CN117572159A

CN117572159A - Power failure detection method and system based on big data analysis

Info

Publication number: CN117572159A
Application number: CN202410064161.0A
Authority: CN
Inventors: 莫曦; 范丹; 廖晋; 高阳; 刘宏伟
Original assignee: Chengdu Chinaelite Technology Co ltd
Current assignee: Chengdu Chinaelite Technology Co ltd
Priority date: 2024-01-17
Filing date: 2024-01-17
Publication date: 2024-02-20
Anticipated expiration: 2044-01-17
Also published as: CN117572159B

Abstract

The embodiment of the application provides a power failure detection method and a system based on big data analysis, wherein through acquiring a sample power system operation data cluster sequence, data comprising a trigger reference power failure state label and system state monitoring data comprising other trigger power failure state labels are respectively loaded into a power failure detection network for knowledge learning, and a first failure detection error parameter and a second failure detection error parameter are determined, so that the power failure detection network is optimized and adjusted, and an optimized and adjusted target power failure detection network is generated. And finally, carrying out power failure detection on the input target system state monitoring data based on the optimized network. Therefore, the accuracy and the efficiency of power failure detection are improved, the running state of the power system is monitored in real time, and the power failure is found and processed in time, so that the stable and safe running of the power system is ensured.

Description

Power failure detection method and system based on big data analysis

Technical Field

The application relates to the technical field of smart grids, in particular to a power failure detection method and system based on big data analysis.

Background

The power system is an infrastructure for modern society operation, and its stability and safety have a crucial impact on socioeconomic performance. However, due to the complexity and large scale of the power system, and the variability of its operating environment, various faults within the power system may occur, such as overloads, short circuits, equipment damage, etc. These faults may cause interruption of the power supply and even cause serious socioeconomic problems.

The traditional power failure detection method mainly relies on manual inspection or fixed threshold value setting for alarming, and the mode can discover and process power failures to a certain extent. However, the manual inspection is low in efficiency and long in time consumption, is easily affected by human factors, and causes larger error; the fixed threshold alarm mode has the problems of difficult threshold selection, low sensitivity and the like. In addition, these conventional methods often cannot realize real-time monitoring of the state of the power system, and cannot predict possible faults in the future, so that it is difficult to meet the requirements of the current power system operation management.

Disclosure of Invention

In order to at least overcome the above-mentioned shortcomings in the prior art, an object of the present application is to provide a method and a system for detecting a power failure based on big data analysis.

In a first aspect, the present application provides a power failure detection method based on big data analysis, applied to a power failure detection system based on big data analysis, the method comprising:

acquiring a sampling example power system operation data cluster sequence; the sample power system operation data cluster sequence comprises a plurality of first sample power system operation data clusters and a plurality of second sample power system operation data clusters, wherein each first sample power system operation data cluster comprises system state monitoring data of a reference power system unit triggering a reference power fault state label, and each second sample power system operation data cluster comprises other system state monitoring data except the reference power system unit triggering the reference power fault state label;

loading the first sample power system operation data cluster into a power failure detection network to perform knowledge learning, and determining a first failure detection error parameter in the knowledge learning process;

loading the second sample power system operation data cluster into the power failure detection network, classifying the power failure state labels triggered by the reference power system units in the second sample power system operation data cluster according to the power failure detection network, and generating a first failure classification diagram; the first fault classification map includes a first confidence that the power fault detection network estimates that the reference power system unit triggered each of the reference power fault status tags;

Determining a second fault detection error parameter in the knowledge learning process based on the first confidence coefficient and the threshold confidence coefficient;

and optimizing and adjusting the power failure detection network based on the first failure detection error parameter and the second failure detection error parameter, generating an optimized and adjusted target power failure detection network, and performing power failure detection on the input target system state monitoring data based on the target power failure detection network.

In a possible implementation manner of the first aspect, the sequence of sample power system operation data clusters further includes fault labeling data corresponding to the first sample power system operation data cluster, where the fault labeling data reflects an actual fault state label of a power fault state label triggered by the reference power system unit in the first sample power system operation data cluster;

the loading the first sample power system operation data cluster into a power failure detection network for knowledge learning, and determining a first failure detection error parameter in the knowledge learning process comprises the following steps:

loading the first sample power system operation data cluster into the power failure detection network, classifying the power failure state labels triggered by the reference power system units in the first sample power system operation data cluster according to the power failure detection network, and generating a second failure classification diagram; the second fault classification map includes a second confidence that the power fault detection network estimates that the reference power system unit triggered each of the reference power fault status tags;

And determining a first fault detection error parameter in the knowledge learning process based on the fault annotation data and the second fault classification map.

