The Research And Application Of Neural Network Optimization Method Based On Co-evolution

Like weather forecast, the issue of evaluation result about air quality situation, increases a channel understanding air quality situation for the general populace in a period of time, through this channel, the resident may understand the air quality situation, in order to arrange own work and the journey reasonably. In addition, various enterprises may act appropriate readjustment to production according to the air quality situation, so as to make the environmental protection department of country and local government take measure to reduce and control pollutant, prevent or reduce pollution event's occurrence. There are many factors that decide air quality situation, according to our country's air pollution characteristic and pollution preventing emphasis, at present, the projects included in the air pollution index are: absorbable particles in the air, the sulfur dioxide and the nitrogen oxide compound, because there are no direct linear relations between these three targets and the evaluation result, simultaneously, there are obvious non-linear characteristics, but the neural network can be used in the non-linear problem exactly, and the experimental effect is good, thus carries on evaluation in of the air quality situation using neural network model.Artificial neural network is the complex network system by interactive connection of massive neurons and simulating brain's method in parallel processing and the non-linearity transform, has very strongly auto-adapted, auto-organization, auto-learning capability. In recent years, artificial neural network is widely applied in breakdown diagnosis, pattern recognition, hydrology forecast and so on. Neural network especially BP neural network has the random non-linear mapping ability in particular approaches, applies the neural network in the nonlinear system modeling and the identification, obtains the network's the intrinsic pattern through the massive data training, then carries on the appraisal with the trained network model. In the practical application, the neural network also exposed some own inherent flaws: The weight initialization is stochastic, easy to fall into partial minimum, and the implicit strata node's number and other network parameter need to be determined according to user's experience in the study process, restraining time is excessively long and so on, but the genetic algorithm has the good overall searching ability, and the search does not rely on the gradient information, this article uses the genetic algorithm's superiority to make up the neural network's inherent flaws, carries on the union of the both to solve the evaluation question.The article devotes to researching air quality model's construction and the application of neural network based on the genetic algorithm, the goal is to carry on the coordination evolution using the genetic algorithm to the neural network weight, achieved to neural network's optimizing effect, avoiding study of neural network falling into partial minimum, enhancing the model's appraisal precision.The prime task is:1. Finding the main factors that decide the air quality situation by analyzing many factors that may be related to the air quality situation, constructing solid foundation for the air quality situation 's evaluation.2. Constructing air quality appraisal model based on the BP neural network, and using the matlab software to realize. Constructing air quality situation appraisal model by taking BP neural network as one kind of non-linear mapping of input/output, through training the network many times, extracting the intrinsic non-linear mapping relations between the input and the output, then establishes the BP network model, and uses the trained model in examining data.3. Carrying on the coordination evolution to the neural network using genetic algorithm, constructing neural network air quality situation appraisal model based on the genetic algorithm. To the BP neural network inherent flaw, it is difficult for BP neural network itself to improve fundamentally so as to solve the problem and easy to fall into partially most superior, it appears especially essential for the genetic algorithm's introduction, the genetic algorithm's overall searching function can make up the BP neural network insufficiency exactly.