An intelligent fault monitoring system and method for power transformation and distribution cabinet

Abstract

The application belongs to the technical field of power distribution cabinet monitoring and analysis, and discloses an intelligent fault monitoring system and method for power transmission and distribution cabinets; the method comprises the following steps: extracting the temperature probability distribution of historical sensing temperature data in a window, constructing a temperature evolution model by analyzing the balance relationship between temperature fluctuation and probability dispersion; extracting the compensation coefficient of environmental parameters on sensing temperature by analyzing the correlation of historical environmental data and historical sensing temperature data in data change; monitoring real-time sensing temperature and real-time environmental data, calculating the theoretical sensing temperature according to the compensation coefficient and real-time environmental data, and calculating the real-time confidence; performing time sequence combination on the real-time confidence to obtain a real-time confidence sequence, and generating a fault monitoring result by analyzing the change characteristics of the real-time confidence sequence. The application solves the problem that temperature sensing equipment is affected by the environment to cause data monitoring abnormalities and is difficult to be found in time, and improves the reliability of the fault monitoring result of the power transmission and distribution cabinet.

Description

Technical Field

[0001] This invention relates to the field of power distribution cabinet monitoring and analysis technology, and more specifically, to an intelligent fault monitoring system and method for power transmission and distribution cabinets. Background Technology

[0002] With the continuous advancement of smart grid construction, fault monitoring of power transmission and distribution cabinets has been transformed towards intelligence and automation. By deploying various types of sensing sensor equipment, real-time collection and analysis of equipment operating parameters and status information can be achieved, providing key technical support for the efficient operation and maintenance of power systems.

[0003] In existing fault monitoring of power transmission and distribution cabinets, temperature sensing devices are the most prone to failure and misjudgment. Environmental factors can cause anomalies in monitoring data, which are highly similar to actual faults in the power transmission and distribution cabinets themselves, leading to abnormal data that closely resembles actual faults and is difficult to detect. Furthermore, due to insufficient detail in the monitoring scenarios, the analysis of historical data lacks accuracy in capturing different operating conditions, resulting in discrepancies between subsequent data analysis results and the actual scenarios, easily leading to misjudgments of power transmission and distribution cabinet faults. For example, the following fault modes directly simulate real fault characteristics: high overall ambient temperature of the distribution cabinet and poor ventilation; high dust concentration inside the cabinet covering the sensor surface, causing temperature deviations; and electromagnetic interference causing sensor signal fluctuations. These problems severely affect the reliability of fault monitoring results for power transmission and distribution cabinets. Therefore, an intelligent fault monitoring system and method for power transmission and distribution cabinets are needed to overcome these issues. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent fault monitoring method for power transmission and distribution cabinets, comprising:

[0005] Acquire multiple historical sensor temperature data for the power transmission and distribution cabinet;

[0006] Extract the temperature probability distribution of historical sensing temperature data within a window, combine the temperature probability distributions of different windows, and construct a temperature evolution model by analyzing the balance between temperature fluctuation and probability dispersion.

[0007] Historical environmental data is collected, and the correlation between historical environmental data and historical sensing temperature data in terms of data changes is analyzed to extract the compensation coefficient of environmental parameters on sensing temperature.

[0008] Monitor real-time sensing temperature and real-time environmental data, calculate the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculate the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature.

[0009] The real-time confidence scores are combined in a time series to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

[0010] Preferably, acquiring multiple historical sensor temperature data of the power transmission and distribution cabinet includes:

[0011] Obtain historical load data of power transmission and distribution cabinets, construct load vectors corresponding to historical load data, identify several stable load vector clusters, and generate corresponding load scenarios based on each stable load vector cluster.

[0012] Based on the time sequence of the load scenarios in which the power transmission and distribution cabinet is located, a time sequence of load scenarios is constructed.

[0013] Construct corresponding scenario change events based on the time sequence changes of adjacent load scenarios in the load scenario time sequence;

[0014] Multiple historical sensor temperature sequences are obtained after each scene change event occurs, and time-series alignment and interpolation are performed on the different historical sensor temperature sequences.

[0015] Preferably, the construction of the temperature evolution model includes:

[0016] The historical temperature sensing sequence is divided into multiple temperature subsequences by a sliding window.

[0017] Integrate temperature subsequences within the same window to obtain a subsequence set, and statistically analyze the temperature probability distribution within the subsequence set;

[0018] Based on the differences in temperature probability distribution within different windows, a stochastic temperature evolution path is constructed.

[0019] The temperature smoothness and probability dispersion of the random temperature evolution path are extracted, and the random temperature evolution path is optimized based on the temperature smoothness and probability dispersion to obtain the optimal temperature evolution path.

[0020] A temperature evolution model is constructed based on the optimal temperature evolution path under different scenario change events.

[0021] Preferably, the construction of the temperature stochastic evolution path includes:

[0022] With the sensed temperature as the horizontal axis and the probability value as the vertical axis, construct the temperature probability density curve within the window;

[0023] Using any axis as the target axis, the temperature probability density curves are stitched together according to the time sequence of the window along the target axis to obtain a probability density time series diagram.

[0024] For each temperature probability density curve, a curve coordinate is randomly selected and designated as the coordinate to be optimized for the window.

[0025] On the probability density time series plot, connect the coordinates to be optimized for each window to obtain a piecewise curve;

[0026] The piecewise curves are smoothed using the least squares fitting method, and the smoothing result is used as the stochastic temperature evolution path.

[0027] Preferably, optimizing the stochastic temperature evolution path includes:

[0028] Extract the sensing temperature amplitude within each window of the random temperature evolution path, and calculate the mean amplitude and standard deviation of amplitude for all windows;

[0029] The temperature smoothness of the random temperature evolution path is obtained by weighted fusion of the mean amplitude and standard deviation of amplitude.

