An adaptive learning method and system for underground cable breakage risk assessment

Abstract

The application discloses a kind of adaptive learning method and system for underground cable breakage risk assessment, it is related to cable breakage risk assessment field;The method comprises: constructing the state evolution vector of M cable breakage events, the first similarity of K state evolution clusters and real-time evolution vector is determined to the near neighbor clustering of M state evolution vectors, generates K state evolution cluster, according to K first similarity, constructs similar cluster sequence, in similar cluster sequence, the variance of several relative time area numbers in Q similar clusters is calculated, according to the variance of several relative time area numbers in Q similar clusters, determine adaptive learning cluster, determine the second similarity in adaptive learning cluster, and calculate breakage risk index based on second similarity;The application generates breakage risk index based on the weighted relationship of the L2 norm of each historical state evolution vector in adaptive learning cluster and second similarity, realizes the intuitive quantification of breakage risk assessment.

Description

Technical Field

[0001] This invention relates to the field of cable damage risk assessment, specifically to an adaptive learning method and system for underground cable damage risk assessment. Background Technology

[0002] In the operation and maintenance of underground cables, accurately assessing their damage risk is crucial to ensuring the safe and stable operation of the power system. Existing risk assessment methods mostly rely on preset thresholds or machine learning models to classify and predict historical fault data. However, these methods have the following limitations:

[0003] First, a single threshold judgment is difficult to adapt to complex and ever-changing environments and load conditions. Supervised learning methods rely on a large number of labeled fault samples, but in actual engineering, cable breakage events are rare and labeling costs are high.

[0004] Secondly, some methods attempt to use historical similar events for analogical reasoning, but they usually directly calculate the similarity between the current state and all historical events without summarizing the historical events, which makes the matching results susceptible to interference from individual abnormal events and lacks stable interpretability.

[0005] Furthermore, existing technologies generally ignore the periodic patterns of cable breakage events over time. If only state similarity matching is relied upon, events that occur by chance and are not reproducible may be selected, thus weakening the reliability of risk warnings.

[0006] Therefore, there is an urgent need for an adaptive learning method that can integrate state evolution paths with temporal patterns, which can extract typical risk patterns from historical data and identify highly reliable evolution paths with periodic recurrence characteristics, thereby achieving stable and interpretable underground cable damage risk assessment without the need for a large number of labeled samples. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an adaptive learning method and system for assessing the risk of damage to underground cables. It solves the technical problems mentioned in the background by introducing an adaptive learning cluster of state evolution vectors and relative time zone numbers.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] An adaptive learning method for assessing the risk of damage to underground cables includes the following steps:

[0010] S1. Construct the state evolution vectors for M cable breakage events;

[0011] S2. Perform nearest neighbor clustering on the M state evolution vectors to generate K state evolution clusters;

[0012] Among them, the state evolution cluster consists of the cluster center vector and the multiple state evolution vectors contained therein, which are used to represent similar state evolution paths;

[0013] S3. Determine the first similarity between the K state evolution clusters and the real-time evolution vector;

[0014] S4. Construct a sequence of similar clusters based on the K first similarity scores;

[0015] S5. In a sequence of similar clusters, calculate the variance of several relative time zones within Q similar clusters;

[0016] S6. Determine the adaptive learning clusters based on the variance of several relative time zone numbers within Q similar clusters;

[0017] S7. Determine the second similarity within the adaptive learning cluster, and calculate the damage risk index based on the second similarity;

[0018] Determining the second similarity within the adaptive learning cluster includes:

[0019] S7-A1: Extract the J state evolution vectors corresponding to the adaptive learning cluster;

[0020] S7-A2, Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

[0021] In some specific embodiments, a state evolution vector for M cable breakage events is constructed, including:

[0022] S1-1. On the historical timeline, anchor the timestamps of M cable breakage events;

[0023] S1-2. For any broken timestamp, take it as the event endpoint, trace back forward on the historical timeline, and extract a broken interval window with a time length of L.

[0024] S1-3. Construct state evolution vectors within M damage time windows.

[0025] In some specific embodiments, state evolution vectors within M damage time windows are constructed, including:

[0026] S1-3-1. Within the damaged section window, collect several cable status parameters at L-1 timestamps.

