A multi-index highway camera equipment status detection and fault warning method

Through unsupervised learning and timing prediction models, clustering and predicting the power remote sensing data of highway camera equipment, the window period problem of traditional monitoring equipment status monitoring is solved, real-time monitoring and early warning of equipment status is realized, and safety is improved and efficiency.

CN114676930BActive Publication Date: 2024-11-22浙江永基智能科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210371744.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-11
Publication Date
2024-11-22
Estimated Expiration
2042-04-11

AI Technical Summary

Technical Problem

The status monitoring of traditional highway monitoring equipment relies on regular inspections, which leads to a monitoring window period when equipment failures, increasing safety hazards and monitoring costs. How to achieve real-time monitoring and early warning of equipment status has become an urgent problem.

Method used

Using a time series prediction model based on unsupervised learning, the power remote sensing data is classified through the K-mean clustering algorithm, and the prediction model is constructed with the quadratic exponential smoothing method, and the absolute error and average absolute error are calculated, the equipment status is judged and the equipment is early discovered. Aging failure.

Benefits of technology

It improves the accuracy of equipment status prediction, reduces the monitoring window period, ensures timely maintenance of highway equipment, and reduces monitoring costs and safety hazards.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114676930B_ABST
    Figure CN114676930B_ABST
Patent Text Reader

Abstract

The present invention provides a multi-index highway camera equipment state detection and fault warning method, and the specific steps are as follows: step S1, collecting power telemetry data of the highway camera equipment during its entire life cycle when it is working; step S2, classifying the collected power remote sensing data through a K-means clustering algorithm to obtain cluster divisions of all data; step S3, for each part in the cluster division, constructing a corresponding highway camera equipment power prediction model based on an exponential smoothing method, inputting power remote sensing data, and obtaining corresponding prediction data; step S4, according to the collected power remote sensing data and the predicted data, the absolute error of the two and the average absolute error within a period of time can be obtained, and compared with the set corresponding absolute error and average absolute error thresholds; step S5, according to the comparison result, judging the state of the equipment and whether the equipment is faulty and aging.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of highway equipment detection, and in particular relates to a multi-index highway camera equipment state detection and fault warning method based on a time series prediction model integrated with unsupervised learning. Background Art

[0002] Nowadays, my country's expressways have formed a complex expressway traffic network, which has promoted rapid economic development on the one hand, but also greatly increased the frequency of traffic accidents on the other hand. In order to effectively manage the expressways, the most common way adopted by the transportation department is to use a large number of monitoring equipment to monitor the highways in real time, so as to grasp the status of the expressways at any time and provide accurate road information and rescue for those in need. Therefore, ensuring the normal operation of the monitoring equipment is a prerequisite for ensuring the above functions. Traditional monitoring equipment status monitoring is achieved through regular inspections and inspections by maintenance personnel, which greatly increases the cost and efficiency of equipment monitoring, and the equipment can only be discovered by maintenance personnel when it fails. During this period, there will inevitably be a period of monitoring gaps, causing certain safety hazards. Therefore, how to monitor the status of monitoring equipment and give early warnings has become a problem that needs to be solved urgently. Summary of the invention

[0003] In response to the problems existing in the above background technology introduction, the purpose of the present invention is to provide a multi-indicator highway camera equipment status detection and fault warning method based on a time series prediction model that integrates unsupervised learning. By analyzing the collectible data to analyze the current status of the equipment, and using past historical data, it detects equipment failures caused by chronic problems such as aging, and issues an alarm before the equipment is damaged, so that maintenance personnel can replace the equipment in time.

[0004] The technical solution adopted by the present invention is:

[0005] A multi-index highway camera equipment state detection and fault warning method, the specific steps of which are as follows:

[0006] Step S1, collecting power telemetry data of the highway camera equipment during its entire life cycle;

[0007] Step S2: classify the collected power remote sensing data using a K-means clustering algorithm to obtain cluster divisions for all data;

[0008] Step S3: for each part in the cluster division, construct a corresponding highway camera equipment power prediction model based on the exponential smoothing method, input the power remote sensing data, and obtain the corresponding prediction data;

[0009] Step S4: According to the collected power remote sensing data and the predicted data, the absolute error between the two and the average absolute error within a period of time can be obtained, and compared with the set corresponding absolute error and average absolute error thresholds;

[0010] Step S5: According to the comparison result, determine the state of the device and whether the device is faulty or aged.

