Transformer state prediction method fusing vibration and temperature data

Abstract

The invention discloses a transformer state prediction method fusing vibration and temperature data, and relates to the technical field of data processing. The method comprises the following steps: interacting fault operation and maintenance records of a transformer, and mining a fault associated network; arranging a front-end sensing array, carrying out operation monitoring sampling on a transformer operation scene, and determining sensing sampling data; a state prediction model is supervised and trained by taking a fault association network as a criterion and subjective and objective weighting based on fault features as a criterion, prediction is carried out based on a time sequence data chain, and the time sequence data chain is updated through average weakening processing of data nodes; sending back the sensing sampling data, and combining the state prediction model to update a time sequence data chain and predict the state of the transformer to determine the state of the transformer; and carrying out operation, maintenance, debugging and control on the transformer by taking the transformer state as a reference. The technical problem of low operation and maintenance efficiency caused by inaccurate transformer state prediction in the prior art is solved, and the technical effect of improving the transformer state prediction precision is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a transformer state prediction method that fuses vibration and temperature data. Background Art

[0002] In the power system, as a key device, the operating state of the transformer is directly related to the safety and reliability of the entire system. However, during long-term operation, the transformer may be affected by various complex factors, including vibration, temperature changes, electrical overload, etc. These factors can easily lead to equipment failures and even major accidents. Existing transformer state monitoring and prediction methods mainly rely on single data sources or simple monitoring means, and cannot comprehensively reflect the actual operating state of the transformer, resulting in lagging fault diagnosis and insufficient prediction accuracy. In addition, traditional methods lack real-time and intelligent support in operation and maintenance management, and it is difficult to meet the requirements of modern power systems for efficient and accurate equipment management. Therefore, there is an urgent need for a transformer state prediction method that fuses multi-source data and has high prediction accuracy and real-time performance to improve equipment management efficiency and system operation safety. Summary of the Invention

[0003] The present application provides a transformer state prediction method that fuses vibration and temperature data, and solves the technical problem of low operation and maintenance efficiency caused by inaccurate transformer state prediction in the prior art.

[0004] In view of the above problems, the present application provides a transformer state prediction method that fuses vibration and temperature data.

[0005] The present application provides a transformer state prediction method that fuses vibration and temperature data, and the method includes: Interact with the fault operation and maintenance records of the transformer to mine the fault correlation network, where the fault correlation network is formed based on fault type - fault characteristics; deploy the front-end sensing array, perform operation monitoring and sampling for the transformer operation scenario to determine the sensing sampling data, including periodic sampling and fluctuating sampling by introducing a frequency conversion monitor; use the fault correlation network as a criterion and the subjective and objective weighting based on fault characteristics as a principle to supervise and train the state prediction model, where the prediction is based on the time series data chain, and the time series data chain is updated by the average weakening processing of data nodes; transmit back the sensing sampling data, combine it with the state prediction model, perform time series data chain update and transformer state prediction to determine the transformer state; based on the transformer state, perform transformer operation and maintenance control and management.

[0006] One or more technical solutions provided in the present application have at least the following technical effects or advantages: First, for the fault operation and maintenance records of the interactive transformer, a fault correlation network is mined, where the fault correlation network is formed based on the fault type - fault characteristics. Next, a front-end sensing array is deployed to perform operation monitoring and sampling for the transformer operation scenario to determine the sensing sampling data, including periodic sampling and fluctuation sampling by introducing a frequency conversion monitor. Then, taking the fault correlation network as the criterion and the subjective and objective weighting based on fault characteristics as the principle, a state prediction model is supervised and trained, where the prediction is performed based on the time series data chain, and the time series data chain is updated by the average weakening process of data nodes. Next, the sensing sampling data is transmitted back, combined with the state prediction model, to update the time series data chain and predict the transformer state, and the transformer state is determined. Finally, based on the transformer state, the operation and maintenance control of the transformer is carried out. This solves the technical problem of low operation and maintenance efficiency caused by inaccurate transformer state prediction in the prior art, and achieves the technical effect of improving the accuracy of transformer state prediction. Description of the Drawings

