Mathematical Function Approximation Modeling Mechanism for Farm Irrigation Behavior Based on Large-Scale Power Data

Abstract

Abstract Water resource constraints are undoubtedly a serious challenge for agriculture in China, and the demand for farmland irrigation is an important factor affecting agricultural production. To address this problem, this paper constructs a DQN model modeled by numerical function approximation to optimize the farmland irrigation behavior. Using the Q learning algorithm in the reinforcement learning algorithm, the farmland irrigation behavior action to action value function Q ( s, t ) is calculated iteratively, combining the behavior network and the target network to construct the DQN model. The parameter optimization of the algorithm was achieved by replacing the DQN farmland irrigation behavior network function with the value function of the linear function. The model was used to measure the water use efficiency of farmland irrigation, so as to deduce the electricity demand, and the average value of irrigation electricity efficiency was 0.566, and the minimum and maximum values were 0.096 and 0.934, respectively, which demonstrated that the problem of wastage of electricity used by farmers for irrigation is very serious, and that there is a great potential for saving resources. Under the irrigation method of canal-well combination, the maximum value of daily change of water layer thickness obtained by the model simulation is 50.36mm, and the minimum value is -6.63mm, which is in line with the measured change, and verifies the accuracy of the calculation results of the model designed in this paper.