Monitoring control system of inflatable high-voltage ring main unit based on data analysis

Abstract

This invention provides a monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis, belonging to the field of high-voltage electrical equipment monitoring technology. This invention acquires operating conditions and real-time sensor data to construct a degradation index, compares it with thresholds to classify monitoring levels, and sets the current sampling time window. Within the window, it monitors partial discharge signals and calculates pulse density, while simultaneously calculating the gas dielectric constant and propagation velocity to locate the partial discharge coordinates. Based on sensor data and operating conditions, it calculates the internal heating temperature, obtains hotspot coordinates and temperature rise rate through spatial interpolation and differentiation, and calculates key monitoring locations using weighted averages. Finally, it constructs and trains a fault prediction model based on a long short-term memory network, inputs feature sequences to obtain fault evolution trend curves, and generates early warning signals. This invention eliminates multiple temperature rise interferences, achieving a leap from immediate passive alarm to proactive advanced perception.

Description

Technical Field

[0001] This invention relates to the field of high-voltage electrical equipment monitoring technology, specifically to a monitoring and control system for a gas-filled high-voltage ring main unit based on data analysis. Background Technology

[0002] Gas-insulated high-voltage ring main units (RNBs) are core equipment in urban power distribution networks, ensuring power supply reliability and load switching. With the deepening of smart grid construction, predictive maintenance of electrical equipment throughout its entire lifecycle using data analysis and edge computing has become a mainstream trend. However, current RNB monitoring technology still lacks sufficient depth in hardware and software integration, failing to organically combine the complex multi-physics dynamic evolution mechanisms within the equipment with advanced artificial intelligence prediction models. This results in a lack of precise local targeting capabilities and comprehensive lifecycle prediction methods for addressing microscopic insulation degradation and contact defects under complex operating conditions.

[0003] In existing ring main unit (RMU) monitoring technologies, current solutions calculate operational attenuation and maintenance coefficients by acquiring historical commissioning and maintenance data of the RMU. Spatially, they simply divide the internal space into key and non-key sub-areas based on the geometric idle volume ratio of the internal monitoring sub-areas. During the operational monitoring phase, the system collects static environmental variables such as temperature, humidity, air pressure, and dust concentration within each sub-area to output internal performance coefficients; simultaneously, it collects external parameters such as wind speed, light intensity, and vibration to obtain external performance coefficients. Finally, the internal and external performance coefficients are weighted and summed to generate a comprehensive risk warning value, which is then statically compared with a preset fixed ring main unit deviation monitoring threshold to trigger an alarm.

[0004] Although the above technologies achieve a comprehensive consideration of the characteristics of the internal and external environments of the cabinet, from the perspective of the depth of electrical engineering and physical field evolution, the following technical shortcomings still exist: First, the spatial calibration has physical blind spots. It relies purely on the geometric free volume to divide the area, ignoring the objective law that real defects such as partial discharge and contact loosening and heating frequently occur in physical entities with high concentrations of electromagnetic and mechanical stress. This results in the key areas not matching the core domain of the actual electrical fault evolution. Second, the parameter processing lacks decoupling of deep mechanisms. It only relies on the original environmental characterization values ​​and cannot eliminate the cross-temperature rise interference of load heating under normal operating conditions, shell thermal hysteresis effect, and gas environment fluctuations on the extraction of state characteristics. Finally, the early warning mechanism has a time domain discontinuity. The system is essentially an immediate passive alarm based on transient monitoring data. It lacks a technical mechanism to continuously extrapolate the trends of multi-dimensional characteristics on the time axis, and cannot assess the defect deterioration rate and quantitatively estimate the remaining service life of the equipment.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a monitoring and control system for inflatable high-voltage ring main units based on data analysis, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The monitoring and control system for gas-insulated high-voltage ring main units based on data analysis specifically includes: The sampling and evaluation module is used to acquire the operating conditions and real-time sensor data of the gas-insulated high-voltage ring main unit, construct a deterioration index based on the operating conditions, compare the deterioration index with multiple preset deterioration thresholds, classify the monitoring level of the high-voltage ring main unit, and set the current sampling time window based on the monitoring level. The partial discharge location module is used to monitor the partial discharge signal of the ring network cabinet within the current sampling time window, count the number of valid partial discharge signals exceeding a preset threshold, calculate the partial discharge pulse density based on the number of valid partial discharge signals, calculate the dielectric constant of the gas inside the cabinet based on real-time sensor data, and then calculate the propagation speed of the partial discharge signal. It also obtains the arrival time difference sequence of the partial discharge signal to the sensors at different locations, and locates the partial discharge coordinates based on the arrival time difference sequence and the propagation speed. The fault fusion and location module is used to calculate the internal heating temperature of the ring network cabinet based on real-time sensor data and operating conditions, and to perform spatial interpolation of the heating temperature to obtain the hot spot coordinates. It also performs time-domain change rate analysis of the heating temperature to obtain the internal temperature rise rate. The partial discharge pulse density and the internal temperature rise rate are used as weights to perform weighted calculations on the partial discharge coordinates and hot spot coordinates to obtain the key monitoring locations. The fault trend prediction module is used to build a fault prediction model based on the long short-term memory network algorithm. The model is trained based on historical monitoring data, extracts the number of effective partial discharge signals and heating temperature corresponding to key monitoring locations, and constructs a feature sequence by combining real-time sensor data. The feature sequence is input into the trained fault prediction model to obtain the fault evolution trend curve, and an early warning signal is generated based on the fault evolution trend curve.

[0008] Furthermore, the operating conditions include: operating time, load current, number of switching operations, ambient temperature, and ambient humidity; The real-time sensing data includes: cabinet internal air pressure, cabinet internal gas temperature, and surface temperature; The thermal cumulative stress is obtained by integrating the square of the load current over the operating time; the mechanical cumulative stress is calculated based on the number of switching operations; and the environmental cumulative stress is calculated based on the ambient temperature and ambient humidity. The thermal accumulated stress, mechanical accumulated stress, and environmental accumulated stress are multiplied by preset weighting coefficients and then weighted and fused to calculate the deterioration index of the high-voltage ring main unit. The larger the deterioration index, the higher the degree of aging and the greater the operational risk of the high-voltage ring main unit. Based on the range of the degradation index, the monitoring level is divided and the current sampling time window is set: If the degradation index is less than or equal to the preset first degradation threshold, the monitoring level will be determined to be level three, the current sampling time window will be set to the first preset duration, and the first sampling frequency will be matched. If the degradation index is greater than the first degradation threshold and less than or equal to the preset second degradation threshold, the monitoring level will be determined to be level two, the current sampling time window will be set to the second preset duration, and the second sampling frequency will be matched. If the degradation index is greater than the second degradation threshold, the monitoring level will be determined to be Level 1, the current sampling time window will be set to the third preset duration, and the third sampling frequency will be matched. Wherein, the second degradation threshold is greater than the first degradation threshold, the first preset duration is greater than the second preset duration, the second preset duration is greater than the third preset duration, the third sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the first sampling frequency.

[0009] Furthermore, within the current sampling time window, partial discharge signals are acquired according to the matched sampling frequency, background electromagnetic noise is filtered out, and the amplitude of the partial discharge signal is obtained. The amplitude of each partial discharge signal is compared with a preset threshold, and the number of partial discharge signals with amplitudes greater than the preset threshold is counted to obtain the number of valid partial discharge signals. The partial discharge pulse density is obtained by dividing the number of valid partial discharge signals by the duration of the sampling time window. Divide the air pressure inside the cabinet by the temperature of the gas inside the cabinet to obtain the temperature-pressure ratio. Multiply the temperature-pressure ratio by the fixed polarization constant of the gas inside the cabinet to obtain the first product. Add the first product to the value 1 to obtain the dielectric constant. The square root of the dielectric constant is calculated by taking the square root of the dielectric constant. The propagation speed of the partial discharge signal is obtained by dividing the speed of light in vacuum by the square root of the dielectric constant.

