Monitoring method and device suitable for inspection data, terminal equipment and storage medium

Abstract

The invention discloses a monitoring method and device suitable for inspection data, terminal equipment and a storage medium, and the method comprises the steps: extracting a plurality of equipment features from each piece of inspection data, taking the feature values of the equipment features as the feature values corresponding to data points, carrying out the clustering analysis of each data point, obtaining a plurality of different types of data groups, and carrying out the clustering analysis of the data groups. And further performing density fitting on a clustering result to obtain a feature distribution interval of each data group, thereby accurately identifying abnormal data points through distance comparison or range comparison between the feature values of the data points and the feature distribution intervals, and improving the accuracy of data point identification. As the feature distribution interval obtained by fitting considers the current overall distribution condition of the data, the real condition of the data can be accurately reflected, and the limitation of simple threshold judgment is avoided. Through clustering analysis and density fitting, the abnormal data points can be accurately identified, and timely and accurate early warning can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of processing inspection data, and in particular to a monitoring method, device, terminal equipment and storage medium suitable for inspection data. Background Art

[0002] Inspection data generally refers to a collection of various types of data related to inspection activities, which usually comes from various equipment and tools used in the inspection process of various production areas, such as inspection vehicles, drones, sensors, etc. By monitoring inspection data, potential failures of equipment or systems can be discovered in a timely manner, and early warnings can be given based on abnormal detection, so that preventive measures can be taken to avoid failures. For example, inspection data contains a lot of safety-related information, such as the operating status of equipment, environmental conditions, etc. By monitoring these data, safety hazards can be discovered in a timely manner, and corresponding safety measures can be taken to ensure the safety of equipment and personnel.

[0003] However, traditional data monitoring technology usually relies on manual monitoring and analysis of inspection data, which is not only time-consuming and labor-intensive, but also prone to errors. Moreover, it often uses simple thresholds to judge whether the data during the inspection process is abnormal. Due to the complexity of the equipment operation status and environmental conditions, the data may be affected by a variety of factors and fluctuate, and the fixed threshold judgment may not accurately reflect the true situation of the data, resulting in high false alarm and missed alarm rates, making it impossible to achieve timely and accurate warnings when equipment abnormalities occur. Summary of the invention

[0004] The embodiments of the present invention provide a monitoring method, apparatus, terminal device and storage medium suitable for inspection data. Through cluster analysis and density fitting, abnormal data points can be accurately identified, and timely and accurate early warnings can be achieved. This can effectively solve the problem in the prior art that the use of fixed threshold judgments may not accurately reflect the true situation of the data, resulting in high false alarm and missed alarm rates.

[0005] An embodiment of the present invention provides a monitoring method applicable to inspection data, comprising:

[0006] Obtain inspection data corresponding to each device in the target area;

[0007] For each device, a plurality of different types of device features are extracted based on the inspection data of the device; wherein each device feature corresponds to a feature value;

[0008] The characteristic value of each device characteristic is used as the characteristic value corresponding to each data point; cluster analysis is performed on each data point according to the characteristic value of each data point to generate several different types of data groups;

[0009] For each data group, density fitting is performed on the eigenvalues ​​corresponding to each data point in the data group to generate the characteristic distribution interval corresponding to the data group;

[0010] For each data set, the data points whose characteristic values ​​are not within the characteristic distribution interval are marked as the first abnormal data points;

[0011] According to the first target device corresponding to each first abnormal data point, first warning information including the first target device is generated.

[0012] Preferably, the inspection data includes: appearance image data corresponding to the equipment;

[0013] According to the inspection data of the equipment, several different types of equipment features are extracted, including:

[0014] The noise in the appearance image data is removed by using Gaussian filtering technology to generate preprocessed appearance image data;

[0015] According to the preset image feature extraction algorithm, the image features in the preprocessed appearance image data are extracted to generate edge features, shape features and texture features corresponding to the device.

[0016] Preferably, the inspection data includes: operation data corresponding to the equipment; the operation data includes: voltage, current, temperature and noise;

[0017] According to the inspection data of the equipment, several different types of equipment features are extracted, including:

[0018] The voltage characteristics, current characteristics, temperature characteristics and noise characteristics corresponding to the device are extracted from the operating data corresponding to the device.

[0019] Preferably, performing density fitting on the characteristic values ​​corresponding to each data point in the data group to generate the characteristic distribution interval corresponding to the data group includes:

[0020] According to each characteristic value in the data group, draw the histogram corresponding to the data group;

[0021] Fitting the frequency distribution of each eigenvalue in the histogram according to a preset probability density function to generate a distribution model; wherein the distribution model corresponds to a statistical indicator for characterizing the distribution characteristics of the eigenvalue;

[0022] According to the distribution model and the statistical indicators, the characteristic distribution interval corresponding to the data group is calculated.

