Proactive Water Supply Management Based on Machine Learning, Sensor Reading and Weather Information

Abstract

The global warming and climate change is a one of the main reason for shortage of water. Therefore, smart use of water resources is necessary for long-term sustainability. Among different large-scale water consumer industries, the agriculture is one of the key consumers. The utilized water in agriculture has polluted and not suitable to use. Therefore, optimization in water management systems in agricultural irrigation system is required. In this paper, we proposed an automated crop irrigation system. The system considers the real and live soil sensor readings for predicting the water treatment plan. On the other hand the weather conditions have also been considered before initiating the water supply. Therefore, decision making function has been formulated. The decision making function requires the prediction of water treatment plan and also the future weather conditions. Therefore, for making accurate prediction ANN algorithm has been trained for performing prediction of both the facts. In addition, for making dataset more effective for training with the ML algorithm a pre-processing algorithm for soil moisture sensor data set has been proposed. The experiments have been carried out and a simulation has been done. Based on the obtained performance the proposed technique of weather prediction using Artificial Neural Network (ANN) provides higher accuracy 99.4% as compared to Support Vector Machine (SVM) which provides 95% accuracy. Additionally for soil conditions this algorithm results 88.4% accurate predictions. In addition, the training and decision making time of the system has also been evaluate which demonstrate the decision can be performed in a fraction of seconds. Finally, the training time is also found acceptable and the system only needs once to train on a location specific data.