AI fault identification method and system for power line

Abstract

The invention belongs to the technical field of circuit monitoring, and particularly relates to an AI fault identification method and system for an electric power circuit, and the method comprises the steps: constructing a cross-modal data flow with a space-time mark through the synchronous collection of a thermal imaging sequence and a three-phase current waveform, extracting the characteristics of a multi-physical field through combining the characteristics of a wire material, and carrying out the recognition of an AI fault. And establishing a thermal-electric coupling relationship by using space-time alignment convolution, further analyzing and calculating correlation strength between temperature anomaly and current fluctuation through a sliding window, constructing a dynamic threshold model, generating a fault probability thermodynamic diagram, and finally outputting a graded early warning signal. Compared with the prior art, the method has the advantages that deep fusion analysis of thermal imaging and electrical signals is realized, the recognition precision and response speed of potential fault points are improved, the abnormal state of the wire joint can be found and accurately positioned earlier, the misjudgment rate is effectively reduced, and the safety and stability of power grid operation are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of circuit monitoring, and in particular relates to an AI fault identification method and system for power lines. Background Art

[0002] During power system operation, transmission line conductor joints can easily overheat due to aging, looseness, or poor contact, leading to line failures and even power outages. Currently, power line status monitoring primarily relies on single-mode data analysis, such as detecting temperature anomalies through infrared thermal imaging or determining distortion by sampling current waveforms.

[0003] Existing technologies typically perform fault identification by independently processing thermal images and electrical quantity data. For example, some systems set warning rules based solely on temperature thresholds in thermal images, ignoring the impact of current fluctuations on thermal effects. Others focus on spectral analysis of current signals, failing to integrate spatial location information and thermal behavior characteristics for comprehensive judgment. This single-modal analysis approach suffers from high false alarm rates, delayed response times, and difficulty accurately locating the fault source. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI fault identification method and system for power lines, which realizes the deep fusion analysis of thermal imaging and electrical signals, improves the identification accuracy and response speed of potential fault points, can detect and accurately locate abnormal conditions of wire joints earlier, effectively reduce the misjudgment rate, and improve the safety and stability of power grid operation, so as to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an AI fault identification method for power lines, comprising the following steps:

[0006] Synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and conductor joint coordinates, and generate a spatiotemporally labeled dataset. Dynamic frequency compensation is performed on the dataset, and an interpolation function is constructed to generate a cross-modal data stream of equal time granularity. The thermal imaging temperature gradient tensor and the current transient harmonic energy distribution are extracted, and a feature vector is constructed in combination with the conductor material parameters. The feature vector is subjected to cross-modal interaction through spatiotemporal aligned convolution to generate a thermal-electric coupling map. A sliding window analysis is performed on the thermal-electric coupling map to calculate the phase offset-energy coupling coefficient between temperature anomalies and current fluctuations. A dynamic threshold model is constructed based on the coupling coefficient, and a fault judgment rule is formed by optimizing the threshold boundary through historical samples. The fault judgment rule is matched with the real-time map to generate a fault probability heat map, and a graded warning signal is output based on the fault probability heat map.

[0007] Preferably, generating a spatiotemporal labeling dataset includes: Infrared sensors and current transformers are deployed at the joints of each conductor in the power lines to collect thermal imaging sequences and three-phase current waveforms, while also recording the timestamp and spatial coordinates of each thermal image frame and each current sampling point. Sort the timestamps and determine the continuity of the samples based on the intervals between adjacent time points. If the interval exceeds a preset maximum value, an abnormal flag is triggered and the sample is skipped. Corresponding the spatial coordinates to the node numbers in the power line topology structure, and establishing a mapping relationship between each thermal imaging pixel point and the physical location of the wire connector; Based on the mapping relationship and the timestamp, the thermal imaging sequence and the three-phase current waveform are combined in the time and space dimensions to generate a data set with time and space labels.

[0008] Preferably, generating a cross-modal data stream of equal time granularity includes: Obtain the thermal imaging frame rate and current sampling rate, calculate the proportional relationship between the two, and set a unified time step based on the high-frequency signal; Performing time interpolation processing on the thermal imaging sequence, constructing a temperature change function between two adjacent frames to generate intermediate temperature data under a uniform time step; The three-phase current waveform is processed based on a unified time step, and the original sampling data in each time period is converted into a current value in a unified format; The interpolated temperature data and the adjusted current data are aligned along the time axis and combined into a multimodal data stream with the same time resolution.

