Cable branch box health early warning method based on fusion of insulation resistance and leakage current

Abstract

The invention belongs to the field of power system monitoring and fault diagnosis, and particularly relates to a cable branch box health early warning method based on fusion of insulation resistance and leakage current. The method comprises the following steps: acquiring insulation resistance and leakage current signals which are synchronously aligned by using a sensing monitoring unit, and extracting deterministic characteristics through nonlinear preprocessing; performing phase-space reconstruction on the leakage current signal, and calculating the maximum Lyapunov exponent, correlation dimension and other dynamic characteristic indexes so as to quantify the chaotic evolution degree responded by the insulating medium; and constructing an evaluation model based on fusion of insulation resistance and dynamic indexes, calculating a comprehensive health index, comparing the comprehensive health index with a threshold value, and determining an early warning level. According to the method, the qualitative change omen at the initial stage of insulation degradation can be sensitively captured by fusing the macroscopic electrical parameters and the microdynamics evolution characteristics, the advancement and the anti-interference capability of early warning are improved, reliable support is provided for refined operation and maintenance of the cable branch box, and the power supply reliability of a power system is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and fault diagnosis, specifically relating to a health early warning method for cable branch boxes based on the fusion of insulation resistance and leakage current. Background Technology

[0002] With the continuous expansion of modern power systems, cable distribution boxes, as key nodes in the distribution network, are crucial to the reliability and security of the entire power grid. Traditional cable monitoring methods mainly rely on periodic preventative tests or basic physical parameter monitoring. By collecting real-time data on cable insulation resistance and leakage current, the insulation performance of the equipment can be preliminarily assessed. Against the backdrop of the ongoing development of smart grids, achieving real-time and accurate early warning of the health status of cable distribution boxes has become a core element in preventing power accidents and optimizing operation and maintenance strategies.

[0003] Health early warning technology based on multi-dimensional parameter fusion provides a theoretical basis for identifying insulation aging and defects inside cables by integrating the dynamic changes in insulation resistance and leakage current. This type of technology aims to establish a mathematical model reflecting the degradation trend of the insulation medium through continuous monitoring of electrical parameters, and to issue timely early warning commands before a fault occurs.

[0004] Existing insulation monitoring methods often focus on the absolute value analysis of parameters. In the early stages of insulation degradation, the changes in insulation resistance and leakage current are not significant, making it difficult for traditional threshold-based early warning mechanisms to capture weak initial degradation signals. The complex electromagnetic interference in the power distribution environment generates a large amount of random noise. Existing linear signal processing techniques cannot distinguish between environmental Gaussian noise and deterministic signals generated by molecular chain breakage within the insulation system, affecting the accuracy of early warnings. Traditional analytical models are mostly based on linear dynamics assumptions, lacking in-depth analysis of the evolution of the insulation medium's response from a linear range to a nonlinear chaotic state. This makes it difficult to extract key dynamic features such as the maximum Lyapunov exponent through phase space reconstruction, resulting in insufficient ability of early warning systems to identify nonlinear state changes. Summary of the Invention

[0005] The purpose of this invention is to provide a cable branch box health early warning method based on the fusion of insulation resistance and leakage current, which can solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a cable branch box health early warning method based on the fusion of insulation resistance and leakage current, comprising the following specific steps: Step 1: The raw operating data is acquired in real time by the sensing and monitoring unit deployed inside the cable branch box. The raw operating data includes at least the insulation resistance time series signal and the leakage current time series signal, and the synchronous clock triggering mechanism is used to ensure that the two types of signals are phase aligned on the time axis. Step 2: Perform nonlinear preprocessing on the original running data. Use a combination algorithm based on multi-scale morphological filtering and adaptive threshold noise reduction to remove random Gaussian noise and periodic narrowband interference mixed in the original signal, and extract deterministic feature signals that reflect the evolution state of molecular chains inside the insulation system. Step 3: Perform phase space reconstruction on the preprocessed leakage current time series signal. By calculating the optimal delay time and minimum embedding dimension, the one-dimensional leakage current time series is mapped to a high-dimensional phase space to construct a phase space trajectory matrix that can characterize the dynamic properties of the insulating medium. Step 4: Calculate the nonlinear dynamic characteristic index based on the reconstructed phase space trajectory matrix. The characteristic index includes the maximum Lyapunov exponent and the correlation dimension, which are used to quantify the degree of deterministic chaos in the evolution of the insulating medium response from the linear interval to the nonlinear chaotic state. Step 5: Construct a health evaluation model based on the fusion of insulation resistance and dynamic characteristic indicators. Use real-time insulation resistance values, maximum Lyapunov exponent, and correlation dimensions as multi-dimensional input variables to calculate the comprehensive health index of the cable branch box. Step 6: Compare the comprehensive health index with the preset health status level threshold in real time to determine the current health warning level and output the corresponding warning instruction.

[0007] Preferably, in step 1, high-precision insulation monitoring sensors continuously sample each branch cable of the cable branch box, with the sampling frequency dynamically adjusted according to a preset power grid operating frequency. The synchronization clock triggering mechanism achieves nanosecond-level synchronization between distributed sensing units by receiving BeiDou or Global Positioning System timing signals, ensuring that paired samples of insulation resistance and leakage current are acquired at the same instant. The insulation resistance time series signal reflects the macroscopic impedance characteristics of the cable insulation medium under a specific voltage, while the leakage current time series signal contains information on microscopic disturbances caused by molecular chain breakage or weak partial discharge within the insulation material.

[0008] Preferably, step 2, performing nonlinear preprocessing on the original running data, specifically includes: using structuring elements to perform multi-scale morphological opening and closing operations on the signal to smooth pulse-type interference in the signal; employing adaptive wavelet packet decomposition technology to divide the signal into multiple frequency bands, and calculating the energy distribution of each frequency band; for frequency bands identified as high-frequency random noise, a soft thresholding function is used for shrinkage processing. The threshold of the soft thresholding function is determined based on the root mean square error of the signal and a preset scaling factor. The processed signal needs to be reconstructed through inverse transform to obtain a pure deterministic signal. This process ensures that the input signal in subsequent chaotic analysis has a high degree of physical authenticity and avoids false alarms caused by environmental electromagnetic interference.

