Distribution network line fault early warning analysis method based on deep learning

Abstract

The invention provides a distribution network line fault early warning analysis method based on deep learning, and the method comprises the steps: fusing a differential safety distance standard, seasonal risk assessment parameters and a real-time risk assessment result to obtain a comprehensive risk level, recognizing a fault risk state based on a potential short circuit position, and determining an early warning level; the early warning level is dynamically adjusted based on the fault risk state in combination with geographical environment data, and a targeted early warning strategy for bird excrement and body contact is selected and formulated according to bird migration paths corresponding to different geographical environments; and comparing the early warning strategy suggestions with the actually occurring bird excrement or body contact short-circuit fault event to obtain the matching degree of the actual fault and the early warning result, and dynamically generating a self-adaptive intelligent bird trouble fault early warning system according to the matching degree.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a distribution network line fault early warning analysis method based on deep learning. BACKGROUND

[0002] As an important part of the power system, the safe and stable operation of the distribution network line is directly related to the reliability and economic benefits of power supply. Bird damage fault has become one of the main reasons for the tripping and power failure of the distribution network line, which seriously threatens the safe operation of the power grid, and an effective early warning mechanism is urgently needed to reduce the failure rate. The traditional bird damage protection method mainly relies on physical isolation and regular inspection, and these passive protection measures have obvious shortcomings. The physical protection device has limited coverage and high maintenance cost, and manual inspection is inefficient and difficult to achieve all-weather monitoring, which cannot timely discover abnormal bird activities. More importantly, the existing method lacks in-depth analysis of bird behavior patterns and cannot predict potential fault risk points. The core challenge of distribution network bird damage early warning lies in the accurate identification of the complex spatio-temporal correlation between bird seasonal activity patterns and line faults. The size difference of different bird species leads to significant differences in their requirements for line safety distance. Large birds such as hawks have larger wing spans and wider body contact ranges, requiring a larger safety distance, while small birds have lower safety distance requirements but higher activity frequency and longer habitat time. This differentiated safety distance requirement further leads to seasonal fluctuations in bird droppings pollution and body contact short-circuit probability. During the nesting period in spring and the migration period in autumn, bird activity is intense, and the frequency of long-term habitat near insulators and conductors increases significantly, increasing the risk of insulator surface pollution by bird droppings and direct contact between birds and live equipment. While the habitat behavior is relatively stable in winter, there is still concentrated excretion pollution caused by warming aggregation. The complexity of bird seasonal activity patterns makes it difficult to adapt to the risk characteristics of excretion pollution and contact short-circuit in different periods, and the bird activity patterns in different geographical environments and migration path regions differ significantly, requiring the early warning system to have strong environmental adaptability and dynamic adjustment capability. Therefore, how to build an intelligent bird damage fault early warning system that can accurately identify bird seasonal activity patterns, dynamically analyze the safety distance requirements of different bird species, real-time assess seasonal excretion pollution and body contact short-circuit risk, and adaptively adjust the early warning strategy according to the geographical environmental characteristics, has become a key problem to ensure the safe operation of the distribution network line. SUMMARY

[0003] The present application provides a distribution network line fault early warning analysis method based on deep learning, mainly including:

[0004] The monitoring image of bird activity around the power distribution line is acquired, different bird species are identified and classified and marked through image recognition, a safety distance parameter library of bird body contacting equipment is established according to the body size, a differentiated safety distance threshold is set, and a differentiated safety distance standard based on the bird body size is obtained;

[0005] The bird activity frequency is identified according to historical monitoring data, the activity density change in the spring nesting period and the autumn migration period is identified, a bird seasonal activity mode database is established, the short circuit risk weight coefficient is adjusted according to the bird excretion frequency and habitat duration characteristics in different seasons, and seasonal risk assessment parameters are determined;

[0006] Bird habitat behavior data and flight trajectory are collected in real time, the residence time and excretion behavior of birds near key equipment such as insulators, conductors and transformers are monitored, the probability value of short circuit caused by bird excrement or body contact is calculated, and the potential short circuit position is predicted;

[0007] The comprehensive risk level is obtained by fusing the differentiated safety distance standard, the seasonal risk assessment parameters and the real-time risk assessment result, the fault risk state is identified based on the potential short circuit position, and the warning level is determined;

[0008] The spatio-temporal distribution law of bird excrement and body contact fault and the bird behavior characteristics are obtained by analyzing historical fault data and bird seasonal activity mode, the high-risk period and the regional distribution characteristics are identified, the geographical environment data is acquired, and the bird migration path selection corresponding to different geographical environments is analyzed;

[0009] The warning level is dynamically adjusted based on the fault risk state and the geographical environment data, and the targeted warning strategy for bird excrement and body contact is formulated according to the bird migration path selection corresponding to different geographical environments;

[0010] The matching degree of actual fault and warning result is obtained by comparing the warning strategy suggestion and the actual bird excrement or body contact short circuit fault event, and the adaptive intelligent bird damage fault warning system is dynamically generated according to the matching degree.

