Cable external breakage risk migration prediction method fusing multi-source micro-vibration monitoring
By constructing a time-series feature extraction and transfer module for multi-source micro-vibration monitoring, and using data from multiple source domain monitoring stations to assist in target station prediction, the problems of data scarcity and geographical location differences in cable external damage risk monitoring are solved, and high-precision cable external damage risk prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID ANHUI ELECTRIC POWER CO LTD DONGZHI COUNTY POWER SUPPLY CO
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies for monitoring cable external damage risks suffer from data scarcity, limited model architecture, and insufficient scenario transferability, resulting in poor accuracy and robustness in predicting cable external damage risks. In particular, it is difficult to achieve high-precision prediction when there is limited monitoring data.
A method integrating multi-source micro-vibration monitoring is adopted. By constructing a time-series feature extraction module and a micro-vibration signal time-series transfer module, data from multiple source domain monitoring stations are used to assist in the prediction of the target station, overcoming the influence of geographical location differences. A time-series prediction module for the source domain and the target domain is constructed to achieve knowledge transfer across monitoring points.
With only limited monitoring data, a high-precision prediction of the risk of cable damage in the future period was achieved, improving the accuracy and robustness of the prediction and overcoming the impact of data scarcity and geographical differences.
Smart Images
Figure CN122045653A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an underground cable monitoring technology, and more particularly to a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring. Background Technology
[0002] Underground cables, unlike common overhead lines, are typically buried underground. They consist of one or more insulated conductors encased in an insulating and protective layer, used to transmit electricity or information from one location to another. In modern society, due to limited urban land, heavy traffic, and urban planning, large cities widely adopt underground cable power transmission. Compared to overhead lines, cables offer advantages such as smaller footprint, reliable power transmission, and stronger anti-interference capabilities.
[0003] In recent years, underground cables have served as the "lifeline" for energy transmission in urban power grids and industrial parks, making their safe and stable operation crucial. However, with the rapid development of urban construction, underground engineering activities such as subway construction, road excavation, and pile foundation compaction are becoming increasingly frequent. The resulting vibrations from construction machinery can easily cause minor external damage to nearby underground cables, such as squeezing, abrasion, or even puncture. This type of damage is insidious and cumulative; if early warning and precise location are not achieved, it may ultimately lead to large-scale power outages, equipment damage, and even safety accidents, causing huge economic losses and social impacts.
[0004] Currently, the monitoring of cable damage risk mainly relies on the following methods: manual inspection, traditional vibration sensor monitoring, and deep learning-based intelligent monitoring.
[0005] 1. Manual inspection relies primarily on the experience of inspectors to periodically patrol cable routes. This method is inefficient, cannot achieve real-time monitoring, has extremely poor ability to detect minor underground or hidden damage, and suffers from severe delays in early warning.
[0006] 2. In traditional vibration sensor monitoring methods, vibration sensors are deployed at key locations on the cable, and alarms are triggered by setting fixed acceleration or frequency thresholds. However, construction site environments are complex, with strong background noise interference (such as vehicle traffic and environmental vibrations). Weak, initial external vibration signals are easily drowned out by noise, resulting in high false alarm and false negative rates for this method. Furthermore, fixed thresholds are difficult to adapt to changes in vibration characteristics caused by different construction machinery and geological conditions, resulting in poor flexibility.
[0007] 3. Deep learning-based intelligent monitoring is a current research hotspot. This approach primarily involves constructing neural network models (such as CNN and LSTM) to automatically learn features from vibration data and identify risks. However, in practical applications, deep learning-based intelligent monitoring faces several significant challenges.
[0008] (1) The contradiction between data dependence and scarcity: Deep learning models, especially deep convolutional neural networks or long short-term memory networks, usually require a large amount of complete and labeled vibration event data for model training in order to achieve high-precision prediction. However, for any specific cable monitoring point, its positive sample (i.e. real external damage event) data is extremely scarce, and most of the time it is in a normal or disturbed state, which leads to insufficient training of the model on the target monitoring point, weak generalization ability, and difficulty in achieving accurate prediction of unknown risks.
[0009] (2) Inherent limitations of model architecture: CNN model is good at extracting local spatial features, but it is not good at capturing long temporal dependencies in vibration signals; LSTM model can process sequence data, but it has gradient vanishing / exploding problems, and it does not deeply mine the transient impact features and frequency domain multidimensional coupling features contained in complex vibration signals; the existing Transformer architecture has high computational complexity when processing long time sequences, and it is not sensitive to the unique frequency domain rapid change features of vibration signals, and it is not good at responding to weak transient features that indicate risks.