In a possible implementation manner of the first aspect, the power failure detection network includes a first feature extraction unit and a second feature extraction unit; the loading the first sample power system operation data cluster into the power failure detection network, classifying the power failure state label triggered by the reference power system unit in the first sample power system operation data cluster according to the power failure detection network, and generating a second failure classification diagram, including:

loading the first sample power system operation data cluster to the first feature extraction unit, and performing feature extraction on the first sample power system operation data cluster according to the first feature extraction unit to generate a first power system operation path vector;

loading the first power system running path vector to the second feature extraction unit, and fusing the first power system running path vector by using a tag centroid influence parameter according to the second feature extraction unit to generate a first target fault tag feature, wherein the tag centroid influence parameter comprises a plurality of tag centroid influence factors, each tag centroid influence factor corresponds to one reference power fault state tag, and the dimension of the first target fault tag feature is the same as the number of the reference power fault state tags;

And carrying out smooth extremum conversion on the first target fault label characteristic according to a smooth extremum conversion mechanism, generating a second target fault label characteristic, outputting the second target fault label characteristic as a second fault classification chart, wherein characteristic unit parameters in the second target fault label characteristic represent second confidence that the power fault detection network estimates the reference power system unit triggering corresponding to the reference power fault state label.

In a possible implementation manner of the first aspect, the determining, based on the fault labeling data and the second fault classification map, a first fault detection error parameter in a knowledge learning process includes:

based on an actual fault state label of the power fault state label in the fault labeling data, determining a corresponding first influence characteristic from the label centroid influence factors, and outputting label centroid influence factors except the first influence characteristic in the label centroid influence factors as second influence characteristics;

determining a first similarity parameter value between the first power system operational path vector and the first influencing feature, and determining a second similarity parameter value between the first power system operational path vector and each of the second influencing features;

Calculating the difference value between the first similarity parameter value and a preset parameter value to generate a first target parameter value;

determining a first fault detection error parameter in a knowledge learning process based on the first target parameter value and each of the second similarity parameter values;

the first target parameter value and the first fault detection error parameter are in a reverse association relationship, and the second similarity parameter value and the first fault detection error parameter are in a forward association relationship.

In a possible implementation manner of the first aspect, the performing, according to a smooth extremum conversion mechanism, the smooth extremum conversion on the first target fault tag feature, to generate a second target fault tag feature, includes:

acquiring a preset distribution adjustment factor; the size of the distribution adjustment factor is greater than 1;

performing distribution adjustment on the first target fault tag characteristic according to the distribution adjustment factor to generate a third target fault tag characteristic;

and carrying out smooth extremum conversion on the third target fault label characteristic through a smooth extremum conversion mechanism to generate a second target fault label characteristic.

In a possible implementation manner of the first aspect, the power failure detection network includes a first feature extraction unit and a second feature extraction unit; the loading the second sample power system operation data cluster into the power failure detection network, classifying the power failure state label triggered by the reference power system unit in the second sample power system operation data cluster according to the power failure detection network, and generating a first failure classification diagram, including:

Loading the second sample power system operation data cluster to the first feature extraction unit, and performing feature extraction on the second sample power system operation data cluster according to the first feature extraction unit to generate a second power system operation path vector;

loading the second power system running path vector to the second feature extraction unit, and fusing the second power system running path vector by using a tag centroid influence parameter according to the second feature extraction unit to generate a fourth target fault tag feature, wherein the tag centroid influence parameter comprises a plurality of tag centroid influence factors, each tag centroid influence factor corresponds to one reference power fault state tag, and the dimension of the fourth target fault tag feature is the same as the number of the reference power fault state tags;

performing smooth extremum conversion on the fourth target fault tag feature according to a smooth extremum conversion mechanism, generating a fifth target fault tag feature, outputting the fifth target fault tag feature as a first fault classification chart, wherein feature unit parameters in the fifth target fault tag feature represent first confidence that the power fault detection network estimates the reference power system unit triggering corresponding to the reference power fault state tag;

The determining the second fault detection error parameter in the knowledge learning process includes:

determining a third similarity parameter value between the second power system running path vector and each of the tag centroid influencing factors;

determining a preset parameter value based on the threshold confidence;

calculating the maximum difference value between the third similarity parameter value and the preset parameter value to generate a third target parameter value;

determining a second fault detection error parameter in the knowledge learning process based on the third target parameter value;

and the third target parameter value and the second fault detection error parameter are in a forward association relationship.

