Fault early warning method based on substation operation multi-source data fusion inspection

Abstract

The application discloses a fault early warning method based on substation operation multi-source data fusion inspection, and relates to the technical field of substation operation fault early warning.The method comprises the following steps: obtaining multi-source inspection data of multiple same-type primary devices connected to the same voltage level bus in the same substation; aligning the same time sampling moments to obtain multi-source aligned inspection data; performing load normalization processing to obtain the normalized characteristic vectors of each same-type primary device; obtaining the same time normalized residual error vectors of each same-type primary device; determining that the target device has an early latent fault, and sending a hierarchical early warning signal.The application solves the technical problems of the prior art, such as being easily disturbed by load and working condition disturbance, lacking same-type device horizontal reference, low early latent fault identification accuracy, and frequent false early warning, by relying on single device longitudinal comparison, and achieves the technical effects of same-source device group horizontal comparison and judgment, improved early latent fault identification accuracy of substation primary devices, and reduced false early warning.

Description

Technical Field

[0001] This invention relates to the field of substation operation fault early warning technology, specifically to a fault early warning method based on multi-source data fusion inspection of substation operation. Background Technology

[0002] Currently, most substation primary equipment fault early warning systems rely on longitudinal comparison of historical data for individual equipment. However, equipment operating characteristics are easily affected by load fluctuations, instantaneous on-site interference, and equipment operation disturbances. This makes it difficult to distinguish between actual latent defects in the equipment itself and temporary anomalies caused by external operating conditions. Furthermore, the lack of horizontal reference benchmarks for equipment of the same voltage level and type within the same substation results in insufficient early identification capability for weak latent faults, frequent false alarms, and an inability to meet the actual operation and maintenance needs for accurate early warning of potential hazards in substation equipment.

[0003] Existing technologies rely on vertical comparison of single devices, which is susceptible to load and operating condition disturbances, lacks horizontal reference of similar devices, and suffers from low accuracy in identifying early latent faults and frequent false alarms. Summary of the Invention

[0004] This application provides a fault early warning method based on multi-source data fusion inspection of substation operation, which is used to address the technical problems of existing technologies that rely on vertical comparison of single equipment, are easily affected by load and operating condition disturbances, lack horizontal reference of similar equipment, have low accuracy in early latent fault identification, and have frequent false alarms.

[0005] In view of the above problems, this application provides a fault early warning method based on multi-source data fusion inspection of substation operation.

[0006] This application provides a fault early warning method based on multi-source data fusion inspection of substation operation, the method comprising: Multi-source inspection data of multiple identical primary devices connected to the same voltage level busbar within the same substation are acquired, forming a homogeneous device group. Using a preset synchronization trigger signal as a reference, the multi-source inspection data of each device in the homogeneous device group are synchronized at the same sampling time to obtain synchronized inspection data. Based on the real-time load current and rated parameters of each identical primary device, the synchronized inspection data is subjected to load normalization processing to obtain the normalized feature vector of each identical primary device. The group median vector of the normalized feature vectors of all devices in the homogeneous device group is calculated in each dimension, and the difference between the normalized feature vector of each device and the group median vector is obtained to obtain the synchronous normalized residual vector of each identical primary device. Based on the synchronous normalized residual vector, the residual deviation index of each device is calculated. When the residual deviation index of any target device exceeds a preset deviation threshold, and the residual deviation indices of the remaining devices in the homogeneous device group are all within the normal range, it is determined that the target device has an early latent fault, and a graded early warning signal is sent.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This method acquires multi-source inspection data from multiple identical primary equipment units connected to the same voltage level busbar within the same substation. The multi-source inspection data of each unit in the same equipment group are synchronized at the same sampling time to obtain synchronized multi-source inspection data. Load normalization is applied to the synchronized multi-source inspection data to obtain the normalized feature vector of each identical primary equipment unit. The group median vector of the normalized feature vectors of all units in the same equipment group is calculated in each dimension, and the difference between the normalized feature vector of each unit and the group median vector is used to obtain the synchronized normalized residual vector of each identical primary equipment unit. When the residual deviation index of any target unit exceeds a preset deviation threshold, and the residual deviation indexes of the remaining units in the same equipment group are all within the normal range, it is determined that the target unit has an early latent fault, and a graded early warning signal is sent. This method achieves the technical effect of horizontal comparison and judgment of the same equipment group, improving the accuracy of early latent fault identification in substation primary equipment and reducing false alarms. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the fault early warning method based on multi-source data fusion inspection of substation operation provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the process of aligning synchronous sampling times in the fault early warning method based on multi-source data fusion inspection of substation operation provided in the embodiments of this application. Detailed Implementation

[0010] This application provides a fault early warning method based on multi-source data fusion inspection of substation operation, which is used to address the technical problems of existing technologies that rely on vertical comparison of single equipment, are susceptible to load and operating condition disturbances, lack horizontal reference of similar equipment, have low accuracy in early latent fault identification, and have frequent false alarms.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0012] Examples, such as Figure 1 As shown, this application provides a fault early warning method based on multi-source data fusion inspection of substation operation, the method comprising: Step S100: Obtain multi-source inspection data of multiple primary devices of the same type connected to the same voltage level bus in the same substation, wherein the multiple primary devices of the same type constitute a group of devices of the same source.

