Power equipment failure risk prediction method, system, device and medium

Abstract

This invention relates to the field of power equipment technology and discloses a method, system, device, and medium for predicting power equipment fault risks. The method acquires the maintenance work order text and time-series data of the target power equipment, and standardizes and associates them according to equipment identification and time information to form fused operation and maintenance data. By performing fault semantic mining on the semantic information of the maintenance work orders in the fused operation and maintenance data, fault patterns related to equipment defects, abnormal phenomena, and maintenance handling can be extracted from natural language records. Through time-series correlation analysis of fault patterns and state quantity characteristics, a fault correlation graph is constructed. Based on the fault correlation graph, the current state quantity information and maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period, improving the data completeness, advance notice, accuracy, and interpretability of fault risk prediction.

Description

Technical Field

[0001] This invention relates to the field of power equipment technology, and in particular to a method, system, device and medium for predicting power equipment failure risks. Background Technology

[0002] In the daily operation and maintenance of power systems, a large number of historical maintenance work orders record key information such as the time, location, phenomenon, and handling process of equipment failures. However, this information is mostly in the form of natural language and is difficult to use directly for quantitative analysis and model building.

[0003] Meanwhile, existing solutions often suffer from data fragmentation issues, with unstructured semantic information from maintenance work orders not being effectively integrated with time-series data from status monitoring; fault mode recognition relies on manual rules, making it difficult to cover the implicit fine-grained fault knowledge hidden in massive amounts of data; and prediction models that rely solely on status quantities or work order texts have information bias, resulting in insufficient advance warning and accuracy of risk prediction, making it difficult to provide maintenance personnel with visualized risk warnings and targeted decision-making suggestions. Summary of the Invention

[0004] In view of this, in order to solve the above-mentioned technical problems, the present invention provides a method, system, device and medium for predicting the failure risk of power equipment.

[0005] The first aspect of this invention provides a method for predicting the failure risk of power equipment, the method comprising: The maintenance work order text and status quantity time series data of the target power equipment are obtained, and the maintenance work order text and status quantity time series data are standardized to obtain fused operation and maintenance data associated with equipment identification and time information. Fault semantic mining is performed on the semantic information of maintenance work orders in the fused operation and maintenance data to identify fault modes related to the target power equipment and determine the state quantity characteristics of the time period corresponding to the fault mode. Based on the fault modes and the state variable characteristics, a time-series correlation analysis is performed to construct a fault correlation map to characterize the correlation between state variable changes and fault modes. Based on the fault association map, the current state information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period.

[0006] In one example, the standardization process of the maintenance work order text and the status quantity time-series data yields fused operation and maintenance data associated with equipment identifiers and time information, including: The maintenance work order text is cleaned, segmented, and filtered for stop words to obtain a valid word sequence for the work order. Convert the effective word sequence of the work order into work order semantic features; The state quantity time series data is subjected to time alignment, denoising and normalization to obtain a standardized state quantity sequence; Based on the equipment identifier and the time information corresponding to the maintenance work order, the semantic features of the work order are associated with the standardized state quantity sequence to obtain the fused operation and maintenance data.

[0007] In one example, the step of performing fault semantic mining on the semantic information of maintenance work orders in the fused operation and maintenance data, identifying fault modes related to the target power equipment, and determining the state quantity features of the time period corresponding to the fault mode includes: The semantic information of the maintenance work order in the fused operation and maintenance data is input into the preset semantic recognition model to extract the contextual semantic features in the maintenance work order text; Based on attention weights, key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance procedures is determined from the contextual semantic features. Based on the key semantic content, fault mode classification is performed to obtain the fault modes related to the target power equipment; Based on the maintenance work order time corresponding to the fault mode, determine the time window before the fault and the time window after the fault. State quantity statistical features are extracted from the state quantity time series data within the pre-fault time window and the post-fault time window.

[0008] In one example, after identifying the fault modes associated with the target power equipment, the method further includes: Based on the identification confidence level of the fault modes, filter out the fault modes that meet the confidence level conditions; The selected fault modes, associated maintenance work order identifiers, associated equipment identifiers, and frequency of occurrence are written into the fault mode dictionary. When a new fault mode or a new semantic expression of an existing fault mode appears in a newly added maintenance work order text, the fault mode dictionary is incrementally updated.

[0009] In one example, the step of performing time-series correlation analysis based on the fault mode and the state variable characteristics to construct a fault correlation map characterizing the relationship between state variable changes and fault modes includes: Based on the occurrence time or maintenance work order time corresponding to the fault mode, determine the correlation analysis time window corresponding to the fault mode; Within the correlation analysis time window, anomaly identification and statistical characterization are performed on the state quantity features to obtain state quantity correlation features that have a time-series correspondence with the fault mode; Based on the co-occurrence, sequential, and combined anomaly relationships between the state variable correlation features and the fault modes within the correlation analysis time window, the correlation strength of the state variable correlation features to the fault modes is determined. Using the fault modes and the state variables as graph nodes, and establishing weighted association edges between graph nodes based on the association strength, a fault association graph is obtained to characterize state variable changes, combined anomalies, and fault mode pointing relationships.

[0010] In one example, the step of fusing and predicting the current state information of the target power equipment and the maintenance work order text based on the fault association map to obtain the fault risk result of the target power equipment in the future prediction period includes: Temporal features are extracted from the current state information of the target power equipment to obtain the state quantity temporal features; Semantic features are extracted from the maintenance work order text of the target power equipment to obtain the work order semantic features; The fault association graph is used to enhance the association between the temporal features of the state variables and the semantic features of the work orders, thereby obtaining fused risk features. The fused risk characteristics are input into a preset risk prediction model, and the risk probability of the target power equipment experiencing different failure modes in the future prediction period is output through the risk prediction model. The fault risk level of the target power equipment is determined based on the risk probability.

[0011] In one example, the step of fusing and predicting the current state information of the target power equipment and the maintenance work order text based on the fault association map to obtain the fault risk result of the target power equipment in the future prediction period, further includes: Receive new maintenance work order text and new status quantity timing data based on the aforementioned fault risk results; The newly added maintenance work order text and the newly added status quantity time series data are correlated to obtain the newly added integrated operation and maintenance data; The fault correlation graph and the preset risk prediction model used to obtain the fault risk results are updated based on the newly added integrated operation and maintenance data.

[0012] Secondly, the present invention also provides a power equipment fault risk prediction system, the system comprising: The data fusion module is used to acquire the maintenance work order text and status quantity time series data of the target power equipment, and to standardize the maintenance work order text and the status quantity time series data to obtain fused operation and maintenance data associated with equipment identification and time information. The state quantity feature determination module is used to perform fault semantic mining on the maintenance work order semantic information in the fused operation and maintenance data, identify the fault modes related to the target power equipment, and determine the state quantity features of the time period corresponding to the fault mode. The graph construction module is used to perform time-series correlation analysis based on the fault modes and the state variable characteristics, and to construct a fault correlation graph to characterize the correlation between state variable changes and fault modes. The fault risk prediction module is used to fuse and predict the current status information of the target power equipment and the maintenance work order text based on the fault association map, so as to obtain the fault risk result of the target power equipment in the future prediction period.

[0013] Thirdly, the present invention also provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the power equipment fault risk prediction method as described in the first aspect.

[0014] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the power equipment fault risk prediction method as described in the first aspect.