[0010] (3) Insufficient scene transfer capability: Existing models are usually trained on data from a single monitoring point, ignoring the inherent correlation and common characteristics between vibration data from multiple monitoring points (multi-source) in different spatial locations under the same construction environment. When a prediction model needs to be built for a newly established monitoring point or a monitoring point with scarce data, the existing methods cannot effectively utilize the knowledge already learned from other monitoring points (source domains) with relatively abundant data, resulting in the phenomenon of "data silos" and information waste.
[0011] Therefore, there is an urgent need for a method that can make full use of multi-source monitoring information and overcome the limitation of data scarcity to achieve knowledge transfer and collaborative early warning across monitoring points, thereby significantly improving the accuracy, robustness and foresight of cable external damage risk prediction. Summary of the Invention
[0012] To avoid the shortcomings of the existing technologies, this invention provides a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, so as to achieve a high-precision prediction of cable external damage risk in the future even with only a small amount of monitoring data.
[0013] The present invention adopts the following technical solution to solve the technical problem.
[0014] The present invention provides a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, comprising the following steps: Step 1: Data acquisition and preprocessing steps; acquire historical monitoring data D, and convert the historical monitoring data into a time series H after preprocessing; Step 2: Construct a time series feature extraction module to extract the time series features h of the time series H; Step 3: Construct a micro-vibration signal time-series transfer module. This module is used to analyze the time-series characteristics of multi-source data from any two in-situ micro-vibration signal monitoring stations. The distribution differences are measured; the temporal characteristics of multi-source data are analyzed using the micro-vibration signal time-series transfer module. and target data time series characteristics The distribution differences are measured.
[0015] Step 4: Construct source domain time series prediction module and target domain time series prediction module respectively to predict the micro-vibration signal at future time in the source domain and the micro-vibration signal at future time in the target domain.
[0016] The present invention also features a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, which is characterized by: Furthermore, the historical monitoring data includes multi-source historical monitoring data Ds and target historical monitoring data Dm.
[0017] Further, step 1 includes the following steps: Step 11: Data acquisition steps; acquire multi-source historical monitoring data Ds through vibration sensors at multiple other source domain in-situ micro-vibration signal monitoring stations; acquire target historical monitoring data Dm through vibration sensors at the target area in-situ micro-vibration signal monitoring station.
[0018] Step 12: Data preprocessing step; perform missing value interpolation, outlier handling, and maximum-minimum normalization on the multi-source historical monitoring data Ds and the target historical monitoring data Dm; Step 13: Data partitioning step; The preprocessed multi-source historical monitoring data Ds and target historical monitoring data Dm from Step 12 are partitioned at time intervals. It is divided into time series H.
[0019] Furthermore, the time series H includes time series of multi-source historical monitoring data. and target historical monitoring data time series .
[0020] Furthermore, in step 2, the temporal feature extraction module includes a source domain temporal feature extraction module. F sAnd target domain temporal feature extraction module F t .
[0021] Furthermore, in step 2, the time series feature h includes multi-source data time series features. and target data time series characteristics .
[0022] Further, step 2 includes the following steps: Step 21: Extracting source domain temporal features of micro-vibration signals using the module F s From multi-source historical monitoring data time series Extracting time-series features from multi-source data ; Step 22: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; Step 23: Minimize the parameter difference loss L θ By utilizing sufficient monitoring data in the source domain, the target domain temporal feature extraction module F is assisted. t Training based on differences.
[0023] Furthermore, step 3 includes the following steps: Step 31: Calculate the characteristic distribution difference of the time sequence characteristics of the micro-vibration signal between the multi-source domain and the target domain using formula (5). ; Step 32: Calculate the characteristic distribution difference of the time-series characteristics of the micro-vibration signals from the two multi-source sites. ; Step 33: Calculate migration loss .
[0024] Furthermore, step 4 includes the following steps: Step 41: Construct the source domain future signal prediction module : Step 42: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; Step 43: Establish the target domain future signal prediction module C t And make predictions based on the time-series characteristics of the target domain monitoring data.
[0025] Furthermore, in step 43, a target domain future signal prediction module C is established using a multi-layer fully connected network based on the target domain micro-vibration monitoring data.t .
