A method for adapting a prediction model for predicting a state of a component of a switchgear, a computer program product, a computer-readable storage medium, as well as electronic computing device

Abstract

The invention relates to a method for adapting a prediction model (16) for predicting a state (18) of a component (12) of a switchgear (10) by an electronic computing device (14), comprising the steps of providing an initial prediction model (16) for predicting the state (18) of the component (12) by the electronic computing device (14); predicting the expected state (18) of the component (12) by the electronic computing device (14) by using the initial prediction model (16); receiving an actual state (18) of the component (12) by the electronic computing device (14); comparing the expected state (18) with actual state (18) by the electronic computing device (14); and adapting at least one parameter (20) of the prediction model (16) depending on the comparison by the electronic computing device (14). Furthermore, the present invention relates to a computer program product, a computer-readable storage medium, as well as to an electronic computing device (14).

Description

[0001] Description

[0002] A method for adapting a prediction model for predicting a state of a component of a switchgear, a computer program product, a computer-readable storage medium, as well as an electronic computing device

[0003] The present invention relates to a method for adapting a prediction model for predicting a state of a component of a switchgear by an electronic computing device according to the pending claim 1. Furthermore, the present invention relates to a corresponding computer program product, a corresponding computer-readable storage medium, as well as to a corresponding electronic computing device.

[0004] Asset management of in particular electric grid assets is shifting from time-based maintenance to condition-based maintenance and risk-centered maintenance. To perform efficient conditionbased maintenance and risk-centered maintenance, tools like a health index and a risk matrix are required, giving the users a necessary information, which assets need attention and what kind of measures are the best.

[0005] Such highly abstracted values are only meaningful when sufficient high-quality information is available. This is difficult for the assets of the medium and low voltage distribution level with a lack of sensor data, the lack of very frequent maintenance and the lack of detailed offline analysis, which is too expensive for distribution grid assets, at least compared to assets of the transmission level.

[0006] The rate of change in today’s electricity grid is higher than ever. A health index must include and reflect this change. Risk-centered maintenance must focus on the single asset, using too many statistics based on a class of assets can be dangerous as a simple interpolation gets complicated in the future.

[0007] A current health index is based on statistics and experience. There is research ongoing to further include sensor data, for example with the help of artificial intelligence. Cause and effect relationships are trained based on historical data or defined via expert knowledge.

[0008] It is an object of the present invention to provide a method, a computer program product, a corresponding computer-readable storage medium, as well as a corresponding electronic computing device, by which a prediction of a state of a component can be performed in an improved manner.

[0009] This object is solved by a method, a corresponding computer program product, a corresponding computer-readable storage medium, as well as a corresponding electronic computing device according to the independent claims. Advantageous embodiments are presented in the dependent claims.

[0010] One aspect of the invention relates to a method for adapting a prediction model for predicting a state of a component of a switchgear by an electronic computing device. An initial prediction model for predicting the state of the component is provided by the electronic computing device. The expected state of the component is predicted by the electronic computing device by using the initial prediction model. An actual state of the component is received by the electronic computing device. The expected state is compared with actual state by the electronic computing device, and at least one parameter of the prediction model is adapted depending on the comparison by the electronic computing device.

[0011] Therefore, an improved way for predicting the state of a component, in particular a switching element of a switchgear, can be provided.

[0012] In particular, the state of the component, which can be, for example, a so called health index, uses the digital twin model and is in particular recalibrated after, for example, a service. The idea uses certain basic properties of digital twins, a single source of truth, the representation of an asset’s true state and modelling the behavior of real assets by physical laws.

[0013] In particular, a processing and weighting algorithm, which may be also called an analytic engine, and which is related to the prediction model, uses several input parameters to display a health index and then composes a risk matrix by combining all assets. Depending on the health index, recommendations can be given, for example, determination of the number of switching operations can lead to a recommendation to maintain the switch. Several data sources can be used as input parameter for the health index. This may be an initial data set at the beginning, for example, in form of a questionnaire, service protocols and sensor information. It is important to acknowledge that the input contains the ground truth, so true process values from the field. The algorithm can only generate an estimated guess. It is therefore important that the algorithm contains a recalibration, for example, with the help of the service protocols. If the service (ground truth) is telling that the asset is still fine then your algorithm can no longer assign a bad health to the asset.