[0009] Preferably, in step 3, the phase space reconstruction process specifically involves: calculating the delay time of the leakage current time series signal using the autocorrelation function method, and selecting the moment when the autocorrelation function drops to a specific proportion of its initial value as the optimal delay time. A false nearest neighbor algorithm is used to determine the minimum embedding dimension. By observing the change in the proportion of nearest neighbors as the dimension increases, when this proportion is lower than a preset threshold, it is determined as the final embedding dimension. Based on the delay time and embedding dimension, the one-dimensional sequence is transformed into a multi-dimensional vector group, forming trajectory points in the phase space. The spatial distribution of these trajectory points reflects the attractor characteristics of the insulation system's operating state. If the system is in a healthy state, the trajectory points exhibit strong regularity; if early insulation degradation occurs, the trajectory points will show a trend towards evolving into nonlinear chaotic attractors.

[0010] Preferably, in step 4, the specific process of calculating the maximum Lyapunov exponent is as follows: A reference trajectory point is selected in the reconstructed phase space, its neighboring points are found, and the initial Euclidean distance between the two points is calculated. As time evolves, the trajectories of these two points in the phase space are tracked, and the instantaneous Euclidean distance after a specific evolution step is calculated. The natural logarithm of the ratio of the instantaneous Euclidean distance to the initial Euclidean distance is calculated, and the arithmetic mean of the logarithmic ratios over multiple evolution cycles is obtained. When this average value is greater than 0, it indicates that the system has chaotic characteristics. The magnitude of the maximum Lyapunov exponent reflects the rate at which the insulation degradation signal deviates from the linear equilibrium state; a larger value indicates a more pronounced qualitative change in insulation performance.

[0011] Preferably, in step 4, the specific process for calculating the correlation dimension is as follows: For all trajectory points in the reconstructed phase space, a measurement radius is defined, and the proportion of point pairs whose distance between any two points in the space is less than the measurement radius is calculated to obtain the correlation integral function. By continuously changing the value of the measurement radius, a series of corresponding correlation integral values ​​are obtained. In a double logarithmic coordinate system, the linear regression slope between the natural logarithm of the correlation integral function and the natural logarithm of the measurement radius is calculated. This slope is the correlation dimension. The value of the correlation dimension reflects the degree of spatial occupancy of the insulating medium response signal in the phase space. When microscopic damage or weak discharge occurs inside the insulating material, the complexity of the signal increases, causing fluctuations in the correlation dimension.

[0012] Preferably, in step 5, the construction of the health assessment model involves the normalization of multi-dimensional features. The measured insulation resistance value is compared with a preset standard insulation resistance benchmark value, and the deviation percentage is calculated. The calculated maximum Lyapunov exponent and the correlation dimension are mapped to a standard interval between 0 and 1. A weighted fusion algorithm is used, based on weight coefficients trained from historical fault samples, to weight and sum the insulation resistance deviation, maximum Lyapunov exponent deviation, and correlation dimension volatility to generate a comprehensive health index. The allocation of the weight coefficients tends to assign higher weights to dynamic characteristic indicators in the early stages of degradation to capture weak early signals, while increasing the weight of insulation resistance in the later stages of degradation.

[0013] Preferably, in step 6, the health status level is divided into multiple levels, including normal operation, initial deterioration, early warning and attention status, and severe fault status. The normal operation status corresponds to a comprehensive health index within a preset first interval; the initial deterioration status corresponds to a comprehensive health index within a preset second interval, where although the insulation resistance value has not changed, the maximum Lyapunov exponent has shown a definite chaotic trend. The early warning and attention status corresponds to a comprehensive health index within a preset third interval, meaning that insulation damage has reached an observable level. The severe fault status corresponds to a comprehensive health index exceeding a preset upper threshold. Based on different levels, the system sends alarm information of different priorities to the operation and maintenance platform via the communication network.

[0014] Preferably, the method further includes establishing a long-term trend monitoring database. The system periodically stores the comprehensive health index, dynamic characteristic indicators, and environmental temperature and humidity data obtained from each calculation into a storage unit. A long short-term memory neural network is used to learn the historical trajectory of the health index to construct a trend prediction model. This model predicts the trend of the health index within a specific future time period. If the predicted value will exceed the warning threshold within a specific future time period, a trend warning is triggered in advance. This predictive maintenance mechanism based on in-depth mining of historical data can further extend the maintenance window of cable branch boxes.

[0015] Preferably, the sensing and monitoring unit includes a microcurrent sensor installed at the cable joint and an insulation resistance detection module on the high-voltage side. The microcurrent sensor adopts an open structure made of high-permeability material and has a microampere-level sensing accuracy. The insulation resistance detection module calculates the leakage current component and converts it into real-time insulation resistance by superimposing a specific low-frequency detection signal on the system operating voltage. All sensors are subjected to moisture-proof and corrosion-proof encapsulation to adapt to the potentially harsh environment inside the cable branch box.

[0016] Preferably, the multi-scale morphological filtering process in step 2 includes: defining a sequence of structuring elements of increasing length, and sequentially performing erosion, dilation, opening, and closing operations on the signal. A feature extraction operator is constructed by calculating the difference between the processing results at different scales. This operator can identify and enhance components with deterministic dynamic characteristics in the signal while suppressing uncorrelated random thermal noise. This step is crucial for extracting chaotic signals in the early stages of insulation degradation from a strong noise background and is the foundation for achieving high-sensitivity early warning.

[0017] Preferably, the phase space trajectory matrix in step 3 consists of multiple column vectors, each representing a state point in the phase space. By performing singular value decomposition on the trajectory matrix, singular spectrum features reflecting the main characteristics of the system can be further extracted. Using the principal component energy distribution of the singular spectrum as an auxiliary criterion, it is possible to distinguish between current changes caused by load fluctuations and nonlinear abrupt changes caused by insulation degradation, thereby reducing the false alarm rate of the system.

