WO2023202789A1 - Procédé et système de prise en charge de surveillance d'émissions - Google Patents

Procédé et système de prise en charge de surveillance d'émissions Download PDF

Info

Publication number
WO2023202789A1
WO2023202789A1 PCT/EP2022/077057 EP2022077057W WO2023202789A1 WO 2023202789 A1 WO2023202789 A1 WO 2023202789A1 EP 2022077057 W EP2022077057 W EP 2022077057W WO 2023202789 A1 WO2023202789 A1 WO 2023202789A1
Authority
WO
WIPO (PCT)
Prior art keywords
subgraph
subgraphs
incongruent
aggregated
entities
Prior art date
Application number
PCT/EP2022/077057
Other languages
English (en)
Inventor
Julia Gastinger
Timo SZTYLER
Original Assignee
NEC Laboratories Europe GmbH
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NEC Laboratories Europe GmbH filed Critical NEC Laboratories Europe GmbH
Publication of WO2023202789A1 publication Critical patent/WO2023202789A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/30—Administration of product recycling or disposal

Definitions

  • the present invention relates to a computer-implemented method for supporting emission monitoring, in particular greenhouse gas emission (GHG) monitoring.
  • emission monitoring in particular greenhouse gas emission (GHG) monitoring.
  • GFG greenhouse gas emission
  • the present invention relates to a system for supporting emission monitoring.
  • embodiments of the invention deal with the problem to detect malfunction in sensor-based greenhouse gas (GHG) emission monitoring in industrial applications.
  • GFG greenhouse gas
  • EU ETS EU emissions trading system
  • reference [2] describes that “under the Kyoto Protocol, countries are to keep precise records of the trades carried out. Transfers and acquisitions of carbon credits are tracked and recorded through the registry systems under the protocol.
  • the UN climate Change Secretariat based in Bonn, Germany, keeps an international transaction log to ensure secure transfer of carbon credits between countries and to verify that transactions are consistent with the rules of the protocol.”
  • Reference [3] describes that “the monitoring and reporting of greenhouse gas emissions must be robust, transparent, consistent and accurate for the EU emissions trading system (EU ETS) to operate effectively”. This means, it is crucial for any company, to have a robust, transparent, consistent and accurate emission monitoring.
  • a computer-implemented method for supporting emission monitoring comprising the steps of: a) computing emissions for one or more industrial applications based on sensor measurements; b) transforming emission information into a knowledge graph, wherein the knowledge graph provides a set of triples representing entities, relations between them, and corresponding attributes of the entities; c) computing an embedding for each entity in the knowledge graph based on structural properties and attributes of the entities; d) extracting, based on a predefined selection factor, one or more subgraphs from the knowledge graph, wherein an aggregated embedding and aggregated emission information are computed as an aggregated subgraph-representation for each subgraph; e) assigning to each extracted subgraph an incongruity score that indicates a degree to which
  • a system for supporting emission monitoring comprising: an emission computation module (A) that is configured to compute emissions for one or more industrial applications based on sensor measurements; a graph creation module (B) that is configured to transform emission information into a knowledge graph, wherein the knowledge graph provides a set of triples representing entities, relations between them, and corresponding attributes of the entities; a graph embedding module (C) that is configured to compute an embedding for each entity in the knowledge graph based on structural properties and attributes of the entities; a subgraph extraction module (D) that is configured to extract, based on a selection factor, one or more subgraphs from the knowledge graph, wherein an aggregated embedding and aggregated emission information are computed as an aggregated subgraph-representation for each subgraph; an incongruity score computation module (E) that is configured to assign to each extracted subgraph an incongruity score that indicates a degree to which its aggregated emission information is faulty such that the extracted subgraphs comprise congruent and incongru
  • emissions are computed for one or more industrial applications based on sensor measurements.
  • emission information which may comprise the computed emissions, is transformed into a knowledge graph.
  • the knowledge graph provides a set of triples representing entities, relations between them, and corresponding attributes of the entities.
  • the knowledge graph may also comprise additional information of the entities.
  • sensor measurements such as air pressure, temperature, etc.
  • information about e.g., the sensors or devices
  • health status e.g., meta information, etc.
  • an embedding in particular an embedding vector
  • one or more subgraphs are extracted from the knowledge graph, wherein an aggregated embedding and aggregated (reported) emission information are computed as an aggregated subgraph-representation for each subgraph.
  • An incongruity score is assigned to each extracted subgraph, wherein the incongruity score indicates a degree to which the subgraph’s aggregated emission information is faulty such that the extracted subgraphs comprise congruent and incongruent subgraphs.
  • the incongruity score is computed based on the aggregated subgraphrepresentations of the extracted subgraphs. Then, for each incongruent subgraph of the extracted subgraphs, a decision is made whether to proceed with either investigating the incongruent subgraph on a next hierarchy level in order to detect malfunction of one or more subgraph elements of the incongruent subgraph or to take an action for the incongruent subgraph.
  • the monitoring may be performed with the goal of identifying malfunctioning (groups of) entities in sensor-based emission reporting.
  • the method and the system according to the invention enable that malfunction in sensor-based emission monitoring is detectable in a more efficient way.
  • industrial applications may refer in particular in the claims, preferably in the description, to industrial facilities and/or plants, including but not limited to applications for manufacturing, food processing or storage, chemical, petrochemical, and/or power plants.
  • emission information may refer in particular in the claims, preferably in the description, to emission information values.
  • the emission information may comprise information about emissions that are computed for one or more industrial applications, based on sensor measurements.
  • the emission information may further comprise information on reports and/or company data from available databases such as a carbon report database.
  • the emission information may comprise information about related companies, information about processes in the companies that are related to the emissions, and/or information about entities outside of the companies (such as suppliers) that influence the emission.
  • entity may refer in particular in the claims, preferably in the description, to an element or entry in the graph, which may be also designated as “node”.
  • a knowledge graph may have many of such entities. These nodes or entities can represent different things, for example, sensors, but also parts of a manufacturing plant, people, locations and/or other information.
  • Relations between entities may refer in particular in the claims, preferably in the description, to relations between the entities in a knowledge graph.
  • Relations may describe different edges.