[0027] S1-3-2. For any timestamp within the damaged interval window, preprocess several cable status parameters to generate several cable status features for the timestamp.

[0028] S1-3-3: Concatenate several cable status features of the timestamp to generate a composite status feature corresponding to the timestamp.

[0029] S1-3-4. Concatenate L-1 composite state features to construct the state evolution vector corresponding to the damaged interval window;

[0030] S1-3-5. Traverse the timestamps of the M cable breakage events and repeatedly execute the construction of the state evolution vector until M state evolution vectors are constructed.

[0031] In some specific embodiments, determining the first similarity between K clusters of state evolution and the real-time evolution vector includes:

[0032] S3-1. Calculate the cluster center vectors of K state evolution clusters;

[0033] S3-2. Collect the neighboring state parameters of underground cables at L timestamps in real time and construct them into a real-time evolution vector;

[0034] S3-3. Calculate the similarity between the cluster center vectors of the K state evolution clusters and the real-time evolution vectors to determine the K first similarity values;

[0035] In some specific embodiments, a sequence of similar clusters is constructed based on K first similarities, including:

[0036] S4-1. Based on K first similarities, assign descending numbers to K state evolution clusters, and generate a state evolution cluster sequence based on the descending numbers;

[0037] S4-2. Extract the state evolution clusters corresponding to the first Q descending numbers from the state evolution cluster sequence;

[0038] S4-3. Define the state evolution clusters corresponding to the first Q descending numbers as similar cluster sequences; wherein, the similar cluster sequence inherits the Q descending numbers of its state evolution clusters;

[0039] In some specific embodiments, in a sequence of similar clusters, the variance of several relative time zones within Q similar clusters is calculated, including:

[0040] S5-1. Based on Q descending numbers, obtain several damaged interval windows corresponding to several state evolution vectors within Q similar clusters in the similar cluster sequence, number by number.

[0041] S5-2. Extract several central timestamps from several damaged interval windows and encode them as several relative time zone numbers within the period;

[0042] S5-3. Collect several relative time zone numbers and calculate the variance of the several relative time zone numbers in the set;

[0043] S5-4. Traverse the descending numbers of the Q similar clusters and repeat S5-1 to S5-3 until the Q variances within the Q similar clusters are obtained.

[0044] In some specific embodiments, the adaptive learning clusters are determined based on the variance of several relative time zone numbers within Q similar clusters, including:

[0045] S6-1. Select the smallest variance from the Q variances;

[0046] S6-2. Based on the minimum variance, select the corresponding similar cluster from the Q similar clusters, and denot it as the adaptive learning cluster;

[0047] In some specific embodiments, a damage risk index is calculated based on a second similarity, including:

[0048] S7-B1, Normalize the J second similarities into J similarity weights;

[0049] S7-B2, Calculate the L2 norm of J state evolution vectors;

[0050] S7-B3: Based on the L2 norm weighted summation of J state evolution vectors using J similarity weights, a damage risk index is generated.

[0051] This invention provides an adaptive learning method and system for assessing the risk of damage to underground cables, which has the following advantages:

[0052] This invention generates K state evolution clusters by performing nearest-neighbor clustering on the state evolution vectors of M historical cable breakage events, and uses the cluster center vector to represent the evolution of each state evolution cluster. This reduces the dependence on individual abnormal historical events, and makes the construction of similar cluster sequences based on the evolution path formed by multiple cable breakage events, rather than the individual characteristics of a single event.

[0053] Furthermore, by calculating the variance of the relative time zone number of the damage events within each similar cluster in the similar cluster sequence, and selecting the cluster with the smallest variance as the adaptive learning cluster, the similar cluster with the most concentrated time distribution in the long-term operation cycle was selected. The damage events in the selected cluster occur more concentratedly in the long-term operation cycle, reflecting the evolutionary characteristics of repeated occurrences in a fixed period of time.

[0054] Furthermore, this invention generates a damage risk index based on the weighted relationship between the L2 norm of each historical state evolution vector within the adaptive learning cluster and the second similarity, where the L2 norm reflects the overall intensity of state parameter fluctuations, thus achieving an intuitive quantification of damage risk assessment.

