Transformer data anomaly monitoring method and system based on data processing

Abstract

The present invention relates to the field of data processing technology, and more particularly to a data processing-based transformer data anomaly monitoring method and system. The method divides the state data of a target transformer into key state data and extended state data, determines a target state matrix based on the key state data, performs state prediction using the target state matrix, then performs anomaly classification and identification to obtain an anomaly probability vector, discriminates the anomaly probability vector, and determines whether to update the target state matrix in combination with the extended state data for recalculation based on the discriminant result of the anomaly probability vector. Through a staged processing approach, the method effectively improves monitoring efficiency and ensures monitoring accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a transformer data anomaly monitoring method and system based on data processing. Background Art

[0002] Transformers are core equipment in power system operations, and their operational stability is directly related to the reliable power supply of the grid. With the advancement of smart grid construction, transformers are often equipped with various monitoring devices, such as sensors and online monitoring devices, which can collect massive amounts of status data in real time, covering multiple dimensions such as electrical parameters, oil service data, and mechanical vibration.

[0003] However, the increase in information dimensions also means an increase in computational complexity. When existing technologies use prediction models to predict transformer data and make fault or anomaly judgments based on the predicted data, it is difficult to meet the needs of timely anomaly monitoring.

[0004] Therefore, how to improve the efficiency of transformer data anomaly monitoring while ensuring the accuracy of transformer data anomaly monitoring has become an urgent problem to be solved. Summary of the Invention

[0005] In order to solve the problem that the existing technology of transformer data anomaly monitoring requires multi-dimensional data processing, resulting in high computational complexity and thus low efficiency of transformer data anomaly monitoring, the present invention provides a transformer data anomaly monitoring method and system based on data processing.

[0006] In a first aspect, the present invention provides a transformer data anomaly monitoring method based on data processing, which adopts the following technical solutions:

[0007] A transformer data anomaly monitoring method based on data processing includes: obtaining key state data and extended state data of a target transformer at M preset time points, where M is a positive integer; determining a target state matrix based on the M key state data; inputting the target state matrix into a trained state prediction model to obtain predicted state data corresponding to the M+1th preset time point to the M+Nth preset time point, where N is a positive integer; inputting the M first state data and the N predicted state data into a trained abnormality classification model to obtain an abnormality probability vector formed by predicted abnormality probabilities corresponding to K preset abnormality categories, where K is a positive integer; inputting the abnormality probability vector into a trained discriminant model to obtain a discrimination result; if the discrimination result does not meet a first preset condition, determining transformer abnormality information based on the abnormality probability vector; if the discrimination result meets the first preset condition, updating the target state matrix based on the M key state data and the M extended state data, returning to the step of inputting the target state matrix into the trained state prediction model until a new abnormality probability vector is obtained, and determining the transformer abnormality information based on the new abnormality probability vector.

[0008] The beneficial effects are: dividing the state data of the target transformer into key state data and extended state data, determining the target state matrix based on the key state data, performing state prediction through the target state matrix, and then performing abnormal classification and identification to obtain an abnormal probability vector, discriminating the abnormal probability vector, and determining whether to update the target state matrix in combination with the extended state data for recalculation based on the abnormal probability vector discrimination result. Through the stage processing method, the monitoring efficiency can be effectively improved and the monitoring accuracy can be guaranteed.

[0009] Furthermore, the target transformer corresponds to a number of preset attributes, each of which has a unique corresponding priority level, and the preset attributes at least include voltage, current, power, winding DC point group, oil temperature, oil level, winding temperature, and dissolved gas concentration;

[0010] The key status data includes status values ​​corresponding to a plurality of preset attributes of the first priority level;

[0011] The extended status data includes status values ​​corresponding to a plurality of preset attributes of the second priority level to the Rth priority level, wherein R is a positive integer.

