A power distribution network equipment state diagnosis and prediction method based on big data analysis

Abstract

The application discloses a power distribution network equipment state diagnosis and prediction method based on big data analysis, relates to the intelligent operation and maintenance technical field of a power distribution network, and comprises the following steps: S1, collecting multi-source equipment state related data to generate a power distribution network equipment state original data set; S2, performing standardization processing to generate a power distribution network equipment state feature set; S3, constructing an equipment state portrait and completing state diagnosis; S4, reconstructing equipment degradation segments to generate an equipment degradation segment sequence; S5, performing state trend prediction through an improved PatchTST prediction network; S6, performing degradation risk correction in combination with equipment degradation related information; and S7, generating power distribution network equipment state diagnosis and prediction decision information. The application adopts equipment degradation segment reconstruction and an improved PatchTST prediction network to realize accurate diagnosis and early risk prediction of the power distribution network equipment state.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power distribution networks, and in particular to a method for diagnosing and predicting the condition of power distribution network equipment based on big data analysis. Background Technology

[0002] With the continuous expansion of power distribution networks, the rapid integration of distributed power sources and electric vehicle loads, and the increasing demand for intelligent operation and maintenance of power distribution equipment, multi-source data analysis, condition diagnosis, and fault risk prediction technologies for the operating status of power distribution network equipment have received widespread attention. Existing methods for managing the condition of power distribution network equipment mainly rely on operating threshold alarms, periodic inspection records, or single time series prediction models for equipment anomaly identification. However, these methods commonly suffer from the following problems in practical applications: Data on power distribution network equipment operation, status monitoring, defect repair, fault alarm, environmental conditions, and load changes are scattered across different sources, with inconsistent data encoding, sampling periods, and time bases. Existing methods struggle to uniformly correlate the status records of the same power distribution network equipment across different systems, resulting in incomplete equipment profiles and a lack of continuity in diagnostic results. Traditional threshold alarm methods typically only trigger alarms when indicators such as voltage, current, temperature, or load rate exceed limits, making it difficult to identify the gradual progression of minor fluctuations into continuous degradation. This can lead to delayed risk detection and insufficient pre-fault warning time. Existing time series forecasting methods often input equipment status data according to fixed time windows, easily mixing normal operation segments, abnormal fluctuation segments, degradation development segments, and maintenance recovery segments in the model, resulting in inaccurate representation of equipment status change trends. Furthermore, forecast results are often not corrected for degradation correlation information such as historical defects, maintenance recovery, and load changes, affecting the reliability of power distribution network equipment status forecasts and the relevance of maintenance decisions.

[0003] Therefore, how to provide a method for the diagnosis and prediction of the condition of power distribution network equipment based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a method for diagnosing and predicting the status of distribution network equipment based on big data analysis. This invention employs a combination of equipment degradation segment reconstruction and an improved PatchTST prediction network. By uniformly processing, standardizing modeling, and constructing equipment status profiles from multi-source equipment status correlation data of distribution network equipment, continuous diagnosis of the current operating status of distribution network equipment is achieved. Based on this, degradation segments are reconstructed from the equipment status diagnosis results and equipment status correlation characteristics, distinguishing different state processes such as normal operation, abnormal fluctuations, degradation development, and maintenance recovery. The equipment degradation segment sequence is then input into the improved PatchTST prediction network for status trend prediction, avoiding prediction bias caused by mixing different state stages in traditional fixed-time-window prediction methods. Simultaneously, degradation risk correction is performed on the basic equipment status prediction results using equipment degradation correlation information, improving the consistency between the prediction results and the actual degradation process. This method has advantages such as strong diagnostic continuity, high prediction lead time, accurate risk identification, and applicability to intelligent operation and maintenance of distribution network equipment.

[0005] A method for diagnosing and predicting the condition of power distribution network equipment based on big data analysis, according to an embodiment of the present invention, includes the following steps: S1. Collect multi-source device status association data during the operation of power distribution network equipment, and process the multi-source device status association data in a unified manner to generate a raw dataset of power distribution network equipment status. S2. Standardize the original dataset of the distribution network equipment status and extract the equipment status correlation features to generate a distribution network equipment status feature set. S3. Construct a device status profile based on the device status feature set of the distribution network equipment, and perform status diagnosis on the distribution network equipment based on the device status profile to generate a set of distribution network equipment status profiles and device status diagnosis results. S4. Based on the distribution network equipment status feature set and the equipment status diagnosis results, reconstruct the equipment degradation segment of the distribution network equipment status time series to generate an equipment degradation segment sequence. S5. Input the equipment degradation fragment sequence and the distribution network equipment status profile set into the improved PatchTST prediction network to predict the trend of distribution network equipment status changes and generate basic equipment status prediction results. S6. Extract equipment degradation correlation information based on the equipment status basic prediction results, and perform degradation risk correction based on the equipment degradation correlation information to generate distribution network equipment status prediction results. S7. Based on the equipment status diagnosis results and the distribution network equipment status prediction results, generate distribution network equipment status diagnosis and prediction decision information.

[0006] Optionally, S1 specifically includes: Acquire basic equipment data, operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data, and load change data during the operation of power distribution network equipment, and generate multi-source equipment status correlation data; Read the equipment name, equipment type, voltage level, feeder, transformer area, installation location and asset number from the multi-source equipment status association data, and perform unified coding according to the preset equipment coding rules to generate unified equipment coding results; Read the data acquisition system, data acquisition interface, data record type and data update time corresponding to the multi-source device status association data, and mark the data records from different sources to generate data source identification results; Read the acquisition time, detection time, defect registration time, maintenance completion time, fault occurrence time and load change time from the multi-source equipment status association data, and convert the time format according to a unified time base to generate time stamp results; Based on the unified coding results of the equipment, the operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data and load change data corresponding to the same distribution network equipment are bound to the equipment affiliation to generate equipment status association records; The unified coding result of the device, the data source identification result, the time stamp result, and the device status association record are written into the multi-source device status association data to generate the original dataset of distribution network device status.

[0007] Optionally, S2 specifically includes: Read the device unified code field, timestamp field, data source field and device status record field from the original data set of the distribution network equipment status, and generate a data set of equipment status to be standardized. The device status dataset to be standardized is subjected to field integrity verification, duplicate record merging, format error correction and invalid record removal to generate a cleaned device status dataset. Based on the rated parameters, operating limits, defect registration time, maintenance completion time, and fault occurrence time of the power distribution network equipment, anomaly verification is performed on the cleaned equipment status dataset to generate a verified equipment status dataset. The operation monitoring data, status detection data, defect repair data, fault alarm data and load change data in the verified equipment status dataset are unified in terms of units and converted in terms of time granularity to generate a standardized equipment status dataset. Based on the unified device coding field and the timestamp field, the standardized device status dataset is time-series aligned to generate device status time-series data. Based on the equipment status time series data, extract equipment status correlation features to characterize the current status, historical status changes, and operational disturbance impacts of distribution network equipment; The device status association features are associated and organized according to the device unified coding field, time stamp field, and data source field to generate a distribution network device status feature set.