[0009] Preferably, the step of constructing a feature vector by combining the conductor material parameters includes: Obtain physical property information of the wire, including resistivity, thermal conductivity, and linear expansion coefficient, and store the parameters in a feature construction module; The maximum temperature rise characteristics of the local area are extracted based on thermal imaging data, and the transient distortion energy characteristics are extracted from the three-phase current waveform; The maximum temperature rise characteristics and transient distortion energy are used to construct a composite characteristic factor in combination with the physical properties of the conductor. The composite characteristic factor is fused with the linear expansion coefficient to generate a normalized characteristic quantity, which is then concatenated with the original characteristic to form a characteristic vector.

[0010] Preferably, generating a thermal-electric coupling map comprises: Based on the eigenvectors, a two-dimensional spatial mapping matrix is constructed to map the eigenvalues of each wire joint position to pixel points in a unified coordinate system; Perform weighted diffusion processing on the two-dimensional spatial mapping matrix, introduce interaction weights within the pixel neighborhood, and update the local response value; The local response values are superimposed on the time dimension to form a dynamic thermal-electric distribution map in a continuous period; The high response areas in the dynamic thermo-electric distribution map are clustered and marked, hot spots with high thermo-electric coupling intensity are extracted, and their time and wire segment information are annotated.

[0011] Preferably, the calculation of the phase shift-energy coupling coefficient between the temperature anomaly and the current fluctuation includes: Collect temperature and current measurements at multiple time points to construct a time series dataset; Perform spectrum analysis on the time series data to obtain the frequency distribution of temperature and current, and calculate the phase offset between the two; Based on the phase shift combined with time series data, the correlation strength between temperature anomaly and current fluctuation is evaluated to obtain the energy coupling coefficient.

[0012] Preferably, the forming of the fault determination rule includes: Collect temperature anomalies and current fluctuation events in historical operating data, extract the corresponding energy coupling coefficient sequence, and mark the time when known faults occurred; Performing statistical analysis on the energy coupling coefficient sequence, determining a reference range under a normal state, and setting a preliminary determination threshold; Comparing the preliminary determination threshold with the actual event, if the energy coupling coefficient exceeds the threshold at multiple consecutive time points, triggering a warning flag and updating the determination boundary; Based on the updated decision boundary, multiple intervals are divided, and warning levels are set for different intervals to form fault decision rules with hierarchical differentiation.

[0013] Preferably, generating a failure probability heat map includes: Obtain historical fault frequency data for each conductor joint location, correspond it to the physical coordinates in the current line topology, and construct an initial probability distribution matrix; Mapping the fault determination results obtained in real time to the initial probability distribution matrix, and updating the probability values of the locations marked as abnormal; The updated probability distribution matrix is normalized, and the neighborhood influence factor is introduced to perform diffusion correction on the probability values of adjacent areas. A visual heat map is generated based on the corrected probability distribution.

[0014] Preferably, outputting a graded warning signal based on the heat map includes: Setting multiple warning level standards based on the fault probability heat map, dividing the probability distribution into different intervals, each interval corresponding to a warning level; Check each location in the fault probability heat map and determine its warning level based on the interval it belongs to; Generate a warning signal distribution map covering the entire monitoring area according to the warning level, and merge consecutive areas with the same level to form a unified warning area; The unified warning area and its corresponding warning level are displayed on the monitoring terminal.

[0015] In another aspect, the present invention provides an AI fault identification system for power lines, comprising: Multimodal data acquisition module, used to synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and wire joint coordinates, and generate spatiotemporally labeled datasets; A cross-modal time alignment module, configured to perform dynamic frequency compensation processing on the data set and construct an interpolation function to generate a cross-modal data stream of equal time granularity; The multi-physics feature fusion module is used to extract the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, combine the wire material parameters to construct a feature vector, and perform cross-modal interaction on the feature vector through spatiotemporal alignment convolution to generate a thermal-electric coupling map; An anomaly correlation modeling module is used to perform a sliding window analysis on the thermal-electric coupling map, calculate the phase offset-energy coupling coefficient between the temperature anomaly and the current fluctuation, build a dynamic threshold model based on the coupling coefficient, and optimize the threshold boundary through historical samples to form a fault judgment rule; The early warning module is used to match the fault judgment rules with the real-time map to generate a fault probability heat map, and output a graded early warning signal according to the fault probability heat map.