[0018] Preferably, the multi-dimensional feature fusion process in step 5 employs fuzzy inference logic. The rate of change of insulation resistance, the absolute value of the maximum Lyapunov exponent, and the abrupt change intensity of the correlation dimension are used as input variables for the fuzzy set. Through a preset membership function, continuous physical quantities are transformed into fuzzy linguistic variables. Inference is performed based on a fuzzy rule table established by an expert knowledge base to derive a fuzzy comprehensive evaluation vector reflecting the health status. Through defuzzification calculation, a unique comprehensive health index value is obtained. This processing method enhances the early warning model's tolerance to sensor measurement errors and parameter uncertainties.

[0019] Preferably, the method compensates for the environmental conditions of the cable branch box during execution. By monitoring the operating environment with temperature and humidity sensors, when the ambient humidity exceeds a preset range, the system automatically increases the threshold value of insulation resistance in the health assessment model, increasing the reliance on dynamic indicators such as the maximum Lyapunov exponent, which are unaffected by humidity. This dynamic compensation mechanism ensures that the early warning system maintains stable detection sensitivity under different climatic conditions.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention solves the problem of difficulty in identifying early, subtle degradation by introducing chaotic analysis from nonlinear dynamics into the health monitoring of cable branch boxes. Instead of relying solely on the absolute values ​​of insulation resistance and leakage current, it captures deterministic chaotic signals generated by molecular chain breakage or minute discharges within the insulation material by extracting the maximum Lyapunov exponent and correlation dimensions.

[0021] 2. This enables the system to detect signs of qualitative change from the dynamic characteristics of the signal before a quantitative change in physical quantities occurs, thus improving the predictability of the early warning.

[0022] 3. This invention utilizes phase space reconstruction and nonlinear filtering techniques to enhance the system's ability to resist environmental noise interference. Chaotic analysis can essentially distinguish between random environmental Gaussian noise and the deterministic nonlinear response of the insulation system itself.

[0023] 4. Through multi-scale morphological preprocessing, random fluctuations caused by electromagnetic interference are removed, enabling the early warning system to maintain high accuracy in complex power distribution network environments and reducing the risk of false alarms and missed alarms caused by noise fluctuations.

[0024] 5. This invention constructs a more scientific health assessment model by fusing multi-dimensional features of insulation resistance and dynamic indicators. This model comprehensively considers macroscopic electrical insulation parameters and microscopic nonlinear dynamic evolution characteristics, achieving a holistic description of the health status of cable branch boxes.

[0025] 6. Through dynamic weight allocation and trend prediction mechanisms, not only can the current status be assessed in real time, but also the future fault risks can be predicted, providing reliable data support for the refined operation and maintenance of cable branch boxes, which is of great significance for ensuring the power supply reliability of the power system. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework for identifying early insulation degradation based on nonlinear dynamic chaotic analysis in this invention. Figure 3 This is a flowchart illustrating the logical process of performing nonlinear preprocessing on the original running data based on multi-scale morphological filtering and adaptive threshold denoising in this invention. Figure 4 This is a flowchart illustrating the logical process of performing phase space reconstruction on the leakage current signal and calculating the maximum Lyapunov exponent and the correlation dimension in this invention. Figure 5 This is a logical flowchart of the process in this invention that uses the fusion of insulation resistance and dynamic characteristic indicators, combined with dynamic weight allocation, to generate a comprehensive health index. Detailed Implementation

[0027] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0028] During the operation and maintenance of cable branch boxes, real-time monitoring of their health status is crucial for ensuring the safe operation of the power distribution network. This embodiment provides a cable branch box health early warning method based on the fusion of insulation resistance and leakage current. By deeply integrating macroscopic electrical parameters with microscopic nonlinear dynamic characteristics, it achieves advanced perception of insulation degradation.

[0029] In step 1, raw operating data is acquired in real time by a sensing and monitoring unit deployed inside the cable branch box. This raw operating data includes at least an insulation resistance time-series signal and a leakage current time-series signal. In actual deployment, the sensing and monitoring unit is precisely installed at each cable branch joint within the cable branch box. Leakage current is acquired by a high-precision micro-current sensor. This sensor employs an open structure made of permalloy material with high magnetic permeability, possessing magnetic induction sensitivity capable of capturing microampere-level weak leakage current fluctuations. Its secondary coil is precisely wound and used in conjunction with a multi-stage active filter circuit to ensure that the acquired analog current signal has a good signal-to-noise ratio before being converted to a digital signal. The insulation resistance detection module superimposes a specific low-frequency probe DC signal or a frequency-converted AC signal onto the system's AC operating voltage. By measuring the weak DC or AC leakage component generated by this probe signal, and based on the verbal expression of Ohm's law, the real-time insulation resistance value is calculated.

[0030] To ensure the accuracy of subsequent fusion analysis, step 1 utilizes a synchronous clock triggering mechanism to ensure phase alignment of the two types of signals on the time axis. This synchronous clock triggering mechanism uses BeiDou or GPS timing signals as a global time reference to achieve nanosecond-level synchronous triggering among distributed sensing units. The main controller periodically sends synchronization pulses to each acquisition module, and the acquisition module simultaneously starts sampling the insulation resistance and leakage current upon receiving the pulse. The sampling frequency is dynamically adjusted according to a preset power grid operating frequency. For example, when the power grid frequency fluctuates around 50 Hz, the sampling frequency is set to 256 times or higher to satisfy the Nyquist sampling theorem and retain sufficient high-frequency signal details. The acquired signals are encapsulated according to a preset data structure, with each data packet containing high-precision timestamp information to ensure that the insulation resistance and leakage current acquired at the same instant can form paired samples in subsequent digital signal processing logic.

