Abstract: The improved neural network model with more stable precision and visualization was used to predict the grain yield. The results showed that the correlation coefficient between the predicted value and the measured value of grain yield was 0.991 039 8, the relative error was 0.006 339 629, and the average error of 10 cross tests was 0.001 229 604. The results showed that the relative error between the predicted value and the measured value was small, and the fitting effect was close to a straight line. The relative error was 3.9% and 1.1% higher than that of spss multi-layer perceptron and nnet packet respectively. The results showed that the neural network prediction model had good prediction and feasibility. It can be used as a new technology to balance the utilization and distribution of crops, and promote the continuous growth of agricultural products economy and the sustainable development of the agricultural industry.

Key words: grain yield prediction, artificial neural network, fitting curve, verification, sustainable development