基于全神经网络增强算法的WSNs故障预警与检测

针对现有WSNs故障检测算法存在的故障分类检测率低、耗时长、节点能耗控制差等问题,提出一种全神经网络增强故障预警与检测算法。全神经网络的神经元节点与临近层的节点连接,形成具有强大故障数据训练功能的深度网络结构,选择平滑性更好的sigmoid函数作为模型的激活函数,并基于感知机合理调节相邻两个隐含层之间的阈值权重,降低模型的训练损失；采用Adam优化算法抑制模型的梯度膨胀和梯度消失等异常情况,并消除训练中产生的数据冗余,以降低故障数据训练中产生的虚预警。实验结果显示:提出算法的总体故障检测率和不同类型故障的分类检测率都优于传统算法,此外全神经网络增强算法在节点故障检测耗时和能耗控制方面,也具有显著优势。

WSNs Fault Warning and Detection Based on Full Neural Network Enhancement Algorithm

A fully neural network enhanced fault warning and detection algorithm is proposed to address the problems of low fault classification detection rate, long time consumption, and poor control of node energy consumption in existing WSNs fault detection algorithms. The neuron nodes of the full neural network are connected with the nodes in the adjacent layers to form a deep network structure with strong fault set training function. The sigmoid function with better smoothness is selected as the activation function of the model, and the weight threshold between the adjacent two hidden layers is reasonably adjusted based on the perceptron to reduce the training loss of the model; Using the Adam optimization algorithm to suppress the gradient expansion and vanishing of the model, and eliminate data redundancy generated during training, reducing the false warning generated by fault data training. The experimental results show that the overall fault detection rate and classification detection rate of different types of faults of the proposed algorithm are superior to traditional algorithms. In addition, the full neural network enhancement algorithm also has significant advantages in node fault detection time and energy consumption control.