In a possible implementation manner of the first aspect, the determining, based on the first confidence level and the threshold confidence level, a second fault detection error parameter in a knowledge learning process includes:

calculating the difference value between the first confidence coefficient corresponding to each reference power fault state label and the threshold confidence coefficient, and generating a second target parameter value corresponding to each reference power fault state label;

comparing the magnitudes of the second target parameter values if there are more than 0, determining the second fault detection error parameter based on the largest second target parameter value; or if the second target parameter values are not greater than 0, determining that the magnitude of the second fault detection error parameter is 0.

In a possible implementation manner of the first aspect, the optimizing adjustment is performed on the power failure detection network based on the first failure detection error parameter and the second failure detection error parameter, and the generating the optimized and adjusted target power failure detection network includes:

determining the first cluster number of the first sample power system operation data clusters and the second cluster number of the second sample power system operation data clusters in the sample power system operation data cluster sequence;

determining a first importance coefficient corresponding to the first fault detection error parameter based on the first cluster number, and determining a second importance coefficient corresponding to the second fault detection error parameter based on the second cluster number;

performing weight fusion on the first fault detection error parameter and the second fault detection error parameter based on the first importance coefficient and the second importance coefficient to generate a global fault detection error parameter;

and carrying out optimization adjustment on the power failure detection network based on the global failure detection error parameter to generate an optimized and adjusted power failure detection network.

In a possible implementation manner of the first aspect, the step of performing power failure detection on the input target system state monitoring data based on the target power failure detection network includes:

Acquiring basic system state monitoring data, performing recursive feature elimination on each basic system state monitoring data, generating initial traversal selection features corresponding to each basic system state monitoring data, adding the acquired basic system state monitoring data into a data sequence to be processed, detecting the number of sequence members in the current data sequence to be processed, and if the number of the sequence members is not less than the set number, extracting a first number of basic system state monitoring data from the data sequence to be processed according to a preset number interval parameter, and generating a first system state monitoring data sequence;

outputting initial traversal selection characteristics corresponding to the basic system state monitoring data which are in a dominant position in the current data sequence to be processed as recursive characteristic elimination nodes, and dividing each basic system state monitoring data in the first system state monitoring data sequence based on the recursive characteristic elimination nodes to generate target system state monitoring data;

loading the target system state monitoring data into the target power failure detection network, classifying a power failure state label triggered by a target power system in the target system state monitoring data according to the target power failure detection network, and generating a third failure classification diagram; the third fault classification map includes a third confidence that the target power fault detection network estimates that the target power system triggered each of the reference power fault status tags;

Comparing the third confidence coefficient corresponding to each reference power fault state label with a threshold confidence coefficient, if the third confidence coefficient larger than the threshold confidence coefficient exists, comparing the third confidence coefficient, and determining a target power fault state label based on the reference power fault state label corresponding to the largest third confidence coefficient; or if the third confidence coefficient is not greater than the threshold confidence coefficient, determining that the target power failure state label is an unknown power failure state label.

In a second aspect, embodiments of the present application further provide a power failure detection system based on big data analysis, where the power failure detection system based on big data analysis includes a processor and a machine-readable storage medium having a computer program stored therein, where the computer program is loaded and executed in conjunction with the processor to implement the power failure detection method based on big data analysis of the first aspect above.

By adopting the technical scheme in any aspect, through acquiring the sample power system operation data cluster sequence, the data comprising the trigger reference power failure state label and the system state monitoring data comprising the trigger other power failure state labels are respectively loaded into the power failure detection network for knowledge learning, and the first failure detection error parameter and the second failure detection error parameter are determined, so that the power failure detection network is optimized and regulated, and the optimized and regulated target power failure detection network is generated. And finally, carrying out power failure detection on the input target system state monitoring data based on the optimized network. Therefore, the accuracy and the efficiency of power failure detection are improved, the running state of the power system is monitored in real time, and the power failure is found and processed in time, so that the stable and safe running of the power system is ensured.

Specifically, the method and the device for detecting the power failure in the power system acquire and process the sequence of the sample power system operation data clusters, perform knowledge learning, determine failure detection error parameters, optimize and adjust the power failure detection network, generate an optimized target power failure detection network, and can effectively identify and predict the most probable power failure state from various power failure state labels triggered by the reference power system unit.

Firstly, knowledge learning is carried out by loading a first sample power system operation data cluster to a power failure detection network, and a first failure detection error parameter is determined, so that the identification capability of the power failure detection network to a reference power failure state label can be improved.