[0030] Identify the probability peak on each temperature probability density curve and mark the historical sensing temperature of the coordinate point corresponding to the probability peak as the target temperature;

[0031] The target temperature is identified on the random temperature evolution path, the probability value at the target temperature is marked as the target probability value, and the difference between the target probability value and the corresponding probability peak value is calculated to obtain the probability difference of the corresponding window.

[0032] The standard deviation is calculated based on the probability differences of all windows and is denoted as the probability dispersion.

[0033] By weighted and fused temperature smoothness and probability dispersion, a fluctuation index of the random temperature development path is obtained.

[0034] With the goal of reducing the fluctuation index, the random temperature evolution path is optimized, and the random temperature evolution path corresponding to the minimum value of the fluctuation index is marked as the optimal temperature evolution path.

[0035] Preferably, the compensation coefficient for the extracted environmental parameters to the sensing temperature includes:

[0036] For each scene change event, integrate the historical environmental data of each historical sensor temperature sequence within each window to obtain a historical environmental dataset.

[0037] Cluster the historical environment dataset for each window to obtain multiple stable environment data clusters;

[0038] Calculate the proportion of each environmental data cluster in the historical environmental dataset, and denote it as the cluster weight of the historical environmental data cluster;

[0039] Calculate the mean of each type of environmental parameter in each environmental data cluster, and combine the mean values ​​of each type of environmental parameter to obtain the cluster average data;

[0040] Historical environmental data clusters are weighted and fused based on cluster weights and cluster averages to obtain standard environmental data.

[0041] The sensing temperature of each window in the corresponding optimal temperature evolution path is taken as the desired temperature.

[0042] Linear fitting was performed on the expected temperature of different windows and the corresponding standard environmental data to obtain the compensation coefficient of each environmental parameter for the sensing temperature.

[0043] Preferably, the method for calculating the real-time confidence level includes:

[0044] The real-time environmental data of each window in the most recent monitoring period is obtained, and the real-time environmental data of each window is compensated sequentially using the compensation coefficient to obtain the theoretical sensing temperature of the corresponding window.

[0045] The real-time sensing temperature of each window in the most recent monitoring cycle is obtained, and the deviation between the real-time sensing temperature and the theoretical sensing temperature of each window is calculated to obtain the real-time sensing temperature deviation.

[0046] Identify current scene change events and determine the probability value of each window as the initial confidence level based on the temperature evolution model;

[0047] The temperature deviation amplitude is compared with the preset standard deviation range. If the real-time sensing temperature deviation amplitude is within the standard deviation range, the initial confidence level is used as the real-time confidence level of the real-time sensing temperature.

[0048] If the real-time temperature deviation is not within the standard deviation range, the real-time confidence level is corrected as follows;

[0049] If the real-time sensing temperature deviation is greater than the maximum value of the standard deviation range, the initial confidence level is reduced, and the reduced initial confidence level is used as the real-time confidence level of the real-time sensing temperature.

[0050] If the real-time temperature deviation is less than the minimum value of the standard deviation range, the initial confidence level is increased, and the increased initial confidence level is used as the real-time confidence level of the real-time temperature.

[0051] Preferably, generating fault monitoring results includes:

[0052] The real-time confidence scores are combined according to the time sequence of the window to obtain the real-time confidence score sequence;

[0053] Calculate the mean periodic confidence score of multiple windows in the most recent monitoring period, and calculate the mean of the initial confidence scores corresponding to multiple windows to obtain the reference confidence score;

[0054] Calculate the absolute difference between the mean confidence level of the period and the reference confidence level, and compare the absolute difference with the preset confidence difference threshold.

[0055] If the confidence difference is greater than the preset confidence difference threshold, the temperature sensor data monitoring is determined to be abnormal.

[0056] If the confidence difference is not greater than the preset confidence difference threshold, the temperature sensor data monitoring is considered normal.

[0057] Preferably, the generation of fault monitoring results further includes:

[0058] When the temperature sensing data is abnormal, the deflection angle and rate of change of the real-time confidence sequence between adjacent windows are extracted, and the deflection angle and rate of change of the rate of change are used as two steady-state indicators.

[0059] Preset deflection angle threshold and rate of change threshold, and compare the two steady-state indices with their corresponding thresholds; if any steady-state indices are greater than the corresponding threshold, mark the corresponding window as an unstable window;

[0060] Calculate the frequency of occurrence of the instability window and record it as the instability frequency. Compare the instability frequency with a preset frequency threshold. If the frequency of occurrence of the instability window is not greater than the preset frequency threshold, it is determined that the temperature sensing data has shifted.

[0061] If the frequency of the instability window is greater than the preset frequency threshold, it is determined that the temperature sensing data is still fluctuating.

[0062] Anomaly warnings are issued based on the type of anomaly in temperature sensor data.

[0063] An intelligent fault monitoring system for a power transmission and distribution cabinet, applied to the aforementioned intelligent fault monitoring method for a power transmission and distribution cabinet, includes:

[0064] The temperature data acquisition module acquires multiple historical sensor temperature data of the power transmission and distribution cabinet;

[0065] The temperature data analysis module extracts the temperature probability distribution of historical sensor temperature data within a window, combines the temperature probability distributions of different windows, and constructs a temperature evolution model by analyzing the balance between temperature fluctuations and probability dispersion.

[0066] The environmental compensation analysis module collects historical environmental data and extracts the compensation coefficient of environmental parameters for sensing temperature by analyzing the correlation between historical environmental data and historical sensing temperature data in terms of data changes.

[0067] The real-time confidence generation module monitors real-time sensing temperature and real-time environmental data, calculates the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculates the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature.

