent or incorrelated. So, this paper proposes the RCM model, which removes the requirement of independence among output curves from the ICARC model. The RCM model is simpler and quicker than the ICARC model. Through tests, the RCM model fulfills the requirements and has better performance that the ICARC model.Thirdly, this paper proposes several optimization algorithms based on Evolutionary Algorithm and Swarm Intelligent method to solve the PA model builded above including Multi-area Genetic Algorithm (MGA), Mutli-target Particle Swarm Optimization (MPSO), Deep Search Multi-target Particle Swarm Optimization (DSMPSO), Multi-group Particle Swarm Optimization (MGPSO) and Parallel Nonlinear Least Square (PNLS). Based on the principle of Genetic Algorithm (GA), the MGA introduces the mutli-area optimizing method, which calculates multi optimal solution simultaneously. The MPSO, DSMPSO and MGPSO algorithms are proposed based on the Particle Swarm Optimization (PSO) algorithm by introducing local best particles for all parameters. By introducing accelerator factors for every parameter, PNLS method is proposed based on the Nonlinear Least Square (NLS) algorithm. Through simulations and experiment on HPLC-DAD data set, the methods proposed in this paper can extract chromatogram peaks and spectra for all compounds contained in the complex HPLC-DAD data set without know the number of the compounds in advance even when noise exists.Finnaly, a conclusion of this research is drawn and the future works are listed.