A safe operation and maintenance method and system driven by large internal resistance data

Abstract

The application discloses a kind of internal resistance big data driven safe operation method and system, method includes: the historical and real-time internal resistance data of target equipment is collected, and the internal resistance time series data set after normalization is generated by dynamic time normalization algorithm;The internal resistance time series data set is carried out multi-scale feature extraction and high-dimensional space mapping, and the internal resistance abnormal mode cluster implied in data distribution is identified using adaptive density clustering algorithm;Based on internal resistance abnormal mode cluster, multi-modal fault correlation analysis is carried out, and a fault prediction atlas is generated;According to the fault prediction atlas, a set of differentiated operation and maintenance strategies is generated, and the operation and maintenance management platform is driven to execute corresponding operation and maintenance instructions. Using the embodiment of the application, the timeliness and accuracy of fault early warning can be improved, the operation and maintenance cost is reduced, and the safe and stable operation of equipment is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of operation and maintenance, in particular to a safety operation and maintenance method and system driven by internal resistance big data. BACKGROUND

[0002] In key fields such as power, communication and new energy batteries, the reliability and safety operation of equipment are crucial. As one of the core parameters representing the health status of equipment, the change trend of internal resistance is closely related to potential faults. Existing operation and maintenance methods mostly rely on threshold monitoring of voltage, current, temperature and other conventional parameters, or simple threshold judgment and trend analysis of internal resistance data. Such methods are difficult to effectively mine the deep evolution rules and early abnormal signs of internal resistance data in the time dimension, and lack the ability to predict complex and gradual faults. At the same time, traditional methods usually treat internal resistance data in isolation and fail to deeply integrate and analyze it with multi-dimensional operating conditions of equipment, resulting in high false and missed alarm rates, and being unable to effectively evaluate the evolution path and risk level of faults, which makes the operation and maintenance decision lack of foresight and precision, and the efficiency of operation and maintenance resource allocation is low. SUMMARY

[0003] The purpose of the present application is to provide a safety operation and maintenance method and system driven by internal resistance big data, to solve the problems in the prior art, and to improve the timeliness and accuracy of fault early warning, reduce operation and maintenance costs and ensure safe and stable operation of equipment.

[0004] One embodiment of the present application provides a safety operation and maintenance method driven by internal resistance big data, which comprises:

[0005] Collecting historical and real-time internal resistance data of a target device, and combining with multi-dimensional operating state parameters, generating a set of internal resistance time series data after normalization by a dynamic time warping algorithm;

[0006] Performing multi-scale feature extraction and high-dimensional space mapping on the set of internal resistance time series data, constructing a set of internal resistance feature vectors, and identifying internal resistance abnormal pattern clusters hidden in the data distribution by using an adaptive density clustering algorithm;

[0007] Performing multi-modal fault correlation analysis based on the internal resistance abnormal pattern clusters, predicting the evolution path and risk level of potential faults in combination with the operating history and operating condition labels of the equipment, and generating a fault prediction map;

[0008] Generating a set of differentiated operation and maintenance strategies according to the fault prediction map, and driving an operation and maintenance management platform to execute corresponding operation and maintenance instructions.

[0009] Another embodiment of the present application provides a safety operation and maintenance system driven by internal resistance big data, which comprises:

[0010] The collection module is configured to collect historical and real-time internal resistance data of a target device, combine multi-dimensional running state parameters, and generate a set of internal resistance time series data after time warping through a dynamic time warping algorithm.

[0011] The construction module is configured to perform multi-scale feature extraction and high-dimensional space mapping on the set of internal resistance time series data, construct a set of internal resistance feature vectors, and identify internal resistance abnormal pattern clusters hidden in the data distribution by using an adaptive density clustering algorithm.

[0012] The analysis module is configured to perform multi-modal fault correlation analysis based on the internal resistance abnormal pattern clusters, combine device running history and working condition labels to predict the evolution path and risk level of potential faults, and generate a fault prediction graph.

[0013] The operation and maintenance module is configured to generate a set of differentiated operation and maintenance strategies based on the fault prediction graph, and drive an operation and maintenance management platform to execute corresponding operation and maintenance instructions.

[0014] Another embodiment of the present application provides a storage medium having a computer program stored therein, wherein the computer program is configured to execute the method described in any of the above embodiments when running.

[0015] Another embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory has a computer program stored therein, and the processor is configured to execute the computer program to perform the method described in any of the above embodiments.

[0016] Compared with the prior art, the safe operation and maintenance method driven by internal resistance big data provided by the present application can improve the timeliness and accuracy of fault early warning, reduce operation and maintenance costs, and ensure safe and stable operation of the device. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal of a safe operation and maintenance method driven by internal resistance big data according to an embodiment of the present application;

[0018] Figure 2 FIG. 2 is a flowchart of a safe operation and maintenance method driven by internal resistance big data according to an embodiment of the present application;

[0019] Figure 3 FIG. 3 is a structure diagram of a safe operation and maintenance system driven by internal resistance big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0021] Figure 1A hardware structure block diagram of a computer terminal of an internal resistance big data driven safe operation method provided by the embodiment of the present application is shown in the figure. Figure 1 As shown in the figure, the computer device comprises a processor, a memory and a network interface connected through a system bus, wherein the memory can comprise a non-volatile storage medium and an internal memory.

[0022] Referring to Figure 2 The embodiment of the present application provides an internal resistance big data driven safe operation method, which can comprise the following steps:

[0023] S201, collecting historical and real-time internal resistance data of a target device, and combining multi-dimensional running state parameters to generate a set of normalized internal resistance time series data through a dynamic time warping algorithm;

[0024] Specifically, the internal resistance data of the target device can be collected in real time through a distributed data collection node, and at the same time, multi-dimensional running state parameters including at least voltage, current and temperature are collected to generate a multi-source original monitoring data set;

[0025] Firstly, the hardware configuration and deployment logic of the "distributed data collection node" are determined, and the node needs to be deployed at the key position of the target device (for example, a 10kV distribution transformer, which is a common power equipment to be monitored) to ensure that the data covers the core state of the equipment operation. The node selects an edge computing collection terminal (model NIcDAQ-9178, supporting 8 modular input channels, working temperature -40~70℃, suitable for harsh industrial field environment), and a total of 3 nodes are deployed: No. 1 node is installed at the high-voltage side terminal of the transformer, responsible for collecting internal resistance and voltage data; No. 2 node is installed in the low-voltage side outlet cabinet, collecting current data; No. 3 node is installed at the top of the transformer oil tank, collecting temperature data, and the nodes are connected through industrial Ethernet (PROFINET protocol, transmission rate 100Mbps, delay≤10ms) to realize synchronous data transmission.

[0026] The internal resistance data is collected by a high-precision direct current internal resistance tester, and the collection principle is "four-wire measurement method" (eliminating the influence of wire resistance on measurement), and the collection frequency is set to 1 time / minute. Because the internal resistance of the transformer changes slowly with the load and temperature (the daily average change is less than or equal to 0.5mΩ), too high frequency will lead to data redundancy, and 1 time / minute can capture the change rule and control the data volume (1440 pieces of data are collected per day).

[0027] The multi-dimensional operating state parameter acquisition device and parameters are as follows: a Hall voltage sensor is used for voltage acquisition, and the acquisition frequency is 1 time / s (the voltage is greatly affected by the power grid fluctuation, and high-frequency capture of instantaneous changes is required); a closed-loop Hall current sensor (model ACS758-200A, range 0~200A, accuracy ±0.3% FS, error ≤0.6A) is used for current acquisition, and the acquisition frequency is 1 time / s (synchronous with voltage, facilitating subsequent power calculation); a PT100 platinum resistance sensor (model WZP-230, measurement range -200~800℃, accuracy ±0.1℃) is used for temperature acquisition, and the acquisition frequency is 1 time / 10 seconds (temperature change lags behind voltage and current, and 10-second intervals can balance accuracy and efficiency).

[0028] The multi-source original monitoring data set is stored in the format of "time stamp (format YYYYMMDDHHMMSSfff) + device ID + internal resistance (unit mΩ) + voltage (unit V) + current (unit A) + temperature (unit ℃)", and example data is as follows: "20251001080000123, TR-001, 150.2mΩ, 380.5V, 120.3A, 45.2℃" "20251001080001456, TR-001, -, 380.4V, 120.1A, 45.3℃" (where "-" indicates that the internal resistance data has not been collected at this time, and there is a missing value), and the data set is uploaded in real time to the local database (MySQL, storage capacity 1TB, retaining 30 days of original data) of the edge node, forming a multi-source original monitoring data set.

[0029] The multi-source original monitoring data set is subjected to data cleaning and outlier processing, and a sliding window filtering algorithm is used to smooth data fluctuations to generate a preprocessed multi-dimensional monitoring data set.

[0030] Data cleaning needs to solve the problems of "missing values" and "abnormal values" to ensure data integrity and accuracy. The missing value processing adopts a "classification filling strategy": the internal resistance data is missing (as in the above example, the internal resistance at 08:00:01 is "-"), and the "forward filling method" is used to fill in 150.2mΩ at the previous valid time (08:00:00); the voltage and current data are missing (for example, the current is "-" due to sensor instantaneous failure at a certain time), and the "linear interpolation method" is used to take each of the three valid data points (for example, 120.3A at 08:00:00 and 119.9A at 08:00:02) before and after the missing time to calculate the missing value (the current at 08:00:01 is approximately 120.1A); the temperature data is missing, and the "mean filling method" is used to take the mean temperature of the same time period (around 08:00) in the past 7 days (for example, the mean temperature of 08:00 in the past 7 days is 45.1℃) to fill in, and the missing data rate after filling is ≤0.1%.

[0031] The abnormal value processing adopts the "physical threshold method + 3 sigma principle" double judgment: the physical threshold is set based on the rated parameters of the equipment, the normal operating range of the transformer resistance is 140-160 mΩ (lower than 140 mΩ may be due to loose wiring, higher than 160 mΩ may be due to winding aging), the normal voltage range is 360-400 V (exceeding is power grid overvoltage / undervoltage), the normal current range is 100-150 A (exceeding is overload), and the normal temperature range is 40-60°C (exceeding is heat dissipation failure), and exceeding the range is marked as abnormal; the 3 sigma principle is for data within the physical threshold, the mean value μ and the standard deviation σ of nearly 1000 valid data of a certain parameter are calculated, and if the data point deviates from μ by more than 3σ (|x-μ|>3σ), it is determined as random noise abnormality. Example: the resistance of nearly 1000 data μ=150 mΩ, σ=2 mΩ, 3σ=6 mΩ, a data point 157 mΩ (although it is within 140-160 mΩ, but 157-150=7>6), it is determined as abnormal, and the "adjacent mean value replacement method" is adopted-the mean value of the two valid data before and after the data point (such as 150.4 mΩ, 150.2 mΩ, 149.9 mΩ, 150.1 mΩ, the mean value is 150.15 mΩ) is replaced.

[0032] The sliding window filtering algorithm is used to smooth the data fluctuations (such as voltage transient jump caused by electromagnetic interference), and the selection of window size N needs to be combined with the data acquisition frequency: the resistance data acquisition frequency is 1 time / minute, N=5 (the window covers 5 minutes, that is, 5 data points) is selected to balance the smoothing effect and lag; the voltage and current acquisition frequency is 1 time / second, N=10 (the window covers 10 seconds) is selected; the temperature acquisition frequency is 1 time / 10 seconds, N=3 (the window covers 30 seconds) is selected. The filtering formula is "x_filtered(t)=(x(t)+x(t-1)+…+x(t-N+1)) / N", where x(t) is the original data at time t, and x_filtered(t) is the filtered data.