[0068] The fault monitoring and analysis module performs time-series combination of real-time confidence scores to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

[0069] The technical effects and advantages of the intelligent fault monitoring system and method for power transmission and distribution cabinets of the present invention are as follows:

[0070] (1) By clustering, the clustering trend of different load data is identified, and the corresponding load scenario is constructed based on the stable clustered load vector cluster, thus realizing the detailed classification of the operation status of the power transmission and distribution cabinet; a scenario change event is generated according to the scenario change of every two load scenarios, and multiple historical sensor temperature data are obtained under the scenario change event, which improves the reliability and accuracy of data analysis and avoids the large difference in the trend of sensor temperature data change when the historical sensor temperature is collected under different load scenarios changing to the same target load scenario.

[0071] (2) Multiple historical sensing temperatures are segmented by sliding windows, the sequence is fragmented and a set of window temperature subsequences is formed. The temperature probability distribution of different window temperature subsequence sets is extracted and combined to determine the random temperature development path. The random temperature development path is optimized based on the smoothness and probability dispersion of sensing temperature aggregation in different windows, so as to avoid unreasonable fluctuations in the temperature evolution of the power transmission and distribution cabinet and avoid large temperature deviations from the probability peaks of each window during temperature evolution. This balances temperature fluctuations and probability dispersion and effectively captures the evolution trend of sensing temperature of the power transmission and distribution cabinet under scene change events.

[0072] (3) By extracting and combining the temperature probability distribution of historical temperature data, a temperature evolution model is constructed to obtain the initial confidence level of each window; by analyzing the correlation between historical environmental data and historical sensing temperature data in terms of data changes, the compensation coefficient of the extracted environmental parameters on the sensing temperature is obtained, the theoretical sensing temperature is calculated based on the compensation coefficient, the initial confidence level of the real-time sensing temperature is corrected by the theoretical sensing temperature, and the real-time sensing temperature is converted into real-time confidence level. By analyzing the real-time confidence level, the problem of difficulty in timely detection of data monitoring anomalies caused by environmental influences of temperature sensing equipment is solved, and the reliability of fault monitoring results of power transmission and distribution cabinets is improved. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the intelligent fault monitoring method for power transmission and distribution cabinets according to the present invention.

[0074] Figure 2 This is a schematic diagram of the method for constructing a temperature evolution model according to the present invention.

[0075] Figure 3 This is a schematic diagram of the intelligent fault monitoring system for a power transmission and distribution cabinet according to the present invention. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example 1 of this application provides an intelligent fault monitoring method for power transmission and distribution cabinets. By converting real-time sensed temperature into real-time confidence levels and analyzing these confidence levels, the method solves the problem of difficulty in timely detection of data monitoring anomalies caused by environmental influences in temperature sensing devices.

[0078] Please see Figure 1 and Figure 2 In Embodiment 1 of the present invention, a method for intelligent fault monitoring of power transmission and distribution cabinets is implemented in detail through the following steps:

[0079] In this first embodiment, to address the current limitations of fault monitoring for power transmission and distribution cabinets, which cannot be segmented into multiple scenarios based on load characteristics and lacks accuracy in data acquisition regarding changes in operation under different scenarios, resulting in analysis results that are not sufficiently relevant to the current actual situation, the following method is designed:

[0080] Acquire multiple historical sensor temperature data for the power transmission and distribution cabinet, including:

[0081] Historical load data of power transmission and distribution cabinets is obtained, load vectors corresponding to the historical load data are constructed, several stable load vector clusters are identified, and a corresponding load scenario is generated based on each stable load vector cluster. The historical load data includes at least the load capacity percentage, load type, load fluctuation, and load cycle. For example, taking a household power distribution scenario, the historical load data is as follows:

[0082] {Light load 25%, resistive load, steady state 2%, 3:00-4:00 AM}, {Light load 27%, resistive load, steady state 2.2%, 4:00-5:00 AM}, {Regular load 62%, inductive load, dynamic 13%, 10:00-11:00 AM}, and {Regular load 60%, inductive load, dynamic 12%, 11:00-11:30 AM}; where the load cycle in the example description is a daily cycle; the load vector is obtained by encoding different types of data in historical load data separately, concatenating the codes, and so on. The load vectors corresponding to the six historical load data points are generated as a1, a2, a3, a4, a5, and a6. Several stable load vector clusters are identified as follows: for example, the historical load data corresponding to load vectors a1 and a2 are similar in terms of load capacity percentage, load fluctuation, load type, and load cycle, thus identifying them as a stable load vector cluster. Load vectors a3 and a4 are also identified as a stable load vector cluster. The corresponding load scenarios are also illustrated, such as constructing an ID identifier for each stable load vector cluster.

[0083] Based on the time sequence of the load scenarios in which the power transmission and distribution cabinet is located, a time sequence of load scenarios is constructed.

[0084] Based on the time sequence changes of adjacent load scenarios in the load scenario time sequence, a corresponding scenario change event is constructed. For example, if the current load scenario ID is X1 and the ID of the previous load scenario that is adjacent to the current load scenario in time sequence is X2, then an ID pair is formed based on IDs X1 and X2, and the ID pair is used as scenario change event A1, indicating that the power distribution cabinet is under this scenario change event.

[0085] Multiple historical sensor temperature sequences after each scene change event are obtained, and time-series alignment and interpolation are performed on the different historical sensor temperature sequences. The time-series alignment is achieved using dynamic time warping, and the interpolation is achieved using linear interpolation. Both dynamic time warping and linear interpolation are existing technologies and will not be described in detail here.