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0008] Figure 1 Schematic flow chart of the transformer state prediction method integrating vibration and temperature data provided by the embodiment of the present application; Figure 2 Schematic flow chart of updating the time series data chain and predicting the transformer state in the transformer state prediction method integrating vibration and temperature data provided by the embodiment of the present application. Detailed Embodiments

[0009] The present application provides a transformer state prediction method integrating vibration and temperature data, which solves the technical problem of low operation and maintenance efficiency caused by inaccurate transformer state prediction in the prior art.

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application.

[0011] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products, or devices.

[0012] Examples, such as Figure 1 As shown, the embodiments of the present application provide a transformer status prediction method that fuses vibration and temperature data. Among them, the method includes: Interact with the fault operation and maintenance records of the transformer to mine the fault correlation network, where the fault correlation network is formed based on the fault type - fault characteristics.

[0013] By interacting with the transformer to obtain the fault operation and maintenance records, these records include various fault types (such as insulation failure, winding short circuit, oil temperature too high, etc.) that occurred during the operation of the transformer and the corresponding operation and maintenance activities (such as maintenance, component replacement, operation parameter adjustment, etc.); based on the extracted fault operation and maintenance records, data mining technology is used to construct a fault correlation network, where the fault type is used as the network node, and the corresponding fault characteristics and the association relationship between the characteristics are used as the edges, thus generating a fault correlation network. The fault correlation network is established based on the relationship between the fault type (such as short circuit, overheat, etc.) and the fault characteristics (such as temperature rise, abnormal vibration amplitude, etc.), which is convenient for systematic analysis of the causes and characteristics of faults. For example, the short circuit fault node may be connected to characteristics such as abnormal current and increased winding vibration, while the winding aging fault node may be associated with characteristics such as high temperature and vibration frequency fluctuation. By constructing the fault correlation network, not only can the characteristic distribution law between different fault types be clearly presented, but also it provides direct support for the subsequent training of the prediction model based on the characteristic data.

[0014] Furthermore, mining the fault correlation network includes: Traverse the fault operation and maintenance records, perform clustering processing based on the fault type, and determine multiple clustering clusters; identify the first clustering cluster within the multiple clustering clusters, and determine multiple groups of fault characteristic intervals, where the first clustering cluster corresponds to the first fault type, and the maximum and minimum values of the same fault characteristics within the cluster are used to determine the fault characteristic intervals; traverse the multiple groups of fault characteristic intervals, and determine the characteristic critical values with the risk boundary; based on the first fault type and the characteristic critical values, determine the first correlation network.

[0015] Specifically, traverse the historical fault operation and maintenance records of the transformer, extract the fault types and corresponding characteristic data (such as temperature, vibration amplitude, oil pressure, etc.) in the records, and input these data into clustering algorithms (such as K-means, DBSCAN, or hierarchical clustering) for clustering processing; through clustering analysis, divide the records into multiple clustering clusters according to the fault types, and each clustering cluster corresponds to a type of fault and its related characteristic set; subsequently, among the generated multiple clustering clusters, select the target clustering cluster, that is, the first clustering cluster, which corresponds to the first fault type, such as "overheating fault" or "abnormal winding vibration fault"; for the first clustering cluster, extract all the fault characteristic data included in the cluster (such as temperature and vibration amplitude), calculate the maximum and minimum values of each characteristic, and then determine the characteristic interval. For example, the temperature characteristic interval is [80°C, 100°C], and the vibration amplitude characteristic interval is [1.5g, 2.0g]; then, traverse all the determined fault characteristic intervals, and based on statistical analysis (such as confidence interval calculation) or empirical rules, determine the risk boundary of each characteristic and calculate the critical value of the characteristic. If it is an extremely large characteristic, that is, the greater the value, the greater the risk, take the minimum boundary of the interval as the critical value; if it is an extremely small characteristic, take the maximum boundary of the interval as the critical value. For example, when the temperature exceeds 95°C or the vibration amplitude exceeds 1.8g, it is regarded as entering the high-risk interval, and these values are defined as the characteristic critical values. On this basis, establish a mapping relationship between the first fault type and its corresponding characteristic critical values to generate the first association network. The nodes of this association network include the first fault type and related characteristic nodes (such as "excessively high temperature" or "abnormal vibration"), and the edges represent the association between the fault type and the characteristic. Through this network, the internal connection between a specific fault type and its key characteristics can be visually presented, providing an accurate characteristic basis and judgment criterion for the subsequent training of the fault prediction model, and laying a foundation for intelligent fault diagnosis and operation and maintenance regulation.