[0011] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0012] The application discloses a distribution network line fault early warning analysis method based on deep learning, and aims at the core problems that the safety distance standard is not unified due to the body size difference of different bird species in the distribution network line, and the fault risk assessment is inaccurate due to the seasonal activity law change, the image recognition technology is used to classify and mark the bird body size, and a differentiated safety distance parameter library is established, a seasonal activity mode database is established by combining the activity density change law of the spring nesting period and the autumn migration period identified according to the historical monitoring data, the habitat behavior and flight trajectory data of birds near the key equipment are collected in real time by using the infrared sensor and the video monitoring equipment, the short-circuit probability is calculated and the potential fault position is predicted by fusing the differentiated safety distance standard, the seasonal risk assessment parameter and the real-time monitoring result, the early warning level is dynamically adjusted by analyzing the correlation between the historical fault data and the geographical environmental factors, finally, the adaptive optimization of the risk assessment parameter is realized by comparing the matching degree of the early warning result and the actual fault event, and the accuracy and the pertinence of the bird damage fault early warning are significantly improved BRIEF DESCRIPTION OF DRAWINGS

[0013] Fig. 1 A flowchart of the distribution network line fault early warning analysis method based on deep learning.

[0014] Fig. 2 A schematic diagram of the distribution network line fault early warning analysis method based on deep learning. DETAILED DESCRIPTION

[0015] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in combination with the drawings in the specification. Obviously, the described embodiments are only some of the embodiments of the specification, not all the embodiments. Based on the embodiments in the specification, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the specification.

[0016] As Figs. 1-2 , the distribution network line fault early warning analysis method based on deep learning specifically can include:

[0017] S101, acquire the bird activity monitoring image around the distribution network line, classify and mark different bird species by image recognition, establish the safety distance parameter library of the bird body contacting equipment according to the body size, set the differentiated safety distance threshold, and obtain the differentiated safety distance standard based on the bird body size.

[0018] The bird activity monitoring image of the fixed monitoring point around the distribution network line is acquired, the YOLOv5 target detection algorithm is used to identify the bird contour in the image, the body length and wingspan data of the bird are extracted, the birds are divided into three body size classifications according to the body length, and the bird classification result containing the body size classification mark and the corresponding wingspan data is obtained. For the wingspan data of each body size classification in the bird classification result, the risk coefficient of the bird body contacting the equipment is calculated by the ratio of the wingspan length to the distance between phases combined with the distance between phases and the length of the insulator string, if the risk coefficient is greater than a first preset threshold, the safety distance threshold of the body size classification is set as the wingspan length plus a first margin value, if the risk coefficient is less than the first preset threshold, the safety distance threshold is set as the wingspan length plus a second margin value, and the safety distance threshold corresponding to each body size classification is obtained. According to the safety distance threshold corresponding to each body size classification, a parameter library containing the body size classification mark, the average wingspan length and the safety distance threshold is established, the differentiated safety distance thresholds of the large birds, the medium birds and the small birds are stored respectively, and the differentiated safety distance standard based on the bird body size is formed.

[0019] For example, the bird activity monitoring around the distribution network line needs to accurately identify birds of different body sizes, because the threat degree of different body size birds to power equipment is significantly different. Large birds such as heron, white egret and the like have wingspan of more than 1.5 meters, which can easily cause short circuit between phases during take-off or landing; while small birds such as sparrows, swallows and the like have wingspan usually within 30 centimeters, and the main threat is nest building behavior. By deploying high-definition cameras at key nodes of the distribution network line, bird activity images can be collected all day long.

[0020] In a possible implementation, the YOLOv5 target detection algorithm can accurately identify the bird contour from a complex background. The algorithm extracts image features through a convolutional neural network, identifies the head, torso and double-wing boundaries of the bird, and then calculates the body length and wingspan data of the bird. When a bird is detected, the algorithm will label a rectangular frame in the image, the width of the frame is the wingspan length, and the height reflects the body length. According to the body length data, the birds can be divided into three categories: birds with body length exceeding 50 centimeters are large birds, birds with body length of 20 to 50 centimeters are medium birds, and birds with body length below 20 centimeters are small birds. This classification method fully considers the positive correlation between the body size of the bird and the threat degree to the power equipment.

[0021] It should be noted that the calculation process of the risk coefficient is directly related to the accuracy of the safety distance threshold. The risk coefficient is equal to the ratio of the wing span length of birds to the distance between the distribution network lines, which reflects the probability of short circuit caused by birds flying or nesting. When the average wing span of a certain large bird is 1.2 meters, and the distance between the 10-kilovolt distribution network lines is 1.0 meter, the risk coefficient is 1.2, which is obviously higher than the safety threshold. At this time, a larger safety distance threshold needs to be set, that is, more margin is added to the wing span length to ensure that even if the bird is in the most unfavorable position, it will not touch the two-phase conductor. On the contrary, the risk coefficient of small birds is usually below 0.3, and the safety distance threshold can be relatively small, which not only ensures safety, but also avoids the increase in cost caused by excessive protection.

[0022] Specifically, the establishment process of the parameter library fully embodies the concept of differentiated management. For large birds, the average wing span of 1.3 meters is recorded in the parameter library, and the corresponding safety distance threshold is set to 1.8 meters; the average wing span of medium-sized birds is 0.6 meters, and the safety distance threshold is 0.9 meters; the average wing span of small birds is 0.25 meters, and the safety distance threshold is 0.4 meters. This differentiated safety distance standard not only improves the targeting of protection, but also provides a scientific basis for the subsequent arrangement of bird repelling devices and the selection of insulation protection measures. By calling the data in the parameter library in real time, the operation and maintenance personnel can quickly judge the risk level according to the type of birds monitored, and take appropriate protective measures, thereby significantly reducing the bird damage failure rate.

