RU2819568C1 - Method of detecting training data for machine learning of computer system of industrial internet of things powered by rechargeable battery - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: invention relates to a method of detecting training data for machine learning of a computer system of the industrial Internet of things powered by rechargeable battery. Method comprises interaction between an industrial Internet of things (IIoT) system, a training data detection system and a training data processing and storage system, during which contextualized training data are detected in remote databases in computer systems for processing and storing training data with access via an Internet communication network by analysing the correspondence of the semantic annotation of the training data and the semantic description of objects and parameters of the industrial Internet of things in the form of concepts and attributes of the industrial Internet of things ontology concepts, as a result of which a command signal is generated to increase the charge consumption of the rechargeable battery for the device using machine learning to start the machine learning program in an Internet of things computer system powered by a rechargeable battery.

EFFECT: longer autonomous operation of an industrial Internet of things computer system powered by a current charge of a rechargeable battery using machine learning.

2 cl, 2 dwg, 2 tbl

Description

Field of technology to which the invention relates

The invention relates to the field of computer technology, and more specifically to self-powered computer systems for the industrial Internet of things. The invention solves the technical problem of increasing the battery life of an Industrial Internet of Things computer system powered by the current charge of a rechargeable battery using machine learning. The technical result consists in storing the current value of energy consumption from a rechargeable battery by a computer system of the industrial Internet of things until a signal is received indicating the detection of contextualized training data and with a subsequent increase in the charge consumption of the rechargeable battery to launch a machine learning program.

State of the art

The prior art has been disclosed to the extent known to the applicants. The work of the inventors, according to the description, is devoted to the basics and aspects of the description of the invention, which can be qualified as prior art necessary for understanding the essence of the invention and examination of the application. The level of technology is determined by the following analogues of the invention known to the applicant.

There is a known analogue of the invention “Methods and systems for evaluating learning objects using machine learning algorithms” (RU 2672394 C1, IPC G06N 5/02 (2006.01), G06N 99/00 (2010.01), 07/26/2017), where the group of inventions relates to the field of machine learning learning and can be used to evaluate learning objects. The technical result is to increase the efficiency of the machine learning algorithm while saving computing resources. The method comprises obtaining a first set of training samples containing a plurality of features, iteratively training a first predictive model based on the plurality of features, and generating a corresponding first prediction error indicator; analyzing a corresponding first prediction error measure for each iteration to determine an "overfitting" point, and determining at least one initial estimation point; obtaining data from a new set of training objects and iteratively retraining using at least one training object of the first predictive model, starting from at least one initial evaluation point, to obtain a plurality of retrained first predictive models, and generating a corresponding prediction error measure after retraining. Based on the plurality of prediction error metrics after retraining and the plurality of corresponding first prediction error metrics, one of the first sets of training samples and at least one training object are selected.

The technical problem of increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery when using machine learning when implementing an analogue of the invention cannot be solved, because the analogue of the invention does not disclose a method for detecting the first set of training samples, which does not allow receiving a signal with a message that contextualized training data has been detected and a subsequent increase in the chargeable battery drain to run the machine learning program.

There is a known analogue of the invention “Knowledge transfer in smart environments” (US 9,251,463 B2, IPC G06N 5/02 (2006.01), G06K 9/00 (2006.01), G06K 9/62 (2006.01) 02/2/2016), where device action patterns are generated from one or more existing smart environments (eg, seed spaces) based on sensor data from one or more existing smart environments that respond to known activities. A target activity pattern is then created and generated for a new smart environment, such as a target space. The activity patterns of the source space are then matched with the target action patterns to enable action recognition based on sensory data received from the target space. In this case, generating initial activity patterns from a plurality of source spaces involves obtaining sensory event data from a plurality of source spaces. Spaces: recognizing activities based on sequential sequences and combining similar activities to form patterns of original activities; generating target activity templates for the target space; matching source activity patterns to target activity patterns based on at least the spatial similarity between the activities of the source activity patterns and the activities associated with the target activity patterns, or the temporal similarity between the activities of the source activity patterns and the activities associated with the target activity patterns. Target activity patterns may be tagged to enable activity recognition based on sensory data from the target space.

The technical problem of increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery when using machine learning when implementing an analogue of the invention cannot be solved, because in the analogue, to determine patterns of device actions by type of activity, a combination of similar types of activities or comparison is used activity templates of the source space with target action templates to ensure recognition of actions based on sensory data received from the target space, as a result of which it is impossible to first, at the stage of forming the activity template of the source space, assess the correspondence of the content of the intellectual environment and the content of the subject area of the target activity template, to increase the confidence in obtaining target activity patterns for the target space, which does not allow a signal to be received indicating that contextualized training data has been detected and with a subsequent increase in the charge consumption of the rechargeable battery to run the machine learning program.

There is a known analogue of the invention “Transfer learning for predictive model development” (US 2015/0235143 A1, IPC G06N 5/02 (2006.01), G06N 99/00 (2006.01), 02/07/2015), where methods, systems and devices, including computer programs , are used to transfer learning within the predictive model. The system may receive data to train a predictive model, wherein the training data and one or more training methods are used to train multiple predictive models. The variables and predictive model are then selected from the trained predictive models. The selected variables are then transferred to the selected type of predictive model as long as they reduce the value of the prediction error. This involves obtaining a plurality of predictive modeling training data, dividing the training data into a plurality of subsamples, and training a plurality of different types of predictive models using one or more of the plurality of subsamples and one or more training methods.

The technical problem of increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery when using machine learning when implementing an analogue of the invention cannot be solved, because the analogue of the invention does not disclose a method for obtaining training data for predictive modeling, which does not allow obtaining signal indicating that contextualized training data has been detected and then increasing the charge consumption of the rechargeable battery to run the machine learning program.

There is a known analogue of the invention “Adaptive energy management” (RU 2436144 C2, IPC G06F 1/32 (2006.01), G06F 9/44 (2006.01), G06F 1/26 (2006.01), 03/8/2007), where a method is proposed that implements measures saving energy based on the amount of power that is available from the power source and includes identifying the current amount of energy that is available from the power source. A determination is then made as to whether the current amount of energy is associated with a reduced performance state. If the current amount of power is associated with a reduced performance state, the method reconfigures the power consuming devices to place the computer in a reduced performance state.