[0012] Furthermore, the process of determining the priority level corresponding to the preset attribute includes the following steps:

[0013] Obtain attribute sample data corresponding to each preset attribute;

[0014] The principal component analysis method is used to perform dimensionality reduction processing on the attribute sample data corresponding to each preset attribute to obtain S reference dimension data, where S is a positive integer;

[0015] For any attribute sample data, based on the cosine distances between the attribute sample data and each reference dimension data, determine the minimum value among each cosine distance as the association evaluation value, and determine the reference dimension data corresponding to the association evaluation value as the association dimension data of the attribute sample data;

[0016] Clustering is performed on the associated evaluation values ​​corresponding to each attribute sample data to obtain several cluster sets;

[0017] For any attribute sample data, the priority level of the attribute sample data corresponding to the preset attribute is determined according to the cluster set to which the attribute sample data belongs.

[0018] The beneficial effect is that: by performing dimensionality reduction processing through principal component analysis, reference dimension data is obtained, and cluster analysis is performed according to the cosine distance between the attribute sample data and the closest reference dimension data, so that the priority level of the preset attribute can be determined according to the characterization ability of the attribute sample data corresponding to the preset attribute, so that when using key status data for prediction to improve the efficiency of transformer data anomaly monitoring, the accuracy of transformer data anomaly monitoring can also be guaranteed in most scenarios.

[0019] Furthermore, for any attribute sample data, determining the priority level of the attribute sample data corresponding to the preset attribute according to the cluster set to which the attribute sample data belongs includes:

[0020] According to the cluster center values ​​corresponding to each cluster set, each cluster set is sorted in a preset order to obtain a cluster set sequence;

[0021] For any attribute sample data, if there is no attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence;

[0022] If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has not been determined, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence;

[0023] If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has been determined, then the position of the cluster set to which the attribute sample data belongs in the cluster set sequence is increased by one, and the result of the increase is used as the priority level of the preset attribute corresponding to the attribute sample data.

[0024] The beneficial effects are: by analyzing the cluster sets, more priority levels are divided for the preset attributes, which further improves the flexibility of transformer data anomaly monitoring. At the same time, the dimensions of key status data are further simplified according to the correlation between the preset attributes, thereby further improving the efficiency of transformer data anomaly monitoring with a smaller loss of transformer data anomaly monitoring accuracy.

[0025] Furthermore, the abnormal probability vector is input into the trained discrimination model to obtain a discrimination result, including:

[0026] Inputting the abnormal probability vector into the trained discriminant model to obtain the discrimination result and its corresponding discrimination probability;

[0027] Accordingly, if the discrimination result satisfies the first preset condition, the target state matrix is ​​updated according to the M key state data and the M extended state data, and the step of inputting the target state matrix into the trained state prediction model is returned to, including:

[0028] If the discrimination result satisfies the first preset condition, determining a reference level according to the discrimination probability corresponding to the discrimination result;

[0029] For any preset time point from the first preset time point to the Mth preset time point, determining a target state vector based on the state values ​​of the preset attributes corresponding to the priority levels less than or equal to the reference level in the extended state data at the preset time point and the key state data corresponding to the preset time point;

[0030] Traverse all preset time points from the first preset time point to the Mth preset time point, obtain the target state vectors corresponding to each preset time point, form the target state matrix by the target state vectors corresponding to each preset time point, and return to execute the step of inputting the target state matrix into the trained state prediction model.

[0031] Furthermore, determining the reference level according to the discrimination probability corresponding to the discrimination result includes:

[0032] If the discrimination probability corresponding to the discrimination result is greater than a preset probability threshold, determining the maximum priority level as the reference level;

[0033] Otherwise, a preset mapping function is used to map the discrimination probability, and the reference level is determined according to the mapping result.

[0034] Furthermore, the discrimination result is a probability normal type or a probability abnormal type;

[0035] The first preset condition is that the judgment result is the probabilistic abnormality type.

[0036] Furthermore, determining transformer abnormality information according to the abnormality probability vector includes:

[0037] Determining a reference probability threshold based on K preset abnormality probabilities in the abnormality probability vector;

[0038] The transformer abnormality information is formed by preset abnormality categories corresponding to preset abnormality probabilities greater than the reference probability threshold.