[0010] Furthermore, at least four ultra-high frequency sensors are pre-installed on different walls inside the high-voltage ring main unit in a non-coplanar three-dimensional arrangement to obtain the spatial position coordinates of each ultra-high frequency sensor. Using the first sensor in the ultra-high frequency sensor as a reference point, the time difference between the arrival time of the partial discharge signal at each of the other ultra-high frequency sensors and the arrival time at the reference point is obtained, and an arrival time difference sequence is constructed from all the time difference values. Multiply each time difference value in the arrival time difference sequence by the propagation speed to obtain the corresponding spatial distance difference sequence; each element in the sequence represents the difference between the distance from the partial discharge power source to the corresponding other UHF sensor and the distance to the reference point. Based on the spatial position coordinates of each UHF sensor and the spatial distance difference sequence, a three-dimensional hyperboloid positioning equation set is constructed. By solving the positioning equation set, the partial discharge coordinates are obtained.

[0011] Furthermore, the first-order thermal conduction time constant and the second-order thermal capacity time constant of the high-voltage ring main unit shell are obtained; Multiplying the first time derivative of the surface temperature by the first-order heat conduction time constant yields the first thermal correction term; multiplying the second time derivative of the surface temperature by the second-order heat capacity time constant yields the second thermal correction term. Calculate the temperature difference between the surface temperature and the ambient temperature, and multiply the temperature difference by a preset surface heat dissipation coefficient to obtain an environmental heat dissipation compensation term; The load reference compensation term is calculated based on the Joule heating effect for the load current; The surface temperature, the first thermal correction term, the second thermal correction term, and the environmental heat dissipation compensation term are added together to obtain the actual internal temperature value. The heating temperature is obtained by subtracting the load reference compensation term from the actual internal temperature value. The coordinate information and heating temperature corresponding to multiple different measuring point locations are obtained. Using the radial basis function interpolation algorithm, the coordinate information is used as the independent variable and the heating temperature is used as the dependent variable to perform surface fitting and construct a three-dimensional continuous temperature field model of the monitoring area inside the high-voltage ring network cabinet. Traverse the three-dimensional continuous temperature field model and take the spatial maximum point with the largest function value in the three-dimensional continuous temperature field model as the hot spot coordinates; Calculate the change in the heating temperature within the sampling time window, and divide the change in the temperature by the length of the sampling time window to obtain the internal temperature rise rate.

[0012] Furthermore, the partial discharge pulse density, the internal temperature rise rate, the partial discharge coordinates, and the hot spot coordinates corresponding to the same sampling time window are extracted respectively. The first weighted spatial component is calculated by multiplying each dimension component of the partial discharge coordinates in three-dimensional space by the partial discharge pulse density. The second weighted spatial component is calculated by multiplying the hot spot coordinates in each dimension of the three-dimensional space by the internal temperature rise rate. The first weighted spatial component and the second weighted spatial component are added together to obtain the total weighted component; The total weighted component is divided by the sum of the partial discharge pulse density and the internal temperature rise rate to calculate the three-dimensional coordinates of the weighted average center point, and this center point is used as the key monitoring location.

[0013] Furthermore, historical monitoring data of the high-voltage ring main unit is obtained. The historical monitoring data includes: load current, internal air pressure, internal gas temperature, surface temperature, ambient temperature, partial discharge signal, and actual value of historical degradation index recorded at each historical time step. Based on the spatial coordinates of the key monitoring location, the number of valid partial discharge signals at each historical time stamp is selected and counted from the historical partial discharge signals. The historical heating temperature of the key monitoring location at each historical time stamp is calculated using the historical surface temperature, historical ambient temperature, and historical load current at the corresponding timestamp. The number of effective partial discharge signals, the historical heating temperature, and the historical cabinet air pressure, historical ambient temperature, and historical degradation index actual values ​​in the historical monitoring data are aligned with a unified timestamp to construct the historical feature sequence. The historical feature sequences are used as training samples and normalized. Supervised learning samples are constructed according to a preset time window length and sliding step size. Continuous... The feature vectors at each time step are the input sequence, and the subsequent... The actual value of the degradation index at each time step is the output sequence, where and All are positive integers; The input sequence is used as input to a long short-term memory network model, wherein the long short-term memory network model includes an LSTM layer and a fully connected output layer. The LSTM layer is used to extract the time-dependent features of the feature sequence, and the fully connected output layer is used to output the predicted value of the degradation index for multiple future time steps. Using the output sequence as a supervision label, the model parameters are iteratively updated through the backpropagation algorithm until the loss function converges, thus obtaining a trained fault prediction model.

[0014] Furthermore, the number of valid partial discharge signals and the heating temperature corresponding to the key monitoring location within the current sampling time window are extracted, and combined with the cabinet air pressure and ambient temperature in the real-time sensor data, a feature sequence of the key monitoring location is constructed. The feature sequence is input into the trained fault prediction model to obtain the degradation prediction values ​​for multiple future time steps from the current moment. The predicted degradation index values ​​for multiple future time steps are sequentially fitted onto the time axis to generate a fault evolution trend curve.

[0015] Furthermore, the pre-configured absolute failure threshold of the target high-voltage ring main unit is obtained; Calculate the first intersection point between the fault evolution trend curve and the absolute failure threshold of the equipment, and record the time span from the current moment to the intersection point as the remaining safe operating time; Obtain a preset first time threshold and a second time threshold, wherein the first time threshold is greater than the second time threshold; If the remaining safe operating time is greater than the first time threshold, a level three warning signal is generated; If the remaining safe operating time is less than or equal to the first time threshold and greater than the second time threshold, a level two warning signal is generated. If the remaining safe operating time is less than or equal to the second time threshold, a level one warning signal is generated.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a degradation index based on acquired operating conditions and compares the degradation index with multiple preset degradation thresholds to classify the monitoring level of the high-voltage ring main unit and set the current sampling time window, overcoming the inaccuracy of analysis caused by the uniform standard used in traditional monitoring. Within the current sampling time window, the partial discharge signal of the ring main unit is monitored and the number of effective partial discharge signals exceeding the preset threshold is counted. Based on this, the partial discharge pulse density is calculated. At the same time, the dielectric constant of the gas inside the unit is calculated using real-time sensor data to calculate the propagation speed of the partial discharge signal. Combined with the arrival time difference sequence, the coordinates of the partial discharge are accurately located. Furthermore, the internal heating temperature of the ring main unit is calculated based on real-time sensor data and operating conditions. Spatial interpolation and differentiation of the heating temperature are performed to obtain the hot spot coordinates and the internal temperature rise rate, respectively. Finally, the partial discharge coordinates and hot spot coordinates are weighted by the partial discharge pulse density and the internal temperature rise rate to obtain the key monitoring location. This positioning mode breaks through the physical blind spots caused by the existing technology that relies purely on geometric idle volume to divide the area, and achieves targeted and accurate locking of the main domain of local core fault evolution. This invention also constructs a fault prediction model based on a long short-term memory network algorithm, trains the model on historical monitoring data, extracts the number of effective partial discharge signals and heating temperature corresponding to key monitoring locations, and constructs a feature sequence by combining real-time sensor data. The feature sequence is then input into the trained fault prediction model to obtain a fault evolution trend curve, and finally generates an early warning signal based on the fault evolution trend curve. This technical architecture effectively connects the time domain discontinuity caused by the fact that existing technologies can only use current transient data for immediate threshold exceedance judgment, enabling operation protection to follow the degradation law of the fault evolution trend curve to perform deep time-series inference, realizing a leap from immediate passive alarm to forward-looking evolution inference, and improving the initiative and advanced perception accuracy of high-voltage power distribution equipment in preventing hidden dangers under complex operating conditions. Attached Figure Description

[0017] Fig. 1 This is a schematic diagram of the overall system modules of the present invention; Fig. 2 This is a diagram showing the temperature field distribution of a key section inside a high-voltage ring main unit. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figs. 1-2 The present invention provides a technical solution: The monitoring and control system for gas-insulated high-voltage ring main units based on data analysis specifically includes: The sampling and evaluation module is used to acquire the operating conditions and real-time sensor data of the gas-insulated high-voltage ring main unit. Based on the operating conditions, a degradation index is constructed, and the degradation index is compared with multiple preset degradation thresholds to classify the monitoring level of the high-voltage ring main unit. Based on the monitoring level, the current sampling time window is set.

[0021] In this embodiment, the operating conditions include: operating time, load current, number of switching operations, ambient temperature, and ambient humidity; The real-time sensing data includes: cabinet internal air pressure, cabinet internal gas temperature, and surface temperature; The sampling and evaluation module establishes a communication cascade with the high-voltage ring main unit's power distribution automation terminal and internal sensor array to obtain the operating conditions and real-time sensor data of the high-voltage ring main unit in real time.