  • the relations may include but are not limited to, e.g., “buys”, “consists of’, “contains”, and/or any other type of relation between two entities/entries in a knowledge graph.
  • Entities and relations may form triples, e.g., [convey or belt A, transports, Product C], with “transports” being the relation in this case.
  • attribute of an entity may refer in particular in the claims, preferably in the description, to attributes and/or properties of entities in a knowledge graph: It may be provided that for each entity in a knowledge graph, further information can be stored such as its attributes/properties. For instance, attributes may include but are not limited to the size, the sector, and/or the number of employees of an industrial application.
  • structural properties may refer in particular in the claims, preferably in the description, to properties of graphs that depend only on the abstract structure, not on graph representations such as particular labels or drawings of the graph. Examples for such properties may include but are not limited to the diameter, the connectivity, the size, and/or the degree sequence etc.
  • embedding may refer in particular in the claims, preferably in the description, to a vector representation of entities and relations in the knowledge graph.
  • the embedding (which may be also designated as embedding vector) may be a dense vector representation of the entity, which can be used for many tasks such as classification.
  • a neural network (NN) and/or other machine learning techniques can be used to create the embedding.
  • aggregated embedding may refer in particular in the claims, preferably in the description, to an aggregation of multiple embedding vectors. Different techniques exist to compute an aggregated embedding, e.g., a mean average or sum of the individual embedding vectors.
  • aggregated emission information may refer in particular in the claims, preferably in the description, to the aggregation of a certain or predetermined amount of individual emission information values. For instance, the aggregation may happen by computing a sum across the individual values, by computing the average, and/or by extracting the maximum value.
  • incongruent and “incongruent” - e.g., in the context of a subgraph or an entity - may refer in particular in the claims, preferably in the description, to the following: “Incongruent” may refer to not accordant, lack of consistency, incongruous. “Congruent” is the contrary. In the context of embodiments of the invention, an incongruency score is used to compute how well sensor values match to the expectance.
  • the decision of step f) may be based on a predefined condition.
  • the predefined condition can be defined depending on the use case.
  • the granularity level for the extracted subgraphs can be adjusted and controlled in a flexible way.
  • steps d) to f) are iterated for investigating incongruent subgraphs on the next hierarchy level until the predefined condition is met.
  • the predefined condition can be employed for adjusting a level of granularity that is to be considered when focusing on incongruent subgraphs.
  • An embodiment may start from the top, and then iterates downwards, meaning it subsequently focuses on a more fine-grained level in each iteration through the steps d) to f).
  • the predefined condition is defined in such a way that steps d) to f) are iterated until a single incongruent entity has been detected.
  • steps d) to f) are iterated until a single incongruent entity has been detected.
  • investigations are stopped for congruent subgraphs of the extracted subgraphs.
  • investigations are only focused on incongruent subgraphs, namely as long as incongruence is identified in the knowledge graph by the incongruity score computation module.
  • the predefined selection factor is adjustable, in particular by a user in order to set different selection factors for controlling a scope of investigations for malfunction detection.
  • step d) the predefined selection factor is set to density-based, sector-based, component-based and/or type-based.
  • the extraction of subgraphs can be controlled with respect to subgraph types.
  • step d) the aggregated embedding is computed for each subgraph based on max-pooling across all embeddings of the subgraph.
  • max-pooling the aggregated embeddings may be computed in a very easy way of aggregation.
  • step d) the aggregated emission information is computed for each subgraph by computing the maximum emission information of the subgraph.
  • the emission information can be computed in an advantageous way, as it will show elements with exceptionally high emissions.
  • step e) the computing of the incongruity score for each extracted subgraph is based on an unsupervised machine learning approach that compute clusters of subgraphs, wherein a subgraph j is detected as outlier, if the subgraph j cannot be assigned to a cluster, or if the distance to the next cluster center is higher than a predefined threshold.
  • K-means clustering is used to compute the clusters of subgraphs.
  • clusters of subgraphs can be computed.
  • K-means clustering may have one or more of the following advantages:
  • the action for the incongruent subgraph is performed dependent on the decision of step f), wherein the action can be selected from a predefined set of actions.
  • the predefined set of actions may comprise an action to automatically turn off entities (and/or components thereof), which are marked as having anomalous amounts of emissions.
  • an action module may be provided that is configured to initiate and/or perform the action for the incongruent subgraph.
  • the action module can initiate and/or perform the action for the incongruent subgraph dependent on the decision of step f), wherein the action can be selected from a predefined set of actions.
  • the action for the incongruent subgraphs may include that a list of entities of the incongruent subgraphs, which are detected to be malfunctioning and/or have anomalous amounts of emissions, is provided.
  • an embodiment of the invention may return a list of entities/components, which are detected to be malfunctioning and/or have anomalous amounts of emissions. This output can be used to automatically turn off one or more (respective) entities/components. Furthermore, a responsible person may be informed.
  • the action for the incongruent subgraph may include that an alarm signal is triggered in order to indicate entities of the incongruent subgraphs, which are detected to be malfunctioning and/or have anomalous amounts of emissions.
  • an alarm light may be turned on for malfunctioning entities/sensors, highlighting which of them need to turn off. This can be used to update sensors or other malfunctioning elements in the sensor network.
  • the action for the incongruent subgraph may include that an entity of the incongruent subgraph is turned off automatically.
  • an automated deactivation of malfunctioning industrial components/entities is supported based on malfunction detection in sensor networks for emission reporting.
  • sensor data may be provided by sensors installed in an industrial production line, wherein the sensor data may comprise sensor information about the emissions in a company.
  • the embodiment helps by finding the malfunctioning components in a production, by using a network of sensors.
  • the reported emissions of a company may be based on a sum of the emissions reported by each single sensor.
  • the embodiment will investigate which (sensor) elements have been reporting the problem in a hierarchical way, focusing on the full company, then on certain parts of sensor groups, and on special sensors - belonging to certain parts in the production line - subsequently.
  • the components that are marked as having anomalous amounts of emissions i.e. marked as malfunctioning will be turned off automatically, to avoid running malfunctioning machines. Further, a responsible person in the production line may immediately be informed in order to replace or repair the components.