The technical problem of increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery when using machine learning when implementing an analogue of the invention cannot be solved, since the proposed method using polling of an application program does not reveal the essence and reasons for transmitting a signal to start a machine program learning to identify the amount of remaining power.

There is a known analogue of the invention “Operations of a mobile device with optimization of battery consumption” (RU 2609136 C2, IPC G06F 1/32 (2006.01), G06F 9/44 (2006.01), 06.06.2012), where the technical result is to save battery power in devices by task delays. Disclosed is a method for conserving battery power in a battery-powered device, comprising queuing at least one latency-sensitive task for execution thereafter; detecting the start of a device battery charging event after said queuing; predicting that the charging event is a long-running charging event based on a charging profile for the device, wherein the long-running charging event is a charging event with a duration greater than a predetermined threshold for allowing execution or execution of a delay-tolerant task of the at least one delay-tolerant task during a charging event; and allowing execution of the queued at least one delay-tolerant task during the charging event after a predetermined period of time has elapsed since said detection.

The technical problem of increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery when using machine learning when implementing an analogue of the invention cannot be solved, since the proposed method with queuing the program involves delaying the launch of the machine learning program relative to the period of time from the beginning device battery charging events, rather than relative to the signal, reporting the detection of contextualized training data and subsequently increasing the charge drain of the rechargeable battery to run the machine learning program.

The analogue closest to the invention - the prototype - is the “Distributed embedded data and knowledge management system integrated with the PLC data archive” (RU 2701845 C1, IPC G05B 19/05 (2006.01), G06N 5/02 (2006.02) 12/10/2015) , where the invention relates to computer technology. The technical result is to ensure the availability of data from a local data archive in a distributed data infrastructure. A data storage system in an industrial manufacturing environment comprises a distributed data management system stored on a plurality of intelligent programmable logic controller devices, including a distributed data management component; a contextualization component for generating contextualized data by annotating the contents of a process image area with automation system context information, a data archive component for locally storing the contents of a process image area and contextualized data, and providing access to the content in a distributed data management system through a distributed data management component and a data analytics component for executing one or more reasoning algorithms to analyze data using a distributed data management component. The prototype in question uses contextualized data, obtained by annotating the contents of a process image component with contextual information from the automation system to facilitate its subsequent interpretation. Contextual information can include any information that describes the meaning of the data. For example, the data context in automation systems may include information about the device that generated the data (e.g., a sensor), the structure of the automation system (e.g., plant topology), the operating mode of the system (e.g., idle time), and the automation software and its status at the time of data generation and/or the product/batch that was being produced at the time of data generation. The context can be represented by a standard modeling language (for example, Web Ontology Language, the language for describing web ontologies OWL), Resource Description Framework (resource description environment), where the meaning of language constructs is formally defined. In this case, one or more declarative knowledge models used by the contextualization component of each corresponding smart programmable logic controller device contain ontologies expressed using the ontology description language OWL.

The technical problem of increasing the battery life of an Industrial Internet of Things computer system powered by the current charge of a rechargeable battery when using machine learning when implementing a prototype of the invention cannot be solved, because in the prototype of the invention the data storage system is used only for the specified set of devices of intelligent programmable logic controllers and uses one or more reasoning algorithms to analyze data stored in a distributed system by just this set of smart programmable logic controller devices, which does not allow discovery of contextualized data outside of said industrial storage system, and there are no processes and devices using computational methods that allow learning from contextualized data after receiving a signal that contextualized training data has been detected and then increasing the charge consumption of the rechargeable battery to run the machine learning program.

Disclosure of the invention

The technical result consists in storing the current value of energy consumption from a rechargeable battery by a computer system of the industrial Internet of things until a signal is received indicating the detection of contextualized training data and with a subsequent increase in the charge consumption of the rechargeable battery to launch a machine learning program.

The essence of the proposed method is the preliminary detection of contextualized training data related to the context of the IIoT domain by determining the measure of semantic proximity of the ontology of the IIoT domain and the semantic annotation of training data stored in remote databases of various formats, where the data is previously semantically described based on knowledge where upon detection, a command signal is generated and transmitted to increase the current energy consumption of the rechargeable battery to run a machine learning program by the IIoT computer device.

To detect contextualized training data, remote databases with training data are created manually or semi-automatically in a computer system for processing and storing training data.

The next action is the creation by an expert, based on knowledge of the context of the training data, manually or semi-automatically, of an organized structure from sets of facts of the subject area in the form of a machine-readable ontology of the corresponding subject area of the industrial Internet of things using certain rules and procedures that are not within the scope of this disclosure, and to create Ontologies describe the semantic meaning of data in the following ways.

Certain methods include describing the facts of the subject area using a standard modeling language, for example, Ontology Web Language, (web ontology description language, OWL), Resource Description Framework (RDF resource description environment), but not limited to them, which formally define the meaning of language constructs for a formalized description of the facts of the subject area of the industrial Internet of things IIoT. In this case, facts within the framework of this disclosure are considered to be information contained in a set of data in an objective form of presentation, and the training data is presented in the form of machine-readable databases with semantic annotation, including a description of database objects, object parameters, which within OWL are considered as concepts and attributes respectively.

The next step is to transfer the domain ontology in machine-readable form to a computer training data discovery system for data storage and processing through a software training data controller.

The next action is the automatic creation or semi-automatic creation by an expert of a description of the context of the training data in the form of a semantic annotation in a computer system for processing and storing training data, while within the framework of this disclosure, the creation of a semantic annotation can take place without using a domain ontology.

Semantic annotation is considered within the framework of this disclosure as an information component of the semantic posting of an enterprise specification in accordance with GOST R ISO/IEC 15414-2017 “Information technologies. Open distributed processing. Reference model. Enterprise Description Language”, which allows you to determine that the specific objects modeled in the specifications of the IIoT computer system and the objects described in the specifications of the training data databases are in fact the same object, and such a description of such matching objects refers to contextualized training data.