[0039] In a second aspect, the present invention provides a transformer data anomaly monitoring system based on data processing, which adopts the following technical solutions:

[0040] A transformer data anomaly monitoring system based on data processing includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the transformer data anomaly monitoring method based on data processing is implemented.

[0041] By adopting the above technical solution, the above-mentioned transformer data anomaly monitoring method based on data processing is generated into a computer program and stored in a memory to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0042] The present invention has the following technical effects:

[0043] The state data of the target transformer is divided into key state data and extended state data. The target state matrix is ​​determined according to the key state data. The state prediction is performed through the target state matrix, and then abnormal classification and identification are performed to obtain the abnormal probability vector. The abnormal probability vector is judged. According to the judgment result of the abnormal probability vector, it is determined whether to update the target state matrix in combination with the extended state data for recalculation. Through the stage processing method, the monitoring efficiency can be effectively improved and the monitoring accuracy can be guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding numbers are the same or corresponding parts.

[0045] Figure 1 This is a method flow chart of a transformer data anomaly monitoring method based on data processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0047] It should be understood that when the terms "first," "second," and the like are used in the claims, description, and drawings of the present invention, they are merely used to distinguish between different objects, rather than to describe a specific order. The terms "comprise" and "comprising" used in the description and claims of the present invention indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof.

[0048] The embodiment of the present invention discloses a transformer data anomaly monitoring method based on data processing, referring to Figure 1 , including steps S1 to S7:

[0049] S1: Obtain key status data and extended status data of the target transformer at M preset time points.

[0050] Wherein, M is a positive integer, and the time interval between adjacent preset time points may be a fixed value, such as 5 seconds, 10 seconds, etc. The target transformer may refer to a transformer for which data anomaly monitoring is required.

[0051] Specifically, the target transformer corresponds to several preset attributes, each of which has a unique corresponding priority level. The preset attributes include at least voltage, current, power, winding DC point group, oil temperature, oil level, winding temperature, and dissolved gas concentration.

[0052] The key status data includes status values ​​corresponding to a number of preset attributes of the first priority level;

[0053] The extended status data includes status values ​​corresponding to a plurality of preset attributes of the second priority level to the R priority level.

[0054] Wherein, R is a positive integer, R is less than or equal to the number of preset attributes, and the preset attributes may also include vibration signals, noise, etc. Power can be divided into reactive power and active power, and dissolved gas concentration can be divided into Gas concentration, Gas concentration, Gas concentration, etc.

[0055] The priority level can represent the criticality of subsequent prediction and anomaly classification of the corresponding preset attribute. The lower the priority level, the higher the priority of the corresponding preset attribute.

[0056] Specifically, the process of determining the priority level corresponding to the preset attribute includes the following steps:

[0057] Obtain attribute sample data corresponding to each preset attribute;

[0058] The principal component analysis method is used to perform dimensionality reduction processing on the attribute sample data corresponding to each preset attribute to obtain S reference dimension data, where S is a positive integer;

[0059] For any attribute sample data, based on the cosine distances between the attribute sample data and each reference dimension data, the minimum value among each cosine distance is determined as the associated evaluation value, and the reference dimension data corresponding to the associated evaluation value is determined as the associated dimension data of the attribute sample data;

[0060] Clustering is performed on the associated evaluation values ​​corresponding to each attribute sample data to obtain several cluster sets;

[0061] For any attribute sample data, the priority level of the attribute sample data corresponding to the preset attribute is determined according to the cluster set to which the attribute sample data belongs.

[0062] Among them, the principal component analysis method and the method of determining the number S of reference dimension data based on the threshold method, scree plot method, etc. when using the principal component analysis method are both existing technologies and will not be repeated here.

[0063] Cosine distance can be used to characterize the closeness between the corresponding attribute sample data and the corresponding reference dimension data in high-dimensional space.