[0016] Technical effects and advantages of the present invention: Compared with the existing technology, the AI fault identification method and system for power lines proposed in the present invention have the following advantages: This method constructs a spatiotemporally labeled cross-modal data stream by synchronously capturing thermal imaging sequences and three-phase current waveforms. It then extracts multi-physics field features based on conductor material properties, establishes a thermal-electrical coupling relationship using spatiotemporal alignment convolution, and further calculates the correlation strength between temperature anomalies and current fluctuations through sliding window analysis. It then constructs a dynamic threshold model and generates a fault probability heat map, ultimately outputting a graded warning signal. Compared to existing technologies, this method achieves a deep fusion analysis of thermal imaging and electrical signals, improving the accuracy and response speed of identifying potential fault points. This allows for earlier detection and accurate location of abnormal conductor joint conditions, effectively reducing the misjudgment rate and improving the safety and stability of power grid operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Flowchart of the AI fault identification method for power lines of the present invention; Figure 2 A block diagram of an AI fault identification system for power lines according to the present invention; Figure 3 This is a current change trend diagram after time alignment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] The present invention provides Figure 1 The AI-powered fault identification method shown in the figure constructs a cross-modal data stream with spatiotemporal labels by synchronously acquiring thermal imaging sequences and three-phase current waveforms. It then extracts multi-physics field features based on conductor material properties and uses spatiotemporal alignment convolution to establish a thermal-electrical coupling relationship. Furthermore, a sliding window analysis calculates the correlation strength between temperature anomalies and current fluctuations, constructs a dynamic threshold model, and generates a fault probability heat map. Finally, it outputs a graded warning signal, as detailed below: In this embodiment, an AI fault identification method for a power line is characterized by comprising the following steps: Step 1: Synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and conductor joint coordinates, and generate a spatiotemporal labeled dataset. This includes the following steps: Infrared sensors and current transformers are deployed at the joints of each conductor in the power line to collect thermal imaging sequences and three-phase current waveforms. The built-in GPS module records the timestamp of each thermal image frame and each current sampling point. and spatial coordinates ; Timestamp Sort by time interval Determine the sampling continuity. If Δt exceeds the preset threshold , then the exception mark is triggered and the data segment is skipped; Indicates the maximum allowed time interval; The spatial coordinates Corresponding to the node number in the power line topology, a mapping relationship is established between each thermal imaging pixel and the physical location of the wire joint; Based on mapping relationship and timestamp , combining the thermal imaging sequence with the three-phase current waveform in time and space dimensions to generate a data structure with time and space labels ,in Indicates the temperature value of the jth coordinate at the i-th moment in thermal imaging, Indicates the three-phase current values corresponding to the same moment.

[0020] Step 2: Perform dynamic frequency compensation on the data set and construct an interpolation function to generate a cross-modal data stream with equal time granularity; specifically, the following steps are included: Get thermal imaging frame rate and current sampling rate , calculate the frequency ratio of the two And set a unified time step Δt based on the high-frequency signal; Perform time interpolation processing on the thermal imaging sequence, and Construct a linear temperature change function between: , used to generate the intermediate temperature data of the Δt interval; the formula Temp(t) indicates that the temperature value at any time point t is obtained from the two real sampling points before and after and Linear interpolation is used to supplement the missing information in the time dimension of thermal imaging data, improve the temporal resolution of thermal imaging data, enable more precise temporal alignment with high-frequency current signals, and enhance the consistency and comparability of cross-modal data.

[0021] Based on the unified time step Δt, the three-phase current waveform is downsampled or subjected to sliding average filtering, and the current value within each Δt time period is taken as the window average. , where N is the number of original sampling points in the period; The interpolated temperature data and the adjusted current data are aligned along the time axis to form a multimodal data stream with the same time resolution. , achieving consistent expression of cross-modal data in the time dimension.