[0055] Secondly, the present invention provides an adaptive learning system for assessing the risk of damage to underground cables, comprising:

[0056] A vector construction unit is used to construct state evolution vectors for M cable breakage events;

[0057] Clustering unit, used to perform nearest-neighbor clustering on M state evolution vectors to generate K state evolution clusters;

[0058] A similarity calculation unit is used to determine the first similarity between the K state evolution clusters and the real-time evolution vector;

[0059] Sequence building units are used to construct similar cluster sequences based on K first similarities;

[0060] The variance calculation unit is used to calculate the variance of several relative time zones within Q similar clusters in a similar cluster sequence.

[0061] An adaptive learning unit is used to determine an adaptive learning cluster based on the variance of several relative time zone numbers within Q similar clusters.

[0062] A risk assessment unit is used to determine the second similarity within the adaptive learning cluster and calculate the damage risk index based on the second similarity.

[0063] Determining the second similarity within the adaptive learning cluster includes:

[0064] Extract the J state evolution vectors corresponding to the adaptive learning cluster;

[0065] Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

[0066] Compared with the prior art, the beneficial effects of the adaptive learning system for underground cable damage risk assessment of the present invention are the same as those of the adaptive learning method for underground cable damage risk assessment described above, so they will not be repeated here. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating an adaptive learning method for assessing the risk of damage to underground cables according to the present invention.

[0068] Figure 2 This is a schematic diagram of the calculation process for the second similarity in this invention;

[0069] Figure 3 This is a schematic diagram illustrating the definition process of similar cluster sequences as described in this invention;

[0070] Figure 4 This is a schematic diagram of the variance calculation process for the relative time zone number described in this invention;

[0071] Figure 5 This is a schematic diagram of the process for generating the damage risk index as described in this invention;

[0072] Figure 6This is a structural block diagram of an adaptive learning system for assessing the risk of damage to underground cables, according to the present invention. Detailed Implementation

[0073] 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.

[0074] Please see Figures 1 to 5 This invention provides an adaptive learning method for assessing the risk of damage to underground cables, comprising the following steps:

[0075] S1. Construct the state evolution vectors for M cable breakage events;

[0076] S2. Perform nearest neighbor clustering on the M state evolution vectors to generate K state evolution clusters;

[0077] Among them, the state evolution cluster consists of the cluster center vector and the multiple state evolution vectors contained therein, which are used to represent similar state evolution paths;

[0078] S3. Determine the first similarity between the K state evolution clusters and the real-time evolution vector;

[0079] S4. Construct a sequence of similar clusters based on the K first similarity scores;

[0080] S5. In a sequence of similar clusters, calculate the variance of several relative time zones within Q similar clusters;

[0081] S6. Determine the adaptive learning clusters based on the variance of several relative time zone numbers within Q similar clusters;

[0082] S7. Determine the second similarity within the adaptive learning cluster, and calculate the damage risk index based on the second similarity.

[0083] In this embodiment, by constructing state evolution vectors for M cable damage events, nearest neighbor clustering is performed on the M state evolution vectors to obtain K state evolution clusters. The first similarity between each state evolution cluster and the real-time evolution vector is calculated. The clusters are arranged in descending order according to the first similarity, and the top Q state evolution clusters are selected to form a similar cluster sequence. For each similar cluster in the similar cluster sequence, the variance of the normalized time of multiple damage events within it is calculated, and the similar cluster with the smallest variance is selected as the adaptive learning cluster. Within the adaptive learning cluster, the second similarity between the real-time evolution vector and each historical state evolution vector is calculated. After normalization, the similarity weight is obtained. The similarity weight is then calculated by combining the L2 norm of each state evolution vector and performing a weighted summation to generate a damage risk index, thereby quantifying the current damage risk level.

[0084] In this embodiment, step S1 specifically includes:

[0085] S1-1. On the historical timeline, anchor the timestamps of M cable breakage events;

[0086] For example, the cable damage event refers to a physical damage event to an underground cable confirmed by sensor monitoring, maintenance records, or manual inspection, including local damage, breakage, short circuit, grounding fault, or significant degradation of insulation performance.

