A multi-source detection image intelligent analysis system and method for a power site

Abstract

The application discloses a kind of multi-source detection image intelligent analysis system and method for electric power field, it is related to electric power field image analysis technical field, including the alarm level determination of multi-source detection image, constructs device alarm sequence set;Generation candidate association pair, analyze the time sequence significant degree between the previous alarm type and the subsequent alarm type in candidate association pair, determine the time sequence interval threshold of target association pair, obtain significant time sequence association set;Obtain the multi-source detection image set of electric power equipment in current period, and generate device alarm data in combination with significant time sequence association set and alarm threshold data;According to the device alarm data of electric power equipment, the electric power equipment of current period in electric power field is carried out equipment safety management, not only realizes the real-time detection of electric power equipment exception in electric power field, but also can actively predict the exception of electric power equipment, fundamentally guarantee the safety of electric power equipment.

Description

Technical Field

[0001] This invention relates to the field of power field image analysis technology, specifically a multi-source detection image intelligent analysis system and method for power field applications. Background Technology

[0002] In power field inspections and defect diagnosis of electrical equipment such as insulators, multi-source spectral imaging technologies such as visible light, infrared, and ultraviolet are widely used. Multi-source spectral imaging technologies acquire multi-source detection images. The use of multi-source detection images avoids the shortcomings of traditional single-source images, which cannot effectively detect the internal and external parts of electrical equipment. It can detect external damage to electrical equipment and effectively detect internal electrical defects, which is a detection effect that traditional single-source detection images cannot achieve.

[0003] However, during the development of power equipment failures, the various modal anomalies do not suddenly appear at the same moment, but follow a certain temporal evolution pattern. However, existing technologies treat multi-source alarms as isolated and concurrent events, failing to effectively utilize this temporal evolution pattern and ignoring the cross-spectral correlation of alarm events in the time dimension. This makes it impossible to predict the risk of subsequent more serious modal alarms from early modal alarms, resulting in delayed early warning capabilities. This not only affects the use of power equipment, but may even further affect the safety of the power site. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-source detection image intelligent analysis system and method for power field applications, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent analysis of multi-source detection images in power field, the method comprising: Step S1: Acquire multi-source detection images of abnormal power equipment at the power site, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences; Step S2: Based on the device alarm sequence set, generate candidate association pairs, obtain the mean observation interval of the candidate association pairs, rearrange the preceding alarm types and subsequent alarm types in the candidate association pairs by random distribution at intervals, analyze the temporal significance between the preceding alarm types and subsequent alarm types in the candidate association pairs, determine the temporal interval threshold of the target association pairs, and obtain the significant temporal association set. Step S3: Obtain the multi-source detection image set of the power equipment in the current cycle, and combine it with the significant temporal correlation set and alarm threshold data to generate equipment alarm data; Step S4: Based on the equipment alarm data of the power equipment, perform equipment safety management on the power equipment at the power site within the current cycle.

[0006] Furthermore, step S2 includes: Obtain the device alarm sequence set A of the device models of the power equipment detected at the power site in the current period, obtain the preceding alarm type and the following alarm type of the power equipment and aggregate them to obtain candidate association pairs; Obtain the preceding alarm type B and the following alarm type C from a candidate association pair; identify the marked abnormal power equipment for the candidate association pair; obtain the characteristic occurrence time points of the preceding alarm type B and the following alarm type C in the marked abnormal power equipment; obtain the effective time interval Δt of the candidate association pair in the marked abnormal power equipment; and calculate the mean observation interval μ of a candidate association pair containing the preceding alarm type B and the following alarm type C. (B,C) ; Obtain the first inspection time point T of a certain marked abnormal power equipment min and the final inspection time point T max ; A random offset variable δ is generated based on the observation time range, where the random offset variable δ conforms to the interval (0, T). max -T min The uniform distribution of ) Obtain the characteristic occurrence time t of subsequent alarm type C in a certain marked abnormal power equipment. c Using a random offset variable δ to represent the feature occurrence time point t c Perform random time-series shifts to obtain the time-series shift point t. (△,c) ; Obtain the time-series displacement time points of subsequent alarm type C in each marked abnormal power equipment, and calculate the mean random observation interval of a candidate correlation pair under the random offset variable δ; Set an independent random number R, generate R random offset variables based on the observation time range, obtain the mean of random observation intervals for a candidate association pair under the R random offset variables, and calculate the time-series significance value Q between the preceding alarm type B and the subsequent alarm type C in a candidate association pair. (B,C) ; Set a significance threshold q, when Q (B,C) When Q > q, it is determined that there is a significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and the candidate correlation pair is recorded as a significant correlation pair. (B,C) When ≤q, it is determined that there is no significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and no processing is performed on a certain candidate association pair; Calculate the time interval threshold W between the preceding alarm type and the following alarm type in the significant association pair, obtain the time interval threshold of each significant association pair of power equipment in the power field and aggregate them to obtain the significant time-series association set of power equipment; The above steps innovatively introduce a statistical hypothesis testing framework based on random rearrangement. By randomly cyclically shifting the time series of subsequent events and using the time series significance value to determine the degree of time series correlation between preceding and subsequent alarm types, this fundamentally solves the shortcomings of traditional technical means in distinguishing between true physical causal relationships and statistical accompanying phenomena naturally present due to the high frequency of alarms. It also eliminates a large number of spurious correlations caused by the frequent alarms of the equipment itself, ensuring the accuracy of subsequent trend identification of power equipment anomalies.

