基于正交试验和神经网络的复合固结土强度预测

Strength Prediction of Compound Stabilized Soil Based on Orthogonal Experiment and Neural Network

以固化剂、石灰、水泥复合固结土为研究对象,采用正交试验的方法进行复合固结土7 d无侧限抗压强度试验的配合比设计,并完成正交试验。针对正交试验结果,采用人工神经网络的方法建立复合固结土强度试验结果的预测模型,通过控制强度试验的配合比指标,训练确定预测模型的模拟函数,从而实现对剩余部分试验结果的预测。研究结果表明,采用正交试验和神经网络相结合的方法,预测复合固结土强度,预测结果准确,误差不超过2.3%。

The compound stabilized soil that consists of curing agent, lime, and cement was the research object in the study. The 7-day unconfined compressive strength test was conducted on the compound stabilized soil whose ratio was designed using the orthogonal experiment method. According to the results of orthogonal experiment, the method of artificial neural network was used to build the prediction model of test results of compound stabilized soil strength. By controlling the match ratio indexes of the strength test and training, the simulation function of the prediction model was identified, which can be used to predict the remaining portion of the results in the experiments. The results showed that the predictions using the orthogonal experiment and neural network combination method to predict the strength of compound stabilized soil are accurate with the error less than 2. 3%.