Switching device fault early warning method based on multi-source data fusion

Abstract

The present application relates to the technical field of fault prediction, in particular to a switch device fault early warning method based on multi-source data fusion, comprising the following steps: acquiring channel transition time sorting to generate time interval, marking behavior label to generate time sequence, constructing sliding window to form combined alarm, analyzing cross matrix to extract abnormal period, and identifying dominant channel to issue early warning. In the present application, the time intervals of multi-channel state transition response are aligned for processing, realizing dynamic linkage identification among multiple signals, strengthening the time sequence correlation between abnormal behaviors by means of behavior label combination and sliding statistical strategy, enhancing the sensitivity and accuracy of abnormal identification through the aggregated expression of composite alarm events, combining cross-channel behavior label matrix analysis to determine abnormal label persistence and identify key channels, improving the precision and timeliness of switch device early warning, reducing the risk of false positives and false negatives, and adapting to the actual needs of multi-source perception data fusion analysis of switch devices.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, and in particular to a fault early warning method for switchgear based on multi-source data fusion. Background Technology

[0002] The field of fault prediction technology involves the identification and prediction of potential faults through the analysis of equipment operating status data. This includes fault modeling based on historical data, dynamic comparison of real-time monitoring data, identification of abnormal behavior patterns, and assessment of equipment degradation trends. It typically combines sensor monitoring, data mining, and pattern recognition methods to analyze the operating status of key facilities such as industrial equipment, power systems, and transportation equipment to provide early warning basis. Among these methods, traditional switchgear fault early warning methods refer to the identification of abnormal states and the prediction of fault risks based on a single data source, such as partial discharge signals or infrared temperature data. This involves extracting a single signal type through a specific data acquisition channel and setting an early warning threshold using statistical rules to determine whether the equipment status is abnormal.

[0003] In the current process of fault prediction for switchgear, the main reliance is on a single data source to obtain local signal feature information. There is a lack of time alignment and behavior coordination judgment mechanism between multiple channels. When multiple signals overlap or fault symptoms are relatively hidden, it is difficult to achieve effective correlation of abnormal behaviors. The warning threshold set by a single channel is easily affected by external environmental fluctuations, leading to an increase in misjudgments. The failure to form a continuous behavior label sequence makes the identification of abnormal states deviate from the actual operating situation. In typical scenarios, such as temperature fluctuations accompanied by voltage fluctuations, they cannot be analyzed synchronously, resulting in fault symptoms being treated in isolation and thus delaying equipment maintenance response. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a fault early warning method for switchgear based on multi-source data fusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a fault early warning method for switchgear based on multi-source data fusion, comprising the following steps:

[0006] S1: Obtain the first timestamp corresponding to the state transition point in the voltage channel, temperature channel, and vibration channel of the switching equipment, sort them according to time sequence, and obtain the set of response time intervals;

[0007] S2: Based on the set of response time intervals, the sampled data is horizontally aligned to detect whether there are voltage stagnation, temperature passivation and vibration drop labels, and the corresponding timestamps are recorded to obtain a behavioral label time series group;

[0008] S3: Based on the behavior tag time series group, construct a sliding window, count the number of non-repeating behavior tags in each sliding window, and when the number of tags is greater than or equal to 2, combine the corresponding sliding window number, tag type and corresponding channel number, and uniformly number them as composite alarm event ID to obtain a set of combined trigger windows.

[0009] S4: Based on the combined trigger window set, perform cross matrix analysis on the channel number corresponding to the marker and the corresponding behavior label, and extract all time periods on the time axis where abnormal labels continuously appear to obtain the multi-channel concurrent abnormal time periods;

[0010] S5: Based on the channel numbers involved in the multi-channel concurrent abnormal period, calculate the single-channel abnormal contribution frequency, mark the abnormal dominant channel, and issue a switch fault warning message.

[0011] As a further embodiment of the present invention, the response time interval set includes the channel response sequence, timestamp pairing structure, and time interval arrangement sequence; the behavior tag time series group includes voltage stagnation tag, temperature passivation tag, and vibration drop tag; the combined trigger window set includes sliding window number, composite alarm event ID, and tag type matching result; the multi-channel concurrent abnormal period includes channel number corresponding matrix analysis result, abnormal tag duration segment, and abnormal tag cross-mapping relationship; and the switch fault early warning information includes behavior tag frequency distribution record, channel number and abnormal correlation record, and switch fault early warning real-time record.

[0012] As a further aspect of the present invention, the specific steps for obtaining the response time interval set are as follows:

[0013] S111: Obtain the original signal sequences monitored in the voltage channel, temperature channel, and vibration channel of the switching equipment, identify and extract the first timestamp value corresponding to the state transition point, record them as the transition start time point of each channel, form a corresponding mapping set composed of channel number and timestamp value, and obtain the channel first jump time set;

[0014] S112: Based on the set of first-hop times of the channels, sort all channel numbers in ascending order according to the corresponding first-hop timestamp values, record the original sequence number corresponding to the channel numbers after ascending order as the current sorting sequence number, construct a set of paired value pairs between the sorting sequence number and the first-hop time of the channel, and obtain the time series mapping data group;

[0015] S113: Based on the time series mapping data group, calculate the response time difference between adjacent channels based on the timestamp values ​​between adjacent sorting numbers, sort the time intervals between all channel sorting numbers in order and put them into a unified set to obtain the response time interval set.

[0016] As a further aspect of the present invention, the specific steps for obtaining the behavior label time series group are as follows:

[0017] S211: Based on the set of response time intervals, the sampling time axes of the voltage channel, temperature channel, and vibration channel are uniformly aligned laterally. A uniform sampling time axis is constructed using the minimum time step as the reference step. The sampling amplitude records at the corresponding time points in each channel are extracted. The unsampled areas are filled in using the previous value preservation method to complete the data alignment. After all the sampled data are aligned, a horizontal channel mapping is established to obtain a three-channel aligned data matrix.

[0018] S212: Based on the three-channel aligned data matrix, detect whether the amplitude change range within two adjacent periodic sampling segments in the voltage channel is less than the voltage stability threshold; calculate two sets of temperature differences for every three consecutive sampling points in the temperature channel and compare whether the maximum difference is less than or equal to the reference value; extract the current sampling value for the vibration channel and determine whether it is lower than 30% of the mean; label the sampling points of the channels that meet the above judgment conditions accordingly and generate a label judgment sequence.

