Monitoring data exception diagnosis method based on generative adversarial network

A technology for monitoring data and anomaly diagnosis, applied in biological neural network models, neural learning methods, instruments, etc., can solve problems such as massive and complex information processing capabilities that need to be improved, and achieve automatic classification, simplified learning methods, and small overlap Effect

CN111695598AActive Publication Date: 2020-09-22SOUTHEAST UNIV
5 Cites 11 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Publication Date
2020-09-22

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a monitoring data exception diagnosis method based on a generative adversarial network, which is suitable for exception diagnosis of long-term measured data of a structural health monitoring system; the method includes dividing the recorded data into a plurality of sub-segments by selecting a proper basic time interval; converting the monitoring time sequence data of each sub-segment into a grayscale image by adopting a Gram angle field, training two unsupervised deep artificial neural networks including a generative adversarial network (GANs) and an auto-encoder (AE) according to the grayscale image, and verifying a training effect of the obtained network according to a prediction error of a test data set. Secondly, selecting an optimal index suitable for data anomaly diagnosis according to samples of the training set and the test set, and judging the state of the measured data in combination with a cumulative summation function. According to the method, rapidand accurate diagnosis of abnormal monitoring data can be realized, and effective data support can be provided for structural state diagnosis and abnormal early warning.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention relates to a data abnormal diagnosis method, in particular to the abnormal diagnosis of long-term monitoring data of a structural health monitoring system, and belongs to the technical field of data abnormal diagnosis. Background technique

[0002] The Structural Health Monitoring System (SHMS) includes numerous sensors for monitoring structural responses such as acceleration, velocity, displacement, and strain, as well as environmental factors such as temperature, wind speed, and humidity. Long-term monitoring data can provide reference for structural damage identification and state assessment. Before carrying out damage identification and status assessment, the accuracy of monitoring data must be ensured, otherwise it will lead to wrong assessment results. However, most sensors are located in harsh environments, and it is difficult for sensors to avoid failures. Therefore, the diagnosis and elimination of abnormal data is one of the key...

Examples

Embodiment 1

[0035] Embodiment 1: see Figure 1-Figure 3 , a monitoring data anomaly diagnosis method based on a generative confrontation network, the method comprising the following steps: Step 1: training data preparation stage, step 2: network training stage, and step 3: data anomaly diagnosis stage.

[0036] Step 1: training data preparation stage, this stage includes two steps:

[0037] Step 1-1: Determine the minimum sampling frequency of the required data according to the analysis purpose, perform resampling operation on the data recorded by the structural health monitoring system (SHMS) to be detected, and down-sample the original data without losing the original data. inherent structural information;

[0038] Step 1-2: Set the basic time interval, divide the measured data into several sub-segments, perform normalization operation on the data of each sub-segment according to formula (3), and normalize it to the interval [-1,1]. Then, the dataset of each subsegment is converted in...