A power distribution cable fault detection method and system

Abstract

This invention relates to the field of power distribution cable fault detection technology, specifically disclosing a power distribution cable fault detection method and system. The method involves synchronously collecting surface temperature, humidity, vibration, and partial discharge signals from a multi-modal sensor integrated into the cable body; subsequently, a low-power wireless ad hoc network based on a priority adaptive mechanism is used to aggregate the data with low latency; by performing spatiotemporal alignment and decoupling analysis based on blind source separation on the aggregated data, independent potential fault components are separated and the evolution sequences of key characteristic parameters are identified; then, the characteristic sequences are dynamically correlated with the cable's operating load and ambient temperature to construct a cable health status map that integrates a comprehensive health index and characteristic trajectories; finally, through intelligent matching of the map morphology with a fault characteristic spectrum library, the fault type and risk level are determined, and a structured operation and maintenance decision report is automatically generated.

Description

Technical Field

[0001] This invention relates to the field of power distribution cable fault detection technology, and specifically to a power distribution cable fault detection method and system. Background Technology

[0002] Against the backdrop of the deep integration of smart grids and the Industrial Internet, power distribution cables, as key carriers of power transmission, directly affect the reliability of power supply. Currently, cable condition monitoring technology based on the Internet of Things (IoT) has been widely applied. By deploying various smart sensors on cables, key parameters such as temperature and partial discharge are collected and monitored, aiming to detect potential faults in a timely manner.

[0003] The existing technology has the following shortcomings:

[0004] When cables vibrate under periodic external forces, traditional methods struggle to effectively distinguish between periodic partial discharges caused by vibration and persistent partial discharges resulting from insulation degradation itself. This misjudgment can lead to serious insulation defects being incorrectly attributed to transient external interference, thus delaying maintenance and eventually causing breakdown faults. Existing technologies, due to deficiencies in data acquisition synchronization and the ability to fuse and decouple multimodal data, cannot fundamentally solve this hidden but serious diagnostic challenge. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting faults in power distribution cables to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for detecting faults in power distribution cables includes the following steps:

[0008] S1: Real-time acquisition of multi-modal data of power distribution cables, including temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cables;

[0009] S2: The collected multimodal data is transmitted and aggregated in a low-latency manner in a wireless ad hoc network architecture through a low-power communication protocol;

[0010] S3: Perform fusion and decoupling analysis on the converged multimodal data to separate and identify the characteristic parameters corresponding to different fault types and their evolution sequence over time;

[0011] S4: Correlate the evolution sequence of characteristic parameters with cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple characteristic parameters as the vertical axis;

[0012] S5: Based on the cable health status map, the real-time health status and fault risk level of the cable are determined by matching and intelligent diagnosis with the cable fault feature spectrum library, and early warning and operation and maintenance decision reports are generated.

[0013] As a further aspect of the present invention: the low-latency transmission and aggregation of the collected multimodal data in a wireless ad hoc network architecture via a low-power communication protocol specifically includes:

[0014] Before sending data, a node first listens to the channel status and, based on its own data priority and buffer queue depth, selects one of several preset differentiated contention windows to back off, thus distinguishing the urgency of the service during the channel contention phase.

[0015] After the data is successfully transmitted to the next hop node, the corresponding receiving node immediately replies to the sending node with an acknowledgment frame containing its own energy level and link quality. Based on this, the sending node dynamically adjusts its transmission power to the lowest level of the corresponding neighboring node.

[0016] The routing path in the network is periodically reconstructed locally based on the dynamically updated link quality information in the acknowledgment frame, so that the data packet bypasses the node that is about to fail and is transmitted to the sink node along the optimal path.

[0017] As a further aspect of the present invention: the fusion and decoupling analysis of the aggregated multimodal data specifically includes:

[0018] Spatiotemporal alignment of temperature, humidity, vibration, and partial discharge data from different sensors is performed to unify all data under the same timestamp, forming a synchronized multimodal data matrix;

[0019] The multimodal data matrix is ​​composed of a linear mixture of multiple independent fault source signals. A decoupling matrix is ​​iteratively solved by constructing a cost function that aims to maximize the statistical independence of each output component.

[0020] By applying a decoupling matrix to transform the mixed data, several independent potential fault components are separated, and each fault component initially represents a dominant fault mode.

[0021] As a further aspect of the present invention: the formation of a synchronized multimodal data matrix specifically includes:

[0022] Establish a standard time series with fixed intervals at the data aggregation point;

[0023] Cubic spline interpolation is used for continuous temperature and humidity data, while piecewise linear interpolation that preserves the signal peak value is used for transient partial discharge and vibration data, so as to reconstruct the data with high fidelity at all standard time points.

[0024] All data values ​​reconstructed at the same standard time point are combined into a row vector, and all row vectors are stacked in chronological order to construct a multimodal data matrix with rows corresponding to time and columns corresponding to physical parameters. The data in the multimodal data matrix includes temperature, humidity, vibration and partial discharge data from different sensors.

[0025] As a further aspect of the present invention: the separation and identification of characteristic parameters corresponding to different fault types and their evolution sequences over time specifically includes:

[0026] For each potential fault component obtained after decoupling analysis, the time-domain and frequency-domain statistical characteristics of the fault component at different time scales are calculated to form a high-dimensional feature set.