[0008] Optionally, S3 specifically includes: Read the status feature set of the power distribution network equipment, and collect it according to the unified equipment coding result and time stamp result corresponding to the power distribution network equipment to generate equipment profile construction input data; Based on the equipment profile, the input data is used to determine the basic attributes of the distribution network equipment, the state-related feature sequence, the state change record, and the data source record, and a set of equipment profile fields is generated. Based on the set of equipment profile fields, the associated features of the equipment status of the same power distribution network equipment at different time nodes are associated and organized to generate an equipment status profile. The equipment status profiles of each distribution network device are classified and organized according to equipment type, feeder, transformer area and unified equipment coding results to generate a set of distribution network equipment status profiles. Read the equipment status profiles in the distribution network equipment status profile set, and determine the corresponding operating deviation status, defect association status and status change trend of the distribution network equipment based on the equipment status profiles, and generate equipment status diagnosis input data; Based on the equipment status diagnosis input data, the status level of the power distribution network equipment is determined and the anomaly category is identified, generating a single equipment status diagnosis result. The individual device status diagnosis results corresponding to each distribution network device are written into the distribution network device status profile set according to the unified device coding results, and then summarized and organized to generate device status diagnosis results.

[0009] Optionally, S4 specifically includes: Read the status feature set of the power distribution network equipment and the status diagnosis results of the equipment, and sort them according to the unified equipment coding results and time stamp results corresponding to the power distribution network equipment to generate equipment degradation fragment reconstruction input data; Based on the input data reconstructed from the equipment degradation fragments, the changes in equipment state association features of the same power distribution network equipment at continuous time nodes are extracted, and a sequence of equipment state changes is generated. Based on the equipment status diagnosis results, determine the status level changes, abnormal continuous status and abnormal recovery status of the power distribution network equipment at different time points, and generate a status diagnosis label sequence; The equipment state change sequence and the state diagnosis mark sequence are time-correlated to determine the state stability interval, state fluctuation interval, state degradation interval and state recovery interval in the distribution network equipment state time series, and a set of equipment state segment boundaries is generated. The distribution network equipment state time series is segmented according to the equipment state segment boundary set, and each segmented state segment is labeled with a segment type to generate an initial equipment state segment set. Based on the segment length, state change amplitude, abnormal duration and adjacent segment connection relationship of each state segment in the initial device state segment set, state segments that do not meet the preset segment length requirements are merged or supplemented to generate a reconstructed device state segment set. The reconstructed set of device state fragments is arranged in order according to the unified device coding result and the time stamp result to generate a sequence of device degradation fragments.

[0010] Optionally, S5 specifically includes: Read the equipment degradation fragment sequence and the distribution network equipment status profile set, and match them according to the unified equipment coding result and time stamp result to generate PatchTST prediction input data; Based on the PatchTST prediction input data, determine the equipment degradation segment sequence, equipment status profile and target prediction time range for the same distribution network equipment, and generate the equipment status prediction task. The device state prediction task is input into the improved PatchTST prediction network; The improved PatchTST prediction network is used to embed and predict the degradation fragment sequence of the device in time to generate a basic prediction result for the single device state. The basic prediction results of the individual equipment status corresponding to each distribution network equipment are summarized and organized according to the unified equipment coding results to generate the basic prediction results of equipment status.

[0011] Optionally, the step of embedding and temporally predicting the device degradation fragment sequence using the improved PatchTST prediction network to generate a single-device state-based prediction result specifically involves: The improved PatchTST prediction network includes a degenerate fragment input layer, a fragment attribute encoding layer, a degenerate fragment embedding layer, a device profile constraint layer, a time series prediction encoding layer, and a state prediction output layer. The degradation fragment input layer reads the equipment degradation fragment sequence in the equipment state prediction task, and determines the degradation fragment arrangement order corresponding to the same distribution network equipment according to the unified equipment coding result and time stamp result, thereby generating the degradation fragment input sequence. The fragment attribute encoding layer extracts fragment boundary information, fragment state information, and fragment duration information corresponding to each device degradation fragment from the degradation fragment input sequence, and generates degradation fragment attribute encoding results. The degenerate fragment embedding layer performs embedding processing on the degenerate fragment input sequence and the degenerate fragment attribute encoding result to generate a degenerate fragment embedding representation; The device state profile in the device state prediction task is read through the device profile constraint layer, and the device state constraint is matched with the degenerate fragment embedding representation based on the device state profile to generate a profile constraint fragment representation. The temporal prediction coding layer performs cross-fragment temporal modeling on the portrait constraint fragment representation to determine the state continuity relationship and state change trend of the same distribution network equipment in different equipment degradation fragments, and generates a temporal prediction representation of equipment state. The state prediction output layer outputs a basic prediction result for a single device state based on the device state time-series prediction representation and the target prediction time range.

[0012] Optionally, S6 specifically includes: Read the basic prediction results of the equipment status, and extract data records related to the degradation process of the distribution network equipment from the original dataset of the distribution network equipment status, the feature set of the distribution network equipment status, the equipment status diagnosis results, and the distribution network equipment status profile set to generate equipment degradation association information; Based on the unified coding results and time stamp results of the equipment, the basic prediction results of the equipment status are matched with the equipment degradation association information to generate degradation risk correction input data; Based on the degradation risk correction input data, the degradation risk correction is performed on the basic equipment status prediction results to generate single equipment status prediction results; The individual device status prediction results for each distribution network device are summarized and organized according to the unified device coding results to generate the distribution network device status prediction results.

[0013] Optionally, the step of performing degradation risk correction on the basic equipment condition prediction result based on the degradation risk correction input data to generate a single equipment condition prediction result specifically involves: Read the degradation risk correction input data, and extract the basic prediction results of equipment status and equipment degradation correlation information corresponding to the same distribution network equipment from the degradation risk correction input data to generate single equipment degradation correction data; Based on the single-device degradation correction data, determine the abnormal continuous records, defect recurrence records, maintenance and recovery records, and load change records of the distribution network equipment during historical operation, and generate a set of degradation risk impact records; Based on the degradation risk impact record set, determine the degradation risk enhancement state, degradation risk mitigation state, and degradation risk stability state corresponding to the distribution network equipment, and generate degradation risk state marking results; Based on the degradation risk status labeling results, the predicted status level, predicted abnormal development direction and predicted risk occurrence time in the basic equipment status prediction results are corrected to generate corrected equipment status prediction data. Based on the corrected equipment status prediction data, the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window of the distribution network equipment within the target prediction time range are determined, and single equipment status prediction results are generated.

[0014] Optionally, S7 specifically includes: Read the equipment status diagnosis results and the power distribution network equipment status prediction results, and perform correlation matching according to the unified equipment coding results to generate diagnostic prediction correlation data; Based on the diagnostic prediction association data, determine the current state diagnostic information and future state prediction information of the same power distribution network equipment, and generate a single equipment diagnostic prediction record. Based on the single-device diagnostic prediction records, the risk levels of the power distribution network equipment are classified, and equipment risk level results are generated. Based on the equipment risk level results and the distribution network equipment status prediction results, determine the corresponding fault warning time window for the distribution network equipment and generate equipment fault warning information; Based on the equipment risk level results and the equipment fault early warning information, determine the maintenance priority and status handling suggestions for the corresponding power distribution network equipment, and generate a single equipment status diagnosis prediction decision record. The individual device status diagnosis prediction and decision records corresponding to each distribution network device are summarized and organized according to the unified device coding results to generate distribution network device status diagnosis prediction and decision information.

