A data-driven approach for optimizing the utilization of photovoltaic based water pumping systems

A photovoltaic based water pumping system (PWPS) is a promising application specifically for farmers and people living in remote or rural regions that may have limited or no access to the utility grid. However, the wider application of PWPS is limited due to the less efficient utilization of installed photovoltaic (PV) capacity, resulting in a low return on investment. Further, farmers need assistance in deciding the operational status of PWPS due to PV intermittency. Therefore, optimizing PV utilization based on farmers' irrigation and water pumping requirements is essential. In this paper, a data-driven methodology is proposed to optimize PWPS utilization and help farmers make appropriate operational decisions based on water pumping needs and available PV power. A tree ensemble supervised learning-based PV power prediction model has been developed as a first step. To enhance the performance of the PV power prediction model, a Bayesian hyper-parameter optimization algorithm has been applied. During the second step, the PV power prediction outcome for the upcoming days serves as input to decide the PWPS operation in coordination with the farmer's observations regarding the water pumping needs. Based on the predicted PV power availability and irrigation/water pumping needs, the reference signal for motor pump operation would be estimated. To validate the performance of the proposed methodology, a case study has been performed, considering different operational scenarios by means of five use cases. A close match between the predicted and actual PV power generation has been observed. Better PV utilization and farm irrigation have been observed as compared to conventional PWPS. Further, the need of a long term test validation is required to analyse the stability and robustness of the developed methodology, specifically for remote/rural regions.