[0087] S1-2. For any broken timestamp, take it as the event endpoint, trace back forward on the historical timeline, and extract a broken interval window with a time length of L.

[0088] The damaged interval window represents the historical observation interval of L time units before the damage event occurred.

[0089] S1-3. Construct state evolution vectors within M damage time windows.

[0090] In this embodiment, by anchoring the timestamps of M cable breakage events, and taking each timestamp as the endpoint, a breakage interval window of length L is extracted by backward tracing, and the time series state data within each window is obtained to construct M state evolution vectors. Each state evolution vector represents a continuous state evolution process before the occurrence of the corresponding breakage event.

[0091] Steps S1-3 further include:

[0092] S1-3-1. Within the damaged section window, collect several cable status parameters at L-1 timestamps.

[0093] For example, some cable status parameters also include:

[0094] Environmental condition parameters, such as temperature and humidity, and soil moisture, are used to describe the basic conditions of the environment in which underground cables are located.

[0095] Health status parameters: current, voltage, and other insulation properties of the cable, used to describe the health status of the cable;

[0096] For example, the health status of the cable can be quantified in the following ways:

[0097] Current and voltage parameters are collected through conventional monitoring devices configured with cables, and their normal fluctuation range is determined based on historical normal operation data; these conventional monitoring devices include current transformers (CTs), voltage transformers (PTs), and smart meters.

[0098] The current collected value is compared with the normal range, the deviation is calculated, and the deviation is quantified as an indicator of the cable's health status.

[0099] The indicators of cable health status reflect the degree of abnormality in cable operating parameters. The greater the deviation, the worse the cable health status and the higher the potential risk of damage.

[0100] Construction status parameters: Excavation actions of ground equipment, mechanical vibration signals, etc. By extracting the intensity and duration of relevant actions and signals, corresponding cable status parameters are generated to describe the impact of external construction activities on the cable area.

[0101] S1-3-2. For any timestamp within the damaged interval window, preprocess several cable status parameters to generate several cable status features for the timestamp.

[0102] For example, the preprocessing in this embodiment includes: performing feature construction such as missing value interpolation, outlier filtering, and normalization on the original cable status parameters to eliminate dimensional differences and enhance the characterization of the cable status parameters.

[0103] S1-3-3: Concatenate several cable status features of the timestamp to generate a composite status feature corresponding to the timestamp.

[0104] S1-3-4. Concatenate L-1 composite state features to construct the state evolution vector corresponding to the damaged interval window;

[0105] Specifically, the state evolution vector is a multi-dimensional temporal feature vector, which is composed of composite state features corresponding to L-1 timestamps within the damage interval window, spliced ​​together in chronological order, and is used to characterize the state evolution path of the cable in the period before the damage occurs.

[0106] S1-3-5. Traverse the timestamps of the M cable breakage events and repeatedly execute the construction of the state evolution vector until M state evolution vectors are constructed.

[0107] In this embodiment, for L-1 timestamps within each damage interval window, environmental, health, and construction status parameters are collected, the original parameters are preprocessed accordingly, and cable status features for each timestamp are generated. Multiple status features under the same timestamp are spliced ​​into composite status features, and then L-1 composite status features are spliced ​​in chronological order to form a status evolution vector representing the status evolution process within the window. The construction of M status evolution vectors is completed by traversing M damage events.

[0108] In this embodiment, step S2 specifically includes:

[0109] S2-1. Randomly select K state evolution vectors from M state evolution vectors as the initial cluster centers;

[0110] S2-2. Anchoring any non-initial cluster center among M state evolution vectors is defined as the state anchoring vector;

[0111] S2-3. Calculate the Euclidean distance between the state anchor vector and the K initial cluster centers to obtain the K Euclidean distances of the state anchor vector;

[0112] S2-4. Select the initial cluster center corresponding to the smallest Euclidean distance from the K Euclidean distances of the state anchoring quantity;

[0113] S2-5. Based on the initial cluster centers corresponding to the minimum Euclidean distance, construct the initial clusters for the state anchoring variables;

[0114] S2-6. Traverse the M state evolution vectors and repeat S2-2 to S2-5 until all non-initial cluster centers are assigned to the initial cluster centers with the smallest Euclidean distance to them, so as to form K initial clusters.