[0023] Preferably, it also includes:

[0024] Mark the data points whose characteristic values ​​are within the characteristic distribution interval as the first normal data points;

[0025] Marking a data group where the total number of first normal data points is greater than a preset number threshold as a data group to be processed;

[0026] For each data group to be processed, the data group to be processed is input into a preset binary classification model, so that the binary classification model maps the feature value of each data point in the data group to be processed into the feature space, and generates the distance between each data point and the hyperplane according to the position of each data point in the feature space and the position of the hyperplane; the data points whose distance is less than the preset distance threshold are marked as second normal data points, and the position of each data point in the feature space and the position of the hyperplane are output;

[0027] According to the position of each data point in the feature space and the position of the hyperplane, a feature space plane diagram including a number of data points and the hyperplane is drawn;

[0028] According to the distance between each second normal data point and the hyperplane, and the total number of second normal data points, an average distance value is obtained;

[0029] Based on the feature space plane graph, a boundary region in the feature space plane graph is generated with the center of the hyperplane as the center of the circle and the average distance value as the radius;

[0030] Marking all data points in the feature space plane that do not fall within the boundary area as second abnormal data points;

[0031] generating second warning information including the second target device according to the second target device corresponding to each second abnormal data point;

[0032] The training process of the binary classification model includes:

[0033] Acquire a plurality of data group samples; wherein each data group sample includes: a plurality of normal data points and a plurality of abnormal data points;

[0034] Taking a number of data group samples as input and the predicted position of the hyperplane in the binary classification model to be trained as output, the binary classification model to be trained is iteratively trained until the interval distance between the position of the support vector in the feature space and the predicted position of the hyperplane is the largest, thereby generating a preset binary classification model;

[0035] The support vector is the normal data point closest to the predicted position of the hyperplane.

[0036] Preferably, when it is determined that the inspection data is appearance image data corresponding to each device, the binary classification model maps the feature value of each data point in the data group to be processed to the feature space, including:

[0037] The binary classification model maps the feature value of each data point in the data set to be processed to a high-dimensional feature space through a kernel function.

[0038] Preferably, the cluster analysis is performed on each data point according to the characteristic value of each data point to generate several different types of data groups, including:

[0039] Repeat the following clustering operation. When the difference between the total value of each updated cluster center and the total value of each cluster center during the last clustering operation is less than the difference threshold, output the current data groups of different types:

[0040] Calculate the distance from the feature value of each data point to each current cluster center, classify each data point to the current cluster center corresponding to the minimum distance, and generate the current data group; wherein, initially randomly select the cluster centers corresponding to the preset number as the current cluster centers; each data group corresponds to a type;

[0041] For each data group, generating an updated cluster center of the data group according to the weighted average value corresponding to the characteristic value of each data point in the data group;

[0042] When the difference between the total value of each updated cluster center and the total value of each cluster center when the cluster operation was last performed is determined to be not less than the difference threshold, the updated cluster center is used as the current cluster center when the cluster operation is performed next time. Based on the above method embodiment, the present invention provides a corresponding device embodiment.

[0043] An embodiment of the present invention provides a monitoring device suitable for inspection data, including: an inspection data acquisition module, an equipment feature extraction module, a clustering module, a feature distribution generation module and an early warning module;

[0044] The inspection data acquisition module is used to acquire the inspection data corresponding to each device in the target area;

[0045] The device feature extraction module is used to extract a number of different types of device features for each device based on the inspection data of the device; wherein each device feature corresponds to a feature value;

[0046] The clustering module is used to use the characteristic value of each device characteristic as the characteristic value corresponding to each data point; perform cluster analysis on each data point according to the characteristic value of each data point to generate a number of different types of data groups;

[0047] The feature distribution generating module is used to perform density fitting on the feature values ​​corresponding to each data point in the data group for each data group, so as to generate a feature distribution interval corresponding to the data group;

[0048] The early warning module is used to mark, for each data group, data points whose characteristic values ​​are not within the characteristic distribution interval as first abnormal data points; and generate first early warning information including the first target device according to the first target device corresponding to each first abnormal data point.

[0049] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0050] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, a monitoring method suitable for patrol data described in the above-mentioned embodiment of the invention is implemented.

[0051] Based on the above method embodiments, the present invention provides corresponding storage medium item embodiments.

[0052] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a monitoring method suitable for patrol data described in the above-mentioned embodiment of the invention.