[0015] As can be seen from the above technical solutions, this invention acquires the maintenance work order text and status quantity time-series data of the target power equipment, and standardizes and associates them according to equipment identification and time information. This enables data that was originally scattered on the maintenance management side and the status monitoring side to form fused operation and maintenance data that can be jointly analyzed, thus providing a unified data foundation for subsequent risk assessment. Furthermore, by performing fault semantic mining on the semantic information of maintenance work orders in the fused operation and maintenance data, fault patterns related to equipment defects, abnormal phenomena, and maintenance handling can be extracted from natural language records. The correspondence between semantic information and operating status can be established by combining the status quantity characteristics of the corresponding time period, reducing the limitations of simply relying on manual experience or fixed rules to identify fault patterns. On this basis, by performing time-series correlation analysis on fault patterns and status quantity characteristics and constructing a fault correlation graph, the correlation between status quantity changes, combined anomalies, and fault patterns can be expressed in a structured way. Finally, based on the fault correlation graph, the current status quantity information and maintenance work order text are fused and predicted to obtain the fault risk results of the target power equipment in the future prediction period. Therefore, this technical solution can improve the data integrity, advance notice, accuracy and interpretability of fault risk prediction, and provide a more reliable technical basis for risk prediction and maintenance decisions in power equipment operation and maintenance. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an application environment diagram of a power equipment fault risk prediction method provided in an embodiment of the present invention; Figure 2 A flowchart of a power equipment fault risk prediction method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a power equipment fault risk prediction system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The power equipment fault risk prediction method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102 or placed on a cloud or other network server. Terminal 101 or server 102 executes a power equipment fault risk prediction method, which includes: acquiring maintenance work order text and state quantity time-series data of the target power equipment, and standardizing the maintenance work order text and state quantity time-series data to obtain fused operation and maintenance data associated with equipment identification and time information; performing fault semantic mining on the maintenance work order semantic information in the fused operation and maintenance data to identify fault modes related to the target power equipment and determine the state quantity characteristics of the corresponding time period of the fault mode; performing time-series correlation analysis based on the fault mode and state quantity characteristics to construct a fault correlation graph to characterize the correlation between state quantity changes and fault modes; and performing fused prediction on the current state quantity information and maintenance work order text of the target power equipment according to the fault correlation graph to obtain the fault risk result of the target power equipment in the future prediction period.

[0020] Terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, and tablets.

[0021] Server 102 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0022] In one optional implementation, the target power equipment may include transformers, circuit breakers, switchgear, transmission line equipment, distribution substation equipment, disconnect switches, cable equipment, and other power equipment requiring condition monitoring and maintenance management. The maintenance work order text can originate from a production management system, asset management system, defect management system, inspection system, or mobile maintenance terminal. The maintenance work order text may include information such as equipment name, equipment identification, fault description, defect phenomenon, maintenance measures, handling results, maintenance personnel records, acceptance conclusion, and work order generation time. The status quantity time-series data can originate from an online monitoring system, SCADA system, PMU system, sensor acquisition system, test data management system, or substation monitoring system. The status quantity time-series data may include data such as voltage, current, temperature, vibration, partial discharge, insulation resistance, load rate, voltage deviation, current harmonics, conductor temperature, icing thickness, and tower tilt.

[0023] like Figure 2 As shown in the embodiments of this application, a method for predicting the fault risk of power equipment is provided, which is applied to... Figure 1 Taking terminal 101 or server 102 as an example, the explanation includes the following steps S1 to S4. Wherein: Step S1: Obtain the maintenance work order text and status quantity time series data of the target power equipment, and standardize the maintenance work order text and status quantity time series data to obtain fused operation and maintenance data associated with equipment identification and time information.

[0024] In one optional implementation, historical maintenance work order texts can be obtained from the production management system, and status quantity time-series data can be obtained from the online monitoring system via application programming interfaces (APIs), database query interfaces, or data file import methods. To ensure the accuracy of subsequent correlations, the equipment identifier, work order number, work order generation time, defect discovery time, maintenance completion time, status quantity sampling time, and data source identifier are retained during data collection. The equipment identifier can be an equipment ID, asset number, interval number, line number, or transformer substation number, and the time information can be the original timestamp or a standard timestamp calibrated to a unified time zone.

[0025] For maintenance work order texts, text cleaning is performed first. Text cleaning may include removing HTML tags, special symbols, invalid spaces, repeated punctuation, system template fields and fixed-format statements irrelevant to fault identification. For example, for a text such as "#1 main transformer body has high temperature, infrared temperature measurement review has been arranged; treatment result: check the operating status of the cooler", words with fault or maintenance meanings such as "main transformer, body, high temperature, infrared temperature measurement, cooler, operating status" can be retained. Subsequently, word segmentation and stop word filtering are performed on the cleaned text to remove high-frequency words lacking fault indication significance such as "of", "了 (le)", "conduct", "already" to obtain an effective word sequence of the work order.

[0026] For state quantity time-series data, since the sampling frequencies of different monitoring systems may be different, and the time of the maintenance work order is usually not completely consistent with the sampling time of the state quantity, time alignment processing is required. In an optional method, a linear interpolation method can be used to align the state quantity time-series data to the time point corresponding to the maintenance work order.

[0027] Step S2: performing fault semantic mining on the maintenance work order semantic information in the fused operation and maintenance data, identifying the fault mode related to the target power equipment, and determining the state quantity characteristics of the time period corresponding to the fault mode.

[0028] Wherein, fault semantic mining can be implemented by a text classification model, a sequence labeling model or a combination of the two. The text classification model is used to judge the corresponding fault mode according to the overall semantics of the maintenance work order, for example, classifying the work order into categories such as "overheating fault", "insulation fault", "mechanical mechanism fault", "abnormal conductive circuit", "communication abnormality", "overload risk". The sequence labeling model is used to identify the fault location, fault phenomenon, fault cause and maintenance disposal from the work order text, for example, from "the circuit breaker is not closed in place, it is found that the mechanism is jammed after inspection, and cleaning and lubrication have been carried out", identify "circuit breaker" as the equipment object, "not closed in place" as the fault phenomenon, "mechanism jammed" as the fault cause, and "cleaning and lubrication" as the maintenance disposal.

[0029] After identifying the fault mode, it is necessary to determine the state quantity characteristics of the time period corresponding to the fault mode. Specifically, in the time window before and after the fault, statistical characteristics of the state quantity can be extracted, including mean value, variance, maximum value, minimum value, peak value, change rate, main frequency component, trend slope, mean offset before and after the fault, etc.

[0030] Step S3: performing time-series correlation analysis based on the fault mode and state quantity characteristics, and constructing a fault correlation graph for characterizing the correlation between state quantity changes and fault modes.

[0031] In one optional implementation, a fault correlation graph is used to describe the dynamic correlation between changes in state variables and fault modes. Nodes in the fault correlation graph may include fault mode nodes, state variable feature nodes, state variable abnormal event nodes, equipment nodes, and historical work order nodes. Specifically, fault mode nodes represent fault modes such as overheating, insulation abnormalities, mechanical jamming, overload, and voltage exceeding limits; state variable feature nodes represent characteristics such as average temperature, peak vibration, insulation resistance decrease rate, current harmonic content, and load rate change rate; and state variable abnormal event nodes represent events such as state variables exceeding thresholds, continuous rise in state variables, abrupt changes in state variables, and simultaneous abnormalities in multiple state variables.

[0032] Step S4: Based on the fault association map, fuse and predict the current state information of the target power equipment and the maintenance work order text to obtain the fault risk result of the target power equipment in the future prediction period.

[0033] In one optional implementation, the fusion prediction is achieved using a two-stream feature extraction and graph enhancement approach. The first stream processes the current state information of the target power equipment, extracting temporal features of the state quantities; the second stream processes the maintenance work order text of the target power equipment, extracting semantic features of the work orders; and the fault association graph is used to enhance the association between the temporal features of the state quantities and the semantic features of the work orders, enabling the fused risk features to simultaneously include real-time state changes, historical maintenance semantics, and state-fault association knowledge.

[0034] It should be noted that, in this embodiment, by acquiring the maintenance work order text and status quantity time-series data of the target power equipment, and standardizing and associating them according to equipment identification and time information, the data originally scattered on the maintenance management side and the status monitoring side can be formed into fused operation and maintenance data that can be jointly analyzed, thereby providing a unified data foundation for subsequent risk assessment. Furthermore, by performing fault semantic mining on the semantic information of the maintenance work orders in the fused operation and maintenance data, fault patterns related to equipment defects, abnormal phenomena, and maintenance handling can be extracted from natural language records, and a correspondence between semantic information and operating status can be established by combining the status quantity characteristics of the corresponding time period, reducing the limitations of simply relying on manual experience or fixed rules to identify fault patterns. On this basis, by performing time-series correlation analysis on fault patterns and status quantity characteristics and constructing a fault correlation graph, the correlation between status quantity changes, combined anomalies, and fault patterns can be expressed in a structured way. Finally, based on the fault correlation graph, the current status quantity information and maintenance work order text are fused and predicted to obtain the fault risk results of the target power equipment in the future prediction period. Therefore, this technical solution can improve the data integrity, advance notice, accuracy and interpretability of fault risk prediction, and provide a more reliable technical basis for risk prediction and maintenance decisions in power equipment operation and maintenance.

[0035] In some embodiments, standardization processing is performed on maintenance work order texts and state quantity time-series data to obtain fused operation and maintenance data associated according to device identifiers and time information, which comprises: performing text cleaning, word segmentation and stop word filtering on maintenance work order texts to obtain valid word sequences of work orders; converting the valid word sequences of work orders into work order semantic features; performing time alignment, denoising and normalization processing on the state quantity time-series data to obtain standardized state quantity sequences; and associating the work order semantic features with the standardized state quantity sequences according to the device identifiers and the time information corresponding to the maintenance work orders to obtain the fused operation and maintenance data.

[0036] For maintenance work order texts, text cleaning is performed first. Text cleaning may include deleting HTML tags, special symbols, invalid spaces, repeated punctuation marks, system template fields and fixed-format statements irrelevant to fault identification. Word segmentation processing uses a Chinese word segmentation tool (such as Jieba or HanLP) to segment the work order text into a word sequence, for example, segmenting "transformer overheating maintenance" into ["transformer", "overheating", "maintenance"]. Stop word removal is based on a predefined stop word list (including high-frequency meaningless words such as "of" and "is"), and key terms are retained after filtering. A pre-trained word vector model (such as Word2Vec or FastText) is used to map each word to a vector of fixed dimensions, wherein the word vector is calculated through model parameters: , here is the word embedding matrix, is the word 's one-hot encoding vector. The overall feature vector of the work order is generated through average pooling: , wherein is the number of valid words in the work order, is the -th word's vector. This processing converts unstructured text into structured features, significantly improves the ability to capture semantic similarity, and is beneficial to subsequent fault pattern recognition.

[0037] For state quantity time-series data, since the sampling frequencies of different monitoring systems may be different, and the time of maintenance work orders is usually not completely consistent with the sampling time of state quantities, time alignment processing is required. In an optional mode, a linear interpolation method can be used to align the state quantity time-series data to the time point corresponding to the maintenance work order.

[0038] Specifically, the alignment of state quantity time-series data is based on the work order timestamp , and the linear interpolation method is used to adjust the time points of the state sequence:

[0039] wherein and It is a nearby timestamp. and This corresponds to the state value. The noise reduction process uses a sliding window averaging filter.

[0040] Among them, window size Set according to the sampling frequency (e.g.) (corresponding to 1Hz data) It is time Nearby state values.

[0041] Subsequently, the denoised state variables are normalized or standardized. Normalization uses Z-score standardization: , It is the historical mean of the state variables. It uses standard deviation to ensure consistent data scaling. Correlation operations are performed using device identifiers. and precise timestamp Matching semantic feature vectors With standardized state sequences:

[0042] in This refers to the state dimension. When storing data in the knowledge base, a relational database (such as MySQL) or a time-series database (such as InfluxDB) is used. The table structure includes fields such as device ID, timestamp, feature vector, and state sequence. Normalization eliminates differences in units, which is beneficial for multi-source data fusion analysis and improves the robustness of subsequent prediction models.

[0043] After obtaining the semantic features of the work order and the standardized status quantity sequence, they are associated according to the equipment identifier and time information. Specifically, the equipment identifier can be used as the first association key, and the maintenance work order time, defect discovery time, or maintenance completion time can be used as the time reference. Association windows are set before and after the time reference, and the standardized status quantity sequence within the window is bound to the corresponding work order semantic features to form fused operation and maintenance data.

[0044] By integrating data that was originally scattered across the maintenance management system and online monitoring system according to the same equipment and the same time scale, the fault description in the work order text can correspond to the changes in state quantities before and after the fault occurred, thus providing a data foundation for jointly judging fault risks from both text semantics and equipment status.

[0045] In some embodiments, fault semantic mining is performed on the semantic information of maintenance work orders in the integrated operation and maintenance data to identify fault modes related to the target power equipment and determine the state quantity features of the time period corresponding to the fault mode. This includes: inputting the semantic information of maintenance work orders in the integrated operation and maintenance data into a preset semantic recognition model to extract contextual semantic features from the maintenance work order text; determining key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance handling from the contextual semantic features based on attention weights; classifying fault modes according to the key semantic content to obtain fault modes related to the target power equipment; determining the pre-fault time window and the post-fault time window based on the maintenance work order time corresponding to the fault mode; and extracting state quantity statistical features from the state quantity time series data within the pre-fault time window and the post-fault time window.

[0046] In one specific computational process, the semantic features of the work order can be input into a bidirectional long short-term memory network (BiLSTM) to obtain the contextual semantic features corresponding to each word or phrase.

[0047] The semantic features of the work order include a preprocessed sequence of word embedding vectors. ,in Indicates the first Each word 3D embedding vector, This refers to the length of the work order text sequence. These feature data have been standardized through semantic analysis to ensure input consistency.

[0048] The hidden state of BiLSTM is calculated as follows:

[0049] in It is a time step The hidden state vector, It is the hidden unit dimension. This represents vector concatenation. It captures contextual dependencies, providing a foundation for subsequent attention mechanisms. An attention mechanism is introduced to automatically learn key phrases, and attention weights are calculated. and context vector : ,in: ,in , and These are trainable parameters. It is the attention dimension. Indicates the first The weight of each word is assigned, with higher weights corresponding to key phrases related to equipment defects or anomalies.

[0050] To enable the model to automatically focus on key semantic content relevant to fault identification, attention weights can be introduced. The advantage of the attention mechanism is that it automatically focuses on fine-grained patterns in the work order text, such as "bearing wear" or "voltage fluctuation," without requiring manual rules, significantly improving the accuracy and interpretability of fault identification. For text classification tasks, the model's output layer uses the softmax function:

[0051] in, , , This is the number of fault mode categories (such as "mechanical failure" or "electrical maintenance"). It predicts the probability distribution. The loss function used is cross-entropy loss: For fault mode classification, the attention semantic vector can be input into the classification layer to obtain the probability of each fault mode: p = softmax(W c +b c ) Where p represents the probability distribution of different failure modes, W c and b c This represents the parameters of the classification layer. The cross-entropy loss function can be used during model training.

[0052] in It is the number of training samples. It is a one-hot encoding of the true labels. The training process uses the Adam optimizer, executed in parallel on a GPU cluster, with a batch size of 128, iterating until convergence. The model output is used to classify potential failure modes, filtered by a threshold (e.g., ...). The system extracts high-confidence results and automatically updates the fault mode dictionary. The dictionary is stored as key-value pairs, where the key is a fault mode description and the value is the relevant work order ID and frequency. Incremental learning is supported to handle new work order data. This dynamic update mechanism ensures the continuous extraction of implicit knowledge from massive amounts of historical data, improving the efficiency of operational decision-making.

[0053] By training this model, implicit fault knowledge can be automatically extracted from a large number of historical maintenance work orders, reducing reliance on manual rules and manual keyword configuration.

[0054] After identifying the fault modes, it is necessary to determine the state characteristics of the corresponding time periods. Specifically, this can be done by using the maintenance work order time, defect discovery time, or fault confirmation time as a benchmark to construct pre-fault and post-fault time windows. For example, the fault confirmation time t can be used as a benchmark. fPrevious time interval [t] f -Δt1,t f As the time window before the fault, t f The subsequent time interval [t] f ,t f +Δt2] serves as the post-fault time window, where Δt1 and Δt2 can be determined based on the equipment type and the fault evolution rate. For temperature-related risks, a longer window can be selected; for shock vibration anomalies, a shorter window can be selected.

[0055] Within the time window before and after a fault, statistical features of state variables can be extracted, including mean, variance, maximum value, minimum value, peak value, rate of change, dominant frequency component, trend slope, and mean offset before and after the fault.

[0056] This application not only identifies fault modes from maintenance texts, but also establishes a connection between fault modes and changes in state variables within the corresponding time period. This enables subsequent correlation analysis to move beyond simple text classification or simple state monitoring, and to reflect the correspondence between "fault semantics and state changes".

[0057] In some embodiments, after identifying the fault modes associated with the target power equipment, the method further includes: filtering fault modes that meet the confidence level conditions based on the identification confidence level of the fault modes; writing the filtered fault modes, associated maintenance work order identifiers, associated equipment identifiers, and occurrence frequencies into a fault mode dictionary; and incrementally updating the fault mode dictionary when a new fault mode appears in a newly added maintenance work order text or when a new semantic expression of an existing fault mode is added.

[0058] In one alternative implementation, the fault mode dictionary can be stored as key-value pairs or graph node attributes. The keys in the fault mode dictionary can be fault mode names or standardized fault descriptions, and the values ​​can include information such as associated maintenance work order numbers, associated equipment identifiers, typical keywords, frequency of occurrence, most recent occurrence time, associated status variables, maintenance and handling measures, and manual confirmation status. For example, the fault mode "cooling system abnormality" can be associated with semantic expressions such as "cooler disconnection," "oil temperature rise," "fan abnormality," and "poor heat dissipation," and with status variables such as temperature, load rate, and cooler operating status.

[0059] In the fault mode recognition results, high-confidence samples can be selected based on the recognition confidence level output by the model. Let the predicted probability of fault mode m corresponding to the k-th work order be P(m|T). k If P(m|T) k )≥θ m If so, the fault mode is written into the fault mode dictionary as a high-confidence fault mode, where θ mThis is the confidence threshold for fault mode m. Different thresholds can be set for different fault modes. For example, a higher threshold can be set for severe faults or faults with a high risk of false alarms, while a relatively lower threshold can be set for gradual faults that require early warning.

[0060] When new fault modes or new semantic expressions for existing fault modes appear in newly added maintenance work orders, the fault mode dictionary can be incrementally updated. New semantic expressions refer to descriptions that are the same as or similar in meaning to existing fault modes but with different wording. For example, "closing incomplete," "closing failed," and "closing not reached the end point" may all point to an abnormality in the closing mechanism. By merging new semantic expressions into the corresponding fault modes, the ability of subsequent models to recognize different expressions can be improved.

[0061] In some embodiments, a fault association graph is constructed based on fault modes and state variable features through temporal correlation analysis to characterize the correlation between state variable changes and fault modes. This includes: determining a correlation analysis time window corresponding to the fault mode based on the occurrence time or maintenance work order time corresponding to the fault mode; within the correlation analysis time window, performing anomaly identification and statistical characterization on the state variable features to obtain state variable correlation features that have a temporal correspondence with the fault mode; determining the correlation strength of the state variable correlation features to the fault mode based on the co-occurrence relationship, sequential relationship, and combined anomaly relationship between the state variable correlation features and the fault mode within the correlation analysis time window; and using the fault mode and state variable correlation features as graph nodes, and establishing weighted correlation edges between the graph nodes based on the correlation strength to obtain a fault association graph characterizing the relationship between state variable changes, combined anomalies, and fault modes.

[0062] Here, co-occurrence relationships describe whether state variable anomalies and fault modes occur within the same association analysis window; sequence relationships describe whether state variable anomalies precede fault modes, thus determining whether they have fault precursor significance; and combined anomaly relationships describe the indicative role of multiple state variable anomalies occurring together for a particular fault mode. If a state variable anomaly typically occurs before a fault mode and has high confidence and lift, then this state variable anomaly can be considered an important precursor feature of that fault mode.

[0063] Before constructing the fault correlation map, the correlation analysis time window can be determined based on the occurrence time or maintenance work order time corresponding to the fault mode. Let the fault mode event be F. m Its occurrence time is t f The state variable characteristic is S i Then, an association analysis window can be constructed: W f =[t f -τ1,t f +τ2] Among them, W f Indicates fault mode F m The corresponding correlation analysis time windows are defined as follows: τ1 represents the analysis duration before the failure occurs, and τ2 represents the analysis duration after the failure occurs. For risk prediction, the focus can be placed on the window before the failure occurs [t]. f -τ1,t f This allows for the discovery of early warning signs of faults; for fault attribution, both the pre-fault window and the post-fault window can be analyzed simultaneously to identify the differences in state before and after the fault.

[0064] Within the correlation analysis time window, anomalies can be identified and statistically characterized in state variables. Anomaly identification can be based on fixed thresholds, dynamic thresholds, historical quantile thresholds, or model residual thresholds. For example, for the standardized state variable z... t You can set an exception event A. i for: A i ={z t >γ i} Where, γ i This represents the anomaly threshold corresponding to state variable i. Alternatively, trend events such as continuous increases, continuous decreases, abrupt changes, and increased fluctuations in state variables can be considered as anomaly events to better describe the precursory characteristics of gradual faults.

[0065] After identifying state variable abnormal events and fault mode events, the co-occurrence relationship, sequence relationship, and combined abnormal relationship between state variable abnormal events and fault mode events can be calculated based on a sliding time window.

[0066] Specifically, define fault mode nodes and state variable attribute nodes. Fault mode node ( Each of the fault events identified in step two originates from the fault events identified in step two. This corresponds to a specific fault type, such as bearing wear. State variable attribute node. ( Based on the multidimensional state variables processed in step one and their statistical characteristics, including time-domain characteristics such as the mean. ,variance and frequency domain characteristics such as the dominant frequency component. ,in It can be represented as: The feature dimension is defined. Node type differentiation ensures heterogeneity; event nodes represent discrete fault occurrences, while attribute nodes encode continuous state changes. A time-series association rule mining algorithm is applied to process the time-series data. A sliding time window mechanism is employed, with a window length of... Based on the system sampling frequency setting, for example Seconds. Within each window, detect abnormal events in the state quantity. (like , (for predefined thresholds) and fault mode events The occurrence of this. Calculating association rules. support and confidence level The formula is:

[0067] in This represents the total number of windows. for The number of windows that occur, This represents the number of windows where both co-occur. Introducing time-series analysis can dynamically capture fault precursors, improving early warning accuracy by over 20% compared to static rules—a beneficial effect. Analyze the changes in state variables before and after the occurrence of a fault mode. For each... and Divide the window before the fault occurs and the window after the failure Calculate statistical differences such as mean shift:

[0068] Use the t-test to verify significance. , The standard deviation is the sample value. Combined anomaly detection is achieved through joint rules of multiple state variables, such as... Its improvement:

[0069] Quantify synergistic effects for The marginal probabilities are calculated. Transmission path analysis employs complex network techniques, identifying paths from state variables to faults based on graph structures. A heterogeneous graph network of "state variable-fault mode" is constructed. Vertex set:

[0070] Edge set connect arrive Edge weight Defined as association strength, using normalized confidence:

[0071] Ensure value range .

[0072] To comprehensively measure the strength of association, the weight calculation incorporates the lift value to avoid rare event bias, i.e.:

[0073] in, The weighting factor (default 0.7) accurately quantifies the indicative effect of state anomalies on specific faults, supporting root cause analysis, which is another beneficial effect. Network visualization uses a force-oriented layout, mapping edge weights to line widths, making it easy for engineers to intuitively identify key propagation paths, such as... .

[0074] When constructing a fault association graph, fault modes and state variable association features can be used as graph nodes, and weighted association edges can be established based on the association strength. If the state variable association feature S i With fault mode F m If the correlation strength w(Si,Fm) between the two reaches a preset threshold, then an edge e{i,m} is established between them, and w(S) is set to the threshold value. i ,F m () is used as the edge weight. The fault correlation graph can be represented as: G=(V,E,W) In this graph, G represents the fault association graph, V represents the set of nodes, E represents the set of edges, and W represents the set of edge weights. This graph visually indicates which state variable changes indicate which fault modes, and further allows for analysis of the critical paths through which state variable anomalies propagate to fault modes.

[0075] This application presents a structured representation of state variable change patterns, combined anomaly relationships, and fault mode orientation relationships in a graphical manner. This enables the risk prediction model to not only rely on a single state variable threshold or a single text classification result, but also to utilize the state-fault correlation patterns formed in historical data, thereby improving the interpretability and foresight of fault risk identification.

[0076] In some embodiments, based on a fault association graph, the current state information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period. This includes: extracting time-series features from the current state information of the target power equipment to obtain state time-series features; extracting semantic features from the maintenance work order text of the target power equipment to obtain work order semantic features; using the fault association graph to enhance the association between the state time-series features and the work order semantic features to obtain fused risk features; inputting the fused risk features into a preset risk prediction model, and outputting the risk probability of different fault modes occurring in the target power equipment in the future prediction period through the risk prediction model; and determining the fault risk level of the target power equipment based on the risk probability.

[0077] The "current state quantity information" can be the sequence of state quantities of the target device within a preset time window before the current prediction time, and should not be understood as the state quantity value at a single moment.

[0078] Maintenance work order texts can be recent work order texts, or work order texts corresponding to the current equipment, current status anomalies, or similar historical statuses. To ensure logical correlation with current status information, maintenance work order texts can be filtered by equipment identifier and time window. For example, work order texts from 30 days, 90 days, or one maintenance cycle prior to the current prediction time for the target power equipment can be selected, or historical work order texts with the same equipment type and similar status characteristics as the current status anomaly event can be selected.

[0079] Association enhancement refers to utilizing the correlation between state variable changes and fault modes in the fault association graph to adjust the weights, propagate information, or supplement features of the temporal features of state variables and the semantic features of work orders. For example, when the temperature rise feature is obvious in the temporal features of the current state variable, and the semantic features of the maintenance work order include semantics such as "cooler abnormality" and "poor heat dissipation", the fault association graph can enhance its connection with "overheating faults"; when the partial discharge quantity increases in the current state variable, and "discharge traces on insulating components" appear in the semantic features of the work order, the fault association graph can enhance its connection with "insulation abnormality".

[0080] The "fault risk result" can include risk probability and risk level, as well as the corresponding fault mode, risk source state quantity, risk contribution, associated fault paths, and suggested handling methods. The risk source state quantity can be determined by high-weight state quantity nodes connected to the target fault mode in the graph, and the risk contribution can be determined by combining the state quantity feature strength, graph edge weights, and model attention weights. By outputting the risk source and fault mode, the interpretability of the risk prediction results can be improved, making it easier for operations and maintenance personnel to understand the reasons for the warning.

[0081] In one specific implementation, for the extraction of temporal features of state variables, let the state variable sequence of the target power equipment within the current time window be:

[0082] in Indicates time The state vector. It can be... Inputting a Long Short-Term Memory (LSTM) network, a gated recurrent unit (GRU) network, or a temporal convolutional network yields the temporal features h of the state variables. s : h s =f s (X) Among them, f sh represents the temporal feature extraction function of state variables. s Used to characterize the current state evolution trend of target power equipment. Taking Long Short-Term Memory (LSTM) networks as an example, they can use gating mechanisms to remember the long-term change trends of state variables, and are suitable for extracting time-dependent risk features such as temperature, load rate, and insulation status.

[0083] The hidden state update formula for LSTM is as follows:

[0084] Final moments Hidden state As a temporal feature representation of state variables For the hidden layer dimension, These are the weights and bias parameters for the LSTM.

[0085] For maintenance work order text, the input is a set of recently associated work orders. Each work order Includes text descriptions, encoded into semantic vectors by a pre-trained language model (such as BERT). The device-work order relationship graph constructed using the preceding steps is used. As prior knowledge, the nodes Includes device nodes and work order nodes, edges This indicates the association between equipment and work orders, or the semantic / process association between work orders. As the initial feature of the work order node, the initial feature of the device node is set to zero, and a graph convolutional network (GCN) is applied for message propagation:

[0086] in Given an adjacency matrix with self-loops, Its degree matrix, For the first Layer node features For learnable weights, The activation function is (e.g., ReLU). After... After layer propagation, the feature vector corresponding to the target device node is extracted. As a semantic feature of work orders that integrates knowledge graphs.

[0087] Temporal characteristics of state variables semantic features of work orders The features are concatenated and explicit interaction terms are introduced to enhance feature coupling. Features are then fused. The structure is as follows:

[0088] in This represents vector concatenation. For element-wise multiplication, The absolute value is used for each element. This method of concatenating interactive features can explicitly capture the non-linear correlation pattern between abnormal status and work order semantics, significantly improving the model's ability to identify complex fault causes compared to simple concatenation or addition. Input is fed into a fully connected layer for multi-task risk probability prediction:

[0089] in , For weights and biases, This represents the number of fault categories. Output:

[0090] Indicates the device in the future The probability of various faults occurring within a given time period.

[0091] After model deployment, an incremental update pipeline is established. Newly generated tagged data (work orders and corresponding time period status values) is cached in a cache of size [size missing]. sliding window buffer At fixed time intervals or when accumulation Incremental training is triggered when a new sample is received. Parameters are updated using momentum-based stochastic gradient descent (SGD).

[0092] in For all trainable parameters of the model, For learning rate, The cross-entropy loss function is used. An Elastic Weight Consolidation (EWC) regularization term is introduced to constrain changes in important parameters and prevent catastrophic forgetting. For parameters The diagonal elements of the Fisher information matrix, For regularity, These are historically optimal parameters. This mechanism uses new data to fine-tune the model while retaining the learning results of historical failure modes through EWC, enabling the model to continuously adapt to equipment state drift and maintenance strategy evolution.

[0093] In one exemplary risk level classification method, if the risk probability p m If the risk level is below the first threshold θ1, it is determined to be low risk; if pm If the value is between the first threshold θ1 and the second threshold θ2, it is determined to be of medium risk; if p m If the risk level is above the second threshold θ2, it is considered high-risk. Different risk thresholds can be set for different failure modes. For example, a more sensitive threshold can be set for serious failures that may cause power outages or equipment damage, while a relatively higher threshold can be set for minor defects.

[0094] Among them, the current status trend, maintenance work order semantics and fault association map are used together for risk prediction, so that the prediction results not only reflect the current operating status of the equipment, but also take into account the historical maintenance semantics and status-fault association rules, thereby reducing the one-sidedness caused by single status prediction or single text analysis.

[0095] In some embodiments, based on the fault association graph, the current state quantity information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period. The method further includes: receiving new maintenance work order text and new state quantity time series data for the fault risk result; associating the new maintenance work order text and the new state quantity time series data to obtain new fused operation and maintenance data; and updating the fault association graph and the preset risk prediction model used to obtain the fault risk result based on the new fused operation and maintenance data.

[0096] In one optional implementation, the newly added maintenance work order text may include processing work orders generated after the warning, manual review records, maintenance acceptance records, or equipment defect elimination records; the newly added status quantity time series data may include status monitoring data continuously collected before and after the warning. When associating the newly added maintenance work order text and the newly added status quantity time series data, the equipment identifier and time information can continue to be used as the association basis to generate new integrated operation and maintenance data.

[0097] To achieve dynamic model adaptation, a sliding window buffer B can be established to store recently added integrated operation and maintenance data. Let the buffer size be M. Incremental training is triggered when the number of new samples reaches M, or when the preset update cycle has elapsed since the last model update. In this way, the risk prediction model can be continuously improved using new work orders and new status data.

[0098] The fault correlation graph can also be updated with new data. Specifically, when new fault modes, new abnormal state events, or new combined abnormal relationships appear in the newly added integrated operation and maintenance data, corresponding nodes can be added; when new samples further verify the relationship between existing state variable correlation features and fault modes, the weight of the corresponding correlation edge can be increased; when a warning is confirmed by manual review to be a false alarm, the weight of the corresponding correlation edge can be decreased or the correlation threshold can be adjusted.

[0099] In addition, the above method, after obtaining the fault risk results for the target power equipment within the predicted future time period, also includes: if the risk probability of any fault mode exceeds the corresponding warning threshold, then a warning message is generated. The warning message may include equipment identifier, equipment name, predicted fault mode, risk probability, risk level, risk source status quantity, prediction time period, trigger time, and associated work order number. The warning message can be sent to maintenance personnel via warning work orders, mobile messages, system pop-ups, or API push notifications.

[0100] When generating early warning information, the risk source state variables and associated fault modes can be determined based on the fault association graph. Specifically, state variable association features that are connected to the target fault mode and have high edge weights can be queried from the fault association graph and identified as the primary risk source. For example, for "transformer overheating risk," if the edge weights between "oil temperature rise rate," "high load rate operation," and "cooler abnormal semantics" in the graph and the fault mode are high, then these state variables and semantic contents are used as early warning interpretation information.

[0101] Furthermore, cosine similarity can be used to calculate similarity and match historical similar cases. The Top-K historical similar cases are selected based on their similarity scores, and the corresponding maintenance measures, handling results, and fault causes are used as the basis for generating maintenance recommendations. This is based on node proximity. Filter for highly correlated failure modes and calculate the similarity of historical cases. Recommended top-k cases, among which Let this be the current state vector. This is a vector of historical cases. The front-end interface embeds a feedback form, allowing users to rate the accuracy of alerts or the weight of rules. Labeling (e.g., confirmation / false alarm) and asynchronously writing feedback data to the database drive incremental model training and graph weight optimization, achieving a closed loop. This graph-based targeted recommendation significantly improves maintenance efficiency, reduces false alarm rates by more than 20%, and ensures decision-making accuracy. In terms of engineering, Docker containers are deployed to ensure high availability, and a logging system records all operations for auditing.

[0102] Maintenance recommendations can include inspection suggestions, testing suggestions, repair and handling suggestions, and continuous monitoring suggestions. For example, for temperature-related risks, recommendations could include infrared thermography verification, checking the cooler's operating status, and verifying load changes; for insulation-related risks, recommendations could include partial discharge detection, insulation resistance testing, or humidity environment investigation; and for mechanical risks, recommendations could include checking the transmission mechanism, energy storage mechanism, and switching coils. By combining fault correlation diagrams and historical similar cases, the specificity of maintenance recommendations can be improved.

[0103] Based on the same inventive concept, this application also provides a power equipment fault risk prediction system for implementing the power equipment fault risk prediction method described above.

[0104] The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more power equipment fault risk prediction system embodiments provided below can be found in the limitations of the power equipment fault risk prediction method described above, and will not be repeated here.

[0105] like Figure 3 As shown in the illustration, this application also provides a power equipment fault risk prediction system, which includes: The data fusion module 100 is used to acquire the maintenance work order text and status quantity time series data of the target power equipment, and to standardize the maintenance work order text and status quantity time series data to obtain fused operation and maintenance data associated with equipment identification and time information. The state quantity feature determination module 200 is used to perform fault semantic mining on the maintenance work order semantic information in the integrated operation and maintenance data, identify fault modes related to the target power equipment, and determine the state quantity features of the time period corresponding to the fault mode. The graph construction module 300 is used to perform time-series correlation analysis based on fault modes and state variable characteristics, and to construct a fault correlation graph to characterize the correlation between state variable changes and fault modes. The fault risk prediction module 400 is used to fuse and predict the current status information of the target power equipment and the maintenance work order text based on the fault association map, so as to obtain the fault risk result of the target power equipment in the future prediction period.

[0106] In some embodiments, the data fusion module 100 is used for: The maintenance work order text is cleaned, segmented, and filtered for stop words to obtain the effective word sequence of the work order. Convert the effective word sequence of the work order into work order semantic features; The time series data of state variables are processed by time alignment, denoising and normalization to obtain a standardized state variable sequence; Based on the equipment identification and the time information corresponding to the maintenance work order, the semantic features of the work order are associated with the standardized state quantity sequence to obtain the integrated operation and maintenance data.

[0107] In some embodiments, the state quantity characteristic determination module 200 is configured to: Input the semantic information of maintenance work orders from the integrated operation and maintenance data into the preset semantic recognition model to extract the contextual semantic features in the maintenance work order text; Based on attention weights, key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance procedures is determined from contextual semantic features. Fault mode classification is performed based on key semantic content to obtain the fault modes related to the target power equipment; Based on the maintenance work order time corresponding to the fault mode, determine the time window before the fault and the time window after the fault. Extract statistical features of state variables from the time series data of state variables within the time window before and after the fault.

[0108] In some embodiments, the system further includes an incremental update module, used for: Based on the identification confidence level of the failure mode, filter out the failure modes that meet the confidence level conditions; The selected fault modes, associated maintenance work order identifiers, associated equipment identifiers, and frequency of occurrence are written into the fault mode dictionary. When a new fault mode appears in the newly added maintenance work order text or a new semantic expression of an existing fault mode is added, the fault mode dictionary is updated incrementally.

[0109] In some embodiments, the map construction module 300 is used for: Based on the occurrence time or maintenance work order time corresponding to the failure mode, determine the correlation analysis time window corresponding to the failure mode; Within the correlation analysis time window, anomaly identification and statistical characterization are performed on the state variable characteristics to obtain state variable correlation characteristics that have a temporal correspondence with the fault mode. Based on the co-occurrence, chronological, and combined anomaly relationships between state variable correlation characteristics and fault modes within the correlation analysis time window, the correlation strength of state variable correlation characteristics to fault modes is determined. Using the correlation features between fault modes and state variables as graph nodes, and establishing weighted correlation edges between graph nodes based on the correlation strength, a fault correlation graph is obtained to characterize state variable changes, combined anomalies, and fault mode pointing relationships.

[0110] In some embodiments, the fault risk prediction module 400 includes: Temporal features are extracted from the current state information of the target power equipment to obtain the temporal features of the state. Semantic features are extracted from the maintenance work order text of the target power equipment to obtain the work order semantic features; By using fault association graphs to enhance the association between temporal features of state variables and semantic features of work orders, a fused risk feature is obtained. The risk characteristics are integrated and input into a preset risk prediction model, which then outputs the risk probability of different failure modes occurring in the target power equipment during the future prediction period. The failure risk level of the target power equipment is determined based on the risk probability.

[0111] In some embodiments, the system further includes: a model update module, used for: Receive new maintenance work order text and new status quantity timing data based on the results of fault risk; The newly added maintenance work order text and the newly added status quantity time series data are correlated to obtain the newly added integrated operation and maintenance data; The fault correlation graph and the preset risk prediction model used to obtain fault risk results are updated based on the newly added integrated operation and maintenance data.

[0112] like Figure 4 As shown, this application provides an electronic device 10, which includes a memory 20 and a processor 30. The memory 20 stores a computer program, and when the computer program is executed by the processor 30, the processor 30 performs the following: The maintenance work order text and status quantity time series data of the target power equipment are obtained, and the maintenance work order text and status quantity time series data are standardized to obtain fused operation and maintenance data associated with equipment identification and time information. Perform fault semantic mining on the semantic information of maintenance work orders in the integrated operation and maintenance data, identify fault modes related to the target power equipment, and determine the state quantity characteristics of the time period corresponding to the fault mode. Based on the fault mode and state variable characteristics, a time series correlation analysis is performed to construct a fault correlation map to characterize the correlation between state variable changes and fault modes. Based on the fault association map, the current status information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period.

[0113] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: The maintenance work order text is cleaned, segmented, and filtered for stop words to obtain the effective word sequence of the work order. Convert the effective word sequence of the work order into work order semantic features; The time series data of state variables are processed by time alignment, denoising and normalization to obtain a standardized state variable sequence; Based on the equipment identification and the time information corresponding to the maintenance work order, the semantic features of the work order are associated with the standardized state quantity sequence to obtain the integrated operation and maintenance data.

[0114] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: Input the semantic information of maintenance work orders from the integrated operation and maintenance data into the preset semantic recognition model to extract the contextual semantic features in the maintenance work order text; Based on attention weights, key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance procedures is determined from contextual semantic features. Fault mode classification is performed based on key semantic content to obtain the fault modes related to the target power equipment; Based on the maintenance work order time corresponding to the fault mode, determine the time window before the fault and the time window after the fault. Extract statistical features of state variables from the time series data of state variables within the time window before and after the fault.

[0115] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: Based on the identification confidence level of the failure mode, filter out the failure modes that meet the confidence level conditions; The selected fault modes, associated maintenance work order identifiers, associated equipment identifiers, and frequency of occurrence are written into the fault mode dictionary. When a new fault mode appears in the newly added maintenance work order text or a new semantic expression of an existing fault mode is added, the fault mode dictionary is updated incrementally.

[0116] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: Based on the occurrence time or maintenance work order time corresponding to the failure mode, determine the correlation analysis time window corresponding to the failure mode; Within the correlation analysis time window, anomaly identification and statistical characterization are performed on the state variable characteristics to obtain state variable correlation characteristics that have a temporal correspondence with the fault mode. Based on the co-occurrence, chronological, and combined anomaly relationships between state variable correlation characteristics and fault modes within the correlation analysis time window, the correlation strength of state variable correlation characteristics to fault modes is determined. Using the correlation features between fault modes and state variables as graph nodes, and establishing weighted correlation edges between graph nodes based on the correlation strength, a fault correlation graph is obtained to characterize state variable changes, combined anomalies, and fault mode pointing relationships.

[0117] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: Temporal features are extracted from the current state information of the target power equipment to obtain the temporal features of the state. Semantic features are extracted from the maintenance work order text of the target power equipment to obtain the work order semantic features; By using fault association graphs to enhance the association between temporal features of state variables and semantic features of work orders, a fused risk feature is obtained. The risk characteristics are integrated and input into a preset risk prediction model, which then outputs the risk probability of different failure modes occurring in the target power equipment during the future prediction period. The failure risk level of the target power equipment is determined based on the risk probability.

[0118] In some embodiments, when the computer program is executed by the processor 30, the processor 30 also performs: Receive new maintenance work order text and new status quantity timing data based on the results of fault risk; The newly added maintenance work order text and the newly added status quantity time series data are correlated to obtain the newly added integrated operation and maintenance data; The fault correlation graph and the preset risk prediction model used to obtain fault risk results are updated based on the newly added integrated operation and maintenance data.

[0119] This application provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements: The maintenance work order text and status quantity time series data of the target power equipment are obtained, and the maintenance work order text and status quantity time series data are standardized to obtain fused operation and maintenance data associated with equipment identification and time information. Perform fault semantic mining on the semantic information of maintenance work orders in the integrated operation and maintenance data, identify fault modes related to the target power equipment, and determine the state quantity characteristics of the time period corresponding to the fault mode. Based on the fault mode and state variable characteristics, a time series correlation analysis is performed to construct a fault correlation map to characterize the correlation between state variable changes and fault modes. Based on the fault association map, the current status information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period.

[0120] In some embodiments, the computer program, when executed, also implements: The maintenance work order text is cleaned, segmented, and filtered for stop words to obtain the effective word sequence of the work order. Convert the effective word sequence of the work order into work order semantic features; The time series data of state variables are processed by time alignment, denoising and normalization to obtain a standardized state variable sequence; Based on the equipment identification and the time information corresponding to the maintenance work order, the semantic features of the work order are associated with the standardized state quantity sequence to obtain the integrated operation and maintenance data.

[0121] In some embodiments, the computer program, when executed, also implements: Input the semantic information of maintenance work orders from the integrated operation and maintenance data into the preset semantic recognition model to extract the contextual semantic features in the maintenance work order text; Based on attention weights, key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance procedures is determined from contextual semantic features. Fault mode classification is performed based on key semantic content to obtain the fault modes related to the target power equipment; Based on the maintenance work order time corresponding to the fault mode, determine the time window before the fault and the time window after the fault. Extract statistical features of state variables from the time series data of state variables within the time window before and after the fault.

[0122] In some embodiments, the computer program, when executed, also implements: Based on the identification confidence level of the failure mode, filter out the failure modes that meet the confidence level conditions; The selected fault modes, associated maintenance work order identifiers, associated equipment identifiers, and frequency of occurrence are written into the fault mode dictionary. When a new fault mode appears in the newly added maintenance work order text or a new semantic expression of an existing fault mode is added, the fault mode dictionary is updated incrementally.