[0033] Based on the pre-processed multi-dimensional monitoring data set, the dynamic time warping algorithm is used to align the resistance data and operating state parameters with different sampling frequencies, eliminate the time scale difference, and generate a time-aligned multi-dimensional data sequence;

[0034] The difference in sampling frequency of different parameters will cause "time scale misalignment" (such as resistance 1 time / minute, voltage 1 time / second, 1 minute resistance only 1 data, voltage has 60 data), which cannot be directly associated and analyzed, and needs to be aligned through the dynamic time warping (Dynamic Time Warping, DTW) algorithm. The core principle of DTW is to find a "minimum cumulative distance" path by constructing a distance matrix of two time series, mapping sequences of different lengths and different frequencies to the same time scale, and not destroying the time order of the sequence.

[0035] Firstly, determine the sequence pairs to be aligned: take the "internal resistance sequence" as the reference sequence (low sampling frequency, as the time anchor point), and take the "voltage sequence", "current sequence", and "temperature sequence" as the sequences to be aligned (high sampling frequency). For example: take the preprocessed data within 10 minutes, the internal resistance sequence T = [t0, t1, …, t9] (10 data points, 1 minute interval), the voltage sequence V = [v0, v1, …, v599] (600 data points, 1 second interval), and align the V sequence to the 10 time points of the T sequence.

[0036] The DTW algorithm performs the following steps: first, construct the distance matrix D, the matrix element d(i, j) is the Euclidean distance between the i-th data of the internal resistance sequence (r_i) and the j-th data of the voltage sequence (v_j), d(i, j) = |r_i-v_j| (since the units are different, the original numerical value is calculated first, and then normalized), in the example d(0, 0) = |150.2-380.5| = 230.3, d(0, 1) = |150.2-380.4| = 230.2; second, calculate the cumulative distance matrix D_acc, D_acc(i, j) = d(i, j) + min(D_acc(i-1, j), D_acc(i, j-1), D_acc(i-1, j-1)), the initial value D_acc(0, 0) = d(0, 0) = 230.3; third, trace back from the lower right corner of D_acc (D_acc(9, 599)) to the upper left corner (D_acc(0, 0)), find the minimum cumulative distance path, the voltage data on this path is the optimal data aligned with the internal resistance data; fourth, for each internal resistance time point t_i, take the arithmetic mean of all voltage data on the path as the aligned voltage value, in the example t0 (08:00) corresponds to v0~v59 (60 voltage data), the mean ≈ 380.48V, which is the aligned voltage value at t0.

[0037] Similarly, the current sequence (600 data points) and the temperature sequence (60 data points, 1 / 10 seconds) are aligned according to the same logic: the current sequence corresponds to the mean of 60 data at each internal resistance time point, and the temperature sequence corresponds to the mean of 6 data at each internal resistance time point. The final time-aligned multi-dimensional data sequence contains "internal resistance + aligned voltage + aligned current + aligned temperature" at each time point (1 minute interval), for example: "20251001080000123, TR-001, 150.2mΩ, 380.48V, 120.22A, 45.23℃" "20251001080100789, TR-001, 150.3mΩ, 380.52V, 120.18A, 45.31℃", the time scales of each parameter in the sequence are completely consistent (all 1 / minute), eliminating the correlation deviation caused by the difference in sampling frequency.

[0038] The normalized multi-dimensional data sequence is time-aligned, and the internal resistance data and each operating state parameter are unified to the same dimension, and finally a normalized internal resistance time series data set is generated.

[0039] The time-aligned multi-dimensional data has a "dimension difference" (such as internal resistance unit mΩ, numerical range 140-160; voltage unit V, numerical range 360-400; current unit A, numerical range 100-150; temperature unit ℃, numerical range 40-60), which will cause "numerical large parameter weight too high" (such as voltage 380V, which is much larger than internal resistance 150mΩ, which will mask the influence of internal resistance change) during subsequent feature extraction, and needs to be normalized to map all parameters to the [0, 1] interval and unify the dimension.

[0040] The Min-Max normalization method is selected, and the normalized calculation example is as follows: take the aligned data point (internal resistance 150.2mΩ, voltage 380.48V, current 120.22A, temperature 45.23℃):

[0041] Internal resistance normalization: (150.2-140) / (160-140)=10.2 / 20=0.51; voltage normalization: (380.48-360) / (400-360)=0.512; current normalization: (120.22-100) / (150-100)=0.4044; temperature normalization: (45.23-40) / (60-40)=0.2615.

[0042] The normalized internal resistance time series data set is stored in the format of "timestamp+device ID+normalized internal resistance+normalized voltage+normalized current+normalized temperature", and the data set needs to meet: the data quantity is consistent with the aligned sequence (1440 per day on average), the mean value of each parameter after normalization is around 0.5 (to avoid data bias), and the standard deviation is between 0.1 and 0.2 (to retain the change characteristics of the parameter), which can be directly used for subsequent multi-scale feature extraction and abnormal pattern recognition.

[0043] S202, multi-scale feature extraction and high-dimensional space mapping are performed on the internal resistance time series data set, an internal resistance feature vector set is constructed, and an adaptive density clustering algorithm is used to identify internal resistance abnormal pattern clusters hidden in the data distribution;

[0044] Specifically, multi-scale wavelet transform can be performed on the normalized internal resistance time series data set to extract time domain and frequency domain features and generate multi-scale feature vectors.

[0045] First of all, it is clear that the specific form of the "normalized internal resistance time series data set" - the set is sampled at 1 minute intervals, each data contains "normalized resistance (such as 0.51), normalized voltage (such as 0.512), normalized current (such as 0.4044), normalized temperature (such as 0.2615)", the data volume is taken as an example of 24-hour running data of a target device (10kV distribution transformer TR-001), a total of 1440, which needs to be captured through multiscale wavelet transform (Multiscale Wavelet Transform) under different time scales. The core advantage is to balance "time domain locality" and "frequency domain globality", which can extract the slow trend of the device resistance over time (such as the daily drift caused by temperature) and the instantaneous fluctuation (such as the short-term change caused by current mutation).

[0046] The db4 wavelet basis (Daubechies-4 wavelet) is selected as the transformation core, because db4 wavelet has 8 support points, balancing between smoothness and computational efficiency - too few support points will lead to rough feature extraction, and too many will increase the computational load; The number of decomposition layers is set to 5, each layer corresponds to a different time scale: 1 layer decomposition corresponds to 2 minutes scale (capture 1-2 minutes of instantaneous fluctuation, such as internal resistance change caused by power grid load mutation), 2 layers correspond to 4 minutes scale (capture 2-4 minutes of medium-term change), 3 layers correspond to 8 minutes scale, 4 layers correspond to 16 minutes scale, and 5 layers correspond to 32 minutes scale (capture 32 minutes or more long-term trend, such as internal resistance drift caused by slow rise of oil temperature), 5 layers of decomposition can cover the main time dimension of device operation, avoiding missing key features.

[0047] The execution process of multiscale wavelet transform needs to be divided into "decomposition-feature extraction" two steps: first, input the 1440 regular data into the wavelet transform module in time order, and perform 5-layer tower decomposition through db4 wavelet basis, to get the approximate coefficients (A1-A5, reflecting low-frequency trend) and detail coefficients (D1-D5, reflecting high-frequency fluctuation) of each layer; The second step is to extract time domain and frequency domain features based on the coefficients, and the features need to cover "trend, fluctuation, energy" three types of core information to ensure comprehensiveness.

[0048] Time domain feature extraction is aimed at each layer of coefficients, and 5 types of indicators are selected, such as "mean, variance, peak, valley, kurtosis", the calculation example is as follows (take A5 layer approximate coefficient as an example, this layer corresponds to 32 minutes long-term trend, there are 45 coefficient values in the range of 0.48-0.55):

[0049] Mean (μ): reflects the trend center level, the calculation formula is μ=(1 / N)×Σx_i (N=45), substituting the data gives μ=(0.48+0.51+...+0.55) / 45≈0.52, which represents the long-term mean of internal resistance at this scale close to the median of the normal interval;

[0050] Variance (σ²): Reflects the degree of dispersion of the trend. The calculation formula is σ²=(1 / N)×Σ(x_i-μ)². The result is σ²≈((0.48-0.52)²+...+(0.55-0.52)²) / 45≈0.0008, which indicates that the long-term trend is stable and the fluctuation is small.

[0051] Peak value (x_max) and valley value (x_min): Reflecting the extreme values ​​of the trend, the peak value of layer A5 is 0.55 and the valley value is 0.48, with a difference of 0.07, which does not exceed the normal fluctuation range (0.1).

[0052] Kurtosis (K): Reflects the steepness of the data distribution. The calculation formula is K=[(1 / N)×Σ(x_i-μ)^4] / σ^4. The calculated value is K≈2.9, which is close to a normal distribution (K=3) and has no extreme outliers.

[0053] The frequency domain features are based on the Fourier transform results of the detail coefficients at each layer. Two types of indicators are selected: "scale energy" and "wavelet entropy". The calculation example is as follows (taking the detail coefficients of layer D1 as an example, corresponding to 2 minutes of high-frequency fluctuations, with a total of 720 coefficients):

[0054] Scale energy (E): Reflects the energy intensity of fluctuations at this scale. The calculation formula is E=Σ|x_i|². Substituting the data, we get E≈0.09 (low high-frequency noise energy, which is consistent with the stable operation characteristics of the equipment).

[0055] Wavelet entropy (S): Reflects the uniformity of energy distribution at each scale. The calculation formula is S=-Σ(p_i×lnp_i) (p_i is the proportion of energy at a certain scale to the total energy). The total energy is the sum of the energies of each layer ≈ 1.62. For layer D1, p_i=0.09 / 1.62≈0.056, and for layer A5, p_i=1.25 / 1.62≈0.772. The calculated value is S≈-(0.772×ln0.772+0.056×ln0.056+...)≈0.41. A low entropy value indicates that the energy is concentrated at the low-frequency scale (long-term trend) and there are no abnormal energy abrupt changes.

[0056] Finally, for each of the 5-layer wavelet decomposition, 7 types of features were extracted: mean, variance, peak value, valley value, kurtosis, energy, and entropy, for a total of 5 × 7 = 35 features. At the same time, considering the multi-parameter correlation, 2 additional cross-parameter features were calculated: internal resistance-temperature correlation coefficient (e.g., 0.68) and voltage-current correlation coefficient (e.g., 0.82), for a total of 37 features, forming a multi-scale feature vector, for example, "[0.52,0.0008,0.55,0.48,2.9,1.25,0.41,0.68,0.82,...]". The first 7 features correspond to layer A5, and the subsequent features correspond to layers A4-D1 and the cross-parameter features, respectively, fully covering the multi-scale feature dimensions of equipment operation.

[0057] The multi-scale feature vector is mapped in a high-dimensional space by using a self-encoder, and a deep feature representation is extracted by the encoder to generate a high-dimensional feature representation vector.

[0058] The "autoencoder" is an unsupervised deep learning model, and its core function is to learn the deep abstract features of data through an "encoder-decoder" structure. In this case, the 37-dimensional multi-scale feature vector is mapped to a 64-dimensional high-dimensional space, with the purpose of amplifying the subtle feature differences (such as the difference between normal patterns and slightly aging patterns), and providing stronger discrimination for subsequent clustering.

[0059] The network structure of the autoencoder needs to match the feature dimension and the mapping target, and the specific design is as follows:

[0060] Input layer (InputLayer): The number of neurons is set to 37, which is consistent with the dimension of the multi-scale feature vector. The activation function uses Linear (linear activation) to avoid the initial features being compressed by nonlinearity and ensure the integrity of the original feature information.

[0061] Encoder (Encoder): It is divided into 3 hidden layers. The first layer (H1) has 128 neurons (to expand the input features and capture low-order correlations), and the activation function is ReLU (formula f(x)=max(0,x), to alleviate the gradient vanishing problem and suitable for non-linear feature learning of industrial data). The second layer (H2) has 64 neurons (target high-dimensional dimension, to realize high-dimensional mapping), and the activation function is still ReLU. The third layer (H3) has 64 neurons (to strengthen the stability of deep features), and the activation function is Tanh (to map the feature values to the interval [-1,1], to reduce the influence of numerical fluctuations).

