Method for searching faults of mutual inductor

Abstract

The invention provides a method for searching a mutual inductor fault, which comprises the following steps of: calculating a correlation coefficient of temperature and current according to a temperature sequence and a current sequence, judging temperature abnormity, and determining an abnormal time period; extracting the contact resistance data of the abnormal time period from the time sequence database, and generating a resistance change sequence; smoothing the resistance change sequence to generate a smoothed resistance sequence; according to the dynamic regression parameters, calculating a fault temperature rise probability, judging a contact fault, and obtaining a fault judgment result; predicting a fault development trend according to the fault judgment result, and generating a fault development trend sequence; according to the fault development trend sequence, predicting future temperature anomaly, and generating a predicted temperature sequence; and generating a fault early warning signal according to the predicted temperature sequence, and storing the fault early warning signal in the time sequence database.

Description

Technical Field

[0001] This invention relates to the field of instrument transformer technology, and in particular to a method for locating instrument transformer faults. Background Technology

[0002] Background of the problem: Instrument transformers are indispensable devices in power systems, and their stable operation is crucial for ensuring grid security. Faults in instrument transformers can lead to inaccurate electricity metering, equipment damage, and even system outages. Therefore, developing efficient fault detection methods is of great significance. Currently, instrument transformer fault detection mainly relies on manual inspections or periodic resistance measurements. This method is time-consuming and labor-intensive, and it is difficult to detect hidden problems in real time, especially in the early stages of contact faults, where fault characteristics are often not obvious and easily overlooked.

[0003] The limitation of existing methods lies in their lack of accurate capture and dynamic analysis of early fault signals. Contact faults typically manifest as abnormal temperature increases, but traditional detection methods struggle to monitor temperature changes in real time and cannot effectively distinguish between abnormal temperature rises caused by faults and temperature changes resulting from normal load fluctuations. This limits the accuracy of fault diagnosis and increases the risk of misdiagnosis or missed diagnosis.

[0004] The core challenge stems first from the difficulty of dynamically monitoring contact temperature. Contact temperature is affected by multiple factors, including load current and environmental conditions. A single temperature reading is insufficient to accurately reflect the fault state, necessitating the establishment of a dynamic model of the temperature-load relationship to eliminate interfering factors. The construction of this model, in turn, relies on the accurate measurement of contact resistance, as changes in resistance directly reflect contact loosening, oxidation, or ablation. However, the real-time performance and accuracy of resistance measurements are limited by current technology, making efficient completion under operating conditions difficult, thus hindering the comprehensive analysis of fault characteristics.

[0005] Therefore, how to construct a dynamic model of the relationship between temperature and load by real-time monitoring of contact temperature and combining it with resistance measurement results, so as to accurately distinguish between fault temperature rise and normal temperature rise, has become a key issue in the detection of transformer contact faults. Summary of the Invention

[0006] This invention provides a method for locating faults in current transformers, mainly including: Real-time data streams of contact temperature and load current are acquired via sensors, timestamped, and used to generate temperature and current sequences, which are then stored in a time-series database. Based on these sequences, the correlation coefficient between temperature and current is calculated to identify temperature anomalies and determine the abnormal time period. Contact resistance data for the abnormal time period is extracted from the time-series database to generate a resistance change sequence. This sequence is then smoothed to generate a smoothed resistance sequence. A temperature-resistance-load dynamic regression model is constructed based on the smoothed resistance sequence and the abnormal time period. The model parameters are fitted to obtain dynamic regression parameters. The probability of a fault temperature rise is calculated based on these parameters to determine contact faults and obtain a fault assessment result. The fault assessment result is used to predict the fault development trend, generating a fault development trend sequence. Based on the fault development trend sequence, future temperature anomalies are predicted, generating a predicted temperature sequence. Finally, a fault warning signal is generated based on the predicted temperature sequence and stored in the time-series database.

[0007] Furthermore, the step of acquiring real-time data streams of contact temperature and load current through sensors, adding timestamps, generating temperature and current sequences, and storing them in a time series database includes: acquiring real-time data streams of contact temperature and load current through a four-wire resistance sensor, adding millisecond-level timestamps to generate sensor data; performing noise reduction processing on the sensor data using a Kalman filter algorithm to obtain smoothed sensor data; separating the contact temperature data and load current data based on the smoothed sensor data to generate temperature and current sequences; storing the temperature and current sequences in a preset time series database through a time series database interface to generate a storage index; extracting the temperature and current sequences based on the storage index, calculating timestamp continuity, and determining data integrity; if the timestamp continuity is lower than a preset threshold, filling in missing data using a linear interpolation algorithm to obtain a complete temperature and current sequence; and using the ARIMA algorithm to predict future temperature and current trends based on the complete temperature and current sequences to obtain a predicted sequence.

