Application Of Improved BP Artificial Neural Network Algorithm Model In Heavy Metal Prediction Of Water

Water quality prediction is an important task to ensure the safety of water supply.By using certain means to predict and evaluate the trend of water quality in the water source in the future,it is possible to grasp the possible changes in water quality more clearly.Especially for the future situation where the water quality may be polluted and deteriorated,corresponding measures can be taken in advance through prediction and analysis to minimize the impact of water quality changes on the safety of water supply.Because the water environment is very complex and the influence relationship between various water quality factors is not clear,the traditional water quality prediction methods with modeling as the main feature generally have problems such as tedious operations and low prediction accuracy.In order to solve these problems,this paper proposes using BP artificial neural network algorithm based on intelligent calculation to predict water quality.Rely on the powerful non-linear mapping ability and automatic learning and adaptability of BP artificial neural network algorithm to analyze and deal with complex water quality relationships.However,in actual operation,it is found that the BP artificial neural network algorithm has a large adjustment range and number of adjustments in the weight threshold of each connection layer during the reverse transmission of the error due to the characteristics of its own structure.This shortcoming makes the algorithm in the running process.It is easy to fall into the local optimal solution and reduce the accuracy of prediction.At the same time,when selecting the input layer independent variables of the BP artificial neural network algorithm,because there are many water quality monitoring indicators,there is no clear regulation on the type and number of indicators that need to be selected.The more indicators that are selected,although the more they can reflect the true situation of the water body,But it also adds a lot of unnecessary information,which makes it easy for the algorithm to learn the irrelevant information,which reduces the accuracy of prediction.At the same time,the increase of the original data dimension also makes the operation more difficult.In order to solve the above problems,this paper improves the traditional BP artificial neural network algorithm.Aiming at the problem of large subjective random selection of water quality indicators during the construction of the algorithm model,a combination of Pearson correlation coefficient method and information indicator evaluation method was used to calculate the variance of each water quality indicator,and then calculate the information holding of each indicator.In the end,the information holding degree is greater than 80% as the criterion for selecting independent variables.At the same time,the BP artificial neural network algorithm should be used for the disadvantage of poor final prediction accuracy caused by excessive adjustment of the weight threshold of the connection layer.The genetic algorithm for search and optimization is improved,and a relatively superior initial weight threshold is assigned in advance before the BP artificial neural network algorithm is run,so as to reduce adjustments in subsequent processes and improve prediction accuracy.In order to fully demonstrate that the improved BP artificial neural network algorithm has higher accuracy in water quality prediction,the multiple linear regression method in the traditional water quality prediction method,the unimproved BP artificial neural network algorithm and the improved BP artificial The neural network algorithm predicts the water quality of a certain water body from January 2011 to December 2016(with arsenic as the predicted index).By analyzing the prediction results,we can see that the average error of the traditional multiple linear regression method is 27.92%,the prediction error of the unmodified BP artificial neural network algorithm is 21.72%,and the prediction error of the improved BP artificial neural network algorithm is 15.79%The remaining evaluation indicators also show that the improved BP artificial neural network algorithm has a better prediction effect,indicating that the improved method of this paper is effective.