Predicting photosynthetic rate of sunflowers using back propagation neural network based on uniform design

The new model to predict photosynthetic rate (Pn) using back propagation neural network (BPNN) based on uniform design (UD) was studied. Four parameters of BPNN were designed at six levels individually by UD experiment to optimize the architecture of the BPNN model. The optimal parameters were used to construct an intelligent, feasible BPNN model which could more accurately predict the photosynthetic rate of sunflowers response to environmental factors. The constructed BPNN model had three layers namely input layer, hidden layer with nine neurons and an output layer. Four environment factors including photosynthetic active radiation (PAR), temperature (T), carbon dioxide level (CO(2)) and relative humidity (RH) were input layers, and photosynthetic rate (Pn) as an output layer. Results showed that the predicted values and actual values of Pn fitted very well, with mean absolute percentage error (MAPE) of 3%, mean square error (MSE) of 0.75 mu mol CO(2) m(-2)s(-1) and mean absolute error (MAE) of 0.72 mu molCO(2) m(-2)s(-1). There was no significant difference using significant test between the actual values obtained from portable photosynthetic system and predicted value calculated by models. The conclusion was that the model established by BPNN based on UD was more accurate than stepwise regression to predict Pn of sunflowers giving the environmental factors (PAR, T, CO(2) and RH).