[0019] S213: Based on the tag determination sequence, map them to the tags of three categories: voltage stagnation, temperature passivation, and vibration drop, respectively, integrate them into a structured recording unit in chronological order, obtain the channel number and event timestamp corresponding to each type of behavior, and obtain the behavior tag time sequence group.

[0020] As a further aspect of the present invention, the specific steps for obtaining the combined trigger window set are as follows:

[0021] S311: Obtain the behavior label, channel number and timestamp corresponding to each record in the behavior label time series group, construct a sliding window on the overall time axis according to the time order, filter the label records contained in each sliding window by time, extract all label records within the time window range and construct a label summary table, remove duplicate behavior labels and perform unique counting to generate window label counting results;

[0022] S312: Based on the window label counting results, filter all sliding windows with a label count greater than or equal to 2, extract the label type and corresponding channel number contained in the window, and combine them with the window number to form a unique identifier structure for all windows that meet the conditions. Based on the behavior label and channel number, integrate the information to obtain a composite label identifier set.

[0023] S313: Based on the composite tag identifier set, assign a number to each composite record according to the alarm management specification, construct a unique composite alarm event identifier by combining the time number of the sliding window, and archive and integrate the number with the window start and end time, the involved channels and tag types, construct a unified sequence in chronological order, and generate a set of combined trigger windows.

[0024] As a further aspect of the present invention, the specific steps for obtaining the multi-channel concurrent abnormal time period are as follows:

[0025] S411: Based on the sliding window number and corresponding channel number and behavior label recorded in the combined trigger window set, extract all sliding window numbers and sort them in chronological order. Calculate whether the difference between adjacent numbers in the window number sequence is 1. Merge the window numbers with a difference of 1 continuously to construct a set of time-continuous window segment sequences and obtain a set of continuous window labels.

[0026] S412: Based on the continuous window label set, construct a cross matrix structure with channel number as row index and behavior label as column index. In each cross cell, count the frequency of channel and label combination in the continuous window segment. At the same time, extract the co-occurrence number of each channel-behavior pair, calculate the behavior coupling degree of channel and label in the time window, and determine whether it exceeds the behavior anomaly judgment threshold. Filter all channel-behavior combinations that meet the coupling strength judgment standard and generate a channel label coupling degree matrix.

[0027] S413: Based on the channel label coupling degree matrix, extract the corresponding time index and integrate the continuously distributed segments in the time set to determine the continuous segment sequence in which the abnormal label appears on the time axis. Combine the start and end times of each segment with the corresponding channel and label, summarize and integrate them in chronological order to obtain the multi-channel concurrent abnormal time period.

[0028] As a further aspect of the present invention, the specific steps for obtaining the switch fault early warning information are as follows:

[0029] S511: Based on the channel numbers involved in the multi-channel concurrent anomaly period, count the number of behavior tags corresponding to each channel number in the covered time segment, extract the tag frequency information of the channel in the concurrent anomaly period, calculate the sum of the number of tag records in each channel, and obtain the channel tag frequency data.

[0030] S512: Based on the channel tag frequency data, and combining the total time length of each channel in the concurrent abnormal segment, the number of historical normal behavior tag types, and the number of abnormal co-occurrence tags, calculate the abnormal contribution frequency of the channel to the current concurrent abnormal segment, and statistically obtain the abnormal contribution quantity.

[0031] S513: Based on the number of abnormal contributions, filter all channel numbers with an abnormal contribution number greater than or equal to 2, construct a mapping table with the corresponding switch identification port, mark all channels that meet the conditions as the dominant status number, load the dominant status number and output it synchronously to the switch status identification port, and issue a switch fault warning message.

[0032] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0033] In this invention, the time intervals of multi-channel state transition responses are aligned to achieve dynamic linkage identification between multiple signals. Behavioral label combinations and sliding statistical strategies are used to strengthen the temporal correlation between abnormal behaviors. Aggregated expression of composite alarm events enhances the sensitivity and accuracy of anomaly identification. Cross-channel behavioral label matrix analysis clarifies the persistence of abnormal labels and identifies key channels, improving the accuracy and timeliness of switchgear early warning, reducing the risk of false alarms and missed alarms, and supporting real-time identification of concurrent and progressive complex faults. While improving the comprehensiveness of the identification logic, it also considers the clarity of the decision-making basis for alarm results, adapting to the actual needs of multi-source sensing data fusion analysis in switchgear, and enhancing the traceability and judgment value of fault evolution trends. Attached Figure Description

[0034] Figure 1 This is a flowchart of the main steps of the present invention;

[0035] Figure 2 This is a flowchart of the process for obtaining the response time interval set in this invention;

[0036] Figure 3 This is a flowchart of the process for obtaining time series groups of behavioral labels according to the present invention;

[0037] Figure 4 This is a flowchart illustrating the process of obtaining the combined trigger window set according to the present invention.

[0038] Figure 5 This is a flowchart of the multi-channel concurrent abnormal period acquisition process of the present invention;

[0039] Figure 6 This is a flowchart of the process for obtaining switch fault early warning information according to the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0042] Please see Figure 1 A fault early warning method for switchgear based on multi-source data fusion includes the following steps:

[0043] S1: Obtain the first timestamp corresponding to the state transition point (matching the "Power Quality Monitoring Standard") in the voltage channel, temperature channel, and vibration channel of the switching equipment. Arrange the channel numbers in chronological order and record the sorting sequence number. Combine the sorting sequence number with the first timestamp of the corresponding channel to form a time series tuple (using the TimestampWithQuality structure in the IEC61850 standard) to obtain the set of response time intervals.

[0044] S2: Based on the response time interval set, the sampling data of the three types of channels are horizontally aligned to detect whether the amplitude change of the voltage channel within two adjacent cycles is less than the voltage stability threshold, whether the difference of the temperature channel in three consecutive sampling points is less than or equal to the reference value (using the temperature gradient anomaly in IEEE C57.91-2011: temperature difference between adjacent sampling points ≤ 0.5℃ / min), and whether there is vibration energy decay in the vibration channel (according to ISO 10816-8 standard: ≤ 30% of the average value). If the corresponding conditions are met, they are marked as voltage stagnation, temperature passivation and vibration drop labels respectively and the corresponding timestamps are recorded to obtain the behavior label time series group.