[0016] Deploy a front-end sensing array to conduct operation monitoring and sampling for the transformer operation scenario and determine the sensing sampling data, including periodic sampling and fluctuation sampling by introducing a frequency conversion monitor.

[0017] Deploy a front-end sensing array in the transformer operation scenario to conduct real-time monitoring and data collection on the transformer operation state. The front-end sensing array includes various types of sensors, such as vibration sensors, temperature sensors, oil pressure sensors, and current sensors, etc., to comprehensively sense the operation environment and state parameters of the transformer.

[0018] In the monitoring sampling process, the sampling data is divided into two categories: periodic sampling and fluctuation sampling. Periodic sampling refers to the collection of sensor data at fixed time intervals, such as recording parameters such as temperature, vibration amplitude and oil pressure every 1 minute. This method can ensure the continuity of data and provide stable basic data for subsequent analysis. Fluctuation sampling, on the other hand, introduces a variable frequency monitor to perform real-time high-frequency sampling when abnormal fluctuations occur in the operating state. For example, when the vibration sensor detects an abnormal vibration amplitude or the temperature sensor detects a rapid heating trend, the variable frequency monitor will automatically increase the sampling frequency to capture high-precision data during the abnormal period. This dynamic sampling method can effectively make up for the problem of insufficient time resolution that may occur in periodic sampling under abnormal conditions. Through the combination of periodic sampling and fluctuation sampling, the front-end sensor array can generate comprehensive and accurate sensor sampling data, which not only provides rich feature information for the input of the transformer state prediction model, but also lays a solid foundation for fault warning and operation optimization decision-making.

[0019] Furthermore, a front-end sensor array is deployed to perform operation monitoring sampling for transformer operation scenarios, including: A variable frequency monitor is introduced, and a communication connection is established between the variable frequency monitor and the front-end sensor array; a preset sampling period is set as a first sampling rule; based on the fault operation and maintenance record, a characteristic fluctuation threshold based on operation risk is mined, and a second sampling rule is set based on the characteristic fluctuation threshold; the front-end sensor array is configured based on the first sampling rule, and the variable frequency monitor is configured based on the second sampling rule; wherein, when the characteristic fluctuation threshold is met, the variable frequency monitor generates a sensor sampling instruction.

