Soil phosphorus content prediction based on improved PSO-LSTM algorithm
Longmin Shi, Yong Liu · 2021
Available P is an important index affecting crop growth. Accurate detection of available P content in soil can provide a guarantee for accurate fertilization. Compared with the traditional detection methods, a particle swarm optimization long - short - term memory artificial neural network (PSO-LSTM) algorithm was proposed to predict the soil phosphorus content. In this model, particle swarm optimization algorithm is introduced to find the optimal parameters iteratively, instead of adjusting parameters based on personal experience. The PSO-LSTM model was built under the KERAS framework, and various parameters detected in the planting base of sugar beet in Hulan, Heilongjiang Province were used as input to predict the soil phosphorus content. The experimental results show that the absolute square error, absolute 100% ratio error and R2 of the improved PSO-LSTM model are 0.115, 0.394 and 0.945, respectively. Compared with other neural network models such as LSTM, it has higher prediction accuracy.