Predicting Gate Operation in Open Canal Control With Multitask Sequential Model

Open canal gate control is an essential type of water resources management for irrigation and water diversion. However, accurately predicting gate operation in open canal control remains a vital regulation algorithm in Internet of Water irrigation. One-to-many mapping relationship between water discharge requirement and gate operation strategy makes the predicting problem difficult. Therefore, real-world application is mostly human-experiences depended. This article proposes a multitask sequential neural model with a two-phase learner–evaluator procedure tackling the one-to-many mapping relationship. The evaluator is first trained for water discharge prediction with gate control operations. Then, the learner is trained with the optimized evaluator to tackle the gate control operation prediction with the indicated water discharge requirements. Considering actual gate control scenarios, the change of gate opening heights is introduced as one of the multitask optimization targets. Results show that the proposed model achieves a high accuracy, and has higher efficiency than the numerical calculation method.