[0013] Specifically, multi-source inspection data of multiple primary devices connected to the same voltage level busbar within the same substation are collected. A group of devices from transformers, GIS switchgear, and high-voltage switchgear with the same model, rated parameters, and service life differences not exceeding a preset threshold are selected. The acquired multi-source inspection data includes load current data, cooler status data, infrared thermal imaging data, dissolved gas chromatography data in oil, and ambient temperature and humidity data. Various types of raw inspection data are collected and aggregated through in-station acquisition terminals and online monitoring devices. The timestamps corresponding to each data point are retained, providing complete raw input data for subsequent sampling time alignment and load normalization processing.

[0014] Step S200: Using a preset synchronous trigger signal as a reference, the multi-source inspection data of each device in the same source device group are synchronized at the same sampling time to obtain multi-source aligned inspection data.

[0015] Specifically, using a preset synchronization trigger signal as the time reference, the timestamp sequence of load current data is extracted. The time interval between two adjacent load current samples is determined as the master clock reference period. For infrared thermal imaging data, the acquisition timestamp is extracted. If the timestamp does not fall within the time window of any master clock reference period, data points are interpolated between two adjacent load current data points before and after the timestamp. For dissolved gas chromatography data in oil, the acquisition timestamp is extracted. When the deviation between the acquisition timestamp and the time window of any master clock reference period exceeds a preset deviation threshold, the dissolved gas chromatography data in oil is associated with the alignment time window corresponding to the master clock reference period with the closest time distance. This completes the time dimension alignment processing of all equipment in the same equipment group from different data sources, and outputs multi-source aligned inspection data with unified timestamp matching, providing a time-synchronized dataset for subsequent load normalization processing.

[0016] Step S300: Based on the real-time load current and rated parameters of each primary device of the same type, the multi-source aligned inspection data is subjected to load normalization processing to obtain the normalized feature vector of each primary device of the same type.

[0017] Specifically, the real-time load current of each primary device of the same type is extracted from the multi-source aligned inspection data. The load rate is calculated by combining the rated current in the rated parameters of the device. The measured temperature value is extracted from the infrared thermal imaging data. Based on the preset temperature rise correction coefficient determined by the slope of different load rate intervals, the measured temperature is converted into the equivalent temperature under rated load conditions. The measured gas production rate of each gas component is extracted from the dissolved gas chromatography data in the oil. The equivalent gas production rate under the rated load benchmark is obtained by converting the square inverse of the load rate. Then, the equivalent top oil temperature, equivalent winding hot spot temperature, equivalent hydrogen gas production rate, equivalent acetylene gas production rate, equivalent total hydrocarbon gas production rate, cooler operation ratio, and ambient temperature and humidity compensation factor are arranged and combined according to the preset dimension order to generate the normalized feature vector corresponding to each primary device of the same type. This eliminates the data differences caused by different operating load conditions of the equipment and provides a unified feature input for the subsequent calculation of the group median vector.

[0018] Step S400: Calculate the group median vector of the normalized feature vector of all devices in the same source device group in each dimension, and subtract the normalized feature vector of each device from the group median vector to obtain the synchronous normalized residual vector of each primary device of the same type.

[0019] Specifically, for the normalized feature vectors of all devices within the same source device group under the same alignment time window, the dimensional feature values ​​corresponding to each device are extracted according to each dimension of the feature vector. After sorting all feature values ​​under each dimension, the median is calculated. The medians of each dimension are combined to form the group median vector. Then, the normalized feature vector of each primary device of the same type in the same source device group is subtracted from the group median vector dimension by dimension. The differences after subtraction of each dimension are combined in sequence to generate the corresponding normalized residual vector of each device. This represents the deviation of a single device from the overall operating level of the same group of devices, and provides residual basic data for subsequent calculation of residual deviation index.

[0020] Step S500: Calculate the residual deviation index of each device based on the synchronous normalized residual vector. When the residual deviation index of any target device exceeds the preset deviation threshold, and the residual deviation index of the other devices in the same source device group is in the normal range, it is determined that the target device has an early latent fault, and a graded early warning signal is sent.