[0115] S2-7. Calculate the centroids of all state evolution vectors in the initial cluster;

[0116] S2-8. Use the centroid of all state evolution vectors as the cluster center for the next round.

[0117] S2-9. Calculate the change distance of the cluster centers between the next round cluster centers and the initial cluster centers;

[0118] S2-10. If the change distance of the cluster center is greater than the preset threshold, then traverse the M state evolution vectors again and assign each state evolution vector to the next round cluster center that is the smallest Euclidean distance from it.

[0119] S2-11, Iterate through S2-7 to S2-10 until the cluster center change distance is less than or equal to the preset threshold;

[0120] S2-12. If the change distance of the cluster center is less than or equal to the preset threshold, then the cluster center of the current round and its assigned state evolution vector are defined as the state evolution cluster, until K state evolution clusters are generated.

[0121] In this embodiment, K randomly selected state evolution vectors are used as initial cluster centers. The Euclidean distance between the remaining state evolution vectors and each initial cluster center is calculated, and they are assigned to the corresponding clusters according to the minimum distance principle to form K initial clusters. The centroid of each initial cluster is calculated as the cluster center for the next round. The change distance of the cluster center is compared with that of a preset threshold. If it is greater, the vectors are reassigned and the cluster centers are updated. This process is repeated until the change distance of the cluster centers is less than or equal to the preset threshold. Finally, the stable cluster centers and their assigned state evolution vectors are defined as K state evolution clusters, thereby completing the grouping of M state evolution vectors and obtaining K state evolution clusters that can represent similar state evolution paths.

[0122] In this embodiment, step S3 specifically includes:

[0123] S3-1. Calculate the cluster center vectors of K state evolution clusters;

[0124] Specifically, a state evolution cluster represents a set of state evolution vectors with similar state evolution paths in a multidimensional state evolution vector space. Each cluster represents a typical risk evolution pattern before cable breakage. The cluster center vector, as the centralized representation of the cluster, is obtained by calculating the mean of all state evolution vectors within the cluster and is used to represent the typical state evolution path of this type of risk.

[0125] S3-2. Collect the neighboring state parameters of underground cables at L timestamps in real time and construct them into a real-time evolution vector;

[0126] S3-3. Calculate the similarity between the cluster center vectors of the K state evolution clusters and the real-time evolution vectors to determine the K first similarity values;

[0127] In this embodiment, the first similarity represents the cosine similarity or inverse Euclidean distance between the real-time evolution vector and each cluster center vector, reflecting the similarity between the two in the direction of state evolution; the higher the similarity, the closer the current state change trend of the cable is to this type of historical damage mode.

[0128] In this embodiment, step S4 specifically includes:

[0129] S4-1. Based on K first similarities, assign descending numbers to K state evolution clusters, and generate a state evolution cluster sequence based on the descending numbers;

[0130] S4-2. Extract the state evolution clusters corresponding to the first Q descending numbers from the state evolution cluster sequence;

[0131] S4-3. Define the state evolution clusters corresponding to the first Q descending numbers as similar cluster sequences; wherein, the similar cluster sequence inherits the Q descending numbers of its state evolution clusters;

[0132] Specifically, the similarity cluster sequence is an ordered sequence formed by arranging the first Q state evolution clusters with the highest similarity in descending order of similarity. Each state evolution cluster corresponds to a set of state evolution vectors for cable damage events.

[0133] In this embodiment, step S5 specifically includes:

[0134] S5-1. Based on Q descending numbers, obtain several damaged interval windows corresponding to several state evolution vectors within Q similar clusters in the similar cluster sequence, number by number.

[0135] S5-2. Extract several central timestamps from several damaged interval windows and encode them as several relative time zone numbers within the period;

[0136] Specifically, several central timestamps represent the midpoint between the start timestamp and the damage timestamp of each damage interval window, used to reflect the typical occurrence period of cable damage events during long-term operation; the period represents a complete time cycle unit, preferably one year or the cable's design warranty period, used to map the central timestamps to the relative time position within the period, in order to identify the concentration trend of damage events in seasonality or aging cycles.