[0031] In step 2, nonlinear preprocessing is performed on the raw running data. This process employs a combined algorithm based on multi-scale morphological filtering and adaptive thresholding denoising. Multi-scale morphological opening and closing operations are performed on the signal using structuring elements. A sequence of structuring elements with progressively increasing lengths is defined, forming different detection windows on the time axis. Positive impulse noise is removed by performing erosion, negative impulse noise by performing dilation, peaks are smoothed by opening, and troughs are filled by closing. By calculating the difference between the processing results at different scales, a feature extraction operator is constructed to smooth random spike interference in the signal while preserving the slow evolution trend reflecting insulation degradation. Adaptive wavelet packet decomposition is used to divide the signal into multiple frequency bands, and the energy distribution of each band is calculated. The raw signal is passed through a set of low-pass and high-pass filters to decompose it into approximate and detail components. For frequency bands identified as high-frequency random noise, a soft thresholding function is used for shrinkage processing. The threshold calculation logic of the soft threshold function is as follows: calculate the root mean square error of the signal in the no-signal interval, multiply the error value by a preset scaling factor, which is usually set empirically between 2 and 3. If the absolute value of the wavelet coefficient is less than the threshold, the coefficient is set to 0; if it is greater than the threshold, the threshold is subtracted from or added to the coefficient. The processed signal is reconstructed through inverse wavelet packet transform to obtain a pure deterministic signal. This process not only eliminates random Gaussian noise mixed in the original signal and periodic narrowband interference caused by surrounding power equipment, but also extracts deterministic feature signals that can truly reflect the evolution state of molecular chains inside the insulation system.

[0032] In step 3, the preprocessed leakage current time series signal is reconstructed in phase space. The theoretical basis of phase space reconstruction lies in the delay embedding theorem, whose core objective is to transform a one-dimensional time series into a high-dimensional spatial trajectory that can display the full picture of system dynamics. Specifically, the delay time of the leakage current time series signal is calculated using the autocorrelation function method. In practice, the correlation coefficient between the sequence and its own delayed sequence is calculated, and the change curve of the correlation coefficient is observed as the delay time increases. The optimal delay time is selected when the autocorrelation coefficient decreases from its initial maximum value of 1 to 1 divided by the natural constant (approximately 37%) or to the first local minimum. The spurious nearest neighbor algorithm is used to determine the minimum embedding dimension. In the reconstructed vector space, as the embedding dimension gradually increases, it is examined whether pairs of points that were originally very close will separate due to the increase in dimension. The proportion of spurious nearest neighbors caused by the increase in dimension is statistically analyzed. When this proportion is lower than a preset minimum threshold (e.g., 5%), the current dimension is determined as the minimum embedding dimension. Based on the aforementioned delay time and embedding dimension, the one-dimensional sequence is transformed into a phase space trajectory matrix composed of multiple multi-dimensional vectors. Each column of this trajectory matrix represents a state point in phase space, reflecting the attractor characteristics of the operating state of the insulation system. If the system is in a healthy state, the attractor trajectory points exhibit strong regularity and centripetal force in space; if early degradation occurs inside the insulating material, such as weak discharge caused by molecular chain breakage, the trajectory points will show a trend of evolving into nonlinear chaotic attractors, and their spatial distribution becomes complex and has a certain degree of divergence.

[0033] In the specific implementation of step 3, to further improve adaptability to complex operating conditions, singular value decomposition (SVD) is performed on the phase space trajectory matrix. The trajectory matrix is ​​decomposed into a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix. By extracting the first few principal singular values ​​from the singular value diagonal matrix, a singular spectrum reflecting the main characteristics of the system is constructed. The energy distribution of the principal components of the singular spectrum is used as an auxiliary criterion. Under normal load fluctuations, the energy distribution of the singular spectrum usually remains stable and concentrated in low-frequency components; however, when a nonlinear abrupt change occurs in the insulation, the energy proportion of higher-order singular values ​​increases. In this way, it is possible to distinguish between current changes caused by load fluctuations and dynamic abrupt changes caused by insulation degradation.

[0034] In step 4, nonlinear dynamic characteristic indices are calculated based on the reconstructed phase space trajectory matrix. These indices include the maximum Lyapunov exponent. The specific process for calculating the maximum Lyapunov exponent is as follows: Reference trajectory points are randomly selected in the reconstructed phase space trajectory, and a search algorithm is used to find their nearest Euclidean neighbor. The Euclidean distance between these two points is recorded at the initial moment. As time progresses, the trajectories of these two points in phase space are tracked. After a preset evolution step size, the instantaneous Euclidean distance between the two points at their new positions is calculated. The ratio of this instantaneous Euclidean distance to the initial Euclidean distance is calculated, and the natural logarithm of this quotient is taken. To eliminate the randomness of local fluctuations, the arithmetic mean of the logarithmic ratios over multiple evolution cycles is calculated. When this arithmetic mean is consistently greater than 0, it indicates, in physical logic, that the system exhibits trajectory divergence, i.e., the system possesses deterministic chaotic characteristics. The larger the value of the maximum Lyapunov exponent, the faster the insulation degradation signal deviates from the linear equilibrium state, indicating a more pronounced qualitative change trend in insulation performance.

[0035] In step 4, the feature index also includes the correlation dimension. The specific process for calculating the correlation dimension is as follows: For all trajectory points in the reconstructed phase space, a measurement radius is defined. For a given measurement radius, the number of point pairs in the space whose distance is less than the measurement radius is counted. This number of point pairs is divided by the square of the total number of point pairs to obtain the correlation integral function. By continuously changing the value of the measurement radius, a series of corresponding correlation integral calculation values ​​are obtained. In a double logarithmic coordinate system (i.e., the horizontal axis is the natural logarithm of the measurement radius, and the vertical axis is the natural logarithm of the correlation integral function), linear regression is performed using the least squares method, and the slope of the regression line is calculated. This slope is the correlation dimension. The value of the correlation dimension reflects the spatial occupancy and self-similarity of the insulating medium response signal in the phase space. When microscopic damage occurs inside the insulating material, the complexity of the signal increases, causing the value of the correlation dimension to rise or fluctuate drastically.