[0023] S102, identify the bird activity frequency according to the historical monitoring data, identify the activity density change in the spring nesting period and the autumn migration period, establish a bird seasonal activity mode database, adjust the short circuit risk weight coefficient according to the bird excretion frequency and the nesting time length characteristics in different seasons, and determine the seasonal risk assessment parameters.

[0024] The bird occurrence time stamp and location information are extracted from the historical monitoring data, the number of bird occurrences per day is counted, the monthly average activity frequency is obtained by accumulating the number of daily occurrences and dividing by the number of days, the peak value of the activity frequency in the spring nesting period and the peak value of the activity frequency in the autumn migration period are identified, and the bird seasonal activity frequency distribution data containing the month identifier and the corresponding activity frequency value are obtained. According to the frequency value sequence in the bird seasonal activity frequency distribution data, the time variation law of the activity density is analyzed by using the autoregressive moving average model, the activity density growth rate is calculated by dividing the difference value of the activity density of adjacent months by the time interval, the peak duration parameter is extracted, and the bird seasonal activity mode database containing the season identifier, the activity density value, the growth rate and the change trend is established. The activity density value and the change trend of each season are read from the bird seasonal activity mode database, the excretion frequency is calculated by combining the excretion trace statistical data extracted from the historical monitoring data, the average habitat duration is calculated by the bird residence time record, if the excretion frequency exceeds the first preset threshold value, the short-circuit risk weight coefficient is set to the second preset threshold value, otherwise it is set to the third preset threshold value, and the seasonal risk assessment parameter containing the season identifier and the corresponding risk weight coefficient is determined.

[0025] For example, the processing of the historical monitoring data is the basis for identifying the bird activity law.

[0026] In a possible implementation, the monitoring device records the bird occurrence every hour, including the specific time of occurrence, the residence duration and the number of birds. Through statistical analysis of these raw data, the activity law of the birds can be accurately mastered. The calculation method of the monthly average activity frequency is to accumulate the number of bird occurrences recorded every day, and then divide by the number of days in the month. For example, 930 bird activities are recorded in a month, and there are 31 days in the month, so the monthly average activity frequency is 30 times / day. This calculation method can eliminate the influence of incidental factors and truly reflect the activity level of the birds.

[0027] It should be noted that the spring nesting period and the autumn migration period are two peak periods of bird activity. During the nesting period, birds frequently go back and forth between the nest and the foraging ground, and the activity frequency can reach 3 to 5 times of the usual time; during the migration period, there is a large-scale cluster activity, and a large number of birds will pass through the distribution network line area in a short time. Through analysis of the historical data of many years, it is found that the activity frequency peak values in these two periods have obvious regularity, which provides a reliable basis for subsequent risk assessment.

[0028] Specifically, the autoregressive moving average model plays a key role in analyzing the changes in bird activity density. This model predicts future trends in activity density by analyzing the autocorrelation and moving average characteristics in historical time series data. The model first calculates the correlation coefficient of the activity density of the current month and the previous months to identify the periodic variation pattern of the activity density. Then, through the moving average method, the random fluctuations in the data are smoothed to extract the true trend. The calculation process of the growth rate of activity density is to subtract the activity density of the last month from the activity density of the current month, and then divide by the time interval. This growth rate directly reflects the speed of change of bird activity.

[0029] In one embodiment, the statistics of the excretion frequency are based on the analysis of the insulator surface contamination samples. Maintenance personnel regularly collect contamination samples on the surface of insulators, identify the bird droppings component through component analysis, and calculate the number of excretions per unit time combined with the sampling period. The length of stay is obtained by analyzing the continuous appearance time of birds in the monitoring images. When the same bird appears in the same location in multiple consecutive images, the start and end times of its stay are recorded. These two parameters together determine the probability of bird-caused pollution flashover failure of the line.

[0030] Preferably, the setting of the risk weight coefficient adopts a grading method. When the excretion frequency exceeds the preset threshold, it means that the contamination on the surface of the insulator accumulates quickly and is prone to cause pollution flashover failure, at which time a higher risk weight coefficient needs to be assigned. This risk assessment method based on actual monitoring data is more scientific and accurate than traditional experience-based judgment, and can provide quantitative support for operation and maintenance decisions. By establishing a complete seasonal risk assessment parameter system, precise identification and differentiated management of bird damage risks in different seasons are achieved.

[0031] S103, real-time collection of bird habitat behavior data and flight trajectory, monitoring of the stay time and excretion behavior of birds near key equipment such as insulators, conductors and transformers, calculation of the probability value of short circuit caused by bird excrement or body contact, and prediction of potential short circuit position.