[0015] This invention addresses the problems of scattered data sources, inconsistent time granularity, lagging traditional threshold alarms, and the tendency of fixed-time-window predictions to mix different operational stages by cascading and integrating equipment degradation fragment reconstruction with an improved PatchTST prediction network. It employs unified equipment coding, unified time stamping, and standardized processing methods to collect and organize operational monitoring data, condition detection data, defect repair data, fault alarm data, environmental condition data, and load change data, generating a distribution network equipment condition feature set that characterizes the current state, historical state changes, and impact of operational disturbances. In the condition diagnosis stage, it constructs an equipment condition profile, integrating basic equipment attributes, state-related feature sequences, state change records, and data. The source records are uniformly organized to achieve continuous expression of the historical status and current diagnostic results of the same distribution network equipment. Addressing the issue of mixed normal operation, abnormal fluctuations, degradation development, and maintenance recovery processes in the time series of distribution network equipment status, this invention reconstructs equipment degradation segments based on changes in equipment status correlation characteristics and equipment status diagnostic results, generating a sequence of equipment degradation segments. This sequence is then input into an improved PatchTST prediction network for degradation segment embedding and time-series prediction, improving the targeting of equipment status change trend prediction. After the prediction output, degradation risk correction is performed by combining equipment degradation correlation information, making the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window more consistent with the actual equipment degradation process. Ultimately, this achieves intelligent output of distribution network equipment risk levels, fault early warning time windows, maintenance priorities, and status handling recommendations. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a method for diagnosing and predicting the condition of power distribution network equipment based on big data analysis proposed in this invention. Figure 2 This is a flowchart illustrating the equipment degradation segment reconstruction process of a power distribution network equipment condition diagnosis and prediction method based on big data analysis proposed in this invention. Figure 3 This is a diagram of the improved PatchTST prediction network structure of a big data analysis-based method for predicting the condition of power distribution network equipment. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3A method for condition diagnosis and prediction of power distribution network equipment based on big data analysis includes the following steps: S1. Collect multi-source equipment status association data during the operation of distribution network equipment, process the multi-source equipment status association data in a unified manner, and generate the original dataset of distribution network equipment status. S2. Standardize the original dataset of distribution network equipment status and extract the equipment status correlation features to generate a distribution network equipment status feature set. S3. Construct equipment status profiles based on the status feature set of distribution network equipment, and perform status diagnosis on distribution network equipment based on the equipment status profiles to generate a set of distribution network equipment status profiles and equipment status diagnosis results. S4. Based on the distribution network equipment status feature set and equipment status diagnosis results, reconstruct the equipment degradation segment of the distribution network equipment status time series to generate the equipment degradation segment sequence. S5. Input the equipment degradation fragment sequence and the distribution network equipment status profile set into the improved PatchTST prediction network to predict the trend of distribution network equipment status changes and generate basic equipment status prediction results. S6. Extract equipment degradation correlation information based on the equipment status basic prediction results, and perform degradation risk correction based on the equipment degradation correlation information to generate distribution network equipment status prediction results. S7. Generate distribution network equipment status diagnosis and prediction decision information based on the equipment status diagnosis results and distribution network equipment status prediction results.

[0019] In this embodiment, S1 specifically refers to: Acquire basic equipment data, operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data, and load change data during the operation of power distribution network equipment, and generate multi-source equipment status correlation data; Read the equipment name, equipment type, voltage level, feeder, transformer area, installation location and asset number from the multi-source equipment status association data, and perform unified coding according to the preset equipment coding rules to generate unified equipment coding results; Read the data acquisition system, data acquisition interface, data record type and data update time corresponding to the status association data of multi-source devices, and mark the data records from different sources to generate data source identification results; Read the acquisition time, detection time, defect registration time, maintenance completion time, fault occurrence time and load change time from the multi-source device status association data, and convert the time format according to a unified time base to generate time stamp results; Based on the unified coding results of the equipment, the operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data and load change data corresponding to the same distribution network equipment are bound to the equipment affiliation to generate equipment status association records; Write the unified coding results of the equipment, the data source identification results, the time stamp results, and the equipment status association records into the multi-source equipment status association data to generate the original dataset of distribution network equipment status. In the specific implementation process, the power distribution network equipment includes distribution transformers, switchgear, ring main units, pole-mounted switches, cable branch boxes, and low-voltage distribution boxes; basic equipment data is obtained from the asset management system, operation monitoring data is obtained from the power distribution automation system and online monitoring terminals, status detection data is obtained from the inspection and testing terminals, defect repair data is obtained from the production management system, fault alarm data is obtained from the alarm platform, environmental condition data is obtained from the meteorological interface and station environmental sensors, and load change data is obtained from the electricity consumption acquisition system. The sampling period for operational monitoring data is set to 1 to 15 minutes. Status detection data is imported daily or according to inspection tasks. Defect repair data and fault alarm data are written using an event-triggered method. The sampling period for environmental operating condition data is set to 10 minutes, and the sampling period for load change data is set to 15 minutes. The unified equipment coding result adopts a combination coding method of "regional number + voltage level + feeder number + equipment type + equipment serial number", with a coding length of 18 to 24 bits. The time stamp result is uniformly converted to yyyy-MM-dd HH:mm:ss format, and the clock synchronization error is controlled within 1 second. After completing the equipment affiliation binding, the data records of the same distribution network equipment within the same time window are merged according to the unified equipment coding result. Duplicate records retain the data with the latest update time, and data records with missing key fields are marked as pending completion and written to the original data set of distribution network equipment status.

[0020] In this embodiment, S2 specifically refers to: Read the equipment unified code field, timestamp field, data source field and equipment status record field from the original data set of distribution network equipment status, and generate the equipment status dataset to be standardized. Perform field integrity checks, duplicate record merging, format error correction, and invalid record removal on the standardized equipment status dataset to generate a cleaned equipment status dataset. Based on the rated parameters, operating limits, defect registration time, maintenance completion time and fault occurrence time of the power distribution network equipment, anomaly verification is performed on the cleaned equipment status dataset to generate a verified equipment status dataset. The standardized equipment status dataset is generated by unifying the units and converting the time granularity of the operation monitoring data, status detection data, defect repair data, fault alarm data and load change data in the verified equipment status dataset. Based on the unified equipment coding field and the timestamp field, the standardized equipment status dataset is time-series aligned to generate equipment status time-series data. Based on the equipment status time series data, extract equipment status correlation features to characterize the current status, historical status changes, and impact of operational disturbances of distribution network equipment; The equipment status association features are associated and organized according to the equipment unified coding field, the timestamp field, and the data source field to generate a distribution network equipment status feature set; In the specific implementation process, the field integrity verification is carried out item by item according to the equipment unified code field, the timestamp field, the data source field, and the equipment status record field. Data records with missing key fields are directly marked as invalid records, and data records with missing non-key fields are supplemented by data from adjacent time points of the same distribution network equipment. Duplicate records are judged according to the equipment unified code field, the timestamp field, and the data source field. Duplicate data is retained based on the record with the latest data update time. The time granularity of operational monitoring data is uniformly converted to 15 minutes. Status detection data is aligned to the corresponding equipment time series according to the detection date. Defect repair data and fault alarm data are written to the corresponding time window according to the event occurrence time. During anomaly verification, voltage, current, temperature, partial discharge, and load rate are all verified within the range of the rated parameters and operating limits of the distribution network equipment. Data exceeding the operating limits by more than 20% and without corresponding fault alarm records are marked as suspected anomaly data. After completing the unit unification and time series alignment, equipment status correlation features are extracted based on a 24-hour statistical window to generate a distribution network equipment status feature set.