Semantic annotation is described, but not limited to what follows, in the source Suleymanova, A. M., Yakovlev, N. N. Semantic annotation and multi-aspect data model in requirements management. Software products and systems. No. 2. 2011. pp. 45-48; The ontological approach to annotation is described in the source Le Hoai, Tuzovsky A. F. Semantic annotation of documents in digital libraries. News of Tomsk Polytechnic University. 2013. Volume 323. No. 5. pp. 157-164; a way to compile a semantic annotation, but not limited to it, are simple semantic networks, described in the source Artyushina, L. A. Methods for presenting information in simple semantic networks. Scientific and technical bulletin of information technologies, mechanics and optics. 2020. T. 20. No. 3. P. 382-393; stochastic models are described in the source Voznesenskaya, T.V., Lednov, D.A. System for automatic annotation of texts using a stochastic model. Machine learning and data analysis, 2018. Vol. 4, No. 4. pp. 266-279.

The next action is to transfer the semantic annotation to the software adapters of the training data with a semantic annotation in the form of machine-readable data for storage and processing by a computer system for processing and storing training data, where the training data is stored in remote databases in various formats for representing and encoding data, and the concept of “remote” for the purposes of this disclosure, means that the computer system for processing and storing training data is located outside the local network of the computer system of the Industrial Internet of Things powered by a rechargeable battery, or outside the direct physical connection of the computer system of the Industrial Internet of Things powered by a rechargeable battery to the computer system for processing and storing training data.

The next action is the formation, with the help of an expert or automatically, on a device that uses the result of machine learning as part of an industrial Internet of things computer system powered by a rechargeable battery, information about objects of the industrial Internet of things, the parameters of these objects in an objective form of representation of a machine-readable data processing program.

The next action is the formation by the device using the result of machine learning of a set of information about objects of the industrial Internet of things IIoT, the parameters of these objects in machine-readable form in the form of a request for the detection of contextualized training data, which, but not limited to, corresponds to the software service for adapting the parameters of the preliminary national standard PNST 423-2020 (ISO/IEC 20005:2013 “Information technologies. Sensor networks. Services and interfaces supporting joint data processing in intelligent sensor networks”) to provide the necessary computing power in case of dynamic changes in the requirements of a device using machine learning.

The next step is to send a discovery request for contextualized training data to the IIoT interconnect device of the IIoT computer system powered by a rechargeable battery.

The next action, based on receiving a request to detect contextualized training data, is the generation of a command signal by the IIoT intersystem communication device to transfer request information to detect contextualized training data with a description of the types of objects and their parameters in machine-readable form.

The next action, based on the generation of a command signal by the IIoT interconnect device to carry request information about the detection of contextualized training data, is the transmission of a command signal over the Internet communication network from the IIoT interconnect device in the direction of the intelligent peripheral device of the computer system for controlling access to training data, while an intelligent peripheral device is considered as a device that has the ability to independently perform part of the functions of a central processor for processing information; a command signal is considered as a command message in a machine-readable form of representation, intended for transmission over channels and lines of wired and wireless communications in the form of a telecommunication signal.

The next action, based on the intelligent peripheral device receiving a command signal to transfer request information for the detection of contextualized training data, is the manipulation of the request data for the detection of contextualized training data by the software controller of the training data by correlating the received information with a description of the types of objects, their parameters in machine-readable form with the ontology subject area for detecting concepts and attributes of the ontology concepts of the subject area that match or have similar semantic properties, and the formation of rules for detecting contextualized training data in machine-readable form, and the rules can be implemented by means of descriptive logic, but not limited to it, while the method of manipulation consists in correlating the concept and information of a data processing program with a description of object types, their parameters and the method of forming rules for detecting contextualized training data does not relate to this disclosure, but corresponds, without limitation, to the methods and methods described in the source Palagin, A. V., Petrenko, N. N. On the issue of system-ontological integration of domain knowledge. Mathematical machines and systems. 2007. pp. 63-75; formation of rules for detecting training data, but not limited to, can be formed by the methods in the dissertation Leshevoy, I. A. Method of automated filling of databases of ontological type: dissertation for the degree of candidate of technical sciences in specialty 2.3.5. St. Petersburg State University, 2022, 204 p. with applications.

The next action, based on the formation of rules for detecting contextualized training data, is the generation of a command signal on an intelligent peripheral device with rules for detecting contextualized training data in machine-readable form.

The next action, based on the formation of a command signal with rules for detecting contextualized training data, is the transmission of a command signal over the Internet communication network from an intelligent peripheral device of a computer system for controlling access to training data to the intersystem communication device of the data storage of a computer system for processing and storing training data, while the command the signal is considered as a command message in a machine-readable form of representation, intended for transmission over channels and lines of wired and wireless communications in the form of a telecommunication signal.

The next action, based on transmitting a command signal with rules for detecting contextualized training data over the Internet communication network, is receiving a command signal with rules for detecting contextualized training data on the intersystem communication device of the data warehouse.

The next action, based on the reception by the intersystem communication device of the data store of a command signal with rules for detecting contextualized training data over the Internet communication network, is transferring the rule for detecting contextualized training data with a description of IIoT concepts, attributes of IIoT concepts in machine-readable form to the software adapter of training data with semantic annotation.

The next action, based on the transfer of rules for detecting contextualized training data in machine-readable form, is the execution of the rules for detecting contextualized training data by a software adapter to determine the measure of agreement between the semantic annotation of the IIoT concepts and the attributes of the IIoT concepts, and the method for determining the correspondence is not related to the present disclosure, but may be implemented as, but not limited to, the Semantic Query-based Annotation, P2P Sharing System, as cited by Fernandez-Garcia, N., Blazquez-del-Toro, JM, Sanchez -Fernandez, L., Luque, V. Exploiting User Queries and Web Communities in Semantic Annotation. ^5th International Workshop on Knowledge Markup and Semantic Annotation (SemAnnot 2005), 2005, Galway, Ireland, Vol.185.

The next action, based on the detection of a match between semantic annotation and IIoT concepts, attributes of IIoT concepts, is the formation and transmission by the software adapter of training data with semantic annotation of an inter-machine response message from the computer system for processing and storing sensory data towards the computer system for detecting training data about the detection of contextualized training data or lack of transmission of an inter-machine response message.

The next action, based on the transmission of an inter-machine response message about the detection of contextualized training data, is the transmission of a signal with an inter-machine response message over the Internet communication network from the data storage interconnect device to the intelligent peripheral device.

The next action, based on the transmission of the machine-to-machine response message about the detection of contextualized training data, is for the intelligent peripheral device to receive a signal with the machine-to-machine response message about the detection of contextualized training data.