[0027] By using predefined fault feature templates, high-dimensional feature sets are filtered and matched to identify key feature parameters of specific fault types such as cable overheating, insulation degradation, and mechanical shock.

[0028] Using a fixed time window, the instantaneous values ​​of key feature parameters are extracted by sliding and arranged in chronological order;

[0029] The instantaneous value sequence of each key feature parameter is smoothed and normalized to generate a continuous evolution sequence that changes over time.

[0030] As a further aspect of the present invention: the association of the feature parameter evolution sequence with cable operating state parameters specifically includes:

[0031] The real-time operating status parameters of the cable are obtained, including load current and ambient reference temperature.

[0032] For each feature parameter evolution sequence, an expected normal fluctuation range affected by the operating state is defined. For temperature features, the upper limit of the range is dynamically adjusted based on the sum of the ambient reference temperature and the theoretical temperature rise caused by the load current.

[0033] Each data point in the feature parameter evolution sequence is compared with its corresponding expected normal fluctuation range at that time, and the relative deviation of each data point is calculated.

[0034] As a further aspect of the present invention: the construction of the cable health status map with time as the horizontal axis and multiple feature parameters as the vertical axis specifically includes:

[0035] Establish a two-dimensional coordinate system, where the horizontal axis represents continuous time and the vertical axis represents multiple relative deviations obtained after correlation.

[0036] Multiple relative deviations at the same time point are weighted and superimposed according to the physical meaning of the relative deviation and the fault contribution weight to calculate the comprehensive health index. The curve of the comprehensive health index changing with time is plotted in a two-dimensional coordinate system as the main trend line of the cable health status map.

[0037] Below the main trend line, curves of different colors or line types are used to draw the relative deviation evolution trajectories of several key individual features in parallel, thus obtaining the cable health status map.

[0038] As a further aspect of the present invention: determining the real-time health status and fault risk level of the cable specifically includes:

[0039] Extract the shape of the comprehensive health index curve segment and key individual feature trajectory within a preset time window from the cable health status map;

[0040] Match the curve segments and their shapes with the standard evolution patterns pre-stored in the cable fault feature spectrum library, which are labeled with clear fault types and risk levels.

[0041] Based on the fault type corresponding to the highest similarity obtained from the matching and the preset similarity threshold, the current fault type of the cable is determined, and the real-time health status and risk level of the cable are classified and output based on the value of the highest similarity.

[0042] As a further aspect of the present invention: the generation of the early warning and operation and maintenance decision report specifically includes:

[0043] Based on the determined fault risk level, different levels of early warning signals are automatically triggered, including text prompts, color codes, and audible and visual warnings;

[0044] Based on the identified fault type, a targeted preliminary handling report is retrieved and combined from the preset operation and maintenance strategy knowledge base. The preliminary handling report includes the detection location, the fault phenomenon to be confirmed, and the preliminary handling steps.

[0045] The system automatically integrates fault type, risk level, early warning signal, evolution trend of key characteristic parameters, and preliminary handling report to generate a structured operation and maintenance decision report.

[0046] A power distribution cable fault detection system, comprising:

[0047] A multimodal collaborative sensing module is used to collect multimodal data of power distribution cables in real time. The multimodal data includes temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cable.

[0048] The low-power adaptive transmission module transmits and aggregates the collected multimodal data in a wireless ad hoc network architecture with low latency through a low-power communication protocol.

[0049] The data fusion and feature decoupling module is used to perform fusion and decoupling analysis on the aggregated multimodal data, and to separate and identify the feature parameters corresponding to different fault types and their evolution sequence over time.

[0050] The health status map construction module associates the evolution sequence of feature parameters with the cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple feature parameters as the vertical axis.

[0051] The intelligent diagnosis and decision generation module, based on the cable health status map, determines the real-time health status and fault risk level of the cable by matching and intelligently diagnosing it with the cable fault feature spectrum library, and generates early warning and operation and maintenance decision reports.

[0052] The beneficial effects of this invention are:

[0053] (1) This invention, by simultaneously acquiring multimodal data such as temperature, humidity, vibration, and partial discharge, and employing decoupling analysis based on maximizing negative entropy, can separate independent components corresponding to different fault types from mixed signals. It can effectively distinguish and accurately identify concurrent or coupled fault modes such as insulation degradation (partial discharge characteristics), joint overheating (temperature characteristics), and external force damage (vibration characteristics). By constructing a dynamic cable health status map and intelligently matching it with a fault feature spectrum library, it achieves intuitive characterization of fault evolution trends and accurate diagnosis based on morphological similarity, thereby enabling early warning at the incipient or early development stage of faults, thus gaining valuable time for operation and maintenance intervention. Attached Figure Description

[0054] The invention will now be further described with reference to the accompanying drawings.

[0055] Figure 1 This is a flowchart of the method of the present invention;

[0056] Figure 2 This is a block diagram of the system of the present invention. Detailed Implementation

[0057] 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.