Then, the second sample power system operation data cluster is loaded to a power failure detection network, the power failure state labels triggered by the reference power system units are classified according to the power failure detection network, a first failure classification chart is generated, and the process can further optimize the performance of the power failure detection network.

Based on the first confidence coefficient and the threshold confidence coefficient, a second fault detection error parameter in a knowledge learning process is determined, and the accuracy and the stability of the power fault detection network can be ensured in the process.

And finally, optimizing and adjusting the power failure detection network based on the first failure detection error parameter and the second failure detection error parameter to generate an optimized and adjusted target power failure detection network. The process can enable the power failure detection network to accurately and effectively detect the power failure when the power failure detection network faces actual target system state monitoring data.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated, for the sake of simplicity, and it should be understood that the following drawings only illustrate some embodiments of the present application and should therefore not be considered as limiting the scope, and that other related drawings can be obtained by those skilled in the art without the inventive effort.

Fig. 1 is a schematic flow chart of a power failure detection method based on big data analysis according to an embodiment of the present application;

fig. 2 is a schematic functional block diagram of a power failure detection system based on big data analysis for implementing the power failure detection method based on big data analysis according to an embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the present application. Thus, the present application is not limited to the embodiments described, but is to be accorded the widest scope consistent with the claims.

Referring to fig. 1, the application provides a power failure detection method based on big data analysis, which comprises the following steps.

Step S110, a sample power system operation data cluster sequence is obtained. The sample power system operation data cluster sequence comprises a plurality of first sample power system operation data clusters and a plurality of second sample power system operation data clusters.

In this embodiment, each of the first example power system operation data clusters includes system state monitoring data of a reference power system unit triggering a reference power failure state tag, and each of the second example power system operation data clusters includes other system state monitoring data of the reference power system unit triggering other than the reference power failure state tag.

For example, in practical applications, faults that may occur in a power system tend to be complex and variable, and some faults may be co-initiated by a variety of factors, which faults tend to be not accurately identified by simple threshold comparisons alone.

It is assumed that a complex power system comprising hundreds of generators, transformers and distribution networks is currently being monitored. Each component in a complex power system is provided with a plurality of sensors, and a plurality of parameters such as voltage, current, frequency, temperature, vibration and the like can be monitored. Wherein data of a part of devices where complex faults have occurred (e.g., system instability caused by frequency anomaly fluctuation) can be selected as a first example power system operation data cluster; and the other part of equipment data which does not have the type of faults or has the other type of faults is used as a second sample power system operation data cluster.

By way of example, assume that a system instability fault caused by frequency anomaly fluctuations in a complex power system is under investigation. Such faults may be caused by a combination of factors, such as power instability, excessive load, etc., and generally need to be identified and predicted by machine learning methods.

First example power system operation data cluster: data of some generator sets which have historically had system instability faults caused by abnormal frequency fluctuations can be selected. These data include various parameter variations prior to the occurrence of the fault, such as power output, load size, rotational speed, temperature, etc. Meanwhile, each data point corresponds to a label, which indicates whether a reference fault (namely, system instability caused by abnormal fluctuation of frequency) occurs. These data constitute a first example power system operational data cluster.

Second example power system operation data cluster: other data of the generator set which is not subjected to system instability faults caused by abnormal frequency fluctuation or faults caused by other reasons (such as excessive current, overheat and the like) can be collected. The data also includes various operating parameters, but unlike the first example data cluster, this portion of data does not trigger a reference fault tag, but may trigger other types of fault tags, or does not trigger any fault tags (i.e., the device is operating properly). This portion of data constitutes a second example power system operational data cluster.

By collecting and comparing these two types of data, the power failure detection network can learn what operating conditions may lead to system instability with frequency anomaly fluctuations, and what operating conditions are normal or may cause other types of failures.

Step S120, loading the first sample power system operation data cluster into a power failure detection network to perform knowledge learning, and determining a first failure detection error parameter in the knowledge learning process.

For example, a deep learning model (such as a convolutional neural network) may be used as the power failure detection network. And inputting a first sample power system operation data cluster, and training a power failure detection network to learn how to identify a system state which possibly causes abnormal fluctuation of frequency from multiple monitoring data. In this process, the error between the power failure detection network prediction and the actual failure label, i.e. the first failure detection error parameter, may be calculated.

Step S130, loading the second sample power system operation data cluster into the power failure detection network, and classifying the power failure state label triggered by the reference power system unit in the second sample power system operation data cluster according to the power failure detection network, so as to generate a first failure classification diagram.

In this embodiment, the first fault classification map includes a first confidence that the power fault detection network estimates that the reference power system unit triggers each of the reference power fault status tags.