The next step, based on receiving the machine-to-machine response message about the detection of contextualized training data, is to transmit the machine-to-machine response message about the detection of contextualized training data from the smart peripheral device to the IIoT interconnect device over the Internet communication network.

The next action based on transmitting the machine-to-machine contextualized training data detection response signal is to receive the contextualized training data detection machine-to-machine response signal at the IIoT interconnect device.

The next action, based on receiving the machine-to-machine response message about the detection of contextualized training data, is to transmit an intra-system trigger command from the IIoT device to increase the charge consumption of the rechargeable battery.

The next action, based on the IIoT interconnect device transmitting an in-system trigger signal to increase consumption, is to increase the charge consumption of the rechargeable battery to trigger the device applying machine learning machine learning programs in the Internet of Things computer system powered by the rechargeable battery.

Carrying out the invention

Implementation of the claimed invention, expressed in increasing the battery life of a computer system of the industrial Internet of things powered by the current charge of a rechargeable battery by maintaining the current value of energy consumption from the rechargeable battery by the computer system of the industrial Internet of things and increasing the energy consumption from the rechargeable battery by the computer system of the industrial Internet of things for startup machine learning program only upon receipt of a signal reporting the detection of contextualized training data is performed in one or more of the same or different embodiments.

The most obvious embodiment is designated as “preferred” for the purpose of disclosing essential features, and the terms “comprising,” “including,” “having,” “enabling,” and the like are used with respect to the embodiment and are synonymous.

It should be understood that other embodiments and structural or logical changes of the claimed method may be used, which can be done without going beyond the scope of the preferred embodiment of the invention and the corresponding example of disclosure, other embodiments and disclosures may be described differently, most useful for understanding of what is stated, in a manner and by example. Accordingly, the following detailed description should not be construed in a limiting sense, the set of embodiments being defined by the appended claims.

In the preferred embodiment, the expert uses an ontological approach that makes it possible to correlate the description of IIoT industrial Internet of things objects in the form of IIoT concepts and IIoT attributes with the semantic annotation of the training data, to determine the correspondence of the semantic annotation to the IIoT concepts and the attributes of the IIoT concepts, and then the semantic correspondence of the contextualized training data is automatically determined data and IIoT domain, after which a machine-to-machine message is automatically generated about the detection of contextualized training data, which is automatically converted into an intra-system command signal to operate to increase the charge consumption of the rechargeable battery, which is necessary for running the machine learning program by the machine learning device, Moreover, before receiving the intra-system command signal, the current charge of the rechargeable battery prior to receiving the signal is maintained, which increases the battery life of the computer system of the industrial Internet of things when using machine learning.

For the purposes of this disclosure, but not limited to it, the following is an example of an industrial Internet of things computer system powered by a rechargeable battery, where the device using machine learning takes the form of a device for remote temperature monitoring of electrical network objects, including transformer windings and magnetic circuits, which, after startup machine learning programs receive contextualized training data on winding temperatures via the Internet telecommunications network, and use the machine learning program, for example, but not limited to, to classify the state of a transformer by the temperature of windings and magnetic cores in order to prevent the occurrence of false alarm messages, as well as to predict winding temperature values and magnetic circuits for predicting moments of maximum, minimum and abnormal temperatures to solve problems of proactive management and predictive technical operation.

In the preferred embodiment, the expert initially forms a domain ontology to describe knowledge about objects of the industrial Internet of things, namely, but not limited to, sensors, transducers and their parameters, which are objects of management and control using a machine learning program, for example in terms of output prediction failure, classification to identify abnormal readings and abnormal operation of industrial facilities, but not limited to them.

The domain ontology is formed within the framework of a semantic information approach to the technology of creating knowledge bases from the subject domain of the industrial Internet using task-specific rules and procedures.

In another embodiment, the expert uses a previously developed domain ontology, which is used “as is” or with additions and changes made by the expert.

In a preferred embodiment, definitions of concepts, ontology attributes, and relationships between them can be described using a formal or logical description language, allowing the use of inference to derive facts from existing domain information.

In the preferred embodiment, generalized descriptions of Industrial Internet of Things objects are considered as concepts. In this case, a physical device is considered as an entity that has material existence in the real world, and a concept is an expert’s idea of some technical entity, part of this entity, which has a certain structure expressed by different groups of features, displayed by appropriate linguistic methods and means using ontology. The properties of objects are described through parameters that semantically express the properties of objects according to the technical data sheet of the product.

In a preferred embodiment, the following concepts and their corresponding attributes may be defined, but are not limited to them, collectively required to implement the invention.

A concept attribute is understood as a parameter or characteristic of the object under consideration in the form of a separate information structure.

The Temperature Sensor concept represents an Industrial Internet of Things object that measures the temperature of one or more physical entities and outputs digital data that can be transmitted over the Internet communication network in machine-readable form in the subject area of the disclosure.

The concept “Transformer Winding” is a description of data in machine-readable form, which describes a set of turns that form an electrical circuit in order to obtain one of the transformer voltages and is an object of the Industrial Internet of Things in the subject area of the disclosure.

The “Surface Temperature” concept is a description of data in machine-readable form that quantitatively expresses different degrees of heating of the surface of physical bodies and is an object of the Industrial Internet of Things in the subject area of disclosure.

The “Surface Temperature” concept has the “Temperature Value” attribute.

In a preferred embodiment, describing an Industrial Internet of Things domain requires defining a set of concepts to describe semantic correspondences using a domain ontology. In another embodiment, a ready-made description of the subject area of the industrial Internet of things is used in machine-readable form as is or with changes and additions.

In a preferred embodiment, the set of concepts for describing sensory data, exemplified by "temperature sensor" and "surface temperature", is specified by, but not limited to, a generalized set of conceptual definitions, which is formally written using the sign " $$\equiv$$ ", which means semantic matching:

{sensor $$\equiv$$ temperature sensor, temperature $$\equiv$$ surface temperature, temperature $$\equiv$$ degree Celsius, temperature $$\equiv$$ Celsius, temperature $$\equiv$$ degrees Fahrenheit, temperature $$\equiv$$ Fahrenheit, temperature $$\equiv$$ Kelvin, temperature $$\equiv$$ Kelvin, degrees Celsius $$\equiv$$ °C, degrees Fahrenheit $$\equiv$$ °F, Kelvin $$\equiv$$ K, Celsius $$\equiv$$ °C, Fahrenheit $$\equiv$$ °F Kelvin $$\equiv$$ K},

where “sensor” is a concept corresponding to the concept of “temperature sensor”;

“temperature” is a concept for describing a measured property of a medium corresponding to the concept of “surface temperature”;

“degree Celsius”, “degree Fahrenheit”, “Kelvin”, “Celsius”, “Fahrenheit”, “Kelvin” are attributes that correspond to the concept for describing the derived units of quantities of the attribute “Temperature value”;

“°C”, “°F”, “K” - correspond to the concept for describing units of measurement of derived units of quantities.