[0032] The bird body temperature characteristics are detected by the infrared sensor and the video monitoring device is triggered to record, real-time image of the bird habitat behavior around the insulator, conductor and transformer is collected, the time when the bird appears, the duration of stay and the flight trajectory coordinates are recorded, and the bird activity data containing timestamp, position coordinates and device number are obtained. According to the timestamp and position coordinates in the bird activity data, the cumulative stay time of the bird on each key device is calculated, the excretion action is detected by the image recognition algorithm and the excretion occurrence time is recorded, the excretion frequency in unit time is counted, and the bird behavior characteristic parameters containing device number, stay duration, excretion frequency and excretion position are obtained. The stay duration in the bird behavior characteristic parameters is used as the exposure time, the excretion frequency is used as the pollution intensity, and the Bayesian probability calculation formula is input, wherein the prior probability is the historical short-circuit failure rate, and the likelihood function is the conditional probability of excrement leading to insulation reduction. The short-circuit probability value of each device position is calculated, and if the probability value exceeds the preset threshold, the device position is marked as a potential short-circuit position.

[0033] The cooperative working mechanism of the infrared sensor and the video monitoring device is an example and is a key to realize accurate monitoring of bird behavior.

[0034] In a possible implementation, the infrared sensor continuously scans the monitoring area, and when a body temperature feature of 37 to 42 degrees Celsius is detected, it is determined that a bird appears and a trigger signal is immediately sent to the video monitoring device. This triggering mechanism avoids the storage pressure brought by 24-hour continuous recording of the video device, while ensuring image collection at critical moments. After receiving the trigger signal, the video device starts recording high-definition images at a rate of 25 frames per second, recording the complete process of the bird entering and leaving the monitoring area.

[0035] It should be noted that the acquisition of flight trajectory coordinates depends on the analysis of consecutive frames of images by image processing technology. By identifying the change of bird position in adjacent frames of images, the moving path of the bird in three-dimensional space is calculated. Each coordinate point contains three elements: horizontal position, vertical position and timestamp. These data points are connected to form a complete flight trajectory. When the bird approaches the insulator string, the system will specially mark the closest distance between it and the device, and this distance information is crucial for subsequent risk assessment.

[0036] Specifically, the calculation of stay duration is based on the continuity of timestamp. When the coordinate change range of the bird at the same device position is less than 0.5 meters, it is determined to be in a stay state, and the system starts to accumulate the stay time. This judgment takes into account the small amplitude activity characteristics of the bird when it is staying. The identification of excretion behavior is achieved by analyzing the change of bird tail posture and the falling trajectory of excrement. The image recognition algorithm can capture the characteristic action of the bird lifting its tail and track the falling path of the excrement, so as to accurately record the excretion occurrence time and the falling point position.

[0037] In an embodiment, the application of the Bayesian probability calculation method makes full use of historical data and real-time monitoring information. The prior probability is derived from the bird damage fault statistical data of the line section in the past five years, reflecting the basic risk level of different equipment positions. The likelihood function is determined according to the laboratory test data, and the quantitative relationship between the number of excretion and the decline of insulation performance is established by simulating the influence of different amounts of bird droppings on the surface resistance of insulators. When a certain insulator position records 5 excretion behaviors within one hour, and the cumulative residence time reaches 20 minutes, the Bayesian calculation will consider these factors comprehensively, and output a probability value between 0 and 1.

[0038] Preferably, the marking of the potential short-circuit position adopts a hierarchical early warning mode. Positions with probability values between 0.3 and 0.5 are marked as yellow warning, reminding the operation and maintenance personnel to strengthen the patrol; positions with probability values between 0.5 and 0.7 are marked as orange warning, suggesting to take bird repelling measures; positions with probability values exceeding 0.7 are marked as red warning, requiring immediate cleaning or replacement of insulators. This prediction method based on probability calculation can more accurately identify high-risk positions compared to the traditional regular inspection mode, greatly improving the pertinence and efficiency of distribution network operation and maintenance, and effectively reducing the short-circuit failure rate caused by bird damage.

[0039] S104, a comprehensive risk level is obtained by fusing the differentiated safety distance standard, the seasonal risk assessment parameter and the real-time risk assessment result, a fault risk state is identified based on the potential short-circuit position, and a warning level is determined.

[0040] The distance difference value of the current position of the bird in the differentiated safety distance standard and the safety threshold, the risk weight coefficient in the seasonal risk assessment parameter and the short-circuit probability value in the real-time risk assessment result are obtained, the violation degree coefficient is obtained by dividing the distance difference value by the safety threshold, the violation degree coefficient is multiplied by the first preset weight, the risk weight coefficient is multiplied by the second preset weight, and the short-circuit probability value is multiplied by the third preset weight by using the weighted summation method, and then added, to obtain a comprehensive risk index. According to the comprehensive risk index and the potential short-circuit position marking, the fault risk state of each monitoring point is identified, if the comprehensive risk index exceeds the high-risk threshold and there is a potential short-circuit marking at the position, it is determined as a high-risk fault risk state, if the comprehensive risk index exceeds the medium-risk threshold, it is determined as a medium-risk fault risk state, otherwise, it is a low-risk fault risk state, to obtain a fault risk state record containing a position number and a risk state category. According to the risk state category in the fault risk state record, the corresponding warning level is mapped, the high-risk fault risk state is mapped to a first-level warning, the medium-risk fault risk state is mapped to a second-level warning, and the low-risk fault risk state is mapped to a third-level warning, to determine the warning level of each monitoring position.

[0041] For example, the calculation of the violation degree coefficient is a key link in evaluating the risk of bird activity.