[0008] Furthermore, the step of calculating the correlation coefficient between temperature and current based on the temperature sequence and the current sequence, determining temperature anomalies, and identifying abnormal time periods includes: obtaining synchronous timestamp data from the temperature sequence and the current sequence, preprocessing it using time series analysis methods to obtain standardized temperature and current sequences; dividing the data into windows using a sliding window method based on the standardized temperature and current sequences, calculating the correlation coefficient between temperature and current within each window to obtain a correlation coefficient sequence; if the correlation coefficient of a certain window in the correlation coefficient sequence is lower than a preset threshold, then determining that the time period corresponding to that window has a temperature anomaly, and identifying a set of abnormal time periods; classifying the abnormal time periods using a K-means clustering algorithm for the set of abnormal time periods to obtain a category distribution of abnormal time periods; extracting the time change trend of each category based on the category distribution to obtain trend features of abnormal time periods; determining the severity of the anomaly based on the trend features to determine the priority ranking of abnormal time periods; and fitting the temperature-current relationship of high-priority abnormal time periods using a linear regression algorithm based on the priority ranking to obtain fitting parameters.

[0009] Furthermore, the step of extracting the contact resistance data of the abnormal time period from the time series database to generate a resistance change sequence includes: extracting the contact resistance data of the abnormal time period from the time series database, using a micro-ohm-level high-precision measurement method, sampling multiple times per second to generate a preliminary resistance dataset; using a mean filtering algorithm to denoise the preliminary resistance dataset to obtain a denoised resistance dataset; using a time series interpolation method to fill in missing data points in the denoised resistance dataset to generate a continuous resistance change sequence; extracting timestamps and resistance values ​​from the continuous resistance change sequence, using a sliding window algorithm to calculate the resistance change rate to obtain a resistance change trend; if the resistance change trend exceeds a preset threshold, identifying abnormal points using an anomaly detection algorithm to obtain an abnormal resistance point sequence; and using a clustering analysis method to group the abnormal points according to the abnormal resistance point sequence to obtain an abnormal event classification result.

[0010] Furthermore, the smoothing process for the resistance change sequence to generate a smoothed resistance sequence includes: acquiring the resistance change sequence; collecting raw resistance data from a sensor to obtain an initial resistance sequence; processing the initial resistance sequence using a Kalman filter algorithm to iteratively update the state estimate to obtain a filtered resistance sequence; calculating a residual sequence by comparing the filtered resistance sequence with the initial resistance sequence to determine environmental noise characteristics; if the residual sequence exceeds a preset threshold, performing a secondary smoothing process on the filtered resistance sequence to obtain an optimized resistance sequence; extracting the micro-ohm to milliohm level variation trend based on the optimized resistance sequence to generate a trend feature sequence; detecting abnormal fluctuations in contact resistance using the trend feature sequence to identify potential fault points; and processing the trend feature sequence using statistical analysis methods to generate a smoothed resistance sequence.

[0011] Furthermore, the step of constructing a temperature-resistance-load dynamic regression model based on the smoothed resistance sequence and the abnormal time period, and fitting the model parameters to obtain dynamic regression parameters, includes: acquiring temperature data and load current data of the smoothed resistance sequence and the abnormal time period, constructing a temperature-resistance-load dynamic regression model that includes ambient temperature fluctuations; fitting the parameters of the dynamic regression model using the least squares method to obtain dynamic regression parameters; extracting dynamic relationship features between temperature, resistance, and load based on the dynamic regression parameters; analyzing the influence weight of ambient temperature fluctuations on resistance changes through the dynamic relationship features; optimizing the dynamic regression model based on the influence weights to obtain optimized dynamic regression parameters; verifying the fitting accuracy of the dynamic regression model using the optimized dynamic regression parameters; and adjusting the parameters of the dynamic regression model based on the verification results to generate the final dynamic regression parameters.

