改进的BP神经网络模型预测充填体强度

Strength prediction of backfilling body based on modified BP neural network

摘要:

为了准确评价和预测充填体强度,采用改进的BP神经网络算法,利用实验室做的18组充填体单轴抗压强度试验结果,建立了充填体强度与影响因素之间的5-7-1网络模型结构(输入层为5个神经元,隐含层为7个神经元,输出层为1个神经元,输入为胶砂比及各胶凝材料掺量,输出为充填体28 d单轴抗压强度)．结果表明,改进的BP神经网络对于充填体的强度具有良好的预测能力,建立的网络模型不仅收敛速度快而且训练精度高,对充填体强度的预测结果与训练数据和测试数据的最大相对误差仅为4.23％．

Abstract:

To evaluate and predict the strength of backfilling body, a new method is provided to establish a model of the relationship between backfilling body strength and influence factors. The modified BP neural network algorithm is used to establish the model based on 18 groups results of the compressive strength tests of the backfilling in laboratory. The structure of the model is 5-7-1 type, that is to say 5,7 and 1 neurons are the input, hidden and output layers respectively, where the input is including the cement-sand ratio and quantity of the cemented material and the output is the 28 days compressive strength of the backfilling body. The results show that BP neural network model exhibits excellent prediction ability for the prediction of the strength of backfilling body, the prediction model speeds up the convergence rate of BP network and improves the training accuracy. The maximum relative error between the predicted results and the test data is 4.23％.