[0058] Please see Figure 1As shown, the present invention is a method for detecting faults in power distribution cables, comprising the following steps:

[0059] S1: Real-time acquisition of multi-modal data of power distribution cables, including temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cables;

[0060] S2: The collected multimodal data is transmitted and aggregated in a low-latency manner in a wireless ad hoc network architecture through a low-power communication protocol;

[0061] S3: Perform fusion and decoupling analysis on the converged multimodal data to separate and identify the characteristic parameters corresponding to different fault types and their evolution sequence over time;

[0062] S4: Correlate the evolution sequence of characteristic parameters with cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple characteristic parameters as the vertical axis;

[0063] S5: Based on the cable health status map, the real-time health status and fault risk level of the cable are determined by matching and intelligent diagnosis with the cable fault feature spectrum library, and early warning and operation and maintenance decision reports are generated.

[0064] In S1, multimodal data of the power distribution cable is acquired in real time. This multimodal data includes: temperature, humidity, vibration, and partial discharge signals on the surface of the power distribution cable, specifically including:

[0065] The acquisition of multimodal data is achieved through a multimodal collaborative sensing module fixedly installed at key monitoring points on the power distribution cable body. The multimodal collaborative sensing module integrates a temperature sensing unit, a humidity sensing unit, a vibration sensing unit, and a partial discharge detection unit. The output signals of each unit are initially amplified and filtered by the signal conditioning circuit inside the module before being synchronously acquired and digitized by the microcontroller.

[0066] The temperature signal acquisition process is as follows: the temperature sensing unit directly contacts the cable surface or a special thermal pad through its sensitive element, converting the temperature change of the cable surface into a corresponding change in resistance or voltage parameters. After conditioning, the analog signal is read by the analog-to-digital converter in the microcontroller and finally converted into a digital signal representing the temperature value.

[0067] The process of acquiring humidity signals is as follows: The humidity sensing unit senses the relative humidity of the medium surrounding the cable joint or cable surface through its capacitive or resistive sensitive element. The electrical parameters of the element change linearly with the ambient water vapor content. This change is converted into a standard voltage signal by a dedicated drive circuit and acquired by a microcontroller to finally obtain a digital value of humidity.

[0068] The vibration signal acquisition process is as follows: The vibration sensing unit uses a microelectromechanical system (MEMS) accelerometer, which senses the minute acceleration of the cable caused by external mechanical force or internal fault and outputs a voltage signal proportional to the acceleration. After anti-aliasing filtering, this signal is acquired and digitized by the microcontroller at a set sampling rate to obtain the cable's vibration acceleration data.

[0069] The partial discharge signal acquisition process is as follows: The partial discharge detection unit couples high-frequency pulse current or electromagnetic wave signals generated by insulation defects in the cable joint or cable body through a high-frequency current transformer or an ultra-high-frequency sensor. The transient pulse signal output by the sensor is processed by a high-speed amplifier and a bandpass filter, and then captured and recorded by the high-speed analog-to-digital converter of the microcontroller, thereby obtaining the original pulse waveform of the partial discharge.

[0070] In S2, the collected multimodal data is transmitted and aggregated in a low-latency manner within the wireless ad hoc network architecture using a low-power communication protocol, specifically including:

[0071] The data aggregation node first establishes the network's time synchronization foundation by periodically broadcasting a reference clock signal to all sensor nodes within the network via a flooding time synchronization protocol. Upon receiving the signal, each sensor node calculates and calibrates its local clock based on its transmission hop count and the difference between the signal reception and transmission times, ensuring that all nodes in the network operate on the same time base, with time errors controlled within the millisecond level.

[0072] When preparing to transmit data, the sensor node first performs carrier sensing to check if the wireless channel is idle. If the channel is busy, the node selects a specific window value from several preset differentiated contention windows based on the priority of its data to be transmitted and the number of data packets in the current buffer queue. Data priority is determined according to the sensor type, with partial discharge data having the highest priority, followed by vibration data, and temperature and humidity data having a normal priority. The buffer queue depth is measured by the number of data packets and is divided into three levels: 1 to 3 packets is level 1, 4 to 6 packets is level 2, and 7 packets and above is level 3. The node queries a preset contention window mapping table based on the "priority-queue depth" combination, selects the corresponding minimum and maximum contention window values, and randomly selects an integer value within this range as the backoff time slot number for this transmission. The physical duration of each backoff time slot is defined as 512 microseconds. The node continuously listens to the channel, and after each idle time slot, the backoff counter is decremented by 1. When the counter reaches 0, the node gains access to the channel and begins transmitting data.

[0073] After data is successfully transmitted to the next-hop node, the receiving node immediately generates and replies with an acknowledgment frame. This acknowledgment frame contains confirmation of successful data reception, as well as two key fields: the receiving node's remaining energy percentage and the link quality index (LQ) for this reception. The LQ is calculated by dividing the signal strength of the successfully received data frame by its noise level; this ratio is represented in decimal. Upon receiving the acknowledgment frame, the sending node parses out the remaining energy value and the LQ. Internally, the node maintains a power adjustment mapping table that takes the LQ as input and outputs a suggested transmit power level. The mapping is as follows: a LQ above 40 uses the lowest power level 1; between 20 and 40 uses the intermediate power level 2; and below 20 uses the higher power level 3. The sending node will adjust its transmit power to the new level when communicating with that specific neighbor node accordingly.