[0022] Step 3: Extract the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, and construct a feature vector based on the conductor material parameters. This specifically includes the following steps: Obtain the physical property information of the wire, including resistivity ρ, thermal conductivity λ, and linear expansion coefficient α, and store the parameters in the form of constants in the feature construction module; Based on the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, the maximum temperature rise in the local area is extracted respectively. and current distortion energy , as the basic thermo-electric characteristics; use and , combined with the conductor material parameters to construct the composite characteristic factor F, the calculation formula is: ; This formula takes the thermal effect ( ) and electrical effect ( ) is weightedly fused through the wire material parameters (λ and ρ) to form a comprehensive composite characteristic factor F.

[0023] Normalize and fuse the composite characteristic factor F and the linear expansion coefficient α to generate a comprehensive characteristic quantity , and concatenate it with the original features to form an enhanced multidimensional feature vector By introducing the linear expansion coefficient for normalization and constructing a multi-dimensional feature vector, the thermal, electrical, and material information are effectively integrated, significantly enhancing the comprehensiveness and robustness of feature expression.

[0024] Step 4: Perform cross-modal interaction on the feature vectors through spatiotemporal alignment convolution to generate a thermal-electric coupling map; specifically, the following steps are included: Based on the enhanced multidimensional eigenvector V, a two-dimensional spatial mapping matrix M is constructed to map the eigenvalues of each wire joint position to pixel points in a unified coordinate system. Perform weighted diffusion processing on the mapping matrix M, introduce the interaction weight w of temperature and current characteristics in the pixel neighborhood, and calculate the updated local response value: Considering the possible coupling effects such as heat conduction and electromagnetic interference between adjacent wire segments, this step adopts the weighted neighborhood diffusion mechanism to weight the influence of neighboring pixels on the feature value of each position. The four neighboring points above, below, left and right are averaged, multiplied by the interaction weight w, and then superimposed back to the original value, thereby enhancing the fluidity and correlation of information between regions.

[0025] The response value Response(i,j) is superimposed in the time dimension to form a dynamic thermal-electric distribution map in a continuous period, reflecting the spatial correlation evolution process of temperature mutation and current distortion; The high response areas in the atlas are clustered and marked, and the hot spots where the thermal-electric coupling intensity is more than twice the background mean are extracted, and their corresponding timestamps and wire segment numbers are marked.

[0026] Step 5: Perform sliding window analysis on the thermal-electric coupling spectrum to calculate the phase shift-energy coupling coefficient between the temperature anomaly and the current fluctuation; specifically, the following steps are included: Collect the temperature measurement values Temp and current measurement values I of the power line at multiple time points to construct a time series dataset ; Perform Fourier transform on the temperature and current time series in data set D to obtain their respective spectral distributions and , and then calculate the phase shift φ between the spectral distributions using the formula , where Im represents the imaginary part, Re represents the real part, and conj represents the complex conjugate. Using a fast Fourier transform (FFT), the temperature and current time series are converted to the frequency domain to reveal the underlying periodic and aperiodic components. Subsequently, the phase offset φ between them is calculated based on their spectral distribution. This offset reflects the degree of lag or lead of the temperature change relative to the current fluctuation.

[0027] Based on the phase shift φ, the energy coupling coefficient is calculated by combining the time series data of temperature and current. , through the formula ,in and are the average values of temperature and current measurements, respectively, and n is the number of measurements. The calculated energy coupling coefficient not only reveals the strength of the relationship between temperature and current, but also provides a quantitative basis for detecting potential faults.

[0028] Using the energy coupling coefficient Evaluate the correlation strength between temperature anomalies and current fluctuations and compare the correlation strength with the preset threshold to determine whether there is a potential fault risk. The moment when the value is greater than the preset threshold and the corresponding line segment position.

[0029] Step 6: Construct a dynamic threshold model based on the coupling coefficient, and optimize the threshold boundary through historical samples to form a fault judgment rule; specifically, the following steps are included: Collect temperature anomalies and current fluctuation events in historical operating data and extract the corresponding phase offset-energy coupling coefficient sequence , and mark the time when the known fault occurs; Perform statistical analysis on the coupling coefficient sequence, calculate the baseline mean μ and standard deviation σ under normal conditions, and set the preliminary judgment threshold ; This step uses statistical methods to analyze the historical data Analyze and identify its distribution characteristics under the "normal" state. Under normal circumstances, the data approximates a normal distribution, and the typical fluctuation range can be estimated by calculating its mean μ and standard deviation σ. Based on the three sigma principle (3σ principle), a preliminary fault judgment threshold is set to identify abnormal events that significantly deviate from normal behavior.

