CCWI2017: F29 'APPLYING DEEP LEARNING WITH EXTENDED KALMAN FILTER AND GENETIC ALGORITHM OPTIMIZATION FOR WATER DISTRIBUTION DATA-DRIVEN MODELING'

Data-driven analysis has recently emerged as an important task for smart water management as large amount of various data collected via smart meters, sensors and data loggers. Among the methods developed for data-driven modeling, deep neural network (DNN) is proved as the competitive and generic approach to solve many challenging problems, including but not limited to voice recognition, natural language processing, image classification etc. Deep belief network (DBN) is one of the DNNs and widely used for data analysis. This paper extended authors’ previous research in applying DBN model with the genetic algorithm to integrate with the Extended Kalman Filter (EKF). It results in a comprehensive and generic approach, by which the genetic algorithm is employed to optimize the configuration of the DBN and the EKF is applied to assimilate the newly available data with the trained DBN model so that the model can be updated whenever new data becomes available. The proposed method has been tested in the case studies of different domains, including but not limited to water distribution systems. The results show that the deep learning method integrated with EKF has resulted in good performance.