[0086] Obtain multiple historical sensor temperature sequences after each scene change event occurs, including:

[0087] Select multiple historical periods, and for each historical period, specify the target scenario change event;

[0088] In different historical periods, all historical sensor temperature data from the occurrence to the end of the target scene change event are acquired. Based on the chronological order of the historical sensor temperature data, a historical sensor temperature sequence is generated. For example, based on scene change event A1 generated by IDs X1 and X2, the historical sensor temperature sequences after each scene change event A1 in the previous three months are obtained, with the time periods being 3:00-5:00 AM, 2:58-4:56 AM, 2:54-5:01 AM, 2:59-4:57 AM, and 3:02 AM respectively. For example, if different historical sensing temperature sequences are time-series aligned to ensure that the time span of each historical sensing temperature sequence is the same, then it should be noted that if another scene change event occurs under a scene change event, the target scene change event is considered to have ended. For example, for scene change event A1, if scene change event A1 occurs at 3:00 AM and scene change event A2 occurs at 5:00 AM, then the historical sensing temperature data from 3:00 AM to 5:00 AM is obtained to generate the historical sensing temperature sequence of scene change event A1.

[0089] By clustering to identify the aggregation trends of different load data, and constructing corresponding load scenarios based on the stable clustered load vector clusters, a detailed classification of the operating status of power transmission and distribution cabinets is achieved. A scenario change event is generated based on the scenario change of every two load scenarios. By acquiring multiple historical sensor temperature data under the scenario change event, the reliability and accuracy of data analysis are improved. This avoids data collection and analysis when different load scenarios change to the same target load scenario, and solves the problem of lack of accuracy in scenario control in data acquisition.

[0090] To address the inconsistency in the aggregation trends of different historical temperature sensing data, which makes it difficult to capture the evolution trend of the sensing temperature of power transmission and distribution cabinets under scene change events, the following method is designed:

[0091] Extract the temperature probability distribution of historical sensing temperature data within a window, combine the temperature probability distributions of different windows, and construct a temperature evolution model by analyzing the balance between temperature fluctuation and probability dispersion.

[0092] In this first embodiment, a temperature evolution model is constructed, including:

[0093] Historical temperature sensing sequences are segmented using a sliding window to obtain multiple temperature subsequences. A uniform segmentation window length is set for each historical temperature sensing sequence. For the segmentation window length, the average noise of the window and the overall data computational complexity are extracted. A weighted calculation based on the average noise and data computational complexity is used as an optimization index for the window length. This optimization index is trained on large datasets to determine the final segmentation window length. For example, when the segmentation window length can be set to 5 minutes, for three historical temperature sensing sequences b1, b2, and b3, after time alignment, the sequence time span is 60 minutes, and each will yield 12 temperature subsequences after sliding window segmentation.

[0094] Integrate temperature subsequences within the same window to obtain a subsequence set, and statistically analyze the temperature probability distribution within the subsequence set; an example is given of integrating temperature subsequences within the same window, such as integrating the first temperature subsequences corresponding to historical sensing temperature sequences b1, b2, and b3 in the first 5-minute window to obtain a subsequence set consisting of three temperature subsequences;

[0095] In this first embodiment, the temperature probability distribution of the statistical historical sensing temperature data includes:

[0096] Extract and integrate historical sensor temperature values ​​from the subsequence set to obtain a temperature value sample set;

[0097] Set temperature ranges, count the frequency of temperature values ​​in each range within the temperature value sample set, and construct a frequency histogram as the temperature probability distribution. The number of temperature ranges is limited to 20-30 to avoid increasing the amount of data due to too many ranges, and to avoid making the probability distribution unclear due to too few ranges. The number of temperature ranges is obtained by training on the probability distribution of historical temperature data.

[0098] Based on the differences in temperature probability distribution within different windows, a stochastic temperature evolution path is constructed.

[0099] In this first embodiment, the construction of a stochastic temperature evolution path includes:

[0100] With the sensed temperature as the horizontal axis and the probability value as the vertical axis, construct the temperature probability density curve within the window;

[0101] In this first embodiment, the construction of the temperature probability density curve within the window includes:

[0102] The probability distribution of temperature data in a statistical temperature data sample set is estimated using kernel density estimation; the kernel function type used is Gaussian kernel.

[0103] The distribution pattern of temperature data in the temperature sample dataset is identified, and the bandwidth is optimized based on the distribution pattern to determine the preliminary optimal bandwidth and generate a KDE (kernel density estimation) curve. An example is provided for optimizing the bandwidth based on the distribution pattern of the temperature data. For instance, if the temperature data has a unimodal symmetrical distribution, empirical rules such as Scott's rule are used to optimize the bandwidth; conversely, if the temperature data has a non-normal distribution, cross-validation is used to optimize the bandwidth, such as leave-one-out cross-validation.

[0104] Compare the KDE curve with the frequency histogram, and identify the peak values ​​of the frequency histogram and the KDE curve respectively, which are denoted as the frequency peak value and the curve peak value.

[0105] The process involves determining whether the frequency peak value matches the curve peak value to obtain a judgment result. If the frequency peak value matches the curve peak value, the preliminary optimal bandwidth is taken as the final bandwidth. If the frequency peak value does not match the curve peak value, the preliminary optimal bandwidth is optimized a second time until the frequency peak value matches the curve peak value, and the final bandwidth is determined. For example, if the KDE curve shows multiple small peak types under the preliminary optimal bandwidth, and does not meet the frequency peak value of the frequency histogram, then the KDE curve is determined to have noise, and the preliminary optimal bandwidth is too low. Alternatively, if the curve peak value shows a double-peak fusion phenomenon compared to the frequency peak value, then it is determined that features are lost, and the bandwidth is too high.

[0106] The final KDE curve is regenerated based on the final bandwidth, and the final KDE curve is used as the temperature probability density curve.

[0107] Using any axis as the target axis, the temperature probability density curves are stitched together according to the time sequence of the windows along the target axis to obtain a probability density time series diagram. Obtaining the probability density time series diagram further includes: unifying the data range and range width along the two axes, and performing a linear transformation on the temperature probability density curves of each window. For example, the axis from which the sensing temperature is formed is used as the target axis, and the probability density curves of each window are stitched together along the target axis. The sensing temperature ranges of the temperature probability density curves within different windows are unionized to determine the maximum sensing temperature range, and a linear transformation is performed on the different temperature probability density curves based on the maximum sensing temperature range.