[0053] The following beneficial effects are achieved by implementing the present invention:

[0054] The embodiment of the present invention provides a monitoring method, apparatus, terminal device and storage medium suitable for inspection data. After the inspection data corresponding to each device in the target area is obtained, several different types of device features corresponding to each inspection data can be extracted, so that the characteristic value of each device feature is used as the characteristic value corresponding to each data point, and cluster analysis is performed on each data point to generate several different types of data groups. Then, through cluster analysis, device data points with similar characteristic values ​​can be classified into one category to form different data groups, thereby realizing automatic clustering and combination of inspection data. In addition, based on the clustering results, the present invention can generate a characteristic distribution interval corresponding to each data group through density fitting, so that abnormal data points can be accurately identified through distance comparison or range comparison between the characteristic value of the data point and the characteristic distribution interval. Since the characteristic distribution interval obtained by fitting takes into account the current overall distribution of the data, it can accurately reflect the actual situation of the data, and compares with each characteristic value based on the current overall distribution, which can avoid the limitations of simple threshold judgment. The present invention can more accurately find abnormal data points than the traditional technology that only uses simple threshold judgment. Even if the data fluctuates due to the influence of various factors, the abnormal data points can be accurately identified through cluster analysis and density fitting, and timely and accurate early warning can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of a monitoring method applicable to inspection data provided by an embodiment of the present invention.

[0056] Figure 2 It is a schematic diagram of the structure of a monitoring device suitable for patrol inspection data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] like Figure 1 FIG. 1 is a flow chart of a monitoring method for inspection data provided by an embodiment of the present invention, wherein the monitoring method for inspection data includes:

[0059] Step S1: Obtain inspection data corresponding to each device in the target area;

[0060] Step S2: for each device, extracting a number of different types of device features according to the inspection data of the device; wherein each device feature corresponds to a feature value;

[0061] Step S3: taking the characteristic value of each device characteristic as the characteristic value corresponding to each data point; performing cluster analysis on each data point according to the characteristic value of each data point to generate several different types of data groups;

[0062] Step S4: For each data group, density fitting is performed on the characteristic values ​​corresponding to each data point in the data group to generate a characteristic distribution interval corresponding to the data group;

[0063] Step S5: for each data group, marking the data points whose characteristic values ​​are not within the characteristic distribution interval as the first abnormal data points;

[0064] According to the first target device corresponding to each first abnormal data point, first warning information including the first target device is generated.

[0065] For step S1, in a preferred embodiment, the present invention can process the inspection data in different areas to identify devices with abnormal data. When obtaining the inspection data of the target area, the target area can be flexibly referred to according to the actual application scenario. For example, in the power industry, the target area can refer to a specific power grid area, substation, power plant or even a specific transmission line area, and the selection of the above areas is usually based on factors such as monitoring needs, importance and potential risks.

[0066] By carefully monitoring the designated target area, potential problems or abnormal situations can be discovered in a timely manner, so that appropriate measures can be taken to prevent or repair them.

[0067] In the power industry, the "equipment" related to the inspection data monitoring method usually includes various power equipment, such as transformers, circuit breakers, transmission lines, generators, etc. The above equipment is a key component of the power system, and its operating status directly affects the stability and safety of the entire system.

[0068] Inspection data is usually obtained through drone photography and various sensors and detection instruments carried by drones. It can contain various status information of the equipment, such as operating data or appearance data.

[0069] In a preferred embodiment, the inspection data may be: appearance image data corresponding to the equipment;

[0070] In a preferred embodiment, the inspection data may be: operating data corresponding to the equipment; the operating data includes but is not limited to: voltage, current, temperature and noise.

[0071] It is understandable that through the appearance image data of the inspection, it is possible to identify appearance defects such as cracks, corrosion, dirt, and oil leakage on the surface of the equipment, which are often early signs of equipment aging or failure. Timely detection helps prevent serious accidents. Through image analysis, the overall status of the equipment can also be evaluated, such as the looseness of fasteners and the integrity of connectors, so as to determine whether the equipment needs maintenance or repair.

[0072] Operation data such as voltage, current, and temperature can monitor the operating status of the equipment in real time, so that abnormal conditions can be detected in time. For example, abnormal current fluctuations may indicate that the equipment has a short circuit or overload problem; excessive temperature may indicate that the equipment is overheating or has poor heat dissipation. By analyzing the operation data, the type and time of possible equipment failures can be predicted, and early warnings can be issued in time to avoid downtime or accidents caused by failures.

[0073] For step S2, in a preferred embodiment, before clustering the inspection data of each device, the present invention can first extract several different types of device features from the inspection data of the device; wherein each device feature corresponds to a feature value.

[0074] In a preferred embodiment, when the inspection data is appearance image data corresponding to the device, the process of extracting several different types of device features includes:

[0075] The noise in the appearance image data is removed by using Gaussian filtering technology to generate preprocessed appearance image data;

[0076] According to the preset image feature extraction algorithm, the image features in the preprocessed appearance image data are extracted to generate edge features, shape features and texture features corresponding to the device.