[0026] Compared with existing technologies, the beneficial effects of this invention are reflected in: This invention discloses a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, comprising: data acquisition and preprocessing steps; acquiring historical monitoring data D, and converting the historical monitoring data into a time series H after preprocessing; constructing a time series feature extraction module to extract the time series features h of the time series H; and extracting the time series features h of multi-source data from any two in-situ micro-vibration signal monitoring stations in multiple source domains. Distribution differences and time series characteristics of multi-source data and target data time series characteristics The distribution differences are measured; a source domain time series prediction module and a target domain time series prediction module are constructed respectively to predict the micro-vibration signals at future times in the source domain and the target domain.
[0027] This invention discloses a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring. By using the multi-source migration method, the influence of geographical differences between different monitoring points is overcome. Vibration monitoring data from multiple other micro-vibration monitoring stations with sufficient monitoring data are used to assist in the prediction of cable external damage risk at the target site. This enables a high-accuracy prediction of cable external damage risk in the future even with only a small amount of monitoring data.
[0028] The cable external damage risk migration prediction method integrating multi-source micro-vibration monitoring of the present invention is mainly used for detecting the external damage risk of underground cables caused by excavation vibration. It has the advantages of being able to predict the cable external damage risk in the future period with high accuracy even with only a small amount of monitoring data. Attached Figure Description
[0029] Figure 1 This is a framework diagram of the cable external damage risk migration prediction method integrating multi-source micro-vibration monitoring of the present invention. Detailed Implementation
[0030] See Figure 1 The present invention provides a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, comprising the following steps: Step 1: Data acquisition and preprocessing steps; acquire historical monitoring data D, and convert the historical monitoring data into a time series H after preprocessing; Step 2: Construct a time series feature extraction module to extract the time series features h of the time series H; Step 3: Construct a micro-vibration signal time-series transfer module. This module is used to analyze the time-series characteristics of multi-source data from any two in-situ micro-vibration signal monitoring stations. The distribution differences are measured; the temporal characteristics of multi-source data are analyzed using the micro-vibration signal time-series transfer module. and target data time series characteristics The distribution differences are measured.
[0031] In this invention, a micro-vibration signal timing transfer module is constructed to measure the feature distribution difference of micro-vibration signal timing features between any two multi-source sites, and also to measure the feature distribution difference of micro-vibration signal timing features between any multi-source site and the target site. By minimizing the above two types of feature distribution differences, the cross-domain transfer of micro-vibration signal timing feature knowledge is achieved, thereby realizing the extraction of the essential features of micro-vibration signal timing changes.
[0032] Step 4: Construct source domain time series prediction module and target domain time series prediction module respectively to predict the micro-vibration signal at future time in the source domain and the micro-vibration signal at future time in the target domain.
[0033] Time series prediction modules were constructed for the micro-vibration signals from multi-source in-situ monitoring stations and target monitoring stations, respectively. The time series prediction modules then utilized the time series features of the multi-source data extracted in step 2. and target data time series characteristics Predicting micro-vibration signals at future times. The time-series prediction module includes a source domain future signal prediction module. And target domain future signal prediction module C t .
[0034] like Figure 1 As shown, the present invention provides a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring. This method utilizes historical monitoring data of micro-vibration signals from multiple in-situ monitoring stations of the same cable to assist in the prediction of micro-vibration signals at the target monitoring station. By constructing a time-series feature migration module, the method overcomes the influence of regional location on the differences in micro-vibration signal monitoring at different points. It can achieve high-precision micro-vibration signal prediction even with only a small amount of historical monitoring data. It can utilize multiple source domain in-situ monitoring stations with sufficient micro-vibration signal monitoring data to improve the micro-vibration signal prediction accuracy of target stations with insufficient monitoring data.
[0035] In practice, the historical monitoring data includes multi-source historical monitoring data Ds and target historical monitoring data Dm.
[0036] Multi-source historical monitoring data Ds of micro-vibration signals from multiple source domain in-situ micro-vibration signal monitoring stations (non-target domain monitoring stations used to provide reference data for the target station, also referred to as multi-source stations in this invention) and target historical monitoring data Dm of micro-vibration signals from the target area in-situ micro-vibration signal monitoring station (also referred to as the target station in this invention) are acquired respectively. The multi-source historical monitoring data Ds and the target historical monitoring data Dm are preprocessed and converted into time series according to the time step T.
[0037] In practice, step 1 includes the following steps: Step 11: Data acquisition steps; acquire multi-source historical monitoring data Ds through vibration sensors at multiple other source domain in-situ micro-vibration signal monitoring stations; acquire target historical monitoring data Dm through vibration sensors at the target area in-situ micro-vibration signal monitoring station.