The formation in a machine-readable form of a description of each type of object and parameter of an IIoT object by a data processing program launched by a device using machine learning as part of an industrial Internet of things computer system powered by a rechargeable battery is performed automatically with the participation of an expert or automatically.

The description of each IIoT object type and IIoT object parameter may be in, but not limited to, Extensible Hypertext Markup Language XML format, in the disclosure example:

where item is the designation of an element that is part of an array of elements;

addr - the address parameter of the computer system of the industrial Internet of things (the IP address is indicated in the example);

name - the name of the object as a designation of the field of application of machine learning (in the example, a transformer winding is indicated);

type - type of object of the computer system of the industrial Internet of things (in the example, a temperature sensor is indicated);

value - parameter indicating the value of all temperature sensor readings.

To generate a request for the detection of contextualized training data with a description of the types of IIoT objects and parameters of IIoT objects within the framework of this disclosure, but not limited to it, a request corresponding to the request description format according to PNST 423-2020 (ISO/IEC 20005:2013) is used:

With the contextualized training data discovery request "QoS-PROFILE-ESTABLISH.request", the machine learning device requests the establishment of a Quality of Service (QoS) profile to control the machine learning program in terms of the availability of contextualized training data. The query design of the preferred embodiment proposes using the query parameters in Table 1.

Table 1
Contextualized training data discovery query parameters Query parameter name Description of the contents of the parameter offered within the scope of this disclosure QoSRequestorID Industrial Internet computer system identifier. QoSProfileManagerID An identifier and associated semantic description of the scope of application of the contextualized training sequence (e.g., “Transformer winding temperature”). QoSProfileID Identifier and description of each concept in the context of use of the training sequence (for example, “Temperature Sensor”). QoSProfileObjectList A list of concepts and their associated attributes (for example, sensor metric values).

Another embodiment is to assign values to the parameters “QoSProfileManagerID”, “QoSProfileID”, “QoSProfileObjectList” using information links to a ready-made domain ontology.

In a preferred embodiment, using an IIoT interconnect device, command signals are generated to carry request information for the detection of contextualized training data with a description of the types of objects, their parameters in machine-readable form, and the request to detect contextualized training data is considered as a command segment as part of an inter-machine message, where The command segment consists of a command word, which for the purposes of this disclosure, but is not limited to it, is denoted by "QoS-PROFILE-ESTABLISH.request", and the data words of the command segment are formed by the parameters QoSProfileManagerID, QoSProfileID, QoSProfileObjectList and parameter values according to provisions of GOST R 53074-2003 “Trunk serial interface of electronic module systems. Testing of serial samples of interface modules operating in bus controller mode. General requirements for control methods."

In the preferred embodiment, the preferred method of transmitting signals for transferring information is encoded bit-by-bit transmission through communication means of the Internet communication network, using at the physical level a wired or wireless, including radio, transmission medium with encoding methods, but not limited to, NRZ encoding methods, 2B1Q, 4B/5B; using asynchronous and/or synchronous signaling protocols IEEE 802.x, LLC2, LLC3, HDLC at the data link level, but not limited to them; using IP v4, IP v6 interconnection protocols at the network level within the framework of optical communication networks SDH and DWDM, fixed broadband access networks PON and xTTP, wireless broadband access networks WiFi, cellular networks of 2G/3G/4G/5G generations.

In a preferred embodiment, the preferred embodiment for describing profile metadata is the Resource Description Framework (RDF), as part of the Semantic Web concept, where the entity in the preferred embodiment is a sensor.

The ontological approach in the preferred embodiment is used to represent domain knowledge of the Industrial Internet of Things, as well as to describe the semantic rules for searching a training sequence. Automated processing of information (metainformation) about objects of the industrial Internet of things, presented in accordance with the ontology of the domain, is carried out using various information technologies and tools, namely the language of rules of the semantic web SWRL (Semantic Web Rule Language), the language of restrictions of forms SHACL (Shapes Constraint Language), where restrictions are written in the language and protocol of queries to data presented according to the RDF SPARQL model (Semantic Protocol and RDF Query Language), while the data description is translated into OWL and associated with ontology classes.

Within the scope of this disclosure, to understand the essence of the invention, but not limited to the rules discussed below, the rules for detecting training data are written through constraints in which the method (SELECT) selects data with measurement values of a physical quantity (testresult) only in the resource specified by the method (WHERE) , where, according to the domain ontology used to describe sensory data “Sensor-Measurement-Sampling-Update” SOSA (Sensor-Observation-Sampling-Actuator Ontology), observation values are selected using an object class (OBSERVATION), resulting in a property (hasResult) in in the form of a temperature concept (?Temp) for an object with properties (hasFeatureOfInterest) of a concept denoting the temperature of a transformer winding (?TransformerWinding), wherein the temperature is an object of the ontology class (Result), and a query regarding temperature combines using the (UNION) method samples of contextualized measurement data in as a class of artifacts (SAMPLE), and the sample has the property of inclusion (isSampleOf) in the result of a request for temperature indicators (?Temp) indicating that the temperature relates to a physical object in the form of a transformer winding (?TransformerWinding), which together forms the search rule The training sequence can be written in the following unified form, without being limited to it, and regardless of the specific database format:

In this case, a link of the form “http://www.w3.org/ns/sosa” is an external link to the existing SOSA ontology for use as a domain ontology. In addition to the SOSA ontology, other software technologies, tools, formats and tools other than those stated within the scope of this disclosure can be used to describe the subject area, constraints, build an information model and unified queries.

