Evolving faults in a power grid

1 . A computer implemented method for remote condition monitoring of an electric power grid comprising intelligent electronic devices, the method comprising:

receiving different events with time information from one or more intelligent electronic devices comprised in the electric power grid, an event being generated by a function in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, wherein different functions generate events;

inputting at least the events as input data to a trained predictor, which is a machine learning based model that has been trained using at least a plurality of event patterns extracted from event history data, an event pattern comprising in occurrence order events preceding within a time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event, the trained predictor outputting predictions for occurrence times of disturbance events in the electric power grid;

mapping the predictions for occurrence times of disturbance events, obtained based on the received events that were input to the model trained for the electric power grid, to probabilities of disturbance events in relation to time;

generating, using the mapped probabilities, a report indicating as a function of time probability of any of the disturbance events to occur within at least two predetermined times of different length; and

displaying the report for condition monitoring to be used for determining whether there is any evolving fault in the electric power grid, wherein an evolving fault has not yet resulted to an interruption in the electric power grid but will do so unless a maintenance action is triggered.

2 . The computer implemented method of claim 1 , wherein the generating the report comprises:

generating two or more time windows of different lengths when determined from the same starting time; and

indicating, per a time window to be displayed, whether an occurrence time of a disturbance event is within the time window, or within the next time window, or neither.

3 . The computer implemented method of claim 1 , further comprising:

maintaining information associating events to event categories, wherein the information associates an event with one of the event categories;

determining, before inputting the event, a category of the event; and

inputting with the event also at least the category as input data to the trained predictor.

4 . The computer implemented method of claim 1 , further comprising:

maintaining information associating events to weight values, wherein the information associates an event with one weight value;

determining, before inputting the event, a weight value of the event; and

inputting with the event also at least the weight value as input data to the trained predictor.

5 . The computer implemented method of claim 1 , further comprising:

storing the events with time information at least temporarily;

inputting with the event also at least the time information as input data to the trained predictor; and

performing the inputting periodically.

6 . A computer implemented method comprising:

acquiring a plurality of event patterns extracted from event history data, or acquiring the event history data and extracting the plurality of event patterns from the history data, wherein the event history data comprises different events with time information from one or more intelligent electronic devices comprised in an electric power grid, an event being generated by a function amongst different functions in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, and an event pattern comprising in occurrence order events preceding within a preset time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event;

inputting the plurality of event patterns as a training data to a machine learning model to train the model to predict occurrence times of disturbance events in the electric power grid based on events that are non-disturbance events; and

storing the trained model to be used as a trained predictor for electric power grids for predictive condition monitoring of a power grid.

7 . The computer implemented method of claim 6 , wherein the machine learning model is a model outputting two or more time windows of different lengths, a length of a time window not exceeding the preset time span, the time windows predicting the occurrence times of disturbance events by being classified, per a time window, to one of three classes indicating, whether an occurrence time of a disturbance event is within the time window, or within the next time window, or neither.

8 . The computer implemented method of claim 6 , further comprising:

filtering, before inputting, from the plurality of event patterns, event patterns in which a disturbance event is generated by a function indicating testing or commissioning.

9 . The computer implemented method of claim 6 , wherein an event pattern comprises events generated by protective functions or control functions.

10 . The computer implemented method of claim 6 , further comprising:

acquiring information associating events to event categories, wherein the information associates an event with one of the event categories;

determining, per an event pattern, before inputting, a category of an event for events in the event pattern; and

including event categories to be part of the event pattern as the training data.

11 . The computer implemented method of claim 6 , further comprising:

acquiring information associating events to weight values, wherein the information associates an event with one weight value;

determining, before inputting, per an event pattern, a weight value of an event in the event pattern; and

including weight values to be part of the event pattern as the training data.

12 . The computer implemented method of claim 11 , further comprising:

filtering from the event pattern events having a weight value under a preset threshold.

13 . An apparatus comprising at least:

a processor;

memory coupled with the processor; and

computer program code stored in the memory and operable, when executed by the processor, to cause the apparatus at least to:

receive different events with time information from one or more intelligent electronic devices comprised in an electric power grid, an event being generated by a function in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, wherein different functions generate events;

input at least the events as input data to a trained predictor, which is a machine learning based model that has been trained using at least a plurality of event patterns extracted from event history data, an event pattern comprising in occurrence order events preceding within a time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event, the trained predictor outputting predictions for occurrence times of disturbance events in the electric power grid;

map the predictions for occurrence times of disturbance events, obtained based on the received events that were input to the model trained for the electric power grid, to probabilities of disturbance events in relation to time;

generate, using the mapped probabilities, a report indicating as a function of time probability of any of the disturbance events to occur within at least two predetermined times of different length; and

display the report via a remote condition monitoring interface to be used for determining whether there is any evolving fault in the electric power grid, wherein an evolving fault has not yet resulted to an interruption in the electric power grid but will do so unless a maintenance action is triggered.