[0137] For example, if the period is set to one year, the damage timestamp of a certain damage event is March 15, 2023, and the damage interval window length L is 60 days, then its starting timestamp is January 15, 2023, and the center timestamp is February 14, 2023. Mapping this center timestamp to the "one year" period, we get its relative time zone number as 45 (i.e., February 14 is the 45th day of the year). Perform the same operation on the center timestamps of all cable damage events in similar cluster sequences to obtain a set of relative time zone numbers within the annual period.

[0138] S5-3. Collect several relative time zone numbers and calculate the variance of the several relative time zone numbers in the set;

[0139] S5-4. Traverse the descending numbers of the Q similar clusters and repeat S5-1 to S5-3 until the Q variances within the Q similar clusters are obtained.

[0140] Specifically, the variance calculation formula is as follows:

[0141]

[0142] in, The variance of relative time zone numbers within similar clusters is represented by N, where N represents the number of cable breakage events within the similar clusters, i.e., the number of events in the set of relative time zone numbers. This represents the relative time zone number formed by the central timestamp encoding in the i-th cable breakage event. This represents the average of N relative time zones.

[0143] Specifically, variance represents the degree of concentration of the relative time zone (i.e. the period of occurrence) of each cable damage event within a similar cluster over a long period. The smaller the value, the closer the time distribution of the corresponding damage event is to a certain fixed interval within the period, reflecting a stronger time regularity. This regularity reflects the repeated effect of specific cable condition parameters (such as seasonal freeze-thaw cycles, rainy season infiltration, and annual construction cycles) in historical evolution, and can therefore serve as a basis for identifying typical risk patterns.

[0144] In this embodiment, step S6 specifically includes:

[0145] S6-1. Select the smallest variance from the Q variances;

[0146] S6-2. Based on the minimum variance, select the corresponding similar cluster from the Q similar clusters, and denot it as the adaptive learning cluster;

[0147] Therefore, the adaptive learning cluster containing the minimum variance represents a type of historical damage pattern that has the highest temporal regularity on the basis of a high degree of matching of the state evolution path (i.e., high first similarity). Its corresponding evolution path is most likely to reflect the typical and repeatable risk development process under the current environment and operating conditions of the cable.

[0148] This embodiment combines the similarity of state evolution vectors with the concentration of relative time zone numbers to preferentially select a set of cable breakage events that are highly matched with real-time evolution vectors and regularly reproduced on the historical timeline as an adaptive learning cluster, thereby improving the stability and interpretability of breakage risk assessment.

[0149] In this embodiment, step S7 is used to determine the second similarity within the adaptive learning cluster, including:

[0150] S7-A1: Extract the J state evolution vectors corresponding to the adaptive learning cluster;

[0151] S7-A2, Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

[0152] In this embodiment, the second similarity is used to measure the similarity between the real-time evolution vector and the J historical state evolution vectors within the adaptive learning cluster. Optionally, cosine similarity or Euclidean distance similarity constructed based on the inverse proportional relationship can be used. Through this similarity measure, the degree of matching between the current state evolution process and the complete evolution path of each historical damage event is reflected, so as to identify the specific historical event that is closest to the current risk.

[0153] In this embodiment, step S7 is further used to calculate a damage risk index based on the second similarity, including:

[0154] S7-B1, Normalize the J second similarities into J similarity weights;

[0155] Specifically, the second similarity is normalized using the softmax function or the min-max normalization method, converting it into a similarity weight within the interval [0,1], with the sum of all similarity weights being 1. This weight represents the relative contribution of each cable damage event to the current risk status. Softmax normalization, by exponentially normalizing the similarity, can highlight the weight of high similarity and suppress the influence of low similarity, making it suitable for scenarios with large differences in similarity between events. Min-max normalization is simple to calculate and suitable for scenarios with uniform similarity distribution or noise, but it may weaken the dominance of high similarity. This embodiment selects a suitable normalization method based on the actual data distribution characteristics to balance the stability of risk assessment.

[0156] For example, the formula for min-max normalization is: ;

[0157] For example, the softmax function is: ;

[0158] in, Denotes the second similarity of the j-th element. and These represent the minimum and maximum values ​​of the second similarity, respectively.

