Prediction of Total Phosphorus Based on Grey GA-LM-BP Neural Network

The overall change characteristics of total phosphorus (TP) could be described by grey theory, but its fitting error was large. Thus we used the BP neural network to amend the residual error. LM-BP neural network has strong nonlinear mapping ability, but it is sensitive to the initialized weight and threshold values, and its generalization ability is not strong. We used genetic algorithm (GA) to optimize the initialized weight and threshold values of LM-BP neural network. We established GA-LM-BP based on grey GM(1,1). This method can utilize the advantages of grey theory, BP network and GA, and it can overcome the shortcomings of grey theory and LM-BP network. The TP data of Lun River from XiaoGan segment were fitted, tested and forecasted. Results showed that the testing errors were less than 2.52%. So it testified the validity of this method.