[0021] Specifically, firstly, the standard deviation of the residuals in each dimension of the synchronous normalized residual vector of all equipment in the same source equipment group is calculated. Then, the absolute value of the residual component of each dimension of the synchronous normalized residual vector of each equipment is divided by the standard deviation of the residual in the corresponding dimension to obtain the standardized residual component. Then, the standardized residual components in each dimension are weighted and summed according to the preset weight coefficient to obtain the residual deviation index of each equipment. The percentage of the number of other equipment in the same source equipment group whose residual deviation index is lower than the upper limit of the normal range is counted. If the residual deviation index of the target equipment is greater than the first deviation threshold and less than or equal to the second deviation threshold, and the percentage of this number meets the preset proportion threshold, then an attention-level warning signal is issued. If the residual deviation index of the target equipment is greater than the second deviation threshold, and the residual deviation index of all other equipment is in the normal range, then a severe warning signal is issued, thus completing the early latent fault determination and graded warning signal issuance of equipment.

[0022] In one possible implementation, step S100 further includes: Step S110: The multi-source inspection data includes load current data and cooler status data, infrared thermal imaging data, dissolved gas chromatography data in oil, and ambient temperature and humidity data.

[0023] Specifically, load current data and cooler status data are collected through the SCADA system to obtain real-time electrical load and heat dissipation system operating conditions. This data carries a collection timestamp, which serves as a benchmark for subsequent data time alignment. Infrared thermal imaging data and environmental temperature and humidity data are collected by inspection robots or drones to obtain heating information of key equipment parts and on-site environmental conditions, which is used to eliminate the interference of environmental factors on equipment status judgment. Dissolved gas chromatography data in oil is collected by an online oil chromatography monitoring device configured in the station, which outputs the gas production rate of various characteristic gases to reflect the characteristics of potential insulation faults inside the equipment. All the collected multi-source inspection data are uniformly aggregated and stored, and their respective collection time information is completely preserved, providing multi-dimensional raw data support for subsequent sampling time alignment, load normalization processing, and fault analysis.

[0024] In one possible implementation, step S120 further includes: Step S120: The multiple primary devices of the same type in the same source equipment group are any one of the following: transformer, GIS combined electrical appliance, and high voltage switchgear, which are of the same model, have the same rated parameters, are connected to the same busbar or the same voltage level, and the difference in the commissioning years of each device does not exceed the preset age threshold.

[0025] Specifically, when constructing the equipment group, the primary equipment under the categories of transformers, GIS switchgear, and high-voltage switchgear in the substation is first screened. Multiple devices with completely identical models, the same rated electrical parameters, and physical connection positions on the same busbar or at the same voltage level are selected. Then, the commissioning time of each device is checked by retrieving the equipment ledger records. Devices with a difference in commissioning years exceeding the preset age threshold are eliminated. The remaining devices form a homogeneous equipment group. This avoids the benchmark comparison deviation caused by differences in equipment models, inconsistent parameters, different voltage levels, and significant differences in aging, ensuring the effectiveness of the subsequent group median benchmark calculation and providing conditions for horizontal comparison of fault characteristics between devices.

[0026] In one possible implementation, step S200 further includes: Step S210: Extract the timestamp sequence of the load current data, and use the time interval between two adjacent samples in the timestamp sequence as the master clock reference period.

[0027] Step S220: For the infrared thermal imaging data, extract the acquisition timestamp. When the acquisition timestamp does not fall within the time window of any master clock reference period, add interpolation points between two adjacent load current data points before and after the timestamp.

[0028] Step S230: For dissolved gas chromatographic data in oil, extract the acquisition timestamp. When the time window deviation between the acquisition timestamp and any master clock reference period exceeds a preset deviation threshold, associate the dissolved gas chromatographic data with the alignment time window corresponding to the master clock reference period with the closest time distance.

[0029] Specifically, the full load current data output by the SCADA system is read, and the timestamp corresponding to each load current data is extracted one by one. The timestamps are sorted in chronological order to form a complete timestamp sequence. The time difference between two adjacent timestamps in the sequence is calculated in turn, and the sampling interval is determined as the master clock reference period. This establishes a unified time reference for the entire set of multi-source inspection data alignment processing. Subsequent inspection data of other types are divided into corresponding time windows with reference to this master clock reference period, providing a unified time scale for time synchronization alignment processing between different data sources.

[0030] The raw infrared thermal imaging data is analyzed one by one, and the corresponding acquisition timestamp is extracted. The timestamp is then compared with each time window divided based on the master clock reference period. If the acquisition timestamp does not fall within the coverage of any time window, the two load current sampling data points closest to the timestamp are located. An interpolation algorithm is used to generate supplementary interpolation points between the two load current data points, mapping the infrared thermal imaging observations to the master clock time system. This completes the time adaptation of the infrared thermal imaging data, ensuring that this type of data can match the time windows divided by the master clock, and providing a data foundation for the synchronous alignment of multi-source inspection data.

[0031] The dissolved gas chromatography data in the oil are analyzed line by line, and the acquisition timestamp corresponding to each record is extracted. The time deviation between the acquisition timestamp and the alignment time window corresponding to each master clock reference period is calculated. The calculated time deviation is compared with the pre-configured deviation threshold. If the time deviation exceeds the preset deviation threshold, all alignment time windows are traversed, and the target alignment time window with the smallest time distance from the acquisition timestamp is selected. The dissolved gas chromatography data in the oil is then bound to the target alignment time window, realizing the mapping and adaptation of low-frequency sampled oil chromatography data to the master clock time system. This completes the time alignment processing of this type of data and ensures that multi-source inspection data can be subjected to subsequent normalization operations in the same time dimension.