[0012] Furthermore, the step of calculating the fault temperature rise probability based on the dynamic regression parameters, determining the contact fault, and obtaining the fault judgment result includes: acquiring the dynamic regression parameters, combining the weight of the influence of resistance change on the probability and the load current factor, and constructing a fault temperature rise probability model based on the normal distribution assumption; calculating the fault temperature rise probability using the fault temperature rise probability model; if the fault temperature rise probability is higher than a preset probability threshold, then determining it as a contact fault and generating a fault judgment result; extracting the temperature, resistance, and load data corresponding to the fault based on the fault judgment result; determining the fault feature distribution using statistical analysis methods based on the data corresponding to the fault; generating a fault classification result based on the fault feature distribution; verifying the accuracy of the fault judgment result using the fault classification result, and generating a final fault judgment result.

[0013] Furthermore, the step of predicting future temperature anomalies and generating a predicted temperature sequence based on the fault development trend sequence includes: acquiring the fault development trend sequence; constructing a temperature anomaly prediction model using a long short-term memory network model combined with a preset safe temperature threshold; performing time series analysis on the fault development trend sequence using the temperature anomaly prediction model to generate a predicted temperature sequence; extracting timestamps and predicted temperature values ​​from the predicted temperature sequence; calculating the probability distribution of future temperature anomalies using the predicted temperature values; determining the time period of future temperature anomalies based on the probability distribution; extracting corresponding predicted temperature data for the time period of future temperature anomalies; and verifying the accuracy of the predicted temperature sequence using the predicted temperature data to generate a final predicted temperature sequence.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a contact fault early warning method. It uses a four-wire resistance sensor to collect contact temperature and load current data in real time and stores them in a time-series database. A sliding window is used to calculate the correlation coefficient between temperature and current to identify periods of abnormal temperature. High-precision contact resistance data is extracted for these abnormal periods, and Kalman filtering is applied to eliminate noise. A temperature-resistance-load dynamic regression model is constructed to calculate the probability of fault temperature rise. Database labels are updated based on the fault identification results, and time-series analysis is used to predict fault development trends. A long short-term memory network is used to predict future temperature anomalies, generating an early warning signal when a safety threshold is exceeded. This invention achieves accurate detection, trend prediction, and timely early warning of contact faults, improving the safety and reliability of power systems. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for locating faults in a current transformer according to the present invention.

[0016] Figure 2 This is a schematic diagram of a method for locating faults in a current transformer according to the present invention.

[0017] Figure 3 This is another schematic diagram of a method for locating faults in a current transformer according to the present invention.

[0018] Figure 4 This is another schematic diagram of a method for locating faults in a current transformer according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0020] like Figure 1-4This embodiment of a method for locating faults in a current transformer may specifically include: Step S101: 1. Obtain real-time data streams of contact temperature and load current using a four-wire resistance sensor, accurate to millisecond timestamps, and store them in a time series database to obtain contact temperature sequence and load current sequence.

[0021] Real-time data streams of contact temperature and load current are acquired using a four-wire resistance sensor, with millisecond-level timestamps appended to generate sensor data. Based on this data, a Kalman filter algorithm is used to denoise the data, resulting in smoothed sensor data. From the smoothed sensor data, the contact temperature and load current data are separated to generate temperature and current sequences. These sequences are stored in a pre-defined time-series database via a time-series database interface, generating a storage index. Based on the storage index, the temperature and current sequences are extracted from the time-series database, and the timestamp continuity is calculated to determine data integrity. If the timestamp continuity is below a preset threshold, missing data is padded using a linear interpolation algorithm to obtain complete temperature and current sequences. Based on these complete temperature and current sequences, a sliding window method is used to calculate the correlation coefficient between temperature and current to identify abnormal temperature periods. Based on these abnormal temperature periods, the fault status labels in the time-series database are updated, generating fault records. Based on these fault records, a time-series analysis algorithm is used to predict the fault development trend, updated every minute, resulting in a fault development trend sequence.