Within the scope of the present disclosure, receipt of a command signal to carry contextualized training data discovery request information by a computer training data discovery system causes the learning data controller software to run, which runs the domain ontology by manipulating the data and training data discovery rules, and the smart peripheral device generates the request detection of contextualized training data describing, within the scope of this disclosure, the transformer winding temperature (?TransformerWinding), and the concept of temperature (?Temp) in, but not limited to, the following unified form:

where the SELECT DISTINCT parameter means selecting different or unique meanings of the concept;

the FROM FILE parameter means reading the metadata of available databases from the temperature.rdf file, with the meta-description of the databases located at the http address dataset/sensor;

the WHERE parameter means selecting the values of the temperature concepts ?temp for the concepts of power transformer windings ?transformerwinding.

In the context of the present disclosure, a command signal for carrying contextualized training data discovery request information from an intelligent peripheral device by a computer training data discovery system to a data store interconnection device is generated, in the preferred embodiment, as a message whose command segment consists of a command word that is within the framework of this disclosure, but not limited to it, is designated by the “SELECT” method, and the data words of the command segment are formed by the parameters after “SELECT” and to the end of the form in accordance with the provisions of GOST R 53074-2003 in machine-readable form.

In the preferred embodiment, as part of a computer system for processing and storing training data for analyzing a request for detection of contextualized training data, a software adapter for training data is used with a previously developed by an expert or automatically generated semantic annotation in the format declared within the framework of this disclosure using the RDF language, but not limited to them, namely:

where the RDF parameter indicates the standard and notation to use;

the xmlns parameter contains a hypertext link to the rdf notation used;

the rdf:Description about parameter contains a text description of the data in the database files, in this example the temperature of the transformer;

the sen:type parameter specifies that the data contains temperature information;

the xs:datatype parameter specifies that the temperature is measured in degrees Celsius;

the sen:item parameter specifies that the temperature sensor data is represented as a list;

the sen:time parameter indicates the point in time of data collection;

the sen:value parameter indicates the temperature value;

the http://Dataset/temperature parameter indicates the path for reading the contents of the file describing the temperature databases.

In the preferred embodiment, a file describing temperature databases is generated using the RDF language within the scope of, but not limited to, the disclosure, namely, in the example:

where the parameters Database1:fileX, Database2:fileY define the conventional name of data arrays and data files;

the dc:title parameters define the description of the data array title, in the example the winding temperature and the outside air temperature;

parameters Database1:partno, Database2:partno determine the type and brand of the data source, in the example the brand of the power transformer.

In a preferred embodiment, the computer system for processing and storing training data, upon receiving a command signal from an IIoT interconnect device with a request to detect contextualized training data describing IIoT concepts, attributes of IIoT concepts, launches a software adapter of training data with semantic annotation, which compares in machine-readable form contents of a semantic annotation and a temperature.rdf file with a request to detect contextualized training data describing the types of IIoT concepts, attributes of IIoT concepts in machine-readable form.

As part of the present disclosure, a semantically annotated training data adapter detects, by comparison, a semantic match between the description of the concept ?temp and the semantic annotation, and a specific match between the concept ?transformerwinding and the contents of the data array title dc:title "Winding_Temperature" and the value rdf:Description about "Transformer temperature" ", and the method of determining the semantic match is not relevant to the present disclosure. As a result, the contextualized training data is considered to be found in the Database1:fileX database, and the contents of the Database2:fileY database are considered decontextualized for the purposes of this disclosure.

In a preferred embodiment, the semantically annotated training data adapter generates a machine-to-machine contextualized training data detection response message to the computer training data discovery system, which is generated using XML within the scope of, but not limited to, the disclosure in an example message:

where SEND ON CONVERSATION is an instruction for transmitting data within a dialog with the identifier @annotation_handle (annotation processing), where the specified identifier and dialog are previously described in the software adapter in machine-readable form;

MESSAGE TYPE - an indication of the type of message being sent, in this disclosure this is a message indicating the path for reading the meta-description of the database [//Dataset/temperature] with contextualized training data;

@FindingTrainingDataset is an expression representing the contents of the response message.

As part of the present disclosure, a machine-to-machine response message from a semantically annotated training data adapter software to detect contextualized training data upon detection of a semantic annotation match to IIoT concepts and IIoT concept attributes is transmitted in machine-readable form from the data store interconnect device to the smart peripheral device by the computer-based training data detection system. in the form of a response message, which consists of response words, which within the scope of this disclosure, but not limited to it, are designated by the instructions "SEND ON CONVERSATION" and the instruction "SEND ON CONVERSATION", and data words attached to the instructions [//Dataset/temperature] , @annotation_handle, @FindingTrainingDataset, respectively, in accordance with the provisions of GOST R 53074-2003 in machine-readable form.

In the present disclosure, a signal with a machine-to-machine response message about the detection of contextualized training data transmitted over the Internet communication network from a data storage interconnect device and received by an intelligent peripheral device leads to the operation of the intelligent peripheral device, as a result of the operation, a machine-to-machine response message about the detection of contextualized training data is generated in the direction IIoT interconnection device corresponding within the framework of this disclosure to the request description format in accordance with PNST 423-2020 (ISO/IEC 20005:2013), namely:

Here, QoS-PROFILE-ESTABLISH.confirm confirms the detection of contextualized sensor data for the machine learning device.

The message structure of this disclosure proposes to use QoS parameters with the meaning in Table 2.

table 2
Suggested QoS Management Service Request Options Parameter name A description of the contents of the parameter offered within the scope of this disclosure. QoSRequestorID Industrial Internet computer system identifier. QoSProfileManagerID An identifier and associated semantic description of the scope of application of the contextualized training sequence (e.g., “Transformer winding temperature”). QoSProfileID Identifier and description of each concept in the context of use of the training sequence (for example, “Temperature Sensor”, “Temperature”). QoSProfileDBList Path to read the database with contextualized training data (for example, //Dataset/temperature).

In the context of the present disclosure, a machine-to-machine response message from an intelligent peripheral device about the detection of contextualized training data is transmitted in machine-readable form via signals to the IIoT interconnect device in the form of a response message to control the machine learning program, which consists of a response word that, in the context of the present disclosure, but not limited to it, is denoted by the word “QoS-PROFILE-ESTABLISH.confirm” and the data words “QoSRequestorID”, “QoSProfileManagerID”, “QoSProfileID”, “QoSProfileDBList” attached to it, in accordance with the provisions of GOST R 53074-2003, in machine-readable form.