For example, the power failure detection network may be further trained and tested with a second sample power system operational data cluster, generating a fault classification map based on predictions of the power failure detection network for likely fault conditions for each device. For example, a power failure detection network may predict that a certain device has 60% probability of abnormal fluctuation in frequency, 30% probability of other types of failures, and 10% probability of normal operation.

Step S140, determining a second fault detection error parameter in the knowledge learning process based on the first confidence level and the threshold confidence level.

For example, a threshold value, for example 70%, may be set, and if the confidence of the prediction of the occurrence of the frequency abnormality fluctuation of the equipment by the power failure detection network is lower than this value, the prediction is considered to be possibly inaccurate, and the second failure detection error parameter may be calculated by comparing the prediction result with the actual result.

And step S150, optimizing and adjusting the power failure detection network based on the first failure detection error parameter and the second failure detection error parameter, generating an optimized and adjusted target power failure detection network, and performing power failure detection on the input target system state monitoring data based on the target power failure detection network.

For example, the power failure detection network may be further optimized according to the first failure detection error parameter and the second failure detection error parameter, e.g. by adjusting the weight of the power failure detection network, changing the network structure of the power failure detection network, or adjusting the training strategy. The optimized target power fault detection network can more accurately identify the system state which possibly causes abnormal fluctuation of frequency, so that early warning is carried out in actual operation in advance, and faults are prevented from occurring.

Based on the steps, through obtaining a sample power system operation data cluster sequence, data comprising a trigger reference power failure state label and system state monitoring data comprising other power failure state labels are respectively loaded into a power failure detection network to carry out knowledge learning, and a first failure detection error parameter and a second failure detection error parameter are determined, so that the power failure detection network is optimized and regulated, and an optimized and regulated target power failure detection network is generated. And finally, carrying out power failure detection on the input target system state monitoring data based on the optimized network. Therefore, the accuracy and the efficiency of power failure detection are improved, the running state of the power system is monitored in real time, and the power failure is found and processed in time, so that the stable and safe running of the power system is ensured.

In a possible implementation manner, the sequence of the sample power system operation data clusters further includes fault labeling data corresponding to the first sample power system operation data cluster, where the fault labeling data reflects an actual fault state label of the power fault state label triggered by the reference power system unit in the first sample power system operation data cluster.

Step S120 may include:

step S121, loading the first sample power system operation data cluster into the power failure detection network, and classifying the power failure state label triggered by the reference power system unit in the first sample power system operation data cluster according to the power failure detection network, so as to generate a second failure classification diagram. The second fault classification map includes a second confidence that the power fault detection network estimates that the reference power system element triggered each of the reference power fault status tags.

Step S122, determining a first fault detection error parameter in the knowledge learning process based on the fault labeling data and the second fault classification map.

For example, each step is specifically described in conjunction with the previous complex scenario example:

in current monitored large power systems, equipment such as generators, transformers, etc. are equipped with various sensors that collect the operating parameters of the equipment in real time. The operation data of a part of equipment which has once had a specific fault (for example, system instability caused by frequency abnormality fluctuation) is selected as a first example power system operation data cluster. Meanwhile, when the equipment fails, the failure state of the equipment can be recorded and used as failure marking data. The fault labeling data is an actual fault state label reflecting the power fault state actually triggered by the reference power system unit in the first sample power system operation data cluster.

a) The first example power system operational data cluster is input into a power failure detection network (e.g., a deep learning model). The power failure detection network will learn and understand from these data various factors and conditions that lead to the occurrence of specific failures (system instability caused by frequency anomaly fluctuations). Through this process, the network can generate a second fault classification map. In this second fault classification diagram, the power fault detection network gives a predictive confidence that each reference power system unit (e.g., each generator) triggers each reference power fault status tag (e.g., system instability due to frequency anomaly fluctuations), which is the second confidence.

b) After the second fault classification map is generated, the prediction accuracy of the power fault detection network needs to be verified. For this purpose, the prediction result of the power failure detection network (i.e., the second failure classification map) is compared with the actual failure marking data. By calculating the difference between the predicted result and the actual result, a first fault detection error parameter may be obtained. This first fault detection error parameter can reflect the accuracy of the power fault detection network in the knowledge learning process and can be used for subsequent network optimization.

In one possible embodiment, the power failure detection network comprises a first feature extraction unit and a second feature extraction unit. Step S121 may include:

step S1211, loading the first sample power system operation data cluster to the first feature extraction unit, and performing feature extraction on the first sample power system operation data cluster according to the first feature extraction unit, so as to generate a first power system operation path vector.

