Application of artificial neural network for wheat yield forecasting

A given model of yield forecasting using an artificial neural network connects the wheat crop with amount productive moisture in soil, soil fertility, weather, and factors presence pests, diseases, weeds. The difficulty creating a forecast system is correct choice predictors that have greatest impact on yield. To build model, 100 cm layer content nitrogen, phosphorus, humus, acidity were used as input parameters. precipitation over 4 months, average air temperature for same period, well weeds also taken into consideration. Data 13 districts North Kazakhstan region period from 2008 to 2017 used. output parameter was spring time period. relative importance variables relation variable determine weight values variables. An error backpropagation method. advantage this method quality increases large training data, ability nonlinear relationships between different data sources. After obtaining predictive good results achieved predicting yields (p=0.52, mean absolute percentage (MAPE)=12.02 %, root square (RMSE)=3.368). Thus, it assumed developed based can be easily adapted other crops places will allow adoption right strategies ensure food security