[0077] In a preferred embodiment, the inspection data is: operation data corresponding to the equipment; and the operation data includes: voltage, current, temperature and noise;

[0078] The process of extracting several different types of device features includes:

[0079] The voltage characteristics, current characteristics, temperature characteristics and noise characteristics corresponding to the device are extracted from the operating data corresponding to the device.

[0080] Schematically, for appearance image feature extraction, by extracting edge features, the outline and shape of the device edge can be obtained to identify the integrity, damage or deformation of the device. By extracting shape features, the overall shape and size of the device can be analyzed to determine whether the device is installed correctly or has abnormal deformation. By extracting texture features, texture changes on the surface of the device, such as cracks, corrosion or dirt, can be detected, so as to timely discover potential faults.

[0081] Therefore, by extracting and clustering these features, abnormal images (i.e., abnormal data points) can be automatically screened out, reducing the workload of manual review and improving inspection efficiency.

[0082] For operation data feature extraction, the electrical performance of the equipment can be reflected by extracting voltage features. For example, abnormal voltage fluctuations may indicate that the equipment is faulty or overloaded.

[0083] By extracting the current characteristics, the current changes can be monitored, which helps to detect problems such as circuit short circuit, overload or leakage. By extracting the temperature characteristics, the temperature of the device can be detected. When the temperature of the device rises abnormally, it may mean that the device is overheating or has poor heat dissipation, and timely measures need to be taken. By extracting the noise characteristics, the noise of the device can be detected, and abnormal noise may indicate mechanical failure or wear inside the device.

[0084] Therefore, by extracting the above features, when clustering and analyzing these feature values, it is possible to automatically identify devices with abnormal operation, thereby improving the accuracy of fault detection.

[0085] For step S3, in a preferred embodiment, cluster analysis is performed on each data point according to the characteristic value of each data point to generate several different types of data groups, including:

[0086] Repeat the following clustering operation. When the difference between the total value of each updated cluster center and the total value of each cluster center during the last clustering operation is less than the difference threshold, output the current data groups of different types:

[0087] Calculate the distance from the feature value of each data point to each current cluster center, classify each data point to the current cluster center corresponding to the minimum distance, and generate the current data group; wherein, initially randomly select the cluster centers corresponding to the preset number as the current cluster centers; each data group corresponds to a type;

[0088] For each data group, generating an updated cluster center of the data group according to the weighted average value corresponding to the characteristic value of each data point in the data group;

[0089] When it is determined that the difference between the total value of each updated cluster center and the total value of each cluster center when the clustering operation was performed last time is not less than the difference threshold, the updated cluster center is used as the current cluster center when the clustering operation is performed next time.

[0090] In a preferred embodiment, the present invention performs cluster analysis on each data group through the K-means algorithm, and then:

[0091] Initialization: Randomly select K data points as the initial cluster centers;

[0092] Assignment: For each data point, calculate its distance to the K cluster centers and assign it to the cluster with the closest distance;

[0093] Update: Recalculate the cluster center of each cluster, that is, the mean of all data points in the cluster;

[0094] Iteration: Repeat the assignment step and the update step until the cluster center no longer changes or the preset number of iterations is reached.

[0095] Illustratively, the present invention can discover potential patterns and rules in the data, such as the normal state range of the equipment, the characteristics of the abnormal state, etc., by automatically performing cluster analysis on the inspection data.

[0096] Therefore, through cluster analysis, data points can be divided into several different types of data groups according to the characteristic values ​​of the device characteristics, so that the actual status of the device can be more accurately reflected based on the similarity between the data points. Compared with traditional manual classification or classification methods based on simple rules, the present invention can automatically discover potential patterns and structures in the data through cluster analysis, thereby providing more accurate classification results.

[0097] For step S4, in a preferred embodiment, performing density fitting on the characteristic values ​​corresponding to each data point in the data group to generate the characteristic distribution interval corresponding to the data group includes:

[0098] According to each characteristic value in the data group, draw the histogram corresponding to the data group;

[0099] Fitting the frequency distribution of each eigenvalue in the histogram according to a preset probability density function to generate a distribution model; wherein the distribution model corresponds to a statistical indicator for characterizing the distribution characteristics of the eigenvalue;

[0100] According to the distribution model and the statistical indicators, the characteristic distribution interval corresponding to the data group is calculated.