[0045] S3: Based on the behavior tag time series group, construct a sliding window (set the window length to 1 second and the step size to 0.5 seconds), count the number of non-repeating behavior tags appearing in each sliding window, and when the number of tags is greater than or equal to 2, combine the corresponding sliding window number, tag type and corresponding channel number, and uniformly number them as composite alarm event ID (following the IEC62682 alarm management standard) to obtain the combined trigger window set;

[0046] S4: Based on the window numbers that appear consecutively in the set of combined trigger windows, perform cross-matrix analysis on the corresponding channel number and the corresponding behavior label (using the fault correlation matrix method in "Machine Condition Monitoring and Diagnosis"), and extract all time periods on the time axis where abnormal labels appear continuously to obtain the multi-channel concurrent abnormal time periods.

[0047] S5: Based on the channel number involved in the multi-channel concurrent abnormal period, and combined with the frequency of occurrence of each channel behavior label in the interval, calculate the single-channel abnormal contribution frequency, mark the channel with a contribution number greater than or equal to 2 as the abnormal dominant channel, output to the switch status identification port, and issue a switch fault warning message.

[0048] The response time interval set includes the channel response sequence, timestamp pairing structure, and time interval arrangement sequence. The behavior label time series group includes voltage stagnation label, temperature passivation label, and vibration drop label. The combined trigger window set includes sliding window number, composite alarm event ID, and label type matching result. The multi-channel concurrent abnormal period includes the channel number corresponding matrix analysis result, abnormal label duration segment, and abnormal label cross-mapping relationship. The switch fault early warning information includes behavior label frequency distribution record, channel number and abnormal correlation record, and switch fault early warning real-time record.

[0049] Please see Figure 2 The specific steps of S1 are as follows:

[0050] S111: Obtain the original signal sequences monitored in the voltage channel, temperature channel, and vibration channel of the switching equipment, identify and extract the first timestamp value corresponding to the state transition point, record them as the transition start time point of each channel, form a corresponding mapping set composed of channel number and timestamp value, and obtain the channel first jump time set;

[0051] Based on the raw signal sequences monitored in the voltage, temperature, and vibration channels of the switching equipment, continuous sampling data sequences collected by voltage sensors, thermocouples, and vibration accelerometers are retrieved respectively. Data is recorded at a fixed sampling frequency of once per millisecond, and each set of data undergoes state identification processing. The location of the state transition point is determined by comparing whether the abrupt change in the data between two consecutive time points exceeds the transition identification threshold set for that type of channel. For example, in the voltage channel, if the sampled value is 112.4V at sampling time 14500010ms, and the value jumps to 93.2V at 14500015ms, then because the amplitude change exceeds the set voltage transition threshold of 15V, this point is determined to be a state transition point, and the first timestamp of the transition is recorded as 14500015ms. Similarly, in the temperature channel, if the sampling point is 38.9℃ at 14500030ms and abruptly changes to 41.2℃ at 14500035ms, exceeding the threshold... The set temperature transition threshold is 2℃. A transition point is identified and the transition timestamp is recorded. In the vibration channel, 0.02g is used as the vibration amplitude transition judgment benchmark. If the sampled value at 14500080ms is 0.021g and changes to 0.012g at the next sampled point, the transition timestamp is recorded as 14500095ms. All data must meet the condition that the jump amplitude exceeds the threshold range of this channel to determine the validity of the state transition point. The state transition threshold needs to be set according to the equipment type standard. The threshold is set to 15V for the voltage channel, 2℃ for the temperature channel, and 0.01g for the vibration channel. The transition point detection process should ensure that the detection process in a single channel only identifies the first timestamp that meets the sudden change amplitude condition. Finally, by calculating point by point in the sampled data sequence and according to the threshold setting standard of each channel, the timestamp value corresponding to the first transition point in all channels is called. Combined with the channel number, a corresponding set of channel number and timestamp is formed to obtain the channel first jump time set.

[0052] S112: Based on the first hop time set of the channels, sort all channel numbers in ascending order according to the corresponding first hop timestamp value, record the original sequence number corresponding to the channel number after ascending order as the current sorting sequence number, construct a set of paired value pairs between the sorting sequence number and the first hop time of the channel, and obtain the time series mapping data group;

[0053] Based on the first-hop time set of the channels, the acquired voltage channel 1 time (14500015ms), temperature channel 2 time (14500035ms), and vibration channel 3 time (14500095ms) are first sorted in ascending order, resulting in the order: voltage channel 1, temperature channel 2, and vibration channel 3. The original sequence number 1 of the voltage channel is assigned the sorting sequence number 1, the original sequence number 2 of the temperature channel is assigned the sorting sequence number 2, and the original sequence number 3 of the vibration channel is assigned the sorting sequence number 3. This sorting sequence number is paired with its original transition time to form a value pair record. The timestamp quality identifier value is recorded according to the IEC61850 standard structure. The quality identifier consists of parameters such as synchronization accuracy and time source consistency. According to the above mapping rules, three sets of recorded values ​​are formed: sorting sequence number 1-14500015ms, sorting sequence number 2-14500035ms, and sorting sequence number 3-14500095ms. Finally, a pairing set between the sorting sequence number and the first-hop time is established to obtain the time series mapping data group.

[0054] S113: Based on the time series mapping data group, calculate the response time difference between adjacent channels based on the timestamp values ​​between adjacent sorting numbers, sort the time intervals between all channel sorting numbers in order and put them into a unified set to obtain the response time interval set.

[0055] Based on the obtained sorting sequence number and corresponding timestamp combination in the time series mapping data set, the response time difference is calculated for the timestamp differences between adjacent items of the sorting sequence number, namely 1 and 2, and 2 and 3. The transition time of the channel with sorting sequence number 1 is 14500015ms, and the transition time of the channel with sorting sequence number 2 is 14500035ms. The time interval is 14500035ms - 14500015ms = 20ms. Continuing to calculate, the time difference between sorting sequence numbers 2 and 3 is 14500095ms - 1450003ms. 5ms = 60ms. The two time differences of 20ms and 60ms are recorded sequentially to form a set of response time intervals. Each interval value in this set needs to meet the logical time response setting standard between devices. For example, each channel in the switching equipment should complete the sequence response process within 100ms. Therefore, the response time difference threshold is set to 100ms. When the interval of a certain channel exceeds this value, it is considered to be outside the response window. This value is calibrated according to the device response frequency in the actual system test and obtained by averaging 5 sets of samples. The sample data is shown in Table 1.