[0108] To address the randomness of the sensed temperature within each window, the following method was designed: a curve coordinate was randomly selected from each temperature probability density curve and designated as the coordinate to be optimized for the window; by selecting random coordinates, the randomness of the sensed temperature within the window was simulated.

[0109] To address the stochastic evolution trend of the sensed temperature within different windows, the following method was designed:

[0110] On the probability density time series plot, connect the coordinates to be optimized for each window to obtain a piecewise curve;

[0111] The piecewise curves are smoothed using the least squares fitting method, and the smoothing result is used as the random evolution path of temperature. The piecewise curves combine the randomness of the sensing temperature in different windows to generate the sensing temperature evolution trend. In particular, the piecewise curves are transformed into smooth curves by the least squares fitting method, which makes it easier to highlight the gentle characteristics of the sensing temperature change and avoid the irrationality of linear changes and abrupt changes in the piecewise curves.

[0112] The temperature smoothness and probability dispersion of the random temperature evolution path are extracted, and the random temperature evolution path is optimized based on the temperature smoothness and probability dispersion to obtain the optimal temperature evolution path.

[0113] In this first embodiment, to address the imbalance between temperature fluctuations and probability discreteness in the stochastic temperature evolution path, the stochastic temperature evolution path is optimized, including:

[0114] Extract the sensing temperature amplitude within each window of the random temperature evolution path, and calculate the mean amplitude and standard deviation of amplitude for all windows;

[0115] The mean amplitude and standard deviation amplitude are weighted and fused to obtain the temperature smoothness of the stochastic temperature evolution path. The mean amplitude reflects the overall volatility of the stochastic evolution path in terms of temperature development and probability value changes, while the standard deviation amplitude reflects the stability of amplitude changes in the stochastic temperature evolution path across different windows. The weights of the mean amplitude and standard deviation amplitude are obtained based on large-scale training on historical temperature smoothness data. The lower the temperature smoothness, the more gradual the trend of temperature evolution.

[0116] Identify the probability peak on each temperature probability density curve and mark the historical sensing temperature of the coordinate point corresponding to the probability peak as the target temperature;

[0117] The target temperature is identified on the random temperature evolution path, the probability value at the target temperature is marked as the target probability value, and the difference between the target probability value and the corresponding probability peak value is calculated to obtain the probability difference of the corresponding window.

[0118] The standard deviation is calculated based on the probability difference of all windows and is denoted as the probability dispersion. The probability dispersion reflects the degree of deviation of the target probability value of the stochastic evolution path from the probability peak in each window. The lower the probability dispersion, the closer the stochastic temperature evolution path is to the probability peak of different windows, and the more ideal it is in terms of the probability of temperature evolution.

[0119] By weighted fusion of temperature smoothness and probability dispersion, a fluctuation index of the random temperature development path is obtained; the weights of temperature smoothness and probability dispersion are obtained based on big data training of historical fluctuation indexes.

[0120] With the goal of reducing the fluctuation index, the random temperature evolution path is optimized, and the random temperature evolution path corresponding to the minimum value of the fluctuation index is marked as the optimal temperature evolution path.

[0121] A temperature evolution model is constructed based on the optimal temperature evolution path under different scenario change events; the optimal temperature evolution path for each scenario change event is periodically updated, and the update period can be set to three months.

[0122] Multiple historical sensor temperatures are segmented using a sliding window, and the sequence is fragmented and assembled into window temperature subsequence sets. By extracting and combining the temperature probability distribution of different window temperature subsequence sets, the random temperature development path is determined. The random temperature development path is optimized based on the smoothness and probability dispersion of the sensor temperature aggregation in different windows, avoiding unreasonable fluctuations in the temperature evolution of the power transmission and distribution cabinet, and avoiding large temperature deviations from the probability peaks of each window during temperature evolution. This balances temperature fluctuations and probability dispersion, effectively capturing the evolution trend of the sensor temperature of the power transmission and distribution cabinet under scene change events.

[0123] To address the issue of environmental data affecting sensing devices and causing changes in the sensed temperature data, resulting in a deviation from the actual sensed temperature, the following method was designed:

[0124] Historical environmental data is collected, and the correlation between historical environmental data and historical sensing temperature data in terms of data changes is analyzed to extract the compensation coefficient of environmental parameters on sensing temperature.

[0125] In this first embodiment, the compensation coefficient for the environmental parameters on the sensing temperature is extracted, including:

[0126] For each scene change event, integrate the historical environmental data of each historical sensor temperature sequence within each window to obtain a historical environmental dataset.

[0127] Clustering is performed on the historical environmental dataset of each window to obtain multiple stable environmental data clusters. In this embodiment, the clustering of the historical environmental dataset of each window includes converting each historical environmental data in the historical environmental dataset into a historical environmental vector and clustering the historical environmental vectors. The historical environmental data includes at least ambient temperature, electromagnetic intensity, air humidity, and dust concentration. In this embodiment, the clustering is implemented using an unsupervised clustering algorithm.

[0128] Calculate the proportion of each environmental data cluster in the historical environmental dataset, and denote it as the cluster weight of the historical environmental data cluster;

[0129] Calculate the mean of each type of environmental parameter in each environmental data cluster, and combine the mean values ​​of each type of environmental parameter to obtain the cluster average data;

[0130] Historical environmental data clusters are weighted and fused based on cluster weights and cluster averages to obtain standard environmental data.

[0131] The sensing temperature of each window in the corresponding optimal temperature evolution path is taken as the desired temperature.

[0132] Linear fitting was performed on the expected temperature of different windows and the corresponding standard environmental data to obtain the compensation coefficient of each environmental parameter for the sensing temperature.