[0020] Preferably, a frequency conversion monitor is deployed in the front-end sensing array and communicates with various sensors (such as temperature sensors, vibration sensors, oil pressure sensors, etc.) through wired or wireless communication methods. The frequency conversion monitor serves as the control center and is used to adjust the sampling frequency of the sensors in real time to meet the monitoring requirements under different operating conditions. A basic sampling period is set, for example, collecting sensor data once per minute, as the first sampling rule, which is used to achieve long-term trend monitoring and basic data accumulation of the transformer under normal operating conditions. Subsequently, by analyzing historical fault operation and maintenance records, the key features that may cause faults (such as rapid temperature increase or abnormal vibration amplitude fluctuation) and their fluctuation rules are mined, and the characteristic fluctuation thresholds are extracted, such as the temperature change rate exceeding 5°C / minute or the vibration amplitude fluctuation exceeding 0.2g. These characteristic thresholds are set as the second sampling rule, which is specifically used to identify abnormal operating states. The front-end sensing array is configured based on the first sampling rule (fixed sampling period) to ensure continuous data acquisition under normal operating conditions. At the same time, the frequency conversion monitor is configured according to the second sampling rule (based on characteristic fluctuation thresholds) to enable it to dynamically adjust the sampling frequency. For example, when a certain characteristic exceeds the set threshold, the frequency conversion monitor can immediately issue a high-frequency sampling instruction. During operation, the sensors will monitor the changes of various characteristics in real time. When a certain characteristic value reaches or exceeds the characteristic fluctuation threshold, the frequency conversion monitor will generate a sensing sampling instruction to switch the sampling frequency from the regular sampling to the high-frequency sampling mode. For example, when the temperature sensor detects that the temperature is rising at a rate exceeding 5°C / minute, the frequency conversion monitor immediately increases the sampling frequency to once per second to capture the rapidly changing data. In this way, the front-end sensing array can complete basic monitoring under normal conditions, quickly respond under abnormal conditions, and provide high-resolution data support, providing a comprehensive and reliable data basis for the accurate prediction and fault warning of the transformer operating state.

[0021] Taking the fault correlation network as the criterion and the subjective and objective weighting based on fault characteristics as the principle, supervise and train the state prediction model, where the prediction is based on the time series data chain, and the time series data chain is updated by the average weakening processing of data nodes.

[0022] Taking the fault correlation network as a criterion, using the relationship between the fault types and key features described therein as the training basis of the state prediction model, the fault correlation network contains nodes (such as fault types and features) and the edges between them (representing the correlation between features or between features and fault types), which are used to constrain the feature selection and learning direction of the model input. Then, for the fault feature data, the importance of each feature is weighted by integrating subjective and objective weighting criteria; among them, subjective weighting is based on expert experience. For example, a higher weight is given to the vibration feature because its correlation with the winding aging fault is stronger; objective weighting calculates the weight through data analysis techniques such as information gain or entropy value method according to the influence degree of the feature on fault prediction; after combining the two, a comprehensive weight is generated to guide the model to learn the priorities of different features.

[0023] In the model training stage, a supervised learning method is adopted, such as the long short-term memory network (LSTM) or the temporal convolutional network (TCN), using historical fault records and status data as the training data set to supervise the training of the state prediction model; the model will learn the temporal variation rules of different fault features and establish the dynamic prediction ability of the transformer state; the data input into the model is based on the temporal data chain, which is composed of continuous data collected by the sensing array and contains information on the variation of multiple features over time. During the training process, the model gradually optimizes the parameters according to the feature weights and their performance in the temporal data chain to improve the prediction accuracy.

[0024] To maintain the sensitivity of the model to real-time data, an average weakening mechanism for data nodes is established. Specifically, the weight of historical data nodes in the temporal data chain is processed with a decreasing weight, and the weight of the latest data is preferentially increased. The weakening formula adopts a time-decreasing method. For example, according to W t =W t-1 ·α, the weight is gradually reduced, where W t represents the weight of the data node at the current time point t, W t-1 represents the weight of the data node at the previous time point t-1, and α is the weakening coefficient, whose value range is greater than zero and less than 1, representing the proportion of weight reduction. The closer the value is to 1, the slower the weight weakens, and the smaller the value, the faster the weight weakens. Finally, based on the above fault correlation network, feature weights, temporal data chain, and average weakening mechanism, the training and deployment of the state prediction model are completed. The model can predict the operation state of the transformer in real time, generate fault warning signals, and provide accurate data support and scientific basis for subsequent operation and maintenance decisions.

[0025] Transmit the sensed sampling data back, and in combination with the state prediction model, update the temporal data chain and predict the transformer state to determine the transformer state.