[0038] Step 12: Data preprocessing step; perform missing value interpolation, outlier handling, and maximum-minimum normalization on the multi-source historical monitoring data Ds and the target historical monitoring data Dm; Step 13: Data partitioning step; The preprocessed multi-source historical monitoring data Ds and target historical monitoring data Dm from Step 12 are partitioned at time intervals. It is divided into time series H.
[0039] In specific implementation, the time series H includes time series of multi-source historical monitoring data. and target historical monitoring data time series .
[0040] In practice, the time interval will be... Set to 5 minutes, with T time steps, where T=12. Divide the multi-source historical monitoring data Ds at time t into... ; This represents the time series of multi-source historical monitoring data of the i-th multi-source station at time t. The target historical monitoring data Dm of the target station is divided into... ; This represents the time series of historical monitoring data for the target station at time t, where the time step T = 12. The subscript t indicates time, and the superscript m indicates the m-th target station.
[0041] In specific implementation, step 2, the temporal feature extraction module includes a source domain temporal feature extraction module. F s And target domain temporal feature extraction module F t .
[0042] In specific implementation, in step 2, the time series feature h includes multi-source data time series features. and target data time series characteristics .
[0043] Time series of multi-source historical monitoring data and target historical monitoring data time series A micro-vibration signal time-series feature extraction module was constructed. This module was used to extract time series data from multi-source historical monitoring data. Multi-source data time series characteristics and target historical monitoring data time series Target data time series features .
[0044] In practice, step 2 includes the following steps: Step 21: Extracting source domain temporal features of micro-vibration signals using the module F s From multi-source historical monitoring data time series Extracting time-series features from multi-source data ; For multi-source historical monitoring data Ds from multiple sites, a source domain temporal feature extraction module for multiple sites is constructed using a bidirectional long short-term memory neural network (BiLSTM). F s The neural network parameters are θ s To facilitate the cross-domain transmission of the temporal characteristics of micro-vibration signals, this invention uses a shared parameter model for multi-source historical monitoring data Ds.
[0045] Step 211: Utilize the time series of multi-source historical monitoring data of the i-th multi-source station at time t. The value in Input gate of source domain BiLSTM Forgotten Gate and output gate The value of is shown in the following formula (1).
[0046] (1); In formula (1), , and They represent the input gates respectively. Forgotten Gate and output gate The output weights; , and They represent the input gates respectively. Forgotten Gate and output gate Input weights; , and They represent the input gates respectively. Forgotten Gate and output gate The bias; This represents the hidden state (i.e., the output gate) of the cell body when the data from the i-th multi-source station is used as input at time t-1. The output value (or the output value of the source domain BiLSTM), σ It is the sigmoid non-linear activation function.
[0047] Step 212: Then, according to the above... , Calculate the input node at time t The value and the cell state at time t The value of is shown in the following formula (2).
[0048] (2); In formula (2), For input node The output weights; Indicates input node Input weights; Indicates input node The bias; This represents the Hadema product. This represents the cell state at time t-1.
[0049] Step 213: For the cell body, That is, the predicted value of the i-th multi-source station at time t+1, as shown in the following formula (3).
[0050] (3); For source domain BiLSTM for forward sequence The output obtained by LSTM for reverse sequence The output obtained by the source domain BiLSTM. Therefore, the final sequence output features obtained by BiLSTM are: ; Abbreviated as , This is the function expression for the time-series feature extraction module.
[0051] Step 22: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; Extracting the time-series features of target data from the target site The process of extracting time-series features from multi-source data Step S21 is basically the same. Establish the target domain temporal feature extraction module F for the target site. t The final target data time series features obtained for .
[0052] Step 23: Minimize the parameter difference loss L θ By utilizing sufficient monitoring data in the source domain, the target domain temporal feature extraction module F is assisted. t Training based on differences.
[0053] Considering the geographical differences between multi-source sites and target sites, the distribution of micro-vibration signals monitored by multi-source sites and target sites are also different. Therefore, the source domain temporal feature extraction module F is suitable for multi-source sites. s This approach may not be suitable for the target site. Therefore, a weakly shared parameter mechanism is designed to effectively utilize multi-source historical monitoring data time series. Multi-source data time series characteristics See formula (4) below.