[0022] The operating conditions include the operating time, load current, and number of switching operations recorded by the terminal's built-in clock, current transformer, and auxiliary contact counter, respectively, as well as the ambient temperature and humidity collected by the integrated environmental sensors outside the cabinet. The real-time sensing data includes the cabinet's internal air pressure, cabinet gas temperature, and the surface temperature of the busbar and contact parts collected by the sulfur hexafluoride gas density relay and the contact surface fiber optic thermometer extending into the sealed gas-filled compartment.

[0023] To accommodate the varying degradation rates of different physical fields at the data conversion level and to avoid ambiguity in terminology, the operating time in the aforementioned operating conditions is stored in the software's underlying database as a continuously updated global historical timeline starting from the initial commissioning of the high-voltage ring main unit. When the algorithm enters a specific stress accumulation calculation branch, the monitoring software invokes a heterogeneous time-domain sliding window mechanism to extract historical operating time periods of different spans from the global historical timeline, serving as the time boundaries for specific integration and accumulation calculations.

[0024] The thermal cumulative stress is obtained by integrating the square of the load current over the operating time; the mechanical cumulative stress is calculated based on the number of switching operations; and the environmental cumulative stress is calculated based on the ambient temperature and ambient humidity. During long-term operation under heavy loads, the heat generation of the main circuit conductive contacts and busbar lap surfaces in high-voltage ring main units primarily follows the Joule heating effect, with the heating power proportional to the square of the load current. The continuous accumulation of heat within the sealed gas-filled casing is the root cause of insulation gas degradation and deterioration of contact surface resistance. Because the load current of the main circuit fluctuates dramatically between day and night, to accurately capture the high dynamic thermal fatigue experienced by local contact surfaces during peak electricity consumption periods, the system must extract a short time-domain window that fully covers the daily load cycle for precise integration.

[0025] The formula for calculating cumulative thermal stress is: In the formula, For thermally accumulated stress, For the first Load current at each sampling time, The sampling time interval, The dimensionless factor of the preset nominal reference is set to . .

[0026] The preset rated reference dimensionless factor is derived from the envelope fitting of the ultimate thermal stability capacity and long-term temperature rise saturation limit of the conductive copper busbar of the main circuit of this type of high-voltage ring main unit. Through factory temperature rise type test calibration, when the main circuit operates under overload conditions and the accumulated energy reaches approximately... At that time, the temperature of its contact surface will approach the critical temperature that destroys the insulating gas.

[0027] The connecting rods, springs, and contact spring clips of the circuit breaker and load switch operating mechanisms within the ring main unit have defined mechanical fatigue life limits. Each opening and closing operation causes mechanical shock and friction, resulting in irreversible mechanical deformation and fatigue damage to the operating mechanism. Multiplying the number of switching operations by a preset mechanical wear proportionality factor yields the cumulative mechanical stress. The preset mechanical wear proportionality factor is set to 0.25. According to the linear cumulative damage theory, the damage percentage caused by each opening and closing operation to the mechanism is 0.0025. Multiplying this value by 100 yields the numerical constant 0.25.

[0028] High temperature and humidity in the external environment have a strong chemical synergistic degradation effect, which is the core external factor causing corrosion of the ring main unit's outer shell, electrochemical corrosion of its structure, and creepage marks on the composite insulation surface. Since the irreversible electrochemical damage to high-voltage insulation components caused by external temperature and humidity is an extremely slow and gradual degradation process, if the sampling time span is too short, the calculated increment of environmental cumulative stress will approach zero infinitely after normalization, leading to feature annihilation. Therefore, the algorithm must adaptively select a longer macroscopic time scale to reflect the cumulative effect.

[0029] From the global historical timeline, the most recent second historical operating period is extracted, specifically configured to be 10 days. The average ambient temperature and humidity within this period are extracted. The temperature and humidity values ​​are multiplied by the corresponding standardized number of days, and then multiplied by a preset environmental stress sensitivity coefficient to obtain the dimensionless cumulative environmental stress. The preset environmental stress sensitivity coefficient is set to 0.125. This preset environmental stress sensitivity coefficient relies on pre-conducted accelerated aging tests in an artificial climate chamber under alternating damp heat conditions in the laboratory.

[0030] The thermal accumulated stress, mechanical accumulated stress, and environmental accumulated stress are multiplied by preset weighting coefficients and then weighted and fused to calculate the deterioration index of the high-voltage ring main unit. The larger the deterioration index, the higher the degree of aging and the greater the operational risk of the high-voltage ring main unit. The Joule heating effect generated by the long-term high-load current in the high-voltage main circuit directly accelerates contact oxidation and insulation deterioration, which is the primary cause of catastrophic accidents. Although mechanical switch operation causes wear, switching is infrequent, and while external temperature and humidity accelerate aging, this is a slow-moving background process. Equal weighting of these factors would result in a sluggish response to overheating of internal contact surfaces. Therefore, the absolute dominance of electrothermal stress must be emphasized.

[0031] In this embodiment, the preset thermal cumulative stress weighting coefficient in the storage medium is specifically limited to: =0.50, the mechanical cumulative stress weighting coefficient is specifically limited to 0.50. =0.30, the environmental cumulative stress weighting coefficient is specifically limited to 0.30. =0.20, which satisfies the constraint that the sum of all weight coefficients equals a constant 1.

[0032] Based on the range of the degradation index, the monitoring level is divided and the current sampling time window is set: If the degradation index is less than or equal to the preset first degradation threshold, the monitoring level will be determined to be level three, the current sampling time window will be set to the first preset duration, and the first sampling frequency will be matched. If the degradation index is greater than the first degradation threshold and less than or equal to the preset second degradation threshold, the monitoring level will be determined to be level two, the current sampling time window will be set to the second preset duration, and the second sampling frequency will be matched. If the degradation index is greater than the second degradation threshold, the monitoring level will be determined to be Level 1, the current sampling time window will be set to the third preset duration, and the third sampling frequency will be matched. Wherein, the second degradation threshold is greater than the first degradation threshold, the first preset duration is greater than the second preset duration, the second preset duration is greater than the third preset duration, the third sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the first sampling frequency.

[0033] The specific hierarchical logic is as follows: Level 3 monitoring branch: If the degradation index is less than or equal to 40.0, the monitoring level is determined to be Level 3. The current sampling time window is set to the first preset duration, specifically 12 hours, and matched with the first sampling frequency, specifically set to... ; Level 2 monitoring branch: If the degradation index is greater than 40.0 and less than or equal to 75.0, the monitoring level is determined to be Level 2. The current sampling time window is set to the second preset duration, specifically 6 hours, and matched with the second sampling frequency, specifically set to... ; Level 1 Monitoring Branch: If the degradation index is greater than 75.0, the monitoring level is determined to be Level 1. The current sampling time window is set to the third preset duration, specifically 2 hours, and matched with the third sampling frequency, specifically set to... .

[0034] The partial discharge location module is used to monitor the partial discharge signal of the ring network cabinet within the current sampling time window, count the number of valid partial discharge signals exceeding a preset threshold, calculate the partial discharge pulse density based on the number of valid partial discharge signals, calculate the dielectric constant of the gas inside the cabinet based on real-time sensor data, and then calculate the propagation speed of the partial discharge signal. It also obtains the arrival time difference sequence of the partial discharge signal to different position sensors, and locates the partial discharge coordinates based on the arrival time difference sequence and the propagation speed.

[0035] In this embodiment, within the current sampling time window, the partial discharge signal is acquired according to the matched sampling frequency, background electromagnetic noise is filtered out, and the amplitude of the partial discharge signal is obtained. The raw electromagnetic wave signal captured by the ultra-high frequency sensor is input into the digital noise reduction algorithm. By performing high-order digital bandpass filtering, the signal frequency band is strictly limited to [specific frequency range]. to Within the ultra-high frequency range, it can completely filter out the low-frequency power grid harmonics and high-frequency high-energy white noise outside this range, restoring a pure partial discharge transient pulse waveform. Subsequently, it iterates through all pure pulses within the current entire sampling time window, extracts and locks the maximum peak absolute voltage value of each pulse waveform as the amplitude of each partial discharge signal.