[0007] Furthermore, step S1 includes: The platform retrieves abnormal power equipment that has been identified as abnormal, obtains multi-source detection images of the abnormal power equipment, extracts the values ​​of multi-source abnormality indicators of the abnormal power equipment under various spectral modes from the multi-source detection images, and aggregates them to obtain the abnormality indicator group of the multi-source detection images. The alarm threshold data of abnormal power equipment is obtained from the platform. The alarm threshold data is used to obtain the alarm thresholds of multi-source abnormal indicators under various spectral modes. Based on the alarm thresholds of multi-source abnormal indicators under various spectral modes, the alarm level of multi-source detection images under various spectral modes is determined to obtain the alarm level under various spectral modes. When the alarm level of a certain spectral mode in a multi-source detection image is not zero, that spectral mode is recorded as an abnormal spectral mode. When an abnormal spectral mode exists in a multi-source detection image, the multi-source detection image is recorded as an abnormal detection image. The capture time point of the abnormal detection image and the alarm level of the abnormal spectral mode are obtained and aggregated to obtain the alarm event of the abnormal detection image. The alarm events of each abnormal detection image in the abnormal power equipment are obtained and sorted according to the time sequence of the abnormal detection images to obtain the equipment alarm sequence of the abnormal power equipment. The alarm sequences of each abnormal power device in the power field are obtained and aggregated to obtain the alarm sequence set of the equipment model to which the abnormal power device belongs in the power field.

[0008] Furthermore, step S3 includes: The process involves acquiring multi-source detection images of power equipment in the current period, acquiring several multi-source detection images of power equipment in historical periods, acquiring the capture time points of the multi-source detection images and several multi-source detection images, and sorting the multi-source detection images and several multi-source detection images in chronological order to obtain a set of multi-source detection images of power equipment in the current period. Obtain the significant temporal correlation set of the equipment model to which the power equipment belongs, obtain the values ​​of multi-source anomaly indicators under various spectral modes from the multi-source detection images of the power equipment in the current period, and obtain the alarm thresholds of the multi-source anomaly indicators under various spectral modes from the alarm threshold data. Based on the alarm threshold data, determine the alarm level of abnormal power equipment in the multi-source detection images under each spectral mode in the current period. Combine the spectral modes of power equipment in the multi-source detection images in the current period with the corresponding alarm levels to obtain the current alarm type of power equipment in the current period. The alarm types of power equipment in each multi-source detection image set are obtained from the multi-source detection image set. When the alarm type of a certain multi-source detection image in the multi-source detection image set is the preceding alarm type F, the subsequent alarm type P that is in the same significant correlation pair as the preceding alarm type F is obtained from the significant temporal correlation set. The temporal interval threshold W between the preceding alarm type F and the subsequent alarm type P is obtained from the significant temporal correlation set. (F,P) Obtain the capture time t of the multi-source detection image to which the preceding alarm type F belongs. F Obtain the capture time t´ of the multi-source detection images within the current period, and calculate the predicted time t of the subsequent alarm type P in the power equipment. △ =t´+[W (F,P) -(t´-t F )]; The subsequent alarm type P is denoted as the predicted abnormal alarm type of the power equipment in the current period. The prediction time points of the current alarm type and the predicted abnormal alarm type of the power equipment in the current period are obtained and aggregated to obtain the equipment alarm data of the power equipment in the current period.

[0009] Furthermore, step S4 includes: Obtain equipment alarm data of power equipment in the current period, and extract the current alarm type and the predicted time point of the predicted abnormal alarm type of power equipment in the current period from the equipment alarm data; If the alarm level of the spectral mode of the power equipment is 0 in the current alarm type in the current cycle, it is determined that there is no abnormality in the power equipment in the current cycle. Otherwise, it is determined that there is an abnormality in the power equipment in the current cycle, and the platform will notify the staff at the power site to carry out safety maintenance on the power equipment. When a predicted abnormal alarm type exists in the current cycle of the power equipment, the predicted time point of the abnormal alarm type is obtained, it is determined that the power equipment will have an abnormal risk after the predicted time point, and the predicted time point and the predicted abnormal alarm type are sent to the staff at the power site through the platform to notify the staff to carry out safety maintenance of the power equipment.

[0010] To better implement the above method, a multi-source detection image intelligent analysis system is also proposed. The system includes an alarm sequence construction module, a temporal correlation significant analysis module, a device alarm module, and a device safety management module. The alarm sequence construction module is used to acquire multi-source detection images of abnormal power equipment in the power field, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences. The temporal correlation significance analysis module is used to generate candidate correlation pairs, analyze the temporal significance between the preceding alarm type and the following alarm type in the candidate correlation pairs, and obtain a significant temporal correlation set. The equipment alarm module is used to acquire multi-source detection image sets of power equipment and generate equipment alarm data of power equipment in the current period; The equipment safety management module is used to manage the safety of electrical equipment in the power field based on the equipment alarm data of the power equipment.