[0014] The problem to be solved is in particular, that a service may only happening in certain intervals. Sensor data may exist, but as mentioned, sensors are not on the distribution level, and it is not as common as in the transmission level. There may also be only a simple sensor installed, for example, the switchgear on / off switch, and no temperature sensors. The temperature information may also not give sufficient information as it is not placed on the most critical position. A digital model may now help to solve these issues by providing more accurate data between service intervals, enabling a more accurate health index. It is also important to notice that the digital model, at least as an option, is not only providing information to the analytical engine, but also gets feedback, if the model is wrong. If a digital model is providing data to the analytical engine, causing the analytical engine to generate a health index not matching the service protocols, then the digital model is updated. As the digital model is extrapolating in a physics-based environment, it can either be a physics-based artificial intelligence model or a physics-reduced model. It may be finally noted that the digital twin is not the digital model.

[0015] In particular, the conventional approach according to the state of the art has the danger that it can be wrong for a single specific asset, causing a wrong health index, especially in between. The model without any feedback according to the state of the art continuously helps to update the health index. But over time, like with an integration, the model may be wrong. Only with the feedback loop, the model can compensate its own incompleteness.

[0016] Therefore, the present invention is an extrapolation capability and the resulting improvement of health index performance. Conventional approaches according to the state of the art do not look at the single asset with digital twin glasses. There is only a single source of truth, so all data must match, and there are physical restraints. This also improves the effort to introduce such a model as the problem dimensionality, compared to pure artificial intelligence approaches, is severely reduced. There is a leverage of existing information via the introduced digital model, reducing the need of additional sensors.

[0017] According to an embodiment, depending on a deviation threshold for the comparison the at least one parameter is adapted. In particular, a deviation value between the expected state and the actual state can be determined. If the deviation value is higher than a threshold, the adaptation will be performed. If the deviation value is lower than the threshold, no adaption will be performed. In another embodiment, the actual state of the component is measured by a measuring device for the component. For example, a sensor element can be provided as the measuring device. The measurements of the actual state can be arranged in time intervals.

[0018] In another embodiment, in predefined time intervals the measuring is performed. Therefore, no permanent measuring of the actual state is needed. Just in time intervals the measuring is performed. Therefore, an easy way for predicting the current state of the component is provided.

[0019] In another embodiment, the actual state is measured manually, and the manual actual state is received from an input device of the electronic computing device. For example, a service personal can measure manually the current state. Then, the service personal can input the information about the current state via an input device, and therefore the electronic computing device gets the information about the current state. This is an easy way for adapting the prediction model.

[0020] In another embodiment, a current state of the component is predicted by the adapted prediction model. In particular, for example a health index can be provided for the component. Therefore, a service personal can have a precise look about the current state of the component or a plurality of components in the switchgear.

[0021] In another embodiment, depending on the predicted current state a service message is generated by the electronic computing device. For example, the service message can be provided as the health index. Therefore, a service personal may get the service message, for example, that service has to be provided for the component. Therefore, a failure of the component can be prevented.

[0022] According to another embodiment, the service message is generated such, that a service suggestion is provided. For example, the service suggestion may provide a complete exchange of the component or just the exchange of parts of the component. Therefore, an improved operation of the component can be provided.

[0023] In another embodiment, an initial questionnaire is used for providing the initial prediction model. For example, by installation of the switchgear and the component, the initial questionnaire is provided for the service personal. Depending on the initial questionnaire, the initial prediction model can be provided. Therefore, a detailed prediction model for an initial state is provided. In another embodiment, a digital twin is used by the prediction model. The digital twin can be, for example, an artificial intelligence model, which copies / models the installed component.

[0024] In another embodiment, the actual state is predicted depending on an estimated temperature of the component and / or an electric current at the component and / or an electric voltage at the component. Therefore, actual parameter values of the component can be taken into consideration in order to predict the static, non-switching state of the component.