[0042] In one possible implementation, when a large bird is monitored to be active at a position 1.2 meters away from the conductor, and the safety distance threshold of the bird is 1.8 meters, the distance difference is 0.6 meters. By dividing this difference by the safety threshold 1.8 meters, the violation degree coefficient is 0.33. This coefficient intuitively reflects the degree of danger of the current position of the bird, and the larger the value, the closer to the dangerous area. This standardized calculation allows the risk of different sizes of birds to be compared and integrated on the same scale.

[0043] It should be noted that the weighted summation method plays a role in balancing various risk factors in the calculation of the comprehensive risk index. The distribution of the three preset weights reflects the importance of different risk factors. The violation degree coefficient reflects the immediate risk in the spatial dimension, which is usually given a higher weight; the seasonal risk weight coefficient reflects the periodic risk characteristics in the time dimension, which should be considered during the active period of the bird; and the real-time short-circuit probability value is a dynamic risk assessment result based on behavior monitoring. When the violation degree coefficient of an insulator position is 0.4, the seasonal risk weight coefficient is 0.8, and the short-circuit probability value is 0.6, the comprehensive risk index obtained after weighted calculation can fully reflect the actual risk level of the position.

[0044] Specifically, the identification process of the fault risk state fully considers the multi-dimensional characteristics of the risk. The determination of the high-risk fault risk state needs to satisfy two conditions at the same time: the comprehensive risk index exceeds the high-risk threshold and there is a potential short-circuit mark. This double verification mechanism avoids misjudgment that may be caused by a single indicator. For example, although the comprehensive risk index of a certain position is high, if it is not marked as a potential short-circuit position, it means that the actual short-circuit probability of the position is still low, and therefore only the medium-risk state is determined. This hierarchical identification method not only ensures the focus on high-risk points, but also avoids excessive investment of resources.

[0045] In one embodiment, the mapping relationship of the warning level follows the emergency response specifications of the power industry. The high-risk fault risk state corresponding to the first-level warning requires immediate measures, including emergency cleaning of the insulator or installation of a bird-proof device; the medium-risk state corresponding to the second-level warning needs to strengthen the monitoring frequency and prepare emergency supplies; and the low-risk state corresponding to the third-level warning can be maintained with regular inspection. This hierarchical warning mechanism enables the operation and maintenance resources to be reasonably allocated according to the actual risk level.

[0046] Preferably, the entire risk assessment and warning process forms a complete chain from multi-dimensional risk identification to unified risk quantification and then to hierarchical response. By integrating risk information in three dimensions of spatial distance, time period, and behavior characteristics, a comprehensive assessment of the bird damage risk of the distribution network is achieved.

[0047] S105, analyze the spatiotemporal distribution of bird excrement and body contact faults and the behavior characteristics of birds based on historical fault data and seasonal activity patterns of birds, identify high-risk time periods and regional distribution characteristics, obtain geographic environment data, and analyze the migration path selection of birds corresponding to different geographic environments.

[0048] The fault occurrence time, location coordinates, and fault type information are extracted from historical fault data, the bird activity frequency data for the corresponding period is matched, the occurrence frequency of excrement pollution flashover faults and body contact short circuit faults in different months and different time periods is counted, the coincidence degree of the fault high incidence period and the bird activity peak period is identified through time series analysis, the spatiotemporal distribution heat map is drawn, and the spatiotemporal distribution rule data including the fault concentrated period, the fault dense area coordinates, and the fault type identifier are obtained. According to the fault dense area coordinates and the fault concentrated period in the spatiotemporal distribution rule data, the bird habitat duration, excretion frequency, and flight height monitoring records within the corresponding spatiotemporal range are extracted, the K-means clustering algorithm is used to group the bird behavior parameters, the behavior patterns of long-time habitat accompanied by high-frequency excretion and low-altitude crossing conductors are identified, and the high-risk time period characteristics and high-risk regional distribution characteristics are obtained. The terrain elevation, water system distribution, and vegetation coverage data of the corresponding position of the high-risk regional distribution characteristics are obtained, the correlation between the bird flight height and the terrain elevation difference is analyzed, the low-elevation channel and the waterway flight path preferred by birds are identified, and the migration path selection rules of valley terrain corresponding to valley flight path, river terrain corresponding to coastal flight path, and plain terrain corresponding to straight flight path are determined according to the matching degree of different geographic environment characteristics and bird flight path.

[0049] For example, the drawing of the spatiotemporal distribution heat map is an important means to identify high-risk bird damage areas.

[0050] In one possible implementation, the distribution area of the network configuration line is divided into 500m x 500m grid cells, and the number of historical faults in each grid is counted. When a certain grid has 15 bird damage faults in the past three years, while the adjacent grid has only 2-3 times, the grid shows deep red on the heat map, directly showing the fault concentrated area. In the time dimension, 24 hours are divided into 48 time periods, each time period is 30 minutes, and the fault occurrence frequency of each time period is counted. Through this detailed spatiotemporal analysis, the time window and spatial position of fault high incidence can be accurately located.

[0051] It should be noted that the coincidence degree analysis of the fault high incidence period and the bird activity peak period reveals the internal relationship between the two. By calculating the Pearson correlation coefficient of the fault occurrence time series and the bird activity frequency time series, when the coefficient is above 0.85, it indicates that the two have strong correlation. This correlation analysis not only verifies that bird activity is the main cause of the fault, but also provides a theoretical basis for subsequent risk prediction.