[0074] Network routing path maintenance relies on dynamic information carried in acknowledgment frames. Each node maintains a link quality record for each of its neighbors. This record is a 10-bit first-in-first-out queue storing the link quality index from the acknowledgment frames of the 10 most recent successful communications. The current aggregate link quality of a neighbor node is defined as the arithmetic mean of all values ​​in this queue. Every fixed routing maintenance cycle, a node checks the current aggregate link quality of all its neighbors. If the aggregate link quality of a neighbor node remains below a preset switching threshold of 15 for three consecutive cycles, the node removes it from its preferred next-hop routing table. When forwarding packets, the node selects the path with the fewest hops from all available next-hop nodes with aggregate link quality above the threshold. If the hop counts are the same, the node with higher remaining energy is selected as the next hop, ensuring that data can adaptively bypass degraded nodes and be transmitted to the aggregation node along an optimized path.

[0075] In S3, the converged multimodal data undergoes fusion and decoupling analysis to separate and identify characteristic parameters corresponding to different fault types and their evolution sequences over time, specifically including:

[0076] At the data aggregation point, a standard time series is first established. This time series starts at the beginning of the data segment to be analyzed, with a fixed time interval of 1 second; that is, each standard time point is spaced 1 second apart. This standard time series will serve as a unified benchmark for time alignment of all sensor data.

[0077] For temperature, humidity, vibration, and partial discharge data from different sensors, differentiated interpolation methods are used for time alignment. Temperature and humidity data are continuously varying physical quantities, so cubic spline interpolation is employed. This method constructs a smooth curve using four adjacent raw data points, ensuring that the interpolated curve has continuous first and second derivatives at the data points. Specifically, for each standard time point, it is calculated which four consecutive raw data points it lies between, and then the corresponding temperature or humidity value is calculated based on the interpolation coefficients of these three adjacent intervals.

[0078] For data involving vibration and partial discharge, which may contain transient signals, a piecewise linear interpolation method that preserves signal peak values ​​is employed. This method first identifies all extreme points in the original data, including maxima and minima. Linear interpolation is then performed between two adjacent extreme points to ensure that the interpolated data retains the peak characteristics of the original signal. For each standard time point, it is determined which two extreme points it lies between, and then the value at that time point is calculated according to a linear relationship.

[0079] The temperature, humidity, vibration, and partial discharge characteristics obtained at each standard time point using the interpolation method described above are combined into a row vector. The dimension of this row vector is equal to the number of sensor parameter types used. Then, the row vectors corresponding to each standard time point are stacked sequentially in chronological order to form a two-dimensional data matrix. The rows of this matrix correspond to the standard time points, and the columns correspond to different sensor parameters; this matrix is ​​called a multimodal data matrix.

[0080] The resulting multimodal data matrix is ​​considered as an observed signal composed of a linear mixture of multiple independent fault source signals. The decoupling matrix is ​​solved by constructing a cost function that maximizes the statistical independence of each output component. An independence metric based on negative entropy is used as the optimization objective of the cost function; the negative entropy is obtained through approximation, specifically using an approximation method based on the maximum entropy principle. The decoupling matrix is ​​solved using a fixed-point iterative algorithm, which updates the decoupling matrix through multiple iterations to maximize the sum of the negative entropies of the output components.

[0081] The specific iterative process is as follows: First, the multimodal data matrix is ​​centered by subtracting the mean of each variable. Then, the centered data is whitened using principal component analysis to obtain the whitening matrix. The decoupling matrix is ​​initialized as an identity matrix. In each iteration, for each output component, its gradient update is calculated, based on the nonlinear function of the current component and the weighted difference between the component and the component. The corresponding row vectors of the decoupling matrix are updated. After updating all components, the decoupling matrix is ​​symmetrically orthogonalized. The above iterative process is repeated until the change in the row vectors of the decoupling matrix is ​​less than a set threshold, at which point the iteration stops.

[0082] The final decoupling matrix is ​​multiplied by the whitening matrix to obtain the complete separation matrix. This separation matrix is ​​then applied to perform a linear transformation on the centered multimodal data matrix, i.e., the sensor parameter vector at each time point is multiplied by the separation matrix on the left to obtain an output vector of the same dimension. Each component of the output is called a potential fault component, and these components are statistically independent, with each component initially representing a dominant fault mode.

[0083] For each potential fault component obtained after decoupling analysis, its time-domain and frequency-domain statistical characteristics at different time scales are calculated. The time-domain characteristics include:

[0084] Mean, which is the arithmetic mean of the component data sequence;

[0085] Variance is the average of the squares of the deviations of each data point from the mean.

[0086] Kurtosis is the ratio of the fourth central moment to the square of the variance minus the kurtosis.

[0087] Skewness is the ratio of the third central moment to the cube of the standard deviation.

[0088] The calculation of frequency domain features first involves performing a Fast Fourier Transform on the component data to obtain the spectrum, and then calculating:

[0089] The centroid of the spectrum is the frequency value of the weighted average of the amplitudes of each frequency component.

[0090] Spectral variance is the weighted average of the squares of the deviations of each frequency component from the centroid of the spectrum.

[0091] Spectral entropy is calculated by normalizing the spectrum to a probability distribution and then calculating the Shannon entropy.

