System for supporting ai analysis of large-capacity power generation sensor data and method for distributed processing of large-capacity power generation sensor data

Abstract

A system for supporting AI analysis of large-capacity power generation sensor data of the present invention may comprise: a time-series distributed processing unit for managing distributed indexing for sensor data generated by each sensor installed in a power facility; a development support model for creating, in a form for developer support, a support list for sensor data present in the time-series distributed processing unit; and an association analysis model for analyzing whether each item of the support list is associated with another item.

Description

AI analysis support system for large-capacity power generation sensor data and distributed processing method for large-capacity power generation sensor data.

[0001] The present invention relates to a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on correlation analysis, and more specifically, to a method and device for classifying historical sensor data (tags) from a time-series distributed database according to correlation and providing the same to an OPC UA information model, distributing and storing large-capacity tag data of a power generation company by tag on a daily and monthly basis to provide high-speed extraction performance, and reflecting overhaul information and sensor abnormality information in an OPC UA information model to provide optimal data for AI analysis.

[0002]

[0003] The global big data market size is expected to reach approximately KRW 561 trillion by 2030 (annual average growth rate of 13.9%) (VMR, 2023), and the energy industry digitalization market is estimated to reach approximately KRW 85 trillion by 2025 (annual average growth rate of 3.3%).

[0004] Figure 11 shows the distribution of the big data market in the electric power sector.

[0005] The market size for digitalization of the energy industry is, as shown in Figure 11, in the order of smart meters, thermal power generation O&M, and distribution automation. The thermal power generation O&M sector, which is an IDPP (Intelligent Digital Power Plant) business area, accounts for 17%, and the market size is estimated to be approximately KRW 14.6 trillion as of 2025.

[0006] In particular, the need for flexible operation of large thermal power plants is rapidly increasing due to the rapid increase in renewable energy generation in accordance with carbon neutrality policies, and accordingly, the introduction of a platform for storing and utilizing real-time power generation sensor data for flexible operation is required.

[0007] In other words, environmental regulations to prepare for climate change require power plant efficiency improvements to reduce CO2 emissions and increase operating rates, and aim to improve power generation cost and performance management systems.

[0008] However, with the rapid growth of renewable energy, power plant startups and shutdowns are increasing rapidly, and low-load operation situations are becoming more frequent. Therefore, countermeasures to address this rapid increase in failures are necessary. For example, Unit 1 of a large thermal power plant generates approximately 1.7 billion data points per day, requiring systems for data storage, learning, and analysis.

[0009] Meanwhile, power generation companies want rapid, economical solutions that reflect the opinions of field experts, minimize the burden on personnel and resources due to business improvements, and continuously strengthen business capabilities.

[0010]

[0011] The present invention aims to provide a large-capacity power generation sensor data AI analysis support system and distributed processing method capable of efficiently managing and appropriately providing large-capacity power generation sensor (tag) or IoT data continuously produced to a power generation company.

[0012] The present invention aims to provide a large-capacity power generation sensor data AI analysis support system and distributed processing method that enable AI developers to quickly obtain large-capacity power generation sensor (tag) data suitable for necessary cases.

[0013]

[0014] A large-capacity power generation sensor data AI analysis support system according to one aspect of the present invention may include a time-series distributed processing unit that manages distributed indexing of sensor data generated from each sensor installed in a power facility; a development support model that creates a support list for sensor data existing in the time-series distributed processing unit in a form for developer support; and a correlation analysis model that analyzes whether each item in the support list is related to other items.

[0015] Here, a sensor anomaly management unit may be further included to detect anomalies in each sensor that generates sensor data using the above-mentioned association analysis model.

[0016] Here, a learning unit may be further included to learn the association analysis model by extracting power plant sensor data history information for a certain period from a large-capacity time series database.

[0017] Here, the time series distribution processing unit can store and manage a set of values ​​generated over a predetermined period of time from a sensor mounted on a power generation facility as unit data in a DB.

[0018] Here, the time series distribution processing unit can perform data organization for distributed processing by integrating one day's worth of values ​​generated from the one sensor at a specific point in time during the day, and can perform data organization for distributed processing by integrating one month's worth of values ​​at a specific point in time during the month.

[0019] Here, the development support model may include an asset node composed of data objects for power generation facilities according to a predetermined developer support form; and an association group node composed of group objects representing an association group of sensors mounted on each power generation facility.

[0020] Here, the above asset node can describe overhaul information for each power generation facility.

[0021] Here, the above-mentioned association group node can describe whether a specific sensor has a sensor abnormality.

[0022] Here, sensor data of power plant power equipment is acquired from an OPC Server connected to DCS, and a linkage module that acquires data as a TimeStamp through PTP time synchronization may be further included.

[0023]

[0024] A method for distributing large-capacity power generation sensor data according to another aspect of the present invention may include the steps of: acquiring power generation data (values ​​generated from sensors installed in each power generation facility) from a power plant DCS; storing the acquired data as a temporary tag-based file; integrating values ​​for one day at a specific point in time during the day for values ​​generated from one sensor, organizing the data for distributed processing, and storing the data as a daily file; and integrating values ​​for one month at a specific point in time during the month, organizing the data for distributed processing, and storing the data as a monthly file.

[0025] Here, a step of performing visual synchronization on the acquired development data may be further included.

[0026]

[0027] A method for creating an OPC UA information model for power generation AI support according to another aspect of the present invention may include the steps of: constructing an OPC UA information model having data objects for power generation facilities; securing a large amount of history data as power generation data composed of values ​​generated from sensors mounted on each power generation facility; training an association analysis model using the large amount of history data; deriving an association group by tag for sensors mounted on each power generation facility using the association analysis model; and reflecting information on the association group by tag in the OPC UA information model.

[0028] Here, the step of verifying the accuracy of the association analysis model that has performed learning may be further included; and the step of distributing the association analysis model learned to have a satisfactory level of accuracy for real-time field use may be further included.

[0029] Here, in the step of constructing the OPC UA information model, an asset node of the OPC UA information model consisting of data objects for power generation facilities can be constructed according to a predetermined OPC UA developer support form.

[0030] Here, a step of acquiring overhaul information for each power generation facility; if it is confirmed that an overhaul has been performed on a specific power generation facility, a step of recording the overhaul performance in an object of the specific power generation facility in the asset node may be further included.

[0031] Here, in the step of reflecting information on the association group by tag, an association group node of the OPC UA information model consisting of group objects representing the association group of sensors mounted on each power generation facility can be configured.

[0032] Here, the method may include a step of performing association analysis on real-time data developed in a time-series distributed DB using a learned association analysis model; a step of storing the association analysis result and searching for tag information that can be compared to the data on which the association analysis was performed from tags for each group of the association group node; a step of comparing the searched tag information with the association analysis result; and a step of recording sensor abnormality information for the sensor that generated the tag information if the tag information and the association analysis result do not match.

[0033]

[0034] A method for supporting AI analysis of large-capacity power generation sensor data according to another aspect of the present invention may include the steps of: searching for a facility object having an attribute entered for search by an AI developer in an asset node of an OPC UA information model; extracting a list of tags related to each facility object; searching for an association group for each tag in an association group node of the OPC UA information model to extract an association group object (Obejct); and extracting data for AI analysis from a power generation data storage unit based on objects extracted from the asset node and the association group node.

[0035] Here, a step of utilizing the extracted large-capacity development data for AI analysis can be further included.

[0036] Here, if the equipment object obtained as a result of the search is an overhaul object, a step of filtering it may be further included.

[0037] Here, a step of filtering out a group object (Object) obtained as a result of the search, if it contains a tag of 'sensor abnormality', may be further included.

[0038]

[0039] By implementing a large-capacity power generation sensor data AI analysis support system and / or distributed processing method according to the present invention of the above-described configuration, there is an advantage in that large-capacity power generation sensor (tag) data suitable for a necessary case can be quickly secured.

[0040] More specifically, by applying the present invention, it is possible to quickly extract power plant ultra-large capacity tag data, improve data quality, and immediately secure tag group data suitable for the corresponding CASE during AI analysis without separate preprocessing work.

[0041] Ultimately, the large-capacity power generation sensor data AI analysis support system and / or distributed processing method of the present invention drastically reduces the AI ​​data preprocessing time and learning execution time, and if ultra-large-capacity power generation data is applied to the flexible operation of 20 coal-fired power plants in Korea through real-time AI learning, it can lead to a reduction in power purchase costs.

[0042] In addition, other features and advantages of the present invention may be newly discovered through the embodiments of the present invention.

[0043]

[0044] Figure 1 is a schematic diagram showing the IDPP platform.

[0045] FIG. 2 is a conceptual diagram illustrating an embodiment of a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis according to the idea of ​​the present invention.

[0046] Figure 3 is a conceptual diagram illustrating the large-capacity power generation sensor data AI analysis support system of the present invention from the perspective of association analysis and sensor anomaly detection.

[0047] Figure 4 is a conceptual diagram illustrating the time series distribution processing unit of Figure 2 from the perspective of the role of the development data distribution processor.

[0048] Figure 5 is a conceptual diagram expressing the functions of the development support model of Figure 2 as an OPC UA information model for development AI support.

[0049] Figure 6 is a conceptual diagram schematically illustrating the interrelationships of various support methods that can be performed in a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis.

[0050] Figure 7 is a flowchart illustrating a method (procedure) for distributing development data among the support methods of Figure 6.

[0051] Figure 8 is a flowchart illustrating a method (generation procedure) for creating an information model for development AI support among the support methods of Figure 6.

[0052] FIG. 9 is a flowchart illustrating a sensor anomaly detection process (procedure) based on association analysis that can be included in or performed subsequently to the method for creating an information model for development AI support of FIG. 8.

[0053] Figure 10 is a flowchart illustrating an example of a method for supporting AI analysis of large-capacity power generation data that can be performed in a large-capacity power generation sensor data AI analysis support system.

[0054] Figure 11 is a table showing the distribution of big data markets in the power sector.

[0055] Figure 12 is a table comparing the IDPP data platform and commercial platforms.

[0056] * Explanation of symbols

[0057] 100: Time series distributed processing unit

[0058] 170: Large-scale time series database

[0059] 200: Development Support Model (OPC UA)

[0060] 220: Asset Nodes

[0061] 240: Associated Group Nodes

[0062] 440: Association Analysis Model

[0063] 450: Learning Department

[0064] 480: Sensor Abnormality Cleanup Department

[0065]

[0066] When describing the present invention, terms such as "first" and "second" may be used to describe various components. However, the components may not be limited by these terms. The terms are used solely to distinguish one component from another. For example, without departing from the scope of the present invention, the first component could be referred to as the "second component," and similarly, the second component could also be referred to as the "first component."

[0067] When it is said that a component is connected or connected to another component, it can be understood that it may be directly connected or connected to that other component, but there may also be other components in between.

[0068] The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the present invention. Singular expressions may include plural expressions unless the context clearly dictates otherwise.

[0069] In this specification, terms such as “include” or “have” are intended to indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, and can be understood as not excluding in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0070] Additionally, the shape and size of elements in the drawing may be exaggerated for clearer explanation.

[0071]

[0072] To achieve the recent demands described above in the power plant field, a data platform that can quickly adopt and utilize rapidly developing AI technologies can be introduced.

[0073] Commercial data platforms such as AVEVA (UK) PI System and Capstone (USA) dataPARC have been introduced and are being utilized in industries around the world, including power generation (including power generation), manufacturing, and service. A domestic small and medium-sized enterprise (BNF) has also developed a similar structure for its data platform and introduced it to some domestic power generation companies.

[0074] Existing solutions such as PI system and dataPARC are based on compressed DB technology developed in the 1990s and 2000s, making it difficult to provide the data extraction performance required for AI analysis (requiring the provision of raw data for all sensors at the same time), and the introduction and operation license costs are high.

[0075] Figure 12 is a table comparing the IDPP data platform and commercial platforms.

[0076] Figure 1 is a schematic diagram showing the IDPP platform.

[0077] The IDPP data platform, developed by the Korea Electric Power Research Institute in 2023, is a time-series database-based technology that maintains quality while maintaining storage. It boasts superior raw data extraction performance and, thanks to its open-source software architecture, offers low licensing costs. Accordingly, the company is actively pursuing domestic and international technology commercialization (Figure 12).

[0078] Power plant data for the IDPP data platform consists of time-series data (also called tags) collected from sensors installed on power generation facilities. These data consist of approximately 10,000 to 50,000 tags per unit, with an acquisition cycle of approximately one second. Currently, the five power generation companies where the IDPP platform has been validated use 230,000 tags, requiring approximately 52 TB of storage capacity annually.

[0079] To utilize this type of ultra-large-capacity sensor data for AI analysis, high-speed data processing, high-quality data, and easy data preprocessing are essential.

[0080] To store and extract time-series sensor data at high speed, distributed processing technology based on a large-capacity time-series database is required.

[0081] Specifically, in the case of power plants, in order to secure data quality and ease of preprocessing for use in AI, it is necessary to synchronize the generation time of sensor data acquired by each power plant with high precision (AI requires data from the same time zone), technology to filter out error data caused by abnormal operation of on-site sensors, technology to exclude data generated during the maintenance (overhaul) period of power plant equipment (this causes a decline in learning quality when AI learns using data generated during non-normal operating periods), technology to group sensor data necessary for learning for AI analysis (for accurate AI learning, tasks such as selection and grouping of sensors that have mutual influence are required), etc.

[0082]

[0083] The present invention provides power plant equipment maintenance (Overhaul) information, ultra-high precision data generation time information, and sensor data error (abnormal operation) information to power plant OPC UA. 11 By reflecting it in the OPC Unified Architecture information model standardized in IEC 62541, high-quality data is secured, correlation group information is created for each sensor data through correlation analysis, and this is provided by fusing it with power plant equipment data so that it can be used immediately without additional preprocessing work when performing AI analysis on large amounts of data.

[0084] Furthermore, based on a large-capacity time-series database, it provides high-performance data extraction through standardized distributed storage and extraction. In short, existing power plant time-series sensor data storage systems focus on storage, such as compressed sensor data or standardized sensor information. This requires lengthy preprocessing for AI applications on large-capacity sensor data, making it difficult to utilize. This issue is addressed by this project.

[0085]

[0086] FIG. 2 is a conceptual diagram illustrating an embodiment of a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis according to the idea of ​​the present invention.

[0087] The city sensor (tag) data AI analysis support system may include a time series distributed processing unit (may be equipped with a database) (100) that manages distributed indexing of sensor data generated from each sensor (tag) installed in a power facility; a development support model (OPC UA) (200) that creates a support list (nodes) for sensor data existing in the time series distributed processing unit in a form for developer support; and a correlation analysis model (440) that analyzes whether each item in the support list is related to other items.

[0088] The city's large-capacity power generation sensor data AI analysis support system may further include a sensor abnormality management unit (480) (performing monitoring and reporting) that detects abnormalities in each sensor (tag) that generates sensor data using the above-mentioned association analysis model (440).

[0089]

[0090] The illustrated implementation is a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis according to the idea of ​​the present invention, which functionally performs the functions of classifying historical sensor data (tags) from a time-series distributed database according to association relationships and providing them to an OPC UA information model, distributing and storing large-capacity tag data of a power generation company by tag on a daily and monthly basis to provide high-speed extraction performance, and reflecting overhaul information and sensor abnormality information in the OPC UA information model to provide optimal data for AI analysis.

[0091]

[0092] Figure 3 is a conceptual diagram expressing the large-capacity power generation sensor data AI analysis support system of the present invention from the perspective of association analysis and sensor anomaly detection roles.

[0093] The urban learning unit (450) extracts power plant tag history information for a certain period from a large-capacity time series database (170) to train an association analysis model and performs repetitive learning to achieve a certain level of accuracy.

[0094] The illustrated association analysis model (440) is a model that has completed learning and been distributed. It performs association analysis on power plant data to divide it into groups by tag (grouped according to correlation) and reflects this in the OPC UA information model (200).

[0095] Thereafter, the association analysis model (440) continuously extracts data from a large-capacity time series database (170) and performs association analysis, and the sensor anomaly management unit (480) shown in the figure reflects the result of the association analysis performed, if different from the initial result (or average result), in the corresponding attribute ('sensor anomaly') of the OPC UA information model so that it can be used to determine whether there is a sensor anomaly.

[0096]

[0097] Fig. 4 is a conceptual diagram illustrating the time series distribution processing unit (100) of Fig. 2 from the perspective of the role of a development data distribution processor.

[0098] The illustrated link module (130) acquires power plant data from an OPC Server (10) connected to the DCS, and at this time, performs ultra-precision time synchronization based on a PTP (Precision Time Protocol, IEEE 1588) server, thereby providing a basis for utilizing tag data from different power plants and different units for AI analysis by considering the same occurrence time.

[0099] AI analysis of time-series data, such as power plant sensor data, requires simultaneous training of data generated at the same point in time. The time-series distributed processing unit (100) is based on a large-capacity time-series database. The acquired data is temporarily stored in files by tag, and distributed processing is performed periodically on a daily and monthly basis.

[0100] In summary, the time series distribution processing unit (100) can store and manage a set of values ​​generated from one sensor or tag over a predetermined period of time as unit data (e.g., file by tag) in a time series distribution DB (170).

[0101] For example, the time series distributed processing unit (100) may perform data organization for distributed processing by integrating one-day values ​​generated from a single sensor or tag at a specific point in time during the day, and may perform data organization for distributed processing by integrating one-month values ​​at a specific point in time during the month. Here, as the data organization, only indexing (or segment organization) may be newly performed, or in more advanced cases, the storage space itself may be changed and re-saved.

[0102] The distributed processing / storage described above can provide fast performance through distributed extraction in future data extraction. Data within a reference date (one month) is frequently extracted, so it is stored on a daily basis by tag. Data exceeding one month is stored on a monthly basis by tag. This data can be conveniently utilized for association analysis or AI development.

[0103]

[0104] Figure 5 is a conceptual diagram expressing the function of the development support model (200) of Figure 2 as an OPC UA information model for development AI support.

[0105] The power plant OPC UA information model (200) consists of asset nodes (220) and association group nodes (240).

[0106] The above Asset Nodes (220) construct power plant equipment and tag information, and the Associated Group Nodes (240) reflect group-specific tag information based on association analysis results. Maintenance (overhaul) information for each facility is linked from the power generation operation system and reflected in the properties of the corresponding Asset Node. In other words, Asset Nodes (220) are configured to apply power generation facilities according to the OPC UA protocol.

[0107] The above-mentioned association group Nodes (240) are operated in a manner that constantly performs association analysis to detect sensor abnormalities and reflects them in the properties of the association group Nodes.

[0108]

[0109] Figure 6 is a conceptual diagram schematically illustrating the interrelationships of various support methods that can be performed in a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis.

[0110] Figure 7 is a flowchart illustrating a development data distribution processing method (procedure) among the support methods of Figure 6.

[0111] The method for distributed processing of power generation data may include a step (S110) of acquiring power generation data (values ​​generated from sensors installed in each power generation facility) from a power plant DCS; a step (S140) of storing the acquired data as a temporary tag-based file; a step (S150 to S170) of integrating values ​​for a day at a specific point in time during the day for values ​​generated from one sensor (or tag), organizing the data for distributed processing, and storing it as a daily file; and a step (S160, S180) of integrating values ​​for a month at a specific point in time during the month, organizing the data for distributed processing, and storing it as a monthly file.

[0112] The method for distributing urban development data may further include a step (S120) of performing visual synchronization on the acquired development data.

[0113] For example, in the illustrated S110 step, power generation data can be linked from the power plant DCS through the OPC Server to acquire power generation data.

[0114] For example, the illustrated S120 step can be performed by acquiring data as a synchronized TimeStamp for each linked module through PTP time synchronization.

[0115] For example, in the illustrated S140 step, as an initial step for distributed storage of development data, real-time data can be temporarily stored in file form based on a large-capacity time series database.

[0116] For example, in the illustrated S150 step, a set of values ​​generated from one sensor or tag over a predetermined period of time can be stored as unit data in a time series database (DB).

[0117] According to steps S160 to S180 in the illustrated flowchart, data within a daily standard (one month) can be stored in a daily file by tag, and data exceeding one month can be stored in a distributed manner in a monthly file by tag.

[0118]

[0119] Figure 8 is a flowchart illustrating a method (generation procedure) for creating an information model for development AI support among the support methods of Figure 6.

[0120] A method for creating an information model for AI support of power generation, which is described as a kind of information model creation procedure, may include a step of constructing an OPC UA information model having data objects for power generation facilities (S210); a step of securing a large amount of history data as power generation data composed of values ​​generated from sensors mounted on each power generation facility (S220); a step of training an association analysis model with the large amount of history data (S230); a step of deriving an association group by tag for sensors mounted on each power generation facility using the association analysis model (S250); and a step of reflecting information on the association group by tag in the OPC UA information model (S260).

[0121] The method for creating an information model for urban development AI support may further include a step (S240) of checking the accuracy of the association analysis model that has been trained up to a point in time in step S230; and a step (S245) of distributing the association analysis model trained to have a satisfactory level of accuracy as a result of performing step S240 for real-time field use.

[0122] For example, in the step (S210) of constructing the OPC UA information model, asset Nodes (220) of FIG. 5 consisting of data objects for power generation facilities can be constructed according to a predetermined OPC UA developer support form.

[0123] For example, in the step (S260) of reflecting information on the association group by tag, the association group Nodes (240) of Fig. 5, which are composed of group objects representing the association groups of sensors mounted on each power generation facility, can be configured.

[0124] Depending on the implementation, the method for creating an information model for AI support of urban power generation may further include a step (S320) of acquiring overhaul information for each power generation facility; and a step (S360) of recording the overhaul performance in an object (Object) of the specific power generation facility of the asset Nodes (220) when it is confirmed that an overhaul has been performed on the specific power generation facility (S340).

[0125] Specifically, the process of creating OPC UA asset nodes and reflecting overhaul information in the flowchart of Fig. 8 is exemplified. In step S210, an OPC UA information model including power plant facility information can be constructed. Next, in step S320, overhaul information for each facility is periodically acquired from the power generation operation system on a daily basis, and in step S360, the node of the facility where overhaul has occurred can activate (reflect) the corresponding 'overhaul' attribute.

[0126] Specifically, the OPC UA association group node creation process in the flowchart of Fig. 8 is illustrated. In step S220, large-capacity historical data is extracted from the power generation data distribution storage unit, and in step S230, learning can be performed by performing association analysis using the extracted power plant tag information. At this time, the reference value can be adjusted when performing the association analysis to create an appropriate number of groups.

[0127] In the above step S250, the model for which learning has been completed is distributed for performing association analysis. In the above step S260, the model for which learning has been completed is used to derive an association group for each power plant tag, and OPC UA association group Nodes (240 in Fig. 5) can be created (written).

[0128]

[0129] FIG. 9 is a flowchart illustrating a sensor anomaly detection process (procedure) based on association analysis that can be included in or performed subsequently to the method for creating an information model for development AI support of FIG. 8.

[0130] The sensor anomaly detection process based on the illustrated association analysis may include a step (S270) of performing association analysis on real-time data developed in a time-series distributed DB using a learned association analysis model; a step (S275) of storing the association analysis result and searching for tag information that can be compared to the data on which the association analysis was performed from tags by group of association group Nodes in the OPC UA information model; a step (S280) of comparing the searched tag information with the association analysis result; and a step (S290) of recording sensor anomaly information for the sensor that generated the corresponding tag information if the tag information and the association analysis result do not match (S285).

[0131] The sensor anomaly detection process according to the illustrated flowchart performs real-time data correlation analysis of the development of a time-series distributed DB (S270), stores the correlation analysis results (S275), compares them with group-specific tag information of association group Nodes in the OPC UA information model (S280), and if there is a mismatch (S285), activates the 'sensor anomaly' property of the tag in which a sensor anomaly is detected (S290).

[0132] In the above step S290, for example, as shown in FIG. 5, constant sensor abnormality information can be recorded in the associated group Nodes (240).

[0133]

[0134] Figure 10 is a flowchart illustrating an example of a method for supporting AI analysis of large-capacity power generation data that can be performed in a large-capacity power generation sensor data AI analysis support system.

[0135] A method for supporting AI analysis of large-capacity power generation data may include a step (S420) of searching for a facility object of an attribute (keyword) entered by an AI developer for search (S410) in an asset node of an OPC UA information model; a step (S440) of extracting a list of tags related to each facility object; a step (S450 to S470) of searching for an association group for each tag in an association group node of the OPC UA information model to extract an association group object (Obejct); and a step (S480) of extracting data for AI analysis from a power generation data storage unit based on objects (tag lists) extracted from the asset node and the association group node.

[0136] Depending on the implementation, the method for supporting AI analysis of large-capacity power generation data may further perform a step (S490) of utilizing the extracted large-capacity power generation data for AI analysis after the step S480.

[0137] Depending on the implementation, if the equipment object obtained as a result of the inquiry in step S420 is an overhaul object, a step of filtering it (S430) may be further included; and / or a step of checking whether the inquiry according to the request in step S420 is complete (S435).

[0138] Depending on the implementation, the steps (S450 to S470) of extracting an associated group object by searching the associated group may include a step (S460) of filtering out a group object (Object) obtained as a result of the search in step S450 if the tag 'sensor abnormality' is included therein; and / or a step (S465) of checking whether the search according to the request in step S450 is completed.

[0139] For example, in step S470, the list of tags extracted in step S440 as an OPC UA asset definition can be returned together with the list of tags extracted in step S460 as an association group definition and filtered in step S465.

[0140] For example, a method for supporting AI analysis of large-scale power generation data involves processes related to asset node querying. When an AI developer enters attribute keywords for querying AI learning data (S410), equipment objects are extracted from OPC UA asset nodes using the keywords (S420). At this time, objects identified as "overhaul" can be excluded from the searched objects (S430, S435).

[0141] For example, a method for supporting AI analysis of large-scale urban power generation data includes processes related to searching for associated group Nodes, extracting a list of tags belonging to the searched facility Object (S440), searching for associated group Nodes for each tag (S450), extracting an associated group Object (S460), and extracting tag information included in the extracted group Object (S460). - Among the extracted tags, tags with 'sensor abnormality' can be excluded.

[0142] For example, a method for supporting AI analysis of large-capacity power generation data includes processes for utilizing AI analysis of large-capacity power generation data, returning a list of tags extracted from a process according to OPC UA asset definition and a process according to association group definition (S470), and extracting data for AI analysis at high speed from a distributed storage unit of power generation data based on the returned tag list (S490).

[0143]

[0144] Those skilled in the art should understand that the present invention can be implemented in other specific forms without changing the technical spirit or essential characteristics thereof, and therefore, the embodiments described above are illustrative in all respects and not restrictive. The scope of the present invention is indicated by the claims below rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be interpreted as being included within the scope of the present invention.

[0145]

[0146] The present invention relates to a large-capacity power generation sensor data AI analysis support system equipped with an OPC UA information model based on association analysis, and can provide a means for efficiently managing large-capacity tag or IoT data continuously produced during the power generation process.

Claims

1. A time series distributed processing unit that manages distributed indexing of sensor data generated from each sensor installed in a power facility; A development support model for creating a support list for sensor data existing in the above time series distributed processing unit in a form for developer support; and An association analysis model that analyzes whether each item in the above support list is related to other items. A large-capacity power generation sensor data AI analysis support system including .

2. In paragraph 1, A sensor anomaly management unit that detects anomalies in each sensor that generates sensor data using the above association analysis model. A large-capacity development sensor data AI analysis support system that includes more.

3. In paragraph 1, A learning unit that extracts power plant sensor data history information for a certain period from a large-capacity time series database and trains the above-mentioned association analysis model. A large-capacity development sensor data AI analysis support system that includes more.

4. In paragraph 1, The above time series distribution processing unit is, A large-capacity power generation sensor data AI analysis support system that stores and manages a set of values ​​generated over a set period of time from a single sensor installed in power generation equipment as unit data in a DB.

5. In paragraph 4, The above time series distribution processing unit is, A large-capacity power generation sensor data AI analysis support system that performs data organization for distributed processing by integrating daily values ​​at a specific point in time during the day and integrating monthly values ​​at a specific point in time during the month.

6. In paragraph 1, The above development support model is, Asset nodes consisting of data objects for power generation facilities, according to a prescribed developer support form; and An association group node consisting of group objects representing the association groups of sensors installed in each power generation facility. A large-capacity power generation sensor data AI analysis support system including .

7. In paragraph 6, The above asset node is, A large-capacity power generation sensor data AI analysis support system that records overhaul information for each power generation facility.

8. In paragraph 6, The above association group node is, A large-capacity power generation sensor data AI analysis support system that reports sensor abnormalities in specific sensors.

9. In paragraph 1, A linkage module that acquires sensor data of power plant power equipment from an OPC server connected to DCS, and acquires data as a TimeStamp through PTP time synchronization. A large-capacity power generation sensor data AI analysis support system that includes more.

10. Step of acquiring power generation data (values ​​generated from sensors installed in each power generation facility) from the power plant DCS; A step of saving the acquired data as a temporary tag-based file; A step of integrating the values ​​generated from one sensor at a specific point in time during the day, organizing the data for distributed processing, and storing it as a daily file; and A step to consolidate one month's worth of values ​​at a specific point in time during the month, organize the data for distributed processing, and save it as a monthly file. A method for distributed processing of large-capacity power generation sensor data including:

11. In paragraph 10, A step of performing visual synchronization on the acquired development data; A method for distributing large-capacity power generation sensor data including:

12. Step of building an OPC UA information model with data objects for power generation facilities; A step of securing large-capacity historical data as power generation data consisting of values ​​generated from sensors installed in each power generation facility; A step of training an association analysis model using the above large-capacity history data; A step of deriving a tag-specific association group for sensors installed in each power generation facility using the above association analysis model; and A step of reflecting information about the association group by tag in the above OPC UA information model. How to create an OPC UA information model for development AI support, including .

13. In paragraph 12, A step of verifying the accuracy of the above-mentioned association analysis model that has performed learning; and The step of deploying the trained association analysis model to achieve a satisfactory level of accuracy for real-time field use. How to create an OPC UA information model for advanced AI support that includes more.

14. In paragraph 12, In the step of building the above OPC UA information model, A method for creating an OPC UA information model for power generation AI support, which comprises an asset node of the OPC UA information model consisting of data objects for power generation facilities, according to a prescribed OPC UA developer support form.

15. In paragraph 14, Step of obtaining overhaul information for each power generation facility; When it is confirmed that an overhaul has been performed on a specific power generation facility, a step of recording the overhaul performance on the object of the specific power generation facility in the asset node How to create an OPC UA information model for advanced AI support that includes more.

16. In paragraph 14, In the step of reflecting information about the association group by the above tag, A method for creating an OPC UA information model for power generation AI support, which comprises an association group node of the OPC UA information model composed of group objects representing an association group of sensors mounted on each power generation facility.

17. In paragraph 16, A step of performing association analysis on real-time data developed in a time-series distributed DB using a learned association analysis model; A step of saving the results of association analysis and searching for tag information that can be compared to data on which association analysis was performed from tags for each group of the association group node; A step of comparing the searched tag information with the above association analysis results; and If the above tag information and the above association analysis result do not match, a step of recording sensor abnormality information for the sensor that generated the tag information. How to create an OPC UA information model for development AI support, including .

18. A step for the AI ​​developer to search for the equipment object with the properties entered for search in the asset node of the OPC UA information model; A step of extracting a list of tags related to each facility object; A step of extracting an association group object (Obejct) by searching for an association group for each tag in the association group node of the above OPC UA information model; and A step of extracting data for AI analysis from the development data storage unit based on objects extracted from the above asset node and the above association group node. A method for supporting AI analysis of large-capacity power generation sensor data including .

19. In paragraph 18, Steps to utilize extracted large-scale power generation data for AI analysis A method for supporting AI analysis of large-capacity development sensor data including more.

20. In paragraph 18, Step for filtering out equipment objects obtained as a result of a search if they are overhaul objects. A method for supporting AI analysis of large-capacity development sensor data including more.

21. In paragraph 18, Step for filtering out group objects (Objects) obtained as a result of a search if they contain a tag that is 'sensor abnormal'. A method for supporting AI analysis of large-capacity development sensor data including more.

Images