[0052] Specifically, the application process of the K-means clustering algorithm in bird behavior pattern recognition includes multiple key steps. The algorithm first represents the behavior data of each bird as a three-dimensional vector, including the length of stay, the frequency of excretion, and the flight height. By calculating the Euclidean distance from each data point to the cluster center, birds with similar behaviors are classified into the same category. After multiple iterations, 3-4 typical behavior patterns can usually be identified. Among them, the pattern of long stay accompanied by high frequency of excretion is most likely to cause insulator pollution flashover, while the pattern of low-altitude crossing of conductors directly causes interphase short circuit. The identification of such behavior patterns enables operation and maintenance personnel to take differentiated protection measures against different types of risks.

[0053] In one embodiment, the analysis of the influence of geographical environment on bird migration paths fully considers the guiding role of terrain. Due to the restriction of the two sides of the mountain, the valley terrain forms a natural flight channel, and birds tend to fly along the valley bottom to reduce energy consumption. By analyzing the terrain elevation data, when the height difference between the two sides of the valley exceeds 200 meters, the valley bottom becomes the preferred flight path for birds. River terrain provides obvious navigation signs, and birds are accustomed to flying along the riverbank, which not only facilitates the search for food but also allows for timely replenishment of water. In the plain area, due to the lack of obvious terrain features, birds mostly use straight flight to shorten the migration distance.

[0054] Preferably, superimposed analysis of geographical environment features and distribution network line orientation can accurately identify high-risk line sections. When the distribution network line crosses the valley or is laid along the river, the degree of coincidence with the bird migration path significantly increases. This risk assessment method based on geographical environment is more scientific and accurate compared to traditional experience-based judgment.

[0055] S106, dynamically adjust the warning level based on the fault risk state combined with geographical environment data, and develop targeted warning strategies for bird excrement and body contact according to the bird migration path selection corresponding to different geographical environments.

[0056] The risk level identifier and corresponding position coordinates in the fault risk state are acquired, the terrain feature data and bird flight path type of the position are matched, the warning level is adjusted according to the combination relationship between the terrain feature and the risk level, if the position is in a valley channel and there is a high-risk fault risk state, the warning level is raised to a higher level, if the position is away from the main flight path, the original warning level is maintained, and the adjusted warning level after fusing the geographical environment factors is obtained. According to the combination of the adjusted warning level and the flight path type, the main risk source causing the fault is identified, for the position along the river flight path and the adjusted warning level is high, the excrement pollution flashover is determined as the main risk type, for the position in the valley flight path and the adjusted warning level is high, the body contact short circuit is determined as the main risk type, and the classification warning information containing the position number, the risk type and the protection key point is obtained. The risk type and the protection key point in the classification warning information are used to generate corresponding warning content parameters, the excrement pollution flashover risk corresponds to the insulation protection type warning parameter, and the body contact short circuit risk corresponds to the distance protection type warning parameter. The warning content parameters are associated with the position information, and the targeted warning strategy for bird excrement and body contact is developed.

[0057] For example, the dynamic adjustment mechanism of the geographical environment factor on the warning level embodies the concept of refined management.

[0058] In a possible implementation, the valley channel becomes a necessary route for bird migration due to its special topographic restrictions. When the distribution network line happens to cross the valley, the bird flight height is compressed, and the vertical distance from the line is significantly reduced. At this time, even if the original risk level is medium, considering the influence of the terrain factor, the system will automatically raise the warning level by one level. This adjustment is not a simple addition or subtraction operation, but an empirical rule based on a large amount of historical data statistics. Through analysis, it is found that the fault rate of the line section in the valley terrain is 2.5 times higher than that in the plain area, which fully illustrates the necessity of geographical environment adjustment.

[0059] It should be noted that the identification process of the main risk source needs to consider information in multiple dimensions. The characteristic of the river flight path is that the bird flight height is relatively stable, but the stay time is longer, because the river bank is often the foraging area of birds. In this environment, the frequent excretion behavior of birds becomes the main threat. Statistical data shows that 80% of the faults of the line section along the river are caused by insulator pollution flashover, and only 20% are caused by body contact. On the contrary, the valley flight path has a narrow space, and the probability of birds crossing the line increases significantly, and body contact becomes the dominant risk. This risk classification method based on environmental characteristics makes the subsequent protection measures more targeted.

[0060] Specifically, the setting of insulation protection type early warning parameters fully considers the progressive characteristics of pollution flashover development. This type of parameters includes pollution accumulation rate, critical pollution degree threshold and cleaning cycle recommendation value. When the pollution accumulation rate of a certain insulator reaches 0. 1 mg / cm2 per month , the system will calculate the time to the critical value according to the current pollution level. If the critical value is expected to be reached within 30 days, an orange early warning will be issued; if within 15 days, it will be upgraded to a red early warning. This early warning method based on the accumulation process provides sufficient response time for operation and maintenance personnel.

[0061] In one embodiment, the spacing protection type early warning parameters focus on the risk prevention and control in the spatial dimension. This type of parameters includes bird activity height distribution, conductor spacing margin and protection device coverage range. Through the analysis of bird flight trajectory data, it is found that 90% of the crossing behavior occurs in the height interval of 0. 5 to 1. 5 meters above the conductor. Based on this rule, the system will evaluate whether the existing protection device can effectively cover this dangerous area. When the coverage rate is less than 70%, the early warning parameter will suggest adding bird spikes or insulating sheath and give specific installation position coordinates.

