Abstract: The traditional method of gravelly soil blending quality evaluation mainly uses supervision of the side station to control its blending construction parameters, and takes the P5 content of a randomly sampled point as the blending quality evaluation index. It is not only subjected to human factors but also limited to the poor timeliness and great randomness, and is hard to truly reflect the uniformity of blending. To solve these problems, we present a new dynamic method for evaluating gravelly soil blending quality based on an IBAS-BP neural network. First, a blending uniformity value (BUV) is worked out to reflect blending quality through a comprehensive consideration of the contents of different particle sizes; then, we develop a new method that can achieve real-time monitoring and collection of the construction parameters of gravelly soil blending based on GPS positioning, wireless transmission and so on. Finally, to realize the dynamic evaluation of the whole work unit, we adopt a BP neural network that is optimized by the Beetle Antennae Search (BAS) algorithm using the backtracking idea and the non-linear step adjustment function. Engineering application shows that the blending quality evaluated using BUV agrees with the test results sampled after blending, and our monitoring method is effective and applicable. Comparison of the model predictions with measurements gives a correlation coefficient of 0.953, an average error of 0.23%, and a standard deviation error of 6.17%, thus improving consistency, representativeness and superiority over traditional regression models.

Key words: gravelly soil, blending quality, blending uniformity value, real-time monitoring, IBAS-BP neural network