[0092] These features together constitute a high-dimensional feature set.

[0093] High-dimensional feature sets are filtered and matched using predefined fault feature templates. These templates are a knowledge base containing feature value ranges corresponding to various typical fault types. For example, the feature template for cable overheating faults requires the mean of the temperature-related component to exceed a threshold of 60 degrees Celsius and the spectral centroid to be in the low-frequency range; insulation degradation faults require the kurtosis of the partial discharge component to exceed a threshold of 5 and the spectral entropy to be below a threshold of 2.5; and mechanical shock faults require the variance of the vibration component to exceed a threshold of 0.8 and the spectral variance to exceed a threshold of 1000. By matching the actually calculated high-dimensional features with these templates, key feature parameters corresponding to specific fault types are identified.

[0094] The instantaneous values ​​of key feature parameters are extracted using a sliding window with a fixed time window length of 10 minutes and a sliding step of 1 minute. For each time window, the average value of the key feature parameters within that window is calculated as the feature value at that time point. These feature values ​​are then arranged in chronological order to form the original time series of the feature parameters.

[0095] The original time series of each key feature parameter is smoothed using a moving average method with a window width of 5 data points. The moving average is calculated by taking the arithmetic mean of the data points within the window as the smoothed value at the center point of the window. The smoothed series is then normalized using a min-max normalization method, linearly transforming the data to the interval between 0 and 1. The normalization is specifically calculated by subtracting the minimum value from each data point and then dividing by the difference between the maximum and minimum values. This process generates a continuous evolution sequence of each key feature parameter over time, reflecting the dynamic development of different failure modes.

[0096] In S4, the evolution sequence of characteristic parameters is correlated with cable operating state parameters to construct a cable health status map with time as the horizontal axis and multiple characteristic parameters as the vertical axis, specifically including:

[0097] Real-time operating status parameters of the cable are acquired, including load current and ambient reference temperature. Load current is obtained from smart meters or monitoring terminals in the distribution network, with a sampling interval of 1 minute, and current values ​​are expressed in amperes. Ambient reference temperature is obtained from an independent temperature sensor installed in the cable trench or tunnel, away from the cable itself. This sensor is unaffected by cable heating, and the sampling interval is also 1 minute, with temperature values ​​expressed in degrees Celsius. These operating status parameters and the evolution sequence of characteristic parameters are aligned using a unified timestamp.

[0098] For each characteristic parameter evolution sequence, an expected normal fluctuation range influenced by operating conditions is defined. For temperature-related characteristic parameters, the upper limit of the expected normal fluctuation range is dynamically adjusted based on the sum of the ambient reference temperature and the theoretical temperature rise caused by the load current. The theoretical temperature rise is calculated as follows: First, the resistance per unit length is obtained according to the cable model and specifications. Then, based on the current load current, the heat generation power per unit length is calculated according to Joule's law. Next, the theoretical temperature rise under steady-state conditions is calculated based on the thermal resistance coefficient of the cable laying environment. Specifically, the theoretical temperature rise equals the heat generation power multiplied by the thermal resistance coefficient. The upper limit of the expected normal fluctuation range for temperature-related characteristic parameters equals the current ambient reference temperature plus the theoretical temperature rise, plus a fixed margin value set at 5 degrees Celsius. The lower limit of the range is set as the ambient reference temperature minus 3 degrees Celsius. For non-temperature-related characteristic parameters, the expected normal fluctuation range is obtained statistically from historical normal operation data, using the 5th and 95th percentiles of the historical data as the lower and upper limits of the range, respectively.

[0099] Each data point in the feature parameter evolution sequence is compared with its corresponding expected normal fluctuation range, and the relative deviation of each data point is calculated. The calculation of relative deviation involves two cases: when the feature parameter value exceeds the upper limit of the expected normal fluctuation range, the relative deviation equals (feature parameter value minus the upper limit of the range) divided by (upper limit of the range minus the lower limit of the range); when the feature parameter value is below the lower limit of the expected normal fluctuation range, the relative deviation equals (feature parameter value minus the lower limit of the range) divided by (upper limit of the range minus the lower limit of the range); when the feature parameter value is within the expected normal fluctuation range, the relative deviation is zero. Relative deviation is a dimensionless numerical value; a positive value indicates that the feature parameter is above the normal range, and a negative value indicates that it is below the normal range.

[0100] A two-dimensional coordinate system is established to construct a cable health status map. The horizontal axis of the coordinate system represents continuous time, covering the entire period to be monitored. The time scale is set according to monitoring requirements and can be minutes, hours, or days. The vertical axis represents multiple relative deviations obtained after correlation, including the relative deviations of the comprehensive health index and various key individual characteristics. The numerical range of the vertical axis is from -1 to +1, covering all possible relative deviation values.