[0030] Will Compared with the coupling coefficient of the actual fault event, if the coupling coefficient of k consecutive time points is Exceed , then the warning mark is triggered, and the samples of this type are added to the training set to update the decision boundary. The update formula is: , where m is the number of original statistical samples; when k consecutive time points are detected The value exceeds the current threshold When a new abnormality occurs, it is considered a potential failure event and marked. Subsequently, these newly discovered abnormal samples are included in the training set, and the threshold is updated based on the original statistical data. A weighted average is used to integrate the new and old information, improving the model's generalization ability and real-time adaptability.

[0031] Based on the updated threshold Construct a multi-level decision interval. in to 1.2 It is marked as a level 1 warning when it exceeds 1.2 It is marked as a second-level warning, forming a fault judgment rule with hierarchical distinction.

[0032] Step 7: Match the fault judgment rules with the real-time graph to generate a fault probability heat map; specifically, the following steps are included: Obtain the historical fault frequency data of each joint position of the conductor, correspond it to the physical coordinates in the current line topology, and construct the initial probability distribution matrix , where each element represents the basic failure probability of the corresponding position; Map the fault judgment results obtained in real time to the matrix For each time window marked as abnormal, the weight value w is accumulated in the matrix according to its corresponding wire segment position. The update formula is: ,in Indicates the cumulative number of abnormalities at the location; by fusing historical and real-time information, dynamic updates of fault probabilities are achieved, improving the system's comprehensive recognition capabilities for sudden and cumulative faults.

[0033] Normalize the updated probability distribution matrix, introduce the neighborhood influence factor k, perform diffusion correction on the probability values of adjacent conductor segments, and calculate the final fault probability: By introducing a neighborhood influence factor, k, we take a weighted average of the failure probabilities of each location and its four neighboring points, and then superimpose them back on the original location to simulate the spatial propagation of fault risk. This operation can effectively identify potential fault diffusion paths.

[0034] based on Generate a two-dimensional visual heat map, use color gradients to represent the fault probability levels in different sections, mark high-probability areas with geographic coordinate information, and output it to the monitoring terminal.

[0035] Step 8: Output a graded warning signal based on the heat map; specifically, the following steps are included: Determine the warning level standards and Divided into multiple intervals, each interval corresponds to an early warning level , where n represents the number of different levels; right Each element in the traversal check is performed, and the warning level of each location is determined according to the interval standard. , the following formula is used to calculate the final warning level of each area: ,in is the lower threshold of the nth level warning; this formula means that for each location, find the probability interval to which it belongs and return the corresponding minimum warning level number n.

[0036] Generate a warning signal distribution map covering the entire monitoring area according to Level(i,j), merge consecutive areas with the same warning level to form a unified warning area , and assign a representative warning level to the region ; Will and its corresponding It is converted into an intuitive graphical interface and displayed on the monitoring terminal, using different colors or patterns to identify different levels of warning areas, ensuring that high-risk areas can be quickly identified and corresponding measures can be taken.

[0037] On the other hand, the present invention proposes an AI fault identification system for power lines, such as Figure 2 As shown, including: Multimodal data acquisition module, used to synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and wire joint coordinates, and generate spatiotemporally labeled datasets; A cross-modal time alignment module, configured to perform dynamic frequency compensation processing on the data set and construct an interpolation function to generate a cross-modal data stream of equal time granularity; The multi-physics feature fusion module is used to extract the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, combine the wire material parameters to construct a feature vector, and perform cross-modal interaction on the feature vector through spatiotemporal alignment convolution to generate a thermal-electric coupling map; An anomaly correlation modeling module is used to perform a sliding window analysis on the thermal-electric coupling map, calculate the phase offset-energy coupling coefficient between the temperature anomaly and the current fluctuation, build a dynamic threshold model based on the coupling coefficient, and optimize the threshold boundary through historical samples to form a fault judgment rule; The early warning module is used to match the fault judgment rules with the real-time map to generate a fault probability heat map, and output a graded early warning signal according to the fault probability heat map.