[0054] (4) Formula (4) introduces the parameter difference loss L θ By minimizing the parameter difference loss L θ In order to ensure the source domain temporal feature extraction module F s And target domain temporal feature extraction module F t Under the premise of differences, make full use of the sufficient monitoring data in the source domain to assist the target domain temporal feature extraction module F. t Training the difference. θs is the source domain temporal feature extraction module F. s The network learning parameters are given by θt, where θt is the source domain temporal feature extraction module F. s Network learning parameters.
[0055] In practice, step 3 includes the following steps: Step 31: Calculate the characteristic distribution difference of the time sequence characteristics of the micro-vibration signal between the multi-source domain and the target domain using formula (5). ; Formula for measuring the difference in characteristic distribution See formula (5) below; (5).
[0056] In formula (5), X i and Xj for The input, X i This represents the time series of multi-source historical monitoring data for the i-th multi-source site. Monitoring data in (for) (the i-th time series vector) This represents a dataset containing time series of multi-source historical monitoring data from all multi-source sites in the source domain; X j This represents the time series of historical monitoring data for the target. The j-th monitoring data (for) (the j-th time series vector) This represents the time series of historical monitoring data for the target domain. The dataset; p and q Let n represent the probability distributions of the source domain and the target domain, respectively; s Represents time series of multi-source historical monitoring data The amount of data in n t This represents the time series of historical monitoring data for the target. The amount of data; For mapping kernel function, This represents the regenerated Hilbert space.
[0057] Therefore, the characteristic distribution differences of the temporal features of the micro-vibration signal between the multi-source domain and the target domain are as follows. It can be expressed as the following formula (6).
[0058] (6) In formula (6), N is the total number of multi-source sites, i represents the i-th multi-source site, and 1≤i≤N. is the function that measures the difference in distribution in formula (5).
[0059] By minimizing cross-domain loss (i.e., the difference in the characteristic distribution of the temporal features of the micro-vibration signal between the multi-source domain and the target domain) This not only allows us to obtain the temporal variation characteristics of micro-vibration signals between multiple source and target domains (loss function) It does not extract features itself, but extracts features by optimizing the overall network model through this loss function. It can also select in-situ monitoring stations in the source domain based on the affinity between the distribution of micro-vibration signals in the source and target domains. For source domain monitoring station data with large differences, the weight of the differences is increased, thereby obtaining better consistency.
[0060] Step 32: Calculate the characteristic distribution difference of the time-series characteristics of the micro-vibration signals from the two multi-source sites. ; Differences in the characteristic distribution of the time-series features of micro-vibration signals from any two multi-source sites The measurement is shown in the following formula (7).
[0061] (7) In formula (7), N is the total number of multi-source sites, i represents the i-th multi-source site, and j represents the previous multi-source site of the i-th multi-source site.
[0062] By minimizing cross-source loss This allows us to obtain the most common features among multiple source domains.
[0063] Step 33: Calculate migration loss .
[0064] Final total migration loss For the feature distribution difference Difference from the aforementioned feature distribution The sum is shown in the following formula (8).
[0065] (8).
[0066] In practice, step 4 includes the following steps: Step 41: Construct the source domain future signal prediction module ; For each source domain monitoring station, a multi-layer fully connected network is used to construct a source domain future signal prediction module. , i represents the i-th multi-source site, as shown in the following formula (9).
[0067] (9); In formula (9), For the time series feature extraction module F s Extracted target data time series features . Source Domain Future Signal Prediction Module The predicted future source domain signal of the i-th multi-source site at time t. (·) is the prediction function for future signals in the source domain.
[0068] Step 42: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; The empirical loss is calculated for the time-series data of the micro-vibration signal in each source domain. L CS See formula (10) below.
[0069] (10).
[0070] In formula (10), MSE(·) is the mean square error function; This represents the time series of multi-source historical monitoring data of the i-th multi-source station at time t; Source Domain Future Signal Prediction Module The predicted future source domain signal of the i-th multi-source station at time t.
[0071] The micro-vibration signal time series prediction module of each source domain is used to make predictions with the time series characteristics of the monitoring data of the target domain as input, and the aggregate loss of each prediction module for the target domain is calculated. See formula (11) below.
[0072] (11).
[0073] In formula (11), To convert the historical monitoring data of the target into a time series Input to source domain temporal feature extraction module F s The extracted time-series features of the data; and Let be the prediction function for future signals in the source domain.
[0074] Step 43: Establish the target domain future signal prediction module C t And make predictions based on the time-series characteristics of the target domain monitoring data.
