Application Of BP Neural Network Based Genetic Algorithm In Multi-Objective Optimizing The Drugs Component

There are many multi-factors and multi-levels of complex optimization problem in the fields of pharmaceutical research, for example, determining the optimal extraction conditions of drugs effective ingredients and drawing up the amount of ratio of the various raw materials of different properties and different prescriptions in drug tablets. Due to the influence of factors often more than one, so the design scheme is multi-factorial, at present, the more commonly used are orthogonal design, uniform design, etc. Then use multiple regression to fitting these experimental data (continuum data), but the drugs'research data are very complex and limited, which makes some problems in modeling process. Linear or nonlinear programming are used during the optimization process, commonly used is to weight multiple target into single objective optimization, which is bigger subjectivity and the weighting coefficients are uncertain. A target is often optimality, while in another goal may be poor, it cannot guarantee all the goals are existing optimal solution, and can provide unique solution, which is the problem has been concerned in pharmaceutical research.In recent years, the artificial neural network with its unique simulation, learning, and classification ability has been extensively applied in the preparation design, structure-activity relationship, in vivo and related research, drug kinetics and clinical pharmacy, etc. It has strong ability of nonlinear, self-organizing, adaptive, self-learning ability, high fault-tolerance and stabilization, etc, and is the emollient tool for data modeling in pharmaceutical research. The genetic algorithm shows a lot of advantages in multi-objective optimization, especially the NSGA and NSGA-Ⅱalgorithm are more obvious.Firstly, this topic introduces the methods of BP neural network and NSGA, then the two methods are connected, apply the BP neural network to train the data of drug extract, and multi-objective optimization the network output by non-inferior classification genetic algorithm by the MATLAB cabin toolbox, which provide feasible solution to multi-objective optimization in pharmaceutical research.Part 1: The summary of BP neural network and using BP neural network modeling issues that need attention, such as: the determination of network structure (hidden layers, transfer function, number of nodes in each layer), the choice of the training function, as well as the shortcomings of BP neural network, and the application of combining with the Genetic Algorithms. Part 2: The theory of genetic algorithm and the multi-objective optimization. The concept of Pareto solutions, the characteristics of multi-objective optimization, as well as the development of multi-objective genetic algorithm. NSGA algorithm and NSGA-II algorithm has more obvious advantages, it is applied to different areas, which are more value in many multi-objective optimization algorithm. This chapter focuses on the improvement of NSGA and NSGA-II algorithm principle.Part 3: Application of BP Neural network Based on Genetic Algorithm in Multi-objective optimization of the Extraction Medicine conditions.Using the Cortex Fraxini extraction material of literature, the uniform design data was trained by multiple regression, BP neural network and GA-BP, results show that multiple regression can't explain the relationship between the factors and the evaluations, the fitting correlation coefficient of BP and GA-BP are 0.9872 and 0.9940, but the speed of BP is slow and the results is instable. So in this paper, the GA-BP is used to modeling.The sum of squares of error (SSE) and the fitness value are stable in the 30th generation, and the training error curve achieve the target of 0.001 in 14th generation. Using Elitist Nondominated Sorting Genetic Algorithm to optimize the output of BP network, and the results showed that the temperature is 81℃, ethanol volume fraction is 60%, the liquid-solid ratio is 10, the extracting time is 46min, the part A of Cortex Fraxini extraction is 9.6130 mg/g, the part B of Cortex Fraxini extraction is 2.2032 mg/g. This is a solution by compromising the two goals, which is t better than anyone of experimental conditions, and the solutions searched by NSGA-Ⅱare ideal and the other solution programs can provide a greater choice for workers.The effect Elitist NSGA-Ⅱis ideal. The stem and leaf diagram shows that the maximum value of the part A of Cortex Fraxini is 10.7mg/g, the minimum value is 8.4mg/g, the Pareto solutions are mostly distributed nearby the value of 8.4 mg/g; the maximum amount of the part B of Cortex Fraxini is 2.6mg/g, the minimum value is 1.9mg/g, Pareto solutions are mostly distributed nearby 2.5 mg/g.Using the microspheres pharmaceutical technology of Indapamide, thinking the yielding, drug loadings and the encapsulation efficiency as the inspection index, select the appropriate technology conditions and polymer materials preparation to prepare the microspheres pharmaceutical which has obvious pH respectively and Indapamide as unique as slow-releasing microspheres. Using multiple regression,BP and GA-BP to train the orthogonal design data for simulation, the fitting model are both good. Then using Elitist NSGA-Ⅱto multi-objective optimize the model fitted by multiple regression and the output of network, and the rang of the three objective are 70.9%～84.2%,15.3%～18.8%,87.1%～96.4% by multiple regression; and the rang of the three objective are 69.4%～81.8%,14.8%～20.0%,81.0%～97.8% by GA-BP. The results show the model fitted by GA-BP is good and the Pareto solutions are consistant.The exploratory research shows that the BP neural network and Elitist NSGA-Ⅱcan obtain the reasonable Pareto solutions, which provides a feasible and simple method for the practical application of drug research.