[0101] The comprehensive health index is calculated by weighting and summing multiple relative deviations at the same time point according to their corresponding physical meanings and fault contribution weights. The weights are determined based on the correlation between the characteristic parameters and cable faults, through expert evaluation and historical fault data analysis. Specifically, the weights are allocated as follows: relative deviation of partial discharge characteristics is 0.4, relative deviation of temperature characteristics is 0.3, relative deviation of vibration characteristics is 0.2, and relative deviation of humidity characteristics is 0.1. The comprehensive health index is calculated using a weighted summation method, that is, each relative deviation is multiplied by its corresponding weight and then summed. The formula is: Comprehensive Health Index = Relative Deviation of Partial Discharge × 0.4 + Relative Deviation of Temperature × 0.3 + Relative Deviation of Vibration × 0.2 + Relative Deviation of Humidity × 0.1. In a two-dimensional coordinate system, with time as the x-axis and the comprehensive health index as the y-axis, a curve showing the comprehensive health index changing over time is plotted as the main trend line of the cable health status map.

[0102] Below the main trend line, curves of different colors or line types are used to plot the relative deviation evolution trajectories of several key individual characteristics in parallel. The color scheme is as follows: partial discharge relative deviation is plotted using a solid red line, temperature relative deviation using an orange dashed line, vibration relative deviation using a blue dotted line, and humidity relative deviation using a green dotted line. Each relative deviation is plotted as a curve over time on the same vertical axis scale, forming a detailed feature view corresponding to the main trend line. These curves together constitute a complete cable health status map, which can simultaneously reflect the changing trend of the overall cable health status and the abnormalities of each specific characteristic parameter.

[0103] The cable health status map is updated at the same frequency as the feature parameter evolution sequence, typically once per minute. When a relative deviation exceeds a set threshold, a marker is added to the corresponding time point in the map, recording the time and type of the abnormal event. The map supports zooming on the time axis, facilitating the viewing of health status changes at different time scales. The final cable health status map is stored as an image file, while the original data is also preserved for subsequent querying and analysis.

[0104] In S5, based on the cable health status map, the real-time health status and fault risk level of the cable are determined through matching and intelligent diagnosis with the cable fault feature spectrum library, and early warning and operation and maintenance decision reports are generated, specifically including:

[0105] The comprehensive health index curve segments and key individual feature trajectories within a preset time window are extracted from the cable health status map. The time window length is set to 30 minutes. The extracted content includes the continuous numerical sequence of the comprehensive health index on the main trend line, as well as the numerical sequences of relative deviations in partial discharge, temperature, vibration, and humidity within their respective time windows. The morphological characteristics of these curve segments are also recorded, including whether the overall trend is upward, downward, or stable, whether there are obvious peaks or troughs, and the amplitude and duration of the peaks and troughs.

[0106] The extracted curve segments and their shapes are matched against standard evolution patterns pre-stored in the cable fault feature spectrum library, which are labeled with clearly defined fault types and risk levels. The cable fault feature spectrum library contains standard patterns for various typical faults, each described by a set of feature parameters, including: trend direction parameters (increasing, decreasing, or stable); fluctuation amplitude parameters (defined as the difference between the maximum and minimum values ​​of the sequence); fluctuation duration parameters (the length of time from the start to the end of the fluctuation); and feature combination parameters (describing the correlation between the various feature parameters). The matching process uses a dynamic time warping algorithm to calculate similarity. This algorithm finds the optimal alignment path between two time series by constructing a cumulative distance matrix. The specific calculation process is as follows: First, the two time series are normalized to the same length range. Then, the Euclidean distance between each pair of points is calculated to form a distance matrix. Next, starting from the top left corner of the matrix, the minimum cumulative distance at each position is calculated step by step. Finally, the optimal path is found through backtracking. The average distance along the path is used as the difference between the two sequences, and the similarity is equal to 1 minus the normalized difference.

[0107] The current fault type of the cable is determined based on the fault type corresponding to the highest similarity score obtained from the matching, and a preset similarity threshold. The similarity threshold is set to 0.7; when the highest similarity score reaches or exceeds this threshold, the cable is deemed to have a corresponding fault type. The real-time health status of the cable is divided into four levels based on the value of the highest similarity score: when the highest similarity score is greater than or equal to 0.9, the health status is severe; when the highest similarity score is between 0.8 and 0.9, the health status is warning; when the highest similarity score is between 0.7 and 0.8, the health status is attentive; and when the highest similarity score is less than 0.7, the health status is normal. The fault risk level corresponds to the health status level, also divided into four levels, from high to low: high risk, medium risk, low risk, and normal.

[0108] Based on the determined fault risk level, different levels of early warning signals are automatically triggered. The early warning signals consist of three components: text prompts, color coding, and audible and visual alerts. Text prompts display different descriptions based on the risk level: high risk displays "Emergency Fault," medium risk displays "Severe Anomaly," low risk displays "Minor Anomaly," and normal status displays "Normal Operation." Color coding follows traffic light principles: high risk corresponds to red, medium risk to orange, low risk to yellow, and normal status to green. Audible and visual alerts are implemented through hardware: high risk triggers continuous audible and visual alarms, medium risk triggers intermittent audible and visual alarms, low risk triggers a single light signal alarm, and normal status does not trigger audible and visual alarms.