[0021] In this embodiment, S3 specifically refers to: Read the status feature set of distribution network equipment, and collect it according to the unified equipment coding result and time stamp result of the corresponding distribution network equipment to generate equipment profiles and construct input data; Based on the equipment profile, the input data is used to determine the basic attributes of the distribution network equipment, the state-related feature sequence, the state change record, and the data source record, and a set of equipment profile fields is generated. Based on the set of equipment profile fields, the associated features of the equipment status of the same power distribution network equipment at different time points are associated and organized to generate an equipment status profile. The equipment status profiles of each distribution network device are classified and organized according to equipment type, feeder, transformer area and unified equipment coding results to generate a set of distribution network equipment status profiles. Read the equipment status profiles from the distribution network equipment status profile set, and determine the corresponding operating deviation status, defect association status and status change trend of the distribution network equipment based on the equipment status profiles, and generate equipment status diagnosis input data; Based on the equipment status diagnosis input data, the status level of the power distribution network equipment is determined and the anomaly category is identified, generating single equipment status diagnosis results; The individual device status diagnosis results corresponding to each distribution network device are written into the distribution network device status profile set according to the unified device coding results, and then summarized and organized to generate device status diagnosis results. In the specific implementation process, the equipment status profile is established based on a single distribution network device as the basic modeling object. The input data for equipment profile construction is updated according to a 24-hour rolling window, and the profile update cycle is set to 15 minutes to 1 hour. The equipment profile field set includes equipment basic attribute fields, status-related feature sequence fields, status change record fields, and data source record fields. Among them, the equipment basic attribute fields include equipment type, voltage level, feeder, transformer area, and installation location. The status-related feature sequence fields retain continuous status characteristics from the most recent 7 days to 30 days. The status change record fields retain defect, maintenance, and fault change records from the most recent year. The status level determination adopts a four-level output method: normal, attention, abnormal, and severe abnormal. When the operation deviates from the status for more than three consecutive sampling windows and there are duplicate records of defect-related statuses, the status level of the corresponding distribution network equipment will be determined to be at least abnormal. When the status change trend continues to rise and there are fault alarm records at the same time, the status level of the corresponding distribution network equipment will be determined to be severe abnormal. By binding the equipment status profile and status diagnosis results according to the unified equipment coding results, the historical status, current status, and diagnosis results of the same distribution network equipment can be kept consistent.

[0022] In this embodiment, S4 specifically refers to: Read the status feature set and status diagnosis results of the distribution network equipment, sort them according to the unified equipment coding results and time stamp results corresponding to the distribution network equipment, and generate the equipment degradation fragment reconstruction input data; Based on the input data reconstructed from the equipment degradation fragments, the changes in equipment status correlation features of the same distribution network equipment at continuous time nodes are extracted, and a sequence of equipment status changes is generated. Based on the equipment status diagnosis results, determine the status level changes, abnormal continuous state and abnormal recovery state of the distribution network equipment at different time points, and generate a status diagnosis label sequence; By mapping the equipment state change sequence and the state diagnosis label sequence to time, the state stability interval, state fluctuation interval, state degradation interval and state recovery interval in the state time series of distribution network equipment are determined, and a set of equipment state segment boundaries is generated. The distribution network equipment state time series is segmented based on the equipment state segment boundary set, and each segmented state segment is labeled with a segment type to generate an initial equipment state segment set. Based on the segment length, state change amplitude, abnormal duration and adjacent segment connection relationship of each state segment in the initial device state segment set, state segments that do not meet the preset segment length requirements are merged or supplemented to generate the reconstructed device state segment set. The reconstructed set of device state fragments is arranged in order according to the unified device coding results and time stamp results to generate a sequence of device degradation fragments; In the specific implementation process, the input data for equipment degradation segment reconstruction is processed according to a single distribution network device, the time sorting granularity is set to 15 minutes, and the time range for a single reconstruction is set to the most recent 7 days to 90 days; when the distribution network device is a high-risk device, the reconstruction time range is extended to the most recent 180 days. In the equipment status change sequence, voltage deviation, current change rate, temperature change rate, load rate change rate, partial discharge change, and number of fault alarms are used as the main change indicators; when the change amplitude of the status correlation characteristics at more than 3 consecutive time nodes is less than the preset stable threshold, it is marked as a stable status interval. When any major change indicator exceeds the corresponding operating limit by 10% to 20% within two consecutive time points, it is marked as a status fluctuation range; when the status level gradually changes from normal or attention to abnormal or severe abnormal, and the abnormal duration exceeds 6 hours, it is marked as a status degradation range; when the status-related characteristics recover to the operating limit range within 24 hours after the maintenance completion time, it is marked as a status recovery range.

[0023] The equipment status segment boundary set is determined by the points of status level change, anomaly occurrence, anomaly recovery, maintenance completion, and significant load change. The minimum length of adjacent segments is set to 4 time nodes, and the maximum length is set to 96 time nodes. Short segments below the minimum length are merged according to the type of adjacent segments and the direction of status change, while long segments above the maximum length are filled and segmented according to a 24-hour rolling window. The initial set of equipment state segments is labeled with segment types including stable segments, fluctuating segments, degraded segments, and recovery segments. Each state segment records the start and end times, duration, type, state change amplitude, and duration of anomalies. After merging or completion, the reconstructed set of equipment state segments is arranged according to the unified equipment coding results and time stamp results, generating a sequence of equipment degradation segments. Through the above equipment degradation segment reconstruction process, the PatchTST prediction network can be improved by avoiding the mixing of normal operation, abnormal fluctuation, and maintenance recovery processes into a fixed time window, thus enhancing the ability of subsequent state trend predictions to represent the actual equipment degradation process.

[0024] In this embodiment, S5 specifically refers to: Read the equipment degradation fragment sequence and the distribution network equipment status profile set, and match them according to the unified equipment coding results and time stamp results to generate PatchTST prediction input data; Based on the PatchTST prediction input data, determine the equipment degradation segment sequence, equipment status profile and target prediction time range for the same distribution network equipment, and generate the equipment status prediction task. Input the device status prediction task into the improved PatchTST prediction network; By improving the PatchTST prediction network, degradation fragment embedding and temporal prediction are performed on device degradation fragment sequences to generate basic prediction results for single device state. The basic prediction results of the individual equipment status corresponding to each distribution network equipment are summarized and organized according to the unified equipment coding results to generate the basic prediction results of equipment status. In the specific implementation process, the PatchTST prediction input data is organized according to a single distribution network device. The device degradation segment sequence and the distribution network device status profile set are matched with the unified device coding results, and the input order is determined by the time stamp results. The target prediction time range is set to at least one of the following: 1 hour, 6 hours, 24 hours and 72 hours. Ordinary distribution network devices output the status prediction results for the next 24 hours by default, while high-risk distribution network devices output short-term status prediction results for the next 1 hour and 6 hours simultaneously. In the equipment status prediction task, the input length for a single distribution network device is set to the most recent 96 to 384 time nodes, with a time granularity of 15 minutes. When the equipment degradation segment sequence has fewer than 96 time nodes, it is supplemented with the most recent valid status value under the same segment type. When the equipment degradation segment sequence has more than 384 time nodes, the time interval containing degradation segments, fluctuation segments, and recovery segments is retained first. After the improved PatchTST prediction network completes the prediction, the basic prediction result of a single device status includes the predicted status level, the predicted status change trend, and the predicted anomaly probability. The basic prediction results of the single device status corresponding to each distribution network device are summarized according to the unified coding results of the device and classified and stored according to the device type, the feeder to which it belongs, and the transformer substation to which it belongs, generating the basic prediction result of the equipment status.