[0062] Decoder (Decoder): It is symmetrical with the encoder, with H4 (128 neurons, ReLU) and H5 (37 neurons, Linear). The reconstruction error is used to optimize the feature extraction ability of the encoder, and the reconstruction error uses mean square error (MSE, formula MSE=(1 / N)×Σ(y_i-ŷ_i)², y_i is the input feature, and ŷ_i is the reconstructed feature output by the decoder).

[0063] Training parameters: The optimizer is Adam (adaptive moment estimation, faster convergence than SGD, suitable for small batch data), the learning rate is set to 0.001 (too small will lead to too long training period, too large will lead to oscillation), the batch size is set to 32 (balance training efficiency and gradient stability), and the training rounds (Epoch) are set to 100, and the training is stopped when the reconstruction error is lower than 0.005 for 10 consecutive rounds.

[0064] The model training takes 1000 multi-scale feature vectors as the sample set, and the reconstruction error changes during the training process: the initial error is 0.032 (the features are not fully learned), it decreases to 0.009 after 50 rounds, and it stabilizes at 0.004 after 80 rounds (satisfying the convergence condition). At this time, the output of the H3 layer of the encoder is the high-dimensional feature representation vector, with a dimension of 64, and each dimension corresponds to a set of deep correlation features. The example vector is "[0.35, -0.18, 0.42, -0.25,..., 0.53]", where the 3rd dimension (0.42) corresponds to "the synergy of long-term resistance trend and temperature trend" (coefficient 0.72, highest weight), and the 12th dimension (-0.25) corresponds to "the negative impact of current transient fluctuations on resistance" (coefficient -0.61). These deep features are difficult for humans to directly extract, but they can effectively distinguish between device states - for example, the 3rd dimension value is usually between 0.3 and 0.5 during normal operation, while the value rises to 0.6-0.8 when the winding is slightly aged, and the feature difference is significant.

[0065] Based on the high-dimensional feature representation vector, a set of internal resistance feature vectors is constructed, and principal component analysis algorithm is used for feature dimension reduction to retain the main feature components and generate the reduced feature vector set.

[0066] Although the high-dimensional feature representation vector (64 dimensions) has strong discrimination, it has the problem of "dimension disaster" (large amount of calculation and many redundant features), which needs to be reduced through principal component analysis (PCA). The core is to retain the principal components with high cumulative variance contribution rate through linear transformation, ensuring that the core information of the original features is still retained after dimension reduction.

[0067] First, construct the "internal resistance feature vector set" - arrange the 1000 64-dimensional high-dimensional vectors in chronological order to form a 1000x64 matrix, with each row corresponding to a high-dimensional feature of a data point and each column corresponding to a high-dimensional feature dimension. The PCA dimension reduction execution steps are as follows:

[0068] Data centralization: centralize each column (high-dimensional feature dimension) of the matrix, with the formula x_centered=x-μ_j (μ_j is the mean of the jth column), to eliminate the influence of mean difference on covariance calculation. For example, the 3rd column (resistance-temperature synergy feature) mean μ_j=0.41, and the 0.53 data point is centralized to 0.12;

[0069] Calculate the covariance matrix: The covariance matrix has a dimension of 64×64. The element C_ij represents the covariance between the i-th column and the j-th column. The calculation formula is C_ij=(1 / (n-1))×Σ(x_centered_i×x_centered_j) (n=1000 is the number of samples). For example, the covariance between the 3rd and 12th columns is C_3,12≈(1 / 999)×Σ(0.12×(-0.28)+...)≈-0.03, indicating a weak negative correlation between the two (when the current fluctuation is large, the internal resistance-temperature synergy decreases slightly).

[0070] Finding eigenvalues ​​and eigenvectors: Perform eigenvalue decomposition on the covariance matrix to obtain 64 eigenvalues ​​(λ_1≥λ_2≥...≥λ_64) and corresponding eigenvectors (v_1-v_64). The larger the eigenvalue, the more information the corresponding principal component contains. The calculated values ​​are λ_1=13.2, λ_2=10.5, λ_3=7.8, λ_4=5.6, λ_5=4.1, and Σλ_k=46.8 (the sum of the 64 eigenvalues).

[0071] Principal Component Selection: The screening criterion is set as "cumulative variance contribution rate ≥ 85%" (a commonly used industry threshold, balancing information retention and dimensionality reduction). The variance contribution rate of each principal component is calculated as η_j = λ_j / Σλ_k: η_1 = 13.2 / 46.8 ≈ 28.2%, η_2 = 10.5 / 46.8 ≈ 22.4%, η_3 = 7.8 / 46.8 ≈ 16.7%, η_4 = 5.6 / 46.8 ≈ 12.0%, η_5 = 4.1 / 46.8 ≈ 8.8%. The cumulative contribution rate is 28.2% + 22.4% + 16.7% + 12.0% + 8.8% = 88.1% (≥ 85%). Therefore, the first 5 principal components are selected (m = 5).

[0072] Generate dimensionality reduction vectors: Multiply the centered 1000×64 matrix by the 64×5 matrix composed of the first 5 eigenvectors to obtain a 1000×5 dimensionality reduction feature matrix. Each dimensionality reduction vector has a dimension of 5. The calculation formula is PC=X_centered×V (X_centered is the centered matrix, and V is the eigenvector matrix).

[0073] The dimensionality-reduced feature vectors need to explain the physical meaning of each principal component to ensure information traceability:

[0074] The first principal component (PC1): The coefficients of "internal resistance-temperature synergistic feature" and "temperature trend feature" in the eigenvector are 0.75 and 0.65, respectively, indicating that PC1 mainly reflects the "long-term synergistic change of internal resistance and temperature", contributing 28.2% of the information;

[0075] The second principal component (PC2) has a coefficient of 0.68 and -0.62 for "current fluctuation characteristics" and "internal resistance short-term fluctuation characteristics", indicating that PC2 mainly reflects the "short-term impact of current transient changes on internal resistance" and contributes 22.4% of the information.

[0076] The third to fifth principal components correspond to "voltage stability and internal resistance correlation", "synchronization of intraday load cycle and internal resistance", and "noise energy distribution", respectively, and collectively contribute 37.5% of the information.

[0077] An example dimensionality reduction vector is "[0.58, -0.32, 0.29, 0.16, -0.09]", where PC1=0.58 (normal internal resistance and temperature correlation) and PC2=-0.32 (small current fluctuation impact), corresponding to a normal device state. Another vector "[1.32, -0.85, 0.73, 0.54, -0.36]" has PC1=1.32 (far beyond the normal range of 0.8), indicating abnormal internal resistance and temperature correlation, possibly indicating winding aging.

[0078] An adaptive density clustering algorithm is designed to automatically adjust clustering parameters based on data distribution characteristics, and to perform clustering analysis on the reduced feature vector set to generate preliminary clustering results.

[0079] Traditional density clustering algorithms (such as DBSCAN) require manual setting of "neighborhood radius ε" (distance threshold for judging core points) and "minimum core point number MinPts" (minimum number of points in the neighborhood), and improper parameters can easily lead to poor clustering results. The "adaptive density clustering algorithm" designed here automatically optimizes parameters based on data distribution, with the core logic being "K-distance graph determines ε, density distribution determines MinPts".

[0080] 1. Adaptive parameter adjustment

[0081] Automatic determination of ε: Calculate the Euclidean distance between all vectors after dimensionality reduction (formula d(x,y)=√Σ(x_i-y_i)²), for each vector, select the K nearest points (K=5, empirical value is 0.5% of the sample size, 5 for 1000 samples), calculate the K-distance (average of these K distances); Sort all K-distances in descending order and draw a K-distance graph (horizontal axis is sample number, vertical axis is K-distance), the K-distance corresponding to the "slope break point" in the graph is ε. For example, the K-distance of the first 820 samples in the K-distance graph is stable at 0.16-0.27, and the K-distance of the 821st sample jumps to 0.43, the K-distance corresponding to the slope break point is 0.29, so ε=0.29;

[0082] Automatic determination of MinPts: Calculate the local density of each vector (the number of samples in the ε neighborhood), count the density distribution histogram, and take the "number of points corresponding to the density peak" as MinPts. For example, the histogram shows that the number of samples is the most (210) when the local density is 10, indicating that the density of most areas is suitable for MinPts=10, so set MinPts=10.

[0083] 2. Clustering execution process

[0084] The core steps of the adaptive density clustering algorithm are consistent with DBSCAN, but the parameters are automatically optimized values:

[0085] Core point judgment: For the reduced dimension vector x, if the number of samples in its ε neighborhood (distance ≤ 0.29) is ≥ MinPts=10, then x is a core point. For example, a certain vector neighborhood contains 12 samples, which is determined as a core point;

[0086] Density reachable judgment: If point y is in the ε neighborhood of core point x, then y and x are density reachable; if point z is density reachable with y, then z is density reachable with x. For example, the neighborhood of core point x contains y, and the neighborhood of y contains z, so z is density reachable with x;

[0087] Cluster generation: all density-reachable core points and non-core points are grouped into a cluster, and samples with a point count < MinPts in the ε neighborhood are grouped as noise points.

[0088] Taking 1000 reduced dimension vectors as an example, the clustering results are as follows:

[0089] Cluster 1: contains 820 samples, 790 core points, principal component range PC1 ∈ [0.3, 0.8], PC2 ∈ [-0.5, 0.2], PC3 ∈ [-0.3, 0.3], corresponding to the normal operation mode of the device - sample ratio 82%, consistent with the rule of "normal operation time accounts for the majority";

[0090] Cluster 2: contains 130 samples, 105 core points, principal component range PC1 ∈ [0.8, 1.1], PC2 ∈ [-0.8, -0.5], corresponding to the light abnormal mode - PC1 slightly exceeds the normal range, and is suspected to be a slight aging of the winding;

[0091] Cluster 3: contains 40 samples, 28 core points, principal component range PC1 ∈ [1.1, 1.5], PC2 ∈ [-1.2, -0.8], corresponding to the moderate abnormal mode - PC1 is significantly out of range, and is suspected to be a local overheating of the winding;

[0092] Noise points: 10 samples, neighborhood point count < 10, no fixed rule, suspected to be abnormal data caused by transient interference of the power grid (such as lightning, device start-stop).

[0093] The cluster effect verification adopts the "silhouette coefficient" to calculate the silhouette coefficient of cluster 1 as 0.83, the silhouette coefficient of cluster 2 as 0.76, the silhouette coefficient of cluster 3 as 0.69, and the overall average as 0.79 (>0.5, good clustering effect); the distance between clusters (cluster center Euclidean distance): cluster 1-cluster 2=0.68, cluster 2-cluster 3=0.85, cluster 1-cluster 3=1.52, the separation degree is high, and there is no cluster overlap.

[0094] The preliminary clustering result is subjected to abnormal mode recognition, and the cluster obviously separated from other clusters is extracted as the internal resistance abnormal mode cluster, and the internal resistance abnormal mode cluster identification is finally output.

[0095] This step needs to be accurately identified by "reference anchoring-multi-dimensional judgment-screening classification-identification output", to avoid misjudgment and omission, and the core is to establish an "abnormal judgment matrix" to comprehensively quantify the indicators and equipment mechanism.

[0096] 1. Determine the normal cluster reference

[0097] Through "historical data verification" and "cluster feature statistics" two-dimensional anchoring:

[0098] The past 3 months of fault-free data (7200) of the equipment are retrieved, and after being processed in the same process, 98% (7056) are classified into cluster 1, and the main component range of cluster 1 completely coincides with the historical normal data;

[0099] The proportion of cluster 1 samples is 82%, the local density is uniform (variance 0.9), which meets the stability characteristics of the normal operation of the equipment, so cluster 1 is determined as the "normal operation mode cluster".