[0064] The associated evaluation value may represent the degree of proximity between the corresponding attribute sample data and the closest reference dimension data in the high-dimensional space. The associated dimension data may refer to the reference dimension data closest to the corresponding attribute sample data in the high-dimensional space.

[0065] Clustering processing can use the DBSCAN clustering algorithm, without the implementer having to set the number of clusters, which would affect the clustering effect.

[0066] Specifically, for any attribute sample data, determining the priority level of the attribute sample data corresponding to the preset attribute according to the cluster set to which the attribute sample data belongs includes:

[0067] According to the cluster center values ​​corresponding to each cluster set, each cluster set is sorted in a preset order to obtain a cluster set sequence;

[0068] For any attribute sample data, if there is no attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence;

[0069] If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has not been determined, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence;

[0070] If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has been determined, then the position of the cluster set to which the attribute sample data belongs in the cluster set sequence is increased by one, and the result of the increase is used as the priority level of the preset attribute corresponding to the attribute sample data.

[0071] Among them, the cluster center value can be determined according to the mean of the associated evaluation values ​​corresponding to all attribute sample data in the corresponding cluster set, and the preset order is the order of cluster center values ​​from small to large. Accordingly, in the cluster set sequence, the cluster set containing attribute sample data with stronger representation ability is higher in the front.

[0072] When determining the priority level, when there is only one attribute sample data in the cluster set that uses a certain reference dimension data as associated dimension data, it can be determined that the attribute sample data can best represent the reference dimension data. When there are multiple attribute sample data in the cluster set that use a certain reference dimension data as associated dimension data, only one of them is selected to represent the reference dimension data, thereby avoiding setting all strongly correlated preset attributes as high priority, further streamlining the number of high-priority preset attributes, and thus improving the efficiency of subsequent model reasoning.

[0073] S2: Determine the target state matrix based on M key state data.

[0074] Here, the key state data is assumed to be P×1 size data, where P is the number of preset attributes of the first priority level, and the target state matrix can be Q×M size data, where Q is the number of all preset attributes.

[0075] Initialize all elements in the target state matrix to 0. For any key state data, expand it into Q×1-sized intermediate data. Each row of the intermediate data uniquely corresponds to a preset attribute. For any row in the key state data, assign the value of the element in that row to the element in the corresponding row of the intermediate data according to its corresponding preset attribute. Unassigned elements in the intermediate data are set to 0. Then, assign the corresponding column of the key state data in the target state matrix based on the key state data. By traversing all key state data, the target state matrix can be obtained.

[0076] S3: Input the target state matrix into the trained state prediction model to obtain the predicted state data corresponding to the M+1th preset time point to the M+Nth preset time point.

[0077] Wherein, N is a positive integer and can be set by the implementer. The state prediction model can adopt a time domain convolution model, a long short-term memory network model, a recurrent network model, etc.

[0078] In this embodiment, the time domain convolution model is used as an example for description. The target state matrix determined according to the key state data corresponding to the 1st preset time point to the Mth preset time point is input into the trained state prediction model to predict the predicted state data corresponding to the M+1th preset time point. Then, the target state matrix is ​​updated according to the key state data corresponding to the 2nd preset time point to the Mth preset time point and the predicted state data corresponding to the M+1th preset time point, and is re-input into the trained state prediction model to predict the predicted state data corresponding to the M+2th preset time point. By analogy, the predicted state data corresponding to the M+1th preset time point to the M+Nth preset time point can be obtained.

[0079] It should be noted that the size of the predicted state data is also Q×1, so when updating the target state matrix during the prediction process, it can be directly achieved through the splicing operation.

[0080] The time-domain convolution model can be trained using conventional training methods, which will not be described in detail here.

[0081] Since the target state matrix has only P rows of valid data, and P is usually much smaller than Q, the model reasoning efficiency of prediction based on the target state matrix is ​​high and can meet the timeliness of transformer data anomaly monitoring.