[0101] Schematically, when a data set contains a certain characteristic value (such as temperature) of 100 devices, the process of generating the characteristic distribution interval corresponding to this data set is as follows:

[0102] First, a histogram is drawn based on each eigenvalue in the data set. The temperature eigenvalues ​​can be divided into 10 intervals (for example, from 20°C to 30°C, 1°C in each interval). Then, the number of devices in each interval is counted, and a histogram containing 10 columns is obtained. The height of each column represents the number of devices in the corresponding interval.

[0103] Next, the frequency distribution of each eigenvalue in the histogram is fitted using a preset probability density function. It is understandable that a probability density function is a mathematical model for describing the probability of a random variable taking a value. In this example, a Gaussian distribution (normal distribution) can be selected as a probability density function.

[0104] In order to fit a Gaussian distribution, it is necessary to calculate the mean (μ) and standard deviation (σ) of the histogram. The mean (μ) is the sum of all eigenvalues ​​divided by the number of eigenvalues; the standard deviation (σ) is the square root of the average of the squares of the differences between the eigenvalues ​​and the mean.

[0105] With the mean and standard deviation, we can draw a Gaussian distribution curve and overlay it with the histogram to observe the fitting effect. Finally, the characteristic distribution interval corresponding to the data set is calculated based on the distribution model (Gaussian distribution in this example) and statistical indicators (such as mean and standard deviation). The characteristic distribution interval can be one or more continuous intervals used to describe the range of characteristic values.

[0106] In this example, you can use the mean plus or minus a number of standard deviations to define the feature distribution interval. For example, you can choose the mean plus or minus 1 standard deviation, 2 standard deviations, or 3 standard deviations as the boundaries of the feature distribution interval. In this way, you get one or more intervals containing different numbers of devices, which reflect the distribution of the feature values.

[0107] The characteristic distribution interval is calculated based on the characteristic value of the entire data set, which takes into account all the information in the data, including the central tendency, dispersion degree and distribution form of the data. Therefore, it can more comprehensively reflect the real situation of the data.

[0108] As data continues to accumulate, the feature distribution interval will also change accordingly, so that it can adapt to changes in data and always maintain an accurate reflection of the true situation of the data. Simple threshold judgments are often based on fixed numerical limits, ignoring individual differences in the data. The feature distribution interval can take into account the overall distribution of the data and respond to individual differences more flexibly. When the data distribution changes, the simple threshold judgment may not be adjusted in time, resulting in a decrease in the accuracy of the judgment result. The present invention, by generating a feature distribution interval, can be updated as the data changes, maintaining the accuracy and adaptability of the judgment. That is, the feature distribution interval can set boundaries based on the overall distribution of the data, reducing the risk of misjudgment.

[0109] Schematically, after K-means clustering, for each data group, the data points in the data group can be further processed and the corresponding feature distribution can be generated. Specifically, the key features of the data group (such as current, voltage, temperature, etc.) can be collected to calculate the statistical indicators of each feature, including mean, variance, skewness and kurtosis. Schematically, these statistical indicators depict the overall distribution of the data. Next, the quantiles of each feature (such as 25%, 50%, 75% quantile values) are also calculated to establish the interval range of the feature distribution. On this basis, a multidimensional density estimation algorithm (such as kernel density estimation) can also be used to further fit the density distribution of the feature values ​​to generate a specific distribution model of the data in the feature space. Taking the current feature as an example, the current value of a certain cluster may be concentrated in the range of [100A, 150A], while the abnormal points may be significantly offset outside this range.

[0110] For step S5, in a preferred embodiment, in the temperature monitoring scenario of the device, the characteristic distribution interval obtained by fitting can be used to determine whether the device temperature is normal. If the temperature of the device falls within the characteristic distribution interval, the temperature of the device is considered normal; if it exceeds the interval, it is considered that the device may be abnormal or faulty. This method is more accurate and reliable than simple threshold judgment.

[0111] By using the characteristic distribution interval, abnormal data points that are inconsistent with the overall data distribution can be quickly identified. These abnormal data points often represent abnormal conditions of the device or system, so the present invention can discover and handle them in a timely manner.

[0112] Moreover, for each identified abnormal data point, the corresponding device can be marked as a target device, thereby generating early warning information including the first target device. Schematically, the early warning information can include information such as the name of the device, the characteristics of the abnormal data, and the time when the abnormality occurred.

[0113] In a preferred embodiment, the present invention can further refine the identification process of abnormal data points through the SVM algorithm model. Especially when the abnormal points are more complex or unevenly distributed, the SVM algorithm can further optimize the classification performance based on clustering, and improve the classification accuracy and recall rate through higher-dimensional hyperplane segmentation.

[0114] For example, the original data is first input into the K-means clustering algorithm for clustering and feature analysis to mark normal data points and abnormal data points.

