Research On The Optimized Prediction Model Of Water Quality Based On RBF Neural Network

As a municipal infrastructure, city pipeline network is obliged for residents to providehigh quality and pollution-free drinking water. However, in the massive complicated pipelinenetwork, the water is likely to be contaminated, causing the quality declining and threatingpeople’s health. Therefore, the studies on predicting the quality change of pipeline water andthe corresponding measures have attracted more and more attentions. The traditionalprediction methods, like fuzzy mathematics and grey system, are falling into disuse due totheir limitations and large errors. Recently, thanks to the infinite approximation ofdifferentiable functions, the RBF neural network is widely applied in the water qualityprediction, which can improve the accuracy and overcome the flaws in other methods.Nonetheless, because of certain drawbacks existing in RBF neural network and thecomplexity of water quality variation, it is urgent to optimize the RBF prediction model. Withthis purpose, this paper thoroughly studied the RBF algorithm and its application. Theresearch emphasized on the following aspects:（1）First we introduce the basic theory of RBF neural network, summarizing threeshortcomings in traditional RBF prediction model and giving corresponding improvements asfollows. The golden section is adopted instead of manual selection based on expertexperiences to determine the node number in hidden layer, therefore, improving the datareliability. The adoption of L-M algorithm to determine RBF neural network, speeding up itsconvergence. The genetic algorithm is introduced to determine initial weights and thresholds,avoiding the local minimums. Then we establish the pipeline water quality prediction modelbased on the optimized RBF, which is simulated and compared with traditional model withsample data provided by Guangzhou University City to prove its advantage.（2）Based on the research above, a user interface of the model is created by MATLABR2007b. Not only can the optimized model be used for non-technical staff to evaluate waterquality by monitoring sample data, but also build more requirement-fitting water qualityprediction model easily. It is possible that improved RBF model can be applied in real life.