In a preferred embodiment, signals with a machine-to-machine response message about the detection of contextualized training data are received at the IIoT machine-to-machine communication device.

In the preferred embodiment, the IIoT interconnect device, having received signals with the response word “QoS-PROFILE-ESTABLISH.confirm”, generates an internal command signal to operate to increase the charge consumption of the rechargeable battery.

In the preferred embodiment, the internal system command signal to operate to increase the charge consumption of the rechargeable battery is transmitted through physical wires with a value of m with a time duration necessary and sufficient to trigger the battery monitoring and control unit, which increases the charge consumption of the rechargeable battery to be triggered by a device using machine learning ,machine learning programs on an internet of things computer system powered by a rechargeable battery.

In this case, in a preferred embodiment, the battery is a system that includes one or more rechargeable batteries, modules or battery packs in accordance with GOST R IEC 62620-2016 “Rechargeable batteries and batteries containing alkaline or other non-acid electrolytes. Rechargeable batteries and lithium batteries for industrial applications”, wherein the battery included in a rechargeable battery, but not limited to, is intended for electrical recharge.

In a preferred embodiment, the rechargeable battery is conceptualized as two or more cells equipped with devices necessary for use, for example: a housing, terminals, and protection devices; Discharge means a process in which the electrical energy of a current source, under certain conditions, is released into an external electrical circuit, and the electrical energy is generated in the battery in accordance with the requirements of GOST R IEC 6427-2-2016 “Accumulators and storage batteries for renewable energy sources. General requirements and test methods. Part 2. Network application" or GOST R IEC 62620-2016.

In a preferred embodiment, the battery monitoring and control unit is an electronic system associated with the battery, the functions of which are entirely performed by the battery itself to monitor and/or control its condition.

The method proposed in the preferred embodiment makes it possible to describe the representation of the same IIoT objects and IIoT object parameters for different database formats as methods of storing training data, annotate the semantic content of data in different databases, semantically compare training data from different databases to ensure completeness and the integrity of training data without increasing the battery drain of the rechargeable device using machine learning and for various machine learning models.

The described objects, parameters, concepts, attributes, requests and response messages, ontologies used, signals may be implemented in a different order than in the preferred embodiment. There may be various additional object types, object parameters, concepts, attributes, ontologies, operations performed and/or operations described, and machine learning techniques that may be implemented in additional embodiments relative to the preferred embodiment.

Brief description of drawings

In the following detailed description, reference is made to the accompanying drawings, which form a part of the detailed description, in which like numerals denote like parts throughout, and in which an embodiment is shown by way of illustration that most fully reflects the disclosure of the invention, but does not constitute the only possible option that can be implemented in practice.

Figure 1 shows a block diagram of a system for implementing the interconnection of computer systems and the Internet communication network to implement a method for discovering training data for machine learning of an industrial Internet of things computer system powered by a rechargeable battery.

Figure 2 is a functional diagram of a method for discovering training data for machine learning of an Industrial Internet of Things computer system powered by a rechargeable battery.

Figure 1 schematically shows the Internet communication network (101) (hereinafter referred to as “Network 101”), which organizes the connection between a computer system for processing and storing training data (105) (hereinafter referred to as “System 105”), and a computer system for detecting training data (115 ) (hereinafter “System 115”) and an industrial Internet of things computer system powered by a rechargeable battery (125) (hereinafter “System 125”).

"System 105" technically consists of information technology equipment capable of storing and processing training data in a specific database format, receiving and processing signals and requests via "Network 101" in machine-readable form from "System 115", generating and issuing response messages and signals in machine-readable form towards “System 115”, support the functioning of system and application software of “System 105”.

“System 115” technically consists of information technology equipment capable of generating, transmitting requests and signals in machine-readable form to “System 105” via “Network 101”, receiving requests, signals in machine-readable form from “System 125” via “Network 101”, generate and transmit response messages, signals in machine-readable form from “System 125” via “Network 101”, support the functioning of system and application software of “System 115”.

“System 125” technically consists of information technology equipment capable of generating, transmitting requests, signals in machine-readable form via “Network 101” towards “System 115”, receiving and processing response messages, signals in machine-readable form from “System 115”, supporting functioning of System 125 system and application software.

In fig. 2 is a functional diagram of a method for detecting training data for machine learning of an Industrial Internet of Things computer system powered by a rechargeable battery.

"Training Data" (202) (hereinafter "Data 202") represents information technology equipment and data in machine-readable form to form a training sequence in the database format "NoSQL", "MySQL", "PostgreSQL", but not limited to these formats.

“Contextualized training data” (204) (hereinafter referred to as “Data 204”) is information technology equipment and contextualized training data in machine-readable form for the formation of a domain machine learning training sequence in a database format in a “NoSQL”, “MySQL” database format. , "PostgreSQL", but not limited to these formats.

“The Request for Discovery of Contextualized Training Data (210) (hereinafter referred to as “Request 210”) represents signals with a message in machine-readable form, transmitted over “Network 101” to “System 105”, describing IIoT concepts, attributes of IIoT concepts in machine-readable form.

The "Machine-to-Machine Response Message" (212) (hereinafter "Message 212") is a message signal in machine-readable form transmitted over the "Network 101" from the "System 105" indicating the detection of contextualized training data allowing the use of "Data 204" for machine learning.

The "data warehouse interconnect device" (208) is communication equipment for receiving and processing "Request 210", processing and transmitting "Message 212" over "Network 101" in the form of signals.

“Training data software adapter with semantic annotation” (206) is a computer program for determining the correspondence of semantic annotation to IIoT concepts and attributes of IIoT concepts on information technology equipment upon receipt of “Request 210” with the formation of “Message 212” describing the detected contextualized training data or with the absence of formation of “Message 212” when contextualized training data is not detected.

“Training Data Software Controller” (214) is a computer program of information technology equipment to support the industrial Internet of Things domain ontology, semantic rules for detecting a training sequence, generating requests and response messages in machine-readable form.

A “domain ontology” (216) is software and information technology to support the processing of query data in machine-readable form to define IIoT concepts, IIoT attributes, and a semantic description of the relationships between them, including an accurate and objective description of the terms that define each concept.