[0011] Further, step S2 specifically includes the following steps:

[0012] Step S21, selecting appropriate data features and cluster tree k value as the basis for clustering based on prior experience of data;

[0013] Step S22: If the amount of power remote sensing data is relatively large, a portion of the data may be randomly selected as a sample set S = {p1, p2, ... p m}, where p i is the i-th selected random sample, m is the number of random samples; the cluster tree is k, and the output is k cluster centers {μ1,μ2,…,μ k}, and the corresponding cluster partition C = {C1, C2, ... C k}, where μ i is the i-th cluster center vector, C i is the set of samples classified by the i-th cluster center, and k is the number of cluster centers;

[0014] Step S23: Randomly select a sample from the sample set in step S22 as the first cluster center μ1, and compare each point in the sample set with p i , calculate its distance to the nearest cluster center among the selected cluster centers in for Take the minimum value, For p i -μ r The square of the 2-norm, k selected is the number of cluster centers that have been selected; then a new data point in the sample set S is selected as the new cluster center, which needs to meet the following conditions: d(p i ) is larger, p i The greater the probability of being selected as a new cluster center; repeat the above operation until k cluster centers are selected and the cluster center initialization is completed. The k cluster centers are {μ1,μ2,…,μ k};

[0015] Step S24: Initialize the cluster partition C For i=1,2…m, calculate the sample point p i and each cluster center μ j(j=1,2,…k) distance: The p i Use d ij The cluster center μ corresponding to the minimum distance j Classify and then update the cluster division C j =C j ∪{p i}; After all sample points are classified, for j = 1, 2, ..., k, the clusters are divided into C j All sample points in the formula Obtain new cluster centers, where |C j |Partition C into clusters j The number of samples in the cluster; repeat the above operation until the values ​​of the k cluster centers no longer change, and output k cluster centers: {μ1,μ2,…,μ k};

[0016] Step S25: cluster all collected power remote sensing data using k cluster centers to obtain cluster divisions C = {C1, C2, ... C k}.

[0017] Furthermore, the specific steps of establishing the prediction model in step S3 include:

[0018] Step S31, select smoothing coefficient α and quadratic exponential smoothing value and The initial value of

[0019] Step S32, the calculation formula of the prediction model of the quadratic exponential smoothing method is:

[0020]

[0021]

[0022]

[0023]

[0024] F t+T =a t +b t T

[0025] In the formula, is the exponential smoothing value of the tth period, α is the smoothing coefficient, x t is the collected data of period t, is the quadratic exponential smoothing value of the t period, F t+T is the predicted value for the t+T period, where T is the number of periods ahead of the prediction;

[0026] Step S33: input the collected power remote sensing data to obtain corresponding prediction data.

[0027] Furthermore, the absolute error between the power remote sensing data collected in step S4 and the predicted data is

[0028] The mean absolute error (MAE) over a period of time is calculated as follows:

[0029] Where m is the number of samples in a period of time, p i is the true value of the data, is the predicted value.

[0030] Further, the step of determining the device status in step S5 includes:

[0031] The absolute error err of the current data point cur The absolute error err of the previous data point pre Set different thresholds to determine the status of a data point on the device;

[0032] Set a threshold for the mean absolute error over a period of time to determine the status of the device during this period.

[0033] Compared with the prior art, the present invention has the following significant advantages:

[0034] (1) Unsupervised learning is used to cluster the collected power remote sensing data. Compared with directly using the prediction model to predict the data, the power remote sensing data is first clustered and divided into different data sets under different equipment states, and then the prediction model is used for prediction. This can greatly improve the prediction accuracy of the model near the power turning point. Moreover, compared with other classification algorithms, the K-means clustering algorithm converges quickly, does not require training sets, has better clustering effects, and only needs to adjust a parameter cluster tree k to change the classification mode.

[0035] (2) The quadratic exponential smoothing method is used to construct a time series prediction model. Compared with the prediction model constructed by the single exponential smoothing method, the model of the present invention can learn the trend of power data. When the power data increases or decreases, the model can capture this information in time, thereby improving the prediction accuracy of the model. Compared with other prediction models, the model of the present invention has a fast learning speed, does not require long-term training, has a good prediction effect, and only needs to adjust the smoothing parameter α to adjust the prediction model.

[0036] (3) By using multiple indicators to detect the status of the equipment, compared with a single indicator, the present invention can more accurately determine the status of the equipment and predict the aging of the equipment. It can also issue an early warning before the equipment is completely damaged, allowing maintenance personnel to replace the equipment in time to ensure the safety of the highway. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the process of the present invention.

[0038] Figure 2 It is a schematic diagram of the present invention using the absolute error threshold between the predicted value and the true value to judge the device state.