[0082] S4: Input the M first state data and the N predicted state data into the trained abnormality classification model to obtain an abnormality probability vector formed by the predicted abnormality probabilities corresponding to the K preset abnormality categories.

[0083] Here, K is a positive integer. The preset abnormality categories may include winding fault, core fault, insulation fault, tap changer fault, cooling system fault, etc. The abnormality classification model may include a convolutional layer and a fully connected layer. The input data size of the convolutional layer may be Q×(M+N), and the output size is 1×K.

[0084] The training process of the anomaly classification model can adopt the conventional training method based on category labeling and cross entropy loss function, which will not be described here.

[0085] In one embodiment, the implementer can set a sliding window of size Q×L to slide and extract M+NL input matrices from the matrix formed by M first state data and N predicted state data, and then input them into the abnormal classification model respectively. In this case, the input data size of the abnormal classification model is Q×L, and then the abnormal probability vector is determined by averaging, weighting, etc. according to the output probability vector corresponding to each input matrix. This method aims to reduce the number of parameters of the abnormal classification model so as to improve the training effect and efficiency of the abnormal classification model, as well as the inference efficiency of the abnormal classification model, but it will increase the number of inferences. The implementer can design the abnormal classification model according to actual conditions.

[0086] S5: Input the abnormal probability vector into the trained discriminant model to obtain the discrimination result.

[0087] Among them, the discriminant model can be implemented using a classification model. The implementer can obtain a sample probability vector based on the real data of the transformer through the above steps, and then compare the abnormal category indicated by the sample probability vector with the abnormal category of the transformer in the real scene, manually mark the discriminant result label, and then use the cross entropy loss function to train the discriminant model.

[0088] It should be noted that both the anomaly classification model and the discriminant model are classification models with fast inference speed and little impact on the efficiency of transformer data anomaly monitoring, so that the time consumption of transformer data anomaly monitoring can be within an acceptable range.

[0089] Specifically, the abnormal probability vector is input into the trained discriminant model to obtain the discrimination results, including:

[0090] Input the abnormal probability vector into the trained discriminant model to obtain the discrimination result and its corresponding discrimination probability;

[0091] Accordingly, if the judgment result satisfies the first preset condition, the target state matrix is ​​updated according to the M key state data and the M extended state data, and the step of inputting the target state matrix into the trained state prediction model is returned to be executed, including:

[0092] If the discrimination result meets the first preset condition, the reference level is determined according to the discrimination probability corresponding to the discrimination result;

[0093] For any preset time point from the first preset time point to the Mth preset time point, determine the target state vector based on the state values ​​of the preset attributes corresponding to the priority levels less than or equal to the reference level in the extended state data at the preset time point and the key state data corresponding to the preset time point;

[0094] Traverse all preset time points from the 1st preset time point to the Mth preset time point, obtain the target state vectors corresponding to each preset time point, form a target state matrix by the target state vectors corresponding to each preset time point, and return to execute the step of inputting the target state matrix into the trained state prediction model.

[0095] Among them, the first preset condition can be used to judge whether the abnormal probability vector conforms to the norm. If the judgment result meets the first preset condition, it indicates that the abnormal probability vector does not conform to the norm, and the target state matrix needs to be updated according to the extended state data and key state data. The specific update method still adopts the assignment method, which will not be repeated here.

[0096] In order to ensure the efficiency of data anomaly monitoring, the reference level is determined according to the discrimination probability, and the state values ​​of the preset attributes corresponding to the priority levels in the extended state data that are less than or equal to the reference level are assigned. In this way, while ensuring the accuracy of data anomaly monitoring, the inference and calculation pressure of the state prediction model is reduced as much as possible to improve the efficiency of data anomaly monitoring.

[0097] Specifically, the reference level is determined based on the discrimination probability corresponding to the discrimination result, including:

[0098] If the discrimination probability corresponding to the discrimination result is greater than the preset probability threshold, the maximum priority level is determined as the reference level;

[0099] Otherwise, a preset mapping function is used to map the discrimination probability, and the reference level is determined according to the mapping result.