[0133] The compensation coefficients for each environmental parameter to the sensing temperature also include:

[0134] The linear fitting method employs a multivariate piecewise linear regression approach to construct a linear mapping model between the desired temperature and the standard environmental data. The linear mapping model is as follows:

[0135]

[0136] In the formula, T i T0 is the desired temperature value; T0 is the baseline temperature value, calculated based on a linear fit of the desired temperature value and historical environmental data using large-scale data analysis; a i For the coefficients of the corresponding environmental parameters; P t,i For real-time environmental parameters; P sd,i The reference environmental parameter is i; i is the environmental parameter number, i is a positive integer; n is the total number of environmental parameters; the coefficient term is used as the compensation coefficient of the environmental parameter for the sensing temperature;

[0137] By analyzing the compensation coefficient of ambient temperature on the sensing temperature, the occurrence of large deviations in the sensing temperature is avoided, making it easier to obtain a more accurate sensing temperature.

[0138] To address the issue of temperature fluctuations and deviations in sensing equipment due to environmental influences, which closely resemble actual faults in power transmission and distribution cabinets and result in abnormal temperature data that closely mimics real-world faults, thus hindering timely detection of these anomalies and severely impacting the reliability of fault monitoring results, the following method is designed:

[0139] Monitor real-time sensing temperature and real-time environmental data, calculate the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculate the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature.

[0140] In this first embodiment, the method for calculating the real-time confidence level includes:

[0141] The real-time environmental data of each window in the most recent monitoring period is obtained, and the real-time environmental data of each window is compensated sequentially using the compensation coefficient to obtain the theoretical sensing temperature of the corresponding window.

[0142] To further explain the theoretical sensing temperature for the corresponding window, it is necessary to align the window of the most recent monitoring period with the multiple windows obtained by sliding window segmentation of the historical sensing temperature sequence. For example, the most recent monitoring period is between the 200th and 800th windows after the current scene change event occurs. After time alignment, the multiple historical sensing temperature sequences corresponding to the current scene change event are divided into 1000 windows by sliding window segmentation. Then, the theoretical sensing temperature of the 200th to 800th windows corresponding to the current scene change event is obtained according to the temperature evolution model.

[0143] The real-time sensing temperature of each window in the most recent monitoring period is obtained. The deviation between the real-time sensing temperature and the theoretical sensing temperature of each window is calculated to obtain the real-time sensing temperature deviation. The real-time sensing temperature deviation can be calculated using the following formula.

[0144] In the formula, SP is the real-time sensing temperature deviation; ST is the real-time sensing temperature; LT is the theoretical sensing temperature;

[0145] Identify current scene change events and determine the probability value of each window as the initial confidence level based on the temperature evolution model;

[0146] The temperature deviation amplitude is compared with the preset standard deviation range. If the real-time sensing temperature deviation amplitude is within the standard deviation range, the initial confidence level is used as the real-time confidence level of the real-time sensing temperature. The preset standard deviation range is obtained by training based on historical sensing temperature big data. For example, the standard deviation range can be set to (2%, 10%).

[0147] If the real-time temperature deviation is not within the standard deviation range, the real-time confidence level is corrected as follows;

[0148] If the real-time sensing temperature deviation is greater than the maximum value of the standard deviation range, the initial confidence level is reduced, and the reduced initial confidence level is used as the real-time confidence level of the real-time sensing temperature.

[0149] If the real-time sensing temperature deviation is less than the minimum value of the standard deviation range, the initial confidence level is increased, and the increased initial confidence level is used as the real-time confidence level of the real-time sensing temperature.

[0150] Specifically, the expression for calculating real-time confidence is as follows:

[0151]

[0152] In the formula, RC represents the real-time confidence level; IC represents the initial confidence level; SD represents the initial confidence level. min This is the minimum value of the standard deviation range; SD max The maximum value of the standard offset range; α and β are the increase and decrease coefficients of the initial confidence IC, respectively, which are obtained by training based on the confidence correction big data of historical sensing temperature;

[0153] The real-time confidence scores are combined in a time series to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

[0154] In this first embodiment, the fault monitoring results are generated, including:

[0155] The real-time confidence scores are combined according to the time sequence of the window to obtain the real-time confidence score sequence;

[0156] The mean periodic confidence score of multiple windows in the most recent monitoring period is calculated, and the mean of the initial confidence scores corresponding to multiple windows is calculated to obtain the reference confidence score. It should be noted that the reference confidence score is determined when the optimal temperature evolution path under the first analysis of the scene switching event is analyzed. The reference confidence score is not updated when the optimal temperature evolution path is updated periodically in subsequent analysis.

[0157] Calculate the absolute difference between the mean confidence level of the period and the reference confidence level, and compare the absolute difference with the preset confidence difference threshold; wherein, the confidence difference threshold is set to 0.2;

[0158] If the confidence difference is greater than the preset confidence difference threshold, the temperature sensor data monitoring is determined to be abnormal.

[0159] If the confidence difference is not greater than the preset confidence difference threshold, the temperature sensor data monitoring is considered normal.

[0160] When the temperature sensing data is abnormal, the deflection angle and rate of change of the real-time confidence sequence between adjacent windows are extracted, and the deflection angle and rate of change of the rate of change are used as two steady-state indicators.

[0161] Preset deflection angle threshold and rate of change threshold, and compare the two steady-state indices with their corresponding thresholds; if any steady-state indices are greater than the corresponding threshold, the corresponding window is marked as an unstable window; the deflection angle mainly describes the angle between the tangent line of the real-time confidence sequence in one window and the tangent line of another window; the deflection angle threshold and rate of change threshold are obtained by training based on big data of deflection angle and big data of change rate of historical confidence sequences, respectively;

[0162] Calculate the frequency of occurrence of the instability window and record it as the instability frequency. Compare the instability frequency with a preset frequency threshold. If the frequency of occurrence of the instability window is greater than the preset frequency threshold, it is determined that the temperature sensing data is still fluctuating.