[0025] In this embodiment, the PatchTST prediction network is improved to embed degradation fragments and perform temporal prediction on the device degradation fragment sequence, generating a basic prediction result for the single device state, specifically as follows: The improved PatchTST prediction network includes a degenerate fragment input layer, a fragment attribute encoding layer, a degenerate fragment embedding layer, a device profiling constraint layer, a temporal prediction encoding layer, and a state prediction output layer. The degradation fragment sequence in the equipment status prediction task is read by the degradation fragment input layer, and the arrangement order of the degradation fragments corresponding to the same distribution network equipment is determined according to the unified equipment coding result and time stamp result to generate the degradation fragment input sequence. The fragment attribute encoding layer extracts the fragment boundary information, fragment state information, and fragment duration information corresponding to each device's degraded fragment from the input sequence of degraded fragments, and generates the degraded fragment attribute encoding result. The degenerate fragment embedding layer embeds the degenerate fragment input sequence and the degenerate fragment attribute encoding result to generate a degenerate fragment embedding representation. The device state profile in the device state prediction task is read through the device profile constraint layer, and the device state constraint is matched with the degenerate fragment embedding representation based on the device state profile to generate the profile constraint fragment representation. By performing cross-segment temporal modeling on the portrait constraint segment representation through the temporal prediction coding layer, the state continuity relationship and state change trend of the same distribution network equipment in different equipment degradation segments are determined, and the equipment state temporal prediction representation is generated. Based on the device state timing prediction representation and the target prediction time range, the state prediction output layer outputs the basic prediction result of a single device state. In its implementation, the improved PatchTST prediction network inherits the fragmented temporal modeling framework of the traditional PatchTST model in its overall structure, including fragment input, fragment embedding, temporal encoding, and prediction output. However, it introduces a degenerate fragment embedding layer and a device profiling constraint layer in the fragment construction and state constraint stages to replace the traditional PatchTST model's approach of simply dividing fragments and directly predicting them according to a fixed time window. First, the sequence of degenerate fragments in the device state prediction task is input into the degenerate fragment input layer according to the order of the degenerate fragments. The length of a single degenerate fragment is set to 8 to 32 time nodes, and the time granularity is set to 15 minutes. The fragment attribute encoding layer encodes fragment boundary information, fragment state information, and fragment duration information, with the encoding dimension set to 32 dimensions. The degenerate fragment embedding layer maps the degenerate fragment input sequence and the degenerate fragment attribute encoding result into a 128-dimensional degenerate fragment embedding representation. Subsequently, the device profile constraint layer maps the device state profile into a 64-dimensional profile constraint vector and performs device state constraint matching with the degenerate fragment embedding representation to generate a profile constraint fragment representation. The temporal prediction coding layer adopts a 3-layer coding structure with a hidden dimension of 128 and a dropout rate of 0.1. Compared with the traditional PatchTST model, this invention can distinguish between stationary segments, fluctuating segments, degenerate segments, and recovery segments, avoiding the mixing of different state stages by a fixed window, and improving the pertinence and stability of the prediction of the state change trend of distribution network equipment.

[0026] In this embodiment, S6 specifically refers to: Read the basic prediction results of equipment status, and extract data records related to the degradation process of distribution network equipment from the original dataset of distribution network equipment status, the feature set of distribution network equipment status, the equipment status diagnosis results and the set of distribution network equipment status profiles to generate equipment degradation association information; Based on the unified coding results and time stamp results of the equipment, the basic prediction results of the equipment status are matched with the equipment degradation correlation information to generate degradation risk correction input data; Based on the degradation risk correction input data, the basic equipment condition prediction results are corrected for degradation risk, and a single equipment condition prediction result is generated. The state prediction results of each individual device in the distribution network are summarized and organized according to the unified coding results of the devices to generate the state prediction results of the distribution network devices. In the specific implementation process, equipment degradation correlation information is extracted on a per-unit distribution network device, and the extraction range is set to the most recent 30 days to 180 days before the target prediction time range corresponding to the basic prediction result of the device status; ordinary distribution network devices are extracted with the most recent 90 days of data by default, and high-risk distribution network devices are extracted with the most recent 180 days of data; Equipment degradation-related information includes historical anomaly records, defect records, maintenance records, fault records, and load change records related to equipment status changes. Historical anomaly records are statistically analyzed at a 15-minute time granularity; defect records, maintenance records, and fault records are written into corresponding time windows according to the event occurrence time; and load change records are organized according to daily maximum load rate, daily average load rate, and peak-hour load rate. Degradation risk correction input data is matched according to the equipment's unified coding results and time stamp results, with the matching time error controlled within one time window. When multiple equipment degradation-related information records exist for the same distribution network equipment, they are sorted from most recent to oldest event time, with priority given to data records adjacent to the predicted anomaly time range in the equipment status basic prediction results. After completing the degradation risk correction, the single device status prediction result should include at least the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window. The distribution network equipment status prediction results are classified and summarized according to equipment type, feeder, and transformer area, and output as input data for subsequent distribution network equipment status diagnosis and prediction decision information.

[0027] In this embodiment, degradation risk correction is performed on the basic equipment condition prediction results based on the degradation risk correction input data to generate single equipment condition prediction results, specifically as follows: Read the degradation risk correction input data, and extract the equipment status basic prediction results and equipment degradation correlation information corresponding to the same distribution network equipment from the degradation risk correction input data to generate single equipment degradation correction data; Based on the degradation correction data of individual equipment, identify the abnormal continuous records, defect recurrence records, maintenance and recovery records and load change records of the distribution network equipment during historical operation, and generate a set of degradation risk impact records; Based on the degradation risk impact record set, determine the degradation risk enhancement state, degradation risk mitigation state, and degradation risk stability state corresponding to the distribution network equipment, and generate degradation risk state labeling results; Based on the degradation risk status labeling results, the predicted status level, predicted anomaly development direction and predicted risk occurrence time in the basic equipment status prediction results are corrected to generate corrected equipment status prediction data. Based on the corrected equipment status prediction data, determine the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window of the distribution network equipment within the target prediction time range, and generate single equipment status prediction results. In the specific implementation process, the degradation correction data of a single device is processed independently according to the unified coding result of the device. The correction time range is consistent with the target prediction time range, which is set to at least one of the following: 1 hour, 6 hours, 24 hours and 72 hours in the future. The degradation risk impact record set is extracted from historical records from the most recent 30 days to 180 days. For ordinary distribution network equipment, the most recent 90 days' records are used by default, while for high-risk distribution network equipment, the most recent 180 days' records are used. Abnormal persistence records are statistically analyzed based on the continuous duration of the abnormality. When the abnormal state of the same distribution network equipment continues for more than 6 hours, it is marked as an abnormal persistence state. Defect recurrence records are counted based on the number of times the same defect type occurs within 30 days. When the number of recurrences reaches 2 or more, it is marked as a defect recurrence state. Maintenance recovery records are judged based on whether the status correlation characteristics return to the operating limit range within 24 to 72 hours after maintenance is completed. Load change records are counted based on the daily maximum load rate, daily average load rate, and peak period load rate. When the daily maximum load rate exceeds 80% for 3 consecutive days, it is marked as a high load state. In the degradation risk status labeling results, the abnormal continuous state, the defect recurrence state, and the high load state are used to generate the degradation risk enhancement state, the state recovery to stability after maintenance is used to generate the degradation risk mitigation state, and the remaining cases generate the degradation risk stability state. When performing degradation risk correction, the enhanced degradation risk state corresponds to increasing the predicted state level or predicting the risk occurrence time earlier, the mitigated degradation risk state corresponds to reducing the predicted abnormal development direction or delaying the predicted risk occurrence time, and the stable degradation risk state keeps the basic prediction result of the equipment state unchanged. The corrected equipment status prediction data is output according to the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window to generate single equipment status prediction results.