[0032] In one possible implementation, step S300 further includes: Step S310: Extract the real-time load current value of each device from the multi-source aligned inspection data, calculate the ratio of the real-time load current value to the rated current value in the rated parameters, and obtain the load rate.

[0033] Step S320: Extract the measured temperature value from the infrared thermal imaging data in the multi-source aligned inspection data, multiply it by the preset temperature rise correction coefficient corresponding to the load rate, and obtain the equivalent temperature converted to the rated load. The preset temperature rise correction coefficient is determined according to the slope of the historical temperature rise curve of the equipment in different load rate ranges.

[0034] Step S330: Extract the measured gas production rate of each gas component from the dissolved gas chromatography data in the multi-source aligned inspection data, multiply it by the square inverse of the load rate, and obtain the equivalent gas production rate converted to the rated load benchmark.

[0035] Step S340: Arrange the equivalent temperature, equivalent gas production rate, cooler status data and ambient temperature and humidity data in a preset dimension order to obtain the normalized feature vector of each primary device of the same type.

[0036] Specifically, from the multi-source aligned inspection data after time synchronization, the real-time load current measured values ​​of a single primary device of the same type under the same aligned time window are read according to the device identifier. The rated current parameters stored in the equipment's factory ledger are retrieved. The real-time load current value is used as the numerator and the rated current value is used as the denominator to perform a division operation. The ratio of the two is used as the load rate of the equipment at the current moment. The load rate is used to characterize the actual load level of the equipment at the current moment. The calculation result will be used as the core input parameter for subsequent load conversion and normalization of infrared temperature and oil chromatography gas production rate.

[0037] The measured temperature values ​​of the equipment obtained by infrared thermal imaging under the corresponding alignment time window are extracted from the multi-source aligned inspection data. The current load rate of the equipment calculated in the previous step is retrieved. Based on the historical temperature rise curve of the equipment, the intervals are pre-divided according to the temperature rise slope of different load rate intervals and segmented preset temperature rise correction coefficients are configured. The corresponding temperature rise correction coefficient is matched according to the interval of the current load rate. The measured temperature value is directly multiplied with the matched temperature rise correction coefficient, and the equivalent temperature converted to the rated load condition is calculated and output. This eliminates the operating condition interference caused by different real-time load sizes on the equipment temperature measurement results and provides temperature characteristics under a unified load benchmark for constructing normalized feature vectors.

[0038] From the multi-source aligned inspection data that has been aligned to the completion time, the measured gas production rates of each faulty gas component, such as hydrogen, acetylene, and total hydrocarbons, are read from the dissolved gas chromatography data of oil under the same aligned time window. The current load rate of the equipment is calculated, and the load rate is first squared and then the reciprocal is obtained. The measured gas production rate of each gas is multiplied by the reciprocal of the square, and finally the equivalent gas production rate of each gas component is obtained under the rated load benchmark. This eliminates the influence of the actual operating load of the equipment on the gas production rate value, so that the oil chromatography data collected under different load conditions have the conditions for horizontal comparison, and provides standardized oil chromatography feature parameters for the subsequent construction of normalized feature vectors.

[0039] The equivalent temperature and equivalent gas production rate of each gas component are obtained after load conversion. Then, the cooler status data and ambient temperature and humidity data under the same aligned time window are read. According to the pre-set dimension arrangement rules, the equivalent top oil temperature, equivalent winding hot spot temperature, equivalent hydrogen production rate, equivalent acetylene production rate, equivalent total hydrocarbon production rate, cooler operating group ratio, and ambient temperature and humidity compensation factor are arranged in an orderly combination. The scalar features processed above are used as the various dimension components of the vector to assemble the normalized feature vector of the corresponding single primary equipment of the same type. The data dimension and arrangement order of the features of each equipment are unified, the data deviation caused by the difference in load conditions is eliminated, and a structured feature input with a unified format is provided for subsequent calculation of the population median vector and solution of the residual vector.

[0040] In one possible implementation, step S340 further includes: Step S341: The dimensions of the normalized feature vector include: equivalent top oil temperature, equivalent winding hot spot temperature, equivalent hydrogen production rate, equivalent acetylene production rate, equivalent total hydrocarbon production rate, ratio of cooler operating groups, and ambient temperature and humidity compensation factor.

