Fig 1.
Workflow and structure of the insulin pump.
(a) flow chart of insulin pump work structure. (b) PH 300 insulin pump experimental platform for algorithm development.
Fig 2.
Comparison of the transmission structure of the insulin pump.
(a) one-stage screw drive. (b) two-stage screw drive.
Table 1.
Design of insulin infusion regimen for experimental verification.
Fig 3.
Schematic diagram of experimental setup for in vitro study.
(a) model. (b) entity.
Table 2.
Performance comparison between theoretical and actual injection doses.
Fig 4.
Boxplot of theoretical and actual injection doses.
(a) 1U and 2U. (b) 3U and 4U. (c) 5U-7U. (d)8U-10U.
Table 3.
Comparison of single infusion accuracy between single-stage and two-stage pumps.
Table 4.
Accuracy comparison of single-stage and two-stage pumps for low base rate infusion.
Fig 5.
The design of the brushless direct current motor (BLDCM) closed-loop speed regulating system.
(a) BP neural network combined with PID controller structure. (b) BLDCM closed-loop single-speed system is controlled based on BP-PID.
Fig 6.
Simulation performance and control effect of BP-PID.
(a) motor speed waveform with standard PID. (b) motor speed waveform with BP-PID. (c) electromagnetic torque waveform with standard PID. (d) electromagnetic torque waveform with BP-PID. (e) comparison of motor speed overshoot between standard PID and BP-PID. (f) PID adaptive parameter waveform diagram.
Fig 7.
Prediction (a-d) and infusion deviation (e-f) results under different training sets and test sets.
(a) and (e) training set 90%, test set 10%. (b) and (f) training set 85%, test set 15%. (c) and (g) training set 80%, test set 20%. (d) and (h) training set 75%, test set 25%.
Table 5.
LSTM parameter combination by using the grid search method.
Fig 8.
Accuracy and loss of LSTM model training and testing.
(a) LSTM training and testing accuracy. (b) LSTM training and testing losses.