[0159] S7-B2, Calculate the L2 norm of J state evolution vectors;

[0160] S7-B3: Based on the L2 norm weighted summation of J state evolution vectors using J similarity weights, a damage risk index is generated.

[0161] Specifically, the L2 norm of the state evolution vector is the square root of the sum of the squares of the elements in each dimension of the vector, and its magnitude reflects the overall amplitude of the vector in the multidimensional feature space. In the state evolution vector, this vector is composed of cable state characteristics from multiple timestamps, including environmental, health, and construction conditions. Its L2 norm comprehensively characterizes the overall fluctuation intensity and cumulative effect of multi-source parameter changes in the cable over a period before the breakage. The more drastic the parameter changes and the greater the deviation from normal levels, the larger the corresponding L2 norm, indicating that the state evolution process experienced by the historical breakage event was more drastic and more likely to be close to the critical stage of actual breakage. Since the L2 norm only depends on observable state data and requires no additional labels or human judgment, it can serve as a calculable indicator to measure the "intrinsic severity" of historical breakage events, used to quantify the differences in risk intensity during the evolution process of different events.

[0162] This invention also provides an adaptive learning system for assessing the risk of damage to underground cables. This system is used to implement the methods described above, and will not be repeated hereafter. The terms "module," "unit," and "subunit," etc., used below refer to combinations of software and / or hardware that perform a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0163] like Figure 6 As shown, Figure 6 This is a structural block diagram of an adaptive learning system for assessing the risk of damage to underground cables, according to the present invention. The system includes:

[0164] A vector construction unit is used to construct state evolution vectors for M cable breakage events;

[0165] Clustering unit, used to perform nearest-neighbor clustering on M state evolution vectors to generate K state evolution clusters;

[0166] A similarity calculation unit is used to determine the first similarity between the K state evolution clusters and the real-time evolution vector;

[0167] Sequence building units are used to construct similar cluster sequences based on K first similarities;

[0168] The variance calculation unit is used to calculate the variance of several relative time zones within Q similar clusters in a similar cluster sequence.

[0169] An adaptive learning unit is used to determine an adaptive learning cluster based on the variance of several relative time zone numbers within Q similar clusters.

[0170] A risk assessment unit is used to determine the second similarity within the adaptive learning cluster and calculate the damage risk index based on the second similarity.

[0171] Determining the second similarity within the adaptive learning cluster includes:

[0172] Extract the J state evolution vectors corresponding to the adaptive learning cluster;

[0173] Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

[0174] In the above system, a state evolution vector is constructed through a vector construction unit, K state evolution clusters are generated through a cluster clustering unit, the first similarity is determined through a similarity calculation unit, a similar cluster sequence is constructed through a sequence construction unit, the variance within the similar cluster is calculated through a variance calculation unit, an adaptive learning cluster is determined through an adaptive learning unit, and a damage risk index is determined through a risk assessment unit, thus solving the problem that the damage risk of underground cables is difficult to quantify intuitively.

[0175] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0176] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An adaptive learning method for assessing the risk of damage to underground cables, characterized in that, include: S1. Construct the state evolution vectors for M cable breakage events; S2. Perform nearest neighbor clustering on the M state evolution vectors to generate K state evolution clusters; Among them, the state evolution cluster consists of the cluster center vector and the multiple state evolution vectors contained therein, which are used to represent similar state evolution paths; S3. Determine the first similarity between the K state evolution clusters and the real-time evolution vector; S4. Construct a sequence of similar clusters based on the K first similarity scores; S5. In a sequence of similar clusters, calculate the variance of several relative time zones within Q similar clusters; S6. Determine the adaptive learning clusters based on the variance of several relative time zone numbers within Q similar clusters; S7. Determine the second similarity within the adaptive learning cluster, and calculate the damage risk index based on the second similarity; Determining the second similarity within the adaptive learning cluster includes: S7-A1: Extract the J state evolution vectors corresponding to the adaptive learning cluster; S7-A2, Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

2. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, Construct state evolution vectors for M cable breakage events, including: S1-1. On the historical timeline, anchor the timestamps of M cable breakage events; S1-2. For any broken timestamp, take it as the event endpoint, trace back forward on the historical timeline, and extract a broken interval window with a time length of L. S1-3. Construct state evolution vectors within M damage time windows.