[0025] In another embodiment, additionally to the current state a potential amount of switching processes of the component is predicted. In particular, a potentially amount of potential switching processes in the future of the component can be predicted. Therefore, the service personal gets an information about how long the component may not get in a failure or error state.

[0026] In particular, the present invention is a computer-implemented method. Therefore, another aspect of the invention relates to a computer program product comprising program code means for performing a method according to the preceding aspect.

[0027] Furthermore, the present invention relates to a computer-readable storage medium comprising at least the computer program product according to the preceding aspect.

[0028] Another aspect of the invention relates to an electronic computing device for adapting a prediction model for predicting a state of a component of a switchgear, wherein the electronic computing device is configured for performing a method according to the preceding aspect. In particular, the method is performed by the electronic computing device.

[0029] A still further aspect of the invention relates to a component comprising at least the electronic computing device and / or to a switchgear comprising at least the component and / or the electronic computing device.

[0030] Advantageous embodiments of the method are to be regarded as an embodiment of the computer program product, the computer-readable storage medium, as well as the electronic computing device, the component, as well as the switchgear. The electronic computing device, the component, as well as the switchgear comprise means for performing the method according to the preceding aspect. An artificial neural network / artificial intelligence can be understood as a software code or a compilation of several software code components, wherein the software code may comprise several software modules for different functions, for example one or more encoder modules and one or more decoder modules.

[0031] An artificial neural network can be understood as a non-linear model or algorithm that maps an input to an output, wherein the input is given by an input feature vector or an input sequence and the output may be an output category for a classification task or a predicted sequence.

[0032] For example, the artificial neural network may be provided in a computer-readable way, for example, stored on a storage medium of the vehicle, in particular of the at least one computing unit.

[0033] The neural network comprises several modules including the encoder module and the at least one decoder module. These modules may be understood as software modules or respective parts of the neural network. A software module may be understood as software code functionally connected and combined to a unit. A software module may comprise or implement several processing steps and / or data structures.

[0034] The modules may, in particular, represent neural networks or sub-networks themselves. If not stated otherwise, a module of the neural network may be understood as a trainable and, in particular, trained module of the neural network. For example, the neural network and thus all of its trainable modules may be trained in an end-to-end fashion before the method is carried out. However, in other implementations, different modules may be trained or pre-trained individually. In other words, the method according to the invention corresponds to a deployment phase of the neural network.

[0035] A computing unit / electronic computing device may in particular be understood as a data processing device, which comprises processing circuitry. The computing unit can therefore in particular process data to perform computing operations. This may also include operations to perform indexed accesses to a data structure, for example a look-up table, LUT.

[0036] In particular, the computing unit may include one or more computers, one or more microcontrollers, and / or one or more integrated circuits, for example, one or more applicationspecific integrated circuits, ASIC, one or more field-programmable gate arrays, FPGA, and / or one or more systems on a chip, SoC. The computing unit may also include one or more processors, for example one or more microprocessors, one or more central processing units, CPU, one or more graphics processing units, GPU, and / or one or more signal processors, in particular one or more digital signal processors, DSP. The computing unit may also include a physical or a virtual cluster of computers or other of said units.

[0037] In various embodiments, the computing unit includes one or more hardware and / or software interfaces and / or one or more memory units.

[0038] A memory unit may be implemented as a volatile data memory, for example a dynamic random access memory, DRAM, or a static random access memory, SRAM, or as a non-volatile data memory, for example a read-only memory, ROM, a programmable read-only memory, PROM, an erasable programmable read-only memory, EPROM, an electrically erasable programmable read-only memory, EEPROM, a flash memory or flash EEPROM, a ferroelectric random access memory, FRAM, a magnetoresistive random access memory, MRAM, or a phase-change random access memory, PCRAM.

[0039] For use cases or use situations which may arise in a method according to the invention and which are not explicitly described herein, it may be provided that, in accordance with the method, an error message and / or a prompt for user feedback is output and / or a default setting and / or a predetermined initial state is set.

[0040] Independent of the grammatical term usage, individuals with male, female or other gender identities are included within the term.

