Fast-prediction of automotive wiring-harness crosstalk based on GA-BP neural network model

This paper puts forward a kind of prediction model for auto-harness-wire crosstalk based on the GA-BP neural network model. Utilizing its peculiarity to the problem of nonlinear adaptive capacity, and giving full consideration to the influence of each factor on the auto-harnesswire crosstalk, the nonlinear mapping prediction model between 8 factors and auto-harness-wire crosstalk is set up. The experimental results show that, to achieve convergence, the BP network needs 403 times training while GA-BP just 30 times, and the correlation coefficient of GA-BP model prediction value and sample’s actual test value reaches 0.999, showing strong correlation. The MAPE value of BP network prediction model and GA-BP network prediction model respectively are 4.7514 and 1.2370. And by comparing the output value of GA-BP network and the sample’s actual values, it can be found that the auto-harness-wire crosstalk model built by the GA-BP neural network is superior to the traditional one in convergence speed and predicting precision. Finally, the effect weight of every factor to the crosstalk is determined by the weight analysis method, and the prediction model is simplified by reducing the input dimension of the model in maximum on the premise of guaranteeing the prediction precision.The research in this paper not only offers a reliable and effective method for the fast prediction of auto-harness-wire crosstalk, but also has enlightenment significance for the electromagnetic compatibility design of auto-harness-wire.