[0123] In some embodiments, the computer program, when executed, also implements: Based on the occurrence time or maintenance work order time corresponding to the failure mode, determine the correlation analysis time window corresponding to the failure mode; Within the correlation analysis time window, anomaly identification and statistical characterization are performed on the state variable characteristics to obtain state variable correlation characteristics that have a temporal correspondence with the fault mode. Based on the co-occurrence, chronological, and combined anomaly relationships between state variable correlation characteristics and fault modes within the correlation analysis time window, the correlation strength of state variable correlation characteristics to fault modes is determined. Using the correlation features between fault modes and state variables as graph nodes, and establishing weighted correlation edges between graph nodes based on the correlation strength, a fault correlation graph is obtained to characterize state variable changes, combined anomalies, and fault mode pointing relationships.

[0124] In some embodiments, the computer program, when executed, also implements: Temporal features are extracted from the current state information of the target power equipment to obtain the temporal features of the state. Semantic features are extracted from the maintenance work order text of the target power equipment to obtain the work order semantic features; By using fault association graphs to enhance the association between temporal features of state variables and semantic features of work orders, a fused risk feature is obtained. The risk characteristics are integrated and input into a preset risk prediction model, which then outputs the risk probability of different failure modes occurring in the target power equipment during the future prediction period. The failure risk level of the target power equipment is determined based on the risk probability.

[0125] In some embodiments, the computer program, when executed, also implements: Receive new maintenance work order text and new status quantity timing data based on the results of fault risk; The newly added maintenance work order text and the newly added status quantity time series data are correlated to obtain the newly added integrated operation and maintenance data; The fault correlation graph and the preset risk prediction model used to obtain fault risk results are updated based on the newly added integrated operation and maintenance data.

[0126] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, electronic devices, and computer storage media described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0127] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0128] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0129] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0130] In the embodiments provided by this invention, it should be understood that the disclosed systems, electronic devices, computer storage media, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0133] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the various embodiments of the present invention through a computer device (which may be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0134] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the failure risk of power equipment, characterized in that, The method includes: The maintenance work order text and status quantity time series data of the target power equipment are obtained, and the maintenance work order text and status quantity time series data are standardized to obtain fused operation and maintenance data associated with equipment identification and time information. Fault semantic mining is performed on the semantic information of maintenance work orders in the fused operation and maintenance data to identify fault modes related to the target power equipment and determine the state quantity characteristics of the time period corresponding to the fault mode. Based on the fault modes and the state variable characteristics, a time-series correlation analysis is performed to construct a fault correlation map to characterize the correlation between state variable changes and fault modes. Based on the fault association map, the current state information of the target power equipment and the maintenance work order text are fused and predicted to obtain the fault risk result of the target power equipment in the future prediction period.

2. The power equipment fault risk prediction method according to claim 1, characterized in that, The standardization process of the maintenance work order text and the status quantity time-series data yields fused operation and maintenance data associated with equipment identifiers and time information, including: The maintenance work order text is cleaned, segmented, and filtered for stop words to obtain a valid word sequence for the work order. Convert the effective word sequence of the work order into work order semantic features; The state quantity time series data is subjected to time alignment, denoising and normalization to obtain a standardized state quantity sequence; Based on the equipment identifier and the time information corresponding to the maintenance work order, the semantic features of the work order are associated with the standardized state quantity sequence to obtain the fused operation and maintenance data.

3. The power equipment fault risk prediction method according to claim 1, characterized in that, The step of performing fault semantic mining on the semantic information of maintenance work orders in the fused operation and maintenance data, identifying fault modes related to the target power equipment, and determining the state quantity features of the time period corresponding to the fault mode includes: The semantic information of the maintenance work order in the fused operation and maintenance data is input into the preset semantic recognition model to extract the contextual semantic features in the maintenance work order text; Based on attention weights, key semantic content related to equipment defects, abnormal phenomena, fault locations, fault causes, and maintenance procedures is determined from the contextual semantic features. Based on the key semantic content, fault mode classification is performed to obtain the fault modes related to the target power equipment; Based on the maintenance work order time corresponding to the fault mode, determine the time window before the fault and the time window after the fault. State quantity statistical features are extracted from the state quantity time series data within the pre-fault time window and the post-fault time window.

4. The power equipment fault risk prediction method according to claim 1, characterized in that, After identifying the fault modes associated with the target power equipment, the method further includes: Based on the identification confidence level of the fault modes, filter out the fault modes that meet the confidence level conditions; The selected fault modes, associated maintenance work order identifiers, associated equipment identifiers, and frequency of occurrence are written into the fault mode dictionary. When a new fault mode or a new semantic expression of an existing fault mode appears in a newly added maintenance work order text, the fault mode dictionary is incrementally updated.

5. The power equipment fault risk prediction method according to claim 1, characterized in that, The step of performing time-series correlation analysis based on the fault modes and the state variable characteristics to construct a fault correlation map characterizing the relationship between state variable changes and fault modes includes: Based on the occurrence time or maintenance work order time corresponding to the fault mode, determine the correlation analysis time window corresponding to the fault mode; Within the correlation analysis time window, anomaly identification and statistical characterization are performed on the state quantity features to obtain state quantity correlation features that have a time-series correspondence with the fault mode; Based on the co-occurrence, sequential, and combined anomaly relationships between the state variable correlation features and the fault modes within the correlation analysis time window, the correlation strength of the state variable correlation features to the fault modes is determined. Using the fault modes and the state variables as graph nodes, and establishing weighted association edges between graph nodes based on the association strength, a fault association graph is obtained to characterize state variable changes, combined anomalies, and fault mode pointing relationships.

6. The power equipment fault risk prediction method according to claim 1, characterized in that, The step of fusing and predicting the current state information and maintenance work order text of the target power equipment based on the fault association map to obtain the fault risk result of the target power equipment in the future prediction period includes: Temporal features are extracted from the current state information of the target power equipment to obtain the state quantity temporal features; Semantic features are extracted from the maintenance work order text of the target power equipment to obtain the work order semantic features; The fault association graph is used to enhance the association between the temporal features of the state variables and the semantic features of the work orders, thereby obtaining fused risk features. The fused risk characteristics are input into a preset risk prediction model, and the risk probability of the target power equipment experiencing different failure modes in the future prediction period is output through the risk prediction model. The fault risk level of the target power equipment is determined based on the risk probability.

7. The power equipment fault risk prediction method according to claim 1, characterized in that, The step of fusing and predicting the current state information and maintenance work order text of the target power equipment based on the fault association map to obtain the fault risk result of the target power equipment in the future prediction period, further includes: Receive new maintenance work order text and new status quantity timing data based on the aforementioned fault risk results; The newly added maintenance work order text and the newly added status quantity time series data are correlated to obtain the newly added integrated operation and maintenance data; The fault correlation graph and the preset risk prediction model used to obtain the fault risk results are updated based on the newly added integrated operation and maintenance data.

8. A power equipment fault risk prediction system, characterized in that, The system includes: The data fusion module is used to acquire the maintenance work order text and status quantity time series data of the target power equipment, and to standardize the maintenance work order text and the status quantity time series data to obtain fused operation and maintenance data associated with equipment identification and time information. The state quantity feature determination module is used to perform fault semantic mining on the maintenance work order semantic information in the fused operation and maintenance data, identify the fault modes related to the target power equipment, and determine the state quantity features of the time period corresponding to the fault mode. The graph construction module is used to perform time-series correlation analysis based on the fault modes and the state variable characteristics, and to construct a fault correlation graph to characterize the correlation between state variable changes and fault modes. The fault risk prediction module is used to fuse and predict the current state information of the target power equipment and the maintenance work order text based on the fault association map, so as to obtain the fault risk result of the target power equipment in the future prediction period.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the power equipment fault risk prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the steps of the power equipment fault risk prediction method as described in any one of claims 1-7.

Images