[0041] Further features and feature combinations of the invention are obtained from the figures and their description as well as the claims. In particular, further implementations of the invention may not necessarily contain all features of one of the claims. Further implementations of the invention may comprise features or combinations of features, which are not recited in the claims.

[0042] Therefore, the figures show in:

[0043] Fig. 1 a schematic block diagram according to an embodiment of a switchgear comprising an embodiment of a component comprising an embodiment of an electronic computing device; Fig. 2 a schematic block diagram according to an embodiment of an electronic computing device; and

[0044] Fig. 3 a schematic flow chart according to an embodiment of the method.

[0045] In the following, the invention will be explained in detail with reference to specific exemplary implementations and respective schematic drawings. In the drawings, identical or functionally identical elements may be denoted by the same reference signs. The description of identical or functionally identical elements is not necessarily repeated with respect to different figures.

[0046] Fig. 1 shows a schematic block diagram according to an embodiment of a switchgear 10. The switchgear 10 comprises at least one component 12. In the shown embodiment three components 12 are shown. The component 12 maybe for example a switching element. Furthermore, the switchgear 10 comprises an electronic computing device 14 comprising a prediction model 16.

[0047] In particular, Fig. 1 shows the switchgear 10 for performing a method for adapting the prediction model 16. An initial prediction model 16 for predicting the state 8 of the component 12 is provided by the electronic computing device 14. The expected state 18 of the component 12 is predicted by the electronic computing device 14 by using the initial prediction model 16. An actual state of the component 12 is received by the electronic computing device 14. The expected state is compared with the actual state by the electronic computing device 14. At least one parameter 20 of the prediction model 16 is adapted depending on the comparison by the electronic computing device 14.

[0048] In particular, depending on a deviation threshold for the comparison at least the one parameter 20 is adapted.

[0049] Furthermore, the actual state 18 is measured by a measuring device 22 for the component 12. Furthermore, in predefined time intervals the measuring is performed.

[0050] In another embodiment, the actual state 18 is measured manually, and the manual actual state 18 is received from an input device 24 of the electronic computing device 14. Furthermore, a current state 18 of the component 12 is predicted by the adapted prediction model 16. Furthermore, depending on the predicted current state 18 a service message 36 (Fig. 2) is generated by the electronic computing device 14. In particular, the service message 36 may be generated such that a service suggestion is provided. Furthermore, the current state 18 is predicted depending on an estimated temperature of the component 12 and / or electric current at the component 12 and / or an electric voltage at the component 12.

[0051] Furthermore, additionally to the current state 18 a potential amount of switching processes of the component 12 is predicted.

[0052] Fig. 2 shows a schematic block diagram according to an embodiment of the electronic computing device 14. In particular, Fig. 2 shows that the electronic computing device 14 may comprise a so-called analytical engine 26. The analytical engine 26 may be configured for displaying the current state 18 and for generating the service message.

[0053] As an input 38 of the analytical engine 26, a service protocol 28 may be used in order to recalibrate a digital model 30 of the component 12. Furthermore, an initial questionnaire 32 may be used once. Furthermore, sensors / measuring devices 22 may be used for the analytical engine 26 as well as for the digital twin 30.

[0054] Fig. 3 shows a schematic flow chart according to an embodiment of the method. Fig. 3 shows this approach on a specific topic, dielectric health and thermal health. It is an incomplete example as the health index of the switchgear 10 does not only consist of its dielectric health and the thermal health. The target here is to show the possible improvement of the health index, not to provide a complete one with these examples.

[0055] First, the method is explained with the dielectric health. The status quo can be described in the following way: The switchgear 10 has no partial discharge sensor installed. The insulators may be checked by service personal using portable partial discharge sensors, but infrequently. Statistics is not helping to improve the health index as the grid experiences and increases an instability due to the new inverter-based consumers / producers.