[0062] Preferably, the development process of the targeted early warning strategy realizes the transition from passive response to active prevention. Through the deep integration of information in three dimensions of geographical environment, bird behavior and equipment characteristics, a complete risk assessment and early warning system is formed. This strategy not only can accurately identify different types of risks, but also can provide differentiated protection recommendations according to the risk characteristics. Practice has proved that after the implementation of the targeted early warning strategy, the bird damage failure rate has decreased by 65%, fully verifying the effectiveness and practicality of the method.

[0063] S107、Through the comparison of early warning strategy suggestions and actual bird excrement or body contact short circuit failure events, the actual failure and early warning result matching degree is obtained, and an adaptive intelligent bird damage failure early warning system is dynamically generated according to the matching degree.

[0064] The pre-warning position, time and type information in the pre-warning record and the position, time and type information of the actual bird excrement pollution flashover failure or body contact short circuit failure are compared for consistency, the number of failures hit by the pre-warning is counted, the recall rate is calculated by dividing the number of hits by the total number of actual failures, the accuracy rate is calculated by dividing the number of hits by the total number of pre-warnings, and the pre-warning result matching degree is calculated by the harmonic mean of the recall rate and the accuracy rate. According to the pre-warning result matching degree value, the contribution degree of each type of parameter to the pre-warning deviation is analyzed, if the matching degree is lower than the first preset threshold, the specific cases causing the missed report and the false report are extracted, the parameters that need to be adjusted in the safety distance weight, the seasonal risk weight and the real-time evaluation weight are identified, the corresponding weight value is increased or decreased according to the deviation type, and an updated weight coefficient set containing new weight values of each parameter is obtained. The comprehensive risk index is recalculated by using the updated weight coefficient set, the new pre-warning result is continuously compared with the subsequent actual failure, and the gradient descent algorithm is used to iteratively adjust each weight value by taking the matching degree as the objective function, so that the system automatically optimizes the parameter configuration according to the actual feedback, and forms a bird damage failure pre-warning system with a self-adaptive weight adjustment mechanism.

[0065] For example, the calculation of the pre-warning result matching degree uses the harmonic mean instead of the simple average, which is based on the balance between the recall rate and the accuracy rate.

[0066] In one possible implementation, the system issued 50 pre-warnings for a certain distribution network line section in a month, 35 of which actually occurred bird damage failures within 72 hours after the pre-warning, which are the number of hits. During the same period, 40 bird damage failures actually occurred on the line section, 35 of which were pre-warning hits and 5 were missed reports. According to this, the recall rate is 35 / 40 = 0.875, and the accuracy rate is 35 / 50 = 0.7. The harmonic mean is calculated as 2 x recall rate x accuracy rate / (recall rate + accuracy rate), and the matching degree is 0.778. This calculation method can effectively punish extreme cases and avoid excessive pre-warning with high recall rate and low accuracy rate.

[0067] It should be noted that the analysis of the deviation contribution degree needs to go deep into each specific case. When a missed report event is found, the system will backtrack the evaluation parameters of the location before the failure. For example, a pollution flashover failure occurred at an insulator location during the migration period in autumn, but the system failed to pre-warn. Through analysis, it is found that the safety distance evaluation of the location is normal, and the real-time monitoring also captures the bird activity, but the seasonal risk weight is only 0.2, which causes the comprehensive risk index to fail to reach the pre-warning threshold. This indicates that the weight setting of the seasonal factor is too low and needs to be increased. Similarly, if a location frequently misreports, it may be caused by an excessively high geographical environment weight, and the valley terrain makes the location assigned an excessively high basic risk value.

[0068] Specifically, the application process of the gradient descent algorithm in weight optimization embodies the core idea of machine learning. The algorithm takes matching degree as the objective function and various weights as the parameters to be optimized. In each iteration, the algorithm calculates the matching degree under the current weight configuration and then adjusts the weights in the direction where the matching degree increases the fastest. For example, if the current safe distance weight is 0.4, the seasonal risk weight is 0.3, and the real-time evaluation weight is 0.3, the matching degree is 0.75. The algorithm calculates that increasing the seasonal risk weight by 0.05 will result in the largest increase in matching degree, so the corresponding adjustment is made. After multiple iterations, the weight configuration gradually converges to the optimal value.

[0069] In an embodiment, the implementation of the adaptive mechanism relies on a continuous feedback loop. The system calculates the latest matching degree every week by counting the effectiveness of the early warning in that week. If the matching degree is below 0.7 for two consecutive weeks, the system will start the weight adjustment process. This dynamic adjustment mechanism enables the system to adapt to external factors such as seasonal changes and changes in bird behavior patterns. For example, when the spring nesting season begins, bird behavior patterns change significantly, and the original weight configuration may no longer be applicable. Through adaptive adjustment, changes can be quickly identified and the weight configuration can be adjusted accordingly.

[0070] Preferably, this adaptive early warning system realizes the transition from static rules to dynamic optimization. Traditional early warning systems often rely on fixed thresholds and weights, making it difficult to cope with complex and changing real-world situations. By introducing a feedback mechanism and automatic optimization algorithm, the system can continuously learn from practice and continuously improve the effectiveness of early warning. Practical applications show that after adopting the adaptive mechanism, the early warning matching degree has increased from the initial 0.65 to more than 0.85, significantly reducing the false alarm and false alarm rates, and providing more reliable protection for the safe operation of the distribution network.