[0022] Specifically, a four-wire resistance sensor is used to collect contact temperature (range -40℃ to 150℃) and load current (range 0 to 100A) at a sampling rate of 1000 times per second, synchronously recording UNIX timestamps (accurate to milliseconds) to generate a sensor data stream containing timestamps, temperature values, and current values. A Kalman filter algorithm (process noise Q=0.01, measurement noise R=0.1) is used to iteratively filter the sensor data, eliminating temperature fluctuations of ±0.5℃ and current noise of ±0.2A, outputting smoothed temperature-current data pairs. Based on the smoothed data stream, fields are separated by data type to generate two independent time series: a temperature series (timestamp-temperature value) and a current series (timestamp-current value). The InfluxDB HTTP API (port 8086) is called to write 500 data entries to the time series database in batches, returning a storage index containing shardID and blockID. The system queries the most recent hour's data based on the index, checking if the interval between adjacent timestamps is 1 millisecond. If more than 5 consecutive data points are missing (the threshold is set to 5ms interruption), linear interpolation data is inserted at the missing positions (e.g., missing value at time t = (value at time t-1 + value at time t+1) / 2). A Pearson correlation coefficient is calculated using a sliding window with a window width of 60 seconds. When the temperature-current correlation coefficient is below 0.7 (preset threshold), the window is marked as an abnormal period. The start and end times corresponding to the abnormal period (e.g., from 2023-08-20T14:25:30.123 to 2023-08-20T14:26:30.123) are updated to the `fault_status` label field in the database. Based on the past 24 hours of fault records, the ARIMA(5,1,0) model is used to predict the fault probability for the next 60 minutes, outputting a sequence of predicted values ​​updated every minute (e.g., predicted value at 14:30 = 0.83, predicted value at 14:31 = 0.79).

[0023] Step S102,2: Based on the contact temperature sequence and load current sequence, the correlation coefficient between temperature and current is calculated using the sliding window method. If the correlation coefficient is lower than the preset threshold, it is determined that there is a temperature anomaly, and the time period of the temperature anomaly is obtained.

[0024] Temperature and load current sequences are extracted from a time-series database, and synchronization timestamp data is obtained to obtain standardized temperature and current sequences. Based on these sequences, a sliding window method is used to divide the data into windows, and the Pearson correlation coefficient between temperature and current within each window is calculated to obtain a correlation coefficient sequence. If the correlation coefficient of a window in the correlation coefficient sequence is lower than a preset threshold, the corresponding time period is considered to have a temperature anomaly, thus determining a set of anomalous time periods. Based on this set, corresponding temperature and current data are obtained, and K-means clustering is used to classify the anomalous time periods, resulting in a category distribution. Based on this category distribution, time-series analysis methods are used to extract the temporal change trend of each category, obtaining the trend characteristics of the anomalous time periods. The duration and magnitude of change of the anomalous time periods are obtained from these trend characteristics, and the severity of the anomalies is determined using preset rules to prioritize the anomalous time periods. Based on this priority ranking, the temperature and current sequences of high-priority anomalous time periods are obtained, and a linear regression algorithm is used to fit the temperature-current relationship of these periods, obtaining the fitting parameters for each anomalous time period. Based on the fitted parameters, the deviation values ​​of temperature and current within the abnormal time period are calculated. If the deviation value exceeds a preset deviation threshold, a potential fault risk is identified, resulting in a set of potential fault time periods. Using this set of potential fault time periods, the fault status labels in the time series database are updated, generating a fault warning signal and obtaining a warning signal sequence.

[0025] Specifically, the contact temperature and load current sequences are extracted from a time-series database, and synchronization timestamp data is obtained. The temperature and current sequences are normalized using the Z-score standardization method to obtain standardized temperature and current sequences with a mean of 0 and a standard deviation of 1. Based on the standardized temperature and current sequences, a sliding window method is used to divide the data into windows with a size of 60 time points and a step size of 10 time points. The Pearson correlation coefficient between temperature and current within each window is calculated to obtain a correlation coefficient sequence. If the correlation coefficient of a window in the correlation coefficient sequence is lower than a preset threshold of 0.7, it is determined that there is a temperature anomaly in the corresponding time period, thus identifying the set of anomalous time periods. Based on the set of anomalous time periods, the corresponding temperature and current data are obtained, and the anomalous time periods are classified using the K-means clustering algorithm with a cluster size of 3. Sample similarity is calculated using Euclidean distance to obtain the category distribution of the anomalous time periods. Based on the category distribution of the anomalous time periods, time series analysis methods are used to extract the time change trend of each category, and a trend line is calculated using the moving average method to obtain the trend characteristics of the anomalous time periods. The duration and magnitude of abnormal time periods are obtained from trend characteristics. Preset rules are used to determine the severity of the anomalies. If the duration exceeds 5 minutes and the magnitude of change exceeds 10%, it is identified as a high-priority abnormal time period, resulting in a priority ranking of the abnormal time periods. Based on the priority ranking, the temperature and current sequences of the high-priority abnormal time periods are obtained. A linear regression algorithm is used to fit the temperature-current relationship of the abnormal time periods, and the regression coefficients are calculated using the least squares method to obtain the fitting parameters for the abnormal time periods. Based on the fitting parameters, the deviation between temperature and current within the abnormal time periods is calculated. If the deviation exceeds a preset deviation threshold of 0.5, a potential fault risk is identified, resulting in a set of potential fault time periods. Using the set of potential fault time periods, the fault status labels in the time series database are updated, generating fault warning signals, which are stored in the database as percentage floating-point numbers, resulting in a warning signal sequence.