[0115] Then, for data groups with more normal data points, these data groups are input into the input SVM model to further achieve accurate classification of normal data and abnormal data. Finally, a comprehensive classification result based on the two-step progressive algorithm can be obtained, which significantly improves the ability to identify abnormal data and achieves more accurate early warning.

[0116] In a preferred embodiment, the present invention may further include:

[0117] Mark the data points whose characteristic values ​​are within the characteristic distribution interval as the first normal data points;

[0118] Marking a data group where the total number of first normal data points is greater than a preset number threshold as a data group to be processed;

[0119] For each data group to be processed, the data group to be processed is input into a preset binary classification model, so that the binary classification model maps the feature value of each data point in the data group to be processed into the feature space, and generates the distance between each data point and the hyperplane according to the position of each data point in the feature space and the position of the hyperplane; the data points whose distance is less than the preset distance threshold are marked as second normal data points, and the position of each data point in the feature space and the position of the hyperplane are output;

[0120] Illustratively, when it is determined that the inspection data is appearance image data corresponding to each device, the binary classification model can map the feature value of each data point in the data group to be processed to a high-dimensional feature space through a kernel function.

[0121] Further, according to the position of each data point in the feature space and the position of the hyperplane, a feature space plane graph including a number of data points and the hyperplane is drawn;

[0122] According to the distance between each second normal data point and the hyperplane, and the total number of second normal data points, an average distance value is obtained;

[0123] Based on the feature space plane graph, a boundary region in the feature space plane graph is generated with the center of the hyperplane as the center of the circle and the average distance value as the radius;

[0124] Marking all data points in the feature space plane that do not fall within the boundary area as second abnormal data points;

[0125] generating second warning information including the second target device according to the second target device corresponding to each second abnormal data point;

[0126] The training process of the binary classification model includes:

[0127] Acquire a plurality of data group samples; wherein each data group sample includes: a plurality of normal data points and a plurality of abnormal data points;

[0128] Taking a number of data group samples as input and the predicted position of the hyperplane in the binary classification model to be trained as output, the binary classification model to be trained is iteratively trained until the interval distance between the position of the support vector in the feature space and the predicted position of the hyperplane is the largest, thereby generating a preset binary classification model;

[0129] The support vector is the normal data point closest to the predicted position of the hyperplane.

[0130] Schematically, the binary classification model is a classification model corresponding to the support vector machine (SVM); the SVM model is a binary classification model that can find a hyperplane in the feature space during training to maximize the interval between two different types of samples on the hyperplane. The sample points located on the maximum interval boundary are called "support vectors".

[0131] Schematically, the training process of the SVM model can be:

[0132] Data preprocessing: including feature selection, feature scaling, and data cleaning to ensure data quality and consistency.

[0133] Define kernel function: SVM model can use different kernel functions, such as linear kernel function, polynomial kernel function and Gaussian kernel function. The choice of kernel function depends on the characteristics of the data and the complexity of the classification problem.

[0134] Set parameters: You need to set the parameters of the SVM model, such as C value and gamma value.

[0135] Training the model: Using the processed data set, optimize the model parameters by minimizing the loss function.

[0136] During the training process, SVM will find a hyperplane that maximizes the interval between the two types of samples.

[0137] Finally, the trained model is used to make predictions on the new data set, and the prediction results depend on which side of the hyperplane the sample point falls on. That is, the distance between each data point and the hyperplane can be obtained based on the position of each data point in the feature space and the position of the hyperplane; thus, the data points whose distance is less than the preset distance threshold are marked as the second normal data points; schematically, the data points whose distance is not less than the preset distance threshold can also be marked as abnormal data points.

[0138] Different from the traditional SVM model application, the embodiment of the present invention can obtain an average distance value based on the distance between the classified normal data points and the hyperplane and the total number of the second normal data points;

[0139] Therefore, based on the average distance value and the center of the hyperplane, a new boundary area is generated, which can further define the range of normal data and abnormal data more accurately. The embodiment of the present invention utilizes the advantages of the SVM model in classification problems and combines the distance metric to improve the accuracy of anomaly detection.

[0140] In schematic form, since the boundary area is generated based on the average distance value of normal data points, when the distribution of normal data points changes for inspection data in different time periods, the boundary area will also be adjusted accordingly, allowing the model to better adapt to data changes and maintain a high accuracy of anomaly detection.

[0141] The traditional SVM model may require specific parameter adjustments for different data sets when applied. However, the method in the embodiment of the present invention generates a new boundary area by calculating the average distance value, and thus does not require complex parameter adjustments to the model to adapt to the characteristics of different data sets.