[0011] Furthermore, the alarm sequence construction module includes an abnormal indicator group acquisition unit and an alarm sequence construction unit; The abnormal index group acquisition unit is used to acquire multi-source detection images of abnormal power equipment, obtain the values ​​of multi-source abnormal indicators of abnormal power equipment under various spectral modes from the multi-source detection images and collect them to obtain the abnormal index group of the multi-source detection images. The alarm sequence construction unit is used to determine the alarm level of the multi-source detection images in different spectral modes based on the abnormal index group of the multi-source detection images, and to construct the device alarm sequence set.

[0012] Furthermore, the temporal correlation saliency analysis module includes an interval random distribution rearrangement unit and a temporal correlation saliency analysis unit; The interval random distribution rearrangement unit is used to acquire the set of equipment alarm sequences, generate candidate association pairs of power equipment, obtain the average observation interval of the candidate association pairs, and perform interval random distribution rearrangement of the preceding alarm type and the following alarm type in the candidate association pairs. The temporal correlation saliency analysis unit is used to analyze the temporal saliency between preceding and subsequent alarm types in candidate correlation pairs, determine the temporal interval threshold of the target correlation pair, and obtain a significant temporal correlation set.

[0013] Furthermore, the device alarm module includes an image set acquisition unit and a device alarm unit; The image set acquisition unit is used to acquire multi-source detection images of power equipment in the power field during the current cycle, acquire several multi-source detection images of power equipment during historical cycles, and collect them in chronological order to obtain a multi-source detection image set. The device alarm unit is used to generate device alarm data based on a multi-source detection image set, combined with a significant temporal correlation set and alarm threshold data.

[0014] Furthermore, the equipment safety management module includes an equipment safety management unit; The equipment safety management unit is used to acquire equipment alarm data of power equipment in the current cycle, and to perform equipment safety management of power equipment in the power field based on the equipment alarm data.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: Current technologies can only trigger alarms when the index of a single spectral mode exceeds a threshold, by which time the abnormality has already developed to a relatively serious level. However, this invention, by exploring the temporal evolution patterns between alarms across spectral modes such as ultraviolet, infrared, and visible light, can predict the approximate time and alarm level of power equipment anomalies in later stages by using weak signals in early multi-source detection images of power equipment. This allows staff to conduct safety management of power equipment several days or even weeks in advance. It not only enables real-time detection of power equipment anomalies in the power field but also proactively predicts power equipment anomalies, fundamentally preventing power equipment safety accidents in the power field and ensuring power equipment safety. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for intelligent analysis of multi-source detection images in power field according to the present invention; Figure 2 This is a flowchart of the modules of a multi-source detection image intelligent analysis system according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figures 1-2 As shown, this invention provides a technical solution: an intelligent analysis method for multi-source detection images in power field applications, the method comprising: Step S1: Acquire multi-source detection images of abnormal power equipment at the power site, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences; Step S1 includes: The platform retrieves abnormal power equipment that has been identified as abnormal, obtains multi-source detection images of the abnormal power equipment, extracts the values ​​of multi-source abnormality indicators of the abnormal power equipment under various spectral modes from the multi-source detection images, and aggregates them to obtain the abnormality indicator group of the multi-source detection images. For example, each spectral mode specifically refers to visible light, infrared, and ultraviolet light. Among them, the multi-source anomaly index of visible light is the visible light anomaly index, the multi-source anomaly index of infrared light is the infrared anomaly index, and the multi-source anomaly index of ultraviolet light is the ultraviolet anomaly index. For example, when the abnormal power equipment is an insulator, the visible light anomaly index X of the abnormal power equipment... V Infrared anomaly index X I and UV anomaly index X U Specifically: The percentage of damaged insulator area and the difference between the highest temperature of the tested insulator and the temperature of the normal reference body in the same frame are obtained from the multi-source detection images. The original photon count N is also obtained from the multi-source detection images. raw When acquiring multi-source detection images for ultraviolet imaging, the imager gain G, the observation distance d, and the normalized reference distance d´ and reference gain G´ are used. The proportion of the damaged area of ​​the insulator obtained from the multi-source detection image is used as the visible light anomaly index of the insulator in the multi-source detection image; The difference between the highest temperature of the insulator under test in the multi-source detection image and the temperature of the normal reference body in the same frame is used as the infrared anomaly index of the insulator in the multi-source detection image. Calculate the ultraviolet anomaly index X U : , The alarm threshold data of abnormal power equipment is obtained from the platform. The alarm threshold data is used to obtain the alarm thresholds of multi-source abnormal indicators under various spectral modes. Based on the alarm thresholds of multi-source abnormal indicators under various spectral modes, the alarm level of multi-source detection images under various spectral modes is determined to obtain the alarm level under various spectral modes. For example, in a specific embodiment where abnormal power equipment is used as an insulator, the alarm level of the multi-source detection image is determined under each spectral mode to obtain the alarm level under each spectral mode. The specific process is as follows: The alarm thresholds for each multi-source anomaly index under each spectral mode are obtained from the alarm threshold data. Among them, the visible light anomaly index X...V The alarm thresholds for are V1 and V2, where 0 < V1 < V2, and the infrared anomaly index is X I The alarm thresholds for are I1 and I2, where 0 < I1 < I2, and the ultraviolet anomaly index is X U The alarm thresholds for are U1 and U2, where 0 < U1 < U2; For example, V1, V2, I1, I2, U1, and U2 are preset based on historical data statistics and combined with expert experience. In this specific embodiment, the specific values are: Visible light anomaly index X V The alarm thresholds for are V1 and V2, where V1 = 0.02 and V2 = 0.10; Infrared anomaly index X I The alarm thresholds for are I1 and I2, where I1 = 5 o C and I2 = 10 o C; Ultraviolet anomaly index X U The alarm thresholds for are U1 and U2, where U1 = 50 and U2 = 200; Alarm levels for visible light: , Alarm levels for infrared: , Alarm levels for ultraviolet: , When the alarm level of a certain spectral modality in the multi-source detection image is not zero, record that spectral modality as an abnormal spectral modality; When there is an abnormal spectral modality in the multi-source detection image, record the multi-source detection image as an abnormal detection image, obtain the shooting time point of the abnormal detection image, the alarm level of the abnormal spectral modality, and collect them to obtain the alarm event of the abnormal detection image; Obtain the alarm events of each abnormal detection image in the abnormal power equipment and sort them in the order of the shooting time of the abnormal detection image to obtain the equipment alarm sequence of the abnormal power equipment; Obtain and collect the equipment alarm sequences of each abnormal power equipment in the power site to obtain the equipment alarm sequence set of the equipment models to which the abnormal power equipment in the power site belongs.

