Abstract for: Toward an integrated use of system dynamics and deep machine learning; a case of groundwater dynamics

Addressing the continuous need for better modeling practices and consequently better forecasts for the future, the research aims to use both deep machine learning and system dynamics simulation. The focus is the application of simulation results as a feed to train a machine using the artificial neural networks technique (MLP). The case is a vast and large system dynamics model that captured the dynamics of groundwater changes and water scarcity issues in the lower Rio Grande (LRG) region within Dona Ana County, New Mexico in 2019. Focusing on change in groundwater and using the historical (1976 - 2011) data sets (training set and test set) of simulation for 17 variables, ANN replicates properly the data with RMSLE of 4.4E-5 and R-square of 0.97 as performance metrics. It means the future prediction of ANN could be used as more reliable data. The comparison shows a very acceptable fit with the test data set from history while there are some notable differences in future forecasts. Also, there is a new model to understand the change in groundwater while benefiting from an integrated approach of system dynamics and deep machine learning.