[0026] During the operation of the transformer, the sensor data collected in real time (such as temperature, vibration amplitude, oil pressure, etc.) is transmitted back to the data processing center through the sensor array. These data include the basic data collected according to the preset sampling period and the fluctuating data triggered by the variable frequency monitor, forming a complete sampling information set. The transmitted data is integrated into new data nodes and added to the existing time series data chain. Then, the updated time series data chain is input into the trained state prediction model. The model analyzes the characteristic changes of the new data nodes, combines the characteristic relationships and weight information in the fault correlation network, and makes a real-time prediction of the operating state of the transformer, outputting the current operating state of the transformer, such as normal operation, warning state, fault state, etc.

[0027] Furthermore, as Figure 2 shown, the update of the time series data chain and the transformer state prediction include: Set the preset number of nodes, perform time series integration on the transformer operation data, and determine the time series data chain, where the preset number of nodes is N; as the sensing sampling data is transmitted back, update the time series data chain in the state prediction model to determine the updated data chain; perform transformer state trend prediction and fault risk determination on the updated data chain to determine the predicted state of the transformer.

[0028] Specifically, set the preset number of nodes of the time series data chain to N. For example, N = 100, indicating that the time series data chain retains at most the last 100 data nodes; integrate the historical operation data of the transformer in chronological order to form a time series data chain, where each data node contains complete operation characteristic information (such as temperature, vibration amplitude, oil pressure, current, etc.), ensuring the time continuity and integrity of the time series chain. During operation, the sensing sampling data collected by the sensing array in real time is transmitted back to the data processing center and added to the time series data chain as a new node; when the new node makes the length of the data chain exceed the preset N, remove the node with the earliest time (i.e., the oldest data) to keep the data chain length at N. Input the updated time series data chain into the state prediction model, and the model analyzes the change trend of the characteristics in the chain over time to predict the operating state of the transformer. The model outputs the state trend prediction result based on the change law of historical data and the fluctuation trend of current data. For example, when all the characteristic values in the data chain are within the normal range and change smoothly, the model determines that the transformer state is normal operation; when some characteristic values are close to the critical value or show abnormal fluctuations, the model determines the state as the warning state; if some characteristic values exceed the critical value range, the state is determined as the fault state.

[0029] Furthermore, the update of the time series data chain in the state prediction model includes: Generate new data nodes based on the sensed sampling data; traverse the time-series data chain, and determine N-1 groups of buffer operators through the average weakening process of the data; integrate the N-1 groups of buffer operators and the new data nodes for time series to serve as the updated data chain.

[0030] Specifically, integrate the real-time sensed sampling data into new data nodes, which contain the transformer operation characteristic data at the current time point, such as parameters like temperature, vibration amplitude, oil pressure, current, etc.; traverse all N-1 data nodes of the current time-series data chain, perform average weakening processing on the characteristic data of each node to generate N-1 groups of buffer operators, and each group of buffer operators contains the weakened characteristic values, which are used to reflect the decreasing trend of the importance of historical data; integrate the N-1 groups of buffer operators and the new data nodes in the order of time series to generate an updated data chain. Among them, the new data node is inserted at the end of the time-series data chain as the data at the latest time point, replacing the earliest removed node, and keeping the number of nodes in the data chain always N.

[0031] Furthermore, determine N-1 groups of buffer operators through the average weakening process of the data, including: Traverse the time-series data chain, and determine the first node data and the second node data in the time-series growth direction; perform mapping and mean calculation on the first node data and the second node data to determine the first buffer operator, where the first buffer operator is the first link node of the updated data chain; determine the second node data and the third node data and perform mapping mean calculation to determine the second buffer operator; until the calculation of the (N-1)th buffer operator is completed, integrate the first buffer operator, the second buffer operator until the (N-1)th buffer operator to determine the N-1 groups of buffer operators.

[0032] Starting from the first node of the time-series data chain, select two adjacent nodes in sequence as a pair of nodes in time order. For example, the first pair of nodes is the first node (D1) and the second node (D2), the second pair of nodes is the second node (D2) and the third node (D3), and so on, until the last pair of nodes is the (N-1)th node (D N-1 ) and the Nth node (D N). Map and calculate the mean of the feature data of the first pair of nodes; assuming that each node contains multi-dimensional feature data (such as temperature, vibration amplitude, oil pressure, etc.), then pair the corresponding feature values of the two nodes one by one and calculate their mean. For example, the mean of the temperature is the sum of the temperature values of the two nodes divided by two, and the means of the vibration amplitude and oil pressure are calculated in the same way. The calculation result is used as the first buffer operator, representing the first buffer node in the updated data chain. Process other node pairs in the time series data chain in the same way. For example, the second pair of nodes consists of D2 and D3, map and calculate the mean of their feature values to obtain the second buffer operator; repeat this process to calculate the means of all node pairs in turn until the last pair of nodes is calculated to obtain the (N - 1)-th buffer operator. Arrange all the calculated buffer operators in chronological order to form a set containing (N - 1) buffer operators. This set represents the historical data after average weakening processing, retains the change information between adjacent nodes in the time series data chain, and smooths the fluctuation trend. Replace the first N - 1 nodes in the time series data chain with the calculated (N - 1) buffer operators, and add the newly added data node (the latest sensor data collected in real time) to the end of the data chain to form a complete updated data chain.

[0033] Furthermore, after determining the updated data chain, it includes: For the sensing sampling data, perform principal component analysis screening based on fault characteristics to determine fault-related characteristics; identify the fault-related characteristics and subjectively assign weights to determine the first data weight; identify the fault-related characteristics and objectively assign weights to determine the second data weight; fit the first data weight and the second data weight to determine the data weight distribution and label the updated data chain.

[0034] Extract the feature data of all nodes from the updated time series data chain, including multi-dimensional data such as temperature, vibration amplitude, and oil pressure; arrange the feature values of all nodes in chronological order to form an N×M matrix, where N is the number of data chain nodes and M is the feature dimension; perform principal component analysis on the feature matrix, calculate the variance contribution rate of the feature values, and arrange them in descending order of contribution rate. Select the features with a cumulative contribution rate of more than 95% as fault-related characteristics. For example, the PCA analysis result shows that temperature and vibration amplitude are the main features, and the contribution of oil pressure is low and is excluded. Based on expert experience and industry knowledge, subjectively assign weights to the fault-related characteristics to generate the first data weight; calculate the weights of each feature based on the statistical characteristics of the sensing sampling data (such as information gain, entropy value, or variance contribution rate), divide the contribution value of the feature by the contribution values of all features to obtain the objective weight of each feature, and generate the second data weight. Fit the subjective weight and the objective weight through a weighting formula to generate the final weight distribution; attach the final weight distribution to each node of the updated data chain, and each node's data feature is assigned the corresponding weight.

[0035] Based on the transformer status, carry out operation and maintenance regulation and control of the transformer.

[0036] According to the predicted operating status of the transformer, formulate corresponding operation and maintenance debugging and control strategies. For example, when a potential fault is predicted, preventive maintenance measures can be taken in advance to reduce the fault risk and improve the operating efficiency and safety of the system.

[0037] Furthermore, carrying out operation and maintenance regulation and control of the transformer includes: Based on the transformer status, generate an operation and maintenance plan for the transformer; identify the operation and maintenance plan for the transformer, and decompose and determine the automated plan and the manual plan; based on the automated plan, generate operation regulation and control instructions to carry out operation and maintenance debugging management of the transformer, including regulating transformer parameters and regulating components of the transformer connection circuit; based on the manual plan, generate operation and maintenance intervention instructions and transmit them to the terminal devices of the operation and maintenance personnel.