[0038] In addition, when executed, the above module is also used to implement other steps of the above-mentioned AI fault identification method for power lines, as shown in the following example: In this example, a 110kV transmission line was used as the monitoring target. Infrared thermal imaging sensors and current transformers were deployed to achieve real-time status perception of the conductor joint area. Multimodal data analysis was used to determine whether there was a potential fault risk.

[0039] Step 1: Synchronously acquire thermal imaging sequences and three-phase current waveforms to generate a spatiotemporal labeled dataset An infrared thermal imager and a current transformer are installed at each conductor joint to collect temperature images and three-phase current waveforms respectively. At the same time, the system has a built-in GPS module to record the timestamp of each frame of thermal imaging and each current sampling point. and spatial coordinates .

[0040] After the collection is completed, sort the timestamps and calculate the intervals between adjacent time points If Δt exceeds the preset maximum allowable value If the sampling interval is x (e.g. 2 seconds), the sampling is considered interrupted and the data segment is skipped.

[0041] Next, the spatial coordinates A one-to-one correspondence is established between each pixel and the location of the wire joint, corresponding to the node number in the power grid topology.

[0042] Finally, the thermal imaging temperature data The corresponding three-phase current Combined by time and space, a data set with time and space labels is formed: ; This step provides a unified temporal-spatial basis for subsequent analysis.

[0043] Step 2: Dynamic frequency compensation processing to build a cross-modal data stream with equal time granularity Due to the thermal imaging frame rate (e.g. 1Hz) is usually much lower than the current sampling rate (For example, 10kHz), first calculate the ratio of the two =10000 / 1=10000.

[0044] Then, based on the current signal, a uniform time step Δt = 1 second (i.e., the thermal imaging frame rate) is set.

[0045] Interpolation processing is performed on the thermal imaging data, and the two adjacent frames are Construct a linear temperature change function between: ; This allows for continuous temperature estimates to be obtained over time intervals of Δt.

[0046] For the current signal, a sliding average filter is used to merge the original sampling points N = 10,000 within 1 second into a mean value: ; Finally, the interpolated temperature data and the adjusted current data are aligned along the time axis to form a cross-modal data stream in a unified format: .

[0047] like Figure 3 As shown, the vertical axis is current and the horizontal axis is time. The figure shows the current change trend after time alignment. The current data is stable at around 100A without obvious abnormal fluctuations.

[0048] Step 3: Extract features and construct enhanced multidimensional feature vectors Next, extract the temperature gradient tensor from the thermal image to find the area with the largest local temperature rise ; Simultaneously extract transient harmonic energy from the current waveform .

[0049] Assume that the current conductor material parameters are as follows: Resistivity , thermal conductivity , linear expansion coefficient ; Construct the composite characteristic factor F: ; Then normalize and fuse it to get comprehensive features ; Finally, it is concatenated into an enhanced feature vector: .

[0050] Step 4: Constructing a Thermal-Electrical Coupling Map Using the above eigenvector V, a mapping matrix M is constructed in a two-dimensional space to map the eigenvalue of each wire joint to a pixel point in a unified coordinate system.

[0051] Then perform weighted diffusion processing: , where w is the interaction weight, such as 0.5.

[0052] The response values are superimposed over time to form a dynamic thermo-electric map, which is used to observe the evolution process of the abnormal area.

[0053] After high-response areas are clustered, their timestamps and wire segment numbers are marked to preliminarily identify suspected fault points.

[0054] Step 5: Calculate the phase shift-energy coupling coefficient Collect temperature Temp and current I data over a period of time and construct a time series .

[0055] Perform Fourier transform to get the spectrum and , and then calculate the phase shift φ: ;

[0056] Combined with time series data, the energy coupling coefficient is calculated : ; if If it is greater than the preset threshold, it is marked as an abnormal time point and the corresponding wire segment.

[0057] Step 6: Build a dynamic threshold model to form fault judgment rules Statistics of historical normal conditions Data, calculate the mean μ=0.8, standard deviation σ=0.1, and set the initial threshold: =0.8+0.3=1.1; If k=3 consecutive time points > , an alert is triggered and the threshold is updated: ; For example, m=100 samples, ,but: =(1.1*100+3.6) / (100+3)=1.104; Divide the interval according to the new threshold, when ∈[1.104,1.325), it is marked as a first-level warning, and when it is >1.325, it is marked as a second-level warning.

