For reasons of low accuracy of artificial survey leakage, a gas pipeline leakage diagnosis method based on BP neural networks and D-S theory is presented by introducing WSN and information fusion theory. Two sub-neural networks are established at normal node to simplify network structure. The leakage characteristic parameters of negative pressure wave and acoustic emission signals are used as input eigenvector respectively for primary diagnosis. Through making preliminary fusion result s as the basic probability assignment of evidence, the impersonal valuations are realized. Finally, all evidences are aggregated at normal and sink node respectively by using the improved combination rules. The method makes full use of redundant and complementary leakage information. Numerical example shows that the proposed improves the leakage diagnosis accuracy and decreases the recognition uncertainty.

The polyvinyl alcohol (PVA) wastewater was pretreated by the process of flocculation-Fenton oxidation. Raw wastewater was treated by the different flocculants first. The results showed that when the flocculant FeSO4 was used, its effect was better than Fe2(SO4)3 and PAC, the dosage of it was 10g/L, under which COD removal efficiency reached 61.72%. The effects of FeSO4 and H2O2 addition and pH on treatment effect were studied in the follow-up Fenton oxidation. The results showed that when the addition of FeSO4 was 20g/L, the dosage of H2O2 was 250mL/L, pH was 4, the removal efficiency of COD reached over 90%. The experiments of fractionated adding the reagents showed that the removal rate of COD was significantly higher in the situation of fractionated adding FeSO4 required in the flocculation and Fenton oxidation process than one-time added. The removal rate of COD changed little when the H2O2 fractionated addition in the process of Fenton oxidation.

The image definition identification method based on the composite model of wavelet transform and neural networks is stronger in image edge character extraction, nonlinear process, self-adapted study and pattern recognition. The paper puts forward an evaluation method of image definition based on the focusing mechanism of simulating persons eyes by neural networks and on the composite model of wavelet transformation and neural networks. The wavelet component statistics obtained by the wavelet transform are taken as the inputs of the 5 layer RBF neural network model. The model identifies the image definition applying the steepest descent method of the additional momentum in a variable step size to adjust the network weights. The compound model is first trained by 75 images from the training set, and then is tested by 102 images from the testing set. The results show that this is a very effective identification method which can obtain a higher recognition rate.