[0028] In this embodiment, S7 specifically refers to: Read the equipment status diagnosis results and the distribution network equipment status prediction results, and perform correlation matching according to the unified equipment coding results to generate diagnostic prediction correlation data; Based on the diagnostic prediction correlation data, determine the current state diagnostic information and future state prediction information of the same power distribution network equipment, and generate a single equipment diagnostic prediction record. Based on the single-device diagnostic prediction records, the risk levels of distribution network equipment are classified, and equipment risk level results are generated. Based on the equipment risk level results and the distribution network equipment status prediction results, determine the corresponding fault warning time window for the distribution network equipment and generate equipment fault warning information; Based on the equipment risk level results and equipment fault early warning information, determine the maintenance priority and status handling suggestions for the corresponding distribution network equipment, and generate a single equipment status diagnosis prediction decision record; The status diagnosis and prediction decision records of each distribution network device are summarized and organized according to the unified coding results of the devices to generate distribution network device status diagnosis and prediction decision information. In the specific implementation process, diagnostic prediction data is matched according to the unified equipment coding results, with a matching cycle set to 15 minutes to 1 hour. Current status diagnostic information includes the current status level, anomaly category, and anomaly severity. Future status prediction information includes the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window. Equipment risk levels are set to four levels: low risk, medium risk, high risk, and emergency risk. A low-risk level is defined as the current status level being normal and showing no abnormal upward trend within the next 24 hours; a medium-risk level is defined as the current status level being of concern or showing a slight abnormal trend within the next 24 hours; and a high-risk level is defined as the current status level being abnormal or showing a significant upward risk trend within the next 6 to 24 hours. When the current status level is severe anomaly or there is a risk of failure within the next 6 hours, it is classified as an emergency risk. Fault warning time windows are output in tiers of 1 hour, 6 hours, 24 hours, and 72 hours. Maintenance priorities are set from P1 to P4, where P1 corresponds to emergency risk equipment, P2 to high-risk equipment, P3 to medium-risk equipment, and P4 to low-risk equipment. Status handling recommendations include continuous monitoring, scheduling on-site inspections, conducting specialized tests, developing maintenance plans, implementing load transfer, and immediate maintenance. Finally, the results are summarized according to equipment type, feeder, transformer substation, and unified equipment coding to generate distribution network equipment status diagnosis and prediction decision information. By jointly outputting current status diagnosis information and future status prediction information, passive maintenance based solely on the current abnormal state can be avoided. This allows maintenance resources to be prioritized for distribution network equipment with rapid risk escalation and short fault warning time windows, thereby improving the timeliness of distribution network equipment status handling and the pertinence of maintenance arrangements.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to the condition diagnosis and prediction system of power distribution network equipment in a mixed power supply area between an industrial park in the eastern part of a city and surrounding residential areas. The test lasted for 3 months, generating 92 days of valid test data. The test area covered approximately 42.8 km², including 2 110kV substations, 14 10kV feeders, 327 distribution transformers, 186 ring main units, 94 switchgear, 213 pole-mounted switches, 126 cable branch boxes, and several low-voltage distribution boxes. This area simultaneously bears industrial production loads, residential loads, and distributed photovoltaic grid-connected loads, with significant differences in load between day and night. Some feeders experienced rapid increases in load rate during weekday morning and evening peak hours. Some older distribution transformers and ring main units exhibited abnormal temperature rises, partial discharge fluctuations, repeated defect registrations, and recurring status issues after maintenance. Traditional power distribution network equipment status management mainly relies on threshold alarms and manual inspections. Alarms are usually triggered only after equipment indicators exceed limits. It is difficult to identify the process of equipment from minor abnormalities to continuous degradation in a timely manner, and it is also difficult to distinguish between temporary load fluctuations and real equipment degradation. As a result, maintenance resources tend to be concentrated on equipment that has already shown obvious abnormalities, and there is insufficient early handling of potentially high-risk equipment.

[0030] In this scenario, multi-source equipment status correlation data were first collected from the power distribution automation system, asset management system, production management system, fault alarm platform, electricity consumption information collection system, and station environment monitoring terminal. During the test, approximately 28.6 million equipment status records were accessed, including approximately 21.8 million operation monitoring data records, approximately 2.14 million status detection data records, approximately 48,000 defect repair data records, approximately 163,000 fault alarm data records, approximately 3.1 million environmental condition data records, and approximately 1.35 million load change data records. The system generated a unified equipment coding result according to the combination of "regional number + voltage level + feeder number + equipment type + equipment serial number," and uniformly converted the acquisition time, detection time, defect registration time, repair completion time, fault occurrence time, and load change time from different systems into the same time format, with the time synchronization error controlled within 1 second. Subsequently, the collected data underwent field integrity verification, duplicate record merging, format error correction, invalid record removal, unit unification, and time granularity conversion. The operation monitoring data was unified to a 15-minute time granularity, and the equipment status association features were extracted according to a 24-hour rolling window to generate a distribution network equipment status feature set.

[0031] After data standardization, the system constructs a device status profile for each individual distribution network device. Basic device attributes, status-related feature sequences, status change records, and data source records are written into the same profile, and the system uses this profile to perform status diagnosis on the distribution network devices. For devices with significant status fluctuations within the test area, the system further reconstructs device degradation segments based on the device status diagnosis results and the distribution network device status feature set. Fixed time series are divided into stable status intervals, fluctuating status intervals, degradation status intervals, and recovery status intervals, forming a device degradation segment sequence. Taking the 10kV Dongkai Line #07 ring main unit as an example, this device experienced continuous temperature increases and slight fluctuations in partial discharge in mid-March. Traditional threshold methods did not trigger alarms, but this invention classifies this period as a status fluctuation segment. In late March, the partial discharge of this device continued to rise during peak load periods, accompanied by repeated defect registrations. The system further identifies this process as a status degradation segment and inputs it into the improved PatchTST prediction network for trend prediction. After degradation risk correction, the system outputs a high-risk warning approximately 22 hours before the actual fault alarm occurs, and provides suggestions for specific testing and maintenance. On-site inspection revealed poor contact and localized overheating issues at the cable joints of the ring main unit, consistent with the risk type predicted by the system.