[0019] Step S2: Generate candidate association pairs according to the equipment alarm sequence set, obtain the average observation interval of the candidate association pairs, rearrange the pre-alarm type and post-alarm type in the candidate association pairs with interval random distribution, analyze the temporal significance between the pre-alarm type and the post-alarm type in the candidate association pairs, determine the temporal interval threshold of the target association pair, and obtain the significant temporal association set; Among them, step S2 includes: Obtain the device alarm sequence set A of the device models of the power equipment detected at the power site in the current period, obtain the preceding alarm type and the following alarm type of the power equipment and aggregate them to obtain candidate association pairs; For example, the specific process for obtaining the preceding and following alarm types of power equipment is as follows: Obtain the equipment alarm sequence set A of the equipment models of the power equipment detected at the power site in the current period, obtain the alarm level of each spectral mode of the power equipment in the power site, and randomly combine each spectral mode with each alarm level to generate each alarm type of the power equipment. Randomly combine each alarm type in pairs, and record the alarm type with the first combination in the order as the preceding alarm type, and the alarm type with the last combination in the order as the following alarm type. For example, the specific process for generating various alarm types for power equipment is as follows: When the spectral mode is visible light and the alarm level is 1, then the alarm type consisting of visible light alarms has an alarm level of 1. For example, a candidate association pair contains only one preceding alarm type and one following alarm type; Obtain the preceding alarm type B and the following alarm type C from a candidate association pair; identify the marked abnormal power equipment for the candidate association pair; obtain the characteristic occurrence time points of the preceding alarm type B and the following alarm type C in the marked abnormal power equipment; obtain the effective time interval Δt of the candidate association pair in the marked abnormal power equipment; and calculate the mean observation interval μ of a candidate association pair containing the preceding alarm type B and the following alarm type C. (B,C) ; For example, the specific process for obtaining the marked abnormal power equipment of a certain candidate association pair is as follows: Obtain the device alarm sequence of each abnormal power device from the device alarm sequence set A. If the alarm event in the device alarm sequence of a certain abnormal power device includes the preceding alarm type B and the following alarm type C, and the shooting time of the alarm event corresponding to the preceding alarm type B is before the shooting time of the alarm event corresponding to the following alarm type C, then the abnormal power device is marked as a candidate associated pair of marked abnormal power devices. For example, the characteristic occurrence time point of the preceding alarm type B in the abnormal power equipment is: The capture time of the abnormal detection image in the alarm event corresponding to the preceding alarm type B is obtained from the equipment alarm sequence of the marked abnormal power equipment, and recorded as the feature occurrence time of the preceding alarm type B in the marked abnormal power equipment. For example, the specific process for obtaining a candidate association pair within the valid time interval Δt for marking abnormal power equipment is as follows: Subtracting the time point of occurrence of the preceding alarm type B from the time point of occurrence of the subsequent alarm type C within the marked abnormal power equipment, we obtain the effective time interval Δt of a candidate association pair in the marked abnormal power equipment. For example, the mean observation interval μ of a candidate correlation pair containing preceding alarm type B and subsequent alarm type C. (B,C) The specific calculation formula is as follows: , Wherein, △t i is the effective time interval of a candidate association pair in the i-th marked anomalous power device; m is the total number of marked anomalous power devices for a candidate association pair; Obtain the first inspection time point T of a certain marked abnormal power equipment min and the final inspection time point T max ; For example, the first inspection time point T of a certain marked abnormal power equipment min and the final inspection time point T max The specific acquisition process is as follows: Obtain the capture time of the first multi-source image captured for a certain labeled anomalous power equipment in a candidate association pair, and record it as the first detection time of that labeled anomalous power equipment. Obtain the capture time T of the last multi-source image captured for that labeled anomalous power equipment. min And denoted as the last inspection time point T of a certain marked abnormal power equipment. max ; A random offset variable δ is generated based on the observation time range, where the random offset variable δ conforms to the interval (0, T). max -T min The uniform distribution of ) For example, a random offset variable δ is generated based on the observation time range. The specific generation process is as follows: Using pseudo-random number generators such as the Mason twitch algorithm and linear congruence generators, a real number uniformly distributed between [0,1] is generated. Then, through a linear transformation, the mean square distribution of the real number is mapped to the observation time range T, resulting in a number that strictly conforms to the interval [0,T]. max -T min A uniformly distributed random offset variable; Obtain the characteristic occurrence time t of subsequent alarm type C in a certain marked abnormal power equipment. c Using a random offset variable δ to represent the feature occurrence time point t c Perform random time-series shifts to obtain the time-series shift point t. (△,c) ; For example, using a random offset variable δ for the feature occurrence time point t c Perform random time-series shifts to obtain the time-series shift point t.(△,c) The specific formula is as follows: , Where mod represents the modulo operation; Obtain the time-series displacement time points of subsequent alarm type C in each marked abnormal power equipment, and calculate the mean random observation interval of a candidate correlation pair under the random offset variable δ; For example, the specific process for obtaining the time shift point of subsequent alarm type C in each marked abnormal power equipment is as follows: Obtain the time-series offset time point of subsequent alarm type C in the marked abnormal power equipment. Use the random offset variable δ to randomly offset the characteristic occurrence time point in each marked abnormal power equipment to obtain the time-series offset time point of subsequent alarm type C in each marked abnormal power equipment. For example, the mean of random observation intervals for a candidate association pair is obtained as follows: Using the time-series displacement time points of subsequent alarm type C in each marked abnormal power equipment, calculate the average observation interval of a candidate correlation pair, and record it as the average random observation interval of a candidate correlation pair under the random offset variable δ. Set an independent random number R, generate R random offset variables based on the observation time range, obtain the mean of random observation intervals for a candidate association pair under the R random offset variables, and calculate the time-series significance value Q between the preceding alarm type B and the subsequent alarm type C in a candidate association pair. (B,C) ; For example, the time-series significance value Q (B,C) The specific calculation formula is as follows: , Where μ is the overall observed mean of a candidate association pair; σ is the overall observed standard deviation of a candidate association pair; For example, the specific calculation process for the overall observed mean μ of a candidate association pair is as follows: Calculate the average of the random observation intervals of a candidate association pair under R random offset variables, and denot it as the overall observation mean of the candidate association pair; For example, the specific formula for calculating the overall observed standard deviation σ of a candidate association pair is: , in, Let be the mean of random observation intervals for a candidate association pair under the z-th random offset variable; Set a significance threshold q, when Q (B,C) When Q > q, it is determined that there is a significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and the candidate correlation pair is recorded as a significant correlation pair. (B,C)When ≤q, it is determined that there is no significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and no processing is performed on a certain candidate association pair; Calculate the time interval threshold W between the preceding alarm type and the following alarm type in the significant association pair, obtain the time interval threshold of each significant association pair of power equipment in the power field and aggregate them to obtain the significant time-series association set of power equipment; For example, the specific calculation process for the time interval threshold W is as follows: Obtain the standard deviation σ´ and the mean observation interval μ´ of the effective time interval for significant correlation pairs in each marked abnormal power equipment, set the feature time series coefficient k, and calculate the time series interval threshold W=μ´-k×σ´ between the preceding alarm type and the following alarm type in the significant correlation pair; For example, the specific characteristic time series coefficient k is: The characteristic time series coefficient k indicates how many standard deviations σ' downward deviation of the observation interval mean μ´ is still considered normal when describing the natural fluctuations in alarm evolution time; The purpose of setting the characteristic time series coefficient k is to: set the characteristic time series coefficient k, and calculate the time interval threshold W between the preceding alarm type and the following alarm type in the significant correlation pair based on the characteristic time series coefficient k. Through the time interval threshold W, the alarm anomaly risk of the following alarm type in the significant correlation pair can be quickly and accurately determined at a certain time point after the abnormal alarm of the preceding alarm type in the significant correlation pair occurs. Therefore, by setting the characteristic time series coefficient k, the accuracy of predicting abnormal alarms of power equipment can be improved. The characteristic time series coefficient k is an engineering parameter obtained based on statistical principles. In the specific embodiment of this application, the effective time interval between the preceding alarm type and the following alarm type of the significantly associated pair approximately follows a normal distribution. At the same time, according to the 3σ principle, the probability that the effective time interval of the significantly associated pair in each marked abnormal power device falls outside [μ´-3×σ´,μ´+k×σ´] is less than 0.3%. Therefore, 3 is selected as the value of the characteristic time series coefficient k in the specific embodiment. For example, the specific calculation process for the mean observation interval μ´ is as follows: Obtain the average of the effective time intervals for significant correlation pairs in each marked anomalous power device, and denote it as the mean observation interval μ´ of significant correlation pairs.