[0036] In step 5, a health evaluation model based on the fusion of insulation resistance and dynamic characteristic indicators is constructed. This process involves the normalization and fusion logic of multi-dimensional features. The measured insulation resistance value is compared with a preset standard insulation resistance benchmark value (usually the factory test value or the average of historical stable operating values), and the deviation percentage is calculated, i.e., the measured value is subtracted from the benchmark value and then divided by the benchmark value. The calculated maximum Lyapunov index and correlation dimension are mapped to a standard range between 0 and 1 through linear mapping or a sigmoid function. A weighted fusion algorithm is used to generate a comprehensive health index. The specific weight coefficient allocation logic follows the principle of dynamic adjustment: in the early stage of equipment operation, due to the weak changes in physical quantities, the system assigns higher weights to the maximum Lyapunov index and correlation dimension (e.g., weight ratios of 40% and 40% respectively), while the weight of insulation resistance deviation is set to 20% to capture weak early degradation signals. As the overall health index declines, when the system recognizes that the signal has entered a period of significant degradation, it automatically increases the weight of the insulation resistance deviation (e.g., to 60%) to strengthen the monitoring of the quantitative change process.

[0037] In the preferred embodiment of step 5, the multi-dimensional feature fusion process employs fuzzy inference logic. The rate of change of insulation resistance, the absolute value of the maximum Lyapunov exponent, and the abrupt change intensity of the correlation dimension are used as input variables for the fuzzy set. For each input variable, multiple membership functions (e.g., normal, slight deviation, severe fluctuation) are preset. Fuzzy inference is performed using a fuzzy rule table established through an expert knowledge base (e.g., if the maximum Lyapunov exponent increases and the correlation dimension fluctuates drastically, the health level is judged as a warning). The resulting fuzzy evaluation vector is defuzzified using the centroid method or the maximum membership method, ultimately yielding a unique comprehensive health index value ranging from 0 to 100. The closer the value is to 100, the healthier the health; the closer it is to 0, the higher the risk of failure.

[0038] An environmental condition compensation mechanism is also introduced during the implementation of the method. Temperature and humidity sensors deployed within the branch box monitor the operating environment. When the ambient humidity exceeds a preset threshold (e.g., 80%), considering that moisture adsorption on the insulation surface can cause a significant drop in resistance, the system automatically increases the threshold value for insulation resistance deviation in the health evaluation model. The system automatically increases its reliance on the maximum Lyapunov exponent, as this kinetic index primarily reflects the deterministic nonlinear evolution within the system and is far less affected by external humidity interference than the measured resistance value. This dynamic compensation mechanism ensures that the early warning system maintains stable detection sensitivity under harsh conditions such as the rainy season or humid underground environments, avoiding false alarms.

[0039] In step 6, the current health warning level is determined by comparing the comprehensive health index with a preset health status level threshold in real time. Specifically, it is divided into four levels: First, normal operation. When the comprehensive health index is in the first range (e.g., 90 to 100 points), it is considered normal, and the system only maintains routine monitoring. Second, initial degradation. When the comprehensive health index is in the second range (e.g., 75 to 90 points), the insulation resistance may not have decreased by orders of magnitude, but the maximum Lyapunov exponent has shown a definite chaotic trend and the correlation dimension has begun to rise. The system determines this as initial degradation and triggers a maintenance plan recommendation. Third, warning and attention status. When the comprehensive health index is in the third range (e.g., 60 to 75 points), it means that the insulation damage has reached an observable level, and the system sends an orange warning to the operation and maintenance platform, requiring manual inspection. Fourth, severe fault status. When the comprehensive health index is below 60 points or exceeds the preset extreme value upper limit threshold, it is determined that there is an immediate risk of breakdown. Based on different levels, the system sends alarm information of different priorities to the remote operation and maintenance platform via fiber optic Ethernet or 5G industrial communication modules, including the device number, the probability of the fault type, and the estimated remaining lifespan.

[0040] The method in Embodiment 1 further includes establishing a long-term trend monitoring database. The system periodically stores the calculated comprehensive health index, maximum Lyapunov index, correlation dimensions, and environmental temperature and humidity data into local flash memory or cloud storage units, either hourly or daily. A Long Short-Term Memory (LSTM) neural network is used to perform deep learning on the historical trajectory of the health index. This network includes an input layer, forget gate, input gate, unit states, and an output layer. By learning from the evolution patterns of data over the past few months or even years, a trend prediction model is constructed. This model can predict the trend of the health index over the next 30 days. If the predicted value will exceed the warning threshold within a specific future timeframe, the system triggers a trend warning in advance. This predictive maintenance mechanism based on deep mining of historical data can further extend the maintenance window of cable branch boxes, completely transforming post-event maintenance into pre-event prevention.

[0041] Example 2: Based on the cable branch box health early warning method based on the fusion of insulation resistance and leakage current described in Example 1, this example provides an application variation under specific high-voltage multi-branch complex operating conditions. In this example, the deployment scheme of the sensing monitoring unit has been optimized.

[0042] In Example 2, when acquiring the raw operating data in step 1, a master-slave distributed acquisition architecture is adopted to address the multi-branch characteristic of the cable branch box. The master station is installed at the main cable inlet, and the slave stations are installed at each outgoing branch. Each slave station aggregates the collected leakage current and resistance data to the master station via a high-speed internal bus. Based on the synchronous clock triggering mechanism, the master station uses a differential detection algorithm to compare the vector sum of the main inlet current signal and the current signals of each branch in real time. If a difference exists between the two, and the change of this difference exhibits chaotic characteristics after the nonlinear preprocessing in step 2, the system will prioritize calculating the dynamic indicators for this difference.

[0043] In the preprocessing stage of step 2, an adaptive notch filter is added to address the strong electromagnetic interference environment. The center frequency of this filter can automatically lock onto the grid frequency and its higher harmonics (such as 50 Hz, 150 Hz, 250 Hz, etc.). By calculating the energy proportion of the signal at these frequency points, the bandwidth and depth of the notch filter are dynamically adjusted. This operation, performed before morphological filtering, can eliminate periodic interference caused by system imbalance or the operation of high-power frequency converters, providing cleaner basic data for the subsequent extraction of weak deterministic signals reflecting molecular chain evolution.