[0075] In specific implementation, in step 43, a target domain future signal prediction module C is established using a multi-layer fully connected network based on the target domain micro-vibration monitoring data. t .
[0076] Prediction is made based on the time-series characteristics of the target domain monitoring data, as shown in the following formula (12).
[0077] (12) In formula (12), L Ct The loss is used to predict future signals in the target domain.
[0078] Cable external damage risk migration prediction model (including source domain future signal prediction module) And target domain future signal prediction module C t The final loss function is L total See formula (13) below.
[0079] (13) In formula (13), a , β , μ These are all balance coefficients of the final loss function.
[0080] This invention proposes a method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring. By using a multi-source migration method, the influence of geographical differences between different monitoring points is overcome. Vibration monitoring data from multiple other micro-vibration monitoring stations with sufficient monitoring data are used to assist in the prediction of cable external damage risk at the target site. This enables high-accuracy prediction of cable external damage risk in the future even with only a small amount of monitoring data, thus solving the technical problem of high false alarm rate in cable external damage risk prediction in the prior art when there is a small amount of micro-vibration monitoring data.
[0081] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0082] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring, characterized in that, Includes the following steps: Step 1: Data acquisition and preprocessing steps; acquire historical monitoring data D, and convert the historical monitoring data into a time series H after preprocessing; Step 2: Construct a time series feature extraction module to extract the time series features h of the time series H; Step 3: Construct a micro-vibration signal time-series transfer module. This module is used to analyze the time-series characteristics of multi-source data from any two in-situ micro-vibration signal monitoring stations. The distribution differences are measured; the temporal characteristics of multi-source data are analyzed using the micro-vibration signal time-series transfer module. and target data time series characteristics The distribution differences are measured; Step 4: Construct source domain time series prediction module and target domain time series prediction module respectively to predict the micro-vibration signal at future time in the source domain and the micro-vibration signal at future time in the target domain.
2. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 1, characterized in that, The historical monitoring data includes multi-source historical monitoring data Ds and target historical monitoring data Dm.
3. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 2, characterized in that, Step 1 includes the following steps: Step 11: Data acquisition steps; acquire multi-source historical monitoring data Ds through vibration sensors at multiple in-situ micro-vibration signal monitoring stations in other source areas; acquire target historical monitoring data Dm through vibration sensors at in-situ micro-vibration signal monitoring stations in the target area; Step 12: Data preprocessing steps; Missing value interpolation, outlier handling, and maximum-min normalization are performed on the multi-source historical monitoring data Ds and the target historical monitoring data Dm. Step 13: Data partitioning step; The preprocessed multi-source historical monitoring data Ds and target historical monitoring data Dm from step 12 are partitioned into time series H with time interval Δt.
4. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 3, characterized in that, The time series H includes time series of multi-source historical monitoring data. and target historical monitoring data time series .
5. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 1, characterized in that, In step 2, the temporal feature extraction module includes a source domain temporal feature extraction module. F s And target domain temporal feature extraction module F t .
6. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 5, characterized in that, In step 2, the time series feature h includes multi-source data time series features. and target data time series characteristics .
7. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 6, characterized in that, Step 2 includes the following steps: Step 21: Extracting source domain temporal features of micro-vibration signals using the module F s From multi-source historical monitoring data time series Extracting time-series features from multi-source data ; Step 22: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; Step 23: Minimize the parameter difference loss L θ By utilizing sufficient monitoring data in the source domain, the target domain temporal feature extraction module F is assisted. t Training based on differences.
8. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 1, characterized in that, Step 3 includes the following steps: Step 31: Calculate the characteristic distribution difference of the time sequence characteristics of the micro-vibration signal between the multi-source domain and the target domain using formula (5). ; Step 32: Calculate the characteristic distribution difference of the time-series characteristics of the micro-vibration signals from the two multi-source sites. ; Step 33: Calculate migration loss .
9. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 1, characterized in that, Step 4 includes the following steps: Step 41: Construct the source domain future signal prediction module ; Step 42: Extract the target domain temporal features of the micro-vibration signal using the F module. t From the target historical monitoring data time series Extracting time-series features from target data ; Step 43: Establish the target domain future signal prediction module C t And make predictions based on the time-series characteristics of the target domain monitoring data.
10. The method for predicting the migration of cable external damage risk by integrating multi-source micro-vibration monitoring according to claim 1, characterized in that, In step 43, a target domain future signal prediction module C is established using a multi-layer fully connected network based on the target domain micro-vibration monitoring data. t .