The BP Neural Network Structure Optimization Research And Application

In recent years, the development of science and technology in China has brought a lot of convenience to people's life and promote the sustainable development of the economy in our country. At the same time, the rich lifestyle also causes great harm and irreparable damage to our environment. Sea water quality monitoring is the important work of the ocean science, so protecting ocean environment is of great significance to the development of ocean resources. Affected by human activities and climate, the coastal water environment in China is hugely changed, and ocean structure has also changed. Protecting ocean environment is important for maintaining the balance of ecology. Sea area in our country is large, and it has a wide variety of spatial structure and characteristics. Among them, the bohai bay with semi-closed geographical characteristics has loser water purification and circulation ability than other waters. So the water quality of China's bohai bay provides important value to the assessment of water environment. The concentration of chlorophyll a is representative in the assessment of ocean environment. There are already some methods to research it, but for such waters with special structure like Bohai bay, the error is relatively large. Therefore, to establish suitable chlorophyll-a concentration inversion model for bohai bay waters is very necessary.This study is based on the spectral characteristics of bohai bay water, using processing data and the corresponding chlorophyll a concentration data measured under different spectral characteristics, building the chlorophyll a concentration inversion model of bohai bay. I collect and process the four band radiation sensitive value as input data according to the demand of the inversion model. At first the relationship between them and the measured concentration will be established by applying the BP neural network, then I adjust the parameters and contrast the errors, finally BP neural network model willed be established whose square error RMSE is 14.54%, in which input layer contains four nodes, implicit layer contains 10 nodes and the output layer contains one node. Results show that BP neural network model has a good effect.By using high-order neural network simplification form to optimize the structure of BP network, solve the problem of the remote sensing technology in Marine application complicated nonlinear. In order to overcome the defects of neural network easy to fall into local convergence, the initial parameters with genetic algorithm for neural network optimization. Genetic algorithm(GA) to assign genetic parameters to that of the BP neural network as the initial weights and thresholds, accept the return error of and as the evaluation of fitness. Gradually optimized, find out the neural network model parameters of small errors. Results show that the optimized BP network model error was 5.67%, the optimization model has better fitting effect.