[0041] Specifically, the dimensions of the normalized feature vector include: equivalent top-layer oil temperature, equivalent winding hot spot temperature, equivalent hydrogen production rate, equivalent acetylene production rate, equivalent total hydrocarbon production rate, cooler operating group ratio, and ambient temperature and humidity compensation factor. This normalized feature vector has seven fixed dimensions. The equivalent top-layer oil temperature and equivalent winding hot spot temperature are temperature characteristics after load conversion, used to characterize the heating level at different locations on the equipment body. The equivalent hydrogen production rate, equivalent acetylene production rate, and equivalent total hydrocarbon production rate are gas characteristics after load conversion from oil chromatography data. The system reflects the gas generation caused by internal insulation defects in the equipment; the ratio of the number of coolers in operation is calculated by dividing the actual number of coolers in operation by the total number of coolers, representing the actual working level of the heat dissipation system; the environmental temperature and humidity compensation factor is calculated by the collected environmental temperature and humidity according to the preset compensation formula, used to offset the interference of the external environment on the equipment operating parameters. The above seven features are used as vector components in a fixed order, and each source device generates a feature vector with completely consistent dimension number and arrangement order, ensuring that the features of each dimension among the source device groups can be compared and calculated dimension by dimension.

[0042] In one possible implementation, step S500 further includes: Step S510: Calculate the residual standard deviation of the synchronous normalized residual vector of all devices in the same source device group in each dimension.

[0043] Step S520: Divide the absolute value of the residual component in each dimension of the synchronous normalized residual vector of each device by the residual standard deviation to obtain the standardized residual component of each device.

[0044] Step S530: The standardized residual components of each dimension are weighted and summed according to the preset weight coefficients to obtain the residual deviation index of each device.

[0045] Specifically, for the synchronous normalized residual vector of all devices in the same source device group under the same alignment time window, the residual components of all devices in each dimension of the vector are extracted. First, the arithmetic mean of all residual components in that dimension is calculated. Then, the difference between each residual component and the arithmetic mean is calculated and the square value is obtained. The sum of all squared results is divided by the total number of devices to obtain the variance of that dimension. Then, the square root operation is performed on the variance to obtain the standard deviation of the residual corresponding to that dimension. The standard deviation of the residual for each dimension is solved by traversing all feature dimensions. The standard deviation of the residual is used to characterize the dispersion of the residual components of the same source device group in the corresponding dimension, providing a calculation benchmark for the subsequent standardization conversion of the residual components in each dimension.

[0046] Iterates through each device in the same device group and all dimensions corresponding to the normalized feature vector. Extracts the residual component of the normalized residual vector of a single device in the current dimension, calculates the absolute value of the residual component, retrieves the calculated residual standard deviation for that dimension, and performs division with the absolute value of the residual component as the dividend and the residual standard deviation as the divisor to obtain the standardized residual component of the current dimension of the device. This process is repeated for all dimensions of the device, and then iterates through all dimensions of all devices in the device group. By dividing by the residual standard deviation of the dimension, the dimensionless transformation of features with different dimensions is achieved, eliminating the influence of inconsistent data dispersion in each dimension, and providing standardized input data for the weighted calculation of the residual deviation index.

[0047] Based on massive historical operation and fault sample data of equipment, the ability of seven feature dimensions to identify and distinguish latent faults and abnormal defects of different equipment is quantitatively calculated. Preset weight coefficients are matched according to the fault discrimination level of each dimension; the higher the fault discrimination level, the larger the weight coefficient assigned to the dimension. The sum of the weight coefficients of each dimension after normalization is a fixed value. For a single device within a group of source equipment, the standardized residual components corresponding to all dimensions are extracted sequentially. The standardized residual component of each dimension is multiplied one by one with the preset weight coefficient of the corresponding dimension to obtain the weighted residual components of each dimension. Then, the weighted residual components of all dimensions are summed. By weightedly fusing the standardized residual features of multiple dimensions, the final residual deviation index of the device is calculated, achieving accurate quantification of the degree to which the operating status of a single device deviates from the normal level of the group. This provides accurate quantitative indicator support for subsequent abnormal fault judgment and graded early warning of equipment.

[0048] In one possible implementation, step S500 further includes: Step S540: Calculate the percentage of devices in the same group whose residual deviation index is lower than the upper limit of the preset normal range.

[0049] Step S550: When the residual deviation index of the target device is greater than the first deviation threshold and less than or equal to the second deviation threshold, and the proportion of the quantity exceeds the preset ratio threshold, a warning signal at the attention level is sent.

[0050] Step S560: When the residual deviation index of the target device is greater than the second deviation threshold and the proportion of the number reaches the total number of all other devices, a severe warning signal is sent.

[0051] Wherein, the second deviation threshold is greater than the first deviation threshold.

[0052] Specifically, the target device to be judged is removed from the same source device group to obtain the remaining set of devices for comparison. Each device in this set is traversed, and the residual deviation index of each device is read one by one. Each residual deviation index is compared with the upper limit of the pre-set normal range. The number of devices with residual deviation index values ​​less than the upper limit of the normal range is counted. The number of qualified devices obtained from this count is divided by the total number of devices in the set after removing the target device to calculate the percentage of the remaining devices with residual deviation index values ​​lower than the upper limit of the preset normal range. This percentage is used to characterize whether the overall operating status of the same source device group is stable, thereby eliminating misjudgments caused by fluctuations in the overall operating conditions of the device group and providing statistical basis for subsequent latent fault judgment.