[0026] Step S103, 3: For the time period of abnormal temperature, extract the contact resistance measurement data of the corresponding time period from the time series database, and use a high-precision micro-ohm level measurement method, combined with multiple samplings per second, to obtain the contact resistance change sequence.

[0027] Contact resistance measurement data during periods of temperature anomalies were extracted from a time-series database. A micro-ohm-level high-precision measurement method was used to obtain a preliminary resistance dataset. This dataset was sampled multiple times per second using high-frequency data acquisition technology to obtain a high-frequency resistance dataset. A mean filtering algorithm and noise removal techniques were applied to the high-frequency resistance dataset to obtain a denoised resistance dataset. For the denoised resistance dataset, time-series interpolation was used to fill in missing data points, generating a continuous resistance change sequence. Timestamps and resistance values ​​were extracted from the continuous resistance change sequence, and a sliding window algorithm was used to calculate the resistance change rate, obtaining the resistance change trend. If the resistance change trend exceeded a preset threshold, an anomaly detection algorithm was used to analyze the resistance change trend and identify anomaly resistance point sequences. Based on the anomaly resistance point sequences, cluster analysis was used to group the anomaly points, obtaining anomaly event classification results. Timestamp data synchronized with the anomaly event classification results were obtained from the contact temperature sequence and load current sequence. Time-series analysis methods were used for preprocessing to obtain standardized temperature and current sequences. Based on the standardized temperature and current sequences, a sliding window method was used to calculate the correlation coefficient between temperature and current, obtaining a correlation coefficient sequence.

[0028] Specifically, contact resistance data for periods of abnormal temperature (e.g., temperatures exceeding 85°C for more than 30 seconds) were extracted from a time-series database and measured using a Fluke 1587C micro-ohmmeter with an accuracy of ±0.1 μΩ to obtain an initial resistance dataset. The initial dataset was sampled at a high frequency of 1000 times per second, and the raw data was recorded using an NI PXIe-5160 data acquisition card. A mean filtering algorithm with a window size of 5 was used to process the high-frequency data and eliminate random noise interference. For missing points in the filtered data (e.g., sampling intervals greater than 10 ms), cubic spline interpolation was performed to generate a continuous resistance sequence. The resistance change rate was calculated with a 10-second sliding window; a change rate exceeding 0.5 μΩ / s was considered an anomaly. Anomalies were detected using the Isolation Forest algorithm, with an anomaly score threshold of 0.65. The DBSCAN clustering algorithm (eps=0.3, min_samples=5) was used to group anomalies and identify the categories of abnormal events. Time-series data from the PT100 temperature sensor and the current transformer (accuracy class 0.2) were extracted simultaneously and processed using the Z-score normalization method. The Pearson correlation coefficient between temperature and current was calculated in 30-second windows, and windows with a coefficient below 0.7 were marked as abnormal.

[0029] Step S104, 4: Using the contact resistance variation sequence, the Kalman filter algorithm is used to smooth the resistance data ranging from micro-ohms to milliohms, eliminating environmental interference noise and obtaining a smoothed resistance sequence.

[0030] The process involves acquiring a sequence of contact resistance changes. Raw resistance data is collected from sensors to obtain an initial resistance sequence. A Kalman filter algorithm is used to process this initial resistance sequence, iteratively updating the state estimate to obtain a filtered resistance sequence. By comparing the filtered resistance sequence with the initial resistance sequence, a residual sequence is calculated to determine environmental noise characteristics. If the residual sequence exceeds a preset threshold, a secondary smoothing process is performed on the filtered resistance sequence to obtain an optimized resistance sequence. Based on the optimized resistance sequence, the variation trend at the micro-ohm to milliohm level is extracted to generate a trend feature sequence. This trend feature sequence is used to detect abnormal fluctuations in contact resistance and identify potential fault points. Statistical analysis methods are then used to process the trend feature sequence to generate a smoothed resistance sequence.