[0039] Figure 3 It is a schematic diagram of the present invention using the threshold of the mean absolute error (MAE) between the predicted value and the true value within 5000 sample points to judge the normal operation of the equipment.

[0040] Figure 4 It is a schematic diagram of the present invention using the threshold of the mean absolute error (MAE) between the predicted value and the true value within 5000 sample points to judge the chronic damage problem of the equipment. DETAILED DESCRIPTION

[0041] The present invention is further described below in conjunction with specific embodiments, but the present invention is not limited to these specific embodiments. Those skilled in the art should recognize that the present invention covers all possible alternatives, improvements and equivalents within the scope of the claims.

[0042] See also Figure 1 This embodiment provides a multi-index highway camera device state detection and fault warning method. The camera device is a photosensitive fill light on a highway as an example. The specific steps are as follows:

[0043] Step S1, collecting power telemetry data of the highway camera equipment during its entire life cycle;

[0044] Step S2, classifying the collected power remote sensing data by using a K-means clustering algorithm;

[0045] The details are as follows:

[0046] Step S21, based on prior experience of data, the power value of the data can be selected as the data feature of clustering. It can be seen from observation that the power values ​​of all collected data can be roughly divided into two categories: working state and sleeping state, so the cluster tree k is set to 2;

[0047] Step S22: The amount of power remote sensing data collected from the fill light is relatively large, with a total of 56,389 sampled data. Therefore, 4,000 data are randomly selected as the sample set S = {p1, p2, ... p 4000}, the cluster tree of clustering is 2, and the output is 2 cluster centers {μ1,μ2} and cluster partition C = {C1,C2};

[0048] Step S23: Randomly select a sample from the sample set in step S22 as the first cluster center μ1. For each point p in the sample set i , calculate its distance to the nearest cluster center among the selected cluster centers in for Take the minimum value, For p i -μ1 squared 2-norm. Then select a new data point in the sample set S as the new cluster center, which needs to meet the following conditions: d(p i ) is larger, p i The greater the probability of being selected as the new cluster center. After completing the initialization of the cluster centers, we get two cluster centers {μ1,μ2};

[0049] Step S24: Initialize the cluster partition C For i = 1, 2…4000, calculate the sample point p i and each cluster center μ j The distance of (j=1,2): x i Use d ij The cluster center μ corresponding to the minimum distance j Perform classification and then update the cluster division C j =C j ∪{p i After classifying the 4000 sample points, for j = 1, 2, the clusters are divided into C j All sample points in the formula Obtain new cluster centers, where |C j |Partition C into clusters j Repeat the above operation until the values ​​of the two cluster centers no longer change, and output two cluster centers: {μ1,μ2};

[0050] Step S25: cluster all collected power remote sensing data using two cluster centers to obtain cluster division C = {C1, C2} for all data.

[0051] Step S3: for each part in the cluster division, construct a corresponding highway camera equipment power prediction model based on the exponential smoothing method, input the power remote sensing data, and obtain the corresponding prediction value;

[0052] Specifically include:

[0053] Step S31, select smoothing coefficient α and quadratic exponential smoothing value and Initial value of: Because under normal circumstances, the power change of the fill light in two different states is stable, so a smaller smoothing coefficient value should be selected, here α=0.0005. The amount of data collected is relatively large, and the quadratic exponential smoothing value and The initial value of is the first data value;

[0054] Step S32, the calculation formula of the prediction model of the quadratic exponential smoothing method is:

[0055]

[0056]

[0057]

[0058]

[0059] F t+1 =a t +b t

[0060] In the formula, is the exponential smoothing value of the tth period, α is the smoothing coefficient, x t is the collected data of period t, is the quadratic exponential smoothing value of the tth period, F t+1 is the predicted value for the t+1th period, and T=1 is the number of periods ahead of the prediction;

[0061] Step S33: input the collected power remote sensing data to obtain corresponding prediction data.

[0062] Step S4: According to the collected power remote sensing data and the predicted data, the absolute error between the two can be obtained. and the mean absolute error (MAE) within 5000 sample points, where Where p i is the true value of the data, is the predicted value; and is compared with the set corresponding absolute error and mean absolute error thresholds;

[0063] Step S5: According to the comparison result, determine the state of the device and whether the device is faulty or aged.

[0064] Specifically, the absolute error err of the current data point cur The absolute error err with the previous data point pre Set different thresholds to determine the status of the data point on the device, as shown in Table 1 below;

[0065] Table 1

[0066]

[0067] According to the absolute error of different data points, that is Figure 2 The dotted line data in the figure are used to judge each real data point. The threshold value set in Table 1 can be used to divide the real data into three states: normal operation, switching lights and power mutation. The point o in the figure is the power mutation point, and the point x is the switching light state point.