[0100] 2. Multi-dimensional abnormal judgment

[0101] For cluster 2 and cluster 3, the separation degree, abnormal feature intensity, and fault matching degree are quantified:

[0102] Separation degree: formula S=cluster center distance / average distance within the cluster, S of cluster 2 and cluster 1=0.68 / 0.26≈2.62, S of cluster 3 and cluster 1=1.52 / 0.29≈5.24 (threshold S>3 for significant separation), so cluster 3 is significantly separated, and cluster 2 is moderately separated;

[0103] Abnormal feature intensity: calculate the proportion of principal components exceeding the normal range, cluster 2 PC1 exceeds 70%, PC2 exceeds 65%, and the total exceeds 33%; cluster 3 PC1 exceeds 100%, PC2 exceeds 100%, PC3 exceeds 95%, and the total exceeds 64% (threshold >50% for high intensity abnormality), so cluster 3 has high abnormal intensity, and cluster 2 has moderate abnormal intensity;

[0104] Fault matching degree: Establish an "anomaly cluster-fault mode" association library, and calculate the matching degree using cosine similarity (the closer to 1, the more matching). Cluster 2 matches "winding slight aging" with a matching degree of 0.75, and cluster 3 matches "winding local overheating" with a matching degree of 0.92 (threshold > 0.8 is high matching). Therefore, cluster 3 matches clearly, and cluster 2 matches moderately.

[0105] 3. Screening, grading, and identification output

[0106] Set the comprehensive score = separation degree x 0.4 + abnormal feature intensity x 0.3 + fault matching degree x 0.3. Cluster 2 score = 2.62 x 0.4 + 33% x 0.3 + 0.75 x 0.3 ≈ 1.38 (50 points for 55 points), cluster 3 score = 5.24 x 0.4 + 64% x 0.3 + 0.92 x 0.3 ≈ 2.57 (50 points for 86 points). Grading according to the score: 80 points and above is "clear anomaly cluster", 50-79 points is "potential anomaly cluster". Therefore, cluster 3 is a clear anomaly cluster, and cluster 2 is a potential anomaly cluster.

[0107] The final output is "internal resistance anomaly mode cluster identification", which includes:

[0108] Unique identification: "AM-TR-001-clear anomaly-20251001" (AM=anomaly mode, TR-001=device ID, date is the clustering day);

[0109] Detailed description: "Abnormal features: PC1 ∈ [1.1, 1.5] (internal resistance-temperature cooperation severe anomaly), PC2 ∈ [-1.2, -0.8] (current fluctuation affects dramatically); Possible causes: winding local overheating (insulation layer aging and poor heat dissipation); Impact range: internal resistance may exceed 170 mΩ within 72 hours, causing short circuit risk";

[0110] Risk warning: "Moderate risk, suggest shutdown for maintenance within 72 hours, use infrared thermal imager to detect winding temperature";

[0111] Visual results: Dimensionality reduction vector scatter plot (normal cluster blue, clear anomaly cluster red, potential anomaly cluster yellow), cluster center and range are labeled, and the distribution of anomalies is displayed intuitively.

[0112] S203, based on the internal resistance anomaly mode cluster, perform multi-modal fault correlation analysis, combine device operation history and working condition labels to predict the evolution path and risk level of potential faults, and generate a fault prediction map;

[0113] Specifically, at least the feature statistics of the distribution range, density, and outlying degree of the data points in the cluster can be extracted to generate an anomaly mode feature description;

[0114] This step needs to quantify and extract core statistical quantities reflecting the data distribution characteristics of the cluster from the identified internal resistance abnormal pattern cluster (such as "Cluster 2: slight aging of winding" and "Cluster 3: local overheating of winding" in the previous text), to provide "data fingerprints" for subsequent fault correlation and avoid correlation deviation caused by relying solely on qualitative description. The extraction of feature statistics needs to combine the principal component features (PC1-PC5) after dimension reduction with the device operation mechanism to ensure that each statistical quantity has a clear physical meaning.

[0115] 1. Cluster data point distribution range extraction

[0116] The distribution range reflects the coverage area of the abnormal pattern in the feature space, and needs to calculate "maximum value - minimum value - mean value - standard deviation" for the five principal components (PC1-PC5) after dimension reduction, among which PC1 (long-term coordinated change of internal resistance and temperature) and PC2 (short-term influence of current fluctuation on internal resistance) are the core principal components and need to be highlighted. Taking "Cluster 3: local overheating of winding" (40 samples) as an example:

[0117] PC1: maximum value 1.5, minimum value 1.1, mean value 1.3, standard deviation 0.12, indicating that the abnormal degree of this cluster in the "internal resistance-temperature coordination" dimension is concentrated in 1.1-1.5 (far beyond the normal cluster range of 0.3-0.8), and the sample distribution is compact (small standard deviation), and the abnormal characteristics are stable;

[0118] PC2: maximum value -0.8, minimum value -1.2, mean value -1.0, standard deviation 0.08, reflecting that the negative influence of current fluctuation on internal resistance is significant and concentrated, with no obvious dispersion;

[0119] PC3-PC5: PC3 (voltage stability correlation) mean value 0.7, standard deviation 0.05, PC4 (daily load cycle) mean value 0.52, standard deviation 0.06, PC5 (noise energy) mean value -0.35, standard deviation 0.04, all showing "high abnormal degree and concentrated distribution" characteristics.

[0120] Taking "Cluster 2: slight aging of winding" (130 samples) as an example, the distribution range difference is obvious:

[0121] PC1: maximum value 1.1, minimum value 0.8, mean value 0.95, standard deviation 0.09, abnormal degree lower than Cluster 3, and distribution range narrower (0.8-1.1);

[0122] PC2: maximum value -0.5, minimum value -0.8, mean value -0.65, standard deviation 0.07, current fluctuation influence is weak, consistent with the gradual characteristic of "slight aging".

[0123] The calculation of the distribution range excludes noise points (such as the extreme value of PC1=1.6 in cluster 3, which is determined as noise and not included in the statistics because it is more than 3 times the standard deviation (1.3±0.36) from the mean), ensuring that the results reflect the true abnormal distribution.

[0124] 2. Extraction of data point density within the cluster

[0125] Density reflects the degree of sample aggregation of abnormal patterns, and high density indicates that the abnormal pattern occurs frequently during equipment operation and needs to be prioritized. Two indicators, "local density mean" and "core point proportion", are used for quantification:

[0126] Local density mean: Calculate the average of the local density (ε=0.29 number of samples within the neighborhood) of all samples within the cluster, with the formula "density mean=(1 / N)×∑ρ_i" (N is the number of samples within the cluster, and ρ_i is the local density of the ith sample). Among the 40 samples in cluster 3, the local density ranges from 10 to 15, with an average of 12.5, indicating that the samples are highly aggregated in the feature space, and the frequency of the abnormal pattern is stable; the 130 samples in cluster 2 have a local density range of 8-12, with an average of 9.8, and the aggregation degree is lower than that of cluster 3, which is consistent with the rule that "mild abnormalities occur less frequently";

[0127] Core point proportion: The proportion of the number of core points (local density ≥ MinPts=10) to the total number of samples within the cluster, with the formula "core point proportion=core point number / N×100%". Cluster 3 has 32 core points, accounting for 80% (32 / 40), indicating that the cluster is mainly composed of stable core samples, and the abnormal pattern is reliable; cluster 2 has 91 core points, accounting for 70% (91 / 130), and the core sample proportion is slightly lower, with a small number of edge samples (local density 8-9), which needs to be further verified in combination with the working conditions.

[0128] 3. Extraction of the degree of outlying data points within the cluster

[0129] Outlying degree reflects the deviation amplitude of abnormal patterns from normal patterns, and is quantified by "outlying index", with the formula "outlying index=(Euclidean distance between cluster center and normal cluster center) / (average distance of samples within normal cluster)", and the larger the index, the more serious the deviation:

[0130] Cluster center calculation: Take the mean vector of the principal components of all samples within the cluster, and the center vector of cluster 3 is [1.3,-1.0,0.7,0.52,-0.35], and the center vector of the normal cluster (cluster 1) is [0.55,-0.15,0.02,0.05,-0.03];

[0131] Euclidean distance calculation: the distance d between cluster 3 and the center of the normal cluster is 1.44; the average distance within the normal cluster: the average Euclidean distance between 820 samples in the normal cluster is 0.21; Outlier index: the outlier index of cluster 3 is 1.44 / 0.21≈6.86, which is much larger than the outlier index of cluster 2 (0.81 / 0.21≈3.86), indicating that the abnormal deviation degree of cluster 3 is 1.78 times that of cluster 2, and the fault risk is higher.

[0132] 4. Generate abnormal mode feature description

[0133] Integrate the above statistical quantities to generate a description according to the structure of "cluster identification-core principal component distribution-density characteristics-outlier degree-physical meaning", for example:

[0134] Cluster 3 (explicit anomaly, local overheating of winding): "Cluster identification AM-TR-001-explicit anomaly-20251001; Core principal component distribution: PC1 [1.1, 1.5] (mean 1.3, standard deviation 0.12), PC2 [-1.2, -0.8] (mean -1.0, standard deviation 0.08), reflecting severe abnormality of internal resistance and temperature coordination, and intense negative impact of current fluctuation; Density characteristics: local density mean 12.5, core point proportion 80%, high abnormal mode aggregation, stable occurrence frequency; Outlier index 6.86, significant deviation from normal mode; Physical meaning: aging of winding insulation layer leads to heat accumulation, internal resistance increases at a rate much higher than normal, and short-term local overheating failure is easily caused."

[0135] Cluster 2 (potential anomaly, slight aging of winding): "Cluster identification AM-TR-001-potential anomaly-20251001; Core principal component distribution: PC1 [0.8, 1.1] (mean 0.95, standard deviation 0.09), PC2 [-0.8, -0.5] (mean -0.65, standard deviation 0.07), reflecting slight abnormality of internal resistance and temperature coordination, and weak impact of current fluctuation; Density characteristics: local density mean 9.8, core point proportion 70%, moderate abnormal mode aggregation; Outlier index 3.86, moderate deviation from normal mode; Physical meaning: slight oxidation of winding wire, slow increase of internal resistance, no direct fault risk in short term, but may deteriorate in long term."

[0136] Combine the fault records and working condition labels in the device operation history data, use graph neural network to build a fault association model, analyze the association between abnormal mode and fault type, and generate fault association analysis results;

[0137] This step needs to establish a quantitative correlation between "internal resistance anomaly mode and equipment failure type" to avoid the error of subjective experience matching. Graph Neural Network (GNN) can effectively handle the correlation structure data of "node-edge", which is suitable for capturing the complex mapping relationship between anomaly and failure. Here, Graph Attention Network (GAT) is selected to automatically assign correlation weights through attention mechanism and highlight strong correlation.

[0138] 1. Build the input graph structure of the graph neural network

[0139] The definition of nodes and edges of graph needs to be combined with historical data and working condition labels to ensure that all key elements are covered:

[0140] Node type: including "anomaly mode node" (such as cluster 2, cluster 3), "failure type node" (such as winding mild aging, winding local overheating, winding short circuit, terminal loose), "working condition label node" (such as load peak (current 140-150A), load trough (current 100-110A), high temperature environment (oil temperature ≥50℃), low temperature environment (oil temperature ≤40℃)), a total of 2 types of anomaly nodes, 4 types of failure nodes, 4 types of working condition nodes, a total of 10 nodes;

[0141] Edge definition: edge represents the correlation between nodes, including "anomaly-failure edge" (direct correlation between anomaly mode and failure type), "anomaly-working condition edge" (dependent correlation between anomaly mode and working condition), "failure-working condition edge" (correlation between failure occurrence and working condition), and the initial value of edge weight is based on historical data statistics (such as the proportion of co-occurrence times of cluster 3 and winding local overheating in high temperature environment).