[0071] The above description is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, as described in the specification and drawings of the present application, are also included in the patent protection scope of the present application.

Claims

1. A deep learning-based distribution line fault early warning analysis method, characterized in that, The method comprises: acquiring bird activity monitoring images around the distribution network line, identifying different bird body types and classifying and marking, and determining differentiated safety distance standards according to the body types; identifying bird activity frequency according to historical monitoring data, analyzing activity density changes in the spring nesting period and the autumn migration period, establishing a bird seasonal activity mode database, adjusting short-circuit risk weight coefficients according to seasonal bird excretion frequency and habitat duration, and determining seasonal risk assessment parameters; real-time collection of bird habitat behavior data and flight trajectories through sensors and monitoring equipment, monitoring of bird residence time and excretion behavior near key equipment, calculation of short-circuit probability values in combination with the seasonal risk assessment parameters, and prediction of potential short-circuit locations; fusion of differentiated safety distance standards, the seasonal risk assessment parameters and real-time risk assessment results, determination of comprehensive risk levels, identification of fault risk states based on the potential short-circuit locations, and determination of warning levels; analysis of historical fault data and bird seasonal activity modes, identification of bird excrement and body contact fault spatiotemporal distribution rules, and analysis of bird migration path selection in combination with geographic environment data; adjustment of warning levels based on fault risk states and geographic environment data, and development of targeted warning strategies according to the bird migration path selection; calculation of matching degrees by comparing the warning strategies with actual fault events, and adjustment of risk assessment parameter weight coefficients according to the matching degrees. 2.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The acquisition of bird activity monitoring images around the distribution network line, the identification of different bird body types and the classification and marking, and the determination of differentiated safety distance standards comprise: acquiring bird activity images around the monitoring points of the distribution network line, extracting bird body length and wing span data, classifying birds into large, medium and small body types according to the body length, and obtaining classification results containing body type markers; determination of each body type safety distance threshold value by calculating the wing span length to the distance between phases ratio according to the wing span data of each body type in the classification results and in combination with the distance between phases of the distribution network line; establishment of a parameter library containing body type markers and safety distance threshold values according to the safety distance threshold values, and storage of differentiated safety distance standards of large, medium and small birds. 3.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The identification of bird activity frequency according to historical monitoring data, the analysis of activity density changes in the spring nesting period and the autumn migration period, and the establishment of a bird seasonal activity mode database comprise: extraction of bird appearance time and position in historical monitoring data, statistics of daily appearance times, calculation of monthly average activity frequency, determination of activity frequency peaks in the spring nesting period and the autumn migration period, and obtaining of bird seasonal activity frequency distribution data; analysis of activity density time changes according to the frequency distribution data, calculation of activity density difference values between adjacent months, extraction of peak duration, and establishment of a database containing seasonal markers and activity density values. 4.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The real-time collection of bird habitat behavior data and flight trajectories through sensors and monitoring equipment, and the monitoring of bird residence time and excretion behavior near key equipment comprise: triggering of monitoring equipment by sensors to detect bird body temperature, collection of bird habitat behavior images around key equipment, and recording of bird appearance time and residence duration; According to the behavior image, a defecation action is identified, and a defecation frequency in a unit time is counted to obtain a characteristic parameter including a device number and a defecation frequency. 5.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The fusion of the differentiated safety distance standard, the seasonal risk assessment parameter and the real-time risk assessment result determines the comprehensive risk level, including: Obtain the distance difference between the current position of the bird in the differentiated safety distance standard and the threshold value, the weight coefficient in the seasonal risk assessment parameter and the real-time short-circuit probability value, calculate the violation degree coefficient, weighted sum to obtain the comprehensive risk index, form the comprehensive risk level. 6.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The analysis of historical failure data and bird seasonal activity patterns identifies the spatio-temporal distribution rule of bird excrement and body contact failure, and analyzes the bird migration path selection combined with geographic environment data, including: Extract the failure time and location in the historical failure data, match the activity frequency in the bird seasonal activity pattern, count the failure frequency in different periods, identify the coincidence degree of failure high incidence period and bird activity peak, and generate spatio-temporal distribution data; According to the spatio-temporal distribution data, extract the bird behavior parameters, and identify the high-risk behavior mode; Obtain the geographic environment data, analyze the correlation between bird flight height and terrain characteristics, and determine the migration path selection rule corresponding to different terrains. 7.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The adjustment of the early warning level based on the failure risk state and the geographic environment data, including: Obtain the risk level and location in the failure risk state, match the terrain characteristics and bird flight path type, and adjust the early warning level according to the combination of terrain and risk level. 8.The deep learning-based power distribution line fault pre-warning analysis method of claim 1, wherein, The matching degree is calculated by comparing the early warning strategy with the actual failure event, and the weight coefficient of the risk assessment parameter is adjusted according to the matching degree, including: Obtain the position and type information in the early warning strategy, match the actual failure information, count the number of early warning hits, calculate the recall rate and accuracy, and obtain the matching degree; According to the matching degree, analyze the parameter deviation, adjust the safety distance, seasonal risk and real-time evaluation weight, and iteratively optimize the weight configuration.

Images