[0036] The amplitude of each partial discharge signal is compared with a preset threshold, and the number of partial discharge signals with amplitudes greater than the preset threshold is counted to obtain the number of valid partial discharge signals. The specific value of the preset threshold is limited to be displayed. The monitoring software compares the amplitude of each partial discharge signal extracted in the preceding steps with a threshold value. If the amplitude exceeds the threshold, the pulse is automatically identified as a substantial insulation fault discharge event, and the counter is incremented by 1. If the threshold is not met, the signal is discarded. After traversing all signals within the current sampling time window, the final accumulated count output by the counter represents the number of valid partial discharge signals.

[0037] The partial discharge pulse density is obtained by dividing the number of valid partial discharge signals by the duration of the sampling time window. Extract the duration of the sampling time window assigned by the monitoring level, divide the number of valid signals by the duration, and calculate the dimensionless partial discharge pulse density. The severity of insulation degradation inside the high-voltage switchgear is directly correlated with the intensity of discharge activity per unit time.

[0038] Divide the air pressure inside the cabinet by the temperature of the gas inside the cabinet to obtain the temperature-pressure ratio. Multiply the temperature-pressure ratio by the fixed polarization constant of the gas inside the cabinet to obtain the first product. Add the first product to the value 1 to obtain the dielectric constant. The current internal gas pressure and temperature are extracted in real time using a gas density relay and fiber optic thermometer. The fixed polarization constant of the insulating gas, pre-calibrated through factory gas type testing for the specific frequency band, is retrieved. The specific formula for calculating the dielectric constant is as follows: In the formula, Where is the dielectric constant. The fixed polarization constant of the gas inside the cabinet. The air pressure inside the cabinet. This refers to the temperature of the gas inside the cabinet.

[0039] The square root of the dielectric constant is calculated by taking the square root of the dielectric constant. The propagation speed of the partial discharge signal is obtained by dividing the speed of light in vacuum by the square root of the dielectric constant.

[0040] According to Maxwell's fundamental laws of electromagnetism, the actual propagation speed of high-frequency electromagnetic waves in a non-vacuum medium is inversely proportional to the square root of the relative permittivity of that medium. The ultra-high frequency signals generated by partial discharge are essentially high-frequency, high-energy electromagnetic waves. Since density fluctuations caused by temperature and pressure have been seamlessly compensated for in the permittivity, this step, through rigorously applying the physical formula for medium wave velocity, can produce an absolute wave velocity that perfectly matches the current transient physical environment inside the high-voltage cabinet, thus eliminating the distortion of the positioning space's geometric scale caused by drastic environmental changes at its source.

[0041] In this embodiment, at least four ultra-high frequency sensors are pre-set on different walls inside the high-voltage ring main unit in a non-coplanar three-dimensional arrangement to obtain the spatial position coordinates of each ultra-high frequency sensor. In this embodiment, the physical dimensions of the gas-filled compartment inside the high-voltage switchgear are specifically limited to a closed cubic metal shell with a length of 1.0 meter, a width of 1.0 meter, and a height of 1.0 meter. Four ultra-high frequency sensors, labeled as the first sensor, the second sensor, the third sensor, and the fourth sensor, are fixedly arranged non-coplanarly on the inner wall of the shell. Their pre-measured and locked absolute spatial coordinates stored in the configuration register are specifically defined as follows: the coordinates of the first sensor are [0.1, 0.2, 0.3] meters, the coordinates of the second sensor are [0.9, 0.2, 0.4] meters, the coordinates of the third sensor are [0.5, 0.8, 0.2] meters, and the coordinates of the fourth sensor are [0.4, 0.3, 0.9] meters.

[0042] Using the first sensor in the ultra-high frequency sensor as a reference point, the time difference between the arrival time of the partial discharge signal at each of the other ultra-high frequency sensors and the arrival time at the reference point is obtained, and an arrival time difference sequence is constructed from all the time difference values. The transient change in partial discharge electromagnetic waves occurs completely randomly and unpredictably on the time axis, making it impossible to establish an absolute time-synchronized triggering chain between the internal defect point of the high-voltage cabinet and the external measuring point. By setting the first sensor as the global reference point and subtracting this reference point from the pulse leading edge times recorded by the other sensors, the absolute occurrence time—an unknown that cannot be directly measured—can be perfectly eliminated in the control logic using relative subtraction. This transforms the absolute timing location problem into a relative arrival time difference calculation problem, reducing the stringent requirements of the system for absolute hardware synchronization triggering. The matrix mapping relationship of the arrival time difference sequence is as follows: Multiply each time difference value in the arrival time difference sequence by the propagation speed to obtain the corresponding spatial distance difference sequence; each element in the sequence represents the difference between the distance from the partial discharge power source to the corresponding other UHF sensor and the distance to the reference point. The propagation speed is reconstructed online based on the actual temperature and pressure of the gas inside the cabinet, thus ensuring that the distance conversion relationship can adapt to the density evolution of the current fluid field and thermal field inside the high-voltage cabinet, and completely eliminate the geometric scaling distortion caused by drastic changes in the cabinet environment.

[0043] Based on the spatial position coordinates of each UHF sensor and the spatial distance difference sequence, a three-dimensional hyperboloid positioning equation set is constructed. By solving the positioning equation set, the partial discharge coordinates are obtained.

[0044] Let the coordinates of the partial discharge to be solved be... Calculate the geometric sum of squares and constants of each sensor relative to the origin. .

[0045] Introducing auxiliary unknown independent variables , representing the absolute geometric distance from the discharge source to the first reference sensor, expands and linearizes the nonlinear hyperboloid equations into a function of the first-stage unknown observation vector. One-stage matrix equations: In the formula, the first-stage coefficient matrix and the first-stage constant term vector Strictly defined by the known sensor coordinates and distance difference elements respectively: By applying the first-stage weighted least squares formula, the intermediate solution vector can be directly calculated. : in, The initial time-domain noise covariance weight matrix is ​​determined by the standard deviation of the internal clock jitter of the UHF acquisition chip.

[0046] To correct the issue of auxiliary unknown independent variables in the first stage of solution... The software addresses the large constraint of the sum of squares of Euclidean geometry in space, which is ignored due to the separate processing of coordinate components. It then incorporates a second-stage error correction mechanism to construct a quadratic corrected solution vector. Two-stage linear matrix equations: In the formula, the two-stage coefficient matrix and the two-stage constant term vector The output results of each stage are constructed as follows: By applying the second-stage weighted least squares formula, redundant noise caused by forced linearization is completely eliminated, and a quadratic square solution vector that perfectly fits the spatial causality law is calculated. : Perform sign determination and square root extraction on the squared term of the second-stage output to restore the partial discharge coordinates. The three components: Since taking the square root inevitably produces paired bidirectional solutions in both positive and negative directions, the monitoring software, in its final output, substitutes the two sets of coordinates into the previously defined boundary conditions for the geometric dimensions of the ring main unit's gas box, thus satisfying the constraints: Branch comparison is performed. False, physically meaningless solutions located outside the metal cabinet are discarded, and only the unique positive solution located entirely within the enclosed region inside the cabinet is retained and output as the final partial discharge coordinates. This high-precision spatial parameter is then written into a memory register for subsequent multi-field energy centroid weighted fusion.

[0047] The fault fusion and location module is used to calculate the internal heating temperature of the ring main unit based on real-time sensor data and operating conditions, perform spatial interpolation on the heating temperature to obtain hot spot coordinates, perform time-domain rate of change analysis on the heating temperature to obtain the internal temperature rise rate, and perform weighted calculation on the partial discharge coordinates and hot spot coordinates with partial discharge pulse density and internal temperature rise rate as weights to obtain the key monitoring location.

[0048] In this embodiment, the first-order thermal conduction time constant and the second-order thermal capacity time constant of the high-voltage ring main unit shell are obtained; In this embodiment, the specific value of the first-order heat conduction time constant is limited to 120.0 s, and the specific value of the second-order heat capacity time constant is limited to... The specific value of the surface heat dissipation coefficient is limited to 0.15.