[0056] In a first step S1 , a voltage transformer gives certain information regarding the primary voltage quality. For example, the voltage transformer is not located in or on the target switchgear 10. A state estimation of the voltage is, for example, sufficient as the achieved accuracy (+ / - 5%) is fully acceptable. Necessary information like earth fault is provided. In a second step S2, the dielectric twin model of the switchgear 10 is now simulating the voltage behavior of the asset, checking for possible dielectric problems. It then calculates a probability of default PD impacting the health index calculation. There are several possibilities to calculate a PD probability. As an example, a very simple one is selected. With the primary voltage II below a certain inception voltage, no PD may happen, in particular a probability b is zero. Above a certain inception voltage, a PD may start an increase in strength, normally as (step) function of the primary voltage. The knowledge of the inception voltage can be an aging parameter. Calculating the inception voltage V can be a complicated task. A very simple aging model is presented in the equation below having parameters a, d and Fo. Vo is the inception voltage for a completely new switchgear 10, and zcis a coefficient defining the maximum relative decrease of the partial discharge inception voltage due to aging, which can have values given by the relation 0 < K < 1. Without better knowledge the initial value of this parameter is set to 1. It may be noted that this model is just for reference purposes. The parameter d, for example, can be itself a function of the PD probability b:

[0057] The parameter b can be related to the transferred charge, expressed as:

[0058] C is a factor describing the involved discharge capacitances. M is a primary voltage dependent multiplier, in particular the mentioned step function 22. Both, C and M must be precalculated which can be done by a state-of-the-art model.

[0059] In a third step S3, unresolved PD can lead to a total breakdown of the switchgear 10. The weighting function to include PD into the health index must therefore be non-linear. A possible solution there can be an exponential term with the weighting factor a and the output b of the dielectric twin. There are several PD faults possible, therefore an index is used. In a fourth step S4, the health index calculation is performed. In particular, a very high PD probability leads to a high health index value for the specific switchgear 10 and recommends a service of the switchgear 10 using a portable PD device. The service personal may execute this task. They check the switchgear 10 and detect that there is some unexpected condensation, but no PD.

[0060] In a fifth step S5, the health index is provided. This information may be used by the health index. The health index has now two contradiction pieces in an information: No PD and very high PD risk. As the switchgear 10 cannot be faulty and healthy, one piece of information must be assumed as ground truth. This is normally the field measurement, in particular ignoring faulty measurement equipment or furthermore. There is no PD yet, but some condensation, indicating a slower aging, but not too slow. Both pieces are fed back to the PD model which adjusts the aging d. This results in an updated health index. The advantage of the physically constraint and simple model is that, even with only one data point provided, the model can be updated. In this case, only the parameter d has to be updated.

[0061] The approach with the thermal health is very similar. The switchgear 10 has no temperature sensor installed, but an initial service e.g., with the help of an infrared camera, has been performed to ensure that the switchgear 10 is probably problem free.

[0062] A current transformer gives certain information regarding the primary current. It can be imagined that the current is not located on the target switchgear 10. A state estimation of the current is for this example sufficient as the achievable accuracy (+ / -5% is fully acceptable).

[0063] The thermal twin model of the switchgear 10 is now simulating the thermal behavior of the asset. The knowledge of the contact resistance can be an aging parameter. Fully calculating the thermal behavior of the switchgear 10 can be a complicated task. Luckily, for local behavior an approximation with a PT1 (first order time delay element) time constant is sufficient. The necessary parameters for this PT 1 element can be precalculated. Mass and Resistance of a piece of copper will not change over the switchgear’s lifetime. The additional loss PLOSS caused the contact can be described acc. to literature with With the base contact resistance RCOntact at reference temperature To, the temperature coefficient a and the contact temperature T. It can again be assumed that the contact is getting worse over time with the aging parameter f, starting on a base level Ro.

[0064] There are several parameters possible for further evaluation. One may directly use the contact resistance. There it makes sense to normalize the value.

[0065] Another possibility is to utilize PLoSs as this includes already the current information with the argument, no current no problem. As the loss may change with the current, a maximum value hold is necessary.

[0066] Unresolved thermal problems can lead to a total breakdown of the switchgear 10. The weighting function to include thermal problems into the Health Index must therefore be non-linear. A possible solution there can be an exponential term with the weighting factor h and the output g of the thermal twin. If P|OSSis utilized, the non-linearity is already part of the calculation, and a linear term may be sufficient. There are several connections in a switchgear 10, therefore an index is used.