[0163] If the frequency of the instability window does not exceed a preset frequency threshold, the temperature sensing data is determined to have shifted; the preset frequency threshold is obtained by training based on large datasets of instability frequencies from historical confidence sequences.

[0164] Provide early warnings of anomalies based on the types of anomalies in temperature sensor data;

[0165] Supplementary explanation for abnormal temperature sensing data monitoring: When the sensing device is affected by the environment or changes itself, resulting in monitoring offset or data jitter, the optimal temperature evolution path obtained by combining the temperature probability distribution of recent and historical temperature sensing data will have a lower initial confidence level. The initial confidence level is then corrected by combining recent real-time and theoretical temperature sensing data to obtain the periodic confidence average of recent real-time sensing data. Compared with the reference confidence level, the confidence difference is likely to exceed the confidence difference threshold, thus determining that the temperature sensing data monitoring is abnormal.

[0166] By extracting and combining the temperature probability distribution of historical temperature data, a temperature evolution model is constructed to obtain the initial confidence level for each window. By analyzing the correlation between historical environmental data and historical sensor temperature data in terms of data changes, the compensation coefficient of the extracted environmental parameters on the sensor temperature is obtained. The theoretical sensor temperature is calculated based on the compensation coefficient, and the initial confidence level of the real-time sensor temperature is corrected by the theoretical sensor temperature, thus converting the real-time sensor temperature into a real-time confidence level. By analyzing the real-time confidence level, the problem of difficulty in timely detection of data monitoring anomalies caused by environmental influences of temperature sensing equipment is solved, thereby improving the reliability of fault monitoring results of power transmission and distribution cabinets.

[0167] Example 2, please refer to Figure 3 Embodiment 2 of this application provides an intelligent fault monitoring system for power transmission and distribution cabinets, applied to the intelligent fault monitoring method for power transmission and distribution cabinets provided in Embodiment 1, specifically including:

[0168] The temperature data acquisition module acquires multiple historical sensor temperature data of the power transmission and distribution cabinet;

[0169] The temperature data analysis module extracts the temperature probability distribution of historical sensor temperature data within a window, combines the temperature probability distributions of different windows, and constructs a temperature evolution model by analyzing the balance between temperature fluctuations and probability dispersion.

[0170] The environmental compensation analysis module collects historical environmental data and extracts the compensation coefficient of environmental parameters for sensing temperature by analyzing the correlation between historical environmental data and historical sensing temperature data in terms of data changes.

[0171] The real-time confidence generation module monitors real-time sensing temperature and real-time environmental data, calculates the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculates the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature.

[0172] The fault monitoring and analysis module performs time-series combination of real-time confidence scores to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

[0173] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0174] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0175] All formulas in this manual are dimensionless and calculated numerically. 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 and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0176] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent fault monitoring of power transmission and distribution cabinets, characterized in that, include: Acquire multiple historical sensor temperature data for the power transmission and distribution cabinet; The acquisition of multiple historical sensor temperature data of the power transmission and distribution cabinet includes: Obtain historical load data of power transmission and distribution cabinets, construct load vectors corresponding to historical load data, identify several stable load vector clusters, and generate corresponding load scenarios based on each stable load vector cluster. Based on the time sequence of the load scenarios in which the power transmission and distribution cabinet is located, a time sequence of load scenarios is constructed. Construct corresponding scenario change events based on the time sequence changes of adjacent load scenarios in the load scenario time sequence; Acquire multiple historical sensor temperature sequences after each scene change event occurs, and perform time-series alignment and interpolation on different historical sensor temperature sequences; Extract the temperature probability distribution of historical sensing temperature data within a window, combine the temperature probability distributions of different windows, and construct a temperature evolution model by analyzing the balance between temperature fluctuation and probability dispersion. The construction of the temperature evolution model includes: The historical temperature sensing sequence is divided into multiple temperature subsequences by a sliding window. Integrate temperature subsequences within the same window to obtain a subsequence set, and statistically analyze the temperature probability distribution within the subsequence set; Based on the differences in temperature probability distribution within different windows, a stochastic temperature evolution path is constructed. The temperature smoothness and probability dispersion of the random temperature evolution path are extracted, and the random temperature evolution path is optimized based on the temperature smoothness and probability dispersion to obtain the optimal temperature evolution path. Construct a temperature evolution model based on the optimal temperature evolution path under different scenario change events; Historical environmental data is collected, and the correlation between historical environmental data and historical sensing temperature data in terms of data changes is analyzed to extract the compensation coefficient of environmental parameters on sensing temperature. Monitor real-time sensing temperature and real-time environmental data, calculate the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculate the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature. The method for calculating real-time confidence includes: The real-time environmental data of each window in the most recent monitoring period is obtained, and the real-time environmental data of each window is compensated sequentially using the compensation coefficient to obtain the theoretical sensing temperature of the corresponding window. The real-time sensing temperature of each window in the most recent monitoring cycle is obtained, and the deviation between the real-time sensing temperature and the theoretical sensing temperature of each window is calculated to obtain the real-time sensing temperature deviation. Identify current scene change events and determine the probability value of each window as the initial confidence level based on the temperature evolution model; The temperature deviation amplitude is compared with the preset standard deviation range. If the real-time sensing temperature deviation amplitude is within the standard deviation range, the initial confidence level is used as the real-time confidence level of the real-time sensing temperature. If the real-time temperature deviation is not within the standard deviation range, the real-time confidence level is corrected as follows; If the real-time sensing temperature deviation is greater than the maximum value of the standard deviation range, the initial confidence level is reduced, and the reduced initial confidence level is used as the real-time confidence level of the real-time sensing temperature. If the real-time sensing temperature deviation is less than the minimum value of the standard deviation range, the initial confidence level is increased, and the increased initial confidence level is used as the real-time confidence level of the real-time sensing temperature. The real-time confidence scores are combined in a time series to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

2. The intelligent fault monitoring method for power transmission and distribution cabinets according to claim 1, characterized in that, The construction of the temperature stochastic evolution path includes: With the sensed temperature as the horizontal axis and the probability value as the vertical axis, construct the temperature probability density curve within the window; Using any axis as the target axis, the temperature probability density curves are stitched together according to the time sequence of the window along the target axis to obtain a probability density time series diagram. For each temperature probability density curve, a curve coordinate is randomly selected and designated as the coordinate to be optimized for the window. On the probability density time series plot, connect the coordinates to be optimized for each window to obtain a piecewise curve; The piecewise curves are smoothed using the least squares fitting method, and the smoothing result is used as the stochastic temperature evolution path.