[0109] Based on the identified fault type, a targeted preliminary handling report is retrieved and combined from a pre-defined operation and maintenance strategy knowledge base. The knowledge base is indexed by fault type and stores standard handling solutions for various fault types. Each handling solution includes three parts: detection location, fault symptoms to be confirmed, and preliminary handling steps. The detection location clearly indicates the cable section or component requiring focused inspection, such as intermediate joints, terminations, or the cable body itself. The fault symptoms to be confirmed describe typical features that need to be verified on-site, such as carbonization marks on the insulation surface corresponding to partial discharge, or oxidation at connection points corresponding to overheating. The preliminary handling steps provide specific operational guidance, including necessary safety measures, detection methods, and temporary treatment solutions.

[0110] The system automatically integrates fault type, risk level, early warning signals, evolution trends of key characteristic parameters, and preliminary handling reports to generate a structured operation and maintenance decision report. The report uses a fixed format and includes six main parts: a basic information section recording the detection time, cable number, and location information; a diagnostic results section explaining the fault type and risk level; a feature analysis section displaying the evolution trend graph of key characteristic parameters; an early warning information section showing details of the triggered early warning signals; a handling suggestion section providing the content of the preliminary handling report; and a supplementary explanation section containing other relevant information and precautions. After the report is generated, it is automatically saved as a PDF file, and a report summary is pushed to the mobile terminals of relevant operation and maintenance personnel via a message middleware. Simultaneously, a complete diagnostic record, including raw data, intermediate calculation results, and the final report, is added to the database for subsequent querying and analysis.

[0111] Please see Figure 2 As shown, a power distribution cable fault detection system includes:

[0112] A multimodal collaborative sensing module is used to collect multimodal data of power distribution cables in real time. The multimodal data includes temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cable.

[0113] The low-power adaptive transmission module transmits and aggregates the collected multimodal data in a wireless ad hoc network architecture with low latency through a low-power communication protocol.

[0114] The data fusion and feature decoupling module is used to perform fusion and decoupling analysis on the aggregated multimodal data, and to separate and identify the feature parameters corresponding to different fault types and their evolution sequence over time.

[0115] The health status map construction module associates the evolution sequence of feature parameters with the cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple feature parameters as the vertical axis.

[0116] The intelligent diagnosis and decision generation module, based on the cable health status map, determines the real-time health status and fault risk level of the cable by matching and intelligently diagnosing it with the cable fault feature spectrum library, and generates early warning and operation and maintenance decision reports.

[0117] The working principle of this invention is as follows: Multimodal sensors fixedly installed on the cable body synchronously collect surface temperature, humidity, vibration, and partial discharge signals. A low-power wireless ad-hoc network based on priority and link state adaptation is used to reliably transmit and aggregate the collected data with low latency. At the data aggregation point, a synchronous multimodal data matrix is ​​constructed using a differentiated interpolation method. A blind source separation algorithm based on maximizing negative entropy is used to fuse and decouple the data, separating independent potential fault components. The time-domain and frequency-domain features of these components are then extracted, and key feature parameters are identified using fault feature templates, generating their evolutionary sequences over time. By dynamically correlating the feature parameter evolution sequences with operating status parameters such as cable load current and ambient temperature, the relative deviation is calculated, and a cable health status map integrating a comprehensive health index and individual feature trajectories is constructed. Finally, based on a dynamic time warping algorithm, the map shape is matched with a fault feature spectrum library to achieve accurate fault type identification and risk level classification, and a structured operation and maintenance decision report containing early warning information and handling suggestions is automatically generated, thus completing a closed loop from status perception to diagnostic decision-making.

[0118] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A power distribution cable fault detection method characterized by, Includes the following steps: S1: Real-time acquisition of multi-modal data of power distribution cables, including temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cables; S2: The collected multimodal data is transmitted and aggregated in a low-latency manner in a wireless ad hoc network architecture through a low-power communication protocol; S3: Perform fusion and decoupling analysis on the converged multimodal data to separate and identify the characteristic parameters corresponding to different fault types and their evolution sequence over time; S4: Correlate the evolution sequence of characteristic parameters with cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple characteristic parameters as the vertical axis; S5: Based on the cable health status map, the real-time health status and fault risk level of the cable are determined by matching and intelligent diagnosis with the cable fault feature spectrum library, and early warning and operation and maintenance decision reports are generated. The process of separating and identifying the characteristic parameters corresponding to different fault types and their evolution sequences over time specifically includes: For each potential fault component obtained after decoupling analysis, the time-domain and frequency-domain statistical characteristics of the fault component at different time scales are calculated to form a high-dimensional feature set. By using predefined fault feature templates, high-dimensional feature sets are filtered and matched to identify key feature parameters of specific fault types such as cable overheating, insulation degradation, and mechanical shock. Using a fixed time window, the instantaneous values ​​of key feature parameters are extracted by sliding and arranged in chronological order; The instantaneous value sequence of each key feature parameter is smoothed and normalized to generate a continuous evolution sequence that changes over time.

2. A power distribution cable fault detection method according to claim 1, characterised in that, The process of transmitting and aggregating the collected multimodal data in a wireless ad hoc network architecture with low latency via a low-power communication protocol specifically includes: Before sending data, a node first listens to the channel status and, based on its own data priority and buffer queue depth, selects one of several preset differentiated contention windows to back off, thus distinguishing the urgency of the service during the channel contention phase. After the data is successfully transmitted to the next hop node, the corresponding receiving node immediately replies to the sending node with an acknowledgment frame containing its own energy level and link quality. Based on this, the sending node dynamically adjusts its transmission power to the lowest level of the corresponding neighboring node. The routing path in the network is periodically reconstructed locally based on the dynamically updated link quality information in the acknowledgment frame, so that the data packet bypasses the node that is about to fail and is transmitted to the sink node along the optimal path.

