Combined effect of multiple factors on the photoacoustic detection of glucose solution based on neural networks

In this study, the combined effect of multiple factors on the photoacoustic detection of glucose solution was studied by using the artificial neural networks. These multiple factors include the excitation energy, temperature, flow velocity, and glucose concentration. To achieve the aim, a set of photoacoustic detection system of glucose solutions was established by using the OPO pulsed laser, ultrasonic transducer, the solution cycling sub-system, signal pre-processing circuit, data acquisition and controlling system. The peak-to-peak values of glucose solutions with different concentrations, temperatures, flow velocities under different energies of pulsed laser were all obtained by the orthogonal experiments. To get the relationship between the peak-to-peak values of glucose and the multiple factors, as well as between the glucose concentration prediction and the multiple factors, four different artificial neural network (ANN) algorithms, i.e., forward propagation (FP), radial basis function (RBF), back propagation (BP), and recurrent neural network (RNN) were used. The root-mean-square error(RMSE) values of different ANN algorithms for the peak-to-peak values prediction and the glucose concentrations prediction were all obtained and compared. Results show that the concentration prediction effect of RBF-ANN is better than that of RNN, BP-ANN, and FP-ANN. Then, to further verify the concentration prediction performance of RBF-ANN, the RMSE values for RBF-ANN algorithm under different parameters of spread were adjusted and compared. Results show that when spread value is 0.7, the RMSE values of photoacoustic peak-to-peak value and the glucose concentration are lest. When the neurons is 108, spread parameter is 0.7, the RMSE of glucose concentration is about 0.45mg/dl.