[0031] Step S105,5: Based on the smoothed resistance sequence and the time period of temperature anomalies, construct a temperature-resistance-load dynamic regression model that includes ambient temperature fluctuations, and use the least squares method to fit the model parameters to obtain the dynamic regression parameters.

[0032] Step S106, 6: For the dynamic regression parameters, combined with the weight of the influence of resistance change on probability and the load current factor, calculate the fault temperature rise probability based on the normal distribution assumption. If the fault temperature rise probability is higher than the preset probability threshold, it is judged as a contact fault, and the fault judgment result is obtained.

[0033] Step S107,7: Based on the fault judgment results, update the fault status labels in the time series database, compare historical fault data, and use a time series analysis algorithm to predict the fault development trend updated every minute to obtain the fault development trend sequence.

[0034] Step S108,8: Based on the fault development trend sequence, a long short-term memory network model is used, combined with a preset safe temperature threshold, to predict future temperature anomalies and obtain a predicted temperature sequence.

[0035] Step S109, 9: For the predicted temperature sequence, if the predicted temperature exceeds the preset safe temperature threshold, a fault warning signal is generated and stored in the database in percentage floating-point format to obtain the warning signal sequence.

[0036] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method of finding a fault in a mutual inductor, characterized by, The method comprises the following steps: obtaining real-time data streams of contact temperature and load current through sensors, adding time stamps, generating temperature sequences and current sequences, and storing them into a time series database; calculating the correlation coefficient of temperature and current according to the temperature sequences and the current sequences, judging temperature abnormalities, and determining abnormal time periods; extracting contact resistance data of the abnormal time periods from the time series database, and generating resistance change sequences; smoothing the resistance change sequences to generate smoothed resistance sequences; constructing a temperature-resistance-load dynamic regression model according to the smoothed resistance sequences and the abnormal time periods, fitting model parameters, and obtaining dynamic regression parameters; calculating the fault temperature rise probability according to the dynamic regression parameters, judging the contact fault, and obtaining the fault judgment result; predicting the fault development trend according to the fault judgment result, and generating a fault development trend sequence; predicting future temperature abnormalities according to the fault development trend sequence, and generating a predicted temperature sequence; generating a fault warning signal according to the predicted temperature sequence, and storing it into the time series database.

2. The method of claim 1, wherein, The method comprises the following steps: collecting real-time data streams of contact temperature and load current through four-wire method resistance sensors, adding millisecond-level time stamps, and generating sensor data; using Kalman filter algorithm to denoise the sensor data to obtain smoothed sensor data; separating contact temperature data and load current data according to the smoothed sensor data to generate temperature sequences and current sequences; storing the temperature sequences and the current sequences into a preset time series database through a time series database interface to generate a storage index; extracting the temperature sequences and the current sequences according to the storage index, calculating the time stamp continuity, and judging the data integrity; if the time stamp continuity is lower than a preset threshold, then completing the missing data through linear interpolation algorithm to obtain complete temperature sequences and complete current sequences; using ARIMA algorithm to predict future temperature and current trends according to the complete temperature sequences and the complete current sequences to obtain prediction sequences.

3. The method of claim 1, wherein, The method comprises the following steps: obtaining synchronous time stamp data from the temperature sequences and the current sequences, and using time series analysis method for preprocessing to obtain standardized temperature sequences and standardized current sequences; dividing data windows using sliding window method according to the standardized temperature sequences and the standardized current sequences, calculating the correlation coefficient of temperature and current in each window, and obtaining a correlation coefficient sequence; if the correlation coefficient of a certain window in the correlation coefficient sequence is lower than a preset threshold, then judging that the time period corresponding to the window has temperature abnormalities, and determining an abnormal time period set; classifying the abnormal time periods using K-means clustering algorithm for the abnormal time period set to obtain the category distribution of the abnormal time periods; According to the category distribution, the time variation trend of each category is extracted to obtain the trend feature of the abnormal time period; According to the trend feature, the severity of the anomaly is judged to determine the priority ranking of the abnormal time period; According to the priority ranking, a linear regression algorithm is used to fit the temperature-current relationship of the high-priority abnormal time period to obtain the fitting parameters.

