Implementation of Data Jittering Technique for Seepage Analysis of Earth fill Dam Using Ensemble of AI Models

Recently, it has been shown that Artificial Intelligence (AI) methods such as Feed forward neural network and Support vector regression have great capability in modeling of non-linear hydraulic time series. AI methods offer effective approaches for handling large amounts of dynamic, non-linear and noisy data. Hence in this paper, seepage of Sattarkhan earth fill dam using two AI models of Feed forward neural network and Support vector regression was simulated, based on 2 scenarios with different combination of inputs. Afterwards, as a pre-processing method for improving the model performance, normally distributed noises with the mean of zero and various standard deviations were generated and added to the time series to form different jittered training data sets, for AI modeling of seepage. Further, as another method for improving the model performance, an ensemble post-processing model was developed using outputs of sole models. Non-linear neural averaging method was considered for model ensembling. The obtained results indicated that simultaneous application of the both jittering and model ensembling methods improved the model accuracy up to 32% in the verification step.