“Rules for detecting training data” (218) are software commands, data and information technologies to support the formation in machine-readable form of information about the semantic relationships of IIoT concepts, IIoT attributes for interpreting semantic annotations of the content of databases storing “Data 202”, “Data 204” "

"Request for discovery of contextualized training data (222) (hereinafter "Request 222") are command signals requesting the discovery of contextualized training data in machine-readable form, transmitted over "Network 101" from "System 125" to "System 115", with description of IIoT objects, parameters of IIoT objects in machine-readable form.

The "machine-to-machine response message" (224) (hereinafter "Message 224") is the machine-to-machine response message signals in machine-readable form transmitted over the "Network 101" from the "System 115" to the "System 125" about the detection of contextualized training data allowing the use "Data 204" for machine learning.

The "intelligent peripheral" (220) is a processor system with software and information technology to support signals for carrying Request 210, Message 212, Request 222, Message 224 information in machine-readable form for detecting contextualized learning data.

“Machine learning device” (226) is information technology, management and control equipment to support software that uses machine learning methods and algorithms to achieve the goals of the Industrial Internet of Things.

"Data Processing Program" (228) is software and information technology to ensure the performance of "System 125" in the event of dynamic changes in the requirements of the program using machine learning to automatically describe IIoT objects and parameters of IIoT objects.

“Machine Learning Program” (230) (hereinafter referred to as “Program 230”) is software and information technology for ensuring the functioning of computer machine learning models on information technology equipment in “System 125”.

A “rechargeable battery” (236) (hereinafter referred to as “Battery 236”) is one, two or more rechargeable battery cells equipped with devices necessary for use, including a battery monitoring and control unit.

The "in-system command signal" (234) (hereinafter referred to as "Signal 234") is a hardware signal, a software command signal, but not limited to, to operate the "Battery 236" and increase the charge consumption of the "Battery 236" and start the "Program 230".

The "IIoT interconnect device" (232) is a communication equipment for receiving and processing "Request 222", processing and transmitting "Message 224" over "Network 101" in the form of signals, generating and transmitting "Signal 234".

Claims (23)

1. A method for detecting training data for machine learning of an industrial Internet of things computer system powered by a rechargeable battery, characterized in that an increase in the charge consumption of a rechargeable battery to run a machine learning program occurs in response to a signal about the detection of contextualized training data, for which: create remote databases with training data in a computer system for processing and storing training data, and create a description of knowledge about the context of the training data in the form of a machine-readable domain ontology in the software controller of the training data of the computer system for detecting training data, and create a description of the context of the training data in the form of a semantic annotation in a computer system for processing and storing training data, and create a description of each object type and object parameter of the industrial Internet of things IIoT computer system in the form of a machine-readable data processing program for a device using machine learning, and generate a discovery request for contextualized training data describing IIoT object types and IIoT object parameters in machine-readable form, and perform the generation of command signals by the IIoT intersystem communication device with a request to detect contextualized training data with a description of the types of objects, their parameters in machine-readable form, and sending command signals requesting discovery of the contextualized training data over the Internet communication network from the IIoT interconnect device towards the intelligent peripheral device of the computer training data access control system, and receive command signals from the IIoT interconnect device requesting the discovery of contextualized training data describing the types of objects, their parameters in machine-readable form on the smart peripheral device, and form a semantic description of training data in machine-readable form upon request for the detection of contextualized training data with a description of object types in the form of IIoT concepts, object parameters in the form of attributes of IIoT concepts by manipulating domain ontology data in the form of rules for detecting contextualized training data in the computer training data controller training data detection systems, and generating command signals requesting discovery of contextualized training data describing IIoT concepts, attributes of IIoT concepts on the smart peripheral device, and transmitting command signals requesting detection of the contextualized training data from the smart peripheral device to the data storage interconnect device of the computer system for processing and storing training data over an Internet communication network, and receiving command signals requesting discovery of contextualized training data describing IIoT concepts, attributes of IIoT concepts on a data storage interconnect device of the training data processing and storage computer system, and transmitting from the data storage interconnect device a request for discovery of contextualized training data with a description of IIoT concepts, attributes of IIoT concepts in machine-readable form to a software adapter of training data with semantic annotation, and comparing, on the training data adapter with the semantic annotation, the stored semantic annotation and IIoT concepts and the attributes of the IIoT concepts, and generating a machine-to-machine response message from the training data software adapter with a semantic annotation about the detection of contextualized training data upon detection of a match between the semantic annotation and IIoT concepts and attributes of IIoT concepts, and transmit signals with a machine-to-machine response message about the detection of contextualized training data over the Internet communication network from the data store interconnect device to the intelligent peripheral device, and receiving, through the intelligent peripheral, signals with a machine-to-machine response message indicating that the contextualized training data has been detected, and transmitting, using the intelligent peripheral device, signals with a machine-to-machine response message about the detection of contextualized training data to the IIoT interconnect device over the Internet communication network, and receive signals with a machine-to-machine response message about the detection of contextualized training data on the IIoT machine-to-machine communication device, and transmitting from the IIoT interconnect device an in-system actuation command signal to increase the charge consumption of the rechargeable battery, and increase the consumption of a rechargeable battery for a machine learning device to run machine learning programs on an Internet of Things computer system powered by a rechargeable battery. 2. A method for detecting training data for machine learning of a computer system of the industrial Internet of things powered by a rechargeable battery, according to claim 1, characterized in that a machine-to-machine response message is not generated from the software adapter of the training data with a semantic annotation about the detection of contextualized training data when not detected correspondence of semantic annotation and IIoT concepts and attributes of IIoT concepts, and do not transmit signals with an inter-machine response message about the detection of contextualized training data over the Internet communication network from the data storage interconnect device to an intelligent peripheral device, and do not receive signals with an inter-machine device using an intelligent peripheral device response message about the detection of contextualized training data, and do not transmit, using the smart peripheral, signals with a machine-to-machine response message about detection of contextualized training data to the IIoT interconnect device over the Internet communication network, and do not receive signals with a machine-to-machine response message about detection of contextualized training data on IIoT interconnect device, and do not transmit from the IIoT interconnect device an internal operation command signal to increase the charge consumption level of the rechargeable battery, and do not increase the charge consumption of the rechargeable battery for the machine learning device to run the machine learning program in the Internet of Things computer system with powered by a rechargeable battery.