[0053] The system retrieves the residual deviation index of the target device, which has already been calculated, and simultaneously reads the percentage of other devices. It then compares the residual deviation index of the target device with the pre-configured first and second deviation thresholds to determine if it meets the condition of being greater than the first deviation threshold and less than or equal to the second deviation threshold. Next, it checks if the percentage of other devices whose residual deviation index is below the upper limit of the normal range is greater than a preset percentage threshold. Only when both conditions are met can the system determine that the target device has a minor potential anomaly. Subsequently, a warning signal at the attention level is generated and issued. This warning logic avoids false alarms caused by fluctuations in the overall operating conditions of the same group of devices by combining the degree of anomaly of a single device with the overall status of the group, and reminds maintenance personnel to continuously monitor the target device.

[0054] The residual deviation index calculated from the target device and the corresponding quantity ratio of other devices in the same source device group are retrieved. The pre-configured second deviation threshold is set to be greater than the first deviation threshold. First, the residual deviation index of the target device is compared with the second deviation threshold to determine whether the residual deviation index of the target device exceeds the second deviation threshold. At the same time, it is verified whether the quantity ratio of the residual deviation index of the other devices that is lower than the upper limit of the preset normal range reaches the total number of the other devices. That is, all devices in the same source device group except the target device are in normal condition. If both conditions are met, it is determined that the anomaly is only from the target device itself and is not caused by the overall operating condition fluctuation of the device group. Then, a severe warning signal is generated and output to remind the operation and maintenance personnel that the target device has a high-risk latent defect and needs to be dealt with in a timely manner. By relying on the dual constraints of single device anomaly index and group status, false warnings caused by group disturbances are effectively reduced.

[0055] In one possible implementation, step S500 further includes: Step S570: While sending the graded early warning signal, record the identifier of the target device that triggered the early warning and the corresponding alignment time window number.

[0056] Step S580: Send a re-inspection instruction for the target device to the inspection equipment, the re-inspection instruction including a shortened re-inspection time interval.

[0057] Specifically, at the same moment the system outputs a warning signal at the attention level or severity level, the unique device identifier of the target device that triggered the warning is read, and the alignment time window number corresponding to this operation is obtained. The device identifier, the alignment time window number and the warning information are associated and bound together and written into the storage unit for complete retention. This achieves a one-to-one correspondence between the warning event and the device object and the data time dimension, which facilitates the subsequent backtracking and retrieval of the complete set of multi-source inspection data under this time window, supporting post-fault review, problem tracing and maintenance log generation.

[0058] When a target device triggers a tiered warning, a re-inspection command is issued to the inspection robot or drone. This re-inspection command carries a shortened re-inspection time interval compared to the regular inspection cycle. When the timer reaches this re-inspection time interval, the system receives the newly collected multi-source inspection data of the target device from the inspection robot or drone. This newly collected multi-source inspection data replaces the original multi-source inspection data corresponding to the target device. The selection range of the same-source device group remains unchanged. The entire processing flow, including synchronous sampling time alignment, load normalization, normalized feature vector calculation, group median vector calculation, synchronous normalized residual vector calculation, and residual deviation index calculation, is then executed again in sequence. The re-inspection residual deviation index is calculated, and a second comparison is performed between the re-inspection residual deviation index and the preset deviation threshold. At the same time, it is verified whether the residual deviation index of other equipment in the same equipment group is within the normal range. When the re-inspection residual deviation index still exceeds the preset deviation threshold and the conditions that all other equipment is within the normal range are met, it is determined that the anomaly is not caused by instantaneous on-site disturbance. The previously issued graded warning signal is maintained, and a maintenance suggestion work order is generated. The target equipment identifier, anomaly dimension list and suggested maintenance type are entered in the work order, and the maintenance suggestion work order is pushed to the operation and maintenance terminal. This completes the closed-loop verification of the warning re-inspection and effectively reduces false alarms caused by instantaneous interference.

[0059] In one possible implementation, step S580 further includes: Step S581: Within the preset cancellation time window after the issuance of the graded warning signal, if any of the following mutually exclusive contradictory proof conditions are met, the sent graded warning signal will be automatically cancelled.

[0060] Step S582: The operation log shows that the target device performed on-load tap changer or cooler switching operations within the preset operation time window before the warning.

[0061] Step S583: After the re-acquired infrared thermal imaging data is aligned with the synchronous sampling time and normalized by the load, the corresponding normalized feature vector has fallen within the preset percentage range of the dimension value corresponding to the population median vector.

[0062] Step S584: The load current data confirms that the target equipment load experienced a step change exceeding a preset percentage during the warning period.

