基于稀疏惩罚交叉熵因子的DRBM桥梁监测

Sparse Penalty Cross Entropy Factor Based DRBM Bridge Monitoring

针对建筑结构中存在的安全威胁因素呈现的累积特征,实时危险检测的难度较大,并且噪声和结构会形成复杂干扰使得实时监测难度更大,对此提出一种基于稀疏惩罚交叉熵因子的受约束玻尔兹曼机（SDRBM）的桥梁结构监测算法。首先,基于稀疏惩罚交叉熵因子和受约束玻尔兹曼机（RBM）对深度网络的学习过程进行改进,有效解决深度网络学习过程存在的同质化问题;其次,利用前置粗糙集对输入的桥梁健康信号进行预处理,实现数据信息完整保持和有效归约的平衡,简化处理复杂度;最后,通过实验表明,所提稀疏惩罚交叉熵因子的DRBM桥梁结构监测算法,可获得噪声和结构未知条件下桥梁安全监测。

In view of the cumulative characteristics of the security threat factors in the building structure, the real-time danger detection is difficult, and the complex interference caused by noise and structure makes real-time monitoring more difficult. This paper proposes a bridge structure monitoring algorithm based on the constrained Boltzmann machine（SDRBM） based on the sparse penalty cross entropy factor. First of all, the learning process of deep network is improved based on sparse penalty cross entropy factor and constrained Boltzmann machine（RBM）, which can effectively solve the homogenization problem of deep network learning process. Secondly, the pre-rough set is used to preprocess the input bridge health signal, so as to achieve the balance of data information integrity and effective reduction and simplify the processing complexity. Finally, experiments show that the proposed DRBM bridge structure monitoring algorithm with sparse penalty cross entropy factor can obtain bridge safety monitoring under conditions of noise and unknown structure.