An abnormal query method for power plant power generation in a starting process
By combining a large language model and an MCP server to query power plant power generation data from the Hive database, the problem of low efficiency and low accuracy in anomaly querying during power plant startup was solved, enabling fast and accurate anomaly analysis and identification of influencing factors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING ENERGY INVESTMENT HLDG
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-16
Smart Images
Figure CN122220338A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data analysis technology during the power plant startup process, and in particular to a method for querying anomalies during the power plant startup process. Background Technology
[0002] The startup process of a power plant generator unit is a highly complex, time-sensitive, and multi-system coupled transient process. From boiler ignition and turbine startup to generator grid connection, tens of thousands of measuring parameters (such as temperature, pressure, flow rate, vibration, and current) need to change in tandem. Any abnormal deviation of critical parameters may lead to startup failure, equipment damage, or even unplanned shutdowns, resulting in huge economic losses and safety risks.
[0003] Currently, data on the power plant startup process is mostly stored in document or report form. Users need to manually retrieve and organize this data, which is not only time-consuming and labor-intensive but also prone to human error, making it difficult to meet the requirements for real-time performance and accuracy. Furthermore, relying on human experience to analyze abnormal data leads to strong subjectivity and a lack of unified and objective measurement standards, resulting in different personnel potentially reaching different analytical conclusions regarding the same abnormal startup process.
[0004] Therefore, there is an urgent need for a technical solution to analyze abnormal situations that occur during the startup process of power plants. Summary of the Invention
[0005] Based on the above analysis, the present invention aims to provide a method for querying anomalies during the startup process of power plant generation, in order to solve the problems of low efficiency and low accuracy in the prior art for querying anomalies during the startup process of power plant generation.
[0006] This invention provides a method for querying anomalies during the startup process of a power plant, the method comprising:
[0007] Receive the natural language anomaly query statement input by the user, and call the preset large language model to convert the natural language anomaly query statement into an SQL anomaly query instruction; The MCP server is invoked to execute an SQL exception query command to query the power plant power generation data stored in the Hive database and obtain the exception query data. Data analysis was performed on the abnormal query data to obtain multiple key stage evaluation parameters, as well as multiple key influencing factors corresponding to each key stage evaluation parameter.
[0008] Based on the further improvement of the above-mentioned anomaly query method, the preset large language model converts natural language anomaly query statements into SQL anomaly query instructions through the following steps: The natural language abnormal query statement is preprocessed to obtain the preprocessed query statement; The query start and end times are determined based on the preprocessed query statement to obtain the query time period. Generate SQL exception query instructions based on the query time period.
[0009] Based on further improvements to the above-mentioned anomaly query method, the preprocessing includes one or more of the following: Redundant space removal operation; Irrelevant symbol removal operation; Noise removal operation; Standardized identification operations.
[0010] Based on a further improvement to the above-mentioned abnormal query method, the step of calling the MCP server to execute SQL abnormal query commands to query the power plant power generation data stored in the Hive database includes: The MCP server identifies multiple executors based on the SQL exception query command, and each executor corresponds to a query sub-time period in the query time period. Each executor retrieves power plant generation data from Hive data based on the corresponding query sub-time period, and sends the power plant generation data within the query sub-time period to the MCP server to obtain abnormal query data.
[0011] Based on the further improvement of the above-mentioned anomaly query method, the data analysis of the anomaly query data includes: Based on various startup status judgment conditions and the startup phase division conditions corresponding to each startup status, determine the current startup status of the abnormal query data and the multiple current startup phases corresponding to the current startup status. Determine multiple evaluation parameters for each current launch phase, and multiple influencing factor parameters corresponding to each current phase evaluation parameter; By comparing and analyzing each current stage evaluation parameter with its corresponding historical best stage evaluation parameter, multiple key stage evaluation parameters are obtained, as well as at least one key influencing factor corresponding to each key stage evaluation parameter.
[0012] Based on the further improvement of the above-mentioned anomaly query method, the comparative analysis of each current stage evaluation parameter and its corresponding historical best stage evaluation parameter includes: Calculate the difference between each current stage evaluation parameter and the corresponding theoretical optimal stage evaluation parameter to obtain the difference value of each current stage evaluation parameter; Calculate the difference between the historical best stage evaluation parameter and the corresponding theoretical best stage evaluation parameter for each current stage evaluation parameter, and obtain the historical best difference value for each current stage evaluation parameter. Determine whether the difference value of each current stage evaluation parameter is greater than the historical best difference value corresponding to each current stage evaluation parameter; if so, use the current stage evaluation parameter as the key stage evaluation parameter, and at the same time determine at least one key influencing factor parameter for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0013] Based on the further improvement of the above-mentioned anomaly query method, the step of determining at least one key influencing factor parameter that the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter includes: The parameter values of multiple influencing factors corresponding to the current stage evaluation parameters at various time points in the current start-up phase are determined as the first data, and the parameter values of the historical best stage evaluation parameters at various time points are determined as the second data; Based on the first and second data, the similarity of multiple parameters of the influencing factors in this case is calculated respectively; Based on the similarity of multiple influencing factor parameters and the preset influencing factor judgment conditions, at least one key influencing factor parameter is identified as the one for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0014] Based on the further improvement of the above-mentioned anomaly query method, the step of calculating the similarity of multiple influencing factor parameters according to the first data and the second data includes: The first and second data are time-aligned using the Dynamic Time Warping (DTW) method to obtain aligned first and second data. Determine the value of each current influencing factor parameter from the aligned first and second data, and calculate the similarity of each current influencing factor parameter.
[0015] Based on the further improvement of the above-mentioned anomaly query method, the step of determining at least one key influencing factor parameter that has not reached the corresponding historical best stage evaluation parameter, according to the similarity of multiple current influencing factor parameters and preset influencing factor judgment conditions, includes: The similarity of multiple influencing factors parameters is sorted in ascending order to obtain a sorted similarity sequence; The parameters corresponding to the pre-defined proportions of similarity in the sorted similarity sequence are taken as the key influencing factors for the current stage evaluation parameters that have not reached the corresponding historical best stage evaluation parameters.
[0016] Based on the further improvement of the above-mentioned anomaly query method, if the difference value of each current stage evaluation parameter is less than the historical best difference value corresponding to each current stage evaluation parameter, then the current stage evaluation parameter will be removed and will not be used as a key stage evaluation parameter.
[0017] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. By calling the preset large language model, the natural language abnormal query statement is converted into an SQL abnormal query command. The MCP server is called to execute the SQL abnormal query command to query the power plant power generation data stored in the Hive database, and the abnormal query data is obtained quickly and accurately. 2. By combining pre-set judgment conditions for various startup states and startup stage division conditions corresponding to each startup state, the startup process of power plant power generation is divided into multiple startup stages corresponding to the current startup state, multiple current stage evaluation parameters for each current startup stage, and multiple current influencing factor parameters corresponding to each current stage evaluation parameter. By comparing and analyzing these parameters with pre-set historical best stage evaluation parameters, multiple key stage evaluation parameters for this abnormal startup process are determined, as well as at least one key influencing factor corresponding to each key stage evaluation parameter. This greatly improves the accuracy of the analysis of key influencing factors in the startup process of power plant power generation.
[0018] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description
[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.
[0020] Figure 1 A flowchart illustrating a method for analyzing key influencing factors during the power plant startup process, provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the data analysis process for abnormal query data provided in an embodiment of the present invention. Detailed Implementation
[0021] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.
[0022] A specific embodiment of the present invention discloses a method for querying anomalies during the startup process of a power plant, such as... Figure 1 As shown, the anomaly query method includes: Step S1: Receive the natural language anomaly query statement input by the user, and call the preset large language model to convert the natural language anomaly query statement into an SQL anomaly query instruction; Step S2: Call the MCP server to execute the SQL exception query command to query the power plant power generation data stored in the Hive database and obtain the exception query data; Step S3: Perform data analysis on the abnormal query data to obtain multiple key stage evaluation parameters and multiple key influencing factors corresponding to each key stage evaluation parameter.
[0023] Specifically, such as Figure 1 As shown, when users discover an anomaly in the power plant's power generation during the startup process, it is necessary to detect and analyze the anomaly. In existing technologies, the power plant's power generation data is usually stored in the form of document reports, which is inefficient when analysis is required.
[0024] Specifically, such as Figure 1 As shown in this embodiment of the invention, all data generated during the power plant's startup process is stored in a Hive database for easy querying and analysis.
[0025] Specifically, during the operation of the generating units in the power plant, various measuring points set up on-site monitor the unit's operating parameters in real time. The measuring point data is collected uniformly by the on-site distributed control system, processed through the data access service according to the preset data synchronization strategy, and sent to the Hive database in a streaming data transmission mode.
[0026] The data synchronization strategy is used to constrain the method and timing of the transmission of measurement point data from the field system to the data platform. Specifically, the synchronization frequency is set according to the data acquisition cycle of the measurement points; the continuously acquired data is batch-packaged according to the time window; each data is attached with an acquisition time identifier and a measurement point identifier; and sequence control and integrity verification are performed during data transmission to ensure the timing consistency and data integrity of the data during cross-system transmission.
[0027] Specifically, during the execution of the data synchronization strategy, the data access service performs protocol parsing, timestamp alignment, and basic verification on the collected measurement point data. Then, it encapsulates the data into structured data according to a unified data model and writes the processed data into the Hive database for centralized storage.
[0028] Specifically, during the data entry process, the raw collected data undergoes time alignment and sampling control processing based on timestamp windows.
[0029] Specifically, the acquisition timestamps attached to the data of each measurement point are parsed, and a fixed time window is used as the alignment benchmark to map the data falling within the same time window to a unified time identifier, thereby achieving time unification of measurement point data with different sampling periods.
[0030] After time alignment is completed, data within the same time window is aggregated and converted into a data structure with minute-level time granularity. The processed data is then written to the Hive data warehouse to reduce the data scanning range during queries and improve the efficiency of subsequent data queries and analysis.
[0031] Specifically, such as Figure 1 As shown, in step S1, after the user determines that an anomaly has occurred in the current startup process, the user can directly input a natural language anomaly query statement with a query time period. After receiving the natural language anomaly query statement input by the user, the large language model is called to parse the natural language anomaly query statement and convert the natural language anomaly query statement into an SQL anomaly query instruction.
[0032] Preferably, the preset large language model converts natural language anomaly query statements into SQL anomaly query instructions through the following steps: The natural language abnormal query statement is preprocessed to obtain the preprocessed query statement; The query start and end times are determined based on the preprocessed query statement to obtain the query time period. Generate SQL exception query instructions based on the query time period.
[0033] Specifically, the pre-defined large language model preprocesses abnormal natural language query statements, standardizing the preprocessed query statements to improve the accuracy of subsequent semantic understanding.
[0034] Preferably, the preprocessing includes one or more of the following: Redundant space removal operation; Irrelevant symbol removal operation; Noise removal operation; Standardized identification operations.
[0035] Specifically, operations such as redundant space removal, irrelevant symbol removal, noise removal, or normalization recognition can reduce the impact of irrelevant factors on the generation of abnormal SQL query commands and improve the accuracy of query command generation.
[0036] Specifically, the pre-defined large language model determines the query start and end times based on the pre-processed query statement. The pre-defined large language model uses natural language understanding to pre-process the query statement, identify and extract key semantic elements related to the database query, including indicator names, time ranges, numerical conditions, and aggregation requirements.
[0037] Specifically, for metric names, the pre-defined large language model uses a combination of fuzzy matching and semantic constraints to map user-input colloquial or non-standard metric descriptions to identifiable metric fields in the Hive database.
[0038] Specifically, when a preprocessed query mentions multiple metrics, the pre-defined large language model automatically determines their logical relationships based on semantic relationships and generates corresponding logical combination conditions.
[0039] Specifically, for time information parsing, it supports the standardization of various time expression forms, including explicit dates (November 11, 2025), interval dates (November 1-10, 2025), and relative time (last week) descriptions. The parsing results are converted into explicit time interval conditions and finally mapped to Hive query constraints based on partition fields or timestamp fields, thereby ensuring the accuracy and executability of the query range.
[0040] Specifically, the pre-defined large language model determines the query start time and query end time based on the preprocessed query statement to obtain the query time period.
[0041] Specifically, it generates SQL exception query commands based on the query time period. Specifically, the preset large language model can structurally assemble the query time period according to preset SQL generation rules and security constraints, automatically generating query statements that conform to HiveSQL syntax specifications.
[0042] It is worth noting that during the generation process, the pre-defined large language model imposes unified constraints on the query table, field range, aggregation method, and filtering conditions to ensure that the SQL statement only accesses the target data table and avoids the introduction of irrelevant fields or illegal structures.
[0043] Specifically, such as Figure 1 As shown, an SQL exception query instruction was obtained in step S1.
[0044] Specifically, such as Figure 1 As shown, in step S2, the MCP server is called to execute an SQL exception query command to query the power plant power generation data stored in the Hive database and obtain exception query data.
[0045] Preferably, the step of calling the MCP server to execute an SQL exception query command to query the power plant power generation data stored in the Hive database includes: The MCP server identifies multiple executors based on the SQL exception query command, and each executor corresponds to a query sub-time period in the query time period. Each executor retrieves power plant generation data from Hive data based on the corresponding query sub-time period, and sends the power plant generation data within the query sub-time period to the MCP server to obtain abnormal query data.
[0046] Specifically, several QueryRunner executors are pre-configured in the MCP server. As the executor of the Hive database, QueryRunner encapsulates the communication logic for querying the Hive database, establishes session connections, and is responsible for completing connection tests, metadata retrieval, and the actual execution of SQL statements.
[0047] Specifically, when the MCP server is invoked, the MCP server selects multiple QueryRunners from a number of QueryRunners. Each QueryRunner is used to execute a query sub-time period, and the multiple QueryRunners work independently of each other.
[0048] Specifically, each executor, QueryRunner, retrieves the power plant generation data for the corresponding query sub-time period from the Hive database and sends the power plant generation data for the query sub-time period to the MCP server. The data retrieved by all executors are combined to obtain the abnormal query data.
[0049] Specifically, such as Figure 1 As shown, data analysis was performed on the abnormal query data to obtain multiple key stage evaluation parameters, as well as multiple key influencing factors corresponding to each key stage evaluation parameter.
[0050] Preferably, the data analysis of abnormal query data includes: Step S51: Determine the current startup status of the abnormal query data and the multiple current startup stages corresponding to the current startup status based on various startup status judgment conditions and startup stage division conditions corresponding to each startup status; Step S52: Determine multiple current stage evaluation parameters for each current start-up stage, and multiple current influencing factor parameters corresponding to each current stage evaluation parameter; Step S53: Compare and analyze each current stage evaluation parameter with the corresponding historical best stage evaluation parameter to obtain multiple key stage evaluation parameters, as well as at least one key influencing factor corresponding to each key stage evaluation parameter.
[0051] Specifically, before proceeding to step S51, the power plant's power generation data can be preprocessed, for example, by performing noise reduction, missing data filling, or outlier removal operations, to improve the accuracy of the analysis.
[0052] Specifically, when implementing the anomaly query method for power plant power generation during startup provided in this embodiment of the invention, it is necessary to pre-set the possible startup states of power plant power generation during startup.
[0053] Specifically, when setting the possible startup states of a power plant during startup, the types of startup states can be set according to the specific type of power plant power generation.
[0054] Preferably, the various startup states include cold startup state, warm startup state, hot startup state, and extremely hot startup state; The startup status judgment parameter is the turbine regulating stage metal temperature during the power plant's power generation startup process.
[0055] Specifically, different power plants can classify their startup status based on the turbine model and operating characteristics.
[0056] In this embodiment of the invention, the turbine regulating stage metal temperature is set as a start-up state judgment parameter, and the start-up state of power plant generation is divided according to the turbine regulating stage metal temperature.
[0057] Specifically, such as Figure 2 As shown, various start-up status judgment conditions and start-up stage division conditions corresponding to each start-up status are pre-set. Then, based on the various start-up status judgment conditions and the start-up stage division conditions corresponding to each start-up status, the current start-up status of the power plant and the multiple current start-up stages corresponding to the current start-up status are determined.
[0058] Specifically, such as Figure 2 As shown, in step S51, the startup status of this abnormal startup is determined based on the abnormal query data of the power plant's power generation during this startup process.
[0059] Specifically, the various pre-set startup states can be configured with different startup stages or the same startup stage; no specific restrictions are imposed here.
[0060] For example, the same startup phase can be set for different startup states. The startup phase includes ignition preparation phase, ignition start-up phase, start-up and grid connection phase, grid connection to AGC phase, etc.
[0061] Preferably, the step of determining the current startup state of the power plant and the multiple current startup stages corresponding to the current startup state based on various startup state judgment conditions and startup stage division conditions corresponding to each startup state includes: In chronological order, the startup start-up parameters of the power plant at each time point are sequentially checked to see if they meet the startup conditions. If the startup conditions are met, the startup start time point is obtained. After the startup start time point, the startup completion parameters of the power plant at each time point are sequentially checked to see if they meet the startup completion conditions. If the startup completion conditions are met, the startup completion time point is obtained. The current startup status of the power plant is determined based on the startup status judgment parameters and various startup status judgment conditions at the startup start time. Based on the startup phase division criteria of this startup status, the data between the startup start time and startup completion time is divided into phases to obtain the time periods corresponding to multiple startup phases of this startup status.
[0062] Specifically, various measurement data during the power plant's startup process are stored in a Hive database in chronological order. All types of measurement data collected at each time point are stored for analysis of the startup process.
[0063] Specifically, when analyzing the startup process of a power plant, priority should be given to determining which time period of data belongs to the startup process of the power plant.
[0064] It is understandable that the power generation process of a power plant can be divided into a startup process, a stable operation process, and a standby process. In this invention, the startup process is analyzed.
[0065] When analyzing the startup process of a power plant, the startup status is determined based on pre-set startup start-up parameters. For example, the startup start-up parameter is the condensate pump inverter current; the time point at which the condensate pump inverter current changes from less than 5A to greater than 5A is taken as the startup start time point.
[0066] It is understandable that during the stable operation of the power plant, the condensate pump inverter current is always greater than 5A, while during standby, the condensate pump inverter current is always less than 5A. Selecting the condensate pump inverter current as the parameter for judging the start of power plant generation can enable the judgment of the start of the startup process.
[0067] That is, the value of the condensate pump inverter current at the start time is greater than 5A, while the value of the condensate pump inverter current at a time point before the start time is less than 5A.
[0068] It is worth noting that the start-up judgment parameters can be reasonably set according to the type of power generation of the power plant, and no specific restrictions are imposed here.
[0069] Specifically, after determining the start-up time, the start-up completion time is then determined. Data from all time points between the start-up time and the start-up completion time are used as data for the power plant's power generation during the start-up process.
[0070] Understandably, AGC (Automatic Generation Control) uses an automatic control program to automatically reallocate the active power output of each generator set within the control area, in order to maintain the system frequency and tie-line exchange power within the planned target range.
[0071] Specifically, when determining the start-up time, the corresponding start-up completion judgment parameter is selected. For example, the start-up completion judgment parameter is "unit in AGC mode". It can be understood that when the value of "unit in AGC mode" changes from abnormal to normal, it proves that the power plant's power generation start-up process is complete.
[0072] Specifically, based on the start-up completion judgment parameters of the power plant at various time points, when the value of the start-up completion judgment parameter at a certain time point meets the start-up completion condition, that time point is taken as the start-up completion time point.
[0073] Specifically, after determining the start time and completion time of the power plant's power generation startup process, the startup status judgment parameters at the start time are obtained. These parameters are then compared with various pre-set startup status judgment conditions to determine the startup status of the current power plant startup process.
[0074] Preferably, the various start-up state determination conditions include: When the startup status judgment parameter is less than the first temperature threshold, the current startup status is a cold startup status; When the startup status judgment parameter is greater than or equal to the first temperature threshold and less than the second temperature threshold, the current startup status is the temperature-state startup status. When the startup status judgment parameter is greater than or equal to the second temperature threshold and less than the third temperature threshold, the current startup status is a hot startup status. When the startup status judgment parameter is greater than or equal to the third temperature threshold, the startup status is the extremely hot startup status.
[0075] Specifically, in this embodiment of the invention, the startup state is divided into cold startup state, warm startup state, hot startup state and extremely hot startup state by setting a first temperature threshold, a second temperature threshold and a third temperature threshold.
[0076] For example, the first temperature threshold is 204℃, the second temperature threshold is 425℃, and the third temperature threshold is 510℃. When the turbine regulating stage metal temperature is less than 204℃, the current startup state is a cold startup state; when the startup state judgment parameter is greater than or equal to 204℃ and less than 425℃, the current startup state is a warm startup state; when the startup state judgment parameter is greater than or equal to 425℃ and less than 510℃, the current startup state is a hot startup state; and when the startup state judgment parameter is greater than or equal to 510℃, the current startup state is an extremely hot startup state.
[0077] Specifically, after determining the current startup status of the power plant based on the startup status judgment parameters and various startup status judgment conditions at the startup start time, the data between the startup start time and the startup completion time is divided into stages in combination with the pre-set startup stage division conditions for this startup status.
[0078] Specifically, for different startup states, startup phases can be reasonably set according to the specific power generation conditions of the power plant, and reasonable startup phase division conditions can also be set.
[0079] Preferably, the step of dividing the data between the start time and the completion time based on the start phase division conditions of the current start state to obtain multiple time periods corresponding to the current start phase, including: Determine the parameters for dividing each stage of this startup process into its current startup state; In chronological order, the parameters for dividing each current startup phase are compared with the startup phase division conditions of the current startup state to determine the time points for dividing each current startup phase. Based on the time points of each startup phase, the data between the startup start time and startup completion time is divided into phases to obtain the time periods corresponding to multiple startup phases for this startup status.
[0080] Specifically, different division and judgment parameters can be set for different startup stages. For example, the startup status includes the ignition preparation stage, the ignition start-up stage, the start-up and grid connection stage, and the grid connection to AGC stage. Fuel flow rate is selected as the division and judgment parameter for the ignition preparation stage. When the fuel flow rate at each time point exceeds the preset fuel flow rate threshold, that time point is taken as the ignition time point. The time period between the startup start time point and the ignition time point is taken as the ignition preparation stage.
[0081] Specifically, after determining the previous startup phase, the corresponding time point is determined based on the division parameters of the next startup phase, until all time periods between the startup start time point and the startup completion time point are divided, resulting in multiple time periods corresponding to the current startup phase for the current startup state.
[0082] It is worth noting that each of the current startup phases corresponds to a time period consisting of multiple time points, and each time point corresponds to various measurement data.
[0083] Specifically, such as Figure 2 As shown, in step S52, multiple stage evaluation parameters for different startup stages under different startup states and multiple influencing factor parameters corresponding to each stage evaluation parameter are pre-set according to the operating characteristics of power plant power generation.
[0084] It is worth noting that different startup stages can be set for different startup states, or the same startup stage can be set; different startup stages can have the same multiple stage evaluation parameters, or different multiple stage evaluation parameters can be set; different stage evaluation parameters can have the same multiple influencing factor parameters, or the same multiple influencing factor parameters can be set. There are no specific restrictions here, and the specific settings can be based on the operating characteristics of the power plant's power generation.
[0085] For example, the stage evaluation parameters are main steam pressure, main steam temperature, main engine lubricating oil temperature, and rotor eccentricity; the main steam pressure and main steam temperature are set with the same influencing factor parameters, and the multiple influencing factor parameters corresponding to the main steam pressure and main steam temperature can be set as total coal quantity, high bypass degree, and low bypass degree; the multiple influencing factor parameters corresponding to the main engine lubricating oil temperature can be set as feedback of the cooling water regulating valve of the turbine lubricating oil cooler, temperature of the outlet header of the closed circulating cooling pump, and status of the water side inlet valve of the oil cooler; the multiple influencing factor parameters corresponding to rotor eccentricity can be set as main steam temperature, reheat steam temperature, turbine lubricating oil pressure, and turbine lubricating oil temperature.
[0086] It is worth noting that all parameters, including influencing factors, stage evaluation parameters, start-up judgment parameters, start-up completion judgment parameters, start-up status judgment parameters, and division judgment parameters, are recorded in the power plant's power generation measurement data at various time points, and will not be elaborated upon here.
[0087] Specifically, such as Figure 2 As shown, in step S53, each current stage evaluation parameter and its corresponding historical best stage evaluation parameter are compared and analyzed to obtain multiple key stage evaluation parameters, as well as at least one key influencing factor corresponding to each key stage evaluation parameter.
[0088] Preferably, the comparative analysis of each current stage evaluation parameter and its corresponding historical best stage evaluation parameter includes: Calculate the difference between each current stage evaluation parameter and the corresponding theoretical optimal stage evaluation parameter to obtain the difference value of each current stage evaluation parameter; Calculate the difference between the historical best stage evaluation parameter and the corresponding theoretical best stage evaluation parameter for each current stage evaluation parameter, and obtain the historical best difference value for each current stage evaluation parameter. Determine whether the difference value of each current stage evaluation parameter is greater than the historical best difference value corresponding to each current stage evaluation parameter; if so, use the current stage evaluation parameter as the key stage evaluation parameter, and at the same time determine at least one key influencing factor parameter for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0089] Specifically, the theoretical optimal stage evaluation parameter corresponding to each stage evaluation parameter is set in advance and used directly when calculating the difference value of each current stage evaluation parameter. The difference value can be the absolute value of the difference between the two, or other measurement values, as long as they can measure the difference between the stage evaluation parameter and the theoretical optimal stage evaluation parameter.
[0090] Specifically, the difference between each current stage evaluation parameter and its corresponding theoretical optimal stage evaluation parameter is calculated to obtain the difference value of each current stage evaluation parameter. At the same time, the difference between the historical optimal stage evaluation parameter and its corresponding theoretical optimal stage evaluation parameter is calculated to obtain the historical optimal difference value of each current stage evaluation parameter.
[0091] Specifically, the difference value of each current stage evaluation parameter is compared with the historical best difference value corresponding to each current stage evaluation parameter. If the difference value of each current stage evaluation parameter is greater than the historical best difference value corresponding to each current stage evaluation parameter, then the current stage evaluation parameter is designated as a key stage evaluation parameter, and at least one key influencing factor parameter is identified as to why the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0092] It is worth noting that when the difference value of each current stage evaluation parameter is greater than the historical best difference value corresponding to each current stage evaluation parameter, it means that the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter. In this case, the current stage evaluation parameter is used as the key stage evaluation parameter. At the same time, at least one key influencing factor parameter is determined for the current stage evaluation parameter not reaching the corresponding historical best stage evaluation parameter, so as to provide a basis for the next power plant power generation startup and optimize the power plant power generation startup process.
[0093] Preferably, the determination of at least one key influencing factor parameter that indicates the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter includes: The parameter values of multiple influencing factors corresponding to the current stage evaluation parameters at various time points in the current start-up phase are determined as the first data, and the parameter values of the historical best stage evaluation parameters at various time points are determined as the second data; Based on the first and second data, the similarity of multiple parameters of the influencing factors in this case is calculated respectively; Based on the similarity of multiple influencing factor parameters and the preset influencing factor judgment conditions, at least one key influencing factor parameter is identified as the one for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0094] Specifically, each stage evaluation parameter corresponds to multiple influencing factor parameters. The data of these multiple influencing factor parameters at various time points corresponding to the current stage evaluation parameter in this startup phase will be used as the first data. For example, the stage evaluation parameter is the main steam temperature, and the multiple influencing factor parameters corresponding to the main steam temperature are total coal quantity, high bypass degree, and low bypass degree. This startup phase includes data at 100 time points, so the first data will each contain 100 parameter values for total coal quantity, high bypass degree, and low bypass degree.
[0095] Specifically, the parameter values of the historical optimal stage evaluation parameters at various time points may vary in number or may be the same, depending on the actual situation. The parameter values of the historical optimal stage evaluation parameters at various time points are used as secondary data.
[0096] Specifically, based on the first and second data, the similarity of multiple parameters of the influencing factors in this case is calculated respectively.
[0097] Preferably, the step of calculating the similarity of multiple influencing factor parameters based on the first data and the second data includes: The first and second data are time-aligned using the Dynamic Time Warping (DTW) method to obtain aligned first and second data. Determine the value of each current influencing factor parameter from the aligned first and second data, and calculate the similarity of each current influencing factor parameter.
[0098] Specifically, the first and second data are time-aligned based on the Dynamic Time Warping (DTW) method to obtain the aligned first data.
[0099] Specifically, considering that the number of time points included in the first data and the number of time points included in the second data are likely to be different, the first data and the second data are time-aligned based on the Dynamic Time Warping (DTW) method to obtain aligned first data and second data.
[0100] Specifically, DTW (Dynamic Time Warping) is an algorithm for measuring the similarity of time series. Its core is based on the idea of dynamic programming. By constructing a distance matrix between two curves and using path constraints with slopes ranging from 0.5 to 2, it finds the optimal path with the minimum cumulative distance, which will not be elaborated on here.
[0101] Specifically, the value corresponding to each current influencing factor parameter is determined from the aligned first and second data, the similarity of each current influencing factor parameter is calculated, and the similarity of multiple current influencing factor parameters is obtained.
[0102] Specifically, the system sets pre-defined influencing factor judgment conditions. After obtaining the similarity of multiple current influencing factor parameters, it determines at least one key influencing factor parameter that has not reached the corresponding historical best stage evaluation parameter based on the similarity of multiple current influencing factor parameters and the pre-defined influencing factor judgment conditions.
[0103] Specifically, the number of key influencing factor parameters should be set reasonably according to the actual situation.
[0104] It is worth noting that when only one key influencing factor parameter is needed, the current influencing factor corresponding to the minimum similarity among multiple current influencing factor parameters can be selected as a key influencing factor parameter for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
[0105] Specifically, when multiple key influencing factor parameters are needed, influencing factor judgment conditions can be set, and the similarity of multiple influencing factor parameters in this case can be combined to determine the influencing factor.
[0106] Preferably, the step of determining at least one key influencing factor parameter that has not reached the corresponding historical best stage evaluation parameter based on the similarity of multiple current influencing factor parameters and preset influencing factor judgment conditions includes: The similarity of multiple influencing factors parameters is sorted in ascending order to obtain a sorted similarity sequence; The parameters corresponding to the pre-defined proportions of similarity in the sorted similarity sequence are taken as the key influencing factors for the current stage evaluation parameters that have not reached the corresponding historical best stage evaluation parameters.
[0107] The similarity of multiple influencing factor parameters is sorted in ascending order. The influencing factor parameters with higher similarity are significantly different from those with historical best evaluation parameters, while the influencing factor parameters with lower similarity are less different from those with historical best evaluation parameters.
[0108] When a specific number of key influencing factor parameters are needed, multiple key influencing factor parameters can be obtained by selecting them sequentially from the sorted similarity sequence.
[0109] Preferably, if the difference value of each current stage evaluation parameter is less than the historical best difference value corresponding to each current stage evaluation parameter, then the current stage evaluation parameter is removed and not used as a key stage evaluation parameter.
[0110] Specifically, if the difference value of the current stage evaluation parameter is less than the historical best difference value corresponding to the current stage evaluation parameter, it means that the current stage evaluation parameter is closer to the corresponding theoretical best stage evaluation parameter. Therefore, the current stage evaluation parameter is not considered an abnormal cause in the current startup process and will be removed from the evaluation parameters for this stage.
[0111] Compared with existing technologies, the present invention provides an anomaly query method for power plant power generation during startup. This method converts natural language anomaly query statements into SQL anomaly query instructions by calling a preset large language model. The SQL anomaly query instructions are then executed by calling an MCP server to query power plant power generation data stored in a Hive database, quickly and accurately obtaining anomaly query data. Simultaneously, by combining pre-set startup state judgment conditions and startup stage division conditions corresponding to each startup state, the power plant power generation startup process is divided. This yields multiple current startup stages corresponding to the current startup state, multiple current stage evaluation parameters for each current startup stage, and multiple current influencing factor parameters corresponding to each current stage evaluation parameter. By comparing and analyzing these parameters with pre-set historical best stage evaluation parameters, multiple key stage evaluation parameters for the current anomaly startup process are determined, along with at least one key influencing factor corresponding to each key stage evaluation parameter. This significantly improves the accuracy of analyzing key influencing factors during power plant power generation startup.
[0112] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0113] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for anomaly detection during the startup process of a power plant, characterized in that, The anomaly query method includes: Receive the natural language anomaly query statement input by the user, and call the preset large language model to convert the natural language anomaly query statement into an SQL anomaly query instruction; The MCP server is invoked to execute an SQL exception query command to query the power plant power generation data stored in the Hive database and obtain the exception query data. Data analysis was performed on the abnormal query data to obtain multiple key stage evaluation parameters, as well as multiple key influencing factors corresponding to each key stage evaluation parameter.
2. The anomaly query method according to claim 1, characterized in that, The preset large language model converts natural language anomaly query statements into SQL anomaly query commands through the following steps: The natural language abnormal query statement is preprocessed to obtain the preprocessed query statement; The query start and end times are determined based on the preprocessed query statement to obtain the query time period. Generate SQL exception query instructions based on the query time period.
3. The anomaly query method according to claim 2, characterized in that, The preprocessing includes one or more of the following: Redundant space removal operation; Irrelevant symbol removal operation; Noise removal operation; Standardized identification operations.
4. The anomaly query method according to claim 2, characterized in that, The step of calling the MCP server to execute an SQL exception query command to query the power plant power generation data stored in the Hive database includes: The MCP server identifies multiple executors based on the SQL exception query command, and each executor corresponds to a query sub-time period in the query time period. Each executor retrieves power plant generation data from Hive data based on the corresponding query sub-time period, and sends the power plant generation data within the query sub-time period to the MCP server to obtain abnormal query data.
5. The anomaly query method according to claim 1, characterized in that, The data analysis of abnormal query data includes: Based on various startup status judgment conditions and the startup phase division conditions corresponding to each startup status, determine the current startup status of the abnormal query data and the multiple current startup phases corresponding to the current startup status. Determine multiple evaluation parameters for each current launch phase, and multiple influencing factor parameters corresponding to each current phase evaluation parameter; By comparing and analyzing each current stage evaluation parameter with its corresponding historical best stage evaluation parameter, multiple key stage evaluation parameters are obtained, as well as at least one key influencing factor corresponding to each key stage evaluation parameter.
6. The anomaly query method according to claim 5, characterized in that, The comparison and analysis of each current stage evaluation parameter and its corresponding historical best stage evaluation parameter includes: Calculate the difference between each current stage evaluation parameter and the corresponding theoretical optimal stage evaluation parameter to obtain the difference value of each current stage evaluation parameter; Calculate the difference between the historical best stage evaluation parameter and the corresponding theoretical best stage evaluation parameter for each current stage evaluation parameter, and obtain the historical best difference value for each current stage evaluation parameter. Determine whether the difference value of each current stage evaluation parameter is greater than the historical best difference value corresponding to each current stage evaluation parameter; if so, use the current stage evaluation parameter as the key stage evaluation parameter, and at the same time determine at least one key influencing factor parameter for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
7. The anomaly query method according to claim 6, characterized in that, The determination of at least one key influencing factor parameter that indicates the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter includes: The parameter values of multiple influencing factors corresponding to the current stage evaluation parameters at various time points in the current start-up phase are determined as the first data, and the parameter values of the historical best stage evaluation parameters at various time points are determined as the second data; Based on the first and second data, the similarity of multiple parameters of the influencing factors in this case is calculated respectively; Based on the similarity of multiple influencing factor parameters and the preset influencing factor judgment conditions, at least one key influencing factor parameter is identified as the one for which the current stage evaluation parameter has not reached the corresponding historical best stage evaluation parameter.
8. The anomaly query method according to claim 7, characterized in that, The step of calculating the similarity of multiple influencing factor parameters based on the first and second data includes: The first and second data are time-aligned using the Dynamic Time Warping (DTW) method to obtain aligned first and second data. Determine the value of each current influencing factor parameter from the aligned first and second data, and calculate the similarity of each current influencing factor parameter.
9. The anomaly query method according to claim 7, characterized in that, The step of determining at least one key influencing factor parameter that has not reached the corresponding historical best stage evaluation parameter based on the similarity of multiple influencing factor parameters and preset influencing factor judgment conditions includes: The similarity of multiple influencing factors parameters is sorted in ascending order to obtain a sorted similarity sequence; The parameters corresponding to the pre-defined proportions of similarity in the sorted similarity sequence are taken as the key influencing factors for the current stage evaluation parameters that have not reached the corresponding historical best stage evaluation parameters.
10. The anomaly query method according to claim 6, characterized in that, If the difference value of each current stage evaluation parameter is less than the historical best difference value corresponding to each current stage evaluation parameter, then the current stage evaluation parameter will be removed and will not be used as a key stage evaluation parameter.