[0063] Step S585: After the warning is revoked, the weight of the data source corresponding to the mutually exclusive proof condition that triggered the revocation will be temporarily reduced in the subsequent calculation of the residual deviation index.

[0064] Specifically, the timing starts when the graded early warning signal is output, and a pre-configured preset cancellation time window is initiated. During the validity period of this time window, the system continuously monitors and judges the counter-evidence conditions. The counter-evidence conditions are set to mutual exclusion logic. As long as any counter-evidence condition is detected, it can be determined that the early warning was caused by factors other than equipment defects, such as instantaneous interference or temporary operating condition fluctuations. Then, the system automatically cancels the warning, clears the issued graded early warning signal, and records the early warning cancellation time, the type of counter-evidence condition that triggered the cancellation, and the target equipment identifier in the system log to avoid maintenance personnel performing ineffective operations on false alarms.

[0065] The system reads the equipment operation and maintenance log database to locate the trigger time of this graded warning. It then traces back a preset operation time window from the warning trigger time to retrieve all historical operation records of the target equipment within that time window. Each record is compared with the operation type field in the log to determine whether there are on-load tap changer operation records or cooler switching operation records. If the corresponding operation record is found, it is determined that the abnormal fluctuation of the indicator is caused by the equipment's operation disturbance due to human operation, which is a mutually exclusive proof condition. The operation event and the time of operation are written into the warning association log as the basis for revoking the graded warning signal.

[0066] The infrared thermal imaging data of the target equipment, which was re-acquired by the inspection equipment, is retrieved. The data is then aligned with the sampling time, and the load normalization calculation is performed to assemble the updated normalized feature vector. Each dimension component of the normalized feature vector is compared with the corresponding dimension value of the median vector of the same group of equipment. It is determined whether the values ​​of all or key dimensions of the updated feature vector fall within the preset percentage fluctuation range of the corresponding dimension value of the median vector of the group. If the value range condition is met, it indicates that the relevant characteristics of the target equipment have returned to the normal fluctuation range of the group. This situation serves as a mutually exclusive proof condition. The data comparison results are stored in the early warning association log to trigger the automatic cancellation of the graded early warning signal.

[0067] Extract the load current time-series data of the target equipment during the warning period from the start time of the warning to the current time. Select the load current at the start time of the warning as the reference current value. Calculate the maximum change in load current during the warning period. Divide the maximum change by the reference current value to obtain the load change rate. Compare the calculated load change rate with a preset percentage. When the load change rate is greater than the preset percentage, it is determined that the equipment has experienced a load step disturbance. This operating condition fluctuation belongs to the mutually exclusive proof by contradiction condition. Record the reference current, maximum load current, load change rate, and occurrence time to the warning association log, which will serve as the basis for automatically canceling the graded warning signal.

[0068] After completing the automatic cancellation of the graded early warning, the specific mutually exclusive counter-evidence conditions that triggered the early warning cancellation are first accurately matched. The original monitoring data source and corresponding feature dimension weights corresponding to the counter-evidence conditions are located. For data sources whose abnormal fluctuations are not due to equipment failures but caused by external operating conditions or operational disturbances, the preset weight coefficients corresponding to these data sources in the residual deviation index calculation are temporarily lowered. The weight coefficients of other normal data sources remain unchanged. The temporary weight adjustment only applies to the calculation process of the same type of equipment index for a period of time in the future. This is used to weaken the interference of such easily disturbed data sources on the judgment results and avoid false early warnings caused by fluctuations in the same operating conditions. At the same time, the data source type, adjustment range, effective duration, and adjustment time of this weight reduction are completely recorded in the system log to ensure that the model parameter adjustment is traceable and recoverable.

[0069] It should be noted that the order of the embodiments described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. Specific embodiments of this specification have been described above. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0070] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0071] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A fault early warning method based on multi-source data fusion inspection of substation operation, characterized in that, The method includes: Acquire multi-source inspection data of multiple primary devices of the same type connected to the same voltage level busbar in the same substation, wherein the multiple primary devices of the same type constitute a group of devices of the same source. Using a preset synchronous trigger signal as a reference, the multi-source inspection data of each device in the same source device group are synchronized at the same sampling time to obtain multi-source aligned inspection data; Based on the real-time load current and rated parameters of each primary device of the same type, the multi-source aligned inspection data is subjected to load normalization processing to obtain the normalized feature vector of each primary device of the same type. Calculate the group median vector of the normalized feature vector of all devices in the same source device group in each dimension, and subtract the normalized feature vector of each device from the group median vector to obtain the synchronous normalized residual vector of each primary device of the same type. Based on the synchronous normalized residual vector, the residual deviation index of each device is calculated. When the residual deviation index of any target device exceeds the preset deviation threshold, and the residual deviation index of the other devices in the same source device group is in the normal range, it is determined that the target device has an early latent fault, and a graded early warning signal is sent.

2. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 1, characterized in that, The multi-source inspection data includes load current data and cooler status data, infrared thermal imaging data, dissolved gas chromatography data in oil, and ambient temperature and humidity data.

3. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 1, characterized in that, The multiple primary devices of the same type in the same equipment group are any one of the following: transformers, GIS combined electrical appliances, and high-voltage switchgear of the same model, with the same rated parameters, connected to the same busbar or at the same voltage level, and the difference in the commissioning years of each device does not exceed the preset age threshold.

4. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 2, characterized in that, Based on a preset synchronization trigger signal, the multi-source inspection data of each device in the same device group are synchronized at the same sampling time, including: Extract the timestamp sequence of the load current data, and use the time interval between two adjacent samples in the timestamp sequence as the master clock reference period; For the infrared thermal imaging data, the acquisition timestamp is extracted. When the acquisition timestamp does not fall within the time window of any master clock reference cycle, interpolation points are added between two adjacent load current data points before and after the timestamp. For dissolved gas chromatographic data in oil, the acquisition timestamp is extracted. When the time window deviation between the acquisition timestamp and any master clock reference period exceeds a preset deviation threshold, the dissolved gas chromatographic data is associated with the alignment time window corresponding to the master clock reference period with the closest time distance.

5. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 1, characterized in that, Based on the real-time load current and rated parameters of each primary device of the same type, the multi-source aligned inspection data is subjected to load normalization processing, including: Extract the real-time load current value of each device from the multi-source aligned inspection data, calculate the ratio of the real-time load current value to the rated current value in the rated parameters, and obtain the load rate; For the infrared thermal imaging data in the multi-source aligned inspection data, the measured temperature value is extracted and multiplied by the preset temperature rise correction coefficient corresponding to the load rate to obtain the equivalent temperature converted to the rated load. The preset temperature rise correction coefficient is determined according to the slope of the equipment's historical temperature rise curve in different load rate ranges. For the dissolved gas chromatographic data in the oil from the multi-source aligned inspection data, the measured gas production rate of each gas component is extracted, and multiplied by the square inverse of the load rate to obtain the equivalent gas production rate converted to the rated load benchmark. The equivalent temperature, equivalent gas production rate, cooler status data, and ambient temperature and humidity data are arranged in a preset dimension order to obtain the normalized feature vector of each primary device of the same type.

6. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 5, characterized in that, The dimensions of the normalized feature vector include: Equivalent top oil temperature, equivalent winding hot spot temperature, equivalent hydrogen production rate, equivalent acetylene production rate, equivalent total hydrocarbon production rate, ratio of cooler operating groups, and ambient temperature and humidity compensation factor.

7. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 1, characterized in that, The residual deviation index for each device is calculated based on the synchronously normalized residual vector, including: Calculate the residual standard deviation of the synchronous normalized residual vector of all devices in the same source device group in each dimension; The absolute value of the residual component in each dimension of the synchronous normalized residual vector of each device is divided by the residual standard deviation to obtain the standardized residual component of each device. The standardized residual components of each dimension are weighted and summed according to preset weighting coefficients to obtain the residual deviation index for each device.

8. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 7, characterized in that, If the target device is determined to have an early latent fault, a tiered early warning signal is sent, including: The percentage of devices in the same equipment group whose residual deviation index is lower than the upper limit of the preset normal range is calculated. When the residual deviation index of the target device is greater than the first deviation threshold and less than or equal to the second deviation threshold, and the proportion of the quantity exceeds the preset ratio threshold, an attention-level warning signal is sent. When the residual deviation index of the target device is greater than the second deviation threshold, and the proportion of such deviations reaches the total number of all other devices, a severe warning signal is sent. Wherein, the second deviation threshold is greater than the first deviation threshold.

9. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 1, characterized in that, After issuing a tiered warning signal, it also includes: While sending the graded early warning signal, record the identifier of the target device that triggered the early warning and the corresponding alignment time window number; Send a re-inspection instruction to the inspection equipment for the target equipment, the re-inspection instruction including a shortened re-inspection time interval.

10. The fault early warning method based on multi-source data fusion inspection of substation operation as described in claim 2, characterized in that, Within the preset cancellation time window after issuing a tiered warning signal, if any of the following mutually exclusive contradictory proof conditions are met, the sent tiered warning signal will be automatically cancelled: The operation log shows that the target device performed on-load tap changing or cooler switching operations within the preset operation time window before the warning. After the re-acquired infrared thermal imaging data is aligned with the synchronous sampling time and normalized by the load, the corresponding normalized feature vector has fallen within the preset percentage range of the dimension value corresponding to the median vector of the population. The load current data confirms that the target equipment load experienced a step change exceeding a preset percentage during the warning period; After the warning is revoked, the weight of the data source corresponding to the mutually exclusive proof condition that triggered the revocation will be temporarily reduced in the subsequent calculation of the residual deviation index.

Images