The Research Of Residual Chlorine Predictive Model In Oilfield Reinjection Water Based On Optimized BP Neural Network

At present,most of oilfields in our country adopt the water recovery mode,which can protect the oilfield environment from the pollution of the oily sewage.But the quality of injection water has an important influence on the development of oilfield,so how to ensure the quality of injection water is the focus in this paper.The residual chlorine concentration is an important index to measure the whole condition of water quality in water injection system of oilfield.It is closely related with the main bacteriological characteristics of the injection water,and it's also the most effective index to control the breeding in water.Because chlorine is a kind of non stability material,its concentration will decrease with time in the transmission process and the sterilization ability decreased gradually,which causes the deterioration of water quality.Therefore,it's very important to predict the residual chlorine and explore the effective way to prevent pollution in water injection pipe network.The change of the concentration of residual chlorine in water is a nonlinear time-varying process,and the water hydrological environment of water injection network is complex and changeable,so there are some difficulties in the establishment and solution of the machine rational model.BP neural network has strong nonlinear approximation ability,self learning and adaptive characteristics,so it is suited to solve some problems in the field of residual chlorine prediction.But because of the limitation of its training mechanism,it has shortcoming of easy to fall into local minimum,long training time and the unsatisfactory prediction effect.In view of the above defects,the improved BP neural network models were built based on Genetic Algorithm and Simulated Annealing Algorithm.The forecast and simulation experiment of residual chlorine in Changqing Oilfield.The predicted results are compared with the actual data to determine the effectiveness of the improved BP neural network model in predicting the residual chlorine in water reinjection.The experimental results show that the improved BP neural network prediction models based on Genetic Algorithm(GA-BP)and Simulated Annealing Algorithm(GASA-BP)have higher fitting performance and better forecast accuracy and the GASA-BP prediction model is better than GA-BP prediction model in predicting.