[0049] In the pre-shipment temperature rise type test, a rated power heat source was applied to the main circuit. The test results showed that from the moment the internal core temperature experienced a step change, the transient temperature rise at the measuring point on the outer shell surface reached 63.2% of the steady-state saturation temperature rise amplitude in exactly 120.0 seconds. Therefore, the first-order heat conduction time constant was strictly locked and preset to 120.0 seconds at the underlying level of the monitoring software.

[0050] Through dynamic pulse temperature rise type testing, when the main circuit current experiences intermittent heavy load abrupt changes, the dynamic thermal relaxation waveform on the shell surface exhibits obvious nonlinear second-order overdamped characteristics. Through systematic identification of the thermal balance equation using an algorithm, the second-order comprehensive thermal resistance-capacitance product factor under this temperature and pressure structure was precisely aligned to 45.0. .

[0051] In a standard indoor environment without wind convection, through a thermal balance calibration test, the relative heat dissipation parameter in the software temperature compensation chain when the temperature difference between the outer shell and the environment dissipates energy is a fixed proportional constant of 0.15.

[0052] Multiplying the first time derivative of the surface temperature by the first-order heat conduction time constant yields the first thermal correction term; multiplying the second time derivative of the surface temperature by the second-order heat capacity time constant yields the second thermal correction term. Calculate the temperature difference between the surface temperature and the ambient temperature, and multiply the temperature difference by a preset surface heat dissipation coefficient to obtain an environmental heat dissipation compensation term; The load reference compensation term is calculated based on the Joule heating effect for the load current; The surface temperature, the first thermal correction term, the second thermal correction term, and the environmental heat dissipation compensation term are added together to obtain the actual internal temperature value. In actual operation scenarios of ring main units (RNBs), fiber optic thermometers or infrared sensors are often only placed on the outer surface of the RNB or on the outer wall of the gas-filled compartment. The actual sources of heat, such as deteriorated contact resistance, are located in the core conductive components inside the sealed compartment. Due to the significant thermal resistance and capacity of the thick metal casing, the conduction of high-temperature heat from the interior to the casing surface suffers from severe nonlinear time-domain lag and amplitude attenuation. Simultaneously, the heat dissipation rate of the casing to the outside air is highly dependent on the transient temperature difference between the casing surface and the external environment. This heat loss into the air causes a nonlinear drop in the temperature rise measured by the sensor. If the surface temperature measured by the sensor is directly mistaken for the internal temperature, serious hidden dangers will be missed and the temperature will be underestimated.

[0053] The formula for calculating the actual internal temperature is: In the formula, This represents the actual internal temperature value. Surface temperature, For ambient temperature, The first-order thermal conduction time constant, The second-order heat capacity time constant is... The surface heat dissipation coefficient, It is a time variable.

[0054] The heating temperature is obtained by subtracting the load reference compensation term from the actual internal temperature value. Even under perfectly healthy operating conditions, high-voltage ring main units will inevitably experience a normal baseline temperature rise due to the inherent rated conductor resistance of the main circuit busbar and switch contacts when a large load current passes through them, according to the Joule heating effect. This invention must quantitatively calculate the healthy, normal heating baseline under the current load online, form a load baseline compensation term, and completely subtract it using physical subtraction, thereby purifying and extracting the heating temperature purely caused by abnormal deterioration of contact resistance or electrical insulation degradation.

[0055] The coordinate information and heating temperature corresponding to multiple different measuring point locations are obtained. Using the radial basis function interpolation algorithm, the coordinate information is used as the independent variable and the heating temperature is used as the dependent variable to perform surface fitting and construct a three-dimensional continuous temperature field model of the monitoring area inside the high-voltage ring network cabinet. Obtain the known spatial coordinates of multiple measuring points with different geometric positions arranged inside the enclosed high-voltage ring main unit. and the corresponding heating temperature, among which, , The total number of physical temperature measuring points is explicitly limited to at least four in this embodiment. The heating temperature is the pure abnormal temperature rise component obtained by inverse full compensation using the transient heat conduction kinetic equation of the shell and by completely deducting the Joule heating background of the normal operation high current.

[0056] Using three-dimensional absolute coordinates Using the heating temperature at each known measuring point as the dependent variable and taking it as the independent variable, a spatial three-dimensional surface is fitted. The interpolation basis characteristic coefficients are obtained by solving the equations of the known measuring points simultaneously. A three-dimensional continuous temperature field model of the monitoring area inside the high-voltage ring main unit was constructed. The equations of the three-dimensional continuous temperature field model are as follows: In the formula, It is a three-dimensional continuous temperature field model. For three-dimensional absolute coordinates, For the first Coordinate information of each physical temperature sensor, For smooth shape control constants, These are the interpolation basis characteristic coefficients. for arrive Spatial Euclidean geometric distance, This represents the total number of physical temperature sensors. This is the index for the physical temperature sensor.

[0057] Traverse the three-dimensional continuous temperature field model and take the spatial maximum point with the largest function value in the three-dimensional continuous temperature field model as the hot spot coordinates; Calculate the change in the heating temperature within the sampling time window, and divide the change in the temperature by the length of the sampling time window to obtain the internal temperature rise rate.

[0058] By iterating through the fitted continuous temperature field model, the global maximum point of the mathematical function is found. The three-dimensional geometric coordinates of this point represent the true, high-temperature lesion center of the internal fault, ignoring the blind spots of physical sensor placement. Simultaneously, the change in heating temperature at this target location within the current sampling time window is calculated and aligned to the time reference axis. This quantitatively reflects the transient acceleration of thermal runaway within the device, providing the purest and most interference-resistant thermal control parameters for the subsequent fault fusion and localization module to perform cross-dimensional electrothermal multi-field equivalent energy centroid cascade.

[0059] In this embodiment, the partial discharge pulse density, the internal temperature rise rate, the partial discharge coordinates, and the hot spot coordinates corresponding to the same sampling time window are extracted respectively. The latent hazards inside gas-insulated high-voltage ring main units caused by insulation breakdown due to dielectric degradation or deterioration of contact resistance manifest in the physical world as a strong synergistic evolution and interactive coupling of multiple physical fields, including high-frequency partial discharge pulse fields and continuous temperature conduction fields. Due to the discrete uncertainties in the physical spatial distribution of different sensor networks, it is necessary to forcibly lock onto the same transient monitoring time window slice assigned by the sampling and evaluation module, aligning the discrete pulse intensity in the time domain with the continuous spatial fitted temperature rise rate on the same spatiotemporal boundary. This eliminates the phase misalignment caused by the time delays of responses from different physical fields, providing a consistent heterogeneous input tensor for subsequently constructing the global equivalent energy damage centroid.

[0060] The first weighted spatial component is calculated by multiplying each dimension component of the partial discharge coordinates in three-dimensional space by the partial discharge pulse density. The partial discharge coordinates calculated analytically from the UHF arrival time difference are essentially just pure geometric intersection coordinates. Their static orientation cannot quantitatively reflect the current discharge intensity and destructive severity of the insulation defect. By introducing partial discharge pulse density as an electrical dynamic weighting coefficient, and performing a cross-dimensional product of the spatial geometric scale and discharge activity intensity, the subsequently fused centroid point can adaptively tilt in three-dimensional space towards the physical orientation of the entity with the highest discharge incidence and the greatest risk of electrical insulation breakdown. This achieves the first electrical morphological stretching of the spatial electric field coordinates by physical causality.

[0061] The second weighted spatial component is calculated by multiplying the hot spot coordinates in each dimension of the three-dimensional space by the internal temperature rise rate. The hotspot coordinates, optimized using the radial basis function space continuous interpolation function, belong to the maxima of the local temperature field in space and possess static properties, failing to reflect the rate of uncontrolled temperature rise development in the internal core conductor. This step introduces the internal temperature rise rate as a thermal dynamic weighting coefficient. By performing a product operation on the temperature rise acceleration of the hotspot spatial coordinates, the spatial trajectory can be forced to adaptively tend towards the fault core source with drastic heat accumulation and a severe nonlinear overheating deterioration trend, realizing a second thermal morphological stretching of the spatial thermal field coordinates by physical causality.

[0062] The first weighted spatial component and the second weighted spatial component are added together to obtain the total weighted component; The total weighted component is divided by the sum of the partial discharge pulse density and the internal temperature rise rate to calculate the three-dimensional coordinates of the weighted average center point, and this center point is used as the key monitoring location.