  • Embodiments of the present invention may relate to an automated sensor-based (green house gas) emission monitoring method and system with an integrated malfunction detection and alarm system for industrial applications.
  • the embodiments may be based on knowledge graph data representation, and uses a hierarchical approach in combination with an incongruity score computation module, to identify malfunctioning (groups of) entities in sensor-based emission reporting.
  • At least one embodiment of the invention may provide one or more of the following aspects: i. Hierarchical, iterative approach to detect incongruent (groups of) entities in a knowledge graph (KG) in a top-down way, starting with analysing the full knowledge graph and then refining the search space and focusing on incongruent subgraphs only.
  • the computation of the incongruity score may be based on an unsupervised learning approach, taking into account the graph embedding and the reported emissions:
  • the subgraph extraction process may comprise:
  • references [11] and [12] are able to discover anomalies, taking into account the structural and relational information.
  • the system of reference [11] creates a knowledge graph from heterogeneous data, finds hidden relations via link prediction and, for incoming streaming data, extracts information and compares this information to the established triples. Anomalies are for example subjects or objects associated with unusual patterns.
  • the system of reference [12] detects anomalies in industrial settings by analyzing log entries of sensor alarm messages in an industrial production scenario. For this, they use sequential alarm data, and detect anomalies based on Markov chain models.
  • the work described in reference [13] also presents an algorithm for anomaly detection in graphs, using a combination of graph features and unsupervised learning methods, focusing on special types of anomalies, namely global, neighbor-based, and community-based, using different approaches to detect each of them.
  • the goal is to detect anomalous nodes, e.g., in wireless sensor networks.
  • a node is considered anomalous if its attributes are different from neighbors attributes.
  • an outlier is detected by looking at it in its community.
  • an autoencoder is used to build a model for unsupervised data, and anomaly is defined as a point with the highest reconstruction error. It does extract communities, which are slightly similar to extracting subgraphs. However, the extraction of the communities is different to the extraction of the subgraph (not taking into account graph parameters as density, neither any use case specific factors as industrial sectors), and in addition, the usage of the communities is different to the usage of the subgraphs: Reference [13] uses the communities to compare single entities inside their community and thus detect single anomalous entities. However, embodiments of the present invention use the subgraphs, to compare the subgraphs to each other, and detect incongruent subgraphs.
  • references [11], [12] and [13] neither focus on emission reports, nor do they have a hierarchical approach.
  • references [11], [12] and [13] cannot be applied to the problem given in embodiments of the present invention.
  • references [11], [12] and [13] cannot detect both, incongruent groups of entities, as well as single incongruent entities, which leads to less flexibility.
  • having to check for anomaly for each single entity, instead of subsequently focusing on incongruent subgraphs only leads to lower efficiency and thus to problems with the data.
  • embodiments of the present invention can deal with a large dataset, as well as complex data, i.e. it can grasp the relational properties between the different components, and the large amount of sensor measurements.
  • embodiments of the present invention can take into account relational and structural properties between the different operators. Further, as an embodiment of the invention may use an unsupervised technique for computing the incongruity score, it is not dependent on the availability of appropriate (labeled) historical training data for malfunction detection.
  • a method and/or a system is able, to not only compute the malfunctioning (groups) of components based on this actual companies measurements, but also based on other companies measurements, because it can compare similar subgraphs.
  • the method or system takes information not only from the one company, but it also collects information from other companies, with similar properties (e.g. similar entity embeddings, similar emissions). For this reason, it can exploit the information and knowledge that already exists, which leads to better accuracy and insights as compared to systems that cannot exploit structural information based on different criteria.
  • the system according to embodiments of the invention may use a hierarchical approach, so it can first check, whether the overall measured emissions are congruent, and then, if the overall measure emissions are not congruent, detect groups, or also single components, that have irregular measured emissions. This leads to an efficient malfunction detection, and thus ensures more timely identification of, and reaction to problems with reported emissions as compared to a system that does not use such an hierarchical approach.
  • Embodiments may depend on the availability of sensor and company data. Embodiments may be based on the idea that many customers/users will buy the proposed systems, and thus, by using the (anonymized) customer databases in addition to public databases having access to a significant amount of data. This means, the more users use the system, the more data will be available.
  • the data may have to have a relational aspect: It needs to have relations between entities; otherwise a knowledge graph is not a suitable data structure.
  • the stopping condition for the hierarchy level selection should be chosen appropriately, e.g., by the user, i.e. up to which level the system should iterate before taken an action.
  • Embodiments of the present invention may be used in industrial applications, where GHG emissions need to be reported, and where industrial components can be equipped with sensors.
  • a differentiator to classical malfunction detection systems is that a method and/or a system according to embodiments of the invention has the capability to take relational and structural properties in combination with domainspecific attributes into account.
  • the method and the system can capture and understand complex issues, e.g., the interaction of different parts of the supply chain, or the similarity of different components across different companies.
  • embodiments of the invention are able to compare the sensor measurements across companies or across different production lines, if the components have similar structural properties.
  • embodiments can compare the packaging line of a toothpaste production line with the packaging line of food production, if similar components and structure are present. Further, due to the unsupervised computation of the incongruity score, a method and/or a system in accordance with an embodiment of the invention does not need a large labeled training dataset. In addition, due to the hierarchical top- down approach, the method/system is able to deal with large amounts of data. In addition, by computing the aggregated representations for the subgraphs, the method/system is able to deal with groups of operators of arbitrary size.
  • Fig. 1 is a schematic view illustrating an overview of a system architecture in accordance with an embodiment of the invention