[0058] Step 7: Generate a failure probability heat map According to the historical fault frequency, construct the initial probability distribution matrix , then map the real-time detection results to the matrix and update the probability value corresponding to the number of abnormalities: ; Introduce the neighborhood influence factor k=0.2 for diffusion correction: ; Finally, a two-dimensional visual heat map is generated, using colors to represent the failure probability levels of different sections and marking high-probability areas.

[0059] Step 8: Output graded warning signals Set alert levels as follows: : <0.3 → normal; :0.3≤ <0.6→Level 1 warning; : ≥0.6→Level 2 warning.

[0060] For each position (i,j), check its Value, determine: ; Then generate the overall warning distribution map and merge the consecutive areas of the same level into , and assign corresponding warning levels .

[0061] Finally, a graphical interface is displayed on the monitoring terminal, with green representing normal, yellow representing level one warning, and red representing level two warning, making it easy for operation and maintenance personnel to quickly locate high-risk sections.

[0062] This embodiment achieves efficient identification and precise early warning of abnormal power line conductor joint conditions through multiple steps, including synchronous data acquisition, multimodal fusion, feature construction, graph generation, correlation modeling, dynamic threshold optimization, probability assessment, and graded early warning. Compared with traditional single-modal analysis methods, this method significantly improves the accuracy and response speed of fault identification, effectively reduces the false alarm rate, and enhances the safety and stability of power grid operations.

[0063] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An AI fault identification method for power lines, characterized in that: The following steps are involved: Synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and conductor joint coordinates, and generate a spatiotemporally labeled dataset. Performing dynamic frequency compensation processing on the data set, constructing an interpolation function to generate a cross-modal data stream with equal time granularity; Extract the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, combine them with the wire material parameters to construct a feature vector, and perform cross-modal interaction on the feature vector through spatiotemporal alignment convolution to generate a thermal-electric coupling map; Performing a sliding window analysis on the thermal-electric coupling map to calculate the phase offset-energy coupling coefficient between temperature anomalies and current fluctuations, building a dynamic threshold model based on the coupling coefficient, and optimizing the threshold boundary through historical samples to form a fault determination rule; The fault judgment rules are matched with the real-time graph to generate a fault probability heat map, and a graded warning signal is output according to the fault probability heat map.

2. The AI fault identification method for power lines according to claim 1, characterized in that: Generating a spatiotemporal labeling dataset includes: Infrared sensors and current transformers are deployed at the joints of each conductor in the power lines to collect thermal imaging sequences and three-phase current waveforms, while also recording the timestamp and spatial coordinates of each thermal image frame and each current sampling point. Sort the timestamps and determine the continuity of the samples based on the intervals between adjacent time points. If the interval exceeds a preset maximum value, an abnormal flag is triggered and the sample is skipped. Corresponding the spatial coordinates to the node numbers in the power line topology structure, and establishing a mapping relationship between each thermal imaging pixel point and the physical location of the wire connector; Based on the mapping relationship and the timestamp, the thermal imaging sequence and the three-phase current waveform are combined in the time and space dimensions to generate a data set with time and space labels.

3. The AI fault identification method for power lines according to claim 1, characterized in that: The generating of a cross-modal data stream of equal time granularity includes: Obtain the thermal imaging frame rate and current sampling rate, calculate the proportional relationship between the two, and set a unified time step based on the high-frequency signal; Performing time interpolation processing on the thermal imaging sequence, constructing a temperature change function between two adjacent frames to generate intermediate temperature data under a uniform time step; The three-phase current waveform is processed based on a unified time step, and the original sampling data in each time period is converted into a current value in a unified format; The interpolated temperature data and the adjusted current data are aligned along the time axis and combined into a multimodal data stream with the same time resolution.

4. The AI fault identification method for power lines according to claim 1, characterized in that: The step of constructing a feature vector by combining the conductor material parameters includes: Obtain physical property information of the wire, including resistivity, thermal conductivity, and linear expansion coefficient, and store the parameters in a feature construction module; The maximum temperature rise characteristics of the local area are extracted based on thermal imaging data, and the transient distortion energy characteristics are extracted from the three-phase current waveform; The maximum temperature rise characteristics and transient distortion energy are used to construct a composite characteristic factor in combination with the physical properties of the conductor. The composite characteristic factor is fused with the linear expansion coefficient to generate a normalized characteristic quantity, which is then concatenated with the original characteristic to form a characteristic vector.