[0056] Table 1 Test Data on Response Time Interval

[0057] Sample number Voltage channel timestamp (ms) Temperature channel timestamp (ms) Vibration channel timestamp (ms) Response interval 1 (ms) Response interval 2 (ms) 1 14500010 14500035 14500095 25 60 2 14500012 14500030 14500100 18 70 3 14500008 14500033 14500090 25 57 4 14500014 14500034 14500093 20 59 5 14500011 14500032 14500096 21 64

[0058] As shown in Table 1, the average response time intervals between the voltage, temperature, and vibration channels in the test sample were 22.8 ms and 62 ms, respectively. Therefore, the upper limit threshold for the system response time difference was set to 100 ms. The response time used in the calculation must be derived from the timestamp difference of continuous effective transition signals, and the difference of repeated transition signals or invalid disturbance signals cannot be used. Finally, the difference values ​​between the above channels were recorded as the set of response time intervals.

[0059] Please see Figure 3 The specific steps of S2 are as follows:

[0060] S211: Based on the response time interval set, the sampling time axis of the voltage channel, temperature channel and vibration channel is uniformly aligned laterally. The minimum time step is used as the reference step to construct a uniform sampling time axis. The sampling amplitude record at the corresponding time point in each channel is extracted. The unsampled area is filled and the data alignment is completed by the previous value preservation method. After all the sampled data is aligned, a horizontal channel mapping is established to obtain a three-channel aligned data matrix.

[0061] Based on the response time interval set, the sampling time points of the voltage, temperature, and vibration channels within each sampling period are extracted, and the time axes of each channel are unified. The shortest effective sampling time step of 100ms is selected as the alignment benchmark, and the time axes of all channels are reconstructed according to this step. Data with effective sampling at the corresponding time point of each channel is extracted as the basis for alignment data. If data is missing at a certain time point, the gap is filled by the previous value preservation strategy. A complete horizontal channel data mapping table is constructed, and example sample data is extracted to construct the following alignment format, as shown in Table 2:

[0062] Table 2 Three-channel sampling alignment data table

[0063] Channel type Sampling time point 1 Sampling time point 2 Sampling time point 3 Sample value 1 Sample value 2 Sample value 3 Voltage channel 1000 1100 1200 220.0 219.8 220.1 Temperature Channel 1000 1100 1200 38.5 38.9 39.1 Vibration Channel 1000 1100 1200 0.062 0.058 0.017

[0064] As shown in Table 2, the sampling data of the three channels at three time points have been horizontally aligned using a reference step size. The amplitude variation of the voltage channel between the three points is within ±0.3V, the temperature channel difference is concentrated within 0.6℃, and the vibration channel shows a clear downward trend in values. This aligned data matrix serves as the basic data structure for subsequent behavior determination, and finally, the three-channel aligned data matrix is ​​obtained.

[0065] S212: Based on the three-channel aligned data matrix, detect whether the amplitude change range within two adjacent periodic sampling segments in the voltage channel is less than the voltage stability threshold. Calculate two sets of temperature differences for every three consecutive sampling points in the temperature channel and compare whether the maximum difference is less than or equal to the reference value. Extract the current sampling value from the vibration channel and determine whether it is lower than 30% of the mean. Label the sampling points of the channels that meet the above judgment conditions accordingly and generate a label judgment sequence.

[0066] Based on the aligned channel data structure in Table 2, stability and state identification judgment operations were performed on the voltage channel, temperature channel, and vibration channel respectively. First, the difference between the sampled values ​​of the voltage channel from 1000ms to 1100ms was calculated as 220.0V − 219.8V = 0.2V, and the difference between 1100ms and 1200ms was 219.8V − 220.1V = −0.3V. The absolute values ​​are all less than the voltage stability judgment threshold of 0.5V, which meets the voltage stagnation judgment condition. Second, for the temperature channel sampling point sequence of 38.5℃, 38.9℃, and 39.1℃, the continuous differences were calculated as 0.4℃ and 0.2℃, respectively, with the maximum difference being... The temperature is 0.4℃, which is lower than the temperature passivation benchmark value of 0.5℃, indicating that temperature passivation has occurred. For the vibration channel, three points are extracted with values ​​of 0.062g, 0.058g, and 0.017g respectively. The current point (1200ms) has a value of 0.017g. The percentage obtained by comparing it with the average value of the previous two points of 0.06g is 28.3%, which is less than the lower limit of energy attenuation judgment of 30%, and it is judged as a sudden drop in vibration. The above judgment process uses the original sampled data as input, and combines the benchmark threshold to filter and judge the state of the current channel at the current time point. After performing the judgment action in a loop for each time period, all the state labels that meet the conditions are recorded and combined to finally generate a label judgment sequence.

[0067] S213: Based on the tag determination sequence, map them to the tags of three categories: voltage stagnation, temperature passivation, and vibration drop, and integrate them into a structured recording unit in chronological order. Obtain the channel number and event timestamp corresponding to each type of behavior to obtain the behavior tag time sequence group.

[0068] Based on the judgment results of the assigned behavior labels in the label judgment sequence, the sampling time and channel information of all successfully identified behavior states are extracted point by point and mapped to a unified structured record format. The recording unit is composed of three elements: channel number, behavior category, and timestamp. For example, if the voltage channel hits the threshold and the amplitude fluctuation is less than the threshold, it is recorded as voltage stagnation behavior, and the record value V1_S_1100 is constructed. If the temperature channel has a temperature difference lower than the baseline at 1200ms, it constitutes temperature passivation, and it is recorded as T1_D_1200. If the vibration channel has an energy drop of less than 30% of the mean at 1200ms, it constitutes a sudden drop behavior, and it is recorded as A1_F_1200. Finally, all records are sorted by time and summarized into a unified sequence set to obtain the behavior label time series group.

[0069] Please see Figure 4 The specific steps of S3 are as follows:

[0070] S311: Obtain the behavior label, channel number and timestamp corresponding to each record in the behavior label time series group, construct a sliding window on the overall time axis according to the time order, filter the label records contained in each sliding window by time, extract all label records within the time window range and construct a label summary table, remove duplicate behavior labels and perform unique counting to generate window label counting results;

[0071] To construct the basic data structure for sliding window statistics, the behavior label, channel number, and timestamp of each record in the behavior label time series group are obtained. First, based on the timestamp information of the behavior label occurrence time, a time axis is established with the current maximum and minimum timestamp interval as the time range. Continuous time periods are then divided according to a 1-second window length and a 0.5-second step size, constructing continuous sliding window numbers and their start and end time ranges. Within each window range, the set of records falling within the current time period is selected from the behavior label time series. Each label record is then mapped to its corresponding window number, forming the label set corresponding to each sliding window. The behavior label content in the set is deduplicated to obtain a unique label type set. The number of elements in the set is used as the unique behavior label value under the current window. In actual deployment, this operation can be implemented as a sliding traversal based on the specific sampling period. For example, in a system where the sampling start and end time is from 0 seconds to 5 seconds, and the behavior labels appear at 1.2 seconds, 1.5 seconds, 1.9 seconds, 2.7 seconds, and 2.9 seconds respectively, the sliding window will generate 9 window segments by sliding from the start time of 0 seconds to the end time of 5 seconds. The label records within the window time range of each segment are counted, and the unique label type is extracted from each segment. The table is shown below:

[0072] Table 3. Statistics on the Number of Labels in the Sliding Window

[0073] Sliding window number Start time (s) End time (s) Number of tag records Number of unique tag types W1 0.0 1.0 0 0 W2 0.5 1.5 1 1 W3 1.0 2.0 3 2 W4 1.5 2.5 2 2 W5 2.0 3.0 3 2 W6 2.5 3.5 2 1

[0074] As shown in Table 3, by counting the number of non-repeating behavior labels appearing within the sliding window, the trend of label clustering change in each time window can be effectively extracted, and the window label counting result can be generated in the end.

[0075] S312: Based on the window label count results, filter all sliding windows with a label count greater than or equal to 2, extract the label type and corresponding channel number contained in the window, and combine them with the window number to form a unique identifier structure for all windows that meet the conditions. Based on the behavior label and channel number, integrate the information to obtain a composite label identifier set.

[0076] Based on the window label count, all sliding windows with a unique label count greater than or equal to 2 are selected. A combination structure of corresponding window number, label type, and channel number is established. For each record in the selected window, the label type and its corresponding channel number information are extracted, and a label-channel two-field combination table is constructed. Each combination field is concatenated in the format "label_channel number" and then combined with the sliding window number in a three-field structure. The above structure is used for subsequent generation of composite event identifiers. Event record numbers are uniformly assigned using the sliding window number as the index. For example, if window W3 contains labels A1, B1, and B2 belonging to channels V1, T1, and T2 respectively, then the combinations are constructed as A1_V1, B1_T1, and B2_T2. These are combined with the window number W3 to form W3_A1V1, W3_B1T1, and W3_B2T2, which are subsequently used for the unified generation of composite event numbers. A complete composite label table entry is constructed for all windows that meet the conditions, resulting in a composite label identifier set.

[0077] S313: Based on the composite tag identifier set, each composite record is assigned a number according to the alarm management specification. A unique composite alarm event identifier is constructed by combining the time number of the sliding window. The number is archived and integrated with the window start and end time, the involved channels and tag types. A unified sequence is constructed in chronological order to generate a set of combined trigger windows.

[0078] Based on the composite tag identifier set, a number mapping operation is performed on each record. Composite events are structurally encoded according to the IEC 62682 standard, using the ascending sliding window number as the primary number sequence. For each composite combination, a behavior type code and channel number are appended as secondary sequences according to the combination order. A unique event identifier is formed by concatenating these using a unified string structure. This structural numbering is constructed for all determined combinations. The event identifier, along with the start and end times of the original window time period, the behavior types involved in the combination, and the channel number set, are recorded in a structured record sequence. A composite alarm event entry table containing four fields—number, time, tag, and channel information—is constructed. For example, under window W3, an ID of "EVT03_A1V1_B1T1_B2T2" is constructed, corresponding to a time period from 1.0 second to 2.0 seconds, a tag set of {A1, B1, B2}, and a channel set of {V1, T1, T2}. Finally, a composite trigger window set is generated.

[0079] Please see Figure 5 The specific steps of S4 are as follows:

[0080] S411: Based on the sliding window number and corresponding channel number and behavior label recorded in the combined trigger window set, extract all sliding window numbers and sort them in chronological order. Calculate whether the difference between adjacent numbers in the window number sequence is 1. Merge the window numbers with a difference of 1 continuously to construct a set of time-continuous window segment sequences and obtain a set of continuous window labels.

[0081] Based on the sliding window number, channel number, and behavior label content recorded in the combined trigger window set, the sliding window number sequence is extracted, and all numbers are sorted in ascending order. Then, the interval between adjacent numbers is calculated item by item. If the difference between two adjacent numbers is 1, they are determined to form a continuous number segment. Adjacent numbers that meet the condition are grouped into window segment groups. For example, if the number sequence is W2, W3, W4, W6, W7, W9, then a continuous segment is constructed as {W2-W4, W6-W7}. The channel number and behavior label content involved in each segment are horizontally paired and extracted, and summarized into a channel-label combination record table, as shown in Table 4.

[0082] Table 4. Continuous Window Channel Label Combination Table

[0083] Window segment number Channel number Tag type Number of actions (times) Duration (seconds) W2-W4 V1 S 6 3 W2-W4 V2 D 7 3.5 W6-W7 T1 D 9 4 W6-W7 T1 F 3 2

[0084] As shown in Table 4, each continuous window segment contains multiple combinations of channel and behavior labels. This data structure serves as the basic input set for subsequent coupling degree calculation, ultimately yielding a continuous window label set.

[0085] S412: Based on the continuous window label set, construct a cross matrix structure with channel number as row index and row label as column index. Count the frequency of channel and label combinations appearing in the continuous window segment in each cross cell. Simultaneously, extract the co-occurrence count of each channel-row pair and apply the following formula:

[0086] ;

[0087] Calculate the behavioral coupling degree between channels and tags within the time window, determine if it exceeds the behavioral anomaly detection threshold, filter all channel behavior combinations that meet the coupling strength judgment criteria, and generate a channel-tag coupling degree matrix. Indicates channel With behavioral tags The degree of coupling, Indicates channel With tags The set of synchronous co-occurrence times on the timeline , Representing channels With tags The number of times the behavior, , Indicates its duration, , Indicates the frequency of behavior (Hz). , This represents the standard deviation (Hz) of the channel's fluctuation relative to the label.

[0088] Based on the continuous window label set, a two-dimensional cross matrix is ​​constructed, where the channel number is the row index and the behavior label is the column index. Within each cross cell, the frequency and fluctuation range of the corresponding channel and behavior label combination appearing in the continuous window segment are statistically analyzed. Relevant data are extracted from the sample. For example, if the number of occurrences of the behavior in the combination of channel V1 and label S is 6, and the duration is 3 seconds, then the behavior frequency is 2Hz, and the standard deviation is 0.8Hz. Label S itself has a total occurrence of 5 times, a duration of 2.5 seconds, a frequency of 2Hz, a standard deviation of 0.6Hz, and a co-occurrence time of 2 seconds. These values ​​are then substituted into the formula for calculation.

