In this formula, a is a constant and 1 b a b 10. Tests have shown that
when R equals 15, the neural network has a better convergence effect.
Choosing purelin and tansig as the activation function, learngdm as
the weight learning function in the default network, and trainlm as
the network training function (selecting the default value training
parameters). Model parameters were set as follows: initial learning
rate η = 0.5, inertia factor a = 0.9, the largest number of learning iterations
was 15,000, the learning accuracy was 0.001, the largest number
of iterations was 15,000, and the target error was 0.0001.