5. The AI fault identification method for power lines according to claim 1, characterized in that: The generating of the thermal-electric coupling map comprises: Based on the eigenvectors, a two-dimensional spatial mapping matrix is constructed to map the eigenvalues of each wire joint position to pixel points in a unified coordinate system; Perform weighted diffusion processing on the two-dimensional spatial mapping matrix, introduce interaction weights within the pixel neighborhood, and update the local response value; The local response values are superimposed on the time dimension to form a dynamic thermal-electric distribution map in a continuous period; The high response areas in the dynamic thermo-electric distribution map are clustered and marked, hot spots with high thermo-electric coupling intensity are extracted, and their time and wire segment information are annotated.

6. The AI fault identification method for power lines according to claim 1, characterized in that: The calculation of the phase shift-energy coupling coefficient between the temperature anomaly and the current fluctuation includes: Collect temperature and current measurements at multiple time points to construct a time series dataset; Perform spectrum analysis on the time series data to obtain the frequency distribution of temperature and current, and calculate the phase offset between the two; Based on the phase shift combined with time series data, the correlation strength between temperature anomaly and current fluctuation is evaluated to obtain the energy coupling coefficient.

7. The AI fault identification method for power lines according to claim 1, characterized in that: The forming of the fault determination rule includes: Collect temperature anomalies and current fluctuation events in historical operating data, extract the corresponding energy coupling coefficient sequence, and mark the time when known faults occurred; Performing statistical analysis on the energy coupling coefficient sequence, determining a reference range under a normal state, and setting a preliminary determination threshold; Comparing the preliminary determination threshold with the actual event, if the energy coupling coefficient exceeds the threshold at multiple consecutive time points, triggering a warning flag and updating the determination boundary; Based on the updated decision boundary, multiple intervals are divided, and warning levels are set for different intervals to form fault decision rules with hierarchical differentiation.

8. The AI fault identification method for power lines according to claim 1, characterized in that: Generating a failure probability heat map includes: Obtain historical fault frequency data for each conductor joint location, correspond it to the physical coordinates in the current line topology, and construct an initial probability distribution matrix; Mapping the fault determination results obtained in real time to the initial probability distribution matrix, and updating the probability values of the locations marked as abnormal; The updated probability distribution matrix is normalized, and the neighborhood influence factor is introduced to perform diffusion correction on the probability values of adjacent areas. A visual heat map is generated based on the corrected probability distribution.

9. The AI fault identification method for power lines according to claim 1, characterized in that: Output graded warning signals based on the heat map, including: Setting multiple warning level standards based on the fault probability heat map, dividing the probability distribution into different intervals, each interval corresponding to a warning level; Check each location in the fault probability heat map and determine its warning level based on the interval it belongs to; Generate a warning signal distribution map covering the entire monitoring area according to the warning level, and merge consecutive areas with the same level to form a unified warning area; The unified warning area and its corresponding warning level are displayed on the monitoring terminal.

10. An AI fault identification system for power lines for implementing the method according to any one of claims 1 to 9, characterized in that: include: Multimodal data acquisition module, used to synchronously acquire thermal imaging sequences and three-phase current waveforms, record timestamps and wire joint coordinates, and generate spatiotemporally labeled datasets; A cross-modal time alignment module, configured to perform dynamic frequency compensation processing on the data set and construct an interpolation function to generate a cross-modal data stream of equal time granularity; The multi-physics feature fusion module is used to extract the thermal imaging temperature gradient tensor and the current transient harmonic energy distribution, combine the wire material parameters to construct a feature vector, and perform cross-modal interaction on the feature vector through spatiotemporal alignment convolution to generate a thermal-electric coupling map; An anomaly correlation modeling module is used to perform a sliding window analysis on the thermal-electric coupling map, calculate the phase offset-energy coupling coefficient between the temperature anomaly and the current fluctuation, build a dynamic threshold model based on the coupling coefficient, and optimize the threshold boundary through historical samples to form a fault judgment rule; The early warning module is used to match the fault judgment rules with the real-time map to generate a fault probability heat map, and output a graded early warning signal according to the fault probability heat map.

Images