[0032] To verify the effectiveness of this invention, the method of this invention was compared with traditional threshold diagnosis methods and ordinary PatchTST prediction methods under the same test data and equipment range. Traditional threshold diagnosis methods make judgments based on operating limits and alarm thresholds, while ordinary PatchTST prediction methods directly input the equipment status time series according to a fixed time window without reconstructing equipment degradation segments or correcting for degradation risks. This test used manual review results, maintenance records, and actual fault alarms as verification criteria, and statistically analyzed equipment anomalies, defect development, and fault risk events that occurred within 92 days. The results are shown in Table 1.

[0033] Table 1 Comparison of Prediction Effects for Condition Diagnosis of Distribution Network Equipment As shown in Table 1, the method of this invention outperforms traditional threshold diagnosis methods and ordinary PatchTST prediction methods in terms of state diagnosis accuracy, 24-hour risk warning hit rate, severe anomaly false alarm rate, average early warning time, and high-risk equipment maintenance matching rate. While traditional threshold diagnosis methods have simple rules, they are insensitive to continuous degradation processes that have not yet exceeded limits, resulting in some equipment being identified only after a fault alarm occurs. Ordinary PatchTST prediction methods can use time series to predict future trends, but the fixed time window easily mixes normal operation segments, abnormal fluctuation segments, and maintenance recovery segments into the model, leading to insufficient advance warning for some risks. This invention reconstructs equipment degradation segments, forming equipment degradation segment sequences for different state stages. Then, it embeds and predicts degradation segments using an improved PatchTST prediction network, and combines equipment degradation association information for degradation risk correction, making the prediction results more closely reflect the actual process of distribution network equipment from slight fluctuations to continuous degradation and then to risk exposure.

[0034] During implementation, this invention identified 47 high-risk and emergency-risk devices, of which 43 were confirmed to have significant hidden dangers through on-site inspections. These included abnormal temperature rise in distribution transformers, overheating of cable joints in ring main units, jamming of pole-mounted switch operating mechanisms, aging of cable branch box joints, and abnormal partial discharge in switchgear. Based on the distribution network equipment status diagnosis and prediction decision information generated by the system, the maintenance unit adjusted the allocation of maintenance tasks from an average periodic allocation to an allocation based on risk level. Specifically, the average response time for P1 level tasks decreased from 9.6 hours to 3.1 hours, and the average processing time for P2 level tasks decreased from 2.8 days to 1.4 days. During the testing period, the number of reactive repairs following faults in the test area decreased from 26 in the previous quarter to 11, a decrease of approximately 57.7%; the number of repeated defect registrations decreased from 38 to 19, a decrease of approximately 50.0%. The above results show that the present invention can realize a continuous closed loop of equipment condition diagnosis, condition trend prediction and maintenance decision output in distribution network scenarios with complex load fluctuations and inconsistent multi-source condition data. It can effectively improve the accuracy of high-risk equipment identification, extend the early warning time of faults, and enhance the pertinence of maintenance resource allocation.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for condition diagnosis and prediction of power distribution network equipment based on big data analysis, characterized in that, Includes the following steps: S1. Collect multi-source device status association data during the operation of power distribution network equipment, and process the multi-source device status association data in a unified manner to generate a raw dataset of power distribution network equipment status. S2. Standardize the original dataset of the distribution network equipment status and extract the equipment status correlation features to generate a distribution network equipment status feature set. S3. Construct a device status profile based on the device status feature set of the distribution network equipment, and perform status diagnosis on the distribution network equipment based on the device status profile to generate a set of distribution network equipment status profiles and device status diagnosis results. S4. Based on the distribution network equipment status feature set and the equipment status diagnosis results, reconstruct the equipment degradation segment of the distribution network equipment status time series to generate an equipment degradation segment sequence. S5. Input the equipment degradation fragment sequence and the distribution network equipment status profile set into the improved PatchTST prediction network to predict the trend of distribution network equipment status changes and generate basic equipment status prediction results. S6. Extract equipment degradation correlation information based on the equipment status basic prediction results, and perform degradation risk correction based on the equipment degradation correlation information to generate distribution network equipment status prediction results. S7. Based on the equipment status diagnosis results and the distribution network equipment status prediction results, generate distribution network equipment status diagnosis and prediction decision information.

2. The method for condition diagnosis and prediction of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S1 is: Acquire basic equipment data, operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data, and load change data during the operation of power distribution network equipment, and generate multi-source equipment status correlation data; Read the equipment name, equipment type, voltage level, feeder, transformer area, installation location and asset number from the multi-source equipment status association data, and perform unified coding according to the preset equipment coding rules to generate unified equipment coding results; Read the data acquisition system, data acquisition interface, data record type and data update time corresponding to the multi-source device status association data, and mark the data records from different sources to generate data source identification results; Read the acquisition time, detection time, defect registration time, maintenance completion time, fault occurrence time and load change time from the multi-source equipment status association data, and convert the time format according to a unified time base to generate time stamp results; Based on the unified coding results of the equipment, the operation monitoring data, status detection data, defect repair data, fault alarm data, environmental condition data and load change data corresponding to the same distribution network equipment are bound to the equipment affiliation to generate equipment status association records; The unified coding result of the device, the data source identification result, the time stamp result, and the device status association record are written into the multi-source device status association data to generate the original dataset of distribution network device status.

3. The method for condition diagnosis and prediction of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S2 is: Read the equipment unified code field, timestamp field, data source field and equipment status record field from the original data set of the distribution network equipment status, and generate the equipment status dataset to be standardized. The device status dataset to be standardized is subjected to field integrity verification, duplicate record merging, format error correction and invalid record removal to generate a cleaned device status dataset. Based on the rated parameters, operating limits, defect registration time, maintenance completion time, and fault occurrence time of the power distribution network equipment, anomaly verification is performed on the cleaned equipment status dataset to generate a verified equipment status dataset. The operation monitoring data, status detection data, defect repair data, fault alarm data and load change data in the verified equipment status dataset are unified in terms of units and converted in terms of time granularity to generate a standardized equipment status dataset. Based on the unified device coding field and the timestamp field, the standardized device status dataset is time-series aligned to generate device status time-series data. Based on the equipment status time series data, extract equipment status correlation features to characterize the current status, historical status changes, and operational disturbance impacts of distribution network equipment; The device status association features are associated and organized according to the device unified coding field, time stamp field, and data source field to generate a distribution network device status feature set.

4. The method for condition diagnosis and prediction of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S3 is: Read the status feature set of the power distribution network equipment, and collect it according to the unified equipment coding result and time stamp result corresponding to the power distribution network equipment to generate equipment profile construction input data; Based on the equipment profile, the input data is used to determine the basic attributes of the distribution network equipment, the state-related feature sequence, the state change record, and the data source record, and a set of equipment profile fields is generated. Based on the set of equipment profile fields, the associated features of the equipment status of the same power distribution network equipment at different time points are associated and organized to generate an equipment status profile. The equipment status profiles of each distribution network device are classified and organized according to equipment type, feeder, transformer area and unified equipment coding results to generate a set of distribution network equipment status profiles. Read the equipment status profiles in the distribution network equipment status profile set, and determine the corresponding operating deviation status, defect association status and status change trend of the distribution network equipment based on the equipment status profiles, and generate equipment status diagnosis input data; Based on the equipment status diagnosis input data, the status level of the power distribution network equipment is determined and the anomaly category is identified, generating a single equipment status diagnosis result. The individual device status diagnosis results corresponding to each distribution network device are written into the distribution network device status profile set according to the unified device coding results, and then summarized and organized to generate device status diagnosis results.