14 . The apparatus of claim 13 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to perform the generating the report by:

generating two or more time windows of different lengths when determined from the same starting time; and

indicating, per a time window to be displayed, whether an occurrence time of a disturbance event is within the time window, or within the next time window, or neither.

15 . The apparatus of claim 13 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

maintain information associating events to event categories, wherein the information associates an event with one of the event categories;

determine, before inputting the event, a category of the event; and

input with the event also at least the category as input data to the trained predictor.

16 . The apparatus of claim 13 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

maintain information associating events to weight values, wherein the information associates an event with one weight value;

determine, before inputting the event, a weight value of the event; and

input with the event also at least the weight value as input data to the trained predictor.

17 . The apparatus of claim 13 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

store the events with time information at least temporarily;

input with the event also at least the time information as input data to the trained predictor; and

perform inputting periodically.

18 . An apparatus comprising at least:

a processor;

memory coupled with the processor; and

computer program code stored in the memory and operable, when executed by the processor, to cause the apparatus at least to:

acquire a plurality of event patterns extracted from event history data, or acquire the event history data and extract the plurality of event patterns from the history data, wherein the event history data comprises different events with time information from one or more intelligent electronic devices comprised in an electric power grid, an event being generated by a function amongst different functions in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, and an event pattern comprising in occurrence order events preceding within a preset time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event;

input the plurality of event patterns as a training data to a machine learning model to train the model to predict occurrence times of disturbance events in the electric power grid based on events that are non-disturbance events; and

store the trained model to be used as a trained predictor for electric power grids for predictive condition monitoring of a power grid.

19 . The apparatus of claim 18 , wherein the machine learning model is a model outputting two or more time windows of different lengths, a length of a time window not exceeding the preset time span, the time windows predicting the occurrence times of disturbance events by being classified, per a time window, to one of three classes indicating, whether an occurrence time of a disturbance event is within the time window, or within the next time window, or neither.

20 . The apparatus of claim 18 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

filter, before inputting, from the plurality of event patterns, event patterns in which a disturbance event is generated by a function indicating testing or commissioning.

21 . The apparatus of claim 18 , wherein an event pattern comprises events generated by protective functions or control functions.

22 . The apparatus of claim 18 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

acquire information associating events to event categories, wherein the information associates an event with one of the event categories;

determine, per an event pattern, before inputting, a category of an event for events in the event pattern; and

include event categories to be part of the event pattern as the training data.

23 . The apparatus of claim 18 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to:

acquire information associating events to weight values, wherein the information associates an event with one weight value;

determine, before inputting, per an event pattern, a weight value of an event in the event pattern; and

include weight values to be part of the event pattern as the training data.

24 . The apparatus of claim 23 , wherein the at least one memory and computer program code are configured to, with the at least one processor, cause the apparatus further at least to filter from the event pattern events having a weight value under a preset threshold.

25 . A non-transitory computer readable medium comprising program instructions for causing a computing apparatus to perform for remote condition monitoring of an electric power grid comprising intelligent electronic devices at least one of a first functionality or a second functionality,

wherein the first functionality comprises at least:

when receiving different events with time information from one or more intelligent electronic devices comprised in the electric power grid, an event being generated by a function in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, wherein different functions generate events, inputting at least the events as input data to a trained predictor, which is a machine learning based model that has been trained using at least a plurality of event patterns extracted from event history data, an event pattern comprising in occurrence order events preceding within a time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event, the trained predictor outputting predictions for occurrence times of disturbance events in the electric power grid;

mapping the predictions for occurrence times of disturbance events, obtained based on the received events that were input to the model trained for the electric power grid, to probabilities of disturbance events elation to time;

generating, using the mapped probabilities, a report indicating as a function of time probability of any of the disturbance events to occur within at least two predetermined times of different length; and

displaying the report for condition monitoring to be used for determining whether there is any evolving fault in the electric power grid, wherein an evolving fault has not yet resulted to an interruption in the electric power grid but will do so unless a maintenance action is triggered,

wherein the second functionality comprises at least:

acquiring a plurality of event patterns extracted from event history data, or acquiring the event history data and extracting the plurality of event patterns from the history data, wherein the event history data comprises different events with time information from one or more intelligent electronic devices, an event being generated by a function amongst different functions in an intelligent electronic device as a reaction to one or more values measured from the electric power grid, the event indicating at least the function and its output, the event not comprising the one or more values measured from the electric power grid, and an event pattern comprising in occurrence order events preceding within a preset time span a disturbance event, which is an event resulting to an interruption in power supply in the electric power grid, the disturbance event not being caused by a failure of an intelligent electronic device whose function generated the disturbance event;

inputting the plurality of event patterns as a training data to a machine learning model to train the model to predict occurrence times of disturbance events in the electric power grid based on events that are non-disturbance events; and

storing the trained model to be used as a trained predictor for electric power grids for predictive condition monitoring of an electric power grid.