[0142] Taking the historical data of the target device (TR-001) in the past 3 years as an example, a total of 1200 running records are counted, including 28 failures (12 mild aging, 8 local overheating, 5 short circuit, 3 loose wiring), and working condition labels cover 200 days of load peak, 400 days of load trough, 150 days of high temperature environment, and 250 days of low temperature environment. Based on this, the initial edge weight is determined:

[0143] Anomaly-failure edge initial weight: the co-occurrence times of cluster 3 and local overheating is 8 times (cluster 3 appears 10 times, and failure 8 times), initial weight = 8 / 10 = 0.8; the co-occurrence times of cluster 2 and mild aging is 12 times (cluster 2 appears 17 times, and failure 12 times), initial weight = 12 / 17 ≈ 0.71;

[0144] Abnormal-condition edge initial weights: Cluster 3 occurs 9 times under high temperature environment (out of a total of 10 times), weight = 9 / 10 = 0.9; Cluster 2 occurs 10 times under peak load (out of a total of 17 times), weight = 10 / 17 ≈ 0.59.

[0145] 2. Training and Optimization of Graph Neural Networks (GAT)

[0146] The GAT network structure is set as "input layer - attention layer - output layer". Its core is to optimize edge weights through the attention mechanism to improve the accuracy of association.

[0147] Input layer: The feature vector of each node is "node type code + historical statistical features". For example, the feature vector of an abnormal pattern node includes "abnormal type code (cluster 2=01, cluster 3=10) + outlier index + density mean", and the feature vector of a fault node includes "fault type code (local overheating=100) + fault occurrence frequency + average repair time".

[0148] Attention layer: A single-head attention mechanism is adopted to calculate the attention coefficient α_ij of each node to its neighbor nodes. The formula is "α_ij=softmax_j(LeakyReLU(W・[h_i||h_j]))" (W is the weight matrix, h_i and h_j are the feature vectors of nodes i and j, and || means concatenation). The larger the attention coefficient, the stronger the connection between nodes.

[0149] Output layer: Outputs optimized edge weights (attention coefficient × initial weights). The training objective is to minimize the mean squared error between the predicted association result and the actual historical fault association. Training parameters: learning rate 0.001, batch size = 32, training epochs 50. Training stops when the validation set error is below 0.05.

[0150] Example of post-training optimized edge weights:

[0151] Anomaly-Fault Edge: Cluster 3 → Local Overheating Weight 0.92 (0.8 before optimization, due to the strong match between the two in the PC1 anomaly range captured by the attention mechanism), Cluster 3 → Short Circuit Weight 0.68 (indirect correlation, local overheating easily leads to short circuit), Cluster 2 → Slight Aging Weight 0.85 (0.71 before optimization, due to the enhanced correlation under peak load conditions), Cluster 2 → Local Overheating Weight 0.32 (potential correlation, slight aging may worsen).

[0152] Abnormal operating conditions: Cluster 3 → High temperature environment weight 0.95 (0.9 before optimization, high temperature aggravates heat accumulation), Cluster 2 → Peak load weight 0.65 (0.59 before optimization, peak load accelerates conductor oxidation).

[0153] 3. Generate fault correlation analysis results

[0154] The integrated and optimized edge weight is outputted as a result in the structure of "abnormal mode - associated failure - associated working condition - associated strength", for example:

[0155] Cluster 3 (local overheating of winding) associated results: core associated failure: local overheating of winding (associated strength 0.92), secondary associated failure: winding short circuit (associated strength 0.68); key associated working condition: high temperature environment (associated strength 0.95), secondary associated working condition: load peak (associated strength 0.62); associated logic: the PC1 abnormality (serious imbalance of internal resistance-temperature coordination) of cluster 3 is highly matched with the failure feature (winding temperature ≥ 120℃) of local overheating, and in the high temperature environment, the heat cannot be dissipated in time, the probability of abnormal mode transforming into failure is increased by 35%.

[0156] Cluster 2 (mild aging of winding) associated results: core associated failure: mild aging of winding (associated strength 0.85), secondary associated failure: local overheating of winding (associated strength 0.32); key associated working condition: load peak (associated strength 0.65), secondary associated working condition: low temperature environment (associated strength 0.28); associated logic: the mild abnormality (slowly increasing internal resistance) of PC1 of cluster 2 is matched with the failure feature (wire oxide layer thickness ≤ 0.1mm) of mild aging, the current increases at the load peak, the wire heating accelerates the oxidation, and the abnormality may worsen into local overheating within 6 months.

[0157] Based on the failure association analysis results, the hidden Markov model is used to predict the evolution path of potential failure, calculate the occurrence probability of each path, and generate the failure evolution path prediction results;

[0158] The hidden Markov model (HMM) is suitable for processing "state sequence evolution" problems, here the "abnormal mode → failure state → more serious failure" is regarded as a state transition process, the occurrence probability of different evolution paths is predicted by HMM, and the basis is provided for risk assessment. The core elements of HMM include "state set, observation set, transition probability matrix, observation probability matrix, initial state probability", which need to be determined in combination with the failure association results and historical data.

[0159] 1. Define the core elements of HMM

[0160] State set S: contains abnormal state and failure state, a total of 5 states, S1 = cluster 2 (mild aging), S2 = cluster 3 (local overheating), S3 = winding mild aging failure, S4 = winding local overheating failure, S5 = winding short circuit failure;

[0161] Observation set O: based on the feature statistics of abnormal patterns, a total of 3 observations, O1 = outlier index 3-4 (slight abnormal observation), O2 = outlier index 6-7 (severe abnormal observation), O3 = internal resistance > 170 mΩ (fault observation);

[0162] Initial state probability π: based on the proportion of the current abnormal pattern, if the cluster 2 sample accounts for 70% of the abnormal samples and cluster 3 accounts for 30%, then π = [0.7, 0.3, 0, 0, 0] (only abnormal states have initial probabilities, and the initial probability of the fault state is 0);

[0163] Transition probability matrix A: A[i][j] represents the probability of transitioning from state Si to Sj, based on the fault evolution history data of the past 3 years (in 28 faults, 5 of the 12 slight aging deteriorated to local overheating, and 3 of the 8 local overheating deteriorated to short circuit):

[0164] A[S1→S1]=0.35 (maintain cluster 2 abnormality), A[S1→S3]=0.48 (cluster 2 turns into slight aging failure), A[S1→S2]=0.17 (cluster 2 deteriorates to cluster 3 abnormality); A[S2→S2]=0.30 (maintain cluster 3 abnormality), A[S2→S4]=0.65 (cluster 3 turns into local overheating failure), A[S2→S5]=0.05 (cluster 3 directly turns into short circuit failure); A[S3→S3]=0.60 (maintain slight aging failure), A[S3→S4]=0.40 (slight aging deteriorates to local overheating failure); A[S4→S4]=0.70 (maintain local overheating failure), A[S4→S5]=0.30 (local overheating deteriorates to short circuit failure); A[S5→S5]=1.0 (short circuit failure is the terminal state, no transition); Observation probability matrix B: B[i][k] represents the probability of state Si generating observation Ok, calculated based on the matching degree of feature statistics and state:

[0165] B[S1→O1]=0.90 (cluster 2 corresponds to slight abnormal observation), B[S1→O2]=0.10 (cluster 2 occasionally appears severe observation); B[S2→O2]=0.95 (cluster 3 corresponds to severe abnormal observation), B[S2→O3]=0.05 (cluster 3 occasionally appears fault observation); B[S3→O1]=0.85 (slight aging failure corresponds to slight observation), B[S3→O2]=0.15 (slight aging occasionally appears severe observation); B[S4→O2]=0.70 (local overheating failure corresponds to severe observation), B[S4→O3]=0.30 (local overheating appears fault observation); B[S5→O3]=1.0 (short circuit failure corresponds to fault observation).

[0166] 2. Predicting fault evolution path and occurrence probability

[0167] The forward algorithm is used to calculate the occurrence probability of each state sequence (evolution path), focusing on the path from the initial abnormal state to the final fault state:

[0168] Path 1: S1 (Cluster 2) → S3 (Mild Aging Fault) → S4 (Local Overheating Fault) → S5 (Short Circuit Fault);

[0169] Path 2: S1 (Cluster 2) → S3 (Mild Aging Fault) → S3 (Maintain Mild Aging);

[0170] Path 3: S2 (Cluster 3) → S4 (Local Overheating Fault) → S4 (Maintain Local Overheating);

[0171] Path 4: S2 (Cluster 3) → S4 (Local Overheating Fault) → S5 (Short Circuit Fault);

[0172] Path 5: S1 (Cluster 2) → S1 (Maintain Cluster 2) → S1 (Maintain Cluster 2).

[0173] 3. Generate Fault Evolution Path Prediction Results

[0174] Output the results in the structure of "Path Priority (Probability from High to Low) - Path Description - Occurrence Probability - Evolution Period", and highlight the high probability paths (probability > 5%):

[0175] Priority 1: Path S1 → S3 → S3 (Maintain Mild Aging Fault), occurrence probability 14.56%, evolution period 6-12 months, characteristics: Cluster 2 abnormality transforms into mild aging fault, and maintains for a long time without obvious deterioration;

[0176] Priority 2: Path S1 → S1 → S1 (Maintain Cluster 2 Abnormality), occurrence probability 6.95%, evolution period 3-6 months, characteristics: Cluster 2 abnormality does not transform, only appears slight fluctuations during peak load;

[0177] Priority 3: Path S2 → S4 → S4 (Maintain Local Overheating Fault), occurrence probability 2.87%, evolution period 1-3 months, characteristics: Cluster 3 abnormality quickly transforms into local overheating fault, intervention within 72 hours is required;

[0178] Priority 4: Path S1 → S3 → S4 → S5 (Mild Aging → Local Overheating → Short Circuit), occurrence probability 2.4%, evolution period 12-18 months, characteristics: Abnormality gradually deteriorates, short circuit risk increases over time;

[0179] Supplementary Note: Path S2 → S4 → S5 (Local Overheating → Short Circuit) probability 1.76%, although the probability is low, but the evolution period is short (within 1 month), which needs to be closely monitored.

[0180] According to the fault evolution path prediction result, the risk level of each path is evaluated, considering the fault severity and occurrence probability, to generate the risk level evaluation result;

[0181] The risk level evaluation needs to consider both the "destructiveness of the fault (severity)" and the "occurrence probability (probability)", to avoid misjudgment caused by focusing on a single dimension (such as the risk balance between high-probability minor faults and low-probability serious faults). The evaluation process consists of four steps: "defining severity classification-quantifying occurrence probability-calculating risk value-dividing risk level", to ensure that the results are quantifiable and comparable.

[0182] 1. Define the severity classification of the fault

[0183] Based on the impact of the fault on the device operation, repair cost and downtime, the severity is divided into 5 levels (1 is the lowest and 5 is the highest), each level corresponds to specific judgment criteria and quantitative indicators:

[0184] Level 1 (no risk): no fault, only abnormal mode, no downtime, repair cost 0 yuan, such as maintaining cluster 2 anomaly (path 5); Level 2 (minor risk): minor fault, no need to stop, repair cost ≤5000 yuan, downtime 0 hours, such as minor aging fault (path 2); Level 3 (mild risk): moderate fault, need to plan downtime (1-4 hours), repair cost 5000-20000 yuan, such as local overheating fault (path 3); Level 4 (moderate risk): serious fault, need to stop urgently (within 72 hours), repair cost 20000-50000 yuan, such as local overheating fault with short circuit risk (path 4); Level 5 (emergency risk): fatal fault, need to stop immediately (within 1 hour), repair cost >50000 yuan, may cause safety accidents, such as short circuit fault (end of path 1).

[0185] Taking the end state of each path as an example, the severity is marked as follows:

[0186] Path 2 end S3 (minor aging): level 2; path 5 end S1 (cluster 2): level 1; path 3 end S4 (local overheating): level 3; path 1 end S5 (short circuit): level 5; path 4 end S5 (short circuit): level 5.

