Enhanced Dynamic Model of Pneumatic Muscle Actuator with Elman Neural Network

Table 4

The results of neural network training for and . In the table, is normalized mean squared error, MSE is mean squared error, EM is enhanced model, M is model, is neural network input vector, is model joint angle, is model angular velocity, is model angular acceleration, is left model muscle force, is right model muscle force, is left model muscle pressure, and is right model muscle pressure.

Variable	training	MSE; testing EM	MSE; testing M	Convergence
Joint angle,	0.00522	1.607	148.896	Yes (444 it.)	, ,
Left muscle force,	0.004	105.925	6960.1	No	, , ,
Right muscle force,	0.00688	83.412	5278.5	Yes (164 it.)	, , ,
Left muscle pressure,	0.00595	102.757	5401.6	Yes (88 it.)	, , ,
Right muscle pressure,	0.006	113.651	6247.5	Yes (66 it.)	, , ,
Joint angle,	0.00237	4.146	80.721	No	, ,
Left muscle force,	0.00644	220.236	8679.9	No	, , ,
Right muscle force,	0.009	316.135	5857.7	No	, , , ,
Left muscle pressure,	0.00292	100.206	3862.9	Yes (535 it.)	, , ,
Right muscle pressure,	0.00391	123.295	7022.3	Yes (449 it.)	, , ,