[0038] After determining the transformer status, generate an operation and maintenance debugging plan based on the status prediction result. First, output the operating status label of the current transformer through the status prediction model, including normal operation, warning status or fault status, and identify relevant abnormal features (such as temperature rise, abnormal vibration or oil pressure fluctuation). According to the status label, formulate an operation and maintenance debugging plan and decompose it into automated regulation tasks and manual intervention tasks. For automated tasks, such as adjusting the load distribution, enhancing the operation intensity of the cooling system or switching the tap changer of the winding, generate specific operation instructions through the automated control system and send them to the control module of the transformer in real time to achieve rapid regulation. After the regulation is completed, the system collects the adjusted operation data in real time, verifies the regulation effect and judges whether further optimization is required.

[0039] At the same time, for tasks that require manual intervention, such as checking the mechanical connection points of the winding, testing the pipeline pressure of the cooling system or replacing damaged components, generate intervention instructions for the operation and maintenance personnel. These instructions include specific task steps, a list of required tools and safety precautions, and are sent to the on-site personnel through the operation and maintenance terminal (such as a tablet or mobile phone) to assist them in accurately performing the operations. During the operation and maintenance process, the execution results of all regulations and interventions will be uploaded to the system in real time to form a complete operation and maintenance record.

[0040] After the execution is completed, the system rechecks the transformer status based on the regulation and intervention results to determine whether the anomalies have been resolved. If the status returns to "normal operation", the current operating parameters are recorded as the benchmark for subsequent monitoring. If it remains in the warning state or fault state, the system will re-analyze the anomaly characteristics, adjust the regulation strategy, or prompt for a higher level of manual intervention. All operation and maintenance data and debugging results will be synchronously transmitted back to the state prediction model to optimize the model parameters and improve the accuracy of future predictions and operation and maintenance efficiency. Through this closed-loop operation and maintenance regulation and control process, it is ensured that the transformer always maintains an efficient, safe, and controllable state during operation.

[0041] In summary, the embodiments of the present application at least have the following technical effects: First, the fault operation and maintenance records of the interactive transformer are retrieved to mine the fault correlation network, where the fault correlation network is constructed based on the fault type - fault characteristics. Then, a front-end sensing array is deployed to perform operation monitoring sampling for the transformer operation scenario to determine the sensing sampling data, including periodic sampling and fluctuation sampling by introducing a variable-frequency monitor. Then, taking the fault correlation network as the criterion and the subjective and objective weighting based on the fault characteristics as the principle, the state prediction model is supervised and trained, where the prediction is based on the time-series data chain, and the time-series data chain is updated by the average weakening process of the data nodes. Next, the sensing sampling data is transmitted back, combined with the state prediction model, to update the time-series data chain and predict the transformer status to determine the transformer status. Finally, based on the transformer status, the operation and maintenance regulation and control of the transformer are carried out. This solves the technical problem in the prior art of inaccurate transformer state prediction resulting in low operation and maintenance efficiency and achieves the technical effect of improving the accuracy of transformer state prediction.

[0042] It should be noted that the above sequence of the embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the specific embodiments of this specification have been described above. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0044] This specification and the drawings are merely illustrative of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications therein.

Claims

1. A transformer condition prediction method that fuses vibration and temperature data, characterized in that, The method includes: Interact with the fault operation and maintenance records of the transformer, and mine the fault correlation network, where the fault correlation network is formed based on fault types - fault characteristics; Deploy a front - end sensing array, perform operation monitoring sampling for the transformer operation scenario, and determine the sensing sampling data, including periodic sampling and fluctuation sampling by introducing a frequency - conversion monitor; Using the fault correlation network as a criterion and the subjective and objective weighting based on fault characteristics as a principle, supervise and train the state prediction model, where the prediction is based on the time - series data chain, and the time - series data chain is updated by the average weakening process of data nodes; Transmit back the sensing sampling data, combine it with the state prediction model, perform time - series data chain update and transformer state prediction, and determine the transformer state; Based on the transformer state as a benchmark, conduct transformer operation and maintenance control.