3. The adaptive learning method for assessing the risk of damage to underground cables according to claim 2, characterized in that, Construct state evolution vectors within M damage time windows, including: S1-3-1. Within the damaged section window, collect several cable status parameters at L-1 timestamps. S1-3-2. For any timestamp within the damaged interval window, preprocess several cable status parameters to generate several cable status features for the timestamp. S1-3-3: Concatenate several cable status features of the timestamp to generate a composite status feature corresponding to the timestamp. S1-3-4. Concatenate L-1 composite state features to construct the state evolution vector corresponding to the damaged interval window; S1-3-5. Traverse the timestamps of the M cable breakage events and repeatedly execute the construction of the state evolution vector until M state evolution vectors are constructed.

4. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, Determine the first similarity between the K state evolution clusters and the real-time evolution vector, including: S3-1. Calculate the cluster center vectors of K state evolution clusters; S3-2. Collect the neighboring state parameters of underground cables at L timestamps in real time and construct them into a real-time evolution vector; S3-3. Calculate the similarity between the cluster center vectors of the K state evolution clusters and the real-time evolution vectors to determine the K first similarity values.

5. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, Based on the K first similarity scores, construct a sequence of similar clusters, including: S4-1. Based on K first similarities, assign descending numbers to K state evolution clusters, and generate a state evolution cluster sequence based on the descending numbers; S4-2. Extract the state evolution clusters corresponding to the first Q descending numbers from the state evolution cluster sequence; S4-3. Define the state evolution clusters corresponding to the first Q descending numbers as similar cluster sequences; wherein, the similar cluster sequence inherits the Q descending numbers of its state evolution clusters.

6. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, In a sequence of similar clusters, calculate the variance of several relative time zones within Q similar clusters, including: S5-1. Based on Q descending numbers, obtain several damaged interval windows corresponding to several state evolution vectors within Q similar clusters in the similar cluster sequence, number by number. S5-2. Extract several central timestamps from several damaged interval windows and encode them as several relative time zone numbers within the period; S5-3. Collect several relative time zone numbers and calculate the variance of the several relative time zone numbers in the set; S5-4. Traverse the descending numbers of the Q similar clusters and repeat S5-1 to S5-3 until the Q variances within the Q similar clusters are obtained.

7. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, Based on the variance of several relative time zones within Q similar clusters, the adaptive learning clusters are determined, including: S6-1. Select the smallest variance from the Q variances; S6-2. Based on the minimum variance, select the corresponding similar cluster from the Q similar clusters, and denot it as the adaptive learning cluster.

8. The adaptive learning method for assessing the risk of damage to underground cables according to claim 1, characterized in that, The damage risk index is calculated based on the second similarity, including: S7-B1, Normalize the J second similarities into J similarity weights; S7-B2, Calculate the L2 norm of J state evolution vectors; S7-B3: Based on the L2 norm weighted summation of J state evolution vectors using J similarity weights, a damage risk index is generated.

9. An adaptive learning system for assessing the risk of damage to underground cables, characterized in that, The system includes: A vector construction unit is used to construct state evolution vectors for M cable breakage events; Clustering unit, used to perform nearest-neighbor clustering on M state evolution vectors to generate K state evolution clusters; Among them, the state evolution cluster consists of the cluster center vector and the multiple state evolution vectors contained therein, which are used to represent similar state evolution paths; A similarity calculation unit is used to determine the first similarity between K state evolution clusters and the real-time evolution vector; Sequence building units are used to construct similar cluster sequences based on K first similarities. The variance calculation unit is used to calculate the variance of several relative time zones within Q similar clusters in a similar cluster sequence. An adaptive learning unit is used to determine an adaptive learning cluster based on the variance of several relative time zone numbers within Q similar clusters. A risk assessment unit is used to determine the second similarity within the adaptive learning cluster and calculate the damage risk index based on the second similarity. Determining the second similarity within the adaptive learning cluster includes: Extract the J state evolution vectors corresponding to the adaptive learning cluster; Calculate the J second similarities between the real-time evolution vector and the J state evolution vectors.

Images