[0044] In step 4, when calculating the correlation dimension, this embodiment employs a multi-scale correlation dimension calculation method. Instead of simply calculating the slope under a single measurement radius, it divides the measurement radius into macroscopic, mesoscopic, and microscopic scales. The macroscopic scale correlation dimension primarily reflects the system's large-scale load fluctuations; the microscopic scale correlation dimension is extremely sensitive to weak discharges within the insulating material. By comparing the rate of change of the correlation dimension at different scales, the physical nature of the degradation can be more accurately located. For example, if the microscopic scale correlation dimension suddenly increases while the macroscopic scale remains stable, it can be generally confirmed as localized internal damage to the insulating medium.

[0045] In the health assessment model described in step 5, this embodiment introduces a self-organizing map neural network (SOM) for cluster analysis. Insulation resistance, maximum Lyapunov exponent, correlation dimension, and feature vectors such as temperature and current amplitude are input into the SOM network for training. The network automatically divides different input patterns into different neuron regions in a competitive layer. The comprehensive health index is calculated by observing the distance of the mapped position of the input feature vector on the neuron map from the normal center point. This method does not require pre-setting strict physical thresholds; instead, it defines health by learning the device's own operating habits, improving the system's compatibility with cables of different brands and service lives.

[0046] In the specific application scenario of Example 2, assume that the operating environment of a cable branch box is extremely humid and experiences severe load fluctuations. When tiny dendritic carbonization channels begin to appear in the insulation material, pulse currents with fractal characteristics will be mixed into the leakage current. In conventional monitoring methods, these weak pulses will be filtered out as noise, or their amplitude will be too small to trigger a resistance alarm. By reconstructing the phase space in step 3 of this example, it can be observed that the originally regular attractor loop trajectory begins to show burrs and divergence in local areas. The maximum Lyapunov exponent calculated in step 4 rapidly changes from a negative value to a positive value (e.g., from -0.12 to +0.05). Although the insulation resistance only decreases from 1000 megohms to 950 megohms at this time (within the normal fluctuation range), the health evaluation model in step 5, due to the extremely high weight given to the dynamic characteristics in the early stage, will immediately cause the comprehensive health index to drop from 95 points to 78 points, triggering an early deterioration warning. This provides maintenance personnel with valuable time for early intervention.

[0047] Example 3: This example further details the implementation method and data interaction details of the cable branch box health early warning method based on the fusion of insulation resistance and leakage current on the cloud big data platform.

[0048] In Example 3, some of the computational tasks in steps 1 to 4 are migrated to the edge computing gateway for execution, while the complex analysis and trend prediction in steps 5 and 6 are completed on the cloud server. The edge computing gateway is responsible for the real-time acquisition of high-frequency data and the preliminary matrix operations for phase space reconstruction. The edge gateway completes one singular value decomposition per second and encapsulates the extracted singular spectral features, the initial value of the maximum Lyapunov exponent, and the initial value of the correlation dimension, along with the minute-average value of the insulation resistance, into a data packet in MQTT protocol format and uploads it to the cloud platform.

[0049] During step 2, to improve adaptability to different batches of cable signals, the cloud server periodically optimizes the structuring element sequence parameters in morphological filtering based on historical data reported from each region. By calculating the functional relationship between the signal-to-noise ratio and the responsivity of the feature extraction operator, a genetic algorithm searches for the optimal combination of structuring element lengths within the textually described parameter space. The optimized parameters are synchronized to the edge gateway via the downlink channel, enabling the dynamic evolution of the preprocessing logic.

[0050] In the health assessment model of step 5, this embodiment employs an ensemble learning strategy. The results based on fuzzy inference, SOM clustering, and physical threshold determination are weighted by voting. If multiple independent models all point to a warning state, the system increases the confidence level of the alarm. This heterogeneous model fusion approach reduces false alarms caused by single sensor failures or sudden environmental changes.

[0051] Following step 6, the system also provides a fault location auxiliary logic. It utilizes the phase difference of leakage current dynamics characteristics acquired by distributed sensing units. When multiple branches simultaneously exhibit chaotic characteristics, the system calculates the order in which the maximum Lyapunov exponent rises in each branch (with microsecond-level accuracy). Based on the signal propagation speed in the cable, and through textual distance conversion logic, the physical distance of the insulation degradation point relative to the branch box is roughly determined. This provides spatial guidance for precise operation and maintenance.

[0052] Example 3 also introduces an interactive operation and maintenance feedback mechanism. After maintenance personnel complete on-site inspections according to the early warning instructions, they need to input the actual insulation status (such as confirming dampness, loose cable joints, or no obvious abnormalities) into the system via a mobile terminal. The system uses feedback tags to perform reinforcement learning training on the health assessment model. If a false alarm occurs, the model will automatically adjust the weight coefficients in step 5 or the interval thresholds in step 6. Through this closed-loop optimization process, the early warning accuracy continuously approaches the theoretical upper limit as the running time increases.

[0053] In the actual operation simulation of this embodiment, the system detected frequent fluctuations in the correlation dimension of a branch box in an industrial area during the high-temperature period in summer. The cloud analysis module, combining historical meteorological data and load data, predicted through a long short-term memory neural network that if this trend continued, the insulation resistance would fall below the safety threshold within three months. Therefore, when the health index was still at 85 points, the system proactively pushed a recommendation to the local power supply station to replace the cable accessories. Subsequent inspection revealed that the cable joint seal did indeed show slight aging, verifying the superior performance of the method of this invention in proactive early warning.

[0054] Example 4: This example focuses on describing a cable branch box health early warning implementation scheme based on the fusion of insulation resistance and leakage current, which is suitable for extreme environments and has hardware redundancy features.

[0055] In step 1, to address potential sensor failures, this embodiment employs a dual-redundancy sampling mode. Two micro-current sensors with identical parameters are installed at each monitoring point. The system compares the outputs of the two sensors in real time and calculates the standard deviation of their difference. If the standard deviation exceeds a preset tolerance range for multiple consecutive sampling periods, the system automatically switches to the backup channel and sends a sensor hardware failure self-check report to the maintenance center, ensuring the continuity and authenticity of the original operating data.

