Abstract: An intelligent irrigation system for farmland based on long short-term memory networks（LSTM）prediction was studied, and a PSO-LSTM based intelligent irrigation system for farmland was proposed．Taking crop planting area of the North China Plain as a research area, a data set was constructed by extracting relevant sample data on wheat irrigation daily water demand．Software and hardware of intelligent irrigation systems for farmland were designed．Proposed prediction model was tested experimentally．Results showed that during wheat growth period from sowing to jointing, predicted value curve based on PSO-LSTM farmland irrigation water demand prediction model almost perfectly matched actual water demand curve, with a maximum error value of 0.01 mm/d and a minimum error value of 0 mm/d．Predicted value curve of model during heading to maturity stage was very close to actual water demand curve, with a maximum error value of 0.21 mm/d, a minimum error value of 0 mm/d, and MAE, MSE, RMSE, and MAPE evaluation index values were 0.0511, 0.0067, 0.024, and 0.0103, respectively．Compared with LSTM based prediction model, comprehensive performance was significantly improved and prediction accuracy was higher．It could be used in farmland intelligent irrigation systems and provide a reference for promoting agricultural intelligence development．