[0142] It can be understood that, first, each data point is mapped to a high-dimensional feature space through a kernel function (such as a radial basis function RBF) to map nonlinear data, so that the distance from each data point to the hyperplane can be calculated. Since the distribution of support vectors can be determined based on the training results of the SVM model, these support vectors constitute the hyperplane region for data classification. Then, based on the support vector or hyperplane, when constructing a boundary distribution model through a distance range in the feature space, the statistical indicators of the distance (such as the mean, standard deviation) can be used to determine the distribution range of normal data points and the boundary (i.e., the above-mentioned boundary region), that is, the distribution range can be obtained based on the obtained distance average and the marked hyperplane center; finally, for data points beyond the distribution range, they are marked as abnormal points.

[0143] In an embodiment of the present invention, even if the data is nonlinear in the original space, it can be mapped to a high-dimensional space through a kernel function, and then a linearly separable hyperplane is found in the high-dimensional space, so that the model can handle a wider range of data types and enhance the generalization ability of the model.

[0144] Finally, by marking data points that do not fall within the boundary area as abnormal data points and generating warning information containing the corresponding target equipment, a more accurate abnormal warning can be achieved. This is of great significance for application scenarios that require real-time monitoring and warning (such as equipment failure warning, network security monitoring, etc.).

[0145] Therefore, since the boundary area is generated based on the average distance value of normal data points, it can reduce the false alarm caused by data fluctuation or noise, thereby improving the accuracy and reliability of the warning information and reducing the false alarm rate.

[0146] like Figure 2 As shown, based on the above-mentioned various embodiments of the monitoring method applicable to patrol inspection data, the present invention provides a corresponding device embodiment;

[0147] An embodiment of the present invention provides a monitoring device suitable for inspection data, characterized in that it includes: an inspection data acquisition module, an equipment feature extraction module, a clustering module, a feature distribution generation module and an early warning module;

[0148] The inspection data acquisition module is used to acquire the inspection data corresponding to each device in the target area;

[0149] The device feature extraction module is used to extract a number of different types of device features for each device based on the inspection data of the device; wherein each device feature corresponds to a feature value;

[0150] The clustering module is used to use the characteristic value of each device characteristic as the characteristic value corresponding to each data point; perform cluster analysis on each data point according to the characteristic value of each data point to generate a number of different types of data groups;

[0151] The feature distribution generating module is used to perform density fitting on the feature values ​​corresponding to each data point in the data group for each data group, so as to generate a feature distribution interval corresponding to the data group;

[0152] The early warning module is used to mark, for each data group, data points whose characteristic values ​​are not within the characteristic distribution interval as first abnormal data points; and generate first early warning information including the first target device according to the first target device corresponding to each first abnormal data point.

[0153] It should be noted that the device embodiments described above are merely schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without paying creative labor.

[0154] Those skilled in the art can clearly understand that, for the sake of convenience and simplicity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0155] Based on the above-mentioned various embodiments of the monitoring method applicable to patrol inspection data, the present invention provides corresponding embodiments of terminal equipment items.

[0156] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a monitoring method suitable for patrol data as described in any method embodiment of the present invention.

[0157] The terminal device may be a computing terminal device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0158] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0159] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0160] Based on the above-mentioned various embodiments of the monitoring method applicable to patrol inspection data, the present invention provides corresponding storage medium item embodiments.

[0161] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a monitoring method suitable for patrol data as described in any method embodiment of the present invention.

[0162] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0163] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A monitoring method suitable for inspection data, characterized in that: include: Obtain inspection data corresponding to each device in the target area; For each device, a plurality of different types of device features are extracted based on the inspection data of the device; wherein each device feature corresponds to a feature value; The characteristic value of each device characteristic is used as the characteristic value corresponding to each data point; cluster analysis is performed on each data point according to the characteristic value of each data point to generate several different types of data groups; For each data group, density fitting is performed on the eigenvalues ​​corresponding to each data point in the data group to generate the characteristic distribution interval corresponding to the data group; For each data set, the data points whose characteristic values ​​are not within the characteristic distribution interval are marked as the first abnormal data points; According to the first target device corresponding to each first abnormal data point, first warning information including the first target device is generated.

2. A monitoring method suitable for inspection data according to claim 1, characterized in that: The inspection data includes: appearance image data corresponding to the equipment; According to the inspection data of the equipment, several different types of equipment features are extracted, including: The noise in the appearance image data is removed by using Gaussian filtering technology to generate preprocessed appearance image data; According to the preset image feature extraction algorithm, the image features in the preprocessed appearance image data are extracted to generate edge features, shape features and texture features corresponding to the device.

3. A monitoring method suitable for inspection data according to claim 1, characterized in that: The inspection data includes: operation data corresponding to the equipment; the operation data includes: voltage, current, temperature and noise; According to the inspection data of the equipment, several different types of equipment features are extracted, including: The voltage characteristics, current characteristics, temperature characteristics and noise characteristics corresponding to the device are extracted from the operating data corresponding to the device.

