An agc abnormal operation identification method and system based on statistical and correlation analysis

Abstract

This invention discloses an AGC (Automatic Guided Vehicle) abnormal operation identification method and system based on statistical and correlation analysis. The method includes constructing an AGC control area encompassing thermal power, new energy, and energy storage; preprocessing historical operation data and enhancing abnormal samples through synonym replacement; performing fuzzy clustering on the operation data to generate standardized text; extracting features through semantic parsing, and combining rule-based correlation analysis and statistical correlation analysis to extract compliance and abnormal features, constructing an abnormal operation feature library; and collecting operation information in real time and performing feature matching: compliance with rules indicates normal operation, while abnormal operation triggers an alert and secondary confirmation. This invention overcomes the shortcomings of traditional fixed thresholds and static rules, effectively identifying human error operations, preventing the issuance of erroneous commands, and improving power grid frequency security.

Description

Technical Field

[0001] This invention belongs to the field of active power control technology in power systems, and particularly relates to an AGC abnormal operation identification method and system based on statistical and correlation analysis. Background Technology

[0002] Currently, Automatic Generation Control (AGC) systems have become an indispensable core component of power grid dispatch automation, widely used for active power regulation of various regulating resources such as conventional generating units, new energy power plants, and energy storage. With the construction of new power systems, new regulating resources such as new energy sources and energy storage are gradually being incorporated into the unified control scope of AGC due to their large quantity, large capacity, and fast regulation rate. The operations performed by operators through the AGC interface are becoming increasingly frequent, and their impact on power grid frequency stability and power balance security is becoming increasingly prominent.

[0003] Existing technologies for identifying AGC (Automatic Guided Vehicle) operational anomalies mainly fall into two categories: data analysis-based methods and rule engine-based methods. Data analysis-based methods determine whether real-time data exceeds normal fluctuation ranges by setting statistical thresholds (such as rate of change or standard deviation) or performing correlation analysis. Rule engine-based methods rely on expert experience to formulate operational rules and constraints (such as power change rate limits and adjustment boundaries). The system compares operational parameters with these rules in real time, and violations of the rules are identified as anomalies.

[0004] However, existing technologies still have significant shortcomings in practical applications: data analysis-based methods often use fixed thresholds, which are difficult to adapt to the dynamic characteristics of new regulation resources that are "numerous and have a fast regulation rate," and are prone to false alarms or missed alarms; rule engine-based methods rely heavily on static expert experience, lack statistical learning of daily correct operating behaviors and correlation analysis of operating procedures, and cannot form a refined and dynamically updatable compliance feature library for different regulation resources. They are also unable to effectively identify human error settings or malicious misoperations, thus failing to prevent rapid fluctuations in grid power caused by the issuance of erroneous targets, which affects the safety of grid frequency operation. Summary of the Invention

[0005] Purpose of the Invention: The purpose of this invention is to provide an AGC (Automatic Guided Vehicle) anomaly operation identification method and system based on statistical and correlation analysis. This aims to address the problems of existing AGC operation anomaly identification methods, which rely on fixed thresholds and static rules, lack statistical learning of daily correct operating behaviors and correlation analysis with operating procedures, making them difficult to adapt to the dynamic characteristics of new regulatory resources, prone to false alarms and missed alarms, and unable to effectively identify human error. The invention achieves rapid and accurate identification of AGC interface operations and real-time feedback of compliance results, thereby preventing power grid frequency fluctuations caused by erroneous target assignments and improving the safety level of power grid operation.

[0006] Technical solution: The AGC abnormal operation identification method based on statistical and correlation analysis described in this invention includes the following steps:

[0007] S1. Construct an AGC control zone to automatically control the power generation of conventional thermal power units, centralized new energy power plants, distributed new energy power plants, and energy storage regulation resources within the control range;

[0008] S2. Obtain historical AGC interface operation data from dispatchers and automated operation and maintenance personnel, and preprocess the historical AGC interface operation data. The preprocessing includes duplicate record cleaning, missing data processing, abnormal data removal, data reconstruction, and multi-source heterogeneous data synchronization processing.

[0009] S3. Based on the preprocessed historical AGC interface operation data, perform abnormal sample enhancement. Generate extended abnormal operation samples through synonym replacement, random swapping, random insertion and random deletion to construct an abnormal operation dataset.

[0010] S4. The enhanced AGC interface operation data is blurred, and a clustering algorithm based on similarity comparison is used to cluster the operation content. The station name, keyword number and operation object identification information are replaced with the corresponding clustering labels to generate standardized operation text.

[0011] S5. Perform semantic analysis on the standardized operation text, and extract the operation semantic features through text segmentation and stop word removal.

[0012] S6. Based on the semantic parsing results, perform rule association analysis and statistical association analysis, wherein the rule association analysis is used to perform rule matching on interface operations according to preset operation procedures, and the statistical association analysis is used to extract statistical association relationships between different operation behaviors;

[0013] S7. Based on the results of rule-based correlation analysis and statistical correlation analysis, extract the compliance and abnormal features of AGC interface operations, and establish a compliance / abnormal feature model.

[0014] S8. Construct an abnormal operation feature library based on the compliance features and abnormal features. The abnormal operation feature library includes the correlation between interface operations, error factors and phenomenon descriptions.

[0015] S9. Obtain real-time AGC interface operation information, perform semantic parsing, logical relationship analysis and feature extraction on the real-time AGC interface operation information, and match and identify the extracted real-time operation features with the compliance features and abnormal features in the abnormal operation feature library.

[0016] S10. When the real-time operation feature meets the compliance feature conditions, output a normal result. When the real-time operation feature meets the abnormal feature conditions, output an abnormal warning message and send an abnormal warning signal to the AGC control module to suspend the corresponding control operation and trigger a secondary confirmation.

[0017] This technical solution effectively alleviates the problem of difficulty in modeling based on fixed thresholds due to the complex dynamic characteristics of new regulatory resources by constructing an AGC control area (S1) containing multiple types of regulatory resources and systematically processing historical operation data (S2). Combined with anomaly sample enhancement (S3) and fuzzy clustering standardization (S4), it further extracts operational features through semantic parsing (S5) and integrates rule-based correlation analysis and statistical correlation analysis (S6). This allows for the simultaneous extraction of compliance and anomaly features based on the statistical relationship between procedure matching and operational behavior (S7), overcoming the underreporting and false alarms caused by traditional methods that rely solely on static rules and lack statistical learning of normal operations. The established anomaly operation feature library (S8) supports rapid semantic parsing and feature matching identification of real-time operations (S9). When anomalies are detected, timely warnings are triggered and operations are suspended (S10), helping to reduce erroneous target assignments caused by human error, mitigating the risk of power grid frequency fluctuations, and improving the identification accuracy of AGC interface operations and the safety of power grid operation.

[0018] Preferably, the preprocessing in step S2 includes:

[0019] The system detects duplicate and missing records in the historical AGC interface operation data, removes duplicate records, and completes and marks missing records.

[0020] The system identifies explicit abnormal data in the historical AGC interface operation data through range checks and implicit abnormal data in the historical AGC interface operation data through correlation checks.

[0021] Data reconstruction is performed on the data gaps formed after removing missing records, explicit abnormal data, and hidden abnormal data;

[0022] Based on a preset synchronization cycle, historical AGC interface operation data from different sources are processed for time alignment and frequency synchronization.

[0023] The preferred step S2, by removing duplicate records and completing and marking missing records in the historical AGC interface operation data, helps improve the completeness and reliability of the dataset. Based on this, combining range checks to identify explicit anomalies and correlation checks to identify hidden anomalies allows for a more comprehensive discovery of various forms of data quality problems, mitigating the potential omission risks of traditional single-anomaly detection methods. Reconstructing the data gaps created after removal reduces the impact of data breakpoints on the continuity of subsequent analysis. Simultaneously, by using a preset synchronization period to perform time alignment and frequency synchronization processing on multi-source heterogeneous data, it helps reduce analytical biases introduced by asynchronous data sources, providing a more standardized and reliable data foundation for subsequent anomaly sample enhancement and feature extraction.

[0024] Preferably, the abnormal sample enhancement in step S3 includes:

[0025] Identify abnormal operation data from the historical AGC interface operation data after preprocessing in step S2;

[0026] Randomly select target words from the abnormal operation data and generate new abnormal samples using synonym replacement.

[0027] Randomly swap the word positions in the anomalous operation data to generate order perturbation samples;

[0028] Insert synonyms of random words at any position in the abnormal operation data to generate inserted perturbation samples;

[0029] Randomly delete some words from the abnormal operation data to generate a censored perturbation sample.

[0030] The preferred step S3 identifies anomalous operation data from the preprocessed data and, based on this, employs various enhancement methods such as synonym replacement, random word position swapping, random synonym insertion, and random deletion of some words to generate diverse extended anomalous operation samples. This helps enrich the coverage of anomalous operation patterns, alleviates the limitations of insufficient or unevenly distributed original anomalous samples on model training and feature extraction, and improves the adaptability of subsequent clustering analysis and semantic parsing to anomalous operation variants, thereby enhancing the robustness and generalization performance of the anomaly identification method.

[0031] Preferably, the blurring process in step S4 includes:

[0032] Select operation A to be clustered;

[0033] Randomly select multiple operation contents B from existing clusters. k ;

[0034] The following formula is used to calculate operation content A and operation content B. kSimilarity between them:

[0035]

[0036] Where S represents the similarity, and C represents the similarity between operation content A and operation content B. k The number of identical characters in the operation content C, l(C) is the number of characters in the operation content A, l(B) is the number of characters in the operation content A, and l(C) is the number of characters in the operation content A. k ) represents operation content B k The number of characters;

[0037] When operation content A is combined with multiple operation contents B k If the similarity between the elements is greater than a preset threshold, the operation content A will be assigned to a cluster that meets the preset threshold condition; otherwise, a new cluster will be created.

[0038] The preferred step S4 involves selecting the operation content A to be clustered, and then considering multiple operation contents B from the existing clusters. k This method employs a similarity calculation formula that considers both the number of identical characters and the total number of characters in each instance for fuzzy clustering. It determines the similarity between A and multiple Bs. k The similarity threshold used to determine whether an operation is assigned to an existing cluster or created helps alleviate the problem of traditional hard clustering being sensitive to subtle differences in the operation text, leading to rigid classification. This fuzzy clustering process based on similarity comparison can replace information such as site names, keyword numbers, and operation object identifiers with unified cluster labels, generating more standardized operation text. This reduces the interference of redundant diversity in operation descriptions on subsequent semantic parsing and improves the consistency of anomaly identification.

[0039] Preferably, the semantic parsing in step S5 includes:

[0040] The standardized operational text after fuzzification in step S4 is segmented to obtain multiple semantic units;

[0041] Extract the operation object, control target, operation time, and operation type information from the semantic unit;

[0042] The status words, punctuation marks, and invalid descriptive words in the standardized operation text are removed to obtain the operation text after removing stop words;

[0043] An operational semantic feature vector is generated based on the operational text after removing stop words.

[0044] The preferred step S5 involves text segmentation of the standardized operation text after fuzzification to obtain semantic units, and extracting key information such as the operation object, control target, operation time, and operation type from them. This helps transform unstructured operation descriptions into structured semantic elements. Further processing to remove status words, punctuation marks, and invalid descriptive words effectively simplifies redundant information in the operation text and reduces the interference of irrelevant words on understanding the operation intent. Based on this, an operation semantic feature vector is generated, providing a clearer and more compact feature representation for subsequent rule-based association analysis and statistical association analysis, which is beneficial for improving the accuracy and consistency of operation behavior identification.

[0045] Preferably, the rule association analysis in step S6 includes:

[0046] A set of interface operation rules was established based on the power industry's operating procedures, AGC operating specifications, and the company's internal operating regulations.

[0047] The set of interface operation rules is classified according to operation type, operation step, and controlled object;

[0048] The real-time interface operation information is matched item by item with the corresponding category of interface operation rules;

[0049] If any interface operation does not meet the corresponding interface operation rules, the interface operation will be judged as an abnormal operation.

[0050] The statistical association analysis includes:

[0051] Perform statistical analysis on the vector of operation semantic features obtained in step S5 to extract the statistical patterns among operation frequency, operation time period, operation type and control target in historical AGC interface operation data;

[0052] Establish an operation association model based on the correlation between different contents in historical AGC interface operation data;

[0053] The deviation between real-time interface operation information and historical statistical patterns is analyzed based on the operation correlation model.

[0054] When the real-time interface operation information deviates from the historical statistical pattern by more than a preset threshold, the interface operation is judged as an abnormal operation.

[0055] In the preferred step S6, rule-based correlation analysis establishes an operational rule set based on power industry operating procedures, AGC operating specifications, and internal enterprise operating regulations. This set is then categorized and matched according to operation type, stage, and controlled object. This allows for the timely detection and identification of interface operation stages that do not meet the rules, helping to alleviate the problem of relying solely on fixed thresholds being insufficient to adapt to procedural logic. Statistical correlation analysis, on the other hand, performs statistical analysis on the semantic feature vectors of operations, extracting historical statistical patterns between operation frequency, time period, type, and control objectives. It then establishes an operational correlation model, identifying abnormal operations based on the degree of deviation between real-time operations and historical statistical patterns. The combination of these two analysis methods takes into account both rigid procedural constraints and daily operational habits, reducing missed or false alarms caused by relying on a single rule and improving the ability to identify human error.

[0056] Preferably, in step S7, a logical language is used to represent compliance features and abnormal features, wherein the abnormal features are represented as follows:

[0057] ((Number of operations and operation types ∈ Historical statistical patterns) ∧ (Operation types ∈ Sensitive operation types) ∧ (Operation time period ∈ / Regular working hours));

[0058] Used to determine whether there is a risk of abnormal AGC operation in the interface operation.

[0059] The preferred step S7 uses logical language to represent compliance and abnormal features. Abnormal features are characterized by combinations of conditions such as "the number of personnel operations and the operation type conform to historical statistical patterns," "the operation type belongs to a sensitive operation type," and "the operation time is outside of regular working hours." This helps to transform multi-dimensional judgment criteria into structured logical expressions. This logical language-based representation provides a clear, rule-based description for judging the risk of abnormal operations on the AGC interface, alleviating the problem of inconsistent abnormal identification standards caused by ambiguous feature expressions, improving the ability to perceive potential risks when sensitive operations occur outside of normal times, and providing clearer feature criteria for the subsequent construction of an abnormal operation feature library and real-time matching and identification.

[0060] Preferably, the abnormal operation feature library in step S8 is constructed according to the AGC operation process, including:

[0061] The AGC interface operation is divided into operation interface, operator, operation time, controlled object, control target, operation process and operation data verification sub-process;

[0062] Establish a relationship structure between interface operation, error factors, and phenomenon description for each sub-process.

[0063] For each abnormal operation, five types of feature elements are established: natural language expression, logical language description, feature mechanism, feature parameters, and abnormal label.

[0064] The preferred step S8 divides the interface operation into sub-processes according to the AGC operation flow, including the operation interface, operator, operation time, controlled object, control target, operation process, and operation data verification. For each sub-process, a correlation structure of "interface operation—error factor—phenomenon description" is established, which helps to correlate anomaly identification with specific operation steps and alleviates the problem of difficulty in locating anomalies from different stages. Simultaneously, five types of feature elements are constructed for each anomalous operation: natural language expression, logical language description, feature mechanism, feature parameters, and anomaly label. This allows for the characterization of anomalous behavior patterns from multiple dimensions, enriching the information hierarchy of the anomalous operation feature database. This process-based decomposition and multi-dimensional feature association construction method provides a more systematic and structured reference for real-time operation feature matching and identification, improving the ability to locate and interpret anomalous operations.

[0065] Preferably, step S9 includes:

[0066] Get the real-time AGC interface operation text;

[0067] Semantic parsing and logical relationship analysis are performed on the real-time AGC interface operation text;

[0068] Extract real-time operation features and match them with compliance and abnormal features in the abnormal operation feature library;

[0069] The logical expression (¬((Modified object ⋀ Control target) → Station power change) → Power grid frequency change) is used to perform logical reasoning on the control panel;

[0070] When the matching result meets the abnormal characteristic conditions, an abnormal warning result is generated and a warning signal is sent to the AGC control module.

[0071] The preferred step S9 involves semantic parsing and logical relationship analysis of the real-time AGC interface operation text to extract real-time operation features and match them with compliance and abnormal features in the abnormal operation feature library. This enables rapid comparison and identification of the current operation behavior. Based on this, the logical expression (¬((Modified Object ⋀ Control Target) → Station Power Change) → Grid Frequency Change) is used to reason about the operation console, helping to reveal the implicit logical relationship between the operation intent, the controlled object, and the system response. This mitigates misjudgments caused by directly comparing the operation appearance and ignoring the deep causal chain. When the matching result meets the abnormal feature conditions, an abnormal warning is generated and a signal is sent to the AGC control module, providing timely evidence for subsequent operation suspension and secondary confirmation, further enhancing the security protection capabilities of the AGC interface operation.

[0072] The AGC abnormal operation identification system based on statistical and correlation analysis described in this invention includes:

[0073] The data acquisition module is used to acquire historical AGC interface operation data and real-time AGC interface operation data.

[0074] The data preprocessing module is used to perform data cleaning, abnormal data removal, data reconstruction, and asynchronous data synchronization on historical AGC interface operation data.

[0075] The data augmentation module is used to perform synonym replacement, random swapping, random insertion, and random deletion on abnormal operation samples to generate augmented abnormal samples.

[0076] The data blurring module is used to cluster the operation content using a clustering algorithm based on similarity comparison and generate cluster labels;

[0077] The semantic parsing module is used to perform word segmentation, stop word removal, and semantic feature extraction on the AGC interface operation text.

[0078] The correlation analysis module is used to conduct rule-based correlation analysis and statistical correlation analysis, and to extract compliance features and anomaly features;

[0079] The feature library construction module is used to build an abnormal operation feature library;

[0080] The anomaly identification module is used to match real-time operation features with an anomaly operation feature library and output the anomaly identification results.

[0081] The control feedback module is used to send an abnormal warning signal to the AGC control module when an abnormal operation is detected, and to suspend the corresponding AGC control operation.

[0082] Beneficial Effects: Compared with the prior art, the present invention has the following significant advantages: 1. The present invention adopts a method combining rule-based correlation analysis and statistical correlation analysis to extract compliance and abnormal features of AGC interface operations and establish a feature library. This overcomes the shortcomings of traditional fixed thresholds and static rules in adapting to the dynamic characteristics of new regulatory resources, effectively solving the problems of missed and false alarms and the inability to identify human error operations. It achieves rapid and accurate identification and real-time compliance feedback of AGC interface operations, thereby preventing power grid frequency fluctuations caused by erroneous target issuance and significantly improving the safety level of power grid operation; 2. The present invention uses abnormal sample enhancement and fuzzy clustering processing, utilizing synonym replacement and randomization... Extended abnormal samples are generated through methods such as swapping, random insertion, and random deletion. Standardized operation texts are generated based on similarity comparison clustering, which significantly improves the model's ability to identify rare and atypical abnormal operations and effectively reduces the false alarm rate and false alarm rate caused by insufficient samples. 3. This invention uses semantic parsing to extract semantic features of operations and combines logical expressions to perform operation rationality reasoning, constructing an abnormal operation feature library that includes interface operation - error factors - phenomenon description. It supports real-time operation feature matching and abnormal early warning, and suspends abnormal operations and triggers secondary confirmation by sending early warning signals to the AGC control module, forming a closed-loop control mechanism, which effectively ensures the stable operation of the power grid frequency. Attached Figure Description

[0083] Figure 1 This is a flowchart of the AGC abnormal operation identification process of the present invention. Detailed Implementation

[0084] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0085] This invention provides a method for identifying AGC (Automatic Guided Vehicle) anomaly operations based on statistical and correlation analysis, such as... Figure 1 As shown, it includes the following steps:

[0086] (1) In response to the frequency and tie-line control requirements of high-proportion new energy power grids, an AGC control area is generally established in the power grid dispatch control center (hereinafter referred to as the dispatch center). It is mainly responsible for implementing closed-loop control of conventional thermal power units, centralized and distributed new energy power plants, and various types of energy storage and other new regulation resources within the dispatch control area.

[0087] (2) Before conducting AGC operation anomaly identification, the historical operation information of the scheduling and automated operation and maintenance personnel should be preprocessed, that is, the operation data should be cleaned, i.e.:

[0088] (21) Detect duplicate and missing records in the original operation information dataset, remove duplicate records, add missing records, and mark missing data.

[0089] (22) Identify and remove abnormal data in the operation dataset. Part of the process involves preliminary screening and removal of obvious abnormal data through range checks, while the other part involves fine-tuning and removal of more hidden abnormal data through correlation checks.

[0090] (23) Reconstruct the original missing data and the missing data caused by the removal of anomalies.

[0091] (24) If there is a data asynchrony problem, determine the synchronization period according to the usage requirements, and synchronize the multi-source heterogeneous data to achieve consistency in frequency and time.

[0092] (3) When scheduling and automated operation and maintenance personnel perform interface operations, erroneous operations are extremely rare. To build an erroneous operation feature library, it is necessary to augment the scheduling interface operation dataset to increase the proportion of abnormal operations in the scheduling operation dataset. The augmentation method is:

[0093] (31) Synonym replacement method: randomly select a word from each abnormal operation data and replace the word with a randomly selected synonym.

[0094] (32) Random swap: Randomly select two words from each abnormal operation data and swap their positions.

[0095] (33) Random insertion: Find a random word from each abnormal operation data. Insert a random synonym of a random word at any position in the sentence.

[0096] (34) Random deletion: Randomly delete one word from each abnormal operation data.

[0097] (4) Based on the data cleaning and enhancement, considering the redundant information such as station names and keyword numbers in the operation content of the scheduling interface, it is necessary to blur this redundant information to improve the accuracy of feature extraction. The data blurring method is based on the clustering algorithm of similarity comparison, which divides the operation content into multiple clusters and gives a label for each cluster. The operation content is then replaced with the label of the cluster. The specific process is as follows:

[0098] (41) Select an operation to be clustered and name it A.

[0099] (42) Randomly select a cluster and randomly select multiple (e.g., 5) operations from it, naming them B{B1;B2;…; Bk;…;B5}. Here, k is the index variable, and Bk is the k-th operation. If there are fewer than 5 operations in the cluster, all operations will be selected. If there is no cluster, a new cluster will be created directly, and the operations will be placed in that cluster.

[0100] (43) Use the following formula to calculate the similarity between operation content A and operation content Bk.

[0101]

[0102] Where S is the similarity score. l(x) is the number of characters in the operation content x. C is the number of common characters in A and Bk.

[0103] (44) If the similarity between all randomly selected operation content Bk and operation content A in the cluster is greater than the preset threshold, then operation content A will be placed in this cluster. If none of the clusters meet the threshold, then a new cluster will be created and operation content A will be placed in this cluster.

[0104] (45) Repeat steps (41) to (44) for each operation to complete the clustering.

[0105] (5) After cleaning, enhancing and blurring the operation data of the scheduling interface, the next step is to carry out semantic analysis of the operation data information. Semantic analysis is carried out by word segmentation and removal of stop words. Word segmentation is taken into consideration that the scheduling operation is mostly stored as standardized information. Word segmentation can quickly locate the semantics of the operation. Removing stop words is used to remove words that are invalid for data analysis from the text data, such as success, failure, period, semicolon and other information.

[0106] (6) The next step is to conduct rule-based and statistical correlation analysis based on the processed historical operation records:

[0107] (61) Rule-based correlation analysis matches actual interface operations with these established rules. The interface operation rules and procedures are categorized and organized. When analyzing actual interface operations, the corresponding rule categories are selected based on the type and stage of the operation for comparison. Any stage that does not conform to the rules will be marked as an abnormal operation. By checking the operation data according to the rules formulated based on power industry standards and internal company regulations, quantitative analysis can systematically check whether the power system operation complies with regulations, providing a basis for AGC interface operation and risk control.

[0108] (62) Statistical correlation analysis: By conducting in-depth statistical analysis on a large amount of power data, we can explore the inherent statistical characteristics and patterns of the data. At the same time, by analyzing the correlation between different operational contents, we may discover some potential correlations.

[0109] (7) Based on the correlation results of rule-based and statistical information processing, the next step is to extract operational compliance / abnormal features. The rule-based method relies on pre-defined clear rules to judge the compliance or abnormality of features, which has strong logic and certainty. The statistical method, on the other hand, discovers potential compliance / abnormal features from the distribution and trends of data through statistical analysis of a large amount of data.

[0110] (71) Based on the differences in information sources, it is divided into two categories: single information source analysis and multi-source information correlation analysis. Single information source analysis only involves information from a specific channel or system, while multi-source information correlation analysis focuses on integrating information from multiple different sources;

[0111] (72) Use logical language to represent compliance / abnormal results, and represent the compliance / abnormal features extracted from AGC interface operations as: ((number of times, type of personnel operation ∉ historical statistical pattern) ⋀ (operation type ∈ sensitive operation type) ⋀ (operation time period ∉ regular working time)), and determine whether the interface operation meets the characteristics of having AGC operation abnormal risk;

[0112] (73) After organizing the variable relationships through association analysis, the compliance and abnormal characteristics of the operation are obtained. The next step is to carry out feature comparison. The characteristics of the operation that occurs in real time are compared with the compliance and abnormal characteristics extracted through association analysis in the early stage. If the operation characteristics are highly consistent with the compliance characteristics, it can be judged that the operation is reasonable and in compliance with the norms. Otherwise, if the operation characteristics are consistent with the abnormal characteristics or there is a large deviation, it can be promptly indicated that the operation may be abnormal and needs to be confirmed again.

[0113] (8) To improve the real-time performance of AGC interface operation anomaly identification, it is necessary to construct an anomaly operation feature library. Based on the AGC interface operation process, the AGC interface operation is divided into sub-processes such as operation interface, operator, operation time, control object, control target, operation process, and operation data verification. Then, based on the relevant processes, the anomaly operation feature library is constructed with the structure of "interface operation - error factor - phenomenon description". For each specific process, five elements are designed: natural language expression, logical language description, feature mechanism, feature parameters, and anomaly label. The anomaly instruction feature library consists of these five elements forming a complete system.

[0114] (9) Based on the abnormal operation feature library constructed above, conduct AGC abnormal operation identification. The specific method is as follows:

[0115] (91) The AGC system uses the operation functions provided by the platform to obtain the operation information of the dispatcher / automation personnel;

[0116] (92) Perform comprehensive semantic analysis based on the AGC interface operation information, obtain the operation steps included in this process, perform text segmentation, divide it into different semantic units, and filter out key words from the segmented text;

[0117] (93) Then, semantic understanding is performed to analyze the meaning of each word and sentence in depth and understand the specific operational intent described by the interface operation.

[0118] (94) Then conduct logical relationship analysis, sort out the sequence and logical connection between the operation steps, so as to fully extract the core content of the operation;

[0119] (95) Compare the extracted interface operation features with the compliance / abnormal feature library (¬((Modified Object ⋀ Control Target) → Station Power Change) → Power Grid Frequency Change) from step 8. Use rigorous mathematical logic to determine the rationality of the operation. During the reasoning process, carefully analyze whether each step and condition in the operation meets the requirements of the compliance features.

[0120] (96) If the identified operation meets the compliance feature requirements, it will return to normal. Once an abnormal operation is detected, the feature library will trigger the abnormal judgment mechanism, mark the current operation as an abnormal operation, and send a warning signal to the AGC control module.

[0121] (10) After generating the operation information text, the AGC control module waits for the feedback result of the anomaly identification module in real time. If an operation anomaly warning is received, the operation will be paused and the dispatcher will be prompted to confirm again to avoid issuing abnormal instructions due to abnormal operation and to ensure the safety of AGC control.

[0122] To address the frequency and tie-line control requirements of a high-proportion renewable energy power grid, an AGC control area is established in the dispatch center. This area is primarily responsible for implementing closed-loop control of conventional thermal power units, centralized and distributed renewable energy power plants, and various types of energy storage and other new regulatory resources within the dispatch area.

[0123] First, preprocessing is performed based on historical operation information from scheduling and automated operation and maintenance personnel. This involves cleaning the operation data, removing duplicate records, supplementing missing records, marking missing data, identifying and removing abnormal data in the operation dataset, including data reconstruction of the original missing data and the missing data caused by abnormal removal.

[0124] Then, an error operation feature library is constructed by augmenting the scheduling interface operation dataset. Augmentation methods include synonym replacement, random swapping, random insertion, and random deletion.

[0125] After data cleaning and enhancement, considering the redundant information such as station names and keyword numbers in the dispatch interface operation content, the next step is to blur this redundant information to improve the accuracy of feature extraction. The data blurring method is based on a similarity comparison clustering algorithm, which divides the operation content into multiple clusters and gives a label for each cluster. The operation content is then replaced with the label of that cluster.

[0126] After cleaning, enhancing, and obfuscating the operation data of the scheduling interface, the next step is to conduct semantic analysis of the operation data information. Semantic analysis is carried out by word segmentation and removal of stop words. Word segmentation is taken into account that scheduling operations are mostly stored as standardized information, and word segmentation can quickly locate the semantics of the operation. Removal of stop words is used to remove words from the text data that are invalid for data analysis.

[0127] Based on the processed historical operation records, the next step is to conduct rule-based and statistical correlation analysis, including rule-based correlation analysis and statistical correlation analysis. These two types of analysis match actual interface operations with these established rules, and through in-depth statistical analysis of large amounts of electricity data, uncover the inherent statistical characteristics and patterns of the data. Furthermore, by analyzing the correlations between different operation contents, some potential relationships may be discovered.

[0128] Based on the correlation results obtained from rule-based and statistical information processing, the next step is to extract operational compliance / abnormal features. The rule-based method relies on pre-defined, explicit rules to determine the compliance or anomalousness of features, exhibiting strong logic and certainty. The statistical method, on the other hand, performs statistical analysis on large amounts of data to discover potential compliance / abnormal features from aspects such as data distribution and trends.

[0129] First, the compliance / abnormal results are represented in the form of logical language. The compliance / abnormal features extracted from the AGC interface operation are represented as: ((number of times and type of personnel operation ∉ historical statistical pattern) ⋀ (operation type ∈ sensitive operation type) ⋀ (operation time period ∉ regular working time)). Then, it is determined whether the interface operation meets the characteristics of having the risk of AGC operation abnormality.

[0130] Based on the compliance and abnormal characteristics of operations, the characteristics of real-time operations are compared one by one with the compliance and abnormal characteristics extracted in the early stage through correlation analysis. If the operation characteristics are highly consistent with the compliance characteristics, it can be judged that the operation is reasonable and in compliance with the regulations; otherwise, if the operation characteristics are consistent with the abnormal characteristics or there is a large deviation, it can be promptly indicated that the operation may be abnormal and requires secondary confirmation.

[0131] To improve the real-time performance of AGC interface operation anomaly identification, it is necessary to construct an anomaly operation feature library. Based on the AGC interface operation flow, the AGC interface operation is divided into sub-flows such as operation interface, operator, operation time, controlled object, control target, operation process, and operation data verification. Then, based on the relevant flows, the anomaly operation feature library is structured as "interface operation - error factor - phenomenon description". For each specific flow, five elements are designed: natural language expression, logical language description, feature mechanism, feature parameters, and anomaly label. The anomaly command feature library consists of these five elements forming a complete system.

[0132] Then, based on the previously constructed abnormal operation feature library, AGC abnormal operation identification is carried out. The steps are as follows: the AGC system uses the operation functions provided by the platform to obtain the operation information of the dispatching / automation personnel; then, a comprehensive semantic analysis is performed on the AGC interface operation information, obtaining the operation steps included in this process, and performing text segmentation to divide it into different semantic units. Key words are selected from the segmented text, and then semantic understanding is performed to deeply analyze the meaning of each word and sentence, understand the specific operation intention described by the interface operation, and then perform logical relationship analysis to sort out the sequence and logical connection between operation steps, thereby completely extracting the core content of the operation. Finally, the extracted interface operation features are compared with the compliance / abnormal feature library in step 8 (¬((Modified object ⋀ Control target) → Station power change) → Power grid frequency change). The rationality of the operation is judged through rigorous mathematical logic reasoning. In the reasoning process, each step and condition in the operation is carefully analyzed to see if it meets the requirements of compliance features.

[0133] If the identified operation meets the compliance requirements, it returns to normal. Once an abnormal operation is detected, the feature database triggers the anomaly detection mechanism, marking the current operation as abnormal and sending a warning signal to the AGC control module. After generating the operation information text, the AGC control module waits for feedback from the anomaly detection module in real time. If it receives an anomaly warning, it will pause the operation and prompt the dispatcher to confirm again to avoid issuing abnormal commands due to abnormal operations and ensure AGC control safety.

[0134] Based on a similar inventive concept, embodiments of the present invention also provide an AGC abnormal operation identification system based on statistical and correlation analysis, corresponding to the AGC abnormal operation identification method based on statistical and correlation analysis, comprising:

[0135] The data acquisition module is used to acquire historical AGC interface operation data and real-time AGC interface operation data.

[0136] The data preprocessing module is used to perform data cleaning, abnormal data removal, data reconstruction, and asynchronous data synchronization on historical AGC interface operation data.

[0137] The data augmentation module is used to perform synonym replacement, random swapping, random insertion, and random deletion on abnormal operation samples to generate augmented abnormal samples.

[0138] The data blurring module is used to cluster the operation content using a clustering algorithm based on similarity comparison and generate cluster labels;

[0139] The semantic parsing module is used to perform word segmentation, stop word removal, and semantic feature extraction on the AGC interface operation text.

[0140] The correlation analysis module is used to conduct rule-based correlation analysis and statistical correlation analysis, and to extract compliance features and anomaly features;

[0141] The feature library construction module is used to build an abnormal operation feature library;

[0142] The anomaly identification module is used to match real-time operation features with an anomaly operation feature library and output the anomaly identification results.

[0143] The control feedback module is used to send an abnormal warning signal to the AGC control module when an abnormal operation is detected, and to suspend the corresponding AGC control operation.

Claims

1. A method for identifying anomalous AGC operations based on statistical and correlation analysis, characterized in that, Includes the following steps: S1. Construct an AGC control zone to automatically control the power generation of conventional thermal power units, centralized new energy power plants, distributed new energy power plants, and energy storage regulation resources within the control range; S2. Obtain historical AGC interface operation data from dispatchers and automated operation and maintenance personnel, and preprocess the historical AGC interface operation data. The preprocessing includes duplicate record cleaning, missing data processing, abnormal data removal, data reconstruction, and multi-source heterogeneous data synchronization processing. S3. Based on the preprocessed historical AGC interface operation data, perform abnormal sample enhancement. Generate extended abnormal operation samples through synonym replacement, random swapping, random insertion and random deletion to construct an abnormal operation dataset. S4. The enhanced AGC interface operation data is blurred, and a clustering algorithm based on similarity comparison is used to cluster the operation content. The station name, keyword number and operation object identification information are replaced with the corresponding clustering labels to generate standardized operation text. S5. Perform semantic analysis on the standardized operation text, and extract the operation semantic features through text segmentation and stop word removal. S6. Based on the semantic parsing results, perform rule association analysis and statistical association analysis, wherein the rule association analysis is used to perform rule matching on interface operations according to preset operation procedures, and the statistical association analysis is used to extract statistical association relationships between different operation behaviors; S7. Based on the results of rule-based correlation analysis and statistical correlation analysis, extract the compliance and abnormal features of AGC interface operations, and establish a compliance / abnormal feature model. S8. Construct an abnormal operation feature library based on the compliance features and abnormal features. The abnormal operation feature library includes the correlation between interface operations, error factors and phenomenon descriptions. S9. Obtain real-time AGC interface operation information, perform semantic parsing, logical relationship analysis and feature extraction on the real-time AGC interface operation information, and match and identify the extracted real-time operation features with the compliance features and abnormal features in the abnormal operation feature library. S10. When the real-time operation feature meets the compliance feature conditions, output a normal result. When the real-time operation feature meets the abnormal feature conditions, output an abnormal warning message and send an abnormal warning signal to the AGC control module to suspend the corresponding control operation and trigger a secondary confirmation.

2. The method according to claim 1, characterized in that, The preprocessing described in step S2 includes: The system detects duplicate and missing records in the historical AGC interface operation data, removes duplicate records, and completes and marks missing records. The system identifies explicit abnormal data in the historical AGC interface operation data through range checks and implicit abnormal data in the historical AGC interface operation data through correlation checks. Data reconstruction is performed on the data gaps formed after removing missing records, explicit abnormal data, and hidden abnormal data; Based on a preset synchronization cycle, historical AGC interface operation data from different sources are processed for time alignment and frequency synchronization.

3. The method according to claim 1, characterized in that, The abnormal sample enhancement described in step S3 includes: Identify abnormal operation data from the historical AGC interface operation data after preprocessing in step S2; Randomly select target words from the abnormal operation data and generate new abnormal samples using synonym replacement. Randomly swap the word positions in the anomalous operation data to generate order perturbation samples; Insert synonyms of random words at any position in the abnormal operation data to generate inserted perturbation samples; Randomly delete some words from the abnormal operation data to generate a censored perturbation sample.

4. The method according to claim 1, characterized in that, The blurring process described in step S4 includes: Select operation A to be clustered; Randomly select multiple operation contents B from existing clusters. k ; The following formula is used to calculate operation content A and operation content B. k Similarity between them: Where S represents the similarity, and C represents the similarity between operation content A and operation content B. k The number of identical characters in the operation content C, l(C) is the number of characters in the operation content A, l(B) is the number of characters in the operation content A, and l(C) is the number of characters in the operation content A. k ) represents operation content B k The number of characters; When operation content A is combined with multiple operation contents B k If the similarity between the elements is greater than a preset threshold, the operation content A will be assigned to a cluster that meets the preset threshold condition; otherwise, a new cluster will be created.

5. The method according to claim 1, characterized in that, The semantic parsing described in step S5 includes: The standardized operational text after fuzzification in step S4 is segmented to obtain multiple semantic units. Extract the operation object, control target, operation time, and operation type information from the semantic unit; The status words, punctuation marks, and invalid descriptive words in the standardized operation text are removed to obtain the operation text after removing stop words; An operational semantic feature vector is generated based on the operational text after removing stop words.

6. The method according to claim 1, characterized in that, The rule association analysis described in step S6 includes: A set of interface operation rules was established based on the power industry's operating procedures, AGC operating specifications, and the company's internal operating regulations. The set of interface operation rules is classified according to operation type, operation step, and controlled object; The real-time interface operation information is matched item by item with the corresponding category of interface operation rules; If any interface operation does not meet the corresponding interface operation rules, the interface operation will be judged as an abnormal operation. The statistical association analysis includes: Perform statistical analysis on the vector of operation semantic features obtained in step S5 to extract the statistical patterns among operation frequency, operation time period, operation type and control target in historical AGC interface operation data; Establish an operation association model based on the correlation between different contents in historical AGC interface operation data; The deviation between real-time interface operation information and historical statistical patterns is analyzed based on the operation correlation model. When the real-time interface operation information deviates from the historical statistical pattern by more than a preset threshold, the interface operation is judged as an abnormal operation.

7. The method according to claim 1, characterized in that, In step S7, a logical language is used to represent compliance features and abnormal features. The abnormal features are represented as follows: ((Number of operations and operation types ∈ Historical statistical patterns) ∧ (Operation types ∈ Sensitive operation types) ∧ (Operation time period ∈ / Regular working hours)); Used to determine whether there is a risk of abnormal AGC operation in the interface.

8. The method according to claim 1, characterized in that, The abnormal operation feature library mentioned in step S8 is constructed according to the AGC operation process, including: The AGC interface operation is divided into operation interface, operator, operation time, controlled object, control target, operation process and operation data verification sub-process; Establish a relationship structure between interface operation, error factors, and phenomenon description for each sub-process. For each abnormal operation, five types of feature elements are established: natural language expression, logical language description, feature mechanism, feature parameters, and abnormal label.

9. The method according to claim 1, characterized in that, Step S9 includes: Get the real-time AGC interface operation text; Semantic parsing and logical relationship analysis are performed on the real-time AGC interface operation text; Extract real-time operation features and match them with compliance and abnormal features in the abnormal operation feature library; Logical reasoning is used to determine the rationality of the operation using the logical expression (¬((Modified object ⋀ Control target) → Station power change) → Power grid frequency change). When the matching result meets the abnormal characteristic conditions, an abnormal warning result is generated and a warning signal is sent to the AGC control module.

10. An AGC (Automatic Guided Vehicle) abnormal operation identification system based on statistical and correlation analysis, characterized in that, include: The data acquisition module is used to acquire historical AGC interface operation data and real-time AGC interface operation data. The data preprocessing module is used to perform data cleaning, abnormal data removal, data reconstruction, and asynchronous data synchronization on historical AGC interface operation data. The data augmentation module is used to perform synonym replacement, random swapping, random insertion, and random deletion on abnormal operation samples to generate augmented abnormal samples. The data blurring module is used to cluster the operation content using a clustering algorithm based on similarity comparison and generate cluster labels; The semantic parsing module is used to perform word segmentation, stop word removal, and semantic feature extraction on the AGC interface operation text; The correlation analysis module is used to conduct rule-based correlation analysis and statistical correlation analysis, and to extract compliance features and anomaly features; The feature library construction module is used to build an abnormal operation feature library; The anomaly identification module is used to match real-time operation features with an anomaly operation feature library and output the anomaly identification results. The control feedback module is used to send an abnormal warning signal to the AGC control module when an abnormal operation is detected, and to suspend the corresponding AGC control operation.

Images