[0020] Step S3: Obtain the multi-source detection image set of the power equipment in the current cycle, and combine it with the significant temporal correlation set and alarm threshold data to generate equipment alarm data; Step S3 includes: The process involves acquiring multi-source detection images of power equipment in the current period, acquiring several multi-source detection images of power equipment in historical periods, acquiring the capture time points of the multi-source detection images and several multi-source detection images, and sorting the multi-source detection images and several multi-source detection images in chronological order to obtain a set of multi-source detection images of power equipment in the current period. Obtain the significant temporal correlation set of the equipment model to which the power equipment belongs, obtain the values ​​of multi-source anomaly indicators under various spectral modes from the multi-source detection images of the power equipment in the current period, and obtain the alarm thresholds of the multi-source anomaly indicators under various spectral modes from the alarm threshold data. Based on the alarm threshold data, determine the alarm level of abnormal power equipment in the multi-source detection images under each spectral mode in the current period. Combine the spectral modes of power equipment in the multi-source detection images in the current period with the corresponding alarm levels to obtain the current alarm type of power equipment in the current period. The alarm types of power equipment in each multi-source detection image set are obtained from the multi-source detection image set. When the alarm type of a certain multi-source detection image in the multi-source detection image set is the preceding alarm type F, the subsequent alarm type P that is in the same significant correlation pair as the preceding alarm type F is obtained from the significant temporal correlation set. The temporal interval threshold W between the preceding alarm type F and the subsequent alarm type P is obtained from the significant temporal correlation set. (F,P) Obtain the capture time t of the multi-source detection image to which the preceding alarm type F belongs. F Obtain the capture time t´ of the multi-source detection images within the current period, and calculate the predicted time t of the subsequent alarm type P in the power equipment. △ =t´+[W (F,P) -(t´-t F )]; For example, when a significant correlation pair contains a preceding alarm type F and a subsequent alarm type P, the time interval threshold for a significant correlation pair is the time interval threshold W between the preceding alarm type F and the subsequent alarm type P. (F,P) ; The subsequent alarm type P is denoted as the predicted abnormal alarm type of the power equipment in the current period. The prediction time points of the current alarm type and the predicted abnormal alarm type of the power equipment in the current period are obtained and aggregated to obtain the equipment alarm data of the power equipment in the current period.