[0063] Essentially, a spatial equivalent energy damage centroid, dominated by both electrical damage intensity and thermal runaway velocity, is virtually reconstructed within the high-voltage switchgear. This centroid, as a unique three-dimensional spatial trajectory point, highly condenses and summarizes the severity of the multi-field integrated fault threat under the current spatiotemporal cross-section. This allows the subsequent fault trend prediction module to focus solely on the single-time-series lifetime extrapolation of the movement trajectory of this physically purified key monitoring location, rather than focusing on the complex continuous multi-dimensional spatial field across the entire domain. This achieves significant control dimensionality reduction and greatly improves the inference response speed and model convergence stability of the edge computing chip.

[0064] The fault trend prediction module is used to build a fault prediction model based on the long short-term memory network algorithm. The model is trained based on historical monitoring data, extracts the number of effective partial discharge signals and heating temperature corresponding to key monitoring locations, and constructs a feature sequence by combining real-time sensor data. The feature sequence is input into the trained fault prediction model to obtain the fault evolution trend curve, and an early warning signal is generated based on the fault evolution trend curve.

[0065] In this embodiment, historical monitoring data of the high-voltage ring network cabinet is obtained. The historical monitoring data includes: load current, cabinet gas pressure, cabinet gas temperature, surface temperature, ambient temperature, partial discharge signal, and actual value of historical degradation index recorded at each historical time step. Based on the spatial coordinates of the key monitoring location, the number of valid partial discharge signals at the corresponding location at each historical time stamp is selected and counted from the historical partial discharge signals. High-voltage ring main units, as large, multi-component electrical physical entities, often generate stray discharge pulses or background noise from different physical locations and different phase conductive circuits during their long-term operation. Blindly inputting historical partial discharge signals from the entire domain into the model will introduce interference noise, leading to severe overfitting or prediction divergence. This invention extracts the spatial coordinates of currently identified key monitoring locations and introduces a spatial tolerance threshold. This allows for the precise screening and statistical analysis of waveform sequences truly originating from the geometrical location of the centroid of the specific hidden danger's evolution from historical partial discharge signals, thus improving the data purity of the training samples.

[0066] The historical heating temperature of the key monitoring location at each historical time stamp is calculated using the historical surface temperature, historical ambient temperature, and historical load current at the corresponding timestamp. The number of effective partial discharge signals, the historical heating temperature, and the historical cabinet air pressure, historical ambient temperature, and historical degradation index actual values ​​in the historical monitoring data are aligned with a unified timestamp to construct the historical feature sequence. The surface temperatures recorded in the historical database exhibit a highly nonlinear and non-stationary pattern due to the thermal hysteresis effect caused by the thermal resistance and capacity of the thick metal casing, as well as the interference of natural heat dissipation caused by variable weather conditions. Simultaneously, normal load current flowing through the main circuit also generates a normal Joule heating temperature rise. If the raw surface temperature is directly used as the input for time-series prediction, the deep learning network will be unable to distinguish between environmental changes, load changes, and real hidden dangers caused by deteriorating contact resistance. This step, by strongly intervening in the reverse inference mechanism of the thermal network again during the historical data processing stage, completely removes the outer casing thermal resistance, environmental heat dissipation compensation, and normal load heat, thereby extracting the historical heating temperature that represents the true deterioration trajectory of the core entity and is entirely caused by local hidden dangers, providing the most crucial thermal input for the model.

[0067] In actual power distribution network operation, hardware devices such as barometric pressure sensors, temperature optical fibers, and UHF partial discharge acquisition cards have their own independent internal crystal oscillator clocks or heterogeneous non-uniformly spaced sampling frequencies. This asynchronous clocking and multi-rate sampling leads to severe data gaps and time axis misalignment in multidimensional matrices. Long Short-Term Memory (LSTM) networks require that at each discrete time step, all dimensions of the input vector must achieve absolute concurrent alignment on the time axis. By establishing a unified timestamp grid and using interpolation or equal-interval resampling mechanisms to perform forced time alignment, the scattered electrical, thermal, and gaseous multi-field features can be compressed and reconstructed into a concurrently evolving composite multidimensional state matrix, thus providing a compliant matrix input for deep learning to fully reproduce the history of the coordinated deterioration of multiple physics fields.

[0068] The historical feature sequences are used as training samples and normalized. Supervised learning samples are constructed according to a preset time window length and sliding step size. Continuous... The feature vectors at each time step are the input sequence, and the subsequent... The actual value of the degradation index at each time step is the output sequence, where and All are positive integers; Using the historical feature sequences as training samples, according to the time window length The value is 24, representing 24 consecutive historical time steps, and the sliding step size S is 1 to construct supervised learning samples, with future time spans. The prediction step size for the output sequence is set to 6. During the model compilation phase, mean squared error is used as the loss function, backpropagation is performed, and the initial learning rate is set to 0.001. To prevent overfitting, a dropout layer with a deactivation rate of 0.2 is introduced between the LSTM layer and the fully connected output layer.

[0069] The input sequence is used as input to a long short-term memory network model, wherein the long short-term memory network model includes an LSTM layer and a fully connected output layer. The LSTM layer is used to extract the time-dependent features of the feature sequence, and the fully connected output layer is used to output the predicted value of the degradation index for multiple future time steps. The solid insulation degradation and charge accumulation inside high-voltage gas-filled switchgear exhibit long-term physical memory and temporal nonlinear dependence. Traditional recurrent neural networks (RNNs) suffer from devastating gradient vanishing or exploding defects during backpropagation due to matrix multiplication when processing long sequences. This invention introduces an LSTM layer that, through its internal dynamic logic gating mechanisms (input gate, forget gate, and output gate) and long-term cellular state bus, can spontaneously and selectively forget random perturbations and noise in the historical data while strongly preserving and remembering long-distance electrothermal co-degradation trends spanning several hours, perfectly extracting the nonlinear long-series temporal dependence features hidden deep within the time axis. Finally, the cascaded fully connected output layer plays a mathematical role in decoupling the feature mapping to the target physical quantity, performing linear spatial transformation and dimension merging on the abstract state vector extracted from the LSTM layer to reconstruct and output the degradation index prediction values ​​for multiple consecutive time steps in the future.

[0070] Using the output sequence as a supervision label, the model parameters are iteratively updated through the backpropagation algorithm until the loss function converges, thus obtaining a trained fault prediction model.

[0071] By defining a mathematical loss function, the sum of squared absolute errors between the predicted values ​​and the true labels is quantitatively calculated. The backpropagation algorithm propagates the error from the output layer to the weight matrices and bias vectors of each gated level within the LSTM layer, following the mathematical gradient of this loss function, and fine-tunes each internal scalar coefficient using optimization techniques. Only when the loss function stops decreasing for several consecutive cycles does it indicate that the model has fully grasped and understood the physical laws of multi-field deterioration of electricity, heat, and gas contained in historical big data, and the network parameters have entered the mathematically global Pareto optimal stable solution space. At this point, the trained fault prediction model can provide deterministic evolution trend prediction support for operating power distribution equipment.

[0072] In this embodiment, the number of effective partial discharge signals and the heating temperature corresponding to the key monitoring location within the current sampling time window are extracted, and combined with the cabinet air pressure and ambient temperature in the real-time sensing data, a feature sequence of the key monitoring location is constructed. By focusing on monitoring the partial discharge frequency and pure heating temperature at key locations, the spatial thermal background and electromagnetic stray noise of other healthy components inside the ring main unit can be completely eliminated. By combining these with the cabinet internal air pressure, which represents the macroscopic density of insulating gas, and the ambient temperature, which represents the external climate boundary, the constructed feature sequence contains both the destructive intensity of microscopic lesions and the constraint boundary of the macroscopic environment, providing the most causal and cohesive real-time feature raw materials for subsequent high-fidelity life extrapolation.

[0073] The feature sequence is input into the trained fault prediction model to obtain the degradation prediction values ​​for multiple future time steps from the current moment. The weight matrices and bias vectors of each layer of the trained fault prediction model have deeply locked the degradation evolution mechanism of long-distance time dependence of multiple physical fields such as electricity, heat, and gas. By feeding the latest continuous temporal feature sequence into the model, the long short-term memory layer inside the model can spontaneously combine the current evolutionary memory to extrapolate the internal cell state into the future time domain by utilizing its unique input gate, forget gate, and output gating mechanism. It then performs a linear spatial transformation through the fully connected output layer to directly map and produce degradation prediction values ​​for multiple future time steps.

