基于LS-SVM的粉煤灰混凝土强度预测模型

Application of LS-SVM to strength forecast model for fly ash concrete

提出一种基于最小二乘支持向量机(LS-SVM)的粉煤灰混凝土强度智能预测模型,并给出了相应的步骤和算法。通过该模型分析了水胶比、水泥用量、粉煤灰替代率及砂率等因素对粉煤灰混凝土强度的影响。在此基础上,对不同配比所浇注的混凝土强度进行预测,有助于准确认识混凝土强度随配比参数的变化规律。与多元线性回归、神经网络及标准SVM模型比较,该模型的优点为:(1)采用了结构风险最小化准则,在最小化样本误差的同时减小模型泛化误差的上界,提高了模型小样本泛化能力;(2)将迭代学习算法转换为求解线性方程组,使得整个模型仅有一个全局最优点,解决局部最小问题;(3)用等式约束代替标准SVM算法中的不等式约束,将求解二次规划问题转化为直接求解线性矩阵方程,有效提高建模速度。用该模型对混凝土的强度预测实例表明,其建模速度比标准SVM高近1个数量级,预测误差仅为SVM方法的20%、BP神经网络方法的10%左右。

A novel prediction model for the strength of fly ash concrete based on least squares support vector machine （LS-SVM） was proposed, The design steps and learning algorithm were given. Four main factors including the W/B （water-binder） ratio, amount of cement, substitution rate of scoria and sand percentage, which influence the strength of fly ash concrete, were analyzed by the proposed model, Based on this, the strength of fly ash concrete with different proportioning can be predicted that is helpful to correctly understand the variation of concrete strength with proportioning parameters. Compared with forecasting models based on linear regression, artificial neural networks （ANN） and standard SVM, this model possesses prominent advantages： （1）over fitting is unlikely to occur by employing structural risk minimization criterion to minimize the errors at the samples and decrease simultaneously the upper bound of the predict error of the model; （2）the global optimal solution can be uniquely obtained owing to that learning algorithm converts machine learning into linear equations; （3）the constraints of inequalities in the standard SVM approach are replaced by equality-type constraints in LS-SVM and the LS-SVM solution follows directly from solving a set of linear equations instead of quadratic programming. The practical experiment results show that the speed of this LS-SVM model is about one order of magnitude higher than SVM model, while the prediction errors are 20% of SVM model, and about 10% of BP model.