  • Fig. 2 is a schematic view illustrating an exemplary knowledge graph of a graph creation module according to an embodiment of the invention
  • Fig. 3 is a schematic view illustrating the hierarchical concept of a method or a system in accordance with an embodiment of the invention.
  • Fig. 4 is a schematic view illustrating an example of extracting subgraphs from a knowledge graph in accordance with an embodiment of the invention.
  • Fig. 1 shows a schematic view illustrating an overview of a system architecture in accordance with an embodiment of the present invention.
  • Embodiments of the present invention may comprise several components, in particular a Sensor-based Emission Computation Module, a Graph Creation Module, a Graph Embedding Module, a Subgraph Extraction module, an Incongruity Score Computation Module, a Hierarchy Level Selection Module, and an Action Module.
  • Fig. 1 shows an overview of a full system in accordance with an embodiment of the invention.
  • Bold Boxes of Fig. 1 mark the key components ((C), (D), (E), (F)) of the embodiment.
  • Fig. 2 for a bigger view of the knowledge graph in component (B), i.e. the Graph Creation Module.
  • the dashed lines in component (D) and in component (F) indicate subgraphs, extracted from the full knowledge graph, or from a bigger subgraph.
  • the dashed lines between the Action Module (G) and a box “Industrial Application(s)” indicate, that either, actions can be recommended to a User (e.g., by showing an alarm), or actions can directly be applied to the industrial components.
  • Embodiments of the invention have a technical aim and serve a specific technical purpose, i.e., embodiments of the invention aim to influence non-human systems, more specific (components in) industrial applications.
  • the input to the automated sensorbased emission monitoring system is information from one or multiple Database(s), which contains information on company data and on reported carbon and other emissions by each company.
  • databases may the Orbis database (cf. reference [7]), which is a huge public database for company data, and governmental databases that contain the reported carbon emissions.
  • Databases contain information on the real world, e.g., companies, suppliers, attributes of companies and the interactions, as well as the reported emissions.
  • the input to the Databases are either company reports or sensor measurements (e.g., conducted by an embodiment of the invention).
  • the Sensor-based Emission Computation Module computes the emissions for one or multiple industrial application(s) based on the sensor measurements. These computed emissions can be used for automated GHG emission reporting, e.g. for Ell ETS.
  • the Emission Computation Module may receive automatically all sensor measurements, where each sensor belongs to one component in the industrial application, and computes the overall emission based on a sum, e.g. as described in reference [1], The output are the emissions for each industrial application, and for each sensor unit corresponding to each component.
  • the Graph Creation Module (B) reads information on measurements, reports, and company data from the Database(s) and the Emission Computation Module (A). Everything that is read may be transformed into a knowledge graph (KG), i.e. , a set of triples. This set of triples represents entities, relations between them, and corresponding attributes.
  • KG knowledge graph
  • An example knowledge graph that is created by the Graph Creation Module is shown in Fig. 2.
  • Fig. 2 is a schematic view illustrating an exemplary knowledge graph of a Graph Creation Module according to an embodiment of the invention. More specifically, Fig. 2 illustrates an exemplary knowledge graph representing the different industrial companies and components (entities), their attributes (white boxes), the sensor units Sil (black boxes, also entities), and the interactions between all entities (relations).
  • Sensors for component 1 and 2 measure very different emissions, although they have same graph embedding and same attributes and belong to the components of the same type. This hints to one sensor or component being malfunctioning.
  • Fig. 2 is a zoomed view of the knowledge graph of the Graph Creation Module (B) as depicted in Fig. 1. Multiple relations of the same type between two nodes are represented as weighted relations.
  • the Graph Embedding Module (C) as depicted in Fig. 1 takes as input the knowledge graph and computes an embedding for each element in the knowledge graph, based on the structural properties and based on the attributes of each entity.
  • the output of the Graph Embedding Module is a vector-embedding E for each entity in the knowledge graph.
  • Fig. 1 - The following three modules (D)-(F) as depicted in Fig. 1 - namely Subgraph Extraction Module (D), Incongruity Score Computation Module (E) and Hierarchy Level Selection Module (F) - may be run iteratively.
  • D Subgraph Extraction Module
  • E Incongruity Score Computation Module
  • F Hierarchy Level Selection Module
  • a hierarchical concept provides a fundamental characteristic for an embodiment of the invention.
  • FIG. 3 is a schematic view illustrating the hierarchical concept of a method or a system in accordance with an embodiment of the invention. The system starts from the top, and then iterates downwards, meaning it subsequently focuses on a more fine-grained level in each iteration.
  • the system first starts to analyze the graph on a high level to detect if there is any incongruency. If yes, the system defines regions of the knowledge graph (subgraphs) based on certain selection factors (step (D)), and computes the incongruency score for those sub-regions, i.e. for the subgraphs (step (E)). If no incongruency is detected for a subgraph, the system will stop investigations for this subgraph. It will thus, in further iterations, be able to focus on the subgraphs with high incongruency score only, and will thus be able to exclude less suspicious subgraphs from the investigations.
  • the top-down approach By the top-down approach, it will be able to identify, step-by-step the malfunctioning elements in a network of elements/entities. In addition, it will be able to identify groups or networks of malfunctioning elements (e.g., the sensors or components belong to a certain area or building in the production line).
  • the knowledge graph and more specific the knowledge graph structure, helps to identify on which areas (i.e. , subgraphs) to focus on, on a more fine-grained level.
  • Fig. 3 is a schematic view illustrating the hierarchical concept of a method or a system in accordance with an embodiment of the invention. More specifically, Fig. 3 is an illustrations of hierarchical levels.
  • the system iterates between steps (D) - (F), focusing on more fine-grained parts of the knowledge graph subsequently.
  • the systems stops investigations (illustrated by a stop sign in Fig. 3) for subgraphs with low incongruency score (illustrated by a check mark in Fig. 3). It continues iterating, or takes an action, for the subgraphs with a high incongruency score (illustrated by an alarm light in Fig. 3).
  • the Subgraph Extraction Module (D) takes as input the knowledge graph as well as the embedding of each entity and the information on computed and reported emissions for each stakeholder (i.e., for each entity representing a company or component). It extracts one or multiple subgraphs from the knowledge graph, and computes the aggregated reported emissions as well as the aggregated embedding for each subgraph. More specifically, the Subgraph Extraction Module may comprise the following steps:
  • the selection of the subgraph-type may be based on a (e.g., hard-coded) selection factor, which can, for example, be selected based on user interest, or based on industrial regulations.