[0021] Step S4: Based on the equipment alarm data of the power equipment, perform equipment safety management on the power equipment at the power site within the current cycle; Step S4 includes: Obtain equipment alarm data of power equipment in the current period, and extract the current alarm type and the predicted time point of the predicted abnormal alarm type of power equipment in the current period from the equipment alarm data; If the alarm level of the spectral mode of the power equipment is 0 in the current alarm type in the current cycle, it is determined that there is no abnormality in the power equipment in the current cycle. Otherwise, it is determined that there is an abnormality in the power equipment in the current cycle, and the platform will notify the staff at the power site to carry out safety maintenance on the power equipment. When a predicted abnormal alarm type exists in the current cycle of the power equipment, the predicted time point of the abnormal alarm type is obtained, it is determined that the power equipment will have an abnormal risk after the predicted time point, and the predicted time point and the predicted abnormal alarm type are sent to the staff at the power site through the platform to notify the staff to carry out safety maintenance of the power equipment.

[0022] To better implement the above method, a multi-source detection image intelligent analysis system is also proposed. The system includes an alarm sequence construction module, a temporal correlation significant analysis module, a device alarm module, and a device safety management module. The alarm sequence construction module is used to acquire multi-source detection images of abnormal power equipment in the power field, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences. The temporal correlation significance analysis module is used to generate candidate correlation pairs, analyze the temporal significance between the preceding alarm type and the following alarm type in the candidate correlation pairs, and obtain a significant temporal correlation set. The equipment alarm module is used to acquire multi-source detection image sets of power equipment and generate equipment alarm data of power equipment in the current period; The equipment safety management module is used to manage the safety of electrical equipment in the power field based on the equipment alarm data of the power equipment.

[0023] The alarm sequence construction module includes an abnormal indicator group acquisition unit and an alarm sequence construction unit. The abnormal index group acquisition unit is used to acquire multi-source detection images of abnormal power equipment, obtain the values ​​of multi-source abnormal indicators of abnormal power equipment under various spectral modes from the multi-source detection images and collect them to obtain the abnormal index group of the multi-source detection images. The alarm sequence construction unit is used to determine the alarm level of the multi-source detection images in different spectral modes based on the abnormal index group of the multi-source detection images, and to construct the device alarm sequence set.

[0024] The time-series association significant analysis module includes an interval random distribution rearrangement unit and a time-series association significant analysis unit; The interval random distribution rearrangement unit is used to acquire the set of equipment alarm sequences, generate candidate association pairs of power equipment, obtain the average observation interval of the candidate association pairs, and perform interval random distribution rearrangement of the preceding alarm type and the following alarm type in the candidate association pairs. The temporal correlation saliency analysis unit is used to analyze the temporal saliency between preceding and subsequent alarm types in candidate correlation pairs, determine the temporal interval threshold of the target correlation pair, and obtain a significant temporal correlation set.

[0025] The device alarm module includes an image set acquisition unit and a device alarm unit. The image set acquisition unit is used to acquire multi-source detection images of power equipment in the power field during the current cycle, acquire several multi-source detection images of power equipment during historical cycles, and collect them in chronological order to obtain a multi-source detection image set. The device alarm unit is used to generate device alarm data based on a multi-source detection image set, combined with a significant temporal correlation set and alarm threshold data.