[0100] Among them, when the discrimination probability corresponding to the discrimination result is greater than the preset probability threshold, it means that the abnormal probability vector is seriously unreasonable and complete status data is needed to support accurate judgment.

[0101] The preset probability threshold can be set to 0.8, and the preset mapping function can be y=(2a-4)×x / (2b-1)+(4b-a-4) / (2b-1), where x is the discrimination probability, y is the mapping result, a is the preset probability threshold, and b is the number of priority levels.

[0102] After obtaining the mapping result, the mapping result is rounded up and the rounded result is used as the reference level.

[0103] S6: If the judgment result does not meet the first preset condition, the transformer abnormality information is determined according to the abnormality probability vector.

[0104] The abnormal information may include the target abnormal category of the target transformer, and the abnormal information may be sent to the manager to serve as an early warning and an instruction for verification.

[0105] Specifically, the discrimination result is a probability normal type or a probability abnormal type;

[0106] The first preset condition is that the judgment result is a probabilistic abnormal type.

[0107] Optionally, determining transformer abnormality information according to the abnormality probability vector includes:

[0108] Determine a reference probability threshold based on K preset abnormality probabilities in the abnormality probability vector;

[0109] The transformer abnormality information is formed by the preset abnormality categories corresponding to the preset abnormality probabilities greater than the reference probability threshold.

[0110] Among them, the method for determining the reference probability threshold can adopt threshold determination methods such as the mean method and the Otsu threshold method, which will not be repeated here.

[0111] The preset abnormality categories corresponding to the preset abnormality probabilities greater than the reference probability threshold are all target abnormality categories to form transformer abnormality information. It can be seen that this embodiment can output multiple target abnormality categories at the same time, which is more convenient for managers to perform abnormality analysis and tracing, and improve the completeness of data abnormality monitoring.

[0112] S7: If the judgment result meets the first preset condition, the target state matrix is ​​updated according to the M key state data and the M extended state data, and the step of inputting the target state matrix into the trained state prediction model is returned to execute until a new abnormal probability vector is obtained, and the transformer abnormal information is determined according to the new abnormal probability vector.

[0113] An embodiment of the present invention further discloses a transformer data anomaly monitoring system based on data processing, comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the transformer data anomaly monitoring method based on data processing according to the present invention is implemented.

[0114] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0115] In the present invention, the aforementioned memory can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium can be part of, accessible to, or connectable to the device.

[0116] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

[0117] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A transformer data anomaly monitoring method based on data processing, characterized in that: include: At M preset time points, key status data and extended status data of the target transformer are respectively obtained, where M is a positive integer, the target transformer corresponds to a plurality of preset attributes, the preset attributes have a unique corresponding priority level, and the extended status data includes status values ​​corresponding to a plurality of preset attributes of the second priority level to the Rth priority level, where R is a positive integer; Determine the target state matrix based on M key state data; Input the target state matrix into the trained state prediction model to obtain the predicted state data corresponding to the M+1th preset time point to the M+Nth preset time point, where N is a positive integer; Inputting M first state data and N predicted state data into the trained abnormality classification model, obtaining an abnormality probability vector formed by the predicted abnormality probabilities corresponding to K preset abnormality categories, where K is a positive integer; Inputting the abnormal probability vector into the trained discriminant model to obtain a discrimination result, wherein inputting the abnormal probability vector into the trained discriminant model to obtain a discrimination result includes: Inputting the abnormal probability vector into the trained discriminant model to obtain the discrimination result and its corresponding discrimination probability; If the discrimination result does not meet the first preset condition, determining the transformer abnormality information according to the abnormality probability vector; If the discrimination result satisfies the first preset condition, the target state matrix is ​​updated according to the M key state data and the M extended state data, and the step of inputting the target state matrix into the trained state prediction model is returned to be executed until a new abnormality probability vector is obtained, and the transformer abnormality information is determined according to the new abnormality probability vector. If the discrimination result satisfies the first preset condition, the target state matrix is ​​updated according to the M key state data and the M extended state data, and the step of inputting the target state matrix into the trained state prediction model is returned to be executed, including: If the discrimination result satisfies the first preset condition, determining a reference level according to the discrimination probability corresponding to the discrimination result; For any preset time point from the first preset time point to the Mth preset time point, determining a target state vector based on the state values ​​of the preset attributes corresponding to the priority levels less than or equal to the reference level in the extended state data at the preset time point and the key state data corresponding to the preset time point; Traverse all preset time points from the first preset time point to the Mth preset time point, obtain the target state vectors corresponding to each preset time point, form the target state matrix by the target state vectors corresponding to each preset time point, and return to execute the step of inputting the target state matrix into the trained state prediction model.