For example, the present embodiment constructs a power failure detection network including two feature extraction units. The first feature extraction unit is used for carrying out feature extraction on input original data to generate an electric power system running path vector. And the second feature extraction unit uses the tag centroid influence parameters to fuse the running path vectors of the first power system and generate the target fault tag features.

First, the collected first sample power system operation data cluster (i.e., the operation data of the equipment in which the system instability fault caused by the frequency abnormal fluctuation has occurred) is input into the first feature extraction unit of the power fault detection network. The first feature extraction unit processes the input first sample power system operation data cluster, extracts key features related to faults, and organizes the features into a vector, and the vector can be understood as a first power system operation path vector.

Illustratively, these first example power system operational data clusters contain a large amount of information, but not all information is useful for the current task. For example, it is possible that only data from a few sensors has a direct relationship with the faults of interest in the present embodiment, while data from other sensors may not be of interest. In addition, the raw data may also contain a lot of noise, such as data fluctuations due to sensor errors, environmental disturbances, etc.

At this time, the first feature extraction unit may process the first example power system operation data cluster, extract useful information therein, and remove irrelevant noise. Specifically, the following operations may be performed:

1. The data of several sensors associated with the fault are selected and the data of the other sensors are ignored.

2. The selected data is subjected to noise reduction processing, for example, high-frequency noise is removed using a filter.

3. Important features in the data are extracted, such as calculating an average, maximum, minimum, etc. of each sensor data.

After these processes, a vector of characteristics is obtained, which is the first power system operation path vector, for example, a large amount of operation data can be collected from various sensors of the generator, including but not limited to: rotational speed, load size, temperature, vibration, etc.

The following steps are selected to extract features: rotational speed: the average, maximum and minimum values of the generator rotational speed are calculated. It is assumed that they are 1000rpm, 1200rpm and 800rpm, respectively. Load size: the average, maximum and minimum values of the generator load size are calculated. They are assumed to be 50%,70% and 30%, respectively. Temperature: the average, maximum and minimum values of the generator temperature are calculated. They were assumed to be 60 ℃, 80 ℃ and 40 ℃, respectively. Vibration: the average, maximum and minimum values of generator vibration are calculated. They are assumed to be 0.5mm, 1mm and 0.2mm, respectively. By the above steps, a vector composed of 12 elements, namely, a first power system operation path vector is obtained:

[1000, 1200, 800, 50%, 70%, 30%, 60℃, 80℃, 40℃, 0.5mm, 1mm, 0.2mm]

This vector is the so-called first power system path vector. Each element represents a key operating characteristic of the generator over time. This first power system operational path vector will be input into the power failure detection network for predicting possible future failures. This first power system path vector captures critical information in the raw data related to faults of interest to the present embodiment, and ignores extraneous details and noise, thereby providing a clearer, more accurate input for subsequent fault detection.

Step S1212, loading the first power system running path vector to the second feature extraction unit, and fusing the first power system running path vector by using a tag centroid influencing parameter according to the second feature extraction unit to generate a first target fault tag feature, where the tag centroid influencing parameter includes a plurality of tag centroid influencing factors, each of the tag centroid influencing factors corresponds to one of the reference power fault state tags, and dimensions of the first target fault tag feature are the same as the number of the reference power fault state tags.

For example, the first power system operational path vector may then be input into a second feature extraction unit of the power failure detection network. This second feature extraction unit uses a set of predefined tag centroid influencing parameters that reflect the extent to which each possible fault condition (i.e. the reference power fault condition tag) influences the system operating condition. And the second feature extraction unit fuses the running path vectors of the first power system through the tag centroid influence parameters to generate first target fault tag features.

By way of example, assume that the goal of the present embodiment is to detect and predict three possible fault conditions in a power system: system instability caused by abnormal fluctuation of frequency, overheat fault caused by overload and current fluctuation fault caused by unstable power supply. These are the reference power failure status labels.

The first power system operation path vector, such as [ average rotation speed, maximum rotation speed, minimum rotation speed, average load size, maximum load size, minimum load size, average temperature, maximum temperature, minimum temperature ], has been obtained by the first feature extraction unit.

Next, this first power system run path vector needs to be input to a second feature extraction unit that fuses the first power system run path vector using the tag centroid influencing parameters.

The tag centroid influencing parameter is a set of predefined parameters, each parameter (or tag centroid influencing factor) corresponding to a reference power failure state tag. For example, there may be a parameter dedicated to measuring the effect of rotational speed on system instability due to frequency anomaly fluctuations, another parameter dedicated to measuring the effect of load size on overload-induced overheat faults, and so on.