  • exemplary factors could be density-based, sector-based, component-based and/or type-based.
  • the selection of factors can be extended, if needed. In each iteration, a different selection factor can be selected.
  • there is a knowledge graph showing different companies and their components as well as their interactions, e.g., like the knowledge graph illustrated in Fig. 2.
  • the selection factor is set to density-based, because the user is interested in groups of components that interact with each other more strongly as compared to other entities of the graph. This could for example illustrate groups of sensors, which measure the emissions for a certain production line in a certain building and communicate the measurements.
  • the subgraph extraction module will automatically extract all subgraphs that fulfill a condition. For the density-based example, it will extract subgraphs, if their density is higher than a predefined threshold, i.e. , it will extract the entities that are more strongly connected than others.
  • a predefined threshold i.e. , it will extract the entities that are more strongly connected than others.
  • Fig. 4 shows an example graph, where, with the method presented in reference [8], for a density threshold of 0.75 two subgraphs (marked by dashed lines in Fig.
  • the density threshold is not predefined, but is learned based on similar cases in the database. For the type-based example, it will for example extract all instantiations of a certain motif (one subgraph represents one motif). In another embodiment (sector-based), it will extract all entities that belong to a certain industrial sector as well as their 1 - hop neighborhood.
  • the subgraphs can be of arbitrary size, including the extraction of single entities as subgraphs.
  • the subgraph extraction module computes the aggregated embedding E Agg for each subgraph. In one embodiment, this is done based on max-pooling across all embeddings in this subgraph. It can be conducted with any aggregation function.
  • the module computes the aggregated reported emissions R Agg .
  • the aggregation is represented by the maximum reported emissions.
  • the Incongruity Score Computation Module (E) takes as input the aggregated embeddings E Agg as well as aggregated measured emissions R Agg for each extracted subgraph.
  • An example of how this module is able to find incongruity in reported emission information may be as follows:
  • Subgraph A has a very similar aggregated embedding to Subgraph B, and Subgraph C, and subgraph D (meaning for example similar suppliers or similar trades of emission credits), but significantly lower measured emissions. This could mean that one part of the production line of subgraph A, with similar sensors and similar components as subgraph B, has significantly lower measured emissions, which could hint to one (or multiple) sensors in this part of the production line failing to measure the emissions.
  • the computation is based on an unsupervised Machine Learning approach, which takes as an input the subgraph-representations S for all subgraphs and computes clusters. If a subgraph j cannot be assigned to a cluster, or the distance to the next cluster center is higher than a certain predefined threshold, the subgraph j is detected as outlier.
  • k-means-clustering may be used to compute the clusters.
  • the Hierarchy Level Selection Module takes as input the subgraphrepresentations S, and the computed incongruity Scores I. The module decides on the next step, with the following two options:
  • This can either be one component (e.g., one sensor or one component in a production line), or a group of components, depending on the subgraph-granularity level that has been focused on.
  • the Hierarchy Level Selection Module (F) decides on one of the aforementioned options based on a predefined condition.
  • the condition can be defined depending on the use case, preferably with close collaboration of the responsible people.
  • the Hierarchy Level Selection Module (F) may decide to keep on iterating until the lowest possible action, i.e. until single incongruent entities have been detected.
  • the Action Module (G) is configured to select and conduct the action based on a user-defined set of actions.
  • the module may act upon the entity, or set of entities with high incongruity score.
  • the actions are dependent on the corresponding use case.
  • a company uses a method and/or a system for supporting emission monitoring in accordance with an embodiment of the invention and thus using a full emission monitoring system, including the sensors and the automated emission report generation.
  • the system may compute the emission report for this company based on the sensor measurements, and in addition, it will check for this company whether the emission report is reasonable, as compared to other (similar) companies’ emission reports.
  • the system will detect the malfunctioning (group) of sensors, and show an alarm for sensors which are expected to be malfunctioning. For this reason, embodiments of the invention help that sensors do not need to be placed redundantly.
  • Data Source Network of sensors in a company. A database with emission reports for different companies, including information on their sensors.
  • a method and/or a system in accordance with an embodiment of the invention helps by finding the malfunctioning sensors in a network of sensors.
  • the reported emissions of a company are based on a sum of the emissions reported by each single sensor. They will be compared to reported emissions of other companies in a knowledge graph. If the reported emissions are marked as incongruent, the invention will investigate which (sensor) elements have been causing the problem in a hierarchical way, focusing on the full company, then on certain parts of sensor groups, and on special sensors subsequently.
  • Output A list of sensors in a company, which seem to be malfunctioning.
  • An alarm light may turn on for the malfunctioning sensors, highlighting which of them need to turn off. For instance, this can be used to update sensors or other malfunctioning elements in the sensor network.
  • a company with industrial production uses a method and/or a system for supporting emission monitoring in accordance with an embodiment of the invention and thus using a emission monitoring system, including the sensors and the automated emission report generation.
  • the system will compute the report for this company based on the sensor measurements, and in addition, check for this company if the emission report is reasonable, as compared to other (similar) company reports.
  • the system can investigate which (groups of) sensors have been causing these non-reasonable emission numbers, and which components in the production they belong to. It may then automatically turn off the components, e.g. parts of the production line, to avoid running malfunctioning machines. This saves the need for manual controls of emissions of every component in the production line.
  • Data Source Network of sensors in a company. A database with emission reports for different companies, including information on their sensors.
  • a method and/or a system in accordance with an embodiment of the invention helps by finding the malfunctioning components in a production, by using a network of sensors.
  • the reported emissions of a company are based on a sum of the emissions reported by each single sensor. If the reported emissions are classified as wrong/faulty, the system may investigate which (sensor) elements have been reporting the problem in a hierarchical way, focusing on the full company, then on certain parts of sensor groups, and on special sensors - belonging to certain parts in the production line - subsequently.
  • Output A list of components in a company, which seem to be malfunctioning or have anomalous amounts of emissions.