[0187] 2. Quantify the fault occurrence probability interval

[0188] Combined with the path occurrence probability, the probability is divided into 3 intervals for risk calculation:

[0189] High probability: P>10%, corresponding to "extremely likely to occur"; medium probability: 3%≤P≤10%, corresponding to "likely to occur"; low probability: P<3%, corresponding to "occasionally occurs".

[0190] Mark the probability interval of each path:

[0191] Path 2 (14.56%): High probability; Path 5 (6.95%): Medium probability; Path 3 (2.87%): Low probability; Path 1 (2.4%): Low probability; Path 4 (1.76%): Low probability.

[0192] 3. Calculate risk value and classify risk level

[0193] The risk value is calculated using the formula: "Risk Value = Severity × Probability Weight × Risk Coefficient". The "Probability Weight" is assigned a range (high probability = 1.0, medium probability = 0.6, low probability = 0.3), and the "Risk Coefficient" is the equipment importance coefficient (target equipment TR-001 is a critical power distribution equipment, coefficient = 1.2; non-critical equipment coefficient = 1.0). A higher risk value indicates a higher risk level.

[0194] Path 2 (S3, Level 2, High Probability): Risk Value = 2 × 1.0 × 1.2 = 2.4; Path 5 (S1, Level 1, Medium Probability): Risk Value = 1 × 0.6 × 1.2 = 0.72; Path 3 (S4, Level 3, Low Probability): Risk Value = 3 × 0.3 × 1.2 = 1.08; Path 1 (S5, Level 5, Low Probability): Risk Value = 5 × 0.3 × 1.2 = 1.8; Path 4 (S5, Level 5, Low Probability): Risk Value = 5 × 0.3 × 1.2 = 1.8.

[0195] Based on the risk value, four risk levels are defined, each corresponding to a different response priority:

[0196] Low risk: Risk value < 1.0, priority 4 (no immediate action required, monthly monitoring), such as path 5 (0.72); Lower risk: 1.0 ≤ risk value < 2.0, priority 3 (quarterly assessment, action as needed), such as path 3 (1.08), path 1 (1.8), path 4 (1.8); Medium risk: 2.0 ≤ risk value < 3.0, priority 2 (monthly assessment, planned action), such as path 2 (2.4); High risk: Risk value ≥ 3.0, priority 1 (immediate action, intervention within 24 hours), no path matches.

[0197] 4. Generate risk level assessment results

[0198] The output should follow the structure of "Path-Destination Status-Severity-Probability Range-Risk Value-Risk Level-Response Priority", as shown in the example below:

[0199] Path 2 (S1→S3→S3): End point state is slight aging fault, severity level 2, probability range is high probability (14.56%), risk value is 2.4, risk level is medium risk, response priority is 2; Recommendation: increase internal resistance monitoring once a month, conduct winding insulation test once every 3 months, and develop a planned shutdown maintenance plan within 6 months to avoid deterioration.

[0200] Path 5 (S1→S1→S1): End state cluster 2 anomaly, severity level 1, probability interval medium probability (6.95%), risk value 0.72, risk level low risk, response priority 4; recommendation: maintain regular monitoring frequency (1 time / minute), quarterly aggregate data, no additional intervention needed.

[0201] Path 3 (S2→S4→S4): End state local overheating fault, severity level 3, probability interval low probability (2.87%), risk value 1.08, risk level lower risk, response priority 3; recommendation: complete infrared thermal imaging detection within 72 hours, evaluate winding temperature distribution, if temperature ≥ 120℃, immediately schedule emergency shutdown.

[0202] Path 1 (S1→S3→S4→S5) and Path 4 (S2→S4→S5): End state short circuit fault, severity level 5, probability interval low probability (2.4% / 1.76%), risk value 1.8, risk level lower risk, response priority 3; recommendation: check winding insulation resistance once a week, if insulation resistance < 100MΩ, immediately shut down to prevent short circuit accidents.

[0203] Integrate the results of fault correlation analysis, fault evolution path prediction, and risk level evaluation to construct a multi-dimensional fault prediction map, and finally output a complete fault prediction map.

[0204] The multi-dimensional fault prediction map needs to visualize the integration of "correlation- evolution path- risk level" three types of information, adopt "hierarchical structure" design, ensure that operation and maintenance personnel can intuitively understand the fault correlation, possible evolution direction and risk level of abnormal patterns, and the map needs to include "basic information area- correlation analysis area- path prediction area- risk assessment area" four modules, each module needs to mark data source and confidence level to ensure the results are traceable.

[0205] 1. Basic information area of the map

[0206] Located at the top of the map, mark the core basic information to ensure the context is clear:

[0207] Target device information: device ID (TR-001), device type (10kV distribution transformer), monitoring period (20251001-20251002), data source (real-time monitoring data + 3 years of historical fault records);

[0208] Abnormal pattern overview: current number of identified abnormal patterns (2: cluster 2, cluster 3), abnormal sample proportion (140 abnormal samples, accounting for 9.7% of the total 1440 samples), data confidence (feature extraction confidence 0.92, clustering confidence 0.89).

[0209] 2. Map correlation analysis area

[0210] Located on the left side of the map, the graphical description of "node-weighted edge" (text instead of graphics) shows the correlation between abnormal patterns and faults, working conditions:

[0211] Abnormal pattern node: Mark "Cluster 2 (mild aging, outlier index 3.86)" "Cluster 3 (local overheating, outlier index 6.86)" with a red box, and mark the core feature statistics below the node;

[0212] Fault node: Mark "mild aging fault (repair cost 5k)" "local overheating fault (repair cost 20k)" "short circuit fault (repair cost 50k)" with a blue ellipse, and mark the severity below the node;

[0213] Working condition node: Mark "high load peak (140-150A)" "high temperature environment (≥50℃)" with a green diamond, and mark the correlation frequency below the node;

[0214] Weighted edge: Use "abnormal node→fault node (weight value)" "abnormal node→working condition node (weight value)" to describe, such as "Cluster 2→mild aging fault (0.85)" "Cluster 3→high temperature environment (0.95)", the greater the weight value, the more prominent the edge description (such as bold font).

[0215] 3. Map path prediction area

[0216] Located in the middle of the map, the evolution paths are arranged vertically according to "path priority", each path is described with "arrow sequence + probability", high probability paths are marked in orange, and low probability paths are marked in gray:

[0217] Priority 1 (medium risk): Cluster 2→mild aging fault→maintain mild aging (probability 14.56%), mark the evolution period "6-12 months"; Priority 2 (low risk): Cluster 2→maintain Cluster 2 anomaly (probability 6.95%), mark the evolution period "3-6 months"; Priority 3 (lower risk): Cluster 3→local overheating fault→maintain local overheating (probability 2.87%), mark the evolution period "1-3 months"; Priority 4 (lower risk): Cluster 2→mild aging→local overheating→short circuit (probability 2.4%), mark the evolution period "12-18 months"; Priority 5 (lower risk): Cluster 3→local overheating→short circuit (probability 1.76%), mark the evolution period "within 1 month".

[0218] 4. Map risk assessment area

[0219] Located on the right side of the map, it is arranged horizontally according to "risk level", each risk level corresponds to a path and a response suggestion, and the risks are distinguished by color (low risk green, lower risk yellow, medium risk orange):

[0220] Low risk (green): Path 5 (risk value 0.72), recommendation "routine monitoring, quarterly summary"; Lower risk (yellow): Path 3, 1, 4 (risk value 1.08-1.8), recommendation "infrared detection, weekly insulation check"; Medium risk (orange): Path 2 (risk value 2.4), recommendation "monthly monitoring, 6-month planned downtime"; Risk summary: Current overall risk level of the device "low-medium risk", no high-risk path, critical intervention nodes "within 72 hours (cluster 3), within 6 months (cluster 2)".

[0221] 5. Output complete failure prediction map

[0222] The bottom of the map is marked with "map update frequency (updated automatically every 24 hours)" "data deviation range (probability deviation ± 3%, risk value deviation ± 0.2)" "contact information of operation and maintenance person in charge", to ensure the timeliness and operability of the map. The final output map can be displayed through the HMI interface of the DCS system, supporting clicking on the node to view detailed data (such as clicking on the cluster 3 node, the characteristic statistics and historical failure cases of the cluster are popped up), providing intuitive and comprehensive decision basis for subsequent operation and maintenance strategy formulation.

[0223] S204, generating a set of differentiated operation and maintenance strategies according to the failure prediction map, and driving the operation and maintenance management platform to execute corresponding operation and maintenance instructions.

[0224] Specifically, the risk level and failure type information in the failure prediction map can be parsed, and the corresponding basic operation and maintenance strategy is matched from the operation and maintenance strategy library to generate a preliminary operation and maintenance strategy set;

[0225] First, the core decision information of the failure prediction map needs to be completely extracted. The map clearly marks two types of key abnormal patterns: one is "cluster 2 (light aging of winding)", the failure type is "slow rise of internal resistance caused by oxidation of winding wire", the risk level is "medium risk", the evolution period is 6-12 months, and the key associated working condition is "high load peak (current 140-150A)"; the second is "cluster 3 (local overheating of winding)", the failure type is "heat accumulation caused by insulation layer aging", the risk level is "low risk", the evolution period is 1-3 months, and the key associated working condition is "high temperature environment (oil temperature ≥ 50℃)". The parsing process needs to ensure that the key constraint "associated working condition" is not missed - the development speed of the same failure under different working conditions is significantly different (such as the monthly change of internal resistance of cluster 2 is 1.5 times that of low load valley), which needs to be used as an important basis for subsequent strategy matching.

[0226] The operation and maintenance policy library needs to be organized according to the three-dimensional structure of "risk level-fault type-working condition label" to ensure the accuracy of matching. Each basic strategy in the library contains four core contents: "monitoring frequency, detection item, warning threshold, basic treatment measures", and the parameters need to be set based on industry standards and historical operation and maintenance experience of the equipment:

[0227] The basic strategy for the "medium risk-winding slight aging-load peak" scenario (strategy number S-002): ① Monitoring frequency: 1-time resistance special monitoring per month (sampling frequency 1 / time per minute, lasting 24 hours), 1-time routine operating parameter (voltage, current, temperature) monitoring per week; ② Detection item: 1-time winding insulation resistance test every 3 months (using KEITHLEY2450 tester, range 0-1000MΩ, accuracy ±0.01MΩ, threshold ≥100MΩ), 1-time infrared thermal imaging detection every 6 months (FLIRT650sc thermal imager, resolution 640×480, temperature measurement range -20~120℃, winding temperature threshold ≤110℃); ③ Warning threshold: trigger SMS warning when the monthly change of resistance is >0.5mΩ, trigger audible and light alarm when the insulation resistance is <100MΩ; ④ Basic treatment: arrange operation and maintenance personnel to review data within 24 hours after warning to exclude measurement errors.

[0228] The basic strategy for the "low risk-winding partial overheating-high temperature environment" scenario (strategy number S-008): ① Monitoring frequency: 1-time resistance monitoring every 2 weeks, 1-time oil temperature monitoring every day (sampling frequency 1 / time per 10 minutes); ② Detection item: 1-time infrared thermal imaging detection every 1 month (focus on scanning the middle area of the winding, temperature threshold ≤120℃), 1-time cooling system function check every 2 months (fan speed ≥2600rpm, no dust accumulation on the cooling fins); ③ Warning threshold: warning when the oil temperature is ≥55℃ for 1 hour or the weekly change of resistance is >0.3mΩ; ④ Basic treatment: clean the cooling system within 48 hours after warning, and replace the fan filter if necessary.