[0074] The predicted degradation index values ​​for multiple future time steps are sequentially fitted onto the time axis to generate a fault evolution trend curve.

[0075] In the curve fitting process of this embodiment, a preset quadratic least squares polynomial fitting algorithm is specifically invoked. Using the future time span as the independent variable and the predicted values ​​of the six discrete degradation indices output by the model as the dependent variable, polynomial coefficient matrix multiplication and adjustment are performed in memory. Finally, an online fitting is performed to generate a smooth and continuous parabolic function trajectory, which is the fault evolution trend curve.

[0076] In this embodiment, the pre-configured absolute failure threshold of the target high-voltage ring main unit is obtained; In actual operation of power systems, the physical core components inside high-voltage gas-insulated switchgear suffer from long-term electrothermal stress-induced damage, reaching an insurmountable physical degradation limit. Once this limit is exceeded, the main circuit will experience instantaneous complete insulation breakdown or metal thermal melting, leading to a catastrophic power outage.

[0077] Calculate the first intersection point between the fault evolution trend curve and the absolute failure threshold of the equipment, and record the time span from the current moment to the intersection point as the remaining safe operating time; The timeline is used to progressively predict future failures in fixed 1-hour increments. Each future time step is substituted into the parabolic function of the failure evolution trend curve for calculation. The calculated future degradation predictions are then compared sequentially with the equipment's absolute failure threshold of 90.0. The first future timestamp where the function value is greater than or equal to 90.0 is identified as the first intersection point. The current timestamp is subtracted from the future timestamp corresponding to this intersection point to quantitatively calculate the remaining safe operating time, which is specifically output as 48 hours.

[0078] This invention incorporates the fault evolution trend curve into a time-domain analysis process, identifying the tangent point where the curve first rises on the future timeline and reaches the absolute failure threshold. This tangent point, physically speaking, precisely indicates the exact future moment when a potential problem transforms from quantitative to qualitative change. By subtracting the current transient moment from this tangent point, the abstract characteristic evolution can be transformed into a concrete, countdown-able remaining safety time span, achieving a control upgrade from transient passive monitoring to long-term forward-looking quantitative life prediction.

[0079] Obtain a preset first time threshold and a second time threshold, wherein the first time threshold is greater than the second time threshold; If the remaining safe operating time is greater than the first time threshold, a level three warning signal is generated; If the remaining safe operating time is less than or equal to the first time threshold and greater than the second time threshold, a level two warning signal is generated. If the remaining safe operating time is less than or equal to the second time threshold, a level one warning signal is generated.

[0080] In this specific power operation and maintenance management configuration, the preset first time threshold is specifically limited to 72 hours. In distribution network operation management, this time span belongs to the planned maintenance window, which is sufficient for the operation and maintenance department to submit a power outage maintenance application to the distribution and dispatch center, allocate spare parts, and organize personnel. The preset second time threshold is specifically limited to 24 hours. In distribution network operation management, this time span belongs to the emergency defect window, which means that the equipment may collapse at any time, and the regular approval process must be skipped to activate the emergency repair command.

[0081] Based on the relationship between the remaining safe operating time and the magnitudes of various time thresholds, the corresponding early warning control branch output is adaptively executed: Level 3 Warning: If the remaining safe operating time exceeds the first time threshold, a Level 3 warning signal is automatically generated. This indicates that the internal defects of the equipment are evolving extremely slowly, and there is still ample time before failure. This signal prompts the backend terminal to mark the equipment as routinely tracked, allowing for normal inspection schedules to be maintained.

[0082] Level 2 Warning: If the remaining safe operating time is less than or equal to the first time threshold and greater than the second time threshold, a Level 2 warning signal is automatically generated. The fault-free operating life of this high-voltage switchgear has been severely shortened, and it must be forcibly intercepted within the next 1 to 3 days. Since the remaining safe operating time calculated and output in this embodiment is specifically 48 hours, it is determined to be a Level 2 warning. The microprocessor immediately terminates the remaining regular polling and automatically outputs a Level 2 warning signal to the distribution automation master station. This signal directly triggers the master station scheduling software to automatically generate an electronic planned dispatch order and automatically places the power outage maintenance plan for the ring network switchgear into the next day's schedule, realizing an adaptive control closed loop of proactive maintenance dispatch for potential faults.

[0083] Level 1 Warning: If the remaining safe operating time is less than or equal to the second time threshold, a Level 1 warning signal is generated and reported to a higher level. This indicates that an internal defect in the ring main unit is causing a malignant accelerated thermal runaway, and the equipment will suffer irreversible physical damage within 24 hours. This signal not only triggers the highest-level audible and visual alarm in the background, but also remotely links to the pre-trip protection logic of the circuit breaker at the next higher level on the substation side, or forcibly controls the electric operating mechanism of the ring main unit to perform an active tripping and isolation action, fundamentally intercepting the occurrence of a malignant distribution network accident in the time domain.

[0084] Table 1: Data Set Generated by the Monitoring and Control System of the Pneumatic High-Voltage Ring Main Unit The data in this table, through multi-dimensional quantitative comparison, confirms the three core technical advantages of this invention under actual power distribution conditions: First, the precise decoupling and anti-interference capability of the fluid thermal field. Comparative data between scenario A and scenario B show that the system can perfectly strip away the normal Joule heating caused by external high temperatures and heavy loads, while also sensitively capturing weak partial discharges under the cover of low temperatures and low loads, fundamentally eliminating false alarms and missed alarms of traditional monitoring. Second, the precise spatial locking capability of multi-field energy centroids. Compared to the physical blind spots caused by blindly locking the geometric center in traditional technologies, the partial discharge and hotspot location coordinates in the table... The system clearly demonstrates how it dynamically cross-fuses partial discharge pulse density and temperature rise rate as weights to target and lock onto the true core fault evolution domain. Finally, it demonstrates the forward-looking interception capability of deep temporal degradation inference. In scenario C, where faults accelerate and deteriorate, traditional monitoring only issues a delayed alarm when transient limits are exceeded and equipment damage occurs. However, this system extracts the time-dependent features of multi-physics fields through an LSTM model, provides accurate safety countdowns in advance, and adaptively triggers graded early warnings. It successfully connects the temporal domain faults and achieves a leap from immediate passive alarms to forward-looking, proactive perception and control.

[0085] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0086] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0087] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A monitoring and control system for an inflatable high-voltage ring main unit based on data analysis, characterized in that, include: The sampling and evaluation module is used to acquire the operating conditions and real-time sensor data of the gas-insulated high-voltage ring main unit, construct a deterioration index based on the operating conditions, compare the deterioration index with multiple preset deterioration thresholds, classify the monitoring level of the high-voltage ring main unit, and set the current sampling time window based on the monitoring level. The partial discharge location module is used to monitor the partial discharge signal of the ring network cabinet within the current sampling time window, count the number of valid partial discharge signals exceeding a preset threshold, calculate the partial discharge pulse density based on the number of valid partial discharge signals, calculate the dielectric constant of the gas inside the cabinet based on real-time sensor data, and then calculate the propagation speed of the partial discharge signal. It also obtains the arrival time difference sequence of the partial discharge signal to the sensors at different locations, and locates the partial discharge coordinates based on the arrival time difference sequence and the propagation speed. The fault fusion and location module is used to calculate the internal heating temperature of the ring network cabinet based on real-time sensor data and operating conditions, and to perform spatial interpolation of the heating temperature to obtain the hot spot coordinates. It also performs time-domain change rate analysis of the heating temperature to obtain the internal temperature rise rate. The partial discharge pulse density and the internal temperature rise rate are used as weights to perform weighted calculations on the partial discharge coordinates and hot spot coordinates to obtain the key monitoring locations. The fault trend prediction module is used to build a fault prediction model based on the long short-term memory network algorithm. The model is trained based on historical monitoring data, extracts the number of effective partial discharge signals and heating temperature corresponding to key monitoring locations, and constructs a feature sequence by combining real-time sensor data. The feature sequence is input into the trained fault prediction model to obtain the fault evolution trend curve, and an early warning signal is generated based on the fault evolution trend curve.

2. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 1, characterized in that: The operating conditions include: operating time, load current, number of switching operations, ambient temperature, and ambient humidity; The real-time sensing data includes: cabinet internal air pressure, cabinet internal gas temperature, and surface temperature; The thermal cumulative stress is obtained by integrating the square of the load current over the operating time; the mechanical cumulative stress is calculated based on the number of switching operations; and the environmental cumulative stress is calculated based on the ambient temperature and ambient humidity. The thermal accumulated stress, mechanical accumulated stress, and environmental accumulated stress are multiplied by preset weighting coefficients and then weighted and fused to calculate the deterioration index of the high-voltage ring main unit. The larger the deterioration index, the higher the degree of aging and the greater the operational risk of the high-voltage ring main unit. Based on the range of the degradation index, the monitoring level is divided and the current sampling time window is set: If the degradation index is less than or equal to the preset first degradation threshold, the monitoring level will be determined to be level three, the current sampling time window will be set to the first preset duration, and the first sampling frequency will be matched. If the degradation index is greater than the first degradation threshold and less than or equal to the preset second degradation threshold, the monitoring level will be determined to be level two, the current sampling time window will be set to the second preset duration, and the second sampling frequency will be matched. If the degradation index is greater than the second degradation threshold, the monitoring level will be determined to be Level 1, the current sampling time window will be set to the third preset duration, and the third sampling frequency will be matched. Wherein, the second degradation threshold is greater than the first degradation threshold, the first preset duration is greater than the second preset duration, the second preset duration is greater than the third preset duration, the third sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the first sampling frequency.

3. The monitoring and control system for an inflatable high-voltage ring main unit based on data analysis according to claim 2, characterized in that: Within the current sampling time window, partial discharge signals are acquired according to the matched sampling frequency, background electromagnetic noise is filtered out, and the amplitude of the partial discharge signal is obtained. The amplitude of each partial discharge signal is compared with a preset threshold, and the number of partial discharge signals with amplitudes greater than the preset threshold is counted to obtain the number of valid partial discharge signals. The partial discharge pulse density is obtained by dividing the number of valid partial discharge signals by the duration of the sampling time window. Divide the air pressure inside the cabinet by the temperature of the gas inside the cabinet to obtain the temperature-pressure ratio. Multiply the temperature-pressure ratio by the fixed polarization constant of the gas inside the cabinet to obtain the first product. Add the first product to the value 1 to obtain the dielectric constant. The square root of the dielectric constant is calculated by taking the square root of the dielectric constant. The propagation speed of the partial discharge signal is obtained by dividing the speed of light in vacuum by the square root of the dielectric constant.

4. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 1, characterized in that: At least four ultra-high frequency sensors are pre-set on different walls inside the high-voltage ring main unit in a non-coplanar three-dimensional arrangement to obtain the spatial position coordinates of each ultra-high frequency sensor. Using the first sensor in the ultra-high frequency sensor as a reference point, the time difference between the arrival time of the partial discharge signal at each of the other ultra-high frequency sensors and the arrival time at the reference point is obtained, and an arrival time difference sequence is constructed from all the time difference values. Multiply each time difference value in the arrival time difference sequence by the propagation speed to obtain the corresponding spatial distance difference sequence; each element in the sequence represents the difference between the distance from the partial discharge power source to the corresponding other UHF sensor and the distance to the reference point. Based on the spatial position coordinates of each UHF sensor and the spatial distance difference sequence, a three-dimensional hyperboloid positioning equation set is constructed. By solving the positioning equation set, the partial discharge coordinates are obtained.

5. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 1, characterized in that: Obtain the first-order thermal conduction time constant and the second-order thermal capacity time constant of the high-voltage ring main unit shell; Multiplying the first time derivative of the surface temperature by the first-order heat conduction time constant yields the first thermal correction term; multiplying the second time derivative of the surface temperature by the second-order heat capacity time constant yields the second thermal correction term. Calculate the temperature difference between the surface temperature and the ambient temperature, and multiply the temperature difference by a preset surface heat dissipation coefficient to obtain an environmental heat dissipation compensation term; The load reference compensation term is calculated based on the Joule heating effect for the load current; The surface temperature, the first thermal correction term, the second thermal correction term, and the environmental heat dissipation compensation term are added together to obtain the actual internal temperature value. The heating temperature is obtained by subtracting the load reference compensation term from the actual internal temperature value. The coordinate information and heating temperature corresponding to multiple different measuring point locations are obtained. Using the radial basis function interpolation algorithm, the coordinate information is used as the independent variable and the heating temperature is used as the dependent variable to perform surface fitting and construct a three-dimensional continuous temperature field model of the monitoring area inside the high-voltage ring network cabinet. Traverse the three-dimensional continuous temperature field model and take the spatial maximum point with the largest function value in the three-dimensional continuous temperature field model as the hot spot coordinates; Calculate the change in the heating temperature within the sampling time window, and divide the change in the temperature by the length of the sampling time window to obtain the internal temperature rise rate.

6. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 1, characterized in that: Extract the partial discharge pulse density, the internal temperature rise rate, the partial discharge coordinates, and the hot spot coordinates corresponding to the same sampling time window; The first weighted spatial component is calculated by multiplying each dimension component of the partial discharge coordinates in three-dimensional space by the partial discharge pulse density. The second weighted spatial component is calculated by multiplying the hot spot coordinates in each dimension of the three-dimensional space by the internal temperature rise rate. The first weighted spatial component and the second weighted spatial component are added together to obtain the total weighted component; The total weighted component is divided by the sum of the partial discharge pulse density and the internal temperature rise rate to calculate the three-dimensional coordinates of the weighted average center point, and this center point is used as the key monitoring location.

7. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 6, characterized in that: The historical monitoring data of the high-voltage ring main unit is obtained. The historical monitoring data includes: load current, internal air pressure, internal gas temperature, surface temperature, ambient temperature, partial discharge signal and actual value of historical degradation index recorded at each historical time step. Based on the spatial coordinates of the key monitoring location, the number of valid partial discharge signals at each historical time stamp is selected from the historical partial discharge signals and counted. The historical heating temperature of the key monitoring location at each historical time stamp is calculated using the historical surface temperature, historical ambient temperature, and historical load current at the corresponding timestamp. The number of effective partial discharge signals, the historical heating temperature, and the historical cabinet air pressure, historical ambient temperature, and historical degradation index actual values ​​in the historical monitoring data are aligned with a unified timestamp to construct the historical feature sequence. The historical feature sequences are used as training samples and normalized. Supervised learning samples are constructed according to a preset time window length and sliding step size. Continuous... The feature vectors at each time step are the input sequence, and the subsequent... The actual value of the degradation index at each time step is the output sequence, where and All are positive integers; The input sequence is used as input to a long short-term memory network model, wherein the long short-term memory network model includes an LSTM layer and a fully connected output layer. The LSTM layer is used to extract the time-dependent features of the feature sequence, and the fully connected output layer is used to output the predicted value of the degradation index for multiple future time steps. Using the output sequence as a supervision label, the model parameters are iteratively updated through the backpropagation algorithm until the loss function converges, thus obtaining a trained fault prediction model.

8. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 7, characterized in that: Extract the number of valid partial discharge signals and the heating temperature corresponding to the key monitoring location within the current sampling time window, and combine them with the cabinet air pressure and ambient temperature in the real-time sensor data to construct the feature sequence of the key monitoring location; The feature sequence is input into the trained fault prediction model to obtain the degradation prediction values ​​for multiple future time steps from the current moment. The predicted degradation index values ​​for multiple future time steps are sequentially fitted onto the time axis to generate a fault evolution trend curve.

9. The monitoring and control system for a gas-insulated high-voltage ring main unit based on data analysis according to claim 1, characterized in that: Obtain the pre-configured absolute failure threshold of the target high-voltage ring main unit; Calculate the first intersection point between the fault evolution trend curve and the absolute failure threshold of the equipment, and record the time span from the current moment to the intersection point as the remaining safe operating time; Obtain a preset first time threshold and a second time threshold, wherein the first time threshold is greater than the second time threshold; If the remaining safe operating time is greater than the first time threshold, a level three warning signal is generated; If the remaining safe operating time is less than or equal to the first time threshold and greater than the second time threshold, a level two warning signal is generated. If the remaining safe operating time is less than or equal to the second time threshold, a level one warning signal is generated.

Images