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Sustainable Development (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

L'invention concerne un procédé mis en œuvre par ordinateur pour prendre en charge une surveillance d'émission, en particulier une surveillance d'émission de gaz à effet de serre, le procédé comprenant les étapes consistant à : a) calculer des émissions pour une ou plusieurs applications industrielles sur la base de mesures de capteur; b) transformer des informations d'émission en un graphe de connaissances, le graphe de connaissances fournissant un ensemble de triplets représentant des entités, des relations entre elles, et des attributs correspondants des entités; c) calculer une incorporation pour chaque entité dans le graphe de connaissances sur la base de propriétés structurelles et des attributs des entités; d) extraire, sur la base d'un facteur de sélection prédéfini, un ou plusieurs sous-graphes à partir du graphe de connaissances, une incorporation agrégée et des informations d'émission agrégées étant calculées sous la forme d'une représentation de sous-graphe agrégée pour chaque sous-graphe; f) attribuer à chaque sous-graphe extrait un score d'incongruité qui indique un degré auquel ses informations d'émission agrégées sont défectueuses de telle sorte que les sous-graphes extraits comprennent des sous-graphes congruents et incongruents, le score d'incongruité étant calculé sur la base des représentations de sous-graphe agrégées des sous-graphes extraits; et g) pour chaque sous-graphe non congruent des sous-graphes extraits, prendre une décision quant au fait de savoir s'il faut procéder à l'étude du sous-graphe incongruent sur un niveau hiérarchique suivant ou entreprendre une action pour le sous-graphe incongruent. L'invention concerne en outre un système correspondant.
PCT/EP2022/077057 2022-04-20 2022-09-28 Procédé et système de prise en charge de surveillance d'émissions WO2023202789A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
EP22169104.1 2022-04-20
EP22169104 2022-04-20