[0229] The strategy matching adopts "double keyword search": first, locate the candidate strategy with "risk level + fault type" (such as "medium risk + slight aging" locates S-002, S-003), and then select the optimal strategy with "associated working condition" (such as cluster 2 is associated with "load peak", exclude S-003 which is suitable for "load trough", and determine S-002). The final generated preliminary operation and maintenance strategy set contains 2 core strategies (S-002, S-008) and 1 supplementary strategy (S-015, for "multiple abnormal mode superposition" scenarios, such as cluster 2 and cluster 3 appear at the same time, preferentially handle partial overheating), and the set needs to be labeled with "applicable scenario, effective time, confidence" of each strategy (such as S-002 confidence 0.92, based on the operation and maintenance effect statistics of 12 similar faults in the past 3 years), to ensure a clear basis for subsequent optimization.

[0230] Based on the real-time running state of the equipment and the historical operation and maintenance effect data, the reinforcement learning algorithm is used to optimize and adjust the preliminary operation and maintenance strategy set, and generate the optimized operation and maintenance strategy;

[0231] The core advantage of reinforcement learning (RL) is to dynamically adapt to the change of equipment state through the "trial and error-reward" mechanism. Here, the deep deterministic policy gradient (DDPG) algorithm is selected because it is suitable for processing continuous action space (such as continuous adjustment of monitoring frequency and early warning threshold), and avoids the parameter jump problem of discrete algorithms. Algorithm design needs to clearly define the three core elements of "agent, environment, reward function" to ensure that the optimization direction is consistent with the operation and maintenance goals (fault prevention, cost control, energy saving).

[0232] 1. Core element definition

[0233] Agent (Agent): "Strategy optimization module" deployed in the operation and maintenance management system, the core function is to output adjusted strategy parameters (such as monitoring frequency, early warning threshold), and the input is the environment state feedback.

[0234] Environment (Environment): Contains real-time running state of equipment and historical operation and maintenance effect data. Real-time state parameters include: current 142A (high load peak interval), oil temperature 48℃ (close to high temperature threshold), internal resistance 152.3mΩ (0.3mΩ higher than last month); historical effect data includes: the results of the past 10 times of executing S-002 strategy (8 times meet the standard, 2 times trigger early warning due to current exceeding 145A leading to internal resistance change exceeding 0.5mΩ, average disposal cost 2000 yuan after early warning), energy consumption data under different monitoring frequencies (5kWh per month, 8kWh every 2 weeks, 15kWh every week).

[0235] Reward function (Reward Function): Comprehensive measure of "fault prevention effect, operation and maintenance cost, energy consumption", formula is R=0.4×R_fault+0.3×R_cost+0.3×R_energy, where:

[0236] R_fault (fault prevention score): If the strategy is executed without triggering early warning, 100 points; Trigger early warning but no fault occurs, 80 points; Fault occurs, 0 points (such as cluster 2 if not monitored in time leading to aging deterioration, 0 points);

[0237] R_cost (Cost Control Score): Based on labor and equipment costs, the benchmark cost is 2000 yuan (S-002 single execution cost), and 15 points are awarded for every 10% reduction, and 10 points are deducted for every 10% increase (e.g. if the cost is reduced to 1800 yuan after optimization, 15 points are awarded).

[0238] R_energy (Energy Saving Score): The benchmark energy consumption is 5 kWh / month, and 20 points are awarded for every 1 kWh reduction, and 10 points are deducted for every 1 kWh increase (e.g. if the energy consumption is 6 kWh after optimization, 10 points are deducted).

[0239] 2. Optimization process and results

[0240] The algorithm training iterates 30 times, with the initial strategy being the original parameters of S-002 and S-008. At the 10th iteration, the key rule is found: when the current is greater than 140A, the daily change of internal resistance of cluster 2 (0.02mΩ / day) is 1.5 times that of when the current is less than or equal to 140A (0.013mΩ / day), and the temperature deviation detected by infrared will increase by 5% (due to current heating interference). Based on this, the agent begins to adjust the parameters:

[0241] Optimization of S-002: ① Monitoring frequency: When the current is greater than 140A, the internal resistance monitoring frequency is adjusted from "once a month" to "once every 2 weeks", and when the current is less than or equal to 140A, it remains once a month; The regular parameter monitoring is changed from "once a week" to "once a day" (only when the current is greater than 140A); ② Warning threshold: The internal resistance monthly change is tightened from 0.5mΩ to 0.4mΩ (due to increased risk during current peak), and the insulation resistance threshold remains 100MΩ (not affected by current); ③ Detection items: In high temperature environment (oil temperature ≥ 50℃), the infrared detection period is shortened from 6 months to 4 months (to avoid heat accelerated aging).

[0242] Optimization of S-008: ① Monitoring frequency: When the oil temperature is greater than or equal to 50℃, the oil temperature monitoring is changed from "once a day" to "once every 2 hours", and the internal resistance monitoring is changed from "once every 2 weeks" to "once every 1 week"; ② Cooling system inspection: The fan speed threshold is increased from 2600rpm to 2700rpm (stronger heat dissipation is needed in high temperature environment), and the cleaning period is shortened from 2 months to 1 month.

[0243] The reward function score of the optimized strategy increased from 75 (S-002) and 72 (S-008) to 92 (optimized S-002) and 89 (optimized S-008). The reasons for the increase are as follows: ① R_fault increased from 80 to 100 (no pre-warning after 2 rounds of adjustment); ② R_cost decreased from 85 to 80 (increased monitoring frequency led to a 5% increase in cost), but R_energy increased from 60 to 90 (increased frequency only in high-risk conditions, and maintained low energy consumption in low-risk conditions), resulting in significant overall benefits. The final output of the optimized strategy needs to be labeled with "adjustment basis" (such as "shorten the monitoring period when the current is greater than 140A, based on historical data showing a 50% increase in aging speed under this condition"), to ensure traceability.

[0244] According to the optimized operation and maintenance strategy, specific executable operation and maintenance instructions are generated, including maintenance time, maintenance content, and required resources, to obtain a set of operation and maintenance instructions.

[0245] Executable instructions need to convert the "abstract parameters" of the optimized strategy into "concrete operation tasks". Each instruction must meet the three requirements of "time feasibility, clear content, and resource allocation" to avoid ambiguous expressions (such as "repair as soon as possible" needs to be specified as "2025-10-12 02:00-04:00"). Instruction generation needs to consider device production plans (to avoid affecting normal power supply), resource availability (such as device calibration status and personnel scheduling), and safety specifications (such as high-voltage device maintenance requiring power outage permission).

[0246] 1. Instructions for optimized S-002 (cluster 2: slight winding aging) (instruction ID: Cmd-TR001-20251005-001)

[0247] Maintenance time: October 12, 2025, 02:00-04:00 (choose the low-usage period in the early morning of the weekend when the average load of the device is only 30% of the peak, the impact on production is minimal after shutdown; submit a power outage application 72 hours in advance to ensure permission from power dispatching);

[0248] Inspection content: ①Winding insulation resistance test: use KEITHLEY2450 insulation resistance tester, disconnect the high voltage side power supply before testing, ground discharge for 10 minutes; test points are high voltage side A, B, C three-phase terminals and ground terminal, each phase test time is 1 minute, record data (require each phase ≥100MΩ, three-phase deviation ≤5%); ②Infrared thermal imaging detection: use FLIRT650sc thermal imager, lens distance from winding surface is 1.5 meters, scanning range covers the entire winding area (including joints, leads), mark areas with temperature >100℃, record the highest temperature (require ≤110℃); ③Internal resistance retest: use the same type of internal resistance tester, measure the internal resistance of phase A on the high voltage side, compare with the same period data last month (152.0mΩ), calculate the change (require ≤0.4mΩ); ④Data archiving: record test data into equipment operation and maintenance archives (system number archives-TR001-202510), generate monthly operation and maintenance report.

[0249] Required resources: ①Equipment: KEITHLEY2450 tester 1, FLIRT650sc thermal imager 1, high voltage electroscope 1 set; ②Personnel: 2 certified high voltage electricians, 1 detection engineer; ③Consumables: anhydrous ethanol, insulating gloves; ④Time allocation: power off for 10 minutes, insulation test for 30 minutes, infrared detection for 40 minutes, internal resistance retest for 20 minutes, data archiving for 20 minutes, total time length is 120 minutes (reserve 20 minutes for unexpected situations).

[0250] 2. Instructions for optimized S-008 (cluster 3: winding local overheating) (instruction ID: Cmd-TR001-20251005-002)

[0251] Maintenance time: October 08, 2025 01:00-03:00 (local overheating risk is high, shorten the execution cycle to 1 month; choose Wednesday morning to avoid weekend maintenance peak and ensure sufficient resources);

[0252] Inspection content: ①Oil temperature and internal resistance monitoring: continuously monitor oil temperature for 30 minutes (record every 5 minutes, require ≤55℃), simultaneously measure internal resistance (require weekly change ≤0.3mΩ); ②Cooling system inspection: remove cooling fan filter, clean dust with compressed air (pressure 0.4MPa), test fan speed (use Instr-012 speed tester, range 0-5000rpm, require ≥2700rpm); ③Winding temperature scanning: infrared thermal imager scans the middle part of the winding (local overheating high-risk area), records temperature distribution (require maximum temperature ≤120℃), if temperature >115℃, apply thermal conductive silicone grease (model ST-80, thermal conductivity 1.2W / (m・K)); ④System debugging: restart the cooling system, monitor for 10 minutes, confirm that the fan is running normally and the oil temperature is not abnormally high.

[0253] Required resources: On the basis of S-002 resources, add ① 1 speed tester, ② 1 compressed air machine, ③ 1 heat-conducting silicone grease; add 1 mechanical maintenance worker, total duration 120 minutes.

[0254] 3. Instruction set integration and verification

[0255] The final operation and maintenance instruction set contains the above 2 instructions, which need to pass the "feasibility verification": ① Time verification: confirm that there is no other equipment maintenance plan during the maintenance period, and the power dispatching has approved the power-off permit; ② Resource verification: all equipment is within the calibration validity period, and the personnel scheduling has no conflict; ③ Safety verification: the maintenance process meets the existing specifications, such as complete power-off, discharge, and electric verification steps. The instruction set is stored in XML format, and each instruction contains "ID, device ID, execution time, content, resource, priority, and responsible person" fields, which is convenient for subsequent API transmission.

[0256] Through the API interface, the operation and maintenance instruction set is sent to the operation and maintenance management platform, which automatically executes the corresponding operation and maintenance tasks and monitors the execution status in real time, finally completing the execution and feedback of the operation and maintenance instruction.

[0257] The API interface is the core channel connecting the "strategy generation system" and the "operation and maintenance management platform", which needs to ensure the safety, real-time performance, and reliability of the transmission. This place adopts the RESTful API architecture (in line with the general standard of industrial systems), combined with Token authentication and data encryption, to avoid tampering or leakage of instructions. The operation and maintenance management platform selects Siemens OpcenterExecution (industrial operation and maintenance management system), which supports automatic generation of work orders, real-time tracking of status, and data archiving.

[0258] 1. API interface design and instruction sending

[0259] Interface parameters: the request method is POST, and the request header contains: ① Authorization: Bearer Token (validity period is 24 hours, Token format is eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9...) before each transmission, Token is obtained by logging in the interface); ② Content-Type: application / xml (adapt to the platform data format); ③ Timeout: 30s (ensure that the timeout retransmission mechanism takes effect). The request body is the instruction set XML data.

[0260] Sending and confirmation mechanism: after the strategy generation system sends the instruction, the platform returns "reception confirmation" (status code 200OK, containing the reception timestamp) within 5 seconds; if the confirmation is not received within the timeout, the system automatically retransmits (up to 3 times, interval 10s), and if the retransmission fails, an email alert is triggered to the operation and maintenance supervisor.

[0261] 2. Task execution and status monitoring of the operation and maintenance platform

[0262] Work order generation: After receiving the instructions, the platform automatically parses the XML data and generates two operation and maintenance work orders (work order number consistent with instruction ID), assigns them to "Operation and Maintenance Group-G03", and pushes them to relevant personnel through the system APP (e.g., electrician-012 receives "Work Order Cmd-TR001-20251005-001 will be executed on October 12, 02:00, please prepare the equipment").