3. A power distribution cable fault detection method according to claim 1, characterised in that, The aforementioned fusion and decoupling analysis of the aggregated multimodal data specifically includes: Spatiotemporal alignment of temperature, humidity, vibration, and partial discharge data from different sensors is performed to unify all data under the same timestamp, forming a synchronized multimodal data matrix; The multimodal data matrix is ​​composed of a linear mixture of multiple independent fault source signals. A decoupling matrix is ​​iteratively solved by constructing a cost function that aims to maximize the statistical independence of each output component. By applying a decoupling matrix to transform the mixed data, several independent potential fault components are separated, and each fault component initially represents a dominant fault mode.

4. A power distribution cable fault detection method according to claim 3, characterised in that, The formation of the synchronized multimodal data matrix specifically includes: Establish a standard time series with fixed intervals at the data aggregation point; Cubic spline interpolation is used for continuous temperature and humidity data, while piecewise linear interpolation that preserves the signal peak value is used for transient partial discharge and vibration data, so as to reconstruct the data with high fidelity at all standard time points. All data values ​​reconstructed at the same standard time point are combined into a row vector, and all row vectors are stacked in chronological order to construct a multimodal data matrix with rows corresponding to time and columns corresponding to physical parameters. The data in the multimodal data matrix includes temperature, humidity, vibration and partial discharge data from different sensors.

5. The method of claim 1, wherein, The association of the feature parameter evolution sequence with cable operating status parameters specifically includes: The real-time operating status parameters of the cable are obtained, including load current and ambient reference temperature. For each feature parameter evolution sequence, an expected normal fluctuation range affected by the operating state is defined. For temperature features, the upper limit of the range is dynamically adjusted based on the sum of the ambient reference temperature and the theoretical temperature rise caused by the load current. Each data point in the feature parameter evolution sequence is compared with its corresponding expected normal fluctuation range at that time, and the relative deviation of each data point is calculated.

6. A power distribution cable fault detection method according to claim 1, characterized in that, The construction of the cable health status map, with time as the horizontal axis and multiple feature parameters as the vertical axis, specifically includes: Establish a two-dimensional coordinate system, where the horizontal axis represents continuous time and the vertical axis represents multiple relative deviations obtained after correlation. Multiple relative deviations at the same time point are weighted and superimposed according to the physical meaning of the relative deviation and the fault contribution weight to calculate the comprehensive health index. The curve of the comprehensive health index changing with time is plotted in a two-dimensional coordinate system as the main trend line of the cable health status map. Below the main trend line, curves of different colors or line types are used to draw the relative deviation evolution trajectories of several key individual features in parallel, thus obtaining the cable health status map.

7. A power distribution cable fault detection method according to claim 1, wherein, The determination of the real-time health status and fault risk level of the cable specifically includes: Extract the shape of the comprehensive health index curve segment and key individual feature trajectory within a preset time window from the cable health status map; Match the curve segments and their shapes with the standard evolution patterns pre-stored in the cable fault feature spectrum library, which are labeled with clear fault types and risk levels. Based on the fault type corresponding to the highest similarity obtained from the matching and the preset similarity threshold, the current fault type of the cable is determined, and the real-time health status and risk level of the cable are classified and output based on the value of the highest similarity.

8. The method of claim 1, wherein, The generation of early warning and operation and maintenance decision reports specifically includes: Based on the determined fault risk level, different levels of early warning signals are automatically triggered, including text prompts, color codes, and audible and visual warnings; Based on the identified fault type, a targeted preliminary handling report is retrieved and combined from the preset operation and maintenance strategy knowledge base. The preliminary handling report includes the detection location, the fault phenomenon to be confirmed, and the preliminary handling steps. The system automatically integrates fault type, risk level, early warning signal, evolution trend of key characteristic parameters, and preliminary handling report to generate a structured operation and maintenance decision report.

9. A power distribution cable fault detection system characterized by, A method for performing a power distribution cable fault detection method as described in any one of claims 1-8, comprising: A multimodal collaborative sensing module is used to collect multimodal data of power distribution cables in real time. The multimodal data includes temperature, humidity, vibration and partial discharge signals on the surface of the power distribution cable. The low-power adaptive transmission module transmits and aggregates the collected multimodal data in a wireless ad hoc network architecture with low latency through a low-power communication protocol. The data fusion and feature decoupling module is used to perform fusion and decoupling analysis on the aggregated multimodal data, and to separate and identify the feature parameters corresponding to different fault types and their evolution sequence over time. The health status map construction module associates the evolution sequence of feature parameters with the cable operating status parameters to construct a cable health status map with time as the horizontal axis and multiple feature parameters as the vertical axis. The intelligent diagnosis and decision generation module, based on the cable health status map, determines the real-time health status and fault risk level of the cable by matching and intelligently diagnosing it with the cable fault feature spectrum library, and generates early warning and operation and maintenance decision reports.

Images