Improved Bp Neural Network In The Underground Water Quality Assessment

Artificial neural network (ANN) is a math-model simulating biological neural network of human brain using computer. It has adaptive, self-organizing and fault-tolerant performance. And it has been widely applied in many-fields such as processing information, automatic control and pattern recognition and so on. The network consists of layers of parallel processing elements called "neurons". The work principle is similar to the course of environment influenced by external stimulus.There are many methods on ground water quality assessment at present, such as single factor pollution index method, integrated index assessment, fuzzy mathematics method and gray system evaluation method, etc. Though each of those methods has certain advantage, most of them need to construct subject functions, to confirm the power values. The designs of those subject functions contain certain artificial factors, and the calculations of power value embody reasonless aspects, as a result much important information will be lost in some circumstance.It can make up for these shortcomings of traditional assessing methods in some certain, most of which need to construct subject functions, can't accurately describe the level interval change, bring certain subjective effects to assessment processes.BP algorithm is the most important and the most popular algorithm in artificial neural network. As BP neural network model can be a good simulation of nonlinear systems, it has a strong ability to learn the network structure, so it can be widely used in scientific predictions. However, the BP neural network is slow to learn, is no guarantee that the convergence of the smallest, the network of learning and memory is not stable, the selection of Network implied stores and the number of hidden layer unit is no unified and other shortcomings. According to the aforementioned shortcomings of BP network, science and technology workers are continuing to explore. The main research directions are: First it is to study generalization, Generalization is the ability to use the less sample to train the network, and the network without a lot of input patterns can also learn to handle correctly; Second it is to study the multi-layered structure of the new network; The third is to optimize network performance. So this paper is an attempt to use the function of the neural network model to improve its weaknesses. The main work of the paper: (1) BP neural network model has been applied to groundwater quality assessment;(2) To improve the BP algorithm. The Changes is to add momentum factor and adaptive factor in the value of the threshold value and the right value, to amend the exciting function, to optimize hidden nodes. I do experiment to compare the improved BP algorithm and the standard BP algorithm, it can be drawn that the improved BP algorithm can effectively accelerate the speed and reduce the training error.(3) In order to better explain the fine of the BP neural network forecasting results, the paper uses single-factor and fuzzy comprehensive evaluation method with which to do a comparison.In short, the paper applied the three-BP neural network model to the data from groundwater samples Kunming, called MATLAB to C # training with BP to develop underground water quality evaluation system. After tests, the model that the paper set up based on artificial neural network to predict the on groundwater quality evaluation model is better than the other. I believe this conclusion on the groundwater evaluation of this paper is a guide, it could also provide for the evaluation of groundwater quality in establishing the groundwater quality evaluation.