[0263] Real-time status tracking: The platform updates the task progress in real time through "status code". Status code definition: ① 00 (preparation pending): work order has been assigned, but personnel have not confirmed; ② 01 (in execution): personnel have arrived on site and started maintenance; ③ 02 (data uploading): maintenance is complete, and test data is being uploaded; ④ 03 (completed): data is archived, and work order is closed; ⑤ 99 (abnormal): problems occur during execution (e.g., device failure, data exceeding standard). The status update frequency is 1 minute / second, and the strategy generation system can query in real time through API.

[0264] Abnormal handling: If an abnormality occurs during execution (e.g., the insulation test data of Cmd-TR001-20251005-001 is 95MΩ < 100MΩ), the platform automatically triggers the "abnormal process": ① suspend the work order and send an alert to the detection engineer; ② the engineer reviews the data (rules out test instrument error), confirms the abnormality, and generates a "supplementary instruction" (Cmd-TR001-20251005-003, content: "replace winding insulation material, execute within 24 hours"), priority is raised to 1; ③ the supplementary instruction is sent according to the same API process, and the original work order is closed after the supplementary instruction is completed.

[0265] 3. Execution feedback and data archiving

[0266] Feedback content: After completing the work order (status code 03), the platform generates an "execution report" containing: ① execution results (e.g., "insulation resistance 125MΩ, qualified; infrared maximum temperature 105℃, qualified"); ② resource consumption (device usage time 120 minutes, personnel working hours 3 person・hours); ③ abnormality record (none / has, including handling process). The report is fed back to the strategy generation system through API and archived to the device operation and maintenance archives (associated with historical data, such as comparing the resistance change trend last month).

[0267] Effect evaluation: The strategy generation system updates the historical effect library of reinforcement learning based on feedback data (e.g., after the execution of Cmd-TR001-20251005-001, the resistance monthly change of cluster 2 is reduced to 0.2mΩ, and the reward function score is increased by 5 points again), providing data support for the next round of strategy optimization.

[0268] Through the above process, the closed loop of "instruction sending-execution-monitoring-feedback" is realized, the operation and maintenance task is ensured to be accurately landed, and finally the safe operation and maintenance target of "preventing faults, reducing costs and improving equipment reliability" is achieved.

[0269] Another embodiment of the application provides a large data driven safe operation and maintenance system of internal resistance, referring to Figure 3 , the system can include:

[0270] The acquisition module 301 is configured to acquire historical and real-time internal resistance data of the target device, combine with multi-dimensional running state parameters, and generate a set of normalized internal resistance time series data through a dynamic time warping algorithm.

[0271] The construction module 302 is configured to perform multi-scale feature extraction and high-dimensional space mapping on the set of internal resistance time series data, construct a set of internal resistance feature vectors, and identify internal resistance abnormal pattern clusters hidden in the data distribution by using an adaptive density clustering algorithm.

[0272] The analysis module 303 is configured to perform multi-modal fault correlation analysis based on the internal resistance abnormal pattern clusters, combine with the device operation history and working condition labels to predict the evolution path and risk level of potential faults, and generate a fault prediction atlas.

[0273] The operation and maintenance module 304 is configured to generate a set of differentiated operation and maintenance strategies according to the fault prediction atlas, and drive the operation and maintenance management platform to execute corresponding operation and maintenance instructions.

[0274] The above embodiments according to the drawings explain the structure, features and effects of the application in detail, and the above description is only the preferred embodiment of the application, but the application is not limited by the drawings, any changes or modifications made according to the concept of the application, or equivalent embodiments with equivalent changes, as long as they are within the scope of the application.

Claims

1. A large data-driven internal resistance safety operation method, characterized in that, The method comprises: Collecting historical and real-time internal resistance data of the target device, combining multi-dimensional running state parameters, and generating a set of normalized internal resistance time series data through a dynamic time warping algorithm; Performing multi-scale feature extraction and high-dimensional space mapping on the set of internal resistance time series data, constructing a set of internal resistance feature vectors, and identifying internal resistance abnormal mode clusters hidden in the data distribution using an adaptive density clustering algorithm; the multi-scale feature extraction and high-dimensional space mapping on the set of internal resistance time series data, the construction of the set of internal resistance feature vectors, and the identification of the internal resistance abnormal mode clusters hidden in the data distribution using the adaptive density clustering algorithm comprise: performing multi-scale wavelet transform on the set of normalized internal resistance time series data, extracting time domain and frequency domain features, and generating a set of multi-scale feature vectors; performing high-dimensional space mapping on the set of multi-scale feature vectors using an autoencoder, extracting deep feature representations through the encoder, and generating a set of high-dimensional feature representation vectors; constructing a set of internal resistance feature vectors based on the set of high-dimensional feature representation vectors, performing feature dimension reduction using a principal component analysis algorithm, retaining main feature components, and generating a set of reduced feature vectors; designing an adaptive density clustering algorithm, automatically adjusting clustering parameters according to data distribution characteristics, performing clustering analysis on the set of reduced feature vectors, and generating a preliminary clustering result; performing abnormal mode identification on the preliminary clustering result, extracting clusters that are obviously separated from other clusters as internal resistance abnormal mode clusters, and finally outputting internal resistance abnormal mode cluster identifiers; Performing multi-modal fault correlation analysis based on the internal resistance abnormal mode clusters, combining device running history and working condition labels to predict the evolution path and risk level of potential faults, and generating a fault prediction graph; Generating a set of differentiated operation and maintenance strategies according to the fault prediction graph, and driving an operation and maintenance management platform to execute corresponding operation and maintenance instructions.

2. The method of claim 1, wherein, The collection of historical and real-time internal resistance data of the target device, the combination of multi-dimensional running state parameters, and the generation of a set of normalized internal resistance time series data through a dynamic time warping algorithm comprise: Real-time collection of internal resistance data of the target device through a distributed data collection node, simultaneous collection of at least multi-dimensional running state parameters including voltage, current, and temperature, and generation of a multi-source original monitoring data set; Data cleaning and outlier processing of the multi-source original monitoring data set, smoothing of data fluctuations using a sliding window filtering algorithm, and generation of a pre-processed multi-dimensional monitoring data set; Based on the pre-processed multi-dimensional monitoring data set, alignment of internal resistance data and running state parameters with different sampling frequencies using a dynamic time warping algorithm, elimination of time scale differences, and generation of a time-aligned multi-dimensional data sequence; Normalization processing of the time-aligned multi-dimensional data sequence, unification of internal resistance data and each running state parameter to the same dimension, and finally generation of a set of normalized internal resistance time series data.

3. The method of claim 1, wherein, The multi-modal fault correlation analysis based on the internal resistance abnormal mode clusters, the combination of device running history and working condition labels to predict the evolution path and risk level of potential faults, and the generation of a fault prediction graph comprise: Extracting feature statistics of the internal resistance abnormal mode clusters including at least the distribution range, density, and outlier degree of the data points in the clusters, and generating abnormal mode feature descriptions; The fault correlation model is constructed by using a graph neural network in combination with the fault records and working condition labels in the equipment operation history data, the correlation between the abnormal patterns and the fault types is analyzed, and a fault correlation analysis result is generated; Based on the fault correlation analysis result, an evolution path of a potential fault is predicted by using a hidden Markov model, the occurrence probability of each path is calculated, and a fault evolution path prediction result is generated; According to the fault evolution path prediction result, the risk level of each path is evaluated, the risk level evaluation result is generated by considering the severity of the fault and the occurrence probability, and the risk level evaluation result is generated; The multi-dimensional fault prediction graph is constructed by integrating the fault correlation analysis result, the fault evolution path prediction result and the risk level evaluation result, and a complete fault prediction graph is finally output.

4. The method of claim 3, wherein, The difference operation and maintenance strategy set is generated according to the fault prediction graph, and the corresponding operation and maintenance instruction is driven to be executed by the operation and maintenance management platform, which comprises: The risk level and fault type information in the fault prediction graph are analyzed, the corresponding basic operation and maintenance strategy is matched from the operation and maintenance strategy library, and a preliminary operation and maintenance strategy set is generated; Based on the real-time operation state and historical operation and maintenance effect data of the equipment, the preliminary operation and maintenance strategy set is optimized and adjusted by using a reinforcement learning algorithm, and an optimized operation and maintenance strategy is generated; According to the optimized operation and maintenance strategy, specific executable operation and maintenance instructions are generated, the instructions include maintenance time, maintenance content and required resources, and an operation and maintenance instruction set is obtained; The operation and maintenance instruction set is sent to the operation and maintenance management platform through an API interface, the platform is driven to automatically execute the corresponding operation and maintenance task, and the execution state is monitored in real time, and finally the execution and feedback of the operation and maintenance instruction are completed.

5. A large data-driven internal resistance safety operation system, characterized in that, The system comprises: The acquisition module is configured to collect historical and real-time internal resistance data of a target device, and combine multi-dimensional operation state parameters to generate a set of internal resistance time series data after dynamic time warping by using a dynamic time warping algorithm; The construction module is configured to perform multi-scale feature extraction and high-dimensional space mapping on the internal resistance time series data set, construct an internal resistance feature vector set, and identify internal resistance abnormal pattern clusters hidden in the data distribution by using an adaptive density clustering algorithm; the multi-scale feature extraction and high-dimensional space mapping on the internal resistance time series data set, the construction of the internal resistance feature vector set, and the identification of the internal resistance abnormal pattern clusters hidden in the data distribution by using the adaptive density clustering algorithm comprise: performing multi-scale wavelet transform on the set of internal resistance time series data after dynamic time warping, extracting time domain and frequency domain features, and generating a multi-scale feature vector; performing high-dimensional space mapping on the multi-scale feature vector by using a self-encoder, extracting deep feature representations by using an encoder, and generating a high-dimensional feature representation vector; constructing an internal resistance feature vector set based on the high-dimensional feature representation vector, performing feature dimension reduction by using a principal component analysis algorithm, retaining main feature components, and generating a set of feature vectors after dimension reduction; designing an adaptive density clustering algorithm, automatically adjusting clustering parameters according to data distribution characteristics, performing clustering analysis on the set of feature vectors after dimension reduction, and generating a preliminary clustering result; performing abnormal pattern recognition on the preliminary clustering result, extracting clusters that are obviously separated from other clusters as internal resistance abnormal pattern clusters, and finally outputting internal resistance abnormal pattern cluster identifiers; The analysis module is configured to perform multi-modal fault correlation analysis based on the internal resistance anomaly pattern cluster, combine device operation history and working condition labels to predict an evolution path and a risk level of a potential fault, and generate a fault prediction graph; The operation and maintenance module is configured to generate a set of differential operation and maintenance strategies according to the fault prediction graph, and drive an operation and maintenance management platform to execute corresponding operation and maintenance instructions.

6. The system of claim 5, wherein, The collection module is specifically configured to: Collect, through a distributed data collection node, internal resistance data of the target device in real time, simultaneously collect multi-dimensional operation state parameters including at least voltage, current, and temperature, and generate a multi-source original monitoring data set; Perform data cleaning and outlier processing on the multi-source original monitoring data set, smooth data fluctuations by using a sliding window filtering algorithm, and generate a pre-processed multi-dimensional monitoring data set; Align internal resistance data and operation state parameters of different sampling frequencies by using a dynamic time warping algorithm based on the pre-processed multi-dimensional monitoring data set, eliminate time scale differences, and generate a time-aligned multi-dimensional data sequence; Perform normalization processing on the time-aligned multi-dimensional data sequence, unify internal resistance data and each operation state parameter to the same dimension, and finally generate a normalized internal resistance time series data set.

7. A storage medium, characterized by The storage medium stores a computer program, and the computer program is configured to execute the method in any one of claims 1-4 when running.

8. An electronic device comprising a memory and a processor, characterized in that The memory stores a computer program, and the processor is configured to execute the method in any one of claims 1-4 by running the computer program.

Images