[0056] In step 2, a nonlinear limiter based on median filtering is incorporated into the preprocessing algorithm to address potential electromagnetic pulse attacks or strong lightning interference. Before performing morphological filtering, the signal undergoes a large-step median scan. If the slope of the numerical change exceeds the physical upper limit of normal insulation physical evolution (described textually as its slope threshold), the signal segment is identified as external shock interference and smoothed. This ensures that the deterministic signal input in subsequent chaotic analysis is caused by the evolution of internal microstructures, rather than triggered by external random events.

[0057] In step 3, the construction of the phase space trajectory matrix incorporates a multi-sampling rate fusion technique. This involves preserving both the transient trajectory from high-frequency sampling and the steady-state trajectory from low-frequency sampling. By reconstructing the phase space at different time scales, millisecond-level partial discharge pulses (represented as isolated points far from the attractor core in phase space) and day-level slow degradation trends (represented as gradual distortion of the attractor's overall shape) can be captured. This multi-scale analysis enhances the early warning system's dual monitoring capability against both sudden and cumulative faults.

[0058] In step 5, this embodiment provides a more detailed mathematical and logical description of the weight allocation algorithm. A weight function with respect to the time evolution period is defined. When the system first starts, the weight function is initialized to a balanced distribution. As the system runs, the system calculates the coefficient of variation (i.e., standard deviation divided by arithmetic mean) of the dynamic characteristic indicators in real time. If the coefficient of variation continues to increase, it indicates that the signal volatility is increasing, and the weight function automatically tilts towards the maximum Lyapunov exponent. The magnitude of the weight increase is directly proportional to the growth rate of the coefficient of variation. This adaptive weighting mechanism enables the model to sensitively perceive subtle nodes in the evolution from a linear interval to a nonlinear chaotic state.

[0059] In the output stage of step 6, the early warning command in this embodiment not only includes level information but also a countermeasures library based on an expert system. For the initial deterioration state, the command suggests conducting live infrared thermography; for the warning-concern state, the command suggests conducting partial discharge field testing. Through this closed-loop management across the entire chain, the method of this invention truly becomes the intelligent core for ensuring the reliability of the power system.

[0060] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A cable branch box health early warning method based on the fusion of insulation resistance and leakage current, characterized in that, Includes the following steps: Step 1: The raw operating data is acquired in real time by the sensing and monitoring unit deployed inside the cable branch box. The raw operating data includes at least the insulation resistance time series signal and the leakage current time series signal. The synchronous clock triggering mechanism is used to ensure that the insulation resistance time series signal and the leakage current time series signal are phase-aligned on the time axis. Step 2: Perform nonlinear preprocessing on the original running data. Use a combination algorithm based on multi-scale morphological filtering and adaptive threshold noise reduction to remove random Gaussian noise and periodic narrowband interference mixed in the original signal, and extract deterministic feature signals that reflect the evolution state of molecular chains inside the insulation system. Step 3: Perform phase space reconstruction on the preprocessed leakage current time series signal. By calculating the optimal delay time and minimum embedding dimension, the one-dimensional leakage current time series is mapped to a high-dimensional phase space to construct a phase space trajectory matrix that can characterize the dynamic properties of the insulating medium. Step 4: Calculate the nonlinear dynamic characteristic index based on the reconstructed phase space trajectory matrix. The nonlinear dynamic characteristic index includes the maximum Lyapunov exponent and the correlation dimension, which are used to quantify the degree of deterministic chaos in the evolution of the insulating medium response from the linear interval to the nonlinear chaotic state. Step 5: Construct a health evaluation model based on the fusion of insulation resistance and dynamic characteristic indicators. Use real-time insulation resistance values, maximum Lyapunov exponent, and correlation dimensions as multi-dimensional input variables to calculate the comprehensive health index of the cable branch box. Step 6: Compare the comprehensive health index with the preset health status level threshold in real time to determine the current health warning level and output the corresponding warning instruction.

2. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 1, characterized in that, In step 1, the high-precision insulation monitoring sensor continuously samples each branch cable of the cable branch box, and the sampling frequency is dynamically adjusted according to the preset power grid operating frequency. The synchronous clock triggering mechanism achieves nanosecond-level synchronization between distributed sensing units by receiving BeiDou timing signals or Global Positioning System timing signals, ensuring that paired samples of insulation resistance and leakage current are acquired at the same instant. The insulation resistance time series signal reflects the macroscopic impedance characteristics of the cable insulation medium under voltage, while the leakage current time series signal contains microscopic disturbance information caused by molecular chain breakage or weak partial discharge inside the insulation material. The sensing and monitoring unit includes a microcurrent sensor installed at the cable joint and an insulation resistance detection module on the high-voltage side. The microcurrent sensor adopts an open structure made of high magnetic permeability material and has a microampere-level sensing accuracy. The insulation resistance detection module calculates the real-time insulation resistance by superimposing a low-frequency detection signal onto the system operating voltage and then calculating the leakage current component.

3. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 2, characterized in that, In step 2, performing nonlinear preprocessing on the original running data specifically includes: firstly, using structuring elements to perform multi-scale morphological opening and closing operations on the signal to smooth out pulse-type interference in the signal; The multi-scale morphological filtering process includes: defining a set of structuring element sequences of increasing length, and performing erosion, dilation, opening and closing operations on the signal in sequence; A feature extraction operator is constructed by calculating the difference between the processing results at different scales; the feature extraction operator identifies and enhances the components with deterministic dynamic characteristics in the signal, while suppressing uncorrelated random thermal noise. Subsequently, adaptive wavelet packet decomposition technology is used to divide the signal into multiple frequency bands, and the energy distribution of each frequency band is calculated; for frequency bands that are identified as high-frequency random noise, a soft threshold function is used for shrinkage processing; The threshold of the soft threshold function is determined based on the root mean square error of the signal and a preset scaling factor; the processed signal is reconstructed through inverse transformation to obtain a pure deterministic signal.

4. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 3, characterized in that, In step 3, the phase space reconstruction process is as follows: First, the delay time of the leakage current time series signal is calculated using the autocorrelation function method, and the moment when the autocorrelation function drops to the ratio of the initial value to the natural constant is selected as the optimal delay time. Next, the spurious nearest neighbor algorithm is used to determine the minimum embedding dimension. By observing the change in the proportion of nearest neighbors as the dimension increases, when the proportion is lower than a preset threshold, it is determined as the final embedding dimension. Based on the aforementioned delay time and embedding dimension, the one-dimensional sequence is transformed into a multi-dimensional vector group, forming trajectory points in phase space; The spatial distribution of the trajectory points reflects the attractor characteristics of the operating state of the insulation system. If the system is in a healthy state, the trajectory points exhibit regularity. If early insulation degradation occurs, the trajectory points tend to evolve into nonlinear chaotic attractors; The phase space trajectory matrix consists of multiple column vectors, each representing a state point in the phase space.

5. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 4, characterized in that, The method further includes performing singular value decomposition on the phase space trajectory matrix to extract singular spectral features that reflect the main characteristics of the system; The principal component energy distribution of the singular spectrum is used as an auxiliary criterion to distinguish between current changes caused by load fluctuations and nonlinear abrupt changes caused by insulation degradation. The singular value decomposition decomposes the trajectory matrix into a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix, and extracts the main singular values ​​ranked in the top column of the singular value diagonal matrix to form the singular spectrum. When the load fluctuates normally, the energy distribution of the singular spectrum remains stable and concentrated in the low-frequency components. However, when the insulation undergoes a nonlinear abrupt change, the energy proportion of higher-order singular values ​​increases.

6. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 5, characterized in that, In step 4, the specific process of calculating the maximum Lyapunov exponent is as follows: select a reference trajectory point in the reconstructed phase space, find its neighboring points, and calculate the initial Euclidean distance between the two points. As time evolves, track the trajectories of these two points in phase space and calculate the instantaneous Euclidean distance after the evolution step. Calculate the natural logarithm of the ratio of the instantaneous Euclidean distance to the initial Euclidean distance, and take the arithmetic mean of the logarithmic ratios over multiple evolution cycles; when the arithmetic mean is greater than 0, it indicates that the system has chaotic characteristics. The magnitude of the maximum Lyapunov exponent reflects the rate at which the insulation degradation signal deviates from the linear equilibrium state; a larger value indicates a more pronounced trend of qualitative change in insulation performance.

7. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 6, characterized in that, In step 4, the specific process for calculating the association dimension is as follows: For all trajectory points in the reconstructed phase space, a measurement radius is defined, and the proportion of point pairs in the space whose distance between any two points is less than the measurement radius is counted to obtain the correlation integral function. By continuously changing the value of the measurement radius, a series of corresponding correlation integral values ​​are obtained; in a double logarithmic coordinate system, the linear regression slope between the natural logarithm of the correlation integral function and the natural logarithm of the measurement radius is calculated; The linear regression slope is the correlation dimension; the value of the correlation dimension reflects the spatial occupancy of the insulating medium response signal in the phase space. When microscopic damage or weak discharge occurs inside the insulating material, the complexity of the signal increases, causing fluctuations in the correlation dimension.

8. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 7, characterized in that, In step 5, the construction of the health assessment model involves the normalization of multidimensional features. First, compare the measured insulation resistance value with the preset standard insulation resistance reference value and calculate the percentage deviation. At the same time, the calculated maximum Lyapunov index and the correlation dimension are mapped to a standard range between 0 and 1; Subsequently, a weighted fusion algorithm is used to sum the insulation resistance deviation, the maximum Lyapunov exponent deviation, and the correlation dimension volatility based on the weight coefficients obtained from training historical fault samples, thereby generating a comprehensive health index. The allocation of the weighting coefficients follows a dynamic adjustment strategy, assigning higher weights to the kinetic characteristic indicators in the early stages of degradation and increasing the weight of the insulation resistance in the later stages of degradation.

9. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 8, characterized in that, The multi-dimensional feature fusion process in step 5 employs fuzzy reasoning logic. The rate of change of insulation resistance, the absolute value of the maximum Lyapunov exponent, and the abrupt change intensity of the correlation dimension are used as input variables for the fuzzy set; through a preset membership function, continuous physical quantities are transformed into fuzzy linguistic variables. Based on the fuzzy rule table established by the expert knowledge base, reasoning is performed to obtain a fuzzy comprehensive evaluation vector reflecting health status; finally, through defuzzification calculation, a unique comprehensive health index value is obtained. This processing method enhances the early warning model's tolerance to sensor measurement errors and parameter uncertainties; the defuzzification calculation is performed using the centroid method or the maximum membership method.

10. The cable branch box health early warning method based on the fusion of insulation resistance and leakage current according to claim 9, characterized in that, In step 6, the health status levels are divided into normal operation status, initial deterioration status, early warning and attention status, and serious fault status. The normal operating state corresponds to a comprehensive health index that is in a preset first range; The initial deterioration state corresponds to a comprehensive health index in the preset second range. At this time, the insulation resistance value does not change significantly, but the maximum Lyapunov index shows a definite chaotic trend. The warning and attention status corresponds to a comprehensive health index that is in the preset third interval, at which point the insulation damage reaches an observable level; The severe fault state corresponds to a comprehensive health index exceeding a preset upper limit threshold; the method also includes monitoring the operating environment through temperature and humidity sensors and performing environmental condition compensation. When the ambient humidity exceeds the preset range, the system automatically increases the threshold value of insulation resistance in the health evaluation model, and at the same time increases the dependence on the maximum Lyapunov index. The method also includes establishing a long-term trend monitoring database, using a long short-term memory neural network to learn the historical trajectory of the health index, constructing a trend prediction model, predicting the trend of the health index in the future, and triggering a trend warning in advance if the predicted value will exceed the warning threshold in a specific future time period.

Images