2. The transformer status prediction method for fusing vibration and temperature data according to claim 1, wherein Deploy a front - end sensing array, perform operation monitoring sampling for the transformer operation scenario, including: Introduce a frequency - conversion monitor and establish a communication connection between the frequency - conversion monitor and the front - end sensing array; Set a preset sampling period as the first sampling rule; Based on the fault operation and maintenance records, mine the characteristic fluctuation threshold based on operation risk, and set the second sampling rule based on the characteristic fluctuation threshold; Configure the front - end sensing array based on the first sampling rule and configure the frequency - conversion monitor based on the second sampling rule; Among them, when the characteristic fluctuation threshold is met, the frequency - conversion monitor generates a sensing sampling instruction.

3. The transformer status prediction method for fusing vibration and temperature data according to claim 1, characterized in that, Perform time - series data chain update and transformer state prediction, including: Set a preset number of nodes, perform time - series integration on the transformer operation data, and determine the time - series data chain, where the preset number of nodes is N; As the sensing sampling data is transmitted back, update the time - series data chain in the state prediction model to determine the updated data chain; Perform transformer state trend prediction and fault risk determination on the updated data chain to determine the predicted transformer state.

4. The transformer status prediction method for fusing vibration and temperature data according to claim 3, wherein Updating the time - series data chain in the state prediction model includes: Based on the sensing sampling data, generate new data nodes; Traverse the time - series data chain, and determine N - 1 groups of buffer operators through the average weakening process of data; Perform time - series integration on the N - 1 groups of buffer operators and the new data nodes as the updated data chain.

5. The transformer condition prediction method for fusing vibration and temperature data according to claim 4, wherein Determine N - 1 groups of buffer operators through the average weakening process of data, including: Traverse the time - series data chain, and determine the first - node data and the second - node data in the time - series growth direction; Perform mapping and mean calculation on the first - node data and the second - node data to determine the first buffer operator, where the first buffer operator is the first link node of the updated data chain; Determine the second - node data and the third - node data and perform mapping mean calculation to determine the second buffer operator; Until the calculation of the (N - 1)th buffer operator is completed, integrate the first buffer operator, the second buffer operator until the (N - 1)th buffer operator to determine the N - 1 groups of buffer operators.

6. The transformer status prediction method for fusing vibration and temperature data according to claim 3, characterized in that, After determining the updated data chain, it includes: Perform principal component analysis screening based on fault characteristics for the sensing sampling data to determine the fault - related characteristics; Identify the fault-related features and subjectively assign weights to determine the first data weight; Identify the fault-related features and objectively assign weights to determine the second data weight; Fit the first data weight and the second data weight to determine the data weight distribution and identify the updated data chain.

7. The transformer state prediction method for fusing vibration and temperature data according to claim 1, characterized in that Mine the fault correlation network, including: Traverse the fault operation and maintenance records, perform clustering processing based on the fault type, and determine multiple clustering clusters; Identify the first clustering cluster within the multiple clustering clusters, and determine multiple groups of fault feature intervals. Among them, the first clustering cluster corresponds to the first fault type, and the maximum and minimum values of the same fault features within the cluster are used to determine the fault feature intervals; Traverse the multiple groups of fault feature intervals and determine the feature critical values based on the risk boundary; Based on the first fault type and the feature critical values, determine the first correlation network.

8. The transformer state prediction method for fusing vibration and temperature data according to claim 1, characterized in that, Carry out the operation and maintenance control of the transformer, including: Generate a transformer operation and maintenance plan based on the transformer status; Identify the transformer operation and maintenance plan and decompose it to determine the automated plan and the manual plan; Generate an operation and control instruction based on the automated plan to perform operation and maintenance debugging management on the transformer, including transformer parameter regulation and component regulation of the transformer connection circuit; Generate an operation and maintenance intervention instruction based on the manual plan and transmit it to the terminal device of the operation and maintenance personnel.

Images