The second feature extraction unit may associate each feature in the first power system path vector with all fault state tags based on the tag centroid influencing parameters. For example, it may calculate how much the possibility of system instability due to occurrence of abnormal fluctuation in frequency increases when the average rotation speed increases, how much the possibility of overheat failure due to occurrence of overload increases, and so on.

In this way, a new feature vector, i.e., the first target failure signature feature, is obtained. The dimensions of this first target fault signature feature are the same as the number of reference power fault state signatures (3 in the example), each dimension representing the likelihood of a particular fault state.

Step S1213, performing smooth extremum conversion on the first target fault label feature according to a smooth extremum conversion mechanism, generating a second target fault label feature, and outputting the second target fault label feature as a second fault classification chart, where a feature unit parameter in the second target fault label feature represents a second confidence level of the power fault detection network in estimating that the reference power system unit triggers the corresponding reference power fault state label.

For example, the first target failure tag feature may be processed using a method known as a smooth extremum shifting mechanism. The method of the smooth extremum conversion mechanism can effectively eliminate noise and abnormal values in data and improve the stability and prediction accuracy of the power failure detection network.

And finally, outputting the second target fault label characteristics after the smooth extremum conversion as a second fault classification chart. In this second fault classification diagram, each device is assigned a confidence value associated with its corresponding reference power fault status label, i.e. the power fault detection network estimates the confidence that the device triggered the fault label, which can help to understand the type of fault that each device may have and its likelihood.

Illustratively, continuing with the above example, first, a first power system travel path vector (e.g., [1000, 1200, 800, 50%, 70%, 30%, 60 ℃, 80 ℃, 40 ℃, 0.5mm, 1mm, 0.2mm ]) is loaded into the second feature extraction unit. This unit will process the vector using predefined tag centroid influencing parameters.

Assume that there are four reference power failure state labels: abnormal frequency, overheating, overload, and mechanical failure. Each fault has a label centroid influencing factor associated with it, e.g. the frequency anomaly may be a factor of 0.8, the superheat 0.6, the overload 0.7 and the mechanical fault 0.5. These factors reflect the extent to which various faults affect the operating state of the system.

The second feature extraction unit uses the tag centroid influencing parameters to fuse the first power system operation path vectors to generate a first target fault tag feature. Specifically, it may multiply each feature value by a corresponding tag centroid influencing factor and then add the results together. For example, for frequency anomalies, it may calculate (1000 x 0.8+1200 x 0.8+800 x 0.8+50% 0.8+70% 0.8+30% 0.8+60 ℃ 0.8+80 ℃ 0.8+40 ℃ 0.8+0.5 mm x 0.8+1 mm x 0.8+0.2 mm x 0.8), resulting in a value. Likewise, it will calculate similar values for the other three faults.

Next, the first target failure tag feature is processed using a smooth extremum shifting mechanism. This mechanism may eliminate extremum and noise in the data, making the generated features more stable. The processed result is the second target failure signature feature.

And finally, outputting the second target fault label characteristic as a second fault classification map. In this second fault classification map, each reference power fault status label corresponds to a confidence value that indicates the confidence that the power fault detection network predicts that the generator will trigger the fault. For example, if the confidence level of the frequency anomaly is 0.9, the confidence level of the overheat is 0.2, the confidence level of the overload is 0.3, and the confidence level of the mechanical fault is 0.1, then it can be concluded that the generator is most likely to have the frequency anomaly.

In one possible implementation, step S122 may include:

step S1221, determining a corresponding first influence feature from the tag centroid influence factors based on the actual fault state tags of the power fault state tags in the fault labeling data, and outputting the tag centroid influence factors except for the first influence feature in the tag centroid influence factors as a second influence feature.

For example, this process is explained in detail with previous complex power system failure scenarios:

first, fault annotation data are checked, which reflect the actual state of each device at the time of the fault. For example, in a system instability fault caused by abnormal fluctuation of frequency, the possible rotation speed is the most critical influence characteristic, and then the system instability fault is determined as the first influence characteristic; while others such as load size, temperature, etc. are considered as second influencing features.

Step S1222, determining a first similarity parameter value between the first power system operational path vector and the first influencing feature, and determining a second similarity parameter value between the first power system operational path vector and each of the second influencing features.

For example, a similarity between a first power system operation path vector (i.e., a feature vector extracted from the device operation data) and a first influencing feature (rotational speed) is calculated, resulting in a first similarity parameter value. Similarly, the similarity between the first power system operation path vector and the second influence characteristic (load size, temperature and the like) is calculated, and a second similarity parameter value is obtained.