Publications (1)

Publication Number Publication Date
WO2023202789A1 true WO2023202789A1 (fr) 2023-10-26

Family

ID=81585175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2022/077057 WO2023202789A1 (fr) 2022-04-20 2022-09-28 Procédé et système de prise en charge de surveillance d'émissions

Country Status (1)

Country Link
WO (1) WO2023202789A1 (fr)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2858015A1 (fr) * 2013-10-04 2015-04-08 Building Research Establishment Ltd Système et procédé de simulation, commande et surveillance des performances de systèmes d'énergie
US20190095806A1 (en) * 2017-09-28 2019-03-28 Siemens Aktiengesellschaft SGCNN: Structural Graph Convolutional Neural Network
US20200082289A1 (en) * 2018-09-11 2020-03-12 Microsoft Technology Licensing, Llc Task scheduling recommendations for reduced carbon footprint
US20200342055A1 (en) * 2019-04-23 2020-10-29 Oracle International Corporation Named entity disambiguation using entity distance in a knowledge graph
US20200372588A1 (en) * 2019-05-20 2020-11-26 Singularity Energy, Inc. Methods and systems for machine-learning for prediction of grid carbon emissions
US20210067527A1 (en) * 2019-08-28 2021-03-04 Nec Laboratories America, Inc. Structural graph neural networks for suspicious event detection
CN113887552A (zh) * 2021-08-25 2022-01-04 苏州数言信息技术有限公司 一种基于传感器的碳计量方法及装置
US20220027707A1 (en) * 2020-07-24 2022-01-27 International Business Machines Corporation Subgraph guided knowledge graph question generation

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2858015A1 (fr) * 2013-10-04 2015-04-08 Building Research Establishment Ltd Système et procédé de simulation, commande et surveillance des performances de systèmes d'énergie
US20190095806A1 (en) * 2017-09-28 2019-03-28 Siemens Aktiengesellschaft SGCNN: Structural Graph Convolutional Neural Network
US20200082289A1 (en) * 2018-09-11 2020-03-12 Microsoft Technology Licensing, Llc Task scheduling recommendations for reduced carbon footprint
US20200342055A1 (en) * 2019-04-23 2020-10-29 Oracle International Corporation Named entity disambiguation using entity distance in a knowledge graph
US20200372588A1 (en) * 2019-05-20 2020-11-26 Singularity Energy, Inc. Methods and systems for machine-learning for prediction of grid carbon emissions
US20210067527A1 (en) * 2019-08-28 2021-03-04 Nec Laboratories America, Inc. Structural graph neural networks for suspicious event detection
US20220027707A1 (en) * 2020-07-24 2022-01-27 International Business Machines Corporation Subgraph guided knowledge graph question generation
CN113887552A (zh) * 2021-08-25 2022-01-04 苏州数言信息技术有限公司 一种基于传感器的碳计量方法及装置

Non-Patent Citations (19)

