Water quality anomaly detection approach based on a neural network prediction model

There is too high false positive rate in water quality anomaly detection in water quality data processing with more impulsive noise, so an approach based on radial basis function neural network and wavelet denoising is presented. It introduces wavelet transform modulus maxima denoising method to process the residual sequence prediction of water quality. The quality anomaly of water is determined by the comparison between the distance from the origin at each moment and special threshold, to achieve anomaly detection with higher accuracy. Due to less abnormal data contained in daily water quality data, we perform simulations with a method of superimposing certain distribution based on actual data, to better simulate the variation of water quality parameters in sudden pollution accident of city. The simulation results indicate the improved detection scheme based on neural network, and wavelet analysis has strong on-line detection ability, especially for low-intensity abnormalities, and the accuracy of detection also achieves significant improvement.