1 11。12 11。26 1。3
2 12。23 11。57 −5。4
3 14。32 14。02 −2。1
4 15。05 16。11 7。0
5 12。58 13。02 3。5
6 13。05 13。65 4。6
The function in the toolbox of MATLAB neural network is called to train the BP network。 After 240 cycles, when the total squared error of the output of samples reaches 0。000 92, the relative error of the simulation results does
not exceed 0。1%, indicating that the fitting accuracy is sufficiently high。 With the reserved six groups of data as the input of the BP network, the output is compared with the actual parameters。 Then the BP network testing is completed。 It can be seen from the comparison (Table 3), the absolute values of the maximum relative error does not exceed 8%, which indicates that the BP network has some predictive ability。 With the increase of training samples, the relative error will continue decreasing。 Therefore, the predictiveability of the BP network will be further strengthened。 It suggests that the BP network represents good applicability and reasonable accuracy for the simulation of cylinder oil consumption model。
4。3 Optimization process and results
Cylinder oil consumption after removing the cylinder liner wall evaporation is selected as the target parameter to optimize。 It is assumed that the nonlinear mapping
relationship from the piston ring parameters to the cylinder oil consumption values after the removal of the cylinder liner wall evaporation is D=f(x1, x2, x3), the optimal parameter combination is the peak value of the discrete points taken from D=f(xi)。 Near the peak value, there may be a better combination than the existing parameter combination。 So the established BP neural network is used to further search for a better parameter combination。 A search method with the gradually reducing step size is adopted for optimizing。