Method and processing system for asset performance management, and electric power system

Abstract

Techniques for performing asset performance management for assets (12, 13, 14) of an electric power system (10) are provided. A processing system (30) is operative to perform natural language processing of a query and to perform a function determined from a set of several functions supported by the processing system (30), with the function being determined based on the query.

Description

[0001] METHOD AND PROCESSING SYSTEM FOR ASSET PERFORMANCE MANAGEMENT, AND ELECTRIC

[0002] POWER SYSTEM

[0003] TECHNICAL FIELD

[0004] Embodiments of the invention relate to methods and processing systems for asset performance management. Embodiments of the invention relate in particular to methods and processing systems that facilitate asset performance management for assets of an electric power system.

[0005] BACKGROUND

[0006] Electric power systems are important infrastructure components. Examples of such electric power systems comprise electric power generation, transmission, and / or a distribution systems. Such electric power systems comprise a plurality of assets. Many of these assets are designed to remain in field operation for longer periods, such as in excess of several years or even several decades. The failure of an asset of an electric power system can have potentially catastrophic consequences. Thus, it is important that the assets be monitored. For illustration, it is desirable to detect potentially adverse effects of aging and / or operating conditions to which some of the assets have been subjected to take corrective action prior to asset failure. Important examples of such electric power system assets comprise transformers, transformer accessories, or other electric power system assets, without being limited thereto.

[0007] Management of the assets is a potentially complex task, owing to the complexity of the electric power systems, the interaction of its assets, and the potentially complex influence of operation conditions and measured parameters on asset state.

[0008] To assist an electric power system operator or engineer in the complex task of performing asset performance management, an asset performance management system may be used. EP 3 514 908 Al, WO 2021 / 170859 Al, and EP 4 133 294 Al disclose techniques useful for performing asset performance management for assets of electric power systems. Such asset performance management systems may provide data-driven health and performance insights to optimize asset lifecycle performance, mitigate a risk of asset failure and reduce unplanned downtime.

[0009] An asset performance management system is generally a complex system that requires a certain level of expertise to be able to interpret all the information provided by it. Conventionally, the asset performance management system does not provide a way to easily interact in a question / answer form with the knowledge base encapsuled in it, which creates a barrier for users (such as electric power system operators or engineers) with less expertise in the interpretation of this information. Thus, it may be challenging for users to easily obtain answers to questions they have about their fleet, laying the burden on the user to know how to correctly operate the asset performance management system.

[0010] This burden also creates a risk of human-induced error in operating an electric power system.

[0011] Thus, there is still a need for improved techniques of performing asset performance management. There is, in particular, a need for techniques that facilitate interaction with an asset performance management system, reducing or eliminating the risk of human-induced error in the interaction with the asset performance management system. There is also a need for techniques that provide access to the large knowledge base encapsulated in an asset performance management system.

[0012] SUMMARY

[0013] It is an object of the invention to provide methods and processing systems that provide enhanced techniques of performing asset performance management. There is, in particular, a need for techniques that facilitate interaction with an asset performance management system, thereby contributing to reducing or eliminating the risk of human-induced error in the interaction with the asset performance management system.

[0014] According to embodiments, methods and processing systems as recited in the claims are provided. The dependent claims define preferred or advantageous embodiments.

[0015] Embodiments of the invention provide a processing system that is operative to allow a user (e.g., an electric power system operator or engineer) to interact therewith in a question / answer basis. This configuration of the processing system allows the user to provide queries to the processing system and obtain answers based on the real-time data from a fleet of assets of an electric power system and on existing engineering models available for monitoring the assets.

[0016] According to an aspect of the invention, there is provided a method of performing asset performance management for assets of an electric power system. The method comprises receiving, by a processing system, a query. The method comprises processing, by the processing system, the query, comprising performing natural language processing of the query, wherein processing the query comprises performing a classification of the query to determine a function to be performed by the processing system responsive to the query, wherein the function is determined from a set of several functions supported by the processing system, wherein the processing system is operative to obtain state data for the assets in an ongoing manner during field operation of the assets to perform some or all of the several functions supported by the processing system. The method comprises processing, by the processing system, data retrieved to perform the determined function responsive to the query. The method comprises generating, by the processing system, output based on a result of the processing of the retrieved data.

[0017] Various effects and advantages are associated with the method. The method facilitates interaction of a user with an asset performance management system. By providing a processing system, which may be comprised by or which may be the asset performance management system, supporting a set of several functions, various different functions may be selected from the set of functions based on the query. With the query being processed using natural language processing (NLP), such as by using a large language model (LLM), the interaction with the processing system may be performed using natural language input, such as spoken input or textual input. This input is referred to as query herein. By providing the processing system that interprets the query, determines which function is to be performed, and performs the determined function using, inter alia, asset status data for the assets that are continually updated during field operation, the risk for human-induced error is reduced. The processing system also provides improve techniques that make it easier for a user to access the large knowledge base of an asset performance management system.

[0018] Thus, the processing system affords an easy way for the user (e.g., an electric power system operator or an engineer) to interact with the processing system that performs asset performance management (APM).

[0019] The processing system may be operative to perform various functions including, without limitation: queries relating to an asset state of one or several of the assets; queries relating to a hypothetical scenario, and more particularly to the asset state of the one or several assets under the assumptions of the hypothetical scenario; queries relating to domain-specific questions; queries relating to assistant tasks (such as setting reminders, alarms, or other asset-specific tasks) for one or several of the assets.

[0020] Thereby, the processing system can provide various functions useful in association with assets of an electric power system, with the functions being selectable by an input (the query) that is processed using NLP (e.g., using an LLM) and which, therefore, may be formulated as a question or other spoken or textual input that is processed by the processing system for determining the function to be performed, including attributes or parameters (such as identifiers for assets and / or identifiers for parameters) to be used in performing the function.

[0021] The processing system may be operative to execute an expert system, such as an expert system for transformer state analysis, for at least perform the functions of providing information on a current asset state and providing information on an asset state for a hypothetical scenario.

[0022] Thereby, the processing system can harness and utilize existing expert systems that are available to the skilled person for a wide variety of electric power system assets. The processing system according to the invention provides an advanced and technically beneficial way of interacting with the processing system, including invoking the expert system for performing functions relating to an asset state and / or accessing a knowledge base present in an APM system.

[0023] The set of several functions supported by the processing system may comprise: an asset state analysis and answering at least one query relating to a hypothetical scenario. Thereby, the processing system can provide functions that relate to both a current, actual asset state and to the asset state that would result for the hypothetical scenario.

[0024] The processing system may be operative such that the asset state analysis provides information on one, several, or all of: cooling; aging; bushing; dissolved gas analysis (DGA); moisture; overload, without being limited thereto.

[0025] Thereby, the processing system can provide, in response to the query, information relevant to the current state of the asset and, optionally, recommendations for maintenance operations that may be required.

[0026] The processing system may be operative such that the asset state analysis provides information on changes in one, several or all of: cooling; aging; bushing; dissolved gas analysis (DGA); moisture; overload; with the changes being respectively determined for the hypothetical scenario as compared to the current (actual) asset state, without being limited thereto.

[0027] Thereby, the processing system can provide, in response to the query, information specifying how various relevant quantities would change for the hypothetical scenario as compared to the current, actual scenario.

[0028] The retrieved data may comprise the state data obtained for the assets, wherein generating the output may comprise using, by the processing system, the state data obtained for the assets to perform the asset state analysis or for answering the at least one query relating to the hypothetical scenario.

[0029] Thereby, the asset state data that is obtained continually and during field operation may be utilized for both providing information on the current state of the asset and for answering at least one query relating to the hypothetical scenario.

[0030] The method may further comprise controlling, by the processing system, a human machine interface (HMI) to enable user input that specifies the hypothetical scenario.

[0031] Thereby, the processing system may determine automatically whether it has all information required for answering the question relating to the hypothetical scenario. If the processing system determines that additional information is required, it can cause the HMI to enable the inputting of the required additional information.

[0032] Controlling the HMI to enable the user input may comprise controlling, by the processing system, the HMI to enable the user input to be provided in a question / answer manner.

[0033] Thereby, the processing system allows the user to interact therewith in a natural manner.

[0034] In this manner, the processing system can assist and guide the user (e.g., the electric power system operator or engineer) to provide all information required to perform the state analysis not only for the current scenario but also for the hypothetical scenario.

[0035] Answering the at least one query relating to the hypothetical scenario may comprise performing a first asset state analysis for a current asset state and performing a second asset state analysis for the hypothetical scenario. Performing the state analyses may respectively comprise using an expert system executed by the processing system to perform the state analyses.

[0036] Thereby, the processing system may utilize and harness the benefits of the expert system that is available to the skilled person for a wide variety of electric power system assets to determine the changes that would result for the hypothetical scenario, as compared to the current scenario. The processing system facilitates this access to the expert system by using NLP (e.g., using an LLM) of the query, thereby providing easier access to the capabilities of the processing system for performing APM.

[0037] Generating the output may comprise generating the output based on a first result of the first asset state analysis and based on a second result of the second asset state analysis.

[0038] Thereby, the processing system can provide an answer to a query relating to the hypothetical scenario, taking into account the changes in various asset-related quantities (such as cooling, bushings, DGA, etc. for a transformer) based on a result of the state analyses.

[0039] Answering the at least one query relating to the hypothetical scenario may further comprise performing a comparison of the first result of the first asset state analysis and the second result of the second asset state analysis.

[0040] Thereby, the processing system can provide an answer to a query relating to the hypothetical scenario, taking into account the comparison of the state analyses, thereby determining the changes in various asset-related quantities (such as cooling, bushings, DGA, etc. for a transformer).

[0041] Generating the output may comprise generating the output based on a result of the comparison.

[0042] Thereby, the processing system can provide an answer to a query relating to the hypothetical scenario, taking into account the comparison of the state analyses, thereby providing the changes in various asset-related quantities (such as cooling, bushings, DGA, etc. for a transformer).

[0043] The processing system may be operative to execute an expert system to perform the asset state analysis. The processing system may be operative to generate input for the expert system based on the NLP of the query and to generate the output based on an expert system output of the expert system.

[0044] Thereby, the processing system can provide access to the capabilities of the expert system in an efficient manner that is less error-prone by interfacing with the user in a manner which enables the user to provide queries in spoken or written form, without requiring the user to adhere to a specific syntax of the expert system. A conversion of the query to the input to the expert system is performed by the processing system, thereby mitigating the risk of human-induced error.

[0045] The processing system may be operative to access historical data relating to the assets and to use the historical data for performing the determined function.

[0046] Thereby, the processing system is enabled to use the historical data that may be collected in an ongoing manner (i.e., continually) during field operation of the electric power system for performing the determined function. The processing system enables use of this historical information by virtue of the novel techniques of interfacing the user with the processing system suggested herein, making it easier for the user to or harness the knowledge included in the historical data. The historical data may comprise historical measurements (such as historical measurements of loads, other electrical characteristics, temperatures, DGA results, etc.) obtained for the assets. The historical data may be used to determine the effects of hypothetical scenarios or for performing the asset state analysis for the current asset state, for example by performing a multi-variate regression or adjusting parameters of a trained artificial intelligence (Al) model.

[0047] The historical data may comprise operating parameters relating to the assets at a time in the past. The operating parameters may comprise measurements (such as historical measurements of loads, other electric characteristics, temperatures, DGA results, etc.) obtained for the assets.

[0048] Thereby, the processing system is enabled to use the historical parameters relating to the assets for performing the set of the functions, while allowing the user to interact with the processing system and thereby harness the historical data in a more natural manner than conventional APM systems.

[0049] The historical parameters may comprise one, several or all of the following historical data: insulation fluid quality; accessory-related data; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions, without being limited thereto.

[0050] Thereby, the processing system is enabled to use the historical parameters that are of particular relevance to transformers (such as transmission transformers, distribution transformers, voltage or current measurement transformers) or other electric power system assets that may or may not have an insulation fluid, such as an insulation oil. The processing system makes it easier for the user to harness the benefits of such historical data by allowing the user to interact with the processing system in a more natural way and without requiring specific syntax that would otherwise be required for legacy APM systems.

[0051] The historical parameters may comprise accessory-related data, comprising data relating to at least one of a bushing, a transformer breather, a tap changer, a cooling system.

[0052] Thereby, the processing system is enabled to use the historical parameters that are of particular relevance to electric power system assets that have accessory components associated therewith.

[0053] The assets may comprise transformers, wherein the asset state analysis and answering the at least one query relating to the hypothetical scenario relate to a transformer state of at least one of the transformers and / or to a transformer accessory state for at least one of the transformers.

[0054] Thereby, the processing system is enabled to perform several functions taking into account the state of the accessory equipment. Retrieving the data may comprise obtaining the state data for at least one asset or at least two assets. Generating the output may comprise generating an output aggregated from the obtained state data for the at least one asset or at least two assets.

[0055] This is of particular relevance of electric power systems, where aging and degradation of accessory components of assets may proceed differently than aging and degradation of the assets with which the accessory components are associated. Thus, the processing system according to the invention and the methods executed using the processing system facilitate performing functions that take into consideration accessory components of assets of the electric power system. In particular, information relating to aging and degradation of such accessory components may be accessed more readily using the processing system and the methods disclosed herein.

[0056] The state data obtained for the assets comprises state data relating to one, several, or all of the following: dissolved gas concentrations of one or several transformers; oil reclaim data for one or several transformers; fault analysis results for one or several assets; nameplate data for one or several assets; repair history data for one or several assets; load data; insulation data; factory acceptance tests; insulation fluid quality; accessory-related data, optionally comprising data relating to at least one of bushings, transformer breathers, tap changers, cooling system; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions.

[0057] Thereby, the processing system is enabled to take into consideration various relevant quantities that are of particular relevance to APM for electric power systems, while affording ease of interaction with the processing system and reducing the risk of human-induced error.

[0058] The set of several functions supported by the processing system may comprise at least one function executable independently of the state data obtained for the assets.

[0059] Thereby, the user of the processing system can trigger one or several functions that do not necessarily require state data obtained for the assets using the query. These functions may still relate to the assets, such as by retrieving domain knowledge for one or several of the assets and / or performing assistance functions that relate to the assets that do not necessarily depend on the asset state.

[0060] The at least one function executable independently of the state data obtained for the assets may comprise provision of domain-specific information.

[0061] Thereby, the retrieval of domain-specific information is afforded by the processing system. This is particularly useful for electric power systems, in view of the wide variety of different asset types and / or the potentially complex interplay of different assets combined in one electric power system. In this case, the generated output may comprise output determined based on the domain-specific information. The at least one function executable independently of the state data obtained for the assets may comprise an assistance function related to at least one of the assets.

[0062] Thereby, assistance functions that relate to at least one of the assets for which the asset state is not indispensable may be accommodated by the processing system responsive to the query that is processed using NLP (e.g., using an LLM). This facilitates performing such assistance functions, which may include the setting of asset-specific reminders, alerts, or notifications, having a bearing on electric power system operation. The risk of human-induced error is mitigated by performing such assistance functions responsive to NLP processing of the query. In this case, the generated output may comprise commands that cause asset-specific reminders, alerts, or notifications to be generated in a database and output at a later stage based on the database entries. The generated output may also comprise an acknowledgment that the assistance function has been performed in accordance with the received query.

[0063] The processing system may be operative to maintain and update a knowledge base of the domainspecific information, comprising updating, by the processing system, the knowledge base of the domain-specific information based on information retrieved via a wide area network (WAN).

[0064] Thereby, the processing system may automatically enhance the knowledge base accessible responsive to NLP (e.g., using an LLM) of the query. The processing system may be operative to perform a crawling operation for resources accessible over the WAN (such as the internet), which may comprise any one, several or all of: new standard documents (such as standard documents), scientific publications, patent literature. The processing system may be operative to enhance the knowledge base based on such publicly available documents, with the processing system being operative to analyze the documents for inclusion in the knowledge base using, e.g., at least one trained Al model.

[0065] The processing system may be operative to use at least one trained Al model (such as a large language model (LLM) or another Al model comprising at least one attention mechanism) to update the knowledge base (e.g., by analyzing documents and generate data for inclusion into the knowledge base). The processing system may also be operative to use at least one trained Al model (such as an LLM or another Al model comprising at least one attention mechanism) to process the information from the knowledge base and generate the output.

[0066] Thereby, the processing system may apply Al-based techniques to provide functionalities such as enabling or facilitating access to a knowledge base of an asset management system.

[0067] Alternatively or additionally, the processing system may be operative to maintain a knowledge base of domain-specific information, comprising updating the knowledge base based on proprietary information of the electric power system operator as the proprietary information becomes available during field operation of the electric power system. Thereby, the processing system may automatically enhance the knowledge base accessible responsive to NLP of the query. The processing system may be operative to monitor proprietary data of the electric power system operator (such as fleet-related data) for inclusion into the knowledge base.

[0068] The processing system may be operative such that the set of several functions supported by the processing system comprises: an asset state analysis; answering at least one query relating to a hypothetical scenario; provision of domain-specific information; an assistance function related to at least one of the assets.

[0069] Thereby, the processing system can offer a multitude of functions that can be performed using, e.g., LLMs for various purposes. A processing system that supports at least these mentioned functions provides significantly improved access to an APM system, e.g., to its knowledge base, and to processing performed using the APM system data.

[0070] The query may comprise spoken input.

[0071] Thereby, the processing system is operative to enable the user to interact with the functions supported by the processing system using spoken input. The performance of one or several functions responsive to such spoken input is particularly time efficient and mitigates the risk of mistyping, thereby further reducing the risk of human-induced errors.

[0072] The query may comprise written input.

[0073] Thereby, the processing system is operative to enable the user to interact with the functions supported by the processing system using textual input. Such a form of interaction may be preferable to various uses, for example in view of a potential background noise levels and / or for security reasons.

[0074] The processing system may be operative to use at least one trained Al model to perform at least one of the supported functions. The at least one trained Al model may comprise or may be an LLM. The at least one trained Al model may comprise a stack of one or several self-attention mechanisms or a stack of attention blocks respectively comprising an attention mechanism and optional additional processing before or after the attention mechanism. The at least one trained Al model may comprise a stack of multi-head self-attention mechanisms or a stack of attention block respectively comprising a multi-head self-attention mechanism and optional additional processing before or after the attention mechanism. The processing system may be operative to use Al to process the query, to retrieve and process data from the knowledge base or otherwise from the asset performance management system, and / or to create the output.

[0075] Thereby, the processing system can use Al-based techniques to perform at least some of the functions supported by the processing system.

[0076] The method may comprise performing asset performance management for at least transformers of the electric power system. Thereby, the advantages and effects disclosed herein are attained specifically for at least the transformers of the electric power system, which are particularly complex assets and failure of which may have particularly detrimental effects on electric power system operation.

[0077] The method may further comprise causing, by the processing system or a control system communicatively coupled to the processing system, execution of at least one control action for at least one of the assets based on the generated output.

[0078] Thereby, the results of the performed determined function may be used to execute one or several control actions acting on the electric power system. The one or several control actions may comprise control actions that affect the primary system equipment of a primary system of the electric power system. The results of the determined function may be used to automatically execute one or several control actions or may be used to request authorization for performing the one or several control actions, optionally depending on the criticality of the control actions to be performed.

[0079] According to another aspect of the invention, there is provided a method of facilitating interaction with an asset performance management (APM) system, wherein the method comprises using at least one large language model (LLM) to facilitate the interaction of the user with the APM system.

[0080] Thereby, the user is enabled to interact with the APM system in a more natural manner.

[0081] Using the at least one LLM may comprise using the LLM to process a query received by the user, accessing data stored by the APM system, processing data stored by the APM system, and / or generating output in response to the query.

[0082] According to another aspect of the invention, there is provided a control method of controlling an electric power system, comprising performing, by a control system, at least one control action based on the output provided by on the output provided by the processing system for performing the APM.

[0083] Thereby, the risk of human-induced errors is mitigated in performing electric power system control operations.

[0084] According to another aspect of the invention, there is provided a processing system for performing asset performance management for assets of an electric power system, the processing system comprising: at least one interface operative to receive a query; and at least one processing circuit operative to process the query, wherein the at least one processing circuit is operative to perform natural language processing of the query, wherein the at least one processing circuit is operative such that processing the query comprises performing a classification of the query to determine a function to be performed by the processing system responsive to the query, wherein the at least one processing circuit is operative such that the function is determined from a set of several functions supported by the processing system, wherein the processing system is operative to obtain state data for the assets in an ongoing manner during field operation of the assets to perform some or all of the several functions supported by the processing system; process data retrieved to perform the determined function responsive to the query; and generate output based on a result of the processing of the retrieved data.

[0085] Various effects and advantages are associated with the processing system for performing APM. The processing system facilitates interaction of a user therewith. By supporting a set of several functions, various different functions may be selected from the set of functions based on the query. With the query being processed using NLP (e.g., using an LLM), the interaction with the processing system may be performed using natural language input, such as spoken input or textual input. This input is referred to as query herein. By providing the processing system that interprets the query, determines which function is to be performed, and performs the determined function using, inter alia, asset status data for the assets that are continually updated during field operation, the risk for human- induced error is reduced.

[0086] The processing system may be operative to perform the method according to any aspect of embodiment disclosed herein. The processing operations may be performed using the at least one processing circuit. The processing operations may be performed automatically by the at least one processing circuit. Accordingly, there is also disclosed the processing system as mentioned above, with the at least one processing circuit being operative to perform the optional features according to the method of any one of the embodiments disclosed herein. The technical effects attained by the optional features of the method correspond to the effects discussed in association with the optional method features.

[0087] The processing system may comprise at least one storage system accessible to the at least one processing circuit. The at least one processing circuit may be operative to store asset state data in the storage system. Alternatively or additionally, the processing system may be operative such that the at least one processing circuit can maintain and update a knowledge base stored in the storage system, with the knowledge base including domain specific information knowledge.

[0088] An electric power system according to an embodiment comprises a plurality of assets and the processing system according to an aspect or embodiment for performing asset performance management for some or all assets of the plurality of assets.

[0089] Thereby, an electric power system is provided in which interaction of a user (such as an electric power system operator or engineer) with a processing system that performs APM functions is facilitated, and the risk of human-induced error is reduced.

[0090] The electric power system may comprise an electric power generation, transmission, and / or distribution system.

[0091] Thereby, an electric power generation, transmission, and / or distribution system is provided in which interaction of a user (such as an electric power system operator or engineer) with a processing system that performs APM functions is facilitated, and the risk of human-induced error is reduced. The electric power system may comprise an electric power grid and / or at least one electric power grid substation.

[0092] Thereby, comprise an electric power grid and / or at least one electric power grid substation is provided in which interaction of a user (such as an electric power system operator or engineer) with a processing system that performs APM functions is facilitated, and the risk of human-induced error is reduced.

[0093] The plurality of assets may comprise a plurality of transformers. The processing system may be operative to perform the asset performance management for at least the plurality of transformers.

[0094] Thereby, the electric power system mitigates the risk of human-induced error at least for monitoring the transformers of the electric power system, which are particularly complex assets and failure of which may have particularly detrimental effects on electric power system operation.

[0095] The electric power system may further comprise an input device operative to capture the query. The input device may be separate from the processing system or integral with the processing system.

[0096] Thereby, an input modality for providing the query to the processing system is provided.

[0097] The input device may comprise a communication interface operative to provide the query to the processing system.

[0098] Thereby, the input device may be implemented separately from the processing system while being communicatively coupled thereto to provide the query to the processing system.

[0099] The input device may comprise an acoustoelectric transducer operative to capture the query. The acoustoelectric transducer may comprise a microphone.

[0100] Thereby, the input device allows the user to provide the query as spoken input.

[0101] The electric power system may further comprise at least one control circuit operative to control the electric power system based on the output provided by the processing system.

[0102] Thereby, a result of performance of the determined function responsive to NLP processing of the query may be used for one or several control actions acting on primary system equipment of the electric power system.

[0103] According to another aspect of the invention, there is provided machine-readable instruction code comprising machine-readable instructions which, when executed by at least one processing circuit, cause the at least one processing circuit to perform the method according to an aspect or embodiment of the invention.

[0104] The effects attained by the machine-readable instruction code correspond to the effects disclosed in association with the methods and processing systems according to various embodiments.

[0105] According to another aspect of the invention, there is provided non-transitory storage medium having stored thereon machine-readable instruction code comprising machine-readable instructions which, when executed by at least one processing circuit, cause the at least one processing circuit to perform the method according to an aspect or embodiment of the invention.

[0106] The effects attained by the non-transitory storage medium correspond to the effects disclosed in association with the methods and processing systems according to various embodiments.

[0107] The processing systems and methods can be used in association with an electric power grid or subsystems thereof, such as a power system substation, without being limited thereto.

[0108] BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Embodiments of the invention will be described with reference to the drawings in which similar or identical reference signs designate elements with similar or identical configuration and / or function.

[0110] Figure 1 is a schematic diagram of a processing system.

[0111] Figure 2 is a block diagram of at least one processing circuit of the processing system.

[0112] Figure 3 is a schematic representation of an electric power system comprising the processing system.

[0113] Figure 4 is a flow chart.

[0114] Figure 5 is a flow chart.

[0115] Figure 6 is a flow chart.

[0116] Figure 7 is a block diagram of an expert system of the processing system.

[0117] Figure 8 is a flow chart.

[0118] Figure 9 is a schematic representation of asset state data.

[0119] Figure 10 is a flow chart.

[0120] Figure 11 shows input and output of an HMI of an input device of or interacting with the processing system.

[0121] Figure 12 shows output of the HMI of the input device of or interacting with the processing system.

[0122] Figure 13 is a flow chart.

[0123] Figure 14 is a flow chart.

[0124] Figure 15 is a flow chart.

[0125] Figure 16 is a flow chart.

[0126] Figure 17 is a schematic partial representation of knowledge base of the processing system.

[0127] Figure 18 is a flow chart.

[0128] Figure 19 is a schematic representation of a fleet of assets.

[0129] Figure 20 is a schematic representation of an artificial intelligence model that may be employed by the processing system.

[0130] Figure 21 is a schematic representation of an asset in perspective view. Figure 22 is a schematic representation of the asset of Figure 21 in cross-sectional view.

[0131] Figure 23 is a view of a system comprising the processing system.

[0132] Figure 24 is a view of a system comprising the processing system.

[0133] Figure 25 is a view of a system comprising the processing system.

[0134] Figure 26 is a view of a system comprising the processing system.

[0135] Figure 27 is a view of a system comprising the processing system.

[0136] Figure 28 is a view of a system comprising the processing system.

[0137] Figure 29 is a block diagram representation of the processing system.

[0138] DETAILED DESCRIPTION OF EMBODIMENTS

[0139] Embodiments of the invention will be described with reference to the drawings. In the drawings, similar or identical reference signs designate elements with similar or identical configuration and / or function.

[0140] Embodiments relate to methods and processing systems useful in association with asset performance management (APM). More particularly, the invention provides processing systems and methods for performing APM that supported several different functions, with a selection of the function to be performed from the several supported functions being responsive to natural language processing (NLP) of a query, e.g., using an LLM. Thereby, the processing system for performing APM and the method performed using the processing system provide versatility in the types of functions that can be performed, while at the same time mitigating the risk of human-induced error in interacting with the APM.

[0141] As used herein, the term "asset performance management" refers to at least monitoring the assets. The APM may also comprise automatically performing and / or suggesting mitigating actions, based on the monitoring of the assets.

[0142] Performing APM may comprise using an expert system. The processing system may be operative to execute various expert systems, each associated with a different type of asset. In the field of electric power systems, expert systems for various types of assets are available to the skilled person. Implementations of APM functions to obtain an asset state are disclosed in, e.g., EP 3 514 908 Al, WO 2021 / 170859 Al, and EP 4 133 294 Al previously cited herein. While expert systems useful for obtaining asset state data are available to the skilled person, the present invention provides a processing system capable of invoking the expert system(s) responsive to NLP of a query (e.g., using a large language model (LLM)), thereby mitigating the risk of human-induced error in electric power system operation. Embodiments of the invention also provide techniques of assessing hypothetical scenarios that use expert systems, such as legacy expert systems, in a novel way based on NLP of a query, to assess the changes brought about by a hypothetical scenario as compared to a current operating scenario for the assets.

[0143] Thus, embodiments of the invention provide a way for the user to more easily interact with a processing system capable of monitoring assets, which is operative to support the interaction on a question / answer basis. This allows the user to simply ask questions or otherwise provide queries to the processing system and obtain answers based on real-time data from a fleet of assets of an electric power system and on the existing engineering models available in a legacy monitoring platform.

[0144] Thus, embodiments of the invention mitigate the challenges encountered by a user when handling a complex APM system that conventionally requires a certain level of expertise to be able to interpret all the information provided by it. Thus, embodiments of the invention mitigate the risk of human errors that results from the fact that conventionally users require a certain level of expertise to interpret the data presented to them, thus, creating a barrier for user who do not have that expertise.

[0145] Embodiments of the invention also make it easier for a user to access the large knowledge base of an asset performance management system.

[0146] Embodiments of the invention also provide advantages as compared to APM systems that wait for a problem to occur and then notify the users, indicating the issue and making recommendations for maintenance actions. This may not be viable in cases where the user is concerned and needs to know what could happen with the evolution of the data and the types of recommendations that would eventually come, as result of that.

[0147] Figure 1 shows a schematic block diagram of a processing system 30 for performing APM for assets of an electric power system.

[0148] The processing system 30 comprises one or several interfaces 31, 32 operative to interact with a human machine interface (HMI) device and the electric power system, and more specifically with secondary system components of the electric power system that may comprise measurement instrumentation or other intelligent electronic devices (lEDs). In the illustrated implementation, the processing system 30 may comprise at least one first interface 31 operative to receive a query 65. The query 65 may comprise or may represent textual or spoken input received via an HMI device communicatively coupled with or comprised by the processing system 30. The processing system 30 may comprise at least one second interface 32 operative to continually (i.e., in an ongoing basis) receive measurements 61 and / or other data from a secondary system of the electric power system, which secondary system comprises measurement instrumentation to capture the measurements. Other examples of data 62 received via the at least one second interface 32 may comprise data from a control system (e.g., from a Supervisory Control and Data Acquisition (SCADA) system) and / or eventbased messages, such as GOOSE messages compatible with or in accordance with IEC 61850 (in particular the version of IEC 61850 as in force on the filing or priority date of the application, whichever is earlier) and / or proprietary event-based messages.

[0149] The processing system 30 comprises a storage system 34. The storage system 30 may be physically integrated into the processing system 30 or may be provided separately therefrom, as long as it is accessible by the processing system 30. The storage system 34 may be operative to store therein asset state data, measurements, and other data related to the electric power system, preferably over a past time horizon to thereby provide historical data. The storage system 34 may be operative to store therein a knowledge base, which may comprise a repository of domain-specific knowledge relating to assets of an electric power system for which the processing system 30 performs APM.

[0150] The storage system 34 may also have stored therein parameter and / or hyperparameters of one or several artificial intelligence (Al) models that can be used to perform one, several, or all of: NLP of the received query 65; adding information to the knowledge base stored in the storage system 34; obtaining asset state data related to one or several assets (using, e.g., techniques such as those discussed in EP 3 514 908 Al).

[0151] The processing system 30 comprises at least one processing circuit 33. The at least one processing circuit 33 may comprise any one or any combination of integrated circuits, integrated semiconductor circuits, processors, controllers, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), circuit(s) including quantum bits (qubits) and / or quantum gates, without being limited thereto, to perform the operations discussed in detail herein.

[0152] The at least one processing circuit 33 is operative to perform NLP 41 of the received query 65. The NLP 41 may be performed using conventional techniques, such as a stack of attention mechanisms, e.g. a stack of self-attention mechanisms, operative to receive the query 65 and provide output that encodes the query 65. The NLP 41 may comprise a first stack of self-attention mechanisms and a feedforward network that, in combination, provide a transformer encoder structure, with the NLP 41 further comprising a transformer decoder structure comprising a second stack of self-attention mechanisms. Other techniques may be used to convert the query 65 into machine-readable data for further use by the at least one processing circuit 33. In particular, and without being limited thereto, the at least one processing circuit 33 may be operative to convert the query 65 into a series of codes, with each code representing a dictionary entry.

[0153] The at least one processing circuit 33 may be operative to determine a function to be performed responsive to the query 65, based on the NLP 41 of the query. This function selection 42 may determine one function for execution, with the function being determined from a predefined set of supported functions that can be performed by the processing system 30. The function determination 42 may be performed using a classifier, which receives the NLP processing 41 of the query 65 as input. Importantly, and as will be discussed in more detail below, the processing system 30 is operative such that any one of a set comprising a plurality of supported functions can be selected based on the query 65, enabling the user to interact with the processing system 30 by using natural textual or spoken input to the processing system 30, without having to worry about a specific syntax that may be required by the internal operations of the processing system 30 to, e.g., obtain an asset state based on an expert system and / or retrieve information from a knowledge base.

[0154] The processing system 30 may be operative to support at least the following set of functions: answering queries relating to an asset state of one or several of the assets; answering queries relating to a hypothetical scenario, and more particularly to the asset state of the one or several assets under the assumptions of the hypothetical scenario; answering queries relating to domain-specific questions; taking action responsive to queries relating to assistant tasks (such as setting reminders, alarms, or other asset-specific tasks) for one or several of the assets.

[0155] The processing system 30 is operative to perform an output generation 44 based on the executed function. The output generation 44 may comprise generating output based on a current asset state, a change in asset state that would occur for hypothetical operating conditions different from the current operating conditions, the provision of domain-specific information, and / or an operation that causes an asset-specific alert, warning, or reminder to be set. The processing system 30 may be operative such that the at least one processing circuit 33 generates the output to provide the output via the at least one first interface 31 and / or the at least one second interface 32. For illustration, the output may comprise control data 66 for controlling the HMI. Alternatively or additionally, the output may comprise controlled to 67 for causing at least one control action to be performed in the electronic power system responsive to a result of the function execution 43.

[0156] Thus, the at least one processing circuit 33 is operative to process the query. The at least one processing circuit 33 is operative to perform NLP 41 of the query. The at least one processing circuit 33 is operative such that processing the query comprises performing a classification 42 of the query to determine a function to be performed by the processing system responsive to the query. The at least one processing circuit 33 is operative such that the function is determined from a set of several functions supported by the processing system 30. The processing system 30 is operative to obtain state data for the assets in an ongoing manner during field operation of the assets to perform some or all of the several functions supported by the processing system 30. The at least one processing circuit 33 is also operative to process data retrieved to perform the determined function (which may in particular comprise the asset state data) responsive to the query (as processed using NLP 41), and generate output 66, 67 based on a result of the processing of the retrieved data. Figure 2 is a functional block diagram to further explain the operation of the at least one processing circuit 33 of the processing system 30. Other implementations may be used Alternatively or in addition to those described in association with Figure 2.

[0157] The NLP 41 of the query may comprise a semantic analysis 45. Techniques of analyzing a spoken or written input using processor-based implementations are available to the skilled person.

[0158] The function selection 42 may comprise a classifier 46 as applied to the NLP result. The classifier 46 may be operative to provide, at its output, a classification that assigns the received query to one of the functions supported by the processing system 30. Alternatively or additionally, the classifier 46 may have an output operative to provide probabilities of the query being associated with each one of the various functions supported by the processing system 30. The probabilities may then be used to determine which of the functions is to be executed.

[0159] The functions supported by the processing system may comprise one or preferably several functions that are asset-state dependent. Execution of these asset-state dependent functions 47 may involve use of at least one expert system 48. Different expert systems 48 may be provided that each have their respective analysis logic, depending on the type of asset.

[0160] The functions supported by the processing system may comprise one or preferably several functions for which determination of an asset state is not indispensable. Provision of domain-specific information and generic asset -related assistance functions are examples for such functions. Execution of such functions 49 that do not necessarily require the asset state may be performed in such a manner that it does not require execution of the expert system 48. Execution of the functions 49 may optionally also use the asset state, for example in order to provide domain-specific information that is specifically tailored to the asset in question (e.g., depending on the type of a transformer and / or its accessories and / or its state) or to provide asset-specific assistance functions that are consistent with the asset state.

[0161] The output generation 44 may comprise an interface control 50 of controlling the at least one first interface 31 to provide the output 66 and / or to control the at least one second interface 32 to provide the output 67.

[0162] Figure 3 is a schematic representation of an electric power electric power system 10 comprising the processing system 30. The electric power system 10 comprises a primary system 11 that may comprise assets of an electric power generation, transmission, and / or distribution system. The primary system 11 may comprise at least part of an electric power grid. The primary system 11 comprises assets that may comprise power transformers 12, 14, switchgear 13, or other assets that may be associated with a power transmission or power distribution line 15. A secondary system of the electric power system 11 comprises measurement instrumentation, such as current and voltage transformers 16, 17, dissolved gas analysis (DGA) sensors, and / or other sensors.

[0163] The electric power system 10 comprises an automation control system 20. The automation control system 20 comprises a plurality of devices 21, 22, 23. At least some of the devices 21, 22, 23 of the automation control system 20 may be operative to perform protection functions or other functions that involve a control of components of the primary system 11, such as switchgear 13 or a transformer 12, 14. At least some of the devices of the automation control system 20 may be operative to execute a decision logic based on measurements, such as measurements received from measurement instrumentation, which may include the current transformer 16, the voltage transformer 17, and / or phasor measurement units.

[0164] The electric power system 10 comprises a communication system 24. The communication system 24 may be or may comprise a communication network. The communication system 24 may comprise a plurality of communication links by which devices of the automation control system 20 communicate with each other and / or with central systems, such as a SCADA system or other control system and the processing system 30. Communication may be performed using communication devices such as gateway devices 25.

[0165] The processing system 30 comprises or is communicatively interfaced with at least one HMI device 38, 39 to receive the query and / or enable additional input from the user. The processing system 30 is also operative to maintain and update a knowledge base stored in storage system 33, and to access the knowledge base, e.g., when determining based on the NLP of the query that domain-specific information is to be provided. The processing system 30 may be operative to retrieve data used by the processing system 30 to update the knowledge base from, e.g., data resources 27 accessible via a wide area network 28 and / or the internet 28. Examples for such retrieved data may comprise new standards (e.g., IEEE standards and / or IEC standards), scientific publications, patent documents, and other data from trustworthy sources, which may be crawled by the processing system 30 to update the knowledge base. Additionally or alternatively, the processing system 30 may be operative to retrieve data from a proprietary data resource 26 of the operator of the electric power system 11. For illustration, the operator-proprietary data resource 26 may comprise fleet data for assets of the electric power system 10, operator-internal knowledge (such as control strategies or other knowledge) that the operator does not wish to make publicly available or share otherwise. The processing system 30, operating under the control of the operator of the electric power system 10, may access this operatorproprietary knowledge from the proprietary data resource 26 to update the knowledge base. In either case, i.e., both for retrieval of data from the data resource 26 that provides confidential knowledge and for retrieval of data from publicly available data resources 27, the processing system 30 may be operative to process retrieved data to identify key words or combinations of key words and to update the knowledge base to include the identified key words or combinations of key words in association with a knowledge bits associated with the key words or combinations of key words.

[0166] Figure 4 is a flow chart of a method 70. The method 70 may be performed automatically by the processing system 30. The method 70 is a method of performing APM.

[0167] At process block 71, the processing system 30 receives measurements during field operation of the electric power system.

[0168] At process block 72, the processing system 30 processes the measurements to continually obtain asset state data during field operation of the electric power system. The processing system 30 may store the asset state data in the storage system 33, e.g. in a state storage thereof. The asset state data may comprise data related to one, several, or all of: dissolved gas concentrations of one or several transformers; oil reclaim data for one or several transformers; fault analysis results for one or several assets; nameplate data for one or several assets; repair history data for one or several assets; load data; insulation data; factory acceptance tests; insulation fluid quality; accessory-related data, optionally comprising data relating to at least one of bushings, transformer breathers, tap changers, cooling system; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions.

[0169] At process block 73, the processing system 30 receives a query from a user (e.g., the electric power system operator or an engineer).

[0170] At process block 74, the processing system 30 performs NLP of the received query.

[0171] At process block 75, the processing system 30 determines, based on the NLP of the received query, which function selected from a set of functions supported by the processing system 30 is to be performed.

[0172] At process block 76, the processing system 30 executes the determined function to generate output. Provision of the output may comprise controlling an HMI and / or generating control signals for effecting control actions in the electric power system. When execution of the determined function requires additional parameters that are not yet included in the query, the processing system 30 may provide assistance to the user and may control the HMI to enable the inputting of the required additional parameters, such as parameters specifying a hypothetical scenario.

[0173] Figure 5 is a flow chart of a method 80. The method 80 may be performed automatically by the processing system 30. The method 80 is a method of performing APM.

[0174] Process blocks 71 to 75 may be implemented as described in association with Figure 4.

[0175] At process block 81, the processing system 30 determines whether the function to be executed is a function for which current and / or historical asset states are indispensable. At process block 82, if the function is a function for which one or several asset states are indispensable (e.g., a function relating to the current asset state or a function relating to the asset state for a hypothetical scenario), the stored asset states may be used to execute the function to generate the output. Thus, execution of the function may be based on an asset state as obtained using the expert system.

[0176] At process block 83, if the function is a function for which the assets states are not indispensable (e.g., the provision of domain-specific information or the performance of general assistance functions relating to one or several assets), and the function is executed to generate the output without necessarily using the asset state as obtained from an expert system.

[0177] Determining which one of several supported functions is to be performed may comprise a classifier based on the NLP of the query. In particular, a semantic analysis of the query may be used to determine which type of function is to be performed to comply with the query.

[0178] Figure 6 is a flow chart of a method 90. A set of functions supported by the processing system 30 is described. It will be appreciated that alternative or additional functions may be implemented.

[0179] At process block 91, the processing system 30 receives a query from a user (e.g., the electric power system operator or an engineer).

[0180] At process block 92, the processing system 30 determines which one of several functions supported by the processing system for performing APM is to be performed. The determination at process block 92 may comprise using a classifier to assign the query to one of the supported functions. Depending on the classification result, the method 90 proceeds to one of the following: process block 93: answering queries relating to an asset state of one or several of the assets: The processing system 30 may use the expert system and historical asset states to provide information on the current asset state. The generation of the output may also comprise generating, based on the historical asset states, recommendations for corrective or mitigating actions to be taken in view of the current asset state. process block 94: answering queries relating to a hypothetical scenario, such as the asset state of the one or several assets under the assumptions of the hypothetical scenario: The processing system 30 may use the expert system to determine both the current asset state and the asset state under the assumption(s) of the hypothetical scenario. The processing system 30 may compare the two asset states to generate output indicative of which changes are brought about by the hypothetical scenario as compared to the current operation scenario. process block 95: answering queries relating to domain-specific questions: The processing system 30 may retrieve information from the knowledge base and provide the domainspecific information in response to the query. As explained above, the processing system 30 may be operative to automatically update the knowledge base in an ongoing manner based on information that becomes available at publicly accessible data resources and / or at proprietary data resources of the electric power system operator. process block 96: taking action responsive to queries relating to assistant tasks (such as setting reminders, alarms, or other asset-specific tasks) for one or several of the assets: The processing system 30 may set alarms, warnings, reminders or other asset-related assistance data for one, several, or all of the assets, responsive to the NLP of the query. The processing system 30 may acknowledge that the assistance function has been implemented and may provide the alarm, warning, and / or reminder at a time or under a condition governed by the query.

[0181] At process block 97, output is generated and provided. The generation and provision of the output is dependent on the function that is to be performed (also in the sense that the logics by which the output is generated is different for the various functions), as explained above. The output may comprise control signals for an HMI and / or control signals for controlling the electric power system assets.

[0182] Several of the functions supported by the processing system 30 may comprise execution of an expert system. It will be appreciated that the skilled person has such expert systems available at his / her disposal for various types of assets.

[0183] Figure 7 shows a generic representation of an expert system 100 that may be employed by the processing system 30. The expert system 100 comprises a logic 101 operative to process various inputs 110 received by the expert system 100. The expert system 100 comprises an interface module 102 operative to interface the logic 101 with historical data 103 that may be stored in the storage system 33.

[0184] The inputs 110 processed by the expert system may comprise, without limitation, any one, any combination, or all of: temperature measurements 111 (such as transformer temperature measurements), DGA results 112 (such as DGA concentrations in an insulation fluid of a transformer), load data 113, measurement of other electric characteristics such as measurements of one or more currents, voltages, or phasor measurement unit (PMU) measurements 114, cooling data 115, data relating to other accessories 116 such as, for a transformer, data relating to bushings etc.

[0185] The expert system 100 may be operative to provide an output 109 that may be indicative of an asset state with respect to one or several aspects relevant for the respective asset type. For a transformer, the expert system 100 may provide output indicative of one, several, or all of the following: dissolved gas concentrations of one or several transformers; oil reclaim data for one or several transformers; fault analysis results for one or several assets; nameplate data for one or several assets; repair history data for one or several assets; load data; insulation data; factory acceptance tests; insulation fluid quality; accessory-related data, optionally comprising data relating to at least one of bushings, transformer breathers, tap changers, cooling system; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions. This information may be provided in a form that assists the performance of control operations, e.g., in accordance with values of an ordinal range that may be represented by colors (such as green, orange, red) to flag whether or not the respective quantity requires further attention in the future or immediate attention.

[0186] The processing system 30 and methods performed using the processing system 30 offer a way to provide the user with the information he / she needs through a personal assistant that can answer questions from the user based on the real-time data analysis from a fleet of assets. The information from the personal assistant may include a daily briefing with highlights over a past time horizon (e.g., of the last 24 hours) and upcoming events, as well as answering specific questions about the status of the fleet of assets.

[0187] Implementations of the various functions performed by the processing system 30 responsive to the NLP of the query will be described with reference to Figure 8 to Figure 18 below.

[0188] For illustration and for ease of reference, aspects of the processing system 30 and its operation will be illustrated for some exemplary use cases.

[0189] Processing the query: Processing the query does not only include determination of the function that is to be performed but may also comprise retrieval of identifiers and / or quantifiers and / or parameters from the query. For illustration, the query may specify a certain asset. The processing system 30 may have stored in the storage system 33 a configuration description (such as a substation configuration language, SCL, configuration description) or a single line diagram of at least the primary system 11 of the electric power system 10 to identify the assets specified by the query.

[0190] Aggregation of information: The processing system 30 may be operative to aggregate information relating to the assets, with the aggregation relating to a time period in the past and / or a further time period in the future. The aggregated information may be based on asset state data for a plurality, preferably all, of the assets. The aggregated information may be provided responsive to a query requesting the aggregated information (e.g., by including a reference to a "daily briefing"). Two examples are provided below:

[0191] Example:

[0192] Query: What's my daily briefing?

[0193] Output generated using processing system 30: In the last 24 hours, Transformer "5" was overloaded for 1 hour. Also, there is a heat wave coming next week, please make sure to check the cooling systems of Transformers 3 and 4 which have underperformed in the past.

[0194] Example:

[0195] Query: What's my daily briefing? Output generated using processing system 30: In the last 24 hours, Transformer "11" reached atypical DGA levels. Also, there is a forecast of a storm coming next week, which would overload the transformers, over-aging them, specifically transformers "3", "7" and "9". There is a significant risk of power outages, I recommend planning for possible power shedding and re-routing to mitigate the issue.

[0196] Provision of information related to maintenance operations: The processing system 30 may maintain, in the historical data, a history of maintenance and repair operations, comprising information on the time at which the respective maintenance or repair was performed and the type of maintenance or repair. This information may be used for providing asset state data and / or for answering queries relating to the maintenance and repair history.

[0197] Example:

[0198] Query: When was the last time we degassed or reclaimed oil in Transformer "8"?

[0199] Output generated using processing system 30: Oil was reclaimed in Transformer 8 on August 23rd of 2009.

[0200] Maintaining a database of asset-related assistance tasks: The processing system 30 may maintain a database of various assistance tasks that are to be performed in accordance with the query received.

[0201] Example:

[0202] Query: Remind me to check this transformer again next week.

[0203] Output generated using processing system 30: Reminder set: check Transformer "6" next Monday.

[0204] Provision of domain-specific info rm at ion: he processing system 30 may use a knowledge base that associates key words with knowledge bits. The processing system 30 may access the knowledge base, using key words identified in the query, to provide an answer to the query. The processing system 30 may be operative to also provide information on the source of the domain-specific knowledge, e.g., the standard from which the knowledge is obtained.

[0205] Example:

[0206] Query: What is the formula for transformer aging?

[0207] Output generated using processing system 30: According to IEEE C57.91, transformer aging is calculated using 'e' to the power 15000 divided by 383 minus 15000 divided by hotspot plus 273.

[0208] Use of historical data for asset state analysis: The processing system 30 may use historical data for the assets to determine when, how frequently, and / or under which circumstances abnormal asset states can be expected to occur.

[0209] Example:

[0210] Query: What is the DGA status for this transformer? Output generated using processing system 30: Currently, Transformer "8" is at level 1, typical levels. Based on an analysis of the readings from the last six months, this transformer will reach atypical levels within 12 months.

[0211] Advanced operations to perform asset state analysis: The processing system 30 may be operative to use the expert system to determine for which scenarios certain asset states are expected to occur.

[0212] Example:

[0213] Query: Run Duval Triangle 1 in this transformer.

[0214] Output generated using processing system 30: According to Duval Triangle 1, Transformer 7 shows Thermal faults above 700 degrees Celsius.

[0215] Determining changes for hypothetical scenarios: The processing system 30 may be operative to use the expert system to determine changes that can be expected for a hypothetical scenario, as compared to the current operating scenario.

[0216] Example:

[0217] Query: What would happen if Ethylene moves to 180 ppm within a month for transformer 3? Output generated using processing system 30: In that case, transformer 3 would move into DGA levels that would require immediate action to mitigate the concern. A problem must be happening inside to be leading to such a rapid increase in combustibles in the oil.

[0218] Provision of additional information on root causes for alarms, warning, or other flags: The processing system 30 may be operative to provide information on root causes for alarms, warnings, or other flags in an asset-specific manner. The processing system 30 may use the expert system and / or the historical data to provide this supplemental information.

[0219] Example:

[0220] Query: You are showing an alarm on Transformer 11, tell me more about it.

[0221] Output generated using processing system 30: Rapid increase of Methane was detected in this transformer. Consider the following:

[0222] This transformer is 30 years old with an LTC that has shown oil leaks to the main tank in the past. The leak was fixed in 2015 but could be happening again.

[0223] Make sure that other combustibles are also present, and it is not a sensor glitch.

[0224] The online DGA sensor on this transformer is Transfix which has shown loss of calibration in the past. The issue was fixed but this could be happening again. Confirm the DGA levels through manual sampling.

[0225] If the previous points are ruled out, Duval Triangles would indicate severe thermal issue, possibly arcing. Given the significance of gas formation, the history and age of this unit, it is highly recommended to plan removal from operation in the short term to perform internal inspection to find the origin of the issue. The processing system 30 provides an assistant specific to power systems and is particularly suitable for APM for transformers. The processing system 30 implements an inference engine, which is operative to receive the query from the user, obtain the information needed to react to the query, map the query to the information required to react to the query, and provides an output in response to the query. In order to implement such an inference engine as described above, the various types of queries supported by the processing system are segregated (Figure 6). Operations that can be executed by the processing system 30 for these various types of queries and corresponding functions are described in more detail with reference to Figure 8 to Figure 18 below.

[0226] Expert System Queries

[0227] The processing system 30 is operative to use an expert system to answer queries that relate to an asset state without necessarily involving a hypothetical scenario. As previously mentioned, the skilled person has expert systems available at his / her disposal for a wide variety of power system assets. The processing system 30 provides an interface that allows the expert system to be used while mitigating the risk of human-induced errors when using the expert system. The expert system(s) may provide functionalities related to a fleet management of the electric power system assets, such as recommendations about what to do for different DGA levels, what is the current diagnosis based on different mechanisms (Duval Triangles / Pentagons, Rogers Ratios, Electric Technology Research Association (ETRA) square, etc.), the state of a transformer's cooling system, hotspot, On-load tap changer (OLTC), etc.

[0228] Queries relating to hypothetical scenarios

[0229] Such queries are based on hypothetical scenarios the user is asking about. For these types of queries, the processing system 30 may be operative to define a state for an asset of the electric power system 10, such as a transformer. The asset state may specify for each of several aspects monitored in the respective asset a classification of a severity level. Examples for the severity levels include colorbased schemes having three or more colors (such as green, yellow, red) or other values taken from an ordinal value set. Examples of monitored conditions comprise DGA concentrations, bushings, cooling, load (e.g., overload situations), aging, moisture, etc.

[0230] In order to be able to answer a query relating to a hypothetical scenario, the processing system 30 must have available to it the parameters specifying the hypothetical scenario. If these parameters are not completely defined by the query received by the processing system 30, the processing system 30 may be operative to control the HMI to enable the user to input the missing parameters defining the hypothetical scenario.

[0231] Figure 8 is a flow chart of a method 120. The method 120 may be performed automatically by the processing system 30. T1

[0232] At process block 121, the processing system 30 classifies a received query as relating to a hypothetical scenario. This classification may be based on the presence of certain key words, such as

[0233] "what if".

[0234] At process block 122, the processing system 30 determines whether it has all information available to it required to determine how the hypothetical scenario affects the asset state. The verification at this process block may comprise determining whether the query includes information on at least one parameter (such as a DGA concentration) that is changed in the hypothetical scenario.

[0235] At process block 123, if the hypothetical scenario is not yet sufficiently defined, the processing system 30 controls the HMI to enable the provision of the information on the hypothetical scenario that is missing. This may comprise guiding the user in providing the still missing information.

[0236] At process block 124, if all information required to define the hypothetical scenario is available to the processing system 30, the processing system 30 determines the asset state for the hypothetical scenario and generate output based thereon. Process block 124 may comprise invoking the expert system and / or using a previously stored asset state data.

[0237] Figure 9 schematically illustrates the asset state data 130 for an asset that is a transformer. The asset state data 130 includes, for each one of several monitored aspects (comprising one, several or all of cooling, aging, bushing, DGA, moisture, overload) a classification result indicating whether the respective aspect is considered to be acceptable (e.g., by mapping the classification result to a class selected from a set of two, three or more than three classes).

[0238] Figure 10 is a flow chart of a method 140. The method 140 may be performed automatically by the processing system 30 to address a query related to a hypothetical scenario. The method 140 may comprise the method steps 122, 123 of Figure 8 to ensure that the hypothetical scenario is sufficiently specified.

[0239] At process block 141, the processing system 30 classifies a received query as relating to a hypothetical scenario. This classification may be based on the presence of certain key words, such as "what if".

[0240] At process block 142, the processing system 30 determines a current asset state. Determining the current asset state may comprise determining, using the expert system, for each of a plurality of aspects monitored for the respective asset an associated severity level (such as a severity level selected from an ordinal range). This determination is performed using the current, i.e. actual real-world conditions, as reflected by measurement data or other data defining the current operating scenario of the asset in the electric power system.

[0241] At process block 143, the processing system 30 determines an asset state for the hypothetical scenario. The hypothetical scenario may be distinguished from the actual real-world operating scenario in only one, two, three or more than three operating parameters (such as DGA concentrations). In this case it is then assumed that the remaining parameters of the hypothetical scenario are the same as in the actual, current real-world operating scenario. Determining the asset state for the hypothetical scenario may comprise determining, using the expert system, for each of a plurality of aspects monitored for the respective asset an associated severity level (such as a severity level selected from an ordinal range). This determination is performed using the hypothetical scenario for the asset in the electric power system.

[0242] At process block 144, the processing system 30 performs a comparison of the assets states determined at process blocks 142 and 143. The comparison may comprise identifying monitored aspects for which the classification result has changed due to the hypothetical parameter changes in the hypothetical scenario.

[0243] At process block 145, the processing system 30 determines, based on the comparison, changes which are caused by the hypothetical parameter changes in the hypothetical scenario. Determining the changes may comprise identifying monitored aspects for which the classification result has changed due to the hypothetical parameter changes in the hypothetical scenario.

[0244] At process block 146, the processing system 30 generates an inference result based on the change is determined processing block 145. The inference result may comprise, for example, information on the monitored aspects that are adversely affected by the hypothetical scenario and suggestions for mitigating such adverse effects.

[0245] At process block 147, the processing system 30 generates and provides a response to the query based on the inference result determined at process block 146.

[0246] Figure 11 and Figure 12 are exemplary representations of output provided via the HMI. The exemplary representations relate to answering a query for a hypothetical scenario. Figure 11 shows a sequence 150 of exemplary screen shots 151 to 155. Figure 12 shows a further sequence 160 of exemplary screen shots 161 to 165.

[0247] The processing may start by the user initially requesting information on a current asset state (in the present example a transformer state) (screen shot 151), while the query also is suggestive of the user's interest in a hypothetical scenario. This triggers the processing system 30 to determine the current asset state for the specified asset. The processing system 30 is operative to provide information on the parameters that can have an effect on the monitored aspects (screen shot 152). The processing system 30 is operative to request parameters to be input, thereby guiding the user in a specifying the hypothetical scenario (screen shot 152). In response to the parameters defining the hypothetical scenario (screen shot 153), the processing system 30 provides information on how the hypothetical scenario affects various monitored aspects of the asset (screen shots 154, 155, 161, 162, 163, 164). The processing system 30 is operative to provide information on root causes (screen shot 164) and / or recommendations suitable for mitigating the potentially problematic issues (screen shot 165). Figure 13 is a flow chart of a process 170. The process 170 may be performed automatically by the processing system 30 to implement process block 142 of the method 140 of Figure 10.

[0248] At process block 171, the processing system 30 obtains data required to determine the current state of the asset. This may comprise receiving measurement data 61 from measurement instrumentation, receiving event-based data 62 from intelligent electronic devices (lEDs), and / or retrieving data from the historical database (e.g., when no measurement is received, and the most recent stored measurement is used instead). Selection of the data that is obtained for processing may be performed based on the NLP of the query, e.g., by obtaining measurements that relate to an asset or asset type specified in the query, as determined by the NLP of the query.

[0249] At process block 172, the processing system 30 processes the obtained data to compute the current asset state. Computing the current asset state may comprise computing, for each one of several aspects monitored for the respective asset, a severity level associated with each of the several aspects. The severity level may be selected from an ordinal range or from a value set that maps to an ordinal range (e.g., red, green, yellow). The processing system 30 may be operative to select the expert system used from a set of available expert systems based on the asset specified in the query, as determined by the NLP of the query. The processing system 30 may be operative to input those data to the expert system that relate to the asset specified in the query.

[0250] At process block 173, the processing system 30 stores the current asset state.

[0251] Figure 14 is a flow chart of a process 180. The process 180 may be performed automatically by the processing system 30 to implement process block 143 of the method 140 of Figure 10.

[0252] At process block 181, the processing system 30 performs NLP of the query and / or possibly additional input provided by the user to determine which parameters are changed in the hypothetical scenario and how the parameters are changed.

[0253] At process block 182, the processing system 30 obtains data required to determine the asset state for the hypothetical scenario. This may comprise using hypothetical parameter values as determined based on the NLP of the query and supplementing the hypothetical parameter values with actual current operating parameters for those parameters for which the hypothetical scenario does not foresee a change.

[0254] At process block 183, the processing system 30 processes the obtained data to compute the asset state for the hypothetical scenario. Computing the asset state for the hypothetical scenario may comprise computing, for each one of several aspects monitored for the respective asset, a severity level associated with each of the several aspects. The severity level may be selected from the same ordinal range or from a value set that maps to an ordinal range (e.g., red, green, yellow) as in process block 172 of Figure 13. The processing system 30 may be operative to select the expert system used from a set of available expert systems based on the asset specified in the query, as determined by the NLP of the query. The processing system 30 may be operative to input those data to the expert system that relate to the asset specified in the query, including parameters that are specified as being hypothetical by the user.

[0255] At process block 184, the processing system 30 stores the asset state for the hypothetical scenario.

[0256] Figure 15 is a flow chart of a process 190. The process 190 may be performed automatically by the processing system 30 to implement process blocks 144-147 of the method 140 of Figure 10.

[0257] At process block 191, the processing system 30 retrieves the stored current asset state and the stored asset state for the hypothetical scenario.

[0258] At process block 192, the processing system 30 identifies changes caused by the hypothetical scenario. This may comprise comparing, for each of the monitored aspects, the severity level in the current asset state to the severity level in the asset state for the hypothetical scenario.

[0259] At process block 193, the processing system 30 may use the expert system to determine root causes and / or recommendations for mitigating those monitored aspects that are adversely affected by the hypothetical scenario.

[0260] At process block 194, the processing system 30 uses NLP to generate output based on the results of the processing operations at processing blocks 192, 193.

[0261] Thus, the processing system 30 is operative to address hypothetical scenarios, using any of the techniques explained with reference to Fig. 8 to Fig. 15.

[0262] When the processing system 30 receives a query relating to a hypothetical scenario, the processing system 30 may be operative to compute the asset state for both the current and hypothetical condition and classify each monitored aspect according to its severity level. The processing system 30 may be operative to determine a difference between the two states and use deltas to analyze the difference between the states. Thereby, the inference engine can draw conclusions relating to the hypothetical scenario. The processing system 30 is operative to create a response for the user based on the inference obtained.

[0263] Queries that trigger provision of domain-specific information

[0264] The processing system 30 is operative to provide answers to queries for domain-specific information. Examples for such domain-specific information include asset knowledge that an expert would have based on documents that describe agreed upon shared knowledge in the asset's community. For assets that are transformers, examples for domain-specific information include limits for DGA based on the pertinent standard (e.g., IEEE standards), the formula for aging, common materials for insulation, etc.

[0265] To answer these types of queries, the processing system 30 is operative to maintain and update an extensible knowledge base that grows as new knowledge becomes available. Updating the extensible knowledge base may be performed by the processing system based on new knowledge that becomes available in the public domain and / or new knowledge proprietary to the operator of the electric power system. Examples for the former include documents that contain knowledge specific to the respective asset (such as standards (e.g., IEEE standards, IEC standards), scientific articles (e.g., Cigre articles), service handbooks, electric power system magazine articles, etc.). The processing system 30 is operative to use the knowledge base to generate output for queries relating to domainspecific information. The correlation between queries and answers can be generated using a selfattention mechanism used by large language models and generative Al.

[0266] Thus, embodiments of the invention provide methods and systems for facilitating interaction with an asset performance management (APM) system, wherein the method comprises using at least one large language model (LLM) to facilitate the interaction of the user with the APM system. Using the at least one LLM may comprise using the LLM to process a query received by the user, accessing data stored by the APM system, processing data stored by the APM system, and / or generating output in response to the query.

[0267] Figure 16 is a flow chart of a method 200. The method 200 may be performed automatically by the processing system 30 to provide an answer to a query for domain-specific information.

[0268] At process block 201, the processing system 30 classifies the query as a request for domain-specific information.

[0269] At process block 202, the processing system 30 identifies key words included in the query. Identification of the key words may comprise identifying positional relations of key words, such as a proximity of key words. The processing system 30 may use at least one trained Al model (e.g., in the form of natural language processing or self— attention) to abstract the key words from the query.

[0270] At process block 203, the processing system 30 maps the identified key words to key words in the knowledge base. This may comprise finding a best match between the key words in the query and the key words used to index the knowledge base.

[0271] At process block 204, the processing system 30 retrieves the knowledge bit from the knowledge base which is associated with key words that provide the best match to the key words abstracted from the query.

[0272] At process block 205, the processing system 30 generates the response using the information retrieved from the knowledge base. To create this response, the processing system 30 may use a large language model (LLM) where a pretrained LLM is fine-tuned to include information from the knowledge base and used to generate output in response to the query.

[0273] Generic assistant functions

[0274] The processing system 30 may be operative to provide generic assistant functions in association with the electric power system 10 and its assets. The processing system 30 may be operative to perform any one or any combination of setting alarms, timers, calendaring events (maintenance of a transformer, etc.) associated with the electric power system 10 or assets thereof. These alarms, timers, events, or other items can all be associated with the specific fleet of assets. Thus, if the processing system 30 receives a query to "set up a reminder to check this asset again next week", the processing system identify that the asset (e.g., transformer) being referenced is the one currently selected.

[0275] Figure 18 is a flow chart of a method 210. The method 210 may be performed automatically by the processing system 30.

[0276] At process block 211, the processing system 30 classifies a query as relating to a generic assistant function. This may comprise identifying key words or key phrases such as "reminder", "remind", "alarm", etc. in the query.

[0277] At process block 212, the processing system 30 identifies the asset to which the query pertains. This may comprise checking, based on the NLP of the query, whether an asset is identified in the query and, if so, determining this asset to be the asset to which the query pertains. Otherwise, an asset currently selected in the processing system 30 may be identified as the asset to which the query pertains.

[0278] At process block 213, the processing system 30 identifies which type of assistance task is to be performed in association with the asset. The identification may be performed based on the NLP of the query.

[0279] Depending on the assistance task identified at process block 213, the processing system 30 creates a reminder (process block 214), creates an alarm (process block 215), or creates an event (process block 216) for the respective asset identified at process block 212.

[0280] At process block 217, the processing system 30 provides an acknowledgment that the task has been completed. At a later time, the processing system 30 may generate further output in accordance with the set reminder, alarm, or event.

[0281] Aggregation of data for several assets

[0282] The processing system 30 may be operative to determine the asset state data for several, optionally all assets (e.g., all transformers), of the electric power system 10. The processing system 30 may aggregate the information obtained for the several assets. This may comprise forming logic groups of assets based on a result of an asset state analysis performed using the expert system, e.g., based on the severity level of one or several monitored aspects. The output may be generated based on the aggregation of information for the several assets.

[0283] For illustration, the processing system 30 may determine all assets of a particular asset type having at least one monitored aspect that requires attention in the future (e.g., flag "yellow") or that requires immediate attention (e.g., flag "red"). The processing system 30 may be operative to provide information on all these assets, optionally including an indication of the aspect that requires attention for each of the aspects and / or a recommendation for mitigating the issue. In one particular example, the processing system 30 may identify, based on the continually monitored asset state data, which transformers have a load, DGA, bushing, cooling, OLTC, or moisture that requires attention and, optionally, a recommendation for how this issue could be investigated further or addressed. The processing system 30 may provide this aggregated information as part of an aggregated report (e.g., in response to a request for a daily briefing) or in response to a query for a report covering all transformer issues.

[0284] Figure 19 schematically illustrates a fleet of assets 200. The processing system may partition a set of assets 221 having the same asset type into several subsets 222, 223, 224, depending on criteria such as the most severe severity level identified for the assets. The processing system may use the asset state data for the assets to generate and provide output, based on the NLP of the query. Thereby, information may be provided to the user in a consolidated manner.

[0285] Use of trained Al model(s)

[0286] As previously mentioned, the processing system 30 may be operative to use at least one trained Al model for performing NLP of the query or for implementing an LLM useful in extending the knowledge base or providing answers to queries for domain-specific information. An Al model that may be used in this context may be or may comprise an Al model having self-attention mechanisms, e.g., multi-head self-attention mechanisms. One such type of Al model comprises an encoder-decoder structure, with both the encoder and the decoder respectively comprising a stack of self-attention mechanisms.

[0287] Figure 20 shows a schematic representation of an Al model useful in performing NLP of the query or for implementing an LLM useful in extending the knowledge base or providing answers to queries for domain-specific information.

[0288] The Al model 230 has an input operative to receive a sequence of words (for NLP of the query) or a sequence of key words (for knowledge base related functions). The Al model 230 is operative to perform a positional encoding 233 to encode the positions of the words or key words, allowing the information on the order of words or key words to be kept when performing the subsequent processing in a stack 232 of self-attention mechanisms of the encoder 231. The Al model 230 may optionally also comprise a feed forward network operative to receive the output of the stack of selfattention mechanisms and a decoder comprising a further stack of self-attention mechanisms. The Al model 230 may be trained with an objective of minimizing a reconstruction loss of the encoderdecoder structure (using, e.g., a L2 or other loss metric).

[0289] The Al model 230 may be trained using a set of training queries. The set of training queries may comprise a set of pre-defined queries and / or may use a set of pre-defined key words, which the processing system 30 shall be capable of handling. The set of training queries may comprise data representing textual input and / or spoken input. Preferably, the set of training queries comprises data representing textual input and data representing spoken input, in order to allow the processing system

[0290] 30 to be responsive to both textual input and spoken input.

[0291] The set of training queries may be divided into training data (e.g., 70% of the set), validation data (e.g., 15% of the set), and test data (e.g., 15% of the set). Training may be performed using supervised, semi-supervised or unsupervised training.

[0292] Types of assets

[0293] The processing system 30 and methods disclosed herein may be used in association with various asset types of an electric power system. The techniques disclosed herein are particularly useful for APM of transformers. Thus, according to preferred embodiments, the processing system 30 and methods are operative to perform APM for at least transformers of the electric power system.

[0294] Figure 21 shows a schematic perspective view of a transformer 240, and Figure 22 shows a schematic cross-section view through a transformer tank of the transformer 240. The transformer 240 is an asset, with the processing system 30 having an expert system for transformers and being operative to provide output in response to transformer-related queries. The processing system 240 may be operative to take into account accessories of the transformer 240 when performing the asset state analysis, such as bushings 241, a transformer breather 242, a cooling system 243, or other accessories such as an OLTC. The transformer 240 has a tank 244 which may be filled with an insulation fluid 245, such as an insulation oil. Sensors, which may comprise sensors 246, 247 installed in the transformer tank, may be operative to perform, e.g., a DGA, moisture sensing in the insulation fluid and / or insulation, temperature measurements in the transformer tank 244, without being limited thereto.

[0295] The processing system 30 may be operative to process the measurements to determine the asset state. This may be done in an ongoing basis (i.e., continually) during operation of the transformer.

[0296] HMI devices and / or trigger technigues

[0297] Queries may be provided to the processing system 30 using various HMIs or various combinations of HMIs, which may comprise optical input / output (I / O) devices or acoustic I / O devices. The processing system 30 may be operative such that the operations disclosed herein may be triggered by any one of a variety of mechanisms, such as by actuation of a soft or hard actuation element and / or by textual or spoken input that includes a trigger term or trigger phrase. This applies irrespective of whether the HMI comprises an optical I / O device, an acoustic I / O device, or a combination thereof.

[0298] Figure 23, Figure 24, Figure 25, and Figure 26 illustrate systems comprising the processing system 30 and an HMI device. While the HMI device is shown separately from the processing system 30, the processing system 30 may comprise the HMI device.

[0299] Figure 23 is a schematic representation of a system 250 comprising the processing system 30 and an HMI device 251. The HMI device 251 may be a portable device, in particular a handheld device or a wearable. The HMI device 251 is communicatively interfaced with the processing system 30 via a communication link 254, which may be implemented as a wireless or wired point-to-point communication link or which may be a communication link established over a local area network (LAN) or WAN.

[0300] The HMI device 251 comprises an HMI 253. The HMI device 251 is operative to provide, on the HMI 253, a virtual actuation element 252, such as a virtual button. The HMI device 251 and processing system 30 are operative such that the HMI device 251 allows the query to be input and transmits the query to the processing system 30 responsive to detecting actuation of the virtual actuation element 252.

[0301] Provision of, e.g., a virtual button 252 integrated into the system allows the operation disclosed herein to be triggered in an efficient manner. There is no need to constantly interpret input to monitor for a trigger phrase or trigger term. The implementation having a virtual actuation element 252 also has technical benefits with regard to privacy and does not require any extra hardware as compared to current techniques in which the user can interface with a APM system using a handheld device.

[0302] Figure 24 is a schematic representation of a system 260 comprising the processing system 30 and an HMI device 261. The HMI device 261 may be wearable. The HMI device 261 is communicatively interfaced with the processing system 30 via a communication link 264, which may be implemented as a wireless or wired point-to-point communication link or which may be a communication link established over a local area network (LAN) or WAN. It is not required that the acoustic I / O components 262, 263 of the HMI device 261 be provided separately from the processing system 30. The acoustic I / O components may also be integrated into the processing system 30.

[0303] The HMI device 261 comprises an electroacoustic transducer 262 (e.g., a loudspeaker integrated into a wearable) and an acoustoelectric transducer 263 (e.g., a microphone integrated into the wearable).

[0304] Irrespective of whether the acoustic I / O components are provided as integral part of the processing system 30 or integrated into a separate HMI device 261, the processing functions disclosed herein may be triggered by a key term or key phrase, which is also referred to as wake-up word or wake-up phrase in the art. The processing system 30 may use the acoustoelectric transducer 263 to continually listen to audio signals, transform the audio signals to text and if the wake-up word or phrase is detected, then the processing system 30 acknowledges detection of the word via an output operation. The processing system 30 is operative to then respond to the following query based on the real-time data from the fleet available to it, optionally also using historical data as previously explained. This implementation is more computationally expensive since the processing system 30 needs to be constantly interpreting language to text and looking for the wake-up word even when no wake-up word is provided. However, this solution provides enhanced convenience by allowing the operations disclosed herein to be triggered by receipt of a key word or key phrase that triggers the operations.

[0305] Figure 25 is a schematic representation of a system 270 comprising the processing system 30 and an HMI device 271. The HMI device 271 is communicatively interfaced with the processing system 30 via a communication link 274, which may be implemented as a wired point-to-point communication connection.

[0306] The HMI device 271 comprises an acoustoelectric transducer 273 (e.g., a microphone integrated into the wearable). The HMI device 271 comprises a physical actuation element 272, such as an actuation button, switch, or other physical actuation element. The HMI device 271 and processing system 30 are operative such that the processing system 30 listens for and processes the query responsive to actuation of the physical actuation element Til.

[0307] Provision of, e.g., a physical actuation element 272 allows the operation disclosed herein to be triggered in an efficient manner. There is no need to constantly interpret input to monitor for a trigger phrase or trigger term. The implementation having a physical actuation element 272 also has technical benefits with regard to privacy and robustness against background noise levels.

[0308] Figure 26 is a schematic representation of a system 280 comprising the processing system 30 and an HMI device 281. The HMI device 281 is communicatively interfaced with the processing system 30 via a wireless communication link 284, which may comprise a wireless communication link.

[0309] The HMI device 281 comprises an acoustic I / O assembly 282 (e.g., a microphone and speaker integrated into the HMI device 281). The HMI device 281 comprises a physical actuation element 283, such as an actuation button, switch, or other physical actuation element. The HMI device 281 and processing system 30 are operative such that the processing system 30 listens for and processes the query responsive to actuation of the physical actuation element 283.

[0310] Provision of, e.g., a physical actuation element 283 allows the operation disclosed herein to be triggered in an efficient manner. There is no need to constantly interpret input to monitor for a trigger phrase or trigger term. The implementation having a physical actuation element 283 also has technical benefits with regard to privacy and robustness against background noise levels.

[0311] In yet other implementations, the processing system and, if present, HMI device may be operative to receive input in textual form and / or provide output in textual form. For illustration, the HMI device 251 may have an HMI 253 comprising a touch- and / or proximity-sensitive display 253 operative to receive the query and provide the output generated by the processing system 30, under the control of the processing system 30. Thus, the system 250 may be operative as a chat system operative to allow the user to write questions rather than talking and obtain answers. It is advantageous because it is the least complex implementations and allows the processing system 30 to provide output comprising graphs and / or links to further resources. Processing system(s) for APM for performing control operations

[0312] The processing system 30 disclosed herein may be used to suggest or automatically perform control operations acting on the primary system 11 and / or secondary system of the electric power system 10.

[0313] Figure 27 and Figure 28 shows systems 280, 290 according to embodiments which respectively comprise a control system 294. The control system 294 may comprise or may be a substation automation system, a national or regional control center, a microgrid control center for a microgrid comprising distributed energy resources (DERs), a high voltage direct current (HVDC) control system, without being limited thereto. The control system 294 comprises at least one control circuit 295 operative to generate and provide control commands 296, which may comprise control commands acting on primary system components and / or secondary system components (such as commands for changing OLTC position, commands relating to transformer cooling, commands acting on switchgear 13, or combinations thereof). The at least one control circuit 295 may comprise any one or any combination of integrated circuits, integrated semiconductor circuits, processors, controllers, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), circuit(s) including quantum bits (qubits) and / or quantum gates, without being limited thereto, to generate and provide the control actions.

[0314] The control system 294 may be operative such that the at least one control circuit 295 generates and output control commands 296 based on the output generated by the processing system 30 disclosed herein. The processing system 30 may be integrated into an APM system 291 separate from the control system 294 and communicatively coupled to the control system 294 (Figure 27) or may be integrated into the control system 294 (Figure 28). The APM system 291 or control system 294 may comprise an HMI 292 to provide output under control of the processing system 30, such as recommendations for actions to be taken.

[0315] Figure 29 shows a functional representation of the processing system 30 that may be used in any of the embodiments disclosed herein.

[0316] The processing system 30 has an application layer 310 that comprises a semantics analytics tool 311 operative to process a query, a query classifier 312 operative to classify the query to determine which one of several functions is to be performed, and an access and alarm manager 313 operative to maintain alarms, reminders, and other events relating to assets. The application layer 310 may also provide I / O control functionalities, such as by controlling the HMI to enable inputting of additional information on a hypothetical scenario.

[0317] The processing system 30 has a processing layer 320 communicatively interfaced with the application layer 310. The processing layer 320 may comprise processing based on standards 321 for power system assets (such as standards for DGA concentrations). The processing layer 320 may comprise proprietary processing 322 specific to the electric power system operator and / or its asset fleet. For illustration, the proprietary processing 322 may be based on or otherwise use decision logics and control strategies deployed in lEDs and / or control systems of the electric power system in which the processing system 30 is provided. The processing layer 320 may comprise an expert system 323 that is accessed using real-time data (such as measurements) obtained from the assets.

[0318] The processing system 30 has a database layer 330. The database layer 330 may be accessed by the application layer 320 for data retrieval and / or data modification in the database layer 330. The database layer 330 may include at least a fleet database 331 storing asset-related data. The database layer may include additional databases 332, which may comprise a knowledge base. The processing system 30 may be operative to extend the knowledge base by performing LLM processing of new information that becomes available over the internet or another WAN or that becomes available in a proprietary database of the electric power system operator.

[0319] Various effects and advantages are attained by the processing system and method according to embodiments. The system and method provide enhanced techniques of performing APM. The processing system and method reduces the risk for human-induced error in asset monitoring and / or control of an electric power system. The processing system and method is operative to enable the user (e.g., electric power system operator or engineer) to easily obtain information in a question / answer basis. The processing system and method is operative to remove barriers of complexity of conventional APM system and makes monitoring more accessible regardless of expertise level.

[0320] While embodiments have been described in detail with reference to the drawings, various modifications may be implemented in other embodiments. For illustration rather than limitation:

[0321] • While embodiments have been described in which the processing system and method are operative to monitor transformers, the techniques may be applied to other electric power system assets.

[0322] • While embodiments have been described in which the processing system and method are operative to monitor power transformers of the primary system, the techniques may be applied to other transformers such as measurement transformers.

[0323] Embodiments may be used in association with a power grid having renewables penetration, such as power grid comprising renewable energy systems (such as DERs).

[0324] This description and the accompanying drawings that illustrate aspects and embodiments of the present invention should not be taken as limiting-the claims defining the protected invention. In other words, while the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative and not restrictive. Various mechanical, compositional, structural, electrical, and operational changes may be made without departing from the spirit and scope of this description and the claims. In some instances, well- known circuits, structures, and techniques have not been shown in detail in order not to obscure the invention. Thus, it will be understood that changes and modifications may be made by those of ordinary skill within the scope and spirit of the following claims. In particular, the present invention covers further embodiments with any combination of features from different embodiments described above and below.

[0325] The disclosure also covers all further features shown in the Figures individually although they may not have been described in the afore or following description. Also, single alternatives of the embodiments described in the Figures and the description and single alternatives of features thereof can be disclaimed from the subject matter of the invention or from disclosed subject matter. The disclosure comprises subject matter consisting of the features defined in the claims or the embodiments as well as subject matter comprising said features.

[0326] The term "comprising" does not exclude other elements or process blocks, and the indefinite article "a" or "an" does not exclude a plurality. A single unit or process block may fulfil the functions of several features recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Components described as coupled or connected may be electrically or mechanically directly coupled, or they may be indirectly coupled via one or more intermediate components. Any reference signs in the claims should not be construed as limiting the scope.

[0327] A machine-readable instruction code may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via a wide area network or other wired or wireless telecommunication systems. Furthermore, a machine-readable instruction code can also be a data structure product or a signal for embodying a specific method such as the method according to embodiments.

Claims

CLAIMS1. A method of performing asset performance management for assets of an electric power system, the method comprising: receiving, by a processing system, a query; processing, by the processing system, the query, comprising performing natural language processing of the query, wherein processing the query comprises performing a classification of the query to determine a function to be performed by the processing system responsive to the query, wherein the function is determined from a set of several functions supported by the processing system, wherein the processing system is operative to obtain state data for the assets in an ongoing manner during field operation of the assets to perform some or all of the several functions supported by the processing system; processing, by the processing system, data retrieved to perform the determined function responsive to the query; and generating, by the processing system, output based on a result of the processing of the retrieved data.

2. The method of claim 1, wherein the processing system is operative such that the set of several functions supported by the processing system comprises: an asset state analysis and answering at least one query relating to a hypothetical scenario.

3. The method of claim 2, wherein the retrieved data comprises the state data obtained for the assets, wherein generating the output comprises using, by the processing system, the state data obtained for the assets to perform the asset state analysis or for answering the at least one query relating to the hypothetical scenario.

4. The method of claim 2 or claim 3, further comprising controlling, by the processing system, a human machine interface, HMI, to enable user input that specifies the hypothetical scenario.

5. The method of any one of claims 2 to 4, wherein answering the at least one query relating to the hypothetical scenario comprises performing a first asset state analysis for a current asset state and performing a second asset state analysis for the hypothetical scenario, andwherein generating the output comprises generating the output based on a first result of the first asset state analysis and based on a second result of the second asset state analysis.

6. The method of claim 5, wherein answering the at least one query relating to the hypothetical scenario further comprises performing a comparison of the first result of the first asset state analysis and the second result of the second asset state analysis, wherein generating the output comprises generating the output based on a result of the comparison.

7. The method of any one of the claims 2 to 6, wherein the processing system is operative to invoke an expert system to perform the asset state analysis.

8. The method of any one of the preceding claims, wherein the processing system is operative to access historical data relating to the assets and to use the historical data for performing the determined function.

9. The method of claim 8, wherein the historical data comprises operating parameters relating to the assets at a time in the past.

10. The method of claim 8 or claim 9, wherein the historical parameters comprise one, several or all of the following historical data: insulation fluid quality; accessory-related data, optionally comprising data relating to at least one of a bushing, a transformer breather, a tap changer, a cooling system; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions.

11. The method of any one of the preceding claims when dependent on claim 2, wherein the assets comprise transformers, wherein the asset state analysis and answering the at least one query relating to the hypothetical scenario relate to a transformer state of at least one of the transformers and / or to a transformer accessory state for at least one of the transformers.

12. The method of any one of the preceding claims, wherein retrieving the data comprises obtaining the state data for at least two assets, and wherein generating the output comprises generating an output aggregated from the obtained state data for the at least two assets.

13. The method of any one of the preceding claims, wherein the state data obtained for the assets comprises state data relating to one, several, or all of the following: dissolved gas concentrations of one or several transformers; oil reclaim data for one or several transformers; fault analysis results for one or several assets; nameplate data for one or several assets; repair history data for one or several assets; load data; insulation data; factory acceptance tests; insulation fluid quality; accessory-related data, optionally comprising data relating to at least one of bushings, transformer breathers, tap changers, cooling system; gases dissolved in an insulation fluid; concentrations of dissolved gases; loading conditions.

14. The method of any one of the preceding claims, wherein the set of several functions supported by the processing system comprises at least one function executable independently of the state data obtained for the assets.

15. The method of claim 14, wherein the at least one function executable independently of the state data obtained for the assets comprises one or both of: provision of domain-specific information; an assistance function related to at least one of the assets.

16. The method of claim 15, wherein the processing system is operative to maintain and update the domain-specific information, comprising updating, by the processing system, the domainspecific information based on information retrieved via a wide area network.

17. The method of claim 1, wherein the processing system is operative such that the set of several functions supported by the processing system comprises: an asset state analysis; answering at least one query relating to a hypothetical scenario; provision of domain-specific information; an assistance function related to at least one of the assets.

18. The method of any one of the preceding claims, wherein the query comprises spoken input.

19. The method of any one of the preceding claims, wherein the query comprises written input.

20. The method of any one of the preceding claims, further comprising causing, by the processing system or a control system communicatively coupled to the processing system, execution of at least one control action for at least one of the assets based on the generated output.

21. A processing system for performing asset performance management for assets of an electric power system, the processing system comprising: at least one interface operative to receive a query; at least one processing circuit operative to process the query, wherein the at least one processing circuit is operative to perform natural language processing of the query, wherein the at least one processing circuit is operative such that processing the query comprises performing a classification of the query to determine a function to be performed by the processing system responsive to the query, wherein the at least one processing circuit is operative such that the function is determined from a set of several functions supported by the processing system, wherein the processing system is operative to obtain state data for the assets in an ongoing manner during field operation of the assets to perform some or all of the several functions supported by the processing system; process data retrieved to perform the determined function responsive to the query; and generate output based on a result of the processing of the retrieved data.

22. The processing system of claim 21, wherein the processing system is operative to perform the method of any one of claims 1 to 21.

23. An electric power system, comprising a plurality of assets, and the processing system of claim 21 or claim 22 for performing asset performance management for some or all assets of the plurality of assets.

24. The electric power system of claim 23, wherein the electric power system comprises an electric power generation, transmission, and / or distribution system.

25. The electric power system of claim 23 or claim 24, wherein the plurality of assets comprises a plurality of transformers, wherein the processing system is operative to perform the asset performance management for at least the plurality of transformers.

26. The electric power system of any one of claims 23 to 25, further comprising an input device operative to capture the query, the input device comprising a communication interface operative to provide the query to the processing system.

27. The electric power system of claim 26, wherein the input device comprises an acoustoelectric transducer operative to capture the query.

28. The electric power system of any one of claims 27, further comprising at least one control circuit operative to control the electric power system based on the output provided by the processing system.

29. Machine-readable instruction code comprising machine-readable instructions which, when executed by at least one processing circuit, cause the at least one processing circuit to perform the method of any one of claims 1 to 20.

30. Non-transitory storage medium having stored thereon machine-readable instruction code comprising machine-readable instructions which, when executed by at least one processing circuit, cause the at least one processing circuit to perform the method of any one of claims 1 to 29.

Images