Application Of An EEMD-GA-BPNN Model In Dam Deformation Prediction

Reservoir dams have a series of functions including water storage irrigation,water power generation,flood control and drought prevention,and play an important role in economic construction and social development.Due to the complicated hydrogeology of the dam and the people’s lack understanding of factors affecting dam safety(such as dam material properties,mechanical structure mechanism,dam construction control,and natural damage),when an earthquake,flood or other anomalous situation that may affect the safety of the dam occurs,the safety hazards of dams will expand.Once danger occurs,it will cause unpredictable loss of life and property.In order to ensure the safe operation of the dam,it is urgent to monitor the safety of the dam.Using monitoring data to predict and evaluate dam safety status and eliminate hidden dangers of dam is the important purpose of monitoring.However,the dam is affected by reservoir water pressure,uplift pressure,reservoir temperature,aging and other uncertain factors.These factors are highly stochastic and complex,resulting in highly nonlinear and non-stationary data of dam deformation.Therefore,it is difficult to use traditional mathematical models to quantitatively describe the relationship between these factors and dam deformation.Many scholars have introduced artificial neural networks to the field of dam safety monitoring.The function approximation capability of artificial neural network model is used to fit the complex function relationship between dam deformation effect and influencing factors.The prediction accuracy of the fitting is greatly improved compared with the traditional model and has achieved good results.However,in the practical application process,neural network has the defects of low generalization ability and easy to fall into local minimum value,and its parameter setting needs to rely on manual judgment,so it has more uncertainties,and there are limitations in practical applications.In view of the above problems,this paper studies the improved method of BPNN based on artificial neural networks,and introduces Genetic Algorithm to optimize the structural connection weights and neuron thresholds of BPNN,which will improve the performance of the model and avoid falling into local minimum.At the same time,considering the non-linear and non-stationary nature of the dam deformation data,the EEMD principle is used to decompose the dam deformation data to reduce its complexity,and each component is learned and predicted by the BPNN.Finally,the component predictions are combined as model predictions.In order to understand the performance of the combined optimization method and the prediction accuracy,a numerical analysis was performed on the traditional model and the combined models.The performance and accuracy were compared from the model training process and the predicted results respectively.It was proved that the GA-BP model based on EEMD decomposition has higher performance.The combined model EEMD-GA-BP can be used as a reference method for the dam prediction model.