5. The method for predicting and diagnosing the condition of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S4 is: Read the status feature set of the power distribution network equipment and the status diagnosis results of the equipment, and sort them according to the unified equipment coding results and time stamp results corresponding to the power distribution network equipment to generate equipment degradation fragment reconstruction input data; Based on the input data reconstructed from the equipment degradation fragments, the changes in equipment state association features of the same power distribution network equipment at continuous time nodes are extracted, and a sequence of equipment state changes is generated. Based on the equipment status diagnosis results, determine the status level changes, abnormal continuous status and abnormal recovery status of the power distribution network equipment at different time points, and generate a status diagnosis label sequence; The equipment state change sequence and the state diagnosis mark sequence are time-correlated to determine the state stability interval, state fluctuation interval, state degradation interval and state recovery interval in the distribution network equipment state time series, and a set of equipment state segment boundaries is generated. The distribution network equipment state time series is segmented according to the equipment state segment boundary set, and each segmented state segment is labeled with a segment type to generate an initial equipment state segment set. Based on the segment length, state change amplitude, abnormal duration and adjacent segment connection relationship of each state segment in the initial device state segment set, state segments that do not meet the preset segment length requirements are merged or supplemented to generate a reconstructed device state segment set. The reconstructed set of device state fragments is arranged in order according to the unified device coding result and the time stamp result to generate a sequence of device degradation fragments.

6. The method for predicting and diagnosing the condition of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S5 is: Read the device degradation fragment sequence and the distribution network device status profile set, and match them according to the device unified coding result and time stamp result to generate PatchTST prediction input data; Based on the PatchTST prediction input data, determine the equipment degradation segment sequence, equipment status profile and target prediction time range for the same distribution network equipment, and generate the equipment status prediction task. The device state prediction task is input into the improved PatchTST prediction network; The improved PatchTST prediction network is used to embed and predict the degradation fragment sequence of the device in a temporal sequence, generating a basic prediction result for the single device state. The basic prediction results of the individual equipment status corresponding to each distribution network equipment are summarized and organized according to the unified equipment coding results to generate the basic prediction results of equipment status.

7. The method for predicting and diagnosing the condition of power distribution network equipment based on big data analysis according to claim 6, characterized in that, The improved PatchTST prediction network is used to embed and predict the degradation fragment sequence of the device, generating a basic prediction result for the single device state. Specifically, this involves: The improved PatchTST prediction network includes a degenerate fragment input layer, a fragment attribute encoding layer, a degenerate fragment embedding layer, a device profile constraint layer, a time series prediction encoding layer, and a state prediction output layer. The degradation fragment input layer reads the equipment degradation fragment sequence in the equipment state prediction task, and determines the degradation fragment arrangement order corresponding to the same distribution network equipment according to the unified equipment coding result and time stamp result, thereby generating the degradation fragment input sequence. The fragment attribute encoding layer extracts fragment boundary information, fragment state information, and fragment duration information corresponding to each device degradation fragment from the degradation fragment input sequence, and generates degradation fragment attribute encoding results. The degenerate fragment embedding layer performs embedding processing on the degenerate fragment input sequence and the degenerate fragment attribute encoding result to generate a degenerate fragment embedding representation; The device state profile in the device state prediction task is read through the device profile constraint layer, and the device state constraint is matched with the degenerate fragment embedding representation based on the device state profile to generate a profile constraint fragment representation. The temporal prediction coding layer performs cross-fragment temporal modeling on the portrait constraint fragment representation to determine the state continuity relationship and state change trend of the same distribution network equipment in different equipment degradation fragments, and generates a temporal prediction representation of equipment state. The state prediction output layer outputs a basic prediction result for a single device state based on the device state time-series prediction representation and the target prediction time range.

8. The method for predicting and diagnosing the condition of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S6 is: Read the basic prediction results of the equipment status, and extract data records related to the degradation process of the distribution network equipment from the original dataset of the distribution network equipment status, the feature set of the distribution network equipment status, the equipment status diagnosis results, and the distribution network equipment status profile set to generate equipment degradation association information; Based on the unified coding results and time stamp results of the equipment, the basic prediction results of the equipment status are matched with the equipment degradation association information to generate degradation risk correction input data; Based on the degradation risk correction input data, the degradation risk correction is performed on the basic equipment status prediction results to generate single equipment status prediction results; The individual device status prediction results for each distribution network device are summarized and organized according to the unified device coding results to generate the distribution network device status prediction results.

9. The method for condition diagnosis and prediction of power distribution network equipment based on big data analysis according to claim 8, characterized in that, The step of performing degradation risk correction on the basic equipment condition prediction results based on the degradation risk correction input data to generate single equipment condition prediction results specifically involves: Read the degradation risk correction input data, and extract the basic prediction results of equipment status and equipment degradation correlation information corresponding to the same distribution network equipment from the degradation risk correction input data to generate single equipment degradation correction data; Based on the single-device degradation correction data, determine the abnormal continuous records, defect recurrence records, maintenance and recovery records, and load change records of the distribution network equipment during historical operation, and generate a set of degradation risk impact records; Based on the degradation risk impact record set, determine the degradation risk enhancement state, degradation risk mitigation state, and degradation risk stability state corresponding to the distribution network equipment, and generate degradation risk state marking results; Based on the degradation risk status labeling results, the predicted status level, predicted abnormal development direction and predicted risk occurrence time in the basic equipment status prediction results are corrected to generate corrected equipment status prediction data. Based on the corrected equipment status prediction data, the predicted status level, predicted anomaly type, predicted risk trend, and predicted risk time window of the distribution network equipment within the target prediction time range are determined, and single equipment status prediction results are generated.

10. The method for predicting and diagnosing the condition of power distribution network equipment based on big data analysis according to claim 1, characterized in that, Specifically, S7 is: Read the equipment status diagnosis results and the power distribution network equipment status prediction results, and perform correlation matching according to the unified equipment coding results to generate diagnostic prediction correlation data; Based on the diagnostic prediction association data, determine the current state diagnostic information and future state prediction information of the same power distribution network equipment, and generate a single equipment diagnostic prediction record. Based on the single-device diagnostic prediction records, the risk levels of the power distribution network equipment are classified, and equipment risk level results are generated. Based on the equipment risk level results and the distribution network equipment status prediction results, determine the corresponding fault warning time window for the distribution network equipment and generate equipment fault warning information; Based on the equipment risk level results and the equipment fault early warning information, determine the maintenance priority and status handling suggestions for the corresponding power distribution network equipment, and generate a single equipment status diagnosis prediction decision record. The individual device status diagnosis prediction decision records corresponding to each distribution network device are summarized and organized according to the unified device coding results to generate distribution network device status diagnosis prediction decision information.

Images