[0089] ;

[0090] Taking channel T1 and tag D as an example again, the co-occurrence time is 3s, the behavior frequencies are 9 / 4=2.25Hz and 7 / 3.5=2Hz respectively, and the standard deviations are 1.1Hz and 0.9Hz respectively. Therefore:

[0091] ;

[0092] Repeat the above process for all combinations to calculate the coupling degree matrix. For coupling degree values ​​greater than or equal to a set threshold,... The channel-label combinations are filtered and extracted to form a set of strong coupling relationships between channels and labels. The threshold is set based on the fact that the top 30% quantiles after sorting all coupling degree calculation results in the sample window segment are taken as the identification boundary. Moreover, this quantile falls in the range of [2.3, 2.7] in all five complete sliding window samples, so it is set to 2.5. Its value fluctuation is highly correlated with the channel behavior frequency and the label appearance density. The higher the channel frequency or the wider the label coverage, the easier it is for this value to rise. If the frequency is low but the synchronization is strong, it will also form a critical coupling. Finally, the channel-label coupling degree matrix is ​​generated.

[0093] Behavioral coupling is used to measure the strength of the synchronous association between a specific channel and a specific behavioral label in a time series. It reflects whether the two frequently occur simultaneously within a unit of time, whether the behavioral intensity is consistent when they occur, and whether their behavioral patterns are stable and consistent. Specifically, a higher coupling value indicates that the channel and the label have a highly overlapping time, high behavioral frequency, and low volatility cooperative relationship, suggesting that the behavioral label is likely a true reflection or dominant feature of the current operating state of the channel. Conversely, if the coupling value is low, it may mean that the appearance of the behavioral label is an accidental disturbance, or that there is a delay or inconsistency with the behavior of the channel itself. Therefore, this indicator can be used to identify channel label combinations with stable behavioral responses and system linkage characteristics in multi-channel concurrent monitoring scenarios, thereby supporting the subsequent identification of fault areas or the establishment of concurrent anomaly triggering mechanisms.

[0094] The formula aims to comprehensively measure the coupling degree between the co-occurrence density, behavioral intensity, and fluctuation stability of channels and behavioral tags over time. The numerator uses the co-occurrence duration of channels and tags on the time axis. With their respective frequency of behavior per unit time , The summation and multiplication represent the product of the degree of overlap between channel and tag behavior and their respective behavior density within the effective time, reflecting the synchronization strength; the denominator introduces... As a joint measure of the volatility of both, the total variability of the stability of the expression channel and the label behavior is considered; a larger variability indicates greater instability. Additionally, the absolute value of the frequency difference between the two is also included. It is used to punish combinations of inconsistent behavior patterns, and comprehensively forms a ratio relationship of "behavioral co-occurrence density to behavioral instability and inconsistency". This enables high scores for channel-label combinations that are highly synchronized and have high intensity but low fluctuation, while suppressing the scores of combinations with large frequency differences or drastic fluctuations, so as to ensure that the coupling degree index has the ability to identify and distinguish in behavioral state recognition.

[0095] S413: Based on the channel label coupling degree matrix, extract the corresponding time index and integrate the continuously distributed segments in the time set to determine the continuous segment sequence of abnormal labels appearing on the time axis. Combine the start and end times of each segment with the corresponding channel and label, summarize and integrate them in chronological order to obtain the multi-channel concurrent abnormal time period.

[0096] The channel label coupling matrix is ​​used to extract the associated timestamp range for channel-label combinations with a coupling value greater than or equal to 2.5. The co-occurrence time segment of each combination is located in the set, and adjacent time points are aggregated. Continuous time indices are merged into complete segments. For example, co-occurrence time points of 1.5s, 2.0s, 2.5s, and 3.0s are merged into the [1.5s–3.0s] segment. The channel number and behavior label type involved are then paired and recorded. All combinations that meet the conditions are integrated to construct a record table consisting of three items: channel number, label type, and time segment. Finally, the concurrent behavior structure data of all coupling segments with significant coupling is output to obtain the multi-channel concurrent abnormal time period.

[0097] Please see Figure 6 The specific steps of S5 are as follows:

[0098] S511: Based on the channel numbers involved in the multi-channel concurrent anomaly period, count the number of behavior tags corresponding to each channel number in the covered time segment, extract the tag frequency information of the channel in the concurrent anomaly period, calculate the sum of the number of tag records in each channel, and obtain the channel tag frequency data.

[0099] Based on the channel numbers involved in the multi-channel concurrent anomaly period, the sliding time interval corresponding to each channel number is extracted and its duration is statistically analyzed. The statistical content requires calling the original behavior tag record sequence within the anomaly period for each channel number, extracting the number of records in the sequence as the tag frequency value, and establishing a one-to-one mapping relationship between the frequency value and the channel number. When performing this process, the number of tag frames recorded per second based on the sliding window number needs to be accumulated. For example, channel G1 has 6 tag records in the anomaly segment W1 to W5, channel G2 has 9, and channel G3 has 4. The frequency data is recorded with the channel number as the index. If there are multiple window segments corresponding to the same channel number, their frequencies need to be accumulated. For example, G1 appears in two segments W1-W3 and W6-W7, with frequency values ​​of 3 and 5 respectively, so the sum of its frequency values ​​is 8. Finally, the above frequency values ​​are structured into key-value pairs to establish a corresponding mapping structure between the channel number and the total frequency of behavior tags, and the channel tag frequency data is obtained.

[0100] S512: Based on channel tag frequency data, and combining the total duration of each channel within the concurrency anomaly segment, the number of historical normal behavior tag types, and the number of anomaly co-occurring tags, the following formula is used:

[0101] ;

[0102] Calculate the frequency of abnormal contributions from the channel to the current concurrent abnormal segment, and statistically obtain the number of abnormal contributions. Indicates channel The frequency (Hz) of the abnormal contribution to the abnormal section. Indicates channel With tags Co-occurrence time (seconds) in the abnormal segment. Indicates channel Coverage time (in seconds) in this segment. Indicates channel With tags The co-occurrence frequency per unit time (Hz). Indicates label Frequency (Hz) identified per unit time in the overall segment. Indicates channel Number of historical normal behavior tag categories Indicates channel The number of historical anomaly co-occurrence tag types, The total number of tags involved in the calculation;

[0103] Based on the channel tag frequency data, the co-occurrence frequency, co-occurrence time, channel duration, tag duration, total tag frequency, number of historical normal tag types, and number of abnormal co-occurrence tag types for each channel and its corresponding tag within the segment are extracted. A time-frequency feature table is constructed for each channel and its associated tags. Taking channel G1 as an example, its co-occurrence time with tag K1 is 2.0 seconds, its own duration is 5.0 seconds, its frequency is 8 times, and its unit frequency is 1.6Hz. Meanwhile, tag K1 appears 12 times in this segment, its duration is 6.0 seconds, and its frequency is 2Hz. The following formula is used to calculate the co-occurrence frequency of G1 with the input data:

[0104] ;

[0105] The calculations for G2 using the input data are as follows:

[0106] ;

[0107] The calculations for G3 using the input data are as follows:

[0108] ;

[0109] Therefore, the abnormal contribution frequencies of each channel are G1=0.16Hz, G2=0.32Hz, and G3=0.23Hz. The relevant parameters are shown in Table 5.

