Support Vector Machine In Sewage Treatment, Energy-saving Applications

Today, the cost of wastewater treatment is expensive, the main reason, is that a lot of electric waste has been generated in the process. For the record, every year, 60%~70% of the electric consumption comes from air-blowers, which usually runs full-load with valve controlling air input to aeration tanks, therefore, how to reduce the electric consumption of air-blowers without effecting the quality of wastewater treatment is an urgent issue to solve, which also conforms the energy-saving policy of our country. Targeting on these issues, with support vector machine as soft sensor modeling method, an energy-saving scheme for air-blowers in wastewater treatment process is proposed in this paper, works we've mainly done include:1. The adjustment for blowers are based on the density of oxygen ( DO ) in wasterwater, due to some time-delay of DO that changes with environment parameters, a soft sensor measurement method is proposed in the paper on the basis of massive data from wastewater treatment factory and numerous experiments. Some prediction experiments are made relying on support vector regression, which results in relatively high accuracy.2. A data pre-processing method based on clustering analysis is proprosed in this paper, which is used to trim the training data for support vector regression, making support vectors more sparse and having higher excution efficiency. To prove the efficiency of this method, clustering-support vector regression is implemented into the prediction of DO,TN and SVI , which has successfully proved the fact that support vectors become more sparce after C - SVR while relatively high accuracy is still guaranteed.3. Some experiments on the comparison among C - SVR, PCA and RS are made in this paper, which have been yielded through 8 different soft sensor models: SVR , C - SVR, PCA - SVR, C - PCA - SVR, RS - SVR, C - RS - SVR,RS - PCA - SVR,C - RS - PCA - SVR, these models are respectively implemented into the prediction of DO,TN and SVI , proving that C - SVR is the most efficient soft sensor modeling method while having high prediction accuracy among these 8 models.4. An brief energy-saving scheme combining C - SVR and fuzzy control for air-blowers is proposed in this paper.