4. The method of claim 1, wherein, The contact resistance data of the abnormal time period is extracted from the time series database to generate a resistance change sequence, including: Extracting the contact resistance data of the abnormal time period from the time series database, using a micro-ohm level high-precision measurement method, and generating a preliminary resistance data set by sampling multiple times per second; Using a mean filter algorithm to denoise the preliminary resistance data set to obtain a denoised resistance data set; For the denoised resistance data set, a time series interpolation method is used to fill in the missing data points to generate a continuous resistance change sequence; From the continuous resistance change sequence, the timestamp and resistance value are extracted, and a sliding window algorithm is used to calculate the resistance change rate to obtain the resistance change trend; If the resistance change trend exceeds a preset threshold, an anomaly detection algorithm is used to judge the abnormal points to obtain an abnormal resistance point sequence; According to the abnormal resistance point sequence, a clustering analysis method is used to group the abnormal points to obtain an abnormal event classification result.

5. The method of claim 1, wherein, The resistance change sequence is smoothed to generate a smoothed resistance sequence, including: Obtain the resistance change sequence, collect the original resistance data from the sensor to obtain the initial resistance sequence; Using a Kalman filter algorithm to process the initial resistance sequence, iteratively updating the state estimation to obtain a filtered resistance sequence; By comparing the filtered resistance sequence with the initial resistance sequence, a residual sequence is calculated to determine the environmental noise characteristics; If the residual sequence exceeds a preset threshold, the filtered resistance sequence is subjected to secondary smoothing processing to obtain an optimized resistance sequence; According to the optimized resistance sequence, a micro-ohm to milli-ohm level change trend is extracted to generate a trend feature sequence; Through the trend feature sequence, the contact resistance abnormal fluctuation is detected to determine the potential fault point; Using a statistical analysis method to process the trend feature sequence to generate a smoothed resistance sequence.

6. The method of claim 1, wherein, According to the smoothed resistance sequence and the abnormal time period, a temperature-resistance-load dynamic regression model is constructed, the model parameters are fitted, and dynamic regression parameters are obtained, including: Obtain the temperature data and load current data of the smoothed resistance sequence and the abnormal time period to construct a temperature-resistance-load dynamic regression model containing environmental temperature fluctuations; Using the least squares method to fit the parameters of the dynamic regression model to obtain dynamic regression parameters; According to the dynamic regression parameters, the dynamic relationship features between temperature, resistance and load are extracted; Through the dynamic relationship features, the influence weight of environmental temperature fluctuation on resistance change is analyzed; According to the influence weight, the dynamic regression model is optimized to obtain optimized dynamic regression parameters; Through the optimized dynamic regression parameters, the fitting accuracy of the dynamic regression model is verified; According to the verification result, the parameters of the dynamic regression model are adjusted to generate the final dynamic regression parameters.

7. The method of claim 1, wherein, The method comprises the following steps: Obtaining the dynamic regression parameter, combining the weight of the influence of resistance change on probability and the load current factor, and constructing a fault temperature rise probability model based on the normal distribution assumption; Calculating the fault temperature rise probability through the fault temperature rise probability model; If the fault temperature rise probability is higher than the preset probability threshold, it is judged as a contact fault, and a fault judgment result is generated; According to the fault judgment result, the temperature, resistance and load data corresponding to the fault are extracted; Through the data corresponding to the fault, the statistical analysis method is used to determine the fault feature distribution; According to the fault feature distribution, a fault classification result is generated; Through the fault classification result, the accuracy of the fault judgment result is verified, and a final fault judgment result is generated.

8. The method of claim 1, wherein, The method comprises the following steps: Obtaining the fault development trend sequence, using a long short-term memory network model, combining a preset safety temperature threshold, and constructing a temperature anomaly prediction model; Through the temperature anomaly prediction model, time series analysis is performed on the fault development trend sequence to generate a predicted temperature sequence; From the predicted temperature sequence, the timestamp and predicted temperature value are extracted; Through the predicted temperature value, the probability distribution of future temperature anomaly is calculated; According to the probability distribution, the time period of future temperature anomaly is determined; For the time period of future temperature anomaly, the corresponding predicted temperature data is extracted; Through the predicted temperature data, the accuracy of the predicted temperature sequence is verified, and a final predicted temperature sequence is generated.

Images