4. A monitoring method suitable for inspection data as claimed in claim 3, characterized in that: The step of performing density fitting on the characteristic values ​​corresponding to each data point in the data group to generate a characteristic distribution interval corresponding to the data group includes: According to each characteristic value in the data group, draw the histogram corresponding to the data group; Fitting the frequency distribution of each eigenvalue in the histogram according to a preset probability density function to generate a distribution model; wherein the distribution model corresponds to a statistical indicator for characterizing the distribution characteristics of the eigenvalue; According to the distribution model and the statistical indicators, the characteristic distribution interval corresponding to the data group is calculated.

5. A monitoring method suitable for inspection data according to claim 1, characterized in that: Also includes: Mark the data points whose characteristic values ​​are within the characteristic distribution interval as the first normal data points; Marking a data group where the total number of first normal data points is greater than a preset number threshold as a data group to be processed; For each data group to be processed, the data group to be processed is input into a preset binary classification model, so that the binary classification model maps the feature value of each data point in the data group to be processed into the feature space, and generates the distance between each data point and the hyperplane according to the position of each data point in the feature space and the position of the hyperplane; the data points whose distance is less than the preset distance threshold are marked as second normal data points, and the position of each data point in the feature space and the position of the hyperplane are output; According to the position of each data point in the feature space and the position of the hyperplane, a feature space plane diagram including a number of data points and the hyperplane is drawn; According to the distance between each second normal data point and the hyperplane, and the total number of second normal data points, an average distance value is obtained; Based on the feature space plane graph, a boundary region in the feature space plane graph is generated with the center of the hyperplane as the center of the circle and the average distance value as the radius; Marking all data points in the feature space plane that do not fall within the boundary area as second abnormal data points; generating second warning information including the second target device according to the second target device corresponding to each second abnormal data point; The training process of the binary classification model includes: Acquire a plurality of data group samples; wherein each data group sample includes: a plurality of normal data points and a plurality of abnormal data points; Taking a number of data group samples as input and the predicted position of the hyperplane in the binary classification model to be trained as output, the binary classification model to be trained is iteratively trained until the interval distance between the position of the support vector in the feature space and the predicted position of the hyperplane is the largest, thereby generating a preset binary classification model; The support vector is the normal data point closest to the predicted position of the hyperplane.

6. A monitoring method suitable for inspection data as claimed in claim 5, characterized in that: In the case where it is determined that the inspection data is the appearance image data corresponding to each device, the binary classification model maps the feature value of each data point in the data group to be processed to the feature space, including: The binary classification model maps the feature value of each data point in the data set to be processed to a high-dimensional feature space through a kernel function.

7. A monitoring method suitable for inspection data according to claim 6, characterized in that: The cluster analysis is performed on each data point according to its characteristic value to generate several different types of data groups, including: Repeat the following clustering operation. When the difference between the total value of each updated cluster center and the total value of each cluster center during the last clustering operation is less than the difference threshold, output the current data groups of different types: Calculate the distance from the feature value of each data point to each current cluster center, classify each data point to the current cluster center corresponding to the minimum distance, and generate the current data group; wherein, initially randomly select the cluster centers corresponding to the preset number as the current cluster centers; each data group corresponds to a type; For each data group, generating an updated cluster center of the data group according to the weighted average value corresponding to the characteristic value of each data point in the data group; When it is determined that the difference between the total value of each updated cluster center and the total value of each cluster center when the clustering operation was performed last time is not less than the difference threshold, the updated cluster center is used as the current cluster center when the clustering operation is performed next time.

8. A monitoring device suitable for inspection data, characterized in that: include: Inspection data acquisition module, equipment feature extraction module, clustering module, feature distribution generation module and early warning module; The inspection data acquisition module is used to acquire the inspection data corresponding to each device in the target area; The device feature extraction module is used to extract a number of different types of device features for each device based on the inspection data of the device; wherein each device feature corresponds to a feature value; The clustering module is used to use the characteristic value of each device characteristic as the characteristic value corresponding to each data point; perform cluster analysis on each data point according to the characteristic value of each data point to generate a number of different types of data groups; The feature distribution generating module is used to perform density fitting on the feature values ​​corresponding to each data point in the data group for each data group, so as to generate a feature distribution interval corresponding to the data group; The early warning module is used to mark, for each data group, data points whose characteristic values ​​are not within the characteristic distribution interval as first abnormal data points; and generate first early warning information including the first target device according to the first target device corresponding to each first abnormal data point.

9. A terminal device, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements a monitoring method suitable for inspection data as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute a monitoring method applicable to patrol inspection data as described in any one of claims 1 to 7.

Images