[0068] The threshold is set for the mean absolute error of 5000 samples to determine the status of the device during this period, as shown in Table 2 below;

[0069] Table 2

[0070] Photosensitive fill light device status MAE Normal operation <0.8 Aging damage alarm >0.9

[0071] according to Figure 3 As shown in the figure, it can be seen that when the equipment is operating normally, the average absolute error of the samples is less than 0.8. Figure 4 As shown in the figure, it can be seen that when the device begins to age, its mean absolute error rises sharply over a period of time, and the peak value is higher than 0.9. Therefore, Table 2 can be used to set the threshold of the mean absolute error to determine whether the device is aged.

Claims

1. A multi-index highway camera equipment status detection and fault warning method, the specific steps of which are as follows: Step S1, collecting power telemetry data of the highway camera equipment during its entire life cycle; Step S2: classify the collected power remote sensing data using the K-means clustering algorithm to obtain cluster divisions for all data; the specific steps of step S2 include: Step S21, selecting appropriate data features and cluster tree k value as the basis for clustering based on prior experience of data; Step S22: If the amount of power remote sensing data is relatively large, a portion of the data may be randomly selected as a sample set S = {p1, p2, ... p m}, where p i is the i-th selected random sample, m is the number of random samples; the cluster tree is k, and the output is k cluster centers {μ1, μ2, ..., μ k}, and the corresponding cluster partition C = {C1, C2, ...C k}, where μ i is the i-th cluster center vector, C i is the set of samples classified by the i-th cluster center, and k is the number of cluster centers; Step S23: Randomly select a sample from the sample set in step S22 as the first cluster center μ1, and compare each point in the sample set with p i , calculate its distance to the nearest cluster center among the selected cluster centers r=1,2,...k selected ,in for Take the minimum value, For p i -μ r The square of the 2-norm, k selected is the number of cluster centers that have been selected; then a new data point in the sample set S is selected as the new cluster center, which needs to meet the following conditions: d(p i ) is larger, p i The greater the probability of being selected as a new cluster center; repeat the above operation until k cluster centers are selected, the cluster center initialization is completed, and the k cluster centers are {μ1, μ2, ..., μ k}; Step S24: Initialize the cluster partition C t=1,2...k, for i=1,2...m, calculate the sample point p i and each cluster center μ j (j=1,2,...k) distance: The p iUse d ij The cluster center μ corresponding to the minimum distance j Perform classification and then update the cluster division C j =C j ∪{p i}; After all sample points are classified, for j = 1, 2, ..., k, the clusters are divided into C j All sample points in the formula Obtain new cluster centers, where |C j |Partition C into clusters j The number of samples in the cluster; repeat the above operation until the values ​​of the k cluster centers no longer change, and output k cluster centers: {μ1, μ2, ..., μ k}; Step S25: cluster all collected power remote sensing data using k cluster centers to obtain cluster divisions C = {C1, C2, ...C k}; Step S3: for each part in the cluster division, construct a corresponding highway camera equipment power prediction model based on the exponential smoothing method, input the power remote sensing data, and obtain the corresponding prediction data; The specific steps of establishing the prediction model in step S3 include: Step S31, select smoothing coefficient α and quadratic exponential smoothing value and The initial value of Step S32, the calculation formula of the prediction model of the quadratic exponential smoothing method is: F t+T =a t +b t T In the formula, is the exponential smoothing value of the tth period, α is the smoothing coefficient, x t is the collected data of period t, is the quadratic exponential smoothing value of the t period, F t+T is the predicted value for the t+T period, where T is the number of periods ahead of the prediction; Step S33, input the collected power remote sensing data to obtain corresponding prediction data; Step S4: According to the collected power remote sensing data and the predicted data, the absolute error and the average absolute error within a period of time can be obtained, and compared with the set corresponding absolute error and average absolute error thresholds; the absolute error of the power remote sensing data collected in step S4 and the predicted data obtained by prediction The mean absolute error (MAE) over a period of time is calculated as follows: Where m is the number of samples in a period of time, pi is the true value of the data, is the predicted value; Step S5: judging the state of the device and whether the device is faulty or aging according to the comparison result; The steps of determining the device status include: The absolute error err of the current data point cur The absolute error err with the previous data point pre Set different thresholds to determine the status of a data point on the device; Set a threshold for the mean absolute error over a period of time to determine the status of the device during this period.