[0026] The equipment safety management module includes an equipment safety management unit; The equipment safety management unit is used to acquire equipment alarm data of power equipment in the current cycle, and to perform equipment safety management of power equipment in the power field based on the equipment alarm data.

[0027] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent analysis of multi-source detection images in power field applications, characterized in that, The method includes: Step S1: Acquire multi-source detection images of abnormal power equipment at the power site, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences; Step S2: Based on the device alarm sequence set, generate candidate association pairs, obtain the mean observation interval of the candidate association pairs, rearrange the preceding alarm types and subsequent alarm types in the candidate association pairs by random distribution at intervals, analyze the temporal significance between the preceding alarm types and subsequent alarm types in the candidate association pairs, determine the temporal interval threshold of the target association pairs, and obtain the significant temporal association set. Step S3: Obtain the multi-source detection image set of the power equipment in the current cycle, and combine it with the significant temporal correlation set and alarm threshold data to generate equipment alarm data; Step S4: Based on the equipment alarm data of the power equipment, perform equipment safety management on the power equipment at the power site within the current cycle.

2. The intelligent analysis method for multi-source detection images in power field according to claim 1, characterized in that, Step S2 includes: Obtain the device alarm sequence set A of the device models of the power equipment detected at the power site in the current period, obtain the preceding alarm type and the following alarm type of the power equipment and aggregate them to obtain candidate association pairs; Obtain the preceding alarm type B and the following alarm type C from a candidate association pair; identify the marked abnormal power equipment for the candidate association pair; obtain the characteristic occurrence time points of the preceding alarm type B and the following alarm type C in the marked abnormal power equipment; obtain the effective time interval Δt of the candidate association pair in the marked abnormal power equipment; and calculate the mean observation interval μ of a candidate association pair containing the preceding alarm type B and the following alarm type C. (B,C) ; Obtain the first inspection time point T of a certain marked abnormal power equipment min and the final inspection time point T max ; A random offset variable δ is generated based on the observation time range, where the random offset variable δ conforms to the interval (0, T). max -T min The uniform distribution of ) Obtain the characteristic occurrence time t of subsequent alarm type C in a certain marked abnormal power equipment. c Using a random offset variable δ to represent the feature occurrence time point t c Perform random time-series shifts to obtain the time-series shift point t. (△,c) ; Obtain the time-series displacement time points of subsequent alarm type C in each marked abnormal power equipment, and calculate the mean random observation interval of a candidate correlation pair under the random offset variable δ; Set an independent random number R, generate R random offset variables based on the observation time range, obtain the mean of random observation intervals for a candidate association pair under the R random offset variables, and calculate the time-series significance value Q between the preceding alarm type B and the subsequent alarm type C in a candidate association pair. (B,C) ; Set a significance threshold q, when Q (B,C) When Q > q, it is determined that there is a significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and the candidate correlation pair is recorded as a significant correlation pair. (B,C) When ≤q, it is determined that there is no significant temporal logical dependency between the preceding alarm type B and the following alarm type C, and no processing is performed on a certain candidate association pair; Calculate the time interval threshold W between the preceding and subsequent alarm types in the significant association pair, obtain the time interval threshold of each significant association pair of power equipment in the power field, and aggregate them to obtain the significant time-series association set of power equipment.

3. The intelligent analysis method for multi-source detection images in power field according to claim 1, characterized in that, Step S1 includes: The platform retrieves abnormal power equipment that has been identified as abnormal, obtains multi-source detection images of the abnormal power equipment, extracts the values ​​of multi-source abnormality indicators of the abnormal power equipment under various spectral modes from the multi-source detection images, and aggregates them to obtain the abnormality indicator group of the multi-source detection images. The alarm threshold data of abnormal power equipment is obtained from the platform. The alarm threshold data is used to obtain the alarm thresholds of multi-source abnormal indicators under various spectral modes. Based on the alarm thresholds of multi-source abnormal indicators under various spectral modes, the alarm level of multi-source detection images under various spectral modes is determined to obtain the alarm level under various spectral modes. When the alarm level of a certain spectral mode in a multi-source detection image is not zero, that spectral mode is recorded as an abnormal spectral mode. When an abnormal spectral mode exists in a multi-source detection image, the multi-source detection image is recorded as an abnormal detection image. The capture time point of the abnormal detection image and the alarm level of the abnormal spectral mode are obtained and aggregated to obtain the alarm event of the abnormal detection image. The alarm events of each abnormal detection image in the abnormal power equipment are obtained and sorted according to the time sequence of the abnormal detection images to obtain the equipment alarm sequence of the abnormal power equipment. The alarm sequences of each abnormal power device in the power field are obtained and aggregated to obtain the alarm sequence set of the equipment model to which the abnormal power device belongs in the power field.