[0110] Table 5 Calculation Parameters for Abnormal Contribution

[0111] Channel number Label number Co-occurrence duration (seconds) Channel duration (seconds) Tag duration (seconds) Co-occurrence frequency (times) Total frequency of tags (times) Normal label types Abnormal label types G1 K1 2.0 5.0 6.0 8 12 4 3 G2 K2 3.5 7.0 6.5 10 15 3 2 G3 K3 1.5 4.0 5.0 5 7 2 1

[0112] As shown in Table 5, all parameters are derived from the sliding window sampling process and behavior record statistics. The frequency is calculated by dividing the frequency by the duration. The total frequency of the tags needs to be summed and counted in all channels to obtain the number of abnormal contributions.

[0113] Anomaly contribution frequency measures the degree of aggregation of abnormal behavior patterns by a channel during a period of concurrent anomalies across multiple channels. Specifically, it reflects the co-occurrence density, frequency deviation, and behavioral response complexity of the channel with multiple anomaly tags per unit time. A higher index indicates that the channel has a significant co-occurrence distribution with anomaly tags in the anomaly segment, and its behavioral frequency deviates significantly from the average behavioral pattern of the tags. This suggests that the channel not only participated in the anomaly process but also possesses dominant behavioral characteristics in terms of time-frequency distribution. In addition, the anomaly contribution frequency also incorporates the number of tag types carried by the channel in its historical behavior, which is used to dynamically adjust the magnitude of contribution assessment and avoid misjudging channels with complex tag structures. Therefore, this index integrates three factors: current behavioral density, frequency difference, and historical tag distribution, and is a key calculation basis for identifying dominant anomaly channels.

[0114] The formula is constructed based on the coordinated behavioral characteristics of channels and labels within anomalous regions, where the numerator is derived by channel... With tags co-occurrence time Total channel time Perform ratio processing to obtain the normalized occurrence ratio of the tag in the channel, and then compare it with the channel's frequency per unit time. Average frequency of labels The absolute values ​​of the sum and difference between the two values ​​are added together to characterize the frequency matching strength and difference sensitivity. This composite feature is used as the coupling effect of the label on the channel, and a time-frequency coupling term is formed by multiplication. Then, for all labels... Summation is performed to form a channel. Multi-label coupling aggregation behavior expression; the denominator uses the channel history normal label count. Number of historical anomaly tags The square root operation is intended to reflect the structural complexity of channel label behavior. The square root operation slows down its growth, which plays a role in normalization and suppressing the influence of historical characteristics, thereby ensuring that the current coupling strength of the channel is the dominant indicator. The overall structure controls the label coverage range within the channel through duration normalization, guides the assessment of the offset strength of channel label pairs with the degree of frequency matching, and makes relative corrections in combination with the background of historical behavior, forming a stable and dynamically balanced abnormal contribution frequency measurement index.

[0115] S513: Based on the number of abnormal contributions, filter all channel numbers with an abnormal contribution of 2 or more, construct a mapping table with the corresponding switch identification port, mark all channels that meet the conditions as the dominant status number, load the dominant status number and output it synchronously to the switch status identification port, and issue a switch fault warning message.

[0116] Based on the number of abnormal contributions, assign a corresponding number to each channel. Values ​​are filtered to determine whether they meet the criteria for an abnormal dominant channel, and a threshold benchmark is set. This value was obtained by comprehensively statistically analyzing the frequency of occurrence and tag coverage in 18 sets of channel fault reproduction experiments. It falls at the critical turning point of the interval [0.2, 0.3]. The average value was selected as the setting benchmark. Combined with the previously calculated G1 value of 0.16Hz, G2 of 0.32Hz, and G3 of 0.23Hz, only G2 meets the setting conditions. Channel number G2 was selected, a mapping table between it and the control port was established, the status signal path was configured, channel number G2 was marked as the dominant channel, and a status identification signal was sent to the downstream identification module. The signal was written to the event stream recording module to obtain the switch fault warning information.

[0117] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for early warning of switchgear faults based on multi-source data fusion, characterized in that, Includes the following steps: S1: Obtain the first timestamp corresponding to the state transition point in the voltage channel, temperature channel, and vibration channel of the switching equipment, sort them according to time sequence, and obtain the set of response time intervals; S2: Based on the set of response time intervals, the sampled data is horizontally aligned to detect whether there are voltage stagnation, temperature passivation and vibration drop labels, and the corresponding timestamps are recorded to obtain a behavioral label time series group; S3: Based on the behavior tag time series group, construct a sliding window, count the number of non-repeating behavior tags in each sliding window, and when the number of tags is greater than or equal to 2, combine the corresponding sliding window number, tag type and corresponding channel number, and uniformly number them as composite alarm event ID to obtain a set of combined trigger windows. S4: Based on the combined trigger window set, perform cross matrix analysis on the channel number corresponding to the marker and the corresponding behavior label, and extract all time periods on the time axis where abnormal labels continuously appear to obtain the multi-channel concurrent abnormal time periods; S5: Based on the channel numbers involved in the multi-channel concurrent abnormal period, calculate the single-channel abnormal contribution frequency, mark the abnormal dominant channel, and issue a switch fault warning message; The specific steps for obtaining the multi-channel concurrent abnormal time period are as follows: S411: Based on the sliding window number and corresponding channel number and behavior label recorded in the combined trigger window set, extract all sliding window numbers and sort them in chronological order. Calculate whether the difference between adjacent numbers in the window number sequence is 1. Merge the window numbers with a difference of 1 continuously to construct a set of time-continuous window segment sequences and obtain a set of continuous window labels. S412: Based on the continuous window label set, construct a cross matrix structure with channel number as row index and behavior label as column index. In each cross cell, count the frequency of channel and label combination in the continuous window segment. At the same time, extract the co-occurrence number of each channel-behavior pair, calculate the behavior coupling degree of channel and label in the time window, and determine whether it exceeds the behavior anomaly judgment threshold. Filter all channel-behavior combinations that meet the coupling strength judgment standard and generate a channel label coupling degree matrix. S413: Based on the channel label coupling degree matrix, extract the corresponding time index and integrate the continuously distributed segments in the time set to determine the continuous segment sequence in which the abnormal label appears on the time axis. Combine the start and end times of each segment with the corresponding channel and label, summarize and integrate them in chronological order to obtain the multi-channel concurrent abnormal time period.

2. The method for early warning of switchgear faults based on multi-source data fusion according to claim 1, characterized in that, The response time interval set includes the channel response sequence, timestamp pairing structure, and time interval arrangement sequence; the behavior label time series group includes voltage stagnation label, temperature passivation label, and vibration drop label; the combined trigger window set includes sliding window number, composite alarm event ID, and label type matching result; the multi-channel concurrent abnormal period includes channel number corresponding matrix analysis result, abnormal label duration segment, and abnormal label cross-mapping relationship; the switch fault early warning information includes behavior label frequency distribution record, channel number and abnormal correlation record, and switch fault early warning real-time record. Voltage stagnation: The voltage channel amplitude change is less than the voltage stability threshold between two adjacent cycles; Temperature passivation: The difference in values ​​of the temperature channel across three consecutive sampling points is less than or equal to the reference value; Sudden vibration drop: Vibration energy attenuation occurs in the vibration channel; Abnormal dominant channel: A channel with a contribution number greater than or equal to 2.

3. The method for early warning of switchgear faults based on multi-source data fusion according to claim 1, characterized in that, The specific steps for obtaining the set of response time intervals are as follows: S111: Obtain the original signal sequences monitored in the voltage channel, temperature channel, and vibration channel of the switching equipment, identify and extract the first timestamp value corresponding to the state transition point, record them as the transition start time point of each channel, form a corresponding mapping set composed of channel number and timestamp value, and obtain the channel first jump time set; S112: Based on the set of first-hop times of the channels, sort all channel numbers in ascending order according to the corresponding first-hop timestamp values, record the original sequence number corresponding to the channel numbers after ascending order as the current sorting sequence number, construct a set of paired value pairs between the sorting sequence number and the first-hop time of the channel, and obtain the time series mapping data group; S113: Based on the time series mapping data group, calculate the response time difference between adjacent channels based on the timestamp values ​​between adjacent sorting numbers, sort the time intervals between all channel sorting numbers in order and put them into a unified set to obtain the response time interval set.

4. The method for early warning of switchgear faults based on multi-source data fusion according to claim 1, characterized in that, The specific steps for obtaining the behavioral label time series group are as follows: S211: Based on the set of response time intervals, the sampling time axes of the voltage channel, temperature channel, and vibration channel are uniformly aligned laterally. A uniform sampling time axis is constructed using the minimum time step as the reference step. The sampling amplitude records at the corresponding time points in each channel are extracted. The unsampled areas are filled in using the previous value preservation method to complete the data alignment. After all the sampled data are aligned, a horizontal channel mapping is established to obtain a three-channel aligned data matrix. S212: Based on the three-channel aligned data matrix, detect whether the amplitude change range within two adjacent periodic sampling segments in the voltage channel is less than the voltage stability threshold; calculate two sets of temperature differences for every three consecutive sampling points in the temperature channel and compare whether the maximum difference is less than or equal to the reference value; extract the current sampling value in the vibration channel and determine whether it is lower than 30% of the mean; and assign corresponding labels to the channel sampling points that satisfy the following conditions: the amplitude change range within two adjacent periodic sampling segments is less than the voltage stability threshold; the maximum difference of the two sets of temperature differences calculated for every three consecutive sampling points in the temperature channel is less than or equal to the reference value; and the current sampling value extracted in the vibration channel is lower than 30% of the mean, thereby generating a label determination sequence. S213: Based on the tag determination sequence, map them to the tags of three categories: voltage stagnation, temperature passivation, and vibration drop, respectively, integrate them into a structured recording unit in chronological order, obtain the channel number and event timestamp corresponding to each type of behavior, and obtain the behavior tag time sequence group.

5. The method for early warning of switchgear faults based on multi-source data fusion according to claim 1, characterized in that, The specific steps for obtaining the combined trigger window set are as follows: S311: Obtain the behavior label, channel number and timestamp corresponding to each record in the behavior label time series group, construct a sliding window on the overall time axis according to the time order, filter the label records contained in each sliding window by time, extract all label records within the time window range and construct a label summary table, remove duplicate behavior labels and perform unique counting to generate window label counting results; S312: Based on the window label counting results, filter all sliding windows with a label count greater than or equal to 2, extract the label type and corresponding channel number contained in the window, and combine them with the window number to form a unique identifier structure for all windows that meet the conditions. Based on the behavior label and channel number, integrate the information to obtain a composite label identifier set. S313: Based on the composite tag identifier set, assign a number to each composite record according to the alarm management specification, construct a unique composite alarm event identifier by combining the time number of the sliding window, and archive and integrate the number with the window start and end time, the involved channels and tag types, construct a unified sequence in chronological order, and generate a set of combined trigger windows.

6. The method for early warning of switchgear faults based on multi-source data fusion according to claim 1, characterized in that, The specific steps for obtaining the switch fault early warning information are as follows: S511: Based on the channel numbers involved in the multi-channel concurrent anomaly period, count the number of behavior tags corresponding to each channel number in the covered time segment, extract the tag frequency information of the channel in the concurrent anomaly period, calculate the sum of the number of tag records in each channel, and obtain the channel tag frequency data. S512: Based on the channel tag frequency data, and combining the total time length of each channel in the concurrent abnormal segment, the number of historical normal behavior tag types, and the number of abnormal co-occurrence tags, calculate the abnormal contribution frequency of the channel to the current concurrent abnormal segment, and statistically obtain the abnormal contribution quantity. S513: Based on the number of abnormal contributions, filter all channel numbers with an abnormal contribution number greater than or equal to 2, construct a mapping table with the corresponding switch identification port, mark all channels that meet the conditions as the dominant status number, load the dominant status number and output it synchronously to the switch status identification port, and issue a switch fault warning message.

Images