Double Q-PI architecture for smart model-free control of canals

Abstract

Nowadays, pressurized irrigation systems have been developing in farms to increase water use efficiency that are successful when the inflow is accurately supplied from water sources. In large irrigation networks, water conveyance and distribution systems are mainly open channels facing several uncertainties. Therefore, water delivery to farms is of the most important tasks that should be done accurately to supply sufficient water to pressurized irrigation farms, causing desired performance of pressurized systems. To this end, regulating structures within irrigation networks should be controlled. Artificial intelligence, as robust and new technology, has been employed for controlling complex systems in the industry. In this research, a new and robust algorithm, namely Double Q-PI (DQ-PI), was developed with the aim of water management in irrigation canals by controlling water depth. It uses a traditional Q-learning algorithm and a double update matrix to tune PI (Proportional-Integral) gains for controlling check gates. It improves its performance using the receiving signals from the irrigation canal model. The approach was tested using several scenarios and evaluated by standard performance indicators. The results showed acceptable accuracy and a reasonable ability in controlling water depth within the canal’s reaches. The maximum and average errors were 11.6% and 10.5%, respectively, resulting in significant improvement in water management. With this degree of accuracy at the canal level, the pressurized systems work properly, and the expected efficiency can be achieved.