* Cited by examiner, † Cited by third party
Title
"International Criminal Police Organisation (INTERPOL", GUIDE TO CARBON TRADING CRIME, 2013
ACTION, EU ETS MONITORING AND REPORTING - QUICK GUIDE FOR STATIONARY INSTALLATIONS, November 2017 (2017-11-01), Retrieved from the Internet <URL:https://ec.europa.eu/clima/system/files/2017-11/quick_guide_operators_en.pdf>
ATZMUELLERMARTINDAVID ARNUANDREAS SCHMIDT: "Data Science-Analytics and Applicati", 2017, SPRINGER VIEWEG, article "Anomaly detection and structural analysis in industrial production environments", pages: 91 - 95
AZIZZENA A. AZIZSIDDEEQ YAMEEN AMEEN: "Air pollution monitoring using wireless sensor Networks", JOURNAL OF INFORMATION TECHNOLOGY AND INFORMATICS, vol. 1, no. 1, 2021, pages 20 - 25, Retrieved from the Internet <URL:https://www.qabasjournals.com/index.php/jiti/article/download/26/37>
BIN JIACAILING DONGZHIJIANG CHENKUO-CHU CHANGNICHOLE SULLIVANGENSHE CHEN: "Pattern Discovery and Anomaly Detection via Knowledge Graph", FUSION, 2018, pages 2392 - 2399, XP033398033, DOI: 10.23919/ICIF.2018.8455737
BUREAU VAN DIJK: "Bureau van Dijk Editions Electroniques S.A", 2019, MOODY'S ANALYTICS, INC, article "Orbis database - Powering the business of certainty The world's most powerful comparable data resource on private companies"
CHENJIEYOUSEF SAAD: "Dense subgraph extraction with application to community detection", IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, vol. 24, no. 7, 2010, pages 1216 - 1230, XP011445000, DOI: 10.1109/TKDE.2010.271
EUROPEAN COMMISSION, MONITORING, REPORTING AND VERIFICATION OF EU ETS EMISSIONS, Retrieved from the Internet <URL:https://ec.europa.eu/clima/eu-action/eu-emissions-trading-system-eu-ets/monitoring-reporting-and-verification-eu-ets-emissions_en>
GARCIA-DURANALBERTOMATHIAS NIEPERT: "Learning graph representations with embedding propagation", IN ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS, 2017, pages 5125 - 5136
HAN XIAO ET AL: "Knowledge Graph Embedding Based on Subgraph-Aware Proximity", 17 November 2018, ADVANCES IN DATABASES AND INFORMATION SYSTEMS; [LECTURE NOTES IN COMPUTER SCIENCE; LECT.NOTES COMPUTER], SPRINGER INTERNATIONAL PUBLISHING, CHAM, PAGE(S) 306 - 318, ISBN: 978-3-319-10403-4, XP047495643 *
NIE BIN ET AL: "Mining Multivariate Discrete Event Sequences for Knowledge Discovery and Anomaly Detection", 2020 50TH ANNUAL IEEE/IFIP INTERNATIONAL CONFERENCE ON DEPENDABLE SYSTEMS AND NETWORKS (DSN), IEEE, 29 June 2020 (2020-06-29), pages 552 - 563, XP033800723, DOI: 10.1109/DSN48063.2020.00067 *
PECHMARTINJAROSLAV VRCHOTAJIRI BEDNAR: "Predictive maintenance and intelligent sensors in smart factory", SENSORS, vol. 21, no. 4, 2021, pages 1470, Retrieved from the Internet <URL:https://mdpi-res.com/d_attachment/sensors/sensors-21-01470/article_deploy/sensors-21-01470-v2.pdf>
QINGYUN SUN ET AL: "SUGAR: Subgraph Neural Network with Reinforcement Pooling and Self-Supervised Mutual Information Mechanism", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 22 March 2021 (2021-03-22), XP081901908 *
SOMOVANDREY ET AL.: "Deployment and evaluation of a wireless sensor network for methane leak detection", SENSORS AND ACTUATORS A: PHYSICAL, vol. 202, 2013, pages 217 - 225, Retrieved from the Internet <URL:https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.723.7644&rep=rep1&type=pdf>
VENGERTSEVDMITRYHEMAL THAKKAR: "Anomaly detection in graph: unsupervised learning, graph-based features and deep architecture", DEPT. COMPUT. SCI., STANFORD UNIV., STANFORD, CA, USA, TECH. REP, 2015
WAN GUOJIA ET AL: "Adaptive knowledge subgraph ensemble for robust and trustworthy knowledge graph completion", WORLD WIDE WEB, BALTZER SCIENCE PUBLISHERS, BUSSUM, NL, vol. 23, no. 1, 15 October 2019 (2019-10-15), pages 471 - 490, XP037044053, ISSN: 1386-145X, [retrieved on 20191015], DOI: 10.1007/S11280-019-00711-Y *
WENFENG PENG ET AL.: "A Sensor Network System for Process Unit Emissions Monitoring", PRESENTED AT THE AIR AND WASTE MANAGEMENT ASSOCIATION AIR QUALITY MEASUREMENT METHODS AND TECHNOLOGY CONFERENCE, 3 April 2019 (2019-04-03), Retrieved from the Internet <URL:https://cfpub.epa.gov/si/si_public_file_download.cfm?p_download_id=538399&Lab=NRMRL>
WILLIAM L HAMILTON ET AL: "Inductive Representation Learning on Large Graphs", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 7 June 2017 (2017-06-07), XP080768222 *
YU ZHENG ET AL: "Generative and Contrastive Self-Supervised Learning for Graph Anomaly Detection", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 23 August 2021 (2021-08-23), XP091036669 *

Similar Documents

Publication Publication Date Title
Lu et al. Digital twin-enabled anomaly detection for built asset monitoring in operation and maintenance
Capozzoli et al. Fault detection analysis using data mining techniques for a cluster of smart office buildings
Hayes et al. Contextual anomaly detection in big sensor data
Ge et al. Plant-wide industrial process monitoring: A distributed modeling framework
US8365019B2 (en) System and method for incident management enhanced with problem classification for technical support services
Marocco et al. Integrating disruptive technologies with facilities management: A literature review and future research directions
Mounce et al. Pattern matching and associative artificial neural networks for water distribution system time series data analysis
Yacout Fault detection and diagnosis for condition based maintenance using the logical analysis of data
US11494690B2 (en) Apparatus and method of high dimensional data analysis in real-time
Perez et al. Forget the myth of the air gap: Machine learning for reliable intrusion detection in SCADA systems
Harinarayan et al. XFDDC: eXplainable Fault Detection Diagnosis and Correction framework for chemical process systems
Lin et al. Predictive analytics for building power demand: Day-ahead forecasting and anomaly prediction
Escobar et al. Combined generative topographic mapping and graph theory unsupervised approach for nonlinear fault identification
Xie et al. Logm: Log analysis for multiple components of hadoop platform
Kameswari et al. Sensor data analysis and anomaly detection using predictive analytics for process industries
Yan et al. A comprehensive survey of deep transfer learning for anomaly detection in industrial time series: Methods, applications, and directions
CN115102834A (zh) 一种变更风险评估方法、设备及存储介质
Chowdhury et al. Internet of Things resource monitoring through proactive fault prediction
Paramkusem et al. Classifying categories of SCADA attacks in a big data framework
Mendia et al. A novel approach for the detection of anomalous energy consumption patterns in industrial cyber‐physical systems
Singh et al. An efficient software fault prediction model using cluster based classification
WO2023202789A1 (fr) Procédé et système de prise en charge de surveillance d&#39;émissions
Merkt Predictive models for maintenance optimization: an analytical literature survey of industrial maintenance strategies
Sun et al. Robust Contrastive Learning and Multi-shot Voting for High-dimensional Multivariate Data-driven Prognostics
Mansouri et al. Explainable fault prediction using learning fuzzy cognitive maps

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22793168

Country of ref document: EP

Kind code of ref document: A1