4. The intelligent analysis method for multi-source detection images in power field according to claim 1, characterized in that, Step S3 includes: The process involves acquiring multi-source detection images of power equipment in the current period, acquiring several multi-source detection images of power equipment in historical periods, acquiring the capture time points of the multi-source detection images and several multi-source detection images, and sorting the multi-source detection images and several multi-source detection images in chronological order to obtain a set of multi-source detection images of power equipment in the current period. Obtain the significant temporal correlation set of the equipment model to which the power equipment belongs, obtain the values ​​of multi-source anomaly indicators under various spectral modes from the multi-source detection images of the power equipment in the current period, and obtain the alarm thresholds of the multi-source anomaly indicators under various spectral modes from the alarm threshold data. Based on the alarm threshold data, determine the alarm level of abnormal power equipment in the multi-source detection images under each spectral mode in the current period. Combine the spectral modes of power equipment in the multi-source detection images in the current period with the corresponding alarm levels to obtain the current alarm type of power equipment in the current period. The alarm types of power equipment in each multi-source detection image set are obtained from the multi-source detection image set. When the alarm type of a certain multi-source detection image in the multi-source detection image set is the preceding alarm type F, the subsequent alarm type P that is in the same significant correlation pair as the preceding alarm type F is obtained from the significant temporal correlation set. The temporal interval threshold W between the preceding alarm type F and the subsequent alarm type P is obtained from the significant temporal correlation set. (F,P) Obtain the capture time t of the multi-source detection image to which the preceding alarm type F belongs. F Obtain the capture time t´ of the multi-source detection images within the current period, and calculate the predicted time t of the subsequent alarm type P in the power equipment. △ =t´+[W (F,P) -(t´-t F )]; The subsequent alarm type P is denoted as the predicted abnormal alarm type of the power equipment in the current period. The prediction time points of the current alarm type and the predicted abnormal alarm type of the power equipment in the current period are obtained and aggregated to obtain the equipment alarm data of the power equipment in the current period.

5. The intelligent analysis method for multi-source detection images in power field according to claim 1, characterized in that, Step S4 includes: Obtain equipment alarm data of power equipment in the current period, and extract the current alarm type and the predicted time point of the predicted abnormal alarm type of power equipment in the current period from the equipment alarm data; If the alarm level of the spectral mode of the power equipment is 0 in the current alarm type in the current cycle, it is determined that there is no abnormality in the power equipment in the current cycle. Otherwise, it is determined that there is an abnormality in the power equipment in the current cycle, and the platform will notify the staff at the power site to carry out safety maintenance on the power equipment. When a predicted abnormal alarm type exists in the current cycle of the power equipment, the predicted time point of the abnormal alarm type is obtained, it is determined that the power equipment will have an abnormal risk after the predicted time point, and the predicted time point and the predicted abnormal alarm type are sent to the staff at the power site through the platform to notify the staff to carry out safety maintenance of the power equipment.

6. A multi-source detection image intelligent analysis system, used to execute the multi-source detection image intelligent analysis method for power field as described in any one of claims 1-5, characterized in that, The system includes an alarm sequence construction module, a time-series correlation significant analysis module, a device alarm module, and a device security management module; The alarm sequence construction module is used to acquire multi-source detection images of abnormal power equipment in the power field, obtain alarm threshold data from the platform, determine the alarm level of the multi-source detection images, and construct a set of equipment alarm sequences. The temporal correlation saliency analysis module is used to generate candidate correlation pairs, analyze the temporal saliency between preceding alarm types and subsequent alarm types in the candidate correlation pairs, and obtain a significant temporal correlation set. The device alarm module is used to acquire multi-source detection image sets of power equipment and generate device alarm data of power equipment in the current period; The equipment safety management module is used to manage the safety of power equipment in the power field based on the equipment alarm data of the power equipment.

7. The multi-source detection image intelligent analysis system according to claim 6, characterized in that, The alarm sequence construction module includes an abnormal indicator group acquisition unit and an alarm sequence construction unit; The abnormal index group acquisition unit is used to acquire multi-source detection images of abnormal power equipment, obtain the values ​​of multi-source abnormal indicators of abnormal power equipment under various spectral modes from the multi-source detection images and collect them to obtain the abnormal index group of the multi-source detection images. The alarm sequence construction unit is used to determine the alarm level of the multi-source detection image under different spectral modes based on the abnormal index group of the multi-source detection image, and construct the device alarm sequence set.

8. The multi-source detection image intelligent analysis system according to claim 6, characterized in that, The time-series association significant analysis module includes an interval random distribution rearrangement unit and a time-series association significant analysis unit; The interval random distribution rearrangement unit is used to acquire the equipment alarm sequence set, generate candidate association pairs of power equipment, acquire the average observation interval of the candidate association pairs, and perform interval random distribution rearrangement of the preceding alarm type and the following alarm type in the candidate association pairs. The temporal correlation saliency analysis unit is used to analyze the temporal saliency between the preceding alarm type and the following alarm type in the candidate correlation pair, determine the temporal interval threshold of the target correlation pair, and obtain a significant temporal correlation set.

9. The multi-source detection image intelligent analysis system according to claim 6, characterized in that, The device alarm module includes an image set acquisition unit and a device alarm unit; The image set acquisition unit is used to acquire multi-source detection images of power equipment in the power field during the current cycle, acquire several multi-source detection images of power equipment during historical cycles, and collect them in chronological order to obtain a multi-source detection image set. The device alarm unit is used to generate device alarm data based on a multi-source detection image set, combined with a significant temporal correlation set and alarm threshold data.

10. The multi-source detection image intelligent analysis system according to claim 6, characterized in that, The equipment safety management module includes an equipment safety management unit; The equipment safety management unit is used to acquire equipment alarm data of power equipment in the current cycle, and to perform equipment safety management of power equipment in the power field based on the equipment alarm data.

Images