2. The transformer data anomaly monitoring method based on data processing according to claim 1 is characterized in that: The preset attributes include at least voltage, current, power, winding DC point group, oil temperature, oil level, winding temperature, and dissolved gas concentration; The key status data includes status values ​​corresponding to a plurality of preset attributes of the first priority level.

3. The transformer data anomaly monitoring method based on data processing according to claim 2 is characterized in that: The process of determining the priority level corresponding to the preset attribute includes the following steps: Obtain attribute sample data corresponding to each preset attribute; The principal component analysis method is used to perform dimensionality reduction processing on the attribute sample data corresponding to each preset attribute to obtain S reference dimension data, where S is a positive integer; For any attribute sample data, based on the cosine distances between the attribute sample data and each reference dimension data, determine the minimum value among each cosine distance as the association evaluation value, and determine the reference dimension data corresponding to the association evaluation value as the association dimension data of the attribute sample data; Clustering is performed on the associated evaluation values ​​corresponding to each attribute sample data to obtain several cluster sets; For any attribute sample data, the priority level of the attribute sample data corresponding to the preset attribute is determined according to the cluster set to which the attribute sample data belongs.

4. The transformer data anomaly monitoring method based on data processing according to claim 3 is characterized in that: The step of determining, for any attribute sample data, the priority level of the attribute sample data corresponding to the preset attribute according to the cluster set to which the attribute sample data belongs includes: According to the cluster center values ​​corresponding to each cluster set, each cluster set is sorted in a preset order to obtain a cluster set sequence; For any attribute sample data, if there is no attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence; If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has not been determined, then the priority level of the preset attribute corresponding to the attribute sample data is determined according to the position of the cluster set to which the attribute sample data belongs in the cluster set sequence; If there is another attribute sample data in the cluster set to which the attribute sample data belongs that has the same associated dimension data as the attribute sample data, and the priority level of the preset attribute corresponding to the other attribute sample data has been determined, then the position of the cluster set to which the attribute sample data belongs in the cluster set sequence is increased by one, and the result of the increase is used as the priority level of the preset attribute corresponding to the attribute sample data.

5. The transformer data anomaly monitoring method based on data processing according to claim 1 is characterized in that: The determining of the reference level according to the discrimination probability corresponding to the discrimination result includes: If the discrimination probability corresponding to the discrimination result is greater than a preset probability threshold, determining the maximum priority level as the reference level; Otherwise, a preset mapping function is used to map the discrimination probability, and the reference level is determined according to the mapping result.

6. The transformer data anomaly monitoring method based on data processing according to claim 1 is characterized in that: The discrimination result is a probability normal type or a probability abnormal type; The first preset condition is that the judgment result is the probabilistic abnormality type.

7. The transformer data anomaly monitoring method based on data processing according to claim 1 is characterized in that: The determining of transformer abnormality information according to the abnormality probability vector includes: Determining a reference probability threshold based on K preset abnormality probabilities in the abnormality probability vector; The transformer abnormality information is formed by preset abnormality categories corresponding to preset abnormality probabilities greater than the reference probability threshold.

8. A transformer data anomaly monitoring system based on data processing, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the transformer data anomaly monitoring method based on data processing according to any one of claims 1 to 7 is implemented.

Images