3. The intelligent fault monitoring method for power transmission and distribution cabinets according to claim 2, characterized in that, The optimization of the stochastic temperature evolution path includes: Extract the sensing temperature amplitude within each window of the random temperature evolution path, and calculate the mean amplitude and standard deviation of amplitude for all windows; The temperature smoothness of the random temperature evolution path is obtained by weighted fusion of the mean amplitude and the standard deviation of amplitude. Identify the probability peak on each temperature probability density curve and mark the historical sensing temperature of the coordinate point corresponding to the probability peak as the target temperature; The target temperature is identified on the random temperature evolution path, the probability value at the target temperature is marked as the target probability value, and the difference between the target probability value and the corresponding probability peak value is calculated to obtain the probability difference of the corresponding window. The standard deviation is calculated based on the probability differences of all windows and is denoted as the probability dispersion. By weighted and fused temperature smoothness and probability dispersion, a fluctuation index of the random temperature development path is obtained. With the goal of reducing the fluctuation index, the random temperature evolution path is optimized, and the random temperature evolution path corresponding to the minimum value of the fluctuation index is marked as the optimal temperature evolution path.

4. The intelligent fault monitoring method for power transmission and distribution cabinets according to claim 3, characterized in that, The compensation coefficient for the extracted environmental parameters on the sensing temperature includes: For each scene change event, integrate the historical environmental data of each historical sensor temperature sequence within each window to obtain a historical environmental dataset. Cluster the historical environment dataset for each window to obtain multiple stable environment data clusters; Calculate the proportion of each environmental data cluster in the historical environmental dataset, and denote it as the cluster weight of the historical environmental data cluster; Calculate the mean of each type of environmental parameter in each environmental data cluster, and combine the mean values ​​of each type of environmental parameter to obtain the cluster average data; Historical environmental data clusters are weighted and fused based on cluster weights and cluster averages to obtain standard environmental data. The sensing temperature of each window in the corresponding optimal temperature evolution path is taken as the desired temperature. Linear fitting was performed on the expected temperature of different windows and the corresponding standard environmental data to obtain the compensation coefficient of each environmental parameter for the sensing temperature.

5. The intelligent fault monitoring method for a power transmission and distribution cabinet according to claim 4, characterized in that, The generated fault monitoring results include: The real-time confidence scores are combined according to the time sequence of the window to obtain the real-time confidence score sequence; Calculate the mean periodic confidence score of multiple windows in the most recent monitoring period, and calculate the mean of the initial confidence scores corresponding to multiple windows to obtain the reference confidence score; Calculate the absolute difference between the mean confidence level of the period and the reference confidence level, and compare the absolute difference with the preset confidence difference threshold. If the confidence difference is greater than the preset confidence difference threshold, the temperature sensor data monitoring is determined to be abnormal. If the confidence difference is not greater than the preset confidence difference threshold, the temperature sensor data monitoring is considered normal.

6. The intelligent fault monitoring method for a power transmission and distribution cabinet according to claim 5, characterized in that, The generation of fault monitoring results also includes: When the temperature sensing data is abnormal, the deflection angle and rate of change of the real-time confidence sequence between adjacent windows are extracted, and the deflection angle and rate of change of the rate of change are used as two steady-state indicators. Preset deflection angle threshold and rate of change threshold, and compare the two steady-state indices with their corresponding thresholds; if any steady-state indices are greater than the corresponding threshold, mark the corresponding window as an unstable window; Calculate the frequency of occurrence of the instability window and record it as the instability frequency. Compare the instability frequency with a preset frequency threshold. If the frequency of occurrence of the instability window is not greater than the preset frequency threshold, it is determined that the temperature sensing data has shifted. If the frequency of the instability window is greater than the preset frequency threshold, it is determined that the temperature sensing data is still fluctuating. Anomaly warnings are issued based on the type of anomaly in temperature sensor data.

7. An intelligent fault monitoring system for a power transmission and distribution cabinet, applied to the intelligent fault monitoring method for a power transmission and distribution cabinet as described in any one of claims 1-6, characterized in that, include: The temperature data acquisition module acquires multiple historical sensor temperature data of the power transmission and distribution cabinet; The temperature data analysis module extracts the temperature probability distribution of historical sensor temperature data within a window, combines the temperature probability distributions of different windows, and constructs a temperature evolution model by analyzing the balance between temperature fluctuations and probability dispersion. The environmental compensation analysis module collects historical environmental data and extracts the compensation coefficient of environmental parameters for sensing temperature by analyzing the correlation between historical environmental data and historical sensing temperature data in terms of data changes. The real-time confidence generation module monitors real-time sensing temperature and real-time environmental data, calculates the theoretical sensing temperature based on the compensation coefficient and real-time environmental data, and calculates the real-time confidence level based on the temperature evolution model and the deviation between the real-time sensing temperature and the theoretical sensing temperature. The fault monitoring and analysis module performs time-series combination of real-time confidence scores to obtain a real-time confidence score sequence. By analyzing the changing characteristics of the real-time confidence score sequence, fault monitoring results are generated.

Images