[0067] A very high thermal fault probability leads to a high (==bad) health index value for the specific switchgear 10 and recommends a service of the switchgear 10, using a portable infrared device or a visual inspection. The service personal will execute this task. They check the switchgear 10 and detect that there is a higher temperature than even predicted, with signs of scorched insulation.

[0068] This information is then used in the health index. The health index has now two contradicting pieces of information: High thermal problems and very high thermal problems. As the switchgear 10 cannot be slightly faulty and near a breakdown, one piece of information must be assumed as ground truth. This is normally the field measurement (ignoring faulty measurement equipment etc.). The model can be updated to include an even faster contact aging. Since the cable connection is already severely damaged, the more proper conclusion is to repair the bad connection with a maintenance action. The contact resistance is updated, and a faster aging factor is assumed for this switchgear 10 as the last one was not fast enough. The advantage of the physically constrained and simple models is that, even with only one datapoint provided, the model can be updated. In this case, only the parameter f must be updated.

[0069] Both examples assumed that no sensor is installed. This is one possibility. The service personal does not have to go to the switchgear 10 if there are sensors installed. It is even be advantageous as then feedback from the switchgear 10 is more common.

[0070] List of ReferenceList

[0071] 10 switchgear

[0072] 12 component

[0073] 14 electronic computing device

[0074] 16 prediction model

[0075] 18 state

[0076] 20 parameter

[0077] 22 measuring device

[0078] 24 input device

[0079] 26 analytical engine

[0080] 28 service protocol

[0081] 30 digital model

[0082] 32 initial questionnaire

[0083] 34 service

[0084] 36 service message

[0085] 38 input

[0086] S1 to S5 steps of the method

Claims

Patent claims1. A method for adapting a prediction model (16) for predicting a state (18) of a component (12) of a switchgear (10) by an electronic computing device (14), comprising the steps of:- providing an initial prediction model (16) for predicting the state (18) of the component (12) by the electronic computing device (14);- predicting the expected state (18) of the component (12) by the electronic computing device (14) by using the initial prediction model (16);- receiving an actual state (18) of the component (12) by the electronic computing device (14);- comparing the expected state (18) with actual state (18) by the electronic computing device (14); and- adapting at least one parameter (20) of the prediction model (16) depending on the comparison by the electronic computing device (14).

2. The method according to claim 1 , wherein depending on a deviation threshold for the comparison the at least one parameter (20) is adapted.

3. The method according to claim 1 or 2, wherein the actual state (18) is measured by a measuring device (20) for the respective component (12).

4. The method according to claim 3, wherein in predefined time intervals the measuring is performed.

5. The method according to any one of claims 1 to 4, wherein the actual state (18) is measured manually and the manual actual state (18) is received from an input device (24) of the electronic computing device (14).

6. The method according to any one of claims 1 to 5, wherein a current state (18) of the component (12) is predicted by the adapted prediction model (16).

7. The method according to claim 6, wherein depending on the predicted current state (18) a service message (36) is generated by the electronic computing device (14).

8. The method according to claim 7, whereinthe service message (36) is generated such, that a service suggestion is provided.

9. The method according to any one of claims 1 to 8, wherein an initial questionnaire (32) is used for providing the initial prediction model (16).

10. The method according to any one of claims 1 to 9, wherein a digital twin is used by the prediction model (16).

11. The method according to any one of claims 1 to 10, wherein the current state (18) is predicted depending on an estimated temperature of the component (12) and / or an electric current at the component (12) and / or an electric voltage at the component (12).

12. The method according to any one of claims 1 to 11 , wherein additionally to the current state a potential amount of switching processes of the component is predicted.

13. A computer program product comprising program code means for performing a method according to any one of claims 1 to 12.

14. A computer-readable storage medium comprising at least the computer program product according to claim 13.

15. An electronic computing device (14) for adapting a prediction model (16) for predicting a state (18) of a component (12) of a switchgear (10), wherein the electronic computing device (14) is configured for performing a method according to any one of claims 1 to 12.

Images