Step S1223, calculating the difference between the first similarity parameter value and the preset parameter value, and generating a first target parameter value.

For example, a difference between a first similarity parameter value (i.e., the similarity between the actual operating state of the device and the fault state) and a preset parameter value (e.g., the similarity value in the ideal state) is calculated to obtain a first target parameter value. This value reflects the deviation between the actual operating state of the device and the fault state.

Step S1224, determining a first fault detection error parameter in the knowledge learning process based on the first target parameter value and each of the second similarity parameter values.

For example, finally, a first fault detection error parameter in the knowledge learning process is determined based on the first target parameter value (i.e., the deviation of the primary fault state) and the respective second similarity parameter value (i.e., the deviation of the other possible fault states). The error parameter can help to know the prediction accuracy of the power failure detection network and provide basis for subsequent optimization.

The first target parameter value and the first fault detection error parameter are in a reverse association relationship, namely the larger the deviation of the main fault state is, the larger the error is; the second similarity parameter value and the first fault detection error parameter are in a forward correlation, namely, the larger the deviation of other possible fault states is, the larger the error is.

In one possible implementation, step S1213 may include:

1. a preset distribution adjustment factor is obtained. The size of the distribution adjustment factor is greater than 1.

2. And carrying out distribution adjustment on the first target fault tag characteristic according to the distribution adjustment factor to generate a third target fault tag characteristic.

3. And carrying out smooth extremum conversion on the third target fault label characteristic through a smooth extremum conversion mechanism to generate a second target fault label characteristic.

For example, this process is explained in the context of a previous complex power system failure scenario:

it is assumed that a first target fault signature is now obtained, e.g. [0.7, 0.2, 0.1], which represents the probability that the current power system state may lead to system instability due to frequency anomaly fluctuations, overheating faults due to overload, and current ripple faults due to supply instability, respectively.

First, a preset distribution adjustment factor needs to be obtained. The magnitude of this distribution adjustment factor is greater than 1 for adjusting the distribution of the various fault conditions. Assume that the acquired distribution adjustment factor is 1.5.

And then, carrying out distribution adjustment on the first target fault tag characteristic according to the distribution adjustment factor. Specifically, each characteristic value is multiplied by a distribution adjustment factor. For example, for system instability caused by frequency anomaly fluctuations, 0.7x1.5=1.05 is calculated. Similarly, similar values would be calculated for the other two faults. In this way, a third target failure signature feature is obtained, e.g., [1.05, 0.3, 0.15].

And finally, processing the third target fault label characteristic by using a smooth extremum conversion mechanism. This mechanism can effectively eliminate extremum and noise in the data, making the generated features more stable. For example, if a feature value deviates too much from other feature values, the smooth extremum shifting mechanism will adjust it back to a reasonable range. After processing, a second target failure signature feature is obtained, e.g., [0.95, 0.27, 0.14]. These values are more stable than the original first target fault signature characteristics and more reflective of the true state of the power system.

In general, the process optimizes the original fault tag characteristics through distribution adjustment and smooth extremum conversion, so that the original fault tag characteristics more accurately reflect various fault states possibly occurring in the power system.

In one possible embodiment, the power failure detection network comprises a first feature extraction unit and a second feature extraction unit. Step S130 may include:

step S131, loading the second sample power system operation data cluster to the first feature extraction unit, and performing feature extraction on the second sample power system operation data cluster according to the first feature extraction unit to generate a second power system operation path vector.

For example, continuing with the previous complex power system as an example, this process is explained in detail:

at this stage, operational data of another set of failed power system equipment (e.g., generators) is collected, referred to as a second sample power system operational data cluster. Then, these data are input to a first feature extraction unit, useful information or features are extracted from the raw data, and a second power system running path vector is generated. For example, it is possible to calculate the average, maximum and minimum values of various parameters such as rotational speed, load size, temperature, etc.

Step S132, loading the second power system operation path vector to the second feature extraction unit, and fusing the second power system operation path vector by using a tag centroid influence parameter according to the second feature extraction unit to generate a fourth target fault tag feature, where the tag centroid influence parameter includes a plurality of tag centroid influence factors, each tag centroid influence factor corresponds to one reference power fault state tag, and the dimension of the fourth target fault tag feature is the same as the number of the reference power fault state tags.

For example, next, the second power system running path vector is input into the second feature extraction unit. This unit processes the vector using the tag centroid influencing parameters to generