Transformer operation anomaly prediction method and system

Abstract

The present application relates to the technical field of anomaly prediction, in particular to a transformer operation anomaly prediction method and system, comprising the following steps: obtaining transformer temperature, current, oil product and vibration parameters, aligning sample sequences and calculating parameter pair Pearson correlation, analyzing time series correlation mutation and extracting feature groups, judging temperature difference and current variability to mark abnormal points, constructing trend sequences to evaluate risk trends, and outputting abnormal trend prediction results. In the present application, by capturing the linkage characteristics between multiple parameter pairs in the time window, sudden change nodes can be identified, multiple parameter statistical characteristics in the time interval before and after the abnormal time point are extracted, a composite feature set representing mutation behavior is constructed, and the abnormal state is determined under the combination condition of maximum difference and coefficient of variation. Combined with the trend time series, risk evolution identification is realized by accumulating abnormal frequency, improving the stability and foresight of transformer operation anomaly prediction.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of anomaly prediction, in particular to a transformer operation anomaly prediction method and system. BACKGROUND

[0002] The technical field of anomaly prediction mainly involves state monitoring, data analysis and fault prediction of equipment or systems through various methods and technical means, so as to identify potential abnormal situations in advance, thereby realizing efficient management and maintenance of equipment. In this technical field, the core tasks include acquisition and analysis of sensor data, construction and optimization of prediction models, application of anomaly detection algorithms, etc. The technical means of anomaly prediction involve signal processing, feature extraction, machine learning, deep learning and other technologies to improve the prediction accuracy and timeliness, and thus realize preventive maintenance and optimized operation of equipment.

[0003] Among them, the traditional transformer operation anomaly prediction method refers to monitoring and analyzing the operation state data of the transformer to identify possible abnormal situations of the transformer. This method usually includes steps such as collection and transmission of transformer operation data, data feature extraction and analysis, and abnormal pattern recognition. In practice, the traditional transformer anomaly prediction method uses statistical analysis method based on historical data or diagnostic model based on experience rules, and monitors the temperature, voltage, current and other operating parameters of the transformer to identify possible faults. However, the traditional method has certain limitations in dealing with complex and variable transformer operating environments, and the accuracy and timeliness of the prediction results need to be improved. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide a transformer operation anomaly prediction method and system.

[0005] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme, a transformer operation anomaly prediction method, comprising the following steps:

[0006] S1: Obtain real-time operating parameters of the transformer, construct parameter pairs of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters, respectively, calculate the correlation value of the parameter pairs in the current window, and obtain the parameter pair time correlation data set;

[0007] S2: Based on the parameter pair time correlation data set, calculate the first-order difference value in the adjacent time window, judge whether the first-order difference value exceeds the preset correlation fluctuation threshold, filter the time nodes with sudden changes and record the corresponding parameter pairs, and obtain the mutation time point and parameter mapping table;

[0008] S3: call the time node information in the mutation time point and parameter mapping table, extract the maximum value and minimum value difference, mean square deviation and coefficient of variation on the time gradient of each parameter, combine into a characteristic combination set corresponding to each group of mutations under the mapping relationship, and obtain a parameter mutation window feature group;

[0009] S4: according to the parameter mutation window feature group, judge whether the maximum value and minimum value difference of the temperature parameter exceeds the preset range threshold, and whether the coefficient of variation of the current parameter is greater than three times the historical sliding mean value of the current window, if it is satisfied, mark the time point as an abnormal state, and obtain an abnormal time index set;

[0010] S5: call the time point determined to be abnormal in the abnormal time index set, extract the order value arranged in time window, construct a trend time sequence, accumulate the frequency according to the number of abnormalities in unit time, calculate the abnormal risk trend in the target period, judge whether it needs to implement risk warning, and obtain the abnormal trend prediction result.

[0011] As a further scheme of the application, the parameter pair time sequence correlation data group includes a correlation curve sequence, a time window label set and a parameter combination index table, the mutation time point and parameter mapping table includes a mutation time set, a parameter pair mapping list and a window index label, the parameter mutation window feature group includes a statistical feature matrix, a mutation parameter mapping unit and a sample time interval set, the abnormal time index set includes an abnormal time point set, an abnormal identification label group and a parameter abnormal positioning cluster, and the abnormal trend prediction result includes a time trend array, an abnormal cumulative frequency value and a risk window index table.

[0012] As a further scheme of the application, the parameter pair time sequence correlation data group is obtained by:

[0013] S111: obtain the real-time running parameters of the transformer, including the continuous monitoring samples of temperature parameters, current parameters, oil parameters and vibration parameters, resample the data frames of each type of parameters according to the same sampling time stamp, perform time axis alignment processing on the sample records in the parameter frame in turn, implement data structure splicing under the same time index, and generate a multi-parameter time alignment sequence;

[0014] S112: call the continuous sample section of each type of parameter in the fixed time window in the multi-parameter time alignment sequence, construct three groups of parameter combinations of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters, and generate each group of parallel sample pairs according to the window position division, calculate the Pearson correlation coefficient value of adjacent time periods in each group of sample pairs, and obtain a parameter combination correlation result set;

[0015] S113: According to the time window index of each set of correlation values in the parameter combination correlation result set, sequentially arrange each set of correlation values in a window sequence, combine the corresponding parameter combination label, correlation value and timestamp field to construct a complete record, form a continuity trend description structure, and generate a parameter pair time sequence correlation data set.

[0016] As a further scheme of the present application, the acquisition step of the mutation time point and parameter mapping table is specifically:

[0017] S211: Based on the continuous numerical sequence of each parameter pair in the parameter pair time sequence correlation data set, sequentially extract the sequence values of each parameter in the adjacent time window, perform difference calculation on the numerical pairs of adjacent windows, construct a first-order difference sequence, and perform difference processing on all parameter pairs to obtain a correlation difference sequence set;

[0018] S212: Call each parameter pair sequence in the correlation difference sequence set, calculate the sliding standard deviation of the correlation value in the same window range and record the statistical value, set the offset judgment threshold to twice the sliding standard deviation, and compare each difference value with the offset judgment threshold, mark the time index of the difference value greater than the offset judgment threshold, and obtain an offset judgment marker set;

[0019] S213: According to the window index of the offset judged in the offset judgment marker set, retrieve the timestamp information of the corresponding parameter pair in the curve set, aggregate the timestamp field and the parameter pair label field to construct a mapping record table, and sequentially arrange the mapping records to obtain a mutation time point and parameter mapping table.

[0020] As a further scheme of the present application, the acquisition step of the parameter mutation window feature set is specifically:

[0021] S311: Call the mutation time index in the mutation time point and parameter mapping table, respectively extend three data sampling intervals forward and backward for each time point, retrieve the continuous data sequence of the temperature parameter, current parameter, oil parameter and vibration parameter in the corresponding period, extract the complete numerical sequence of each parameter in the extended interval, and obtain an extended window parameter sequence set;

[0022] S312: According to each parameter sequence in the extended window parameter sequence set, sequentially perform extreme value extraction on each sequence, calculate the difference between the maximum value and the minimum value, and calculate the mean square error value in the time dimension, obtain the standard deviation based on the current sequence mean value, construct a coefficient of variation set, and obtain window statistical feature information;

[0023] S313: Based on the three statistical feature values of each parameter group in the window statistical feature information, the parameter features at the same mutation time point are combined according to the label mapping rule of the mutation time and the parameter, and the parameter features at the same mutation time point are constructed into a unified structure data set and sequentially numbered to obtain a parameter mutation window feature group.

[0024] As a further scheme of the present application, the step of acquiring the abnormal time index set is specifically:

[0025] S411: According to the parameter feature records in the parameter mutation window feature group, the maximum value and the minimum value difference and the coefficient of variation corresponding to the temperature parameter and the current parameter are extracted in sequence, the two types of statistical quantities are aggregated under the same time index to form a combined quantity sequence, and the time index is arranged and collected to obtain a statistical combined quantity sequence;

[0026] S412: The combined quantity values under the time index in the statistical combined quantity sequence are called, the maximum value and the minimum value difference of the temperature parameter are compared with the preset range threshold value, and the coefficient of variation of the current parameter is compared with three times of the historical sliding mean value of the current window, and the index records that satisfy both judgment conditions are recorded as effective marker points to generate an abnormal marker index set;

[0027] S413: Based on the time index recorded in the abnormal marker index set, the corresponding time field is searched and the indexes that satisfy the abnormal condition are sequentially processed to construct a time index set structure and output as a unique result unit, and an abnormal time index set is established.

[0028] As a further scheme of the present application, the step of acquiring the abnormal trend prediction result is specifically:

[0029] S511: The abnormal time points recorded in the abnormal time index set are called, an equal interval time window is expanded before and after each time point, the continuous time period sample sequence of the temperature parameter, the current parameter, the oil parameter and the vibration parameter is searched, the mean value difference value of each type of parameter is calculated in sequence, and the sliding mean value difference information is generated;

[0030] S512: According to the time sequence index in the sliding mean value difference information, the mean value difference of each parameter under the same abnormal index is extracted, the number of abnormal triggers in each unit of time is accumulated according to the window sequence on the time axis, and the abnormal risk change trend value is calculated and obtained;

[0031] S513: According to the abnormal risk change trend value, it is judged whether there is a situation that the abnormal risk change trend value exceeds the risk trend threshold value in the continuous time period, if it is satisfied, the time period is marked as a high-risk area, a risk warning signal is generated and output, and an abnormal trend prediction result is obtained.

[0032] As a further scheme of the present application, the formula for calculating the abnormal risk trend value is specifically:

[0033] ;

[0034] wherein, represents the abnormal risk trend value in a time window, represents the sliding mean normalized value of the i-th monitoring parameter in the time window, represents the sliding mean normalized value of the i-th monitoring parameter in the time window, represents the abnormal trigger frequency of the i-th monitoring parameter in the time window, represents the abnormal trigger frequency of the i-th monitoring parameter in the time window, represents the abnormal trigger frequency of the i-th monitoring parameter in the time window, represents the normalized variance sum of the sliding means of all parameters in the time window, represents the abnormal trigger frequency of the i-th monitoring parameter in the time window, represents the abnormal trigger frequency of the i-th monitoring parameter in the time window, represents the total number of parameters participating in monitoring.

[0035] A transformer operation abnormality prediction system is used to implement the transformer operation abnormality prediction method, and the system comprises:

[0036] A correlation evaluation module acquires real-time operation parameters of the transformer, constructs parameter pairs of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters, respectively, calculates correlation values of the parameter pairs in a current window, and obtains a parameter pair time series correlation data set;

[0037] A mutation screening module calculates a first-order difference value in adjacent time windows based on the parameter pair time series correlation data set, judges whether the first-order difference value exceeds a preset correlation fluctuation threshold, screens time nodes with mutation behaviors and records corresponding parameter pairs, and obtains a mutation time point and parameter mapping table;

[0038] A mutation feature analysis module calls time node information in the mutation time point and parameter mapping table, extracts a maximum value and a minimum value difference of each parameter, a mean square error on a time gradient, and a coefficient of variation, combines each group of mutation corresponding feature combination sets under a mapping relationship, and obtains a parameter mutation window feature group;

[0039] ​​​​​​​An abnormal state identification module judges whether the difference between the maximum value and the minimum value of the temperature parameter exceeds a preset range threshold and whether the coefficient of variation of the current parameter is greater than three times the historical sliding average of the current window, and if so, marks the time point as an abnormal state to obtain an abnormal time index set;

[0040] An abnormal trend analysis module calls the time points in the abnormal time index set that have been determined to be abnormal, extracts the order values arranged in time windows, constructs a trend time series, performs cumulative frequency processing on the number of abnormalities per unit time, calculates the abnormal risk trend in the target period, judges whether risk warning needs to be implemented, and obtains an abnormal trend prediction result.

[0041] Compared with the prior art, the advantages and positive effects of the present application are that:

[0042] In the present application, by constructing a time sequence correlation data set of temperature, current, oil and vibration and calculating the Pearson coefficient, the linkage characteristics between multiple parameters in the time window are captured, the mutation threshold is set in combination with the first-order difference value and the sliding standard deviation, the sudden change nodes can be identified, the statistical characteristics of multiple parameters in the time interval before and after the abnormal time point are extracted, the composite feature set representing the mutation behavior is constructed, the abnormal state is determined under the combination condition of the maximum difference and the coefficient of variation, the risk evolution identification is realized in combination with the trend time series and the abnormal frequency accumulation, the sensitivity and accuracy of the multi-parameter dynamic abnormality detection are enhanced, and the stability and foresight of the transformer operation abnormality prediction are improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] Figure 1 The workflow diagram of the present application;

[0045] Figure 2 The flowchart of obtaining the parameter pair time sequence correlation data set of the present application;

[0046] Figure 3 The flowchart of obtaining the mutation time point and parameter mapping table of the present application;

[0047] Figure 4 The flowchart of obtaining the parameter mutation window feature set of the present application;

[0048] Figure 5 The flowchart of obtaining the abnormal time index set of the present application;

[0049] Figure 6 Flow chart for obtaining abnormal trend prediction result of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the present application will be described below with reference to the drawings.

[0051] In the embodiments of the present application, the words such as "example", "for example" are used to represent an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0052] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized. "Of", "corresponding" and "corresponding" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0053] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1, and the meanings expressed are consistent when the distinction is not emphasized.

[0054] In order to make the technical problems, technical solutions and advantages of the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0055] Please refer to Figure 1 The present application provides a technical solution, a transformer operation abnormality prediction method, comprising the following steps:

[0056] S1: Obtain real-time operation parameters of the transformer, including continuous monitoring samples of temperature parameters, current parameters, oil parameters and vibration parameters, align the parameters with samples according to a unified time stamp, call parallel sample sequences of each parameter combination in an equal time window, construct parameter pairs of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters, respectively, calculate the correlation value of the parameter pairs in the current window by using the Pearson coefficient, arrange the window to form a time sequence, and obtain a parameter pair time sequence correlation data set;

[0057] S2: Based on the parameter pair time series correlation data set, the continuous value sequence of each parameter pair curve is extracted, the first-order difference value is calculated in the adjacent time window, the offset judgment threshold is set as twice the current parameter pair sliding standard deviation, it is judged whether the first-order difference value exceeds the preset correlation fluctuation threshold, the time nodes with mutation behavior are screened and the corresponding parameter pairs are recorded, and a mutation time point and parameter mapping table is obtained;

[0058] S3: The time node information in the mutation time point and parameter mapping table is called, the temperature parameters, current parameters, oil parameters and vibration parameters of the transformer in the three groups of data intervals extending forward and backward from the mutation time point are retrieved, the maximum value and minimum value difference, the mean square deviation and the variation coefficient on the time gradient of each parameter are extracted, and the maximum value and minimum value difference and the variation coefficient are combined into a feature combination set corresponding to each mutation under the mapping relationship to obtain a parameter mutation window feature group;

[0059]

[0060] S5: The time points determined to be abnormal in the abnormal time index set are called, the sliding mean value difference of each parameter in the time span interval is retrieved, the order value arranged according to the time window is extracted, a trend time series is constructed, the cumulative frequency is processed according to the number of abnormalities in a unit time, the abnormal risk trend in the target period is calculated, it is judged whether risk warning needs to be implemented, and an abnormal trend prediction result is obtained.

[0061] The parameter pair time series correlation data set includes a correlation curve sequence, a time window label set and a parameter combination index table, the mutation time point and parameter mapping table includes a mutation time set, a parameter pair mapping list and a window index label, the parameter mutation window feature group includes a statistical feature matrix, a mutation parameter mapping unit and a sample time interval set, the abnormal time index set includes an abnormal time point set, an abnormal identification label group and a parameter abnormality positioning cluster, and the abnormal trend prediction result includes a time trend array, an abnormal cumulative frequency value and a risk window index table.

[0062] Please refer to Figure 2 The acquisition steps of the parameter pair time series correlation data set are as follows:

[0063] ​S111: Obtain real-time operating parameters of the transformer, including continuous monitoring samples of temperature, current, oil, and vibration parameters. Resample the data frames of each type of parameter according to the same sampling timestamp, perform time axis alignment processing on the sample records in the parameter frames in sequence, and perform data structure splicing under the same time index to generate a multi-parameter time-aligned sequence.

[0064] Data is collected by an array of sensors deployed on and around the transformer, with temperature parameters measured via fiber optic temperature sensors. The top oil temperature and winding hot spot temperature are collected at a high frequency, and the current parameters are transmitted through a high-frequency current transformer. The sampling rate is used to obtain the grounding current of the bushing end screen; oil parameters are obtained by acquiring the dissolved gas concentration in the oil once every 4 hours using an online chromatographic monitoring device; vibration parameters are obtained by using a piezoelectric accelerometer. The frequency of the vibration signal from the enclosure wall is collected, and the four types of raw data streams are transmitted to the central processing unit. A uniform resampling time interval is set. For high-frequency sampled data, namely current and vibration parameters, a moving average downsampling method is used to map every 1000 or 2000 original sample points to a resampling point. For low-frequency sampled data, namely temperature and oil parameters, a linear interpolation method is used to fill in missing time values, thereby unifying the data density at a time granularity of 0.1s. Then, using the first timestamp after resampling as the benchmark, all parameter stream data frames are traversed, and times when valid values ​​cannot be obtained simultaneously in all four dimensions are removed. The time indices where values ​​exist for all four types of parameters are retained. The temperature values ​​at the same time are then processed in ascending order of time indices. Current value Oil quality values and vibration values Horizontal splicing, constructing dimensions as The matrix structure, where The total number of samples is used to form a multi-parameter time-aligned sequence containing complete timestamp information.

[0065] S112: Call the continuous sample segments of each type of parameter in the multi-parameter time-aligned sequence within a fixed time window, construct three sets of parameter combinations: temperature parameter and current parameter, oil parameter and temperature parameter, and vibration parameter and current parameter, and generate parallel sample pairs for each set according to the window position. Calculate the Pearson correlation coefficient values ​​of adjacent time periods in each sample pair to obtain the parameter combination correlation result set.

[0066] Set time window length With 600 sampling points corresponding to a monitoring duration of 60 seconds, and a sliding step size of 60 sampling points, temperature sequences were extracted from the aligned sequence. Current sequence , oil product sequence and vibration sequence , three groups of vector pairs to be analyzed are constructed, i.e. 、 and For each group of vector pairs, such as temperature and current combination, the covariance and the respective standard deviation and are calculated. The Pearson correlation coefficient is solved by the formula , the mean value of the temperature sequence in a certain window is , the mean value of the current sequence is , the calculated covariance is 12.5, the temperature standard deviation is 2.1, and the current standard deviation is 6.5, then the correlation coefficient in this window is

[0067] S113: According to the time window index of each correlation value in the parameter combination correlation result set, each correlation value is sequentially windowed and arranged, the corresponding parameter combination label, correlation value and time stamp field are combined to construct a complete record, form a continuity trend description structure, and generate a parameter pair time sequence correlation data set.

[0068] The start time stamp of the time window is taken as the unique identifier of the window calculation result, a structured data table is created, which contains four column fields, namely time stamp index, parameter combination label, Pearson correlation coefficient value and window state identifier, for example, for the window with time stamp , the correlation value corresponding to the parameter combination label "Temp-Current" is 0.915, the correlation value corresponding to the label "Oil-Temp" is 0.882, and the correlation value corresponding to the label "Vib-Current" is 0.756. The calculation results of subsequent windows are appended row by row to the data table according to the time axis direction, maintaining the time continuity of the records. For the window that cannot calculate the correlation due to sensor failure or data loss, fill in the null identifier in the correlation value field and record the log. Finally, the correlation evolution process of the three groups of parameter pairs in all time periods is integrated into a three-dimensional tensor or multi-column time sequence file, and the parameter pair time sequence correlation data set is generated.

[0069] Please refer to Figure 3 , the acquisition steps of the mutation time point and parameter mapping table are as follows:

[0070] S211: Based on the continuous numerical sequence of each parameter pair in the time correlation data set, the sequence values of each parameter in the adjacent time window are extracted in turn, the difference calculation is performed on the numerical pairs of adjacent windows, the first-order difference sequence is constructed, and the difference processing is performed on all parameter pairs to obtain the correlation difference sequence group;

[0071] The first-order difference operation rule is defined as the correlation value of the current window minus the correlation value of the previous adjacent window, that is , wherein is the window number, for the "Temp-Current" combination, if the correlation value of the first window is 0.915, the correlation value of the first window is 0.912, and the first-order difference value at this moment is 0.003, if the first window value suddenly changes to 0.850, then the difference value at the next moment is , and the subtraction operation is performed on all adjacent points in the sequence in turn. This process eliminates the static reference deviation of the correlation coefficient and only retains the fluctuation amplitude over time. Similarly, the same difference processing is performed on the "Oil-Temp" and "Vib-Current" two groups of sequences in parallel, and the three groups of difference values obtained are rearranged according to the original time index to form the correlation difference sequence group reflecting the correlation change rate.

[0072] S212: Call each parameter pair sequence in the correlation difference sequence group, calculate the sliding standard deviation of the correlation value in the same window range and record the statistical value, set the offset judgment threshold as twice the sliding standard deviation, and compare each difference value with the offset judgment threshold, mark the time index of the difference value greater than the offset judgment threshold, and obtain the offset judgment marker set;

[0073] The sliding statistical window size is set to 10 difference points, the sliding standard deviation of the correlation difference value is calculated in the same window range, for example, for a difference sequence , the standard deviation of this segment of data is 0.0025, the offset judgment threshold is set to twice the sliding standard deviation, that is , in this example , then the absolute value of each difference value in the window is compared with the dynamic threshold, if , for example, a point difference value is , and its absolute value is much larger than , then it is determined that the point is a correlation mutation point, and the time index corresponding to the point is recorded The sequence is scanned point by point and the threshold value is dynamically updated, and the positions of all mutation points meeting the conditions are collected to obtain a set of offset judgment markers.

[0074] S213: According to the window index in the offset judgment marker set that is judged as offset, the corresponding parameter pair in the timestamp information in the curve group is retrieved, the timestamp field and the parameter pair label field are aggregated to construct a mapping record table, and the mapping record is serialized and arranged to obtain a mutation time point and parameter mapping table;

[0075] Each index value in the marker set is traversed, and the exact timestamp corresponding to the index in the original alignment sequence is retrieved, for example, index The corresponding timestamp is , identify which group of parameter pairs triggered the offset, for example, triggered by the "Temp-Current" combination, then build a mapping record If multiple groups of parameter pairs occur at the same time, multiple records are generated, the generated records are sorted according to time sequence, duplicate items are removed and the validity of the timestamp is checked, the arranged records are stored in the form of key-value pair list or relational database table, and the exact time of each mutation event and the involved physical parameter coupling relationship are clear. Obtain a mutation time point and parameter mapping table.

[0076] Please refer to Figure 4 The acquisition steps of the parameter mutation window feature group are as follows:

[0077] S311: Call the mutation time index in the mutation time point and parameter mapping table, respectively extend three data sampling intervals forward and backward for each time point, retrieve the continuous data sequence of temperature parameters, current parameters, oil parameters and vibration parameters in the corresponding period, extract the complete value sequence of each parameter in the extended interval, and obtain the extended window parameter sequence set;

[0078] For example, lock the mutation time point , set the data sampling interval , respectively extend 3 sampling intervals forward and backward, that is, the determined time interval is , which contains a total of 7 sampling points. Retrieve all original data in this time period in the multi-parameter time alignment sequence, extract temperature parameter sequence , current parameter sequence , oil parameter sequence and vibration parameter sequence , for example, the extracted current sequence is (unit: A), the extracted temperature sequence is (unit: ), the above expansion and extraction operations are repeatedly performed for each recorded mutation point, ensuring that the mutation moment and the transient waveform details before and after the mutation moment are captured, the extracted short sequences are stored according to the mutation event ID, and the expansion window parameter sequence set is obtained.

[0079] S312: According to each group of parameter sequences in the expansion window parameter sequence set, extreme value extraction is sequentially performed on each sequence, the difference between the maximum value and the minimum value is calculated, the mean square error value in the time dimension is calculated, the standard deviation is obtained based on the current sequence mean value, a coefficient of variation set is constructed, and window statistical feature information is obtained;

[0080] For the extracted point sequence, the maximum value , the minimum value , the extreme value difference , the time dimension mean square error (i.e. variance) of the sequence is calculated, the mean value is first calculated , and then the variance is calculated , so as to obtain the standard deviation , and the coefficient of variation , and the calculation formula is , and the numerical calculation is , and the extreme value difference and the corresponding coefficient of variation of the temperature sequence are calculated, and the statistical quantity calculation of the oil and vibration parameters is sequentially completed, the extreme value difference, the mean square error and the coefficient of variation of each parameter are packaged as a group of feature vectors, and the window statistical feature information is obtained.

[0081] S313: Based on the three statistical feature values of each parameter group in the window statistical feature information, the feature combination is performed according to the label mapping rule of the mutation time and the parameter, the parameter features under the same mutation time point are constructed into a unified structure data set and are sequentially numbered, and a parameter mutation window feature group is obtained.

[0082] According to the rule in the mapping table, the feature values under a specific mutation point are structurally associated, for example, for the time , the feature vectors of the temperature parameter (corresponding to the extreme value difference, the mean square error and the coefficient of variation, respectively), the feature vectors of the current parameter , and the feature vectors of the oil and vibration are merged to construct a unified feature data set row in the form of , and the unique sequence number of the data set is allocated, for example , ensuring that all feature data correspond to the original mutation event one by one and have traceability, and for a complex event involving multiple parameters, all single-parameter features involved are integrated in the same record to obtain a parameter mutation window feature group.

[0083] Please refer to Figure 5The specific steps for obtaining the abnormal time index set are as follows:

[0084] S411: Based on the parameter feature records in the parameter mutation window feature group, extract the maximum and minimum value differences and coefficient of variation of the temperature parameter and the current parameter in sequence. Aggregate the two types of statistics under the same time index to form a combined quantity sequence, and perform sorting and aggregation on the time index to obtain the statistical combined quantity sequence.

[0085] Key statistics are selectively identified from the full set of features, specifically the difference between the maximum and minimum values ​​of the temperature parameter. and the coefficient of variation of current parameters For example, from Extract and Indexing the statistics of these two specific dimensions at the same time. The pairing and aggregation process is performed to form binary groups. This extraction process is repeated for all mutation windows, and the generated binary sequences are re-aggregated and indexed in chronological order. Irrelevant intermediate calculation variables are removed, and only the core indicators used for final anomaly determination are retained to obtain the statistical combination quantity sequence.

[0086] S412: Call the combined quantity values ​​under the time index in the statistical combined quantity sequence, compare the difference between the maximum and minimum values ​​of the temperature parameter with the preset range threshold, and compare the coefficient of variation of the current parameter with three times the historical sliding average of the current window. Record the indexes that meet both judgment conditions as valid markers and generate an anomaly marker index set.

[0087] Set an absolute threshold for the difference between the maximum and minimum values ​​of the temperature parameter. This threshold is set based on the temperature rise response characteristics of the transformer under a step change in rated load, for example, a value of (That is, the temperature rise should not exceed this value within the short-term extended window), set the current parameter to the historical sliding average of the current window. for Calculate the coefficient of variation criteria and take the mean coefficient of variation during the stable period. (For example Three times the threshold, i.e. The current window's statistical value is logically ANDed with a threshold, and the judgment condition is: Substituting the data from the previous example, , (Not satisfied) , (Satisfied), since the two conditions are not satisfied simultaneously, this point is not marked as an anomaly. If at another time... , and , both conditions are met, the index is recorded as a valid marker point, after traversing the entire sequence, an abnormal marker index set is generated.

[0088] S413: Based on the time index recorded in the abnormal marker index set, retrieve the corresponding time field and serialize the indexes that meet the abnormal conditions, construct a time index set structure and output as a unique result unit, establish an abnormal time index set;

[0089] Retrieve each abnormal time point marked as valid, such as , , check the distribution of these indexes on the original time axis, for continuous abnormal indexes (such as consecutive points with an interval less than 1s), merge them into an abnormal event interval, record the start and end time of the interval, for isolated abnormal indexes, directly retain their time stamps, arrange all confirmed abnormal time points in ascending order, construct a time index set structure without duplicate elements, and output the set in a standardized list format as the input basis for subsequent trend analysis, establish an abnormal time index set.

[0090] Please refer to Figure 6 , the steps for obtaining the abnormal trend prediction result are:

[0091] S511: Call the abnormal time points recorded in the abnormal time index set, expand the equal interval time window before and after each time point, retrieve the continuous time period sample sequence of temperature parameters, current parameters, oil parameters and vibration parameters, and calculate the mean difference value of each type of parameter in turn, generate sliding mean difference information;

[0092] For example, for the abnormal point , expand 10 time windows (e.g. 60s for each window) before and after the point, retrieve the temperature, current, oil and vibration parameter data in this range, for each parameter sequence, calculate the mean difference of adjacent time windows, i.e. , for example, the mean of the temperature parameter in window is , and the mean of the temperature parameter in window is , then the mean difference is , calculate the mean jump of all parameters between each adjacent window in turn, generate sliding mean difference information containing the change rate of four types of parameters.

[0093] S512: According to the time sequence index in the sliding mean difference information, extract the mean difference of each parameter under the same abnormal index, according to the window sequence on the time axis, accumulate the number of abnormal triggers per unit time, use the formula:

[0094] ;

[0095] an abnormal risk trend value is obtained by operation,

[0096] wherein, represents an abnormal risk trend value in a time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a normalized variance sum of the sliding means of all parameters in the time window represents a normalized variance sum of the sliding means of all parameters in the time window represents a normalized variance sum of the sliding means of all parameters in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents a sliding mean normalized value of the mth monitoring parameter in the time window represents the total number of parameters participating in monitoring, which is fixed as 4, corresponding to temperature parameters, current parameters, oil parameters, and vibration parameters, and the risk trend threshold is a determination boundary value for determining whether to enter a warning interval, which can be determined according to the 90th percentile value of the value sequence calculated in the full sample time period, or set as the mean value of the sequence plus 1.5 times the standard deviation according to actual requirements, for identifying the aggregation of abnormal trends in the time dimension; intermediate variables required in the calculation formula, is a normalized value, and the calculation method is

[0097] wherein, and are the historical global mean and standard deviation of the mth parameter, respectively, is the number of times that the mth parameter triggers the S412 abnormal condition in the window is the number of times that the mth parameter triggers the S412 abnormal condition in the window is the number of times that the mth parameter triggers the S412 abnormal condition in the window is the number of times that the mth parameter triggers the S412 abnormal condition in the window is the number of times that the mth parameter triggers the S412 abnormal condition in the window is the number of times that the mth parameter triggers the S412 abnormal condition in the window

[0098] ;

[0099] Explanation of parameters in the formula: For time window The abnormal risk change trend value; Represents four monitoring parameters; Indicates the first The absolute change of the normalized mean of a parameter across adjacent windows is used to quantify the severity of the parameter trend. The frequency of abnormal triggers is used as a weighting factor; the higher the frequency, the greater the contribution of this parameter to the risk. For all parameters in the window The sum of normalized variances of the moving mean within the range, i.e. ( (The mean of the normalized values ​​of all parameters in this window). As a denominator term, it measures the dispersion of the whole; when the overall fluctuation is large ( When the denominator is large, the weight of a single parameter change is diluted; conversely, when the denominator is small, the change in a single parameter is more significant. Used to normalize the cumulative frequency to prevent excessively high frequencies from causing errors. Unlimited increase to ensure numerical convergence. Risk trend threshold setting: Calculated using a full sample of historical normal operation and typical failure periods. Sequence, count the number of the sequence. Percentile. Hypothetical history. The sequence mean is The standard deviation is The threshold can then be set to Example calculation: Selecting a certain time window The parameter data are shown in Table 1.

[0100] Table 1 Time Window Internal parameter statistics table:

[0101] ;

[0102] Refer to Table 1, first calculate :

[0103] The normalized mean of all parameters in the current window is Its average value ;

[0104] ;

[0105] ;

[0106] Calculate the denominator: ;

[0107] Calculate the sum of the terms in the numerator:

[0108] (temperature): ;

[0109] (Current): ;

[0110] (Oil products): ;

[0111] (vibration): ;

[0112] Total of numerators ;

[0113] Final calculation ;

[0114] This result indicates that the intensity of the multi-parameter joint anomaly evolution trend within the current time window is... This value comprehensively reflects the magnitude and frequency of parameter changes and is used in conjunction with the threshold. The risk level is determined by comparison, and the trend value of abnormal risk changes is obtained by calculation.

[0115] S513: Based on the abnormal risk change trend value, determine whether there is a situation where the abnormal risk change trend value exceeds the risk trend threshold within a continuous time period. If it is satisfied, mark the time period as a high-risk area, generate and output a risk warning signal, and obtain the abnormal trend prediction result.

[0116] For continuous time periods The value is monitored, and the previously set risk trend threshold is invoked. If continuous A time window (e.g.) )of The values ​​are respectively All exceeded the threshold. If the transformer is deemed to have entered a high-risk operating state, the corresponding continuous time period is marked as a high-risk area, and an early warning signal is immediately sent to the operation and maintenance management personnel via the SCADA system. The warning content includes the trigger time and the main abnormal parameters involved (based on...). By ranking contributions and determining risk trend values, abnormal trend prediction results are obtained.

[0117] A transformer operation anomaly prediction system is provided, which is used to execute the above-mentioned transformer operation anomaly prediction method. The system includes...

[0118] The correlation evaluation module obtains real-time operation parameters of the transformer, respectively constructs parameter pairs of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters, calculates correlation values of the parameter pairs in a current window, and obtains a parameter pair time correlation data set;

[0119] The mutation screening module calculates a first-order difference value in adjacent time windows based on the parameter pair time correlation data set, judges whether the first-order difference value exceeds a preset correlation fluctuation threshold, screens time nodes where a mutation behavior occurs and records corresponding parameter pairs, and obtains a mutation time point and parameter mapping table;

[0120] The mutation feature analysis module calls time node information in the mutation time point and parameter mapping table, extracts a maximum value and a minimum value difference of each parameter, a mean square error on a time gradient, and a coefficient of variation, combines each group of mutation corresponding feature combination sets under a mapping relationship, and obtains a parameter mutation window feature group.

[0121] The abnormal state recognition module judges whether the maximum value and the minimum value difference of the temperature parameter exceed a preset range threshold, and whether the coefficient of variation of the current parameter is greater than three times of a historical sliding mean value of the current window, and if so, marks the time point as an abnormal state, and obtains an abnormal time index set.

[0122] The abnormal trend analysis module calls time points that have been judged as abnormal in the abnormal time index set, extracts order values arranged in time windows, constructs a trend time sequence, accumulates frequencies according to the number of abnormalities in a unit time, calculates an abnormal risk trend in a target period, judges whether risk early warning needs to be implemented, and obtains an abnormal trend prediction result.

[0123] It should be understood that the term "and / or" herein merely describes an association relationship of associated objects, and represents that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone, wherein A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0124] In the present application, "at least one" means one or more, and "a plurality of" means two or more. "At least one of the following" or the like means any combination of these items, including any combination of single item or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, and c can be single or multiple.

[0125] It should be understood that the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0126] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0127] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0128] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0129] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0130] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0131] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0132] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A transformer abnormal operation prediction method characterized by comprising: The method comprises the following steps: S1: obtaining real-time operation parameters of the transformer, constructing parameter pairs of temperature parameters and current parameters, oil parameters and temperature parameters, and vibration parameters and current parameters respectively, calculating correlation values of the parameter pairs in a current window, and obtaining a parameter pair time correlation data set; S2: based on the parameter pair time correlation data set, calculating a first-order difference value in adjacent time windows, judging whether the first-order difference value exceeds a preset correlation fluctuation threshold, screening time nodes with mutation behaviors and recording corresponding parameter pairs to obtain a mutation time point and parameter mapping table; S3: calling time node information in the mutation time point and parameter mapping table, extracting the maximum value and minimum value difference of each parameter, the mean square error and the variation coefficient on the time gradient, combining each group of mutation corresponding feature combination set under the mapping relationship to obtain a parameter mutation window feature group; S4: according to the parameter mutation window feature group, judging whether the maximum value and minimum value difference of the temperature parameter exceeds a preset range threshold, and whether the variation coefficient of the current parameter is greater than three times of the historical sliding mean value of the current window, if yes, marking the time point as an abnormal state to obtain an abnormal time index set; S5: calling the time points determined to be abnormal in the abnormal time index set, extracting the order value arranged in time windows, constructing a trend time sequence, accumulating the frequency according to the number of abnormalities in a unit time, calculating the abnormal risk trend in the target period, judging whether risk warning needs to be implemented, and obtaining an abnormal trend prediction result.

2. The transformer abnormality prediction method according to claim 1, characterized by, The parameter pair time correlation data set comprises a correlation curve sequence, a time window label set and a parameter combination index table, the mutation time point and parameter mapping table comprises a mutation time set, a parameter pair mapping list and a window index label, the parameter mutation window feature group comprises a statistical feature matrix, a mutation parameter mapping unit and a sample time interval set, the abnormal time index set comprises an abnormal time point set, an abnormal identification label group and a parameter abnormal positioning cluster, and the abnormal trend prediction result comprises a time trend array, an abnormal accumulation frequency value and a risk window index table.

3. The transformer abnormality prediction method according to claim 2, characterized by, The parameter pair time correlation data set comprises a correlation curve sequence, a time window label set and a parameter combination index table, the mutation time point and parameter mapping table comprises a mutation time set, a parameter pair mapping list and a window index label, the parameter mutation window feature group comprises a statistical feature matrix, a mutation parameter mapping unit and a sample time interval set, the abnormal time index set comprises an abnormal time point set, an abnormal identification label group and a parameter abnormal positioning cluster, and the abnormal trend prediction result comprises a time trend array, an abnormal accumulation frequency value and a risk window index table. The parameter pair time correlation data set comprises a correlation curve sequence, a time window label set and a parameter combination index table, the mutation time point and parameter mapping table comprises a mutation time set, a parameter pair mapping list and a window index label, the parameter mutation window feature group comprises a statistical feature matrix, a mutation parameter mapping unit and a sample time interval set, the abnormal time index set comprises an abnormal time point set, an abnormal identification label group and a parameter abnormal positioning cluster, and the abnormal trend prediction result comprises a time trend array, an abnormal accumulation frequency value and a risk window index table. ​ S113: According to the time window index of each set of correlation values in the parameter combination correlation result set, sequentially window-sequencing arrangement is performed on each set of correlation values, a complete record is constructed by combining the corresponding parameter combination label, correlation value and timestamp field under each window, a continuity trend description structure is formed, and a parameter time sequence correlation data set is generated.

4. The transformer abnormality prediction method according to claim 3, characterized by, The acquisition step of the mutation time point and parameter mapping table is specifically: S211: Based on the continuous numerical sequence of each parameter pair in the parameter time sequence correlation data set, the sequence values of each parameter in the adjacent time window are sequentially extracted, the difference value calculation is performed on the numerical pairs of adjacent windows, the first-order difference sequence is constructed, and the difference processing is performed on all parameter pairs to obtain a correlation difference sequence set; S212: The sequence of each parameter pair in the correlation difference sequence set is called, the sliding standard deviation of the correlation value in the same window range is calculated and the statistical value is recorded, the offset judgment threshold is set to twice the sliding standard deviation, each difference value is compared with the offset judgment threshold, the time index of the difference value greater than the offset judgment threshold is marked, and an offset judgment marker set is obtained; S213: According to the window index judged as offset in the offset judgment marker set, the timestamp information of the corresponding parameter pair in the curve set is retrieved, the mapping record table is constructed by aggregating the timestamp field and the parameter pair label field, and the mapping record is sequentially arranged to obtain the mutation time point and parameter mapping table.

5. The transformer abnormality prediction method according to claim 4, characterized by, The acquisition step of the parameter mutation window feature group is specifically: S311: The mutation time index in the mutation time point and parameter mapping table is called, three data sampling intervals are respectively extended forward and backward for each time point, the continuous data sequence of the temperature parameter, the current parameter, the oil parameter and the vibration parameter in the corresponding period is retrieved, the complete numerical sequence of each parameter in the extended window is extracted, and an extended window parameter sequence set is obtained; S312: According to each parameter sequence in the extended window parameter sequence set, extreme value extraction is sequentially performed on each sequence, the difference value of the maximum value and the minimum value is calculated, the mean square error value in the time dimension is calculated, the standard deviation is obtained based on the current sequence mean value, a coefficient of variation set is constructed, and window statistical feature information is obtained; S313: Based on the three statistical feature values of each parameter group in the window statistical feature information, the feature combination is performed according to the mutation time and parameter pair label mapping rule, the parameter features under the same mutation time point are constructed into a unified structure data set and are sequentially numbered, and a parameter mutation window feature group is obtained.

6. The transformer abnormality prediction method according to claim 5, characterized by, The acquisition step of the abnormal time index set is specifically: S411: According to the parameter feature record in the parameter mutation window feature group, the maximum value and minimum value difference and the coefficient of variation of the temperature parameter and the current parameter are sequentially extracted, the two types of statistical quantities are aggregated to form a combined quantity sequence under the same time index, and the time index is arranged and collected to obtain a statistical combined quantity sequence; S412: Call the combination quantity value under the time index in the statistical combination quantity sequence, perform comparison on the temperature parameter maximum value and minimum value difference and the preset range threshold value, and perform comparison on the current window historical sliding mean value and three times of the current window historical sliding mean value, record the index that meets the two judgment conditions at the same time as an effective marker point, generate an abnormal marker index set, and output the abnormal marker index set. S413: Based on the time index recorded in the abnormal marker index set, retrieve the corresponding time field and perform sequence processing on the indexes that meet the abnormal condition, construct a time index set structure and output as a unique result unit, and establish an abnormal time index set.

7. The transformer abnormality prediction method according to claim 6, characterized by, The abnormal trend prediction result acquisition step specifically comprises: S511: Call the abnormal time points recorded in the abnormal time index set, expand an equal interval time window before and after each time point, retrieve the continuous time period sample sequence of the temperature parameter, the current parameter, the oil parameter and the vibration parameter, calculate the adjacent window mean difference value of each type of parameter in turn, and generate a sliding mean difference information; S512: According to the time sequence index in the sliding mean difference information, extract the mean difference of each parameter under the same abnormal index, according to the window sequence on the time axis, accumulate the number of abnormal triggers in each unit of time, and calculate and obtain an abnormal risk change trend value; S513: According to the abnormal risk change trend value, judge whether the abnormal risk change trend value exceeds the risk trend threshold value in the continuous time period, if yes, mark the time period as a high-risk area, generate and output a risk warning signal, and obtain an abnormal trend prediction result.

8. The transformer abnormality prediction method according to claim 7, characterized by, The formula for calculating the abnormal risk change trend value is specifically: ； wherein, represents a trend value of abnormal risk change within a time window, represents a sliding mean normalized value of the th monitoring parameter within a time window, represents a sliding mean normalized value of the th monitoring parameter within a time window, represents a frequency of abnormal triggering of the th monitoring parameter within a time window, represents a sum of normalized variances of sliding means of all parameters within a time window, represents a frequency of abnormal triggering of the th monitoring parameter within a time window, represents a total number of parameters participating in monitoring.​​​​​​ 9. A transformer abnormal operation prediction system characterized by comprising: The system is used to realize the transformer operation abnormality prediction method in any one of claims 1-8, and the system comprises: A correlation evaluation module acquires real-time operation parameters of a transformer, constructs parameter pairs of a temperature parameter and a current parameter, an oil parameter and a temperature parameter, and a vibration parameter and a current parameter respectively, calculates correlation values of the parameter pairs in a current window, and obtains a parameter pair time sequence correlation data group; A mutation screening module calculates a first-order difference value in adjacent time windows based on the parameter pair time sequence correlation data group, judges whether the first-order difference value exceeds a preset correlation fluctuation threshold, screens time nodes with mutation behaviors and records corresponding parameter pairs, and obtains a mutation time point and parameter mapping table; A mutation feature analysis module calls time node information in the mutation time point and parameter mapping table, extracts a maximum value and a minimum value difference, a mean square error on a time gradient, and a variation coefficient of each parameter, combines the features into a feature combination set corresponding to each mutation under a mapping relationship, and obtains a parameter mutation window feature group; An abnormal state recognition module judges whether the temperature parameter maximum value and the minimum value difference exceed a preset range threshold value and whether the current parameter variation coefficient is greater than three times of a current window historical sliding mean value according to the parameter mutation window feature group, and marks the time point as an abnormal state if the conditions are met, and obtains an abnormal time index set. The abnormal trend analysis module calls the time points determined to be abnormal in the abnormal time index set, extracts the order values arranged in time windows, constructs a trend time series, performs cumulative frequency processing according to the number of abnormalities in a unit time, calculates the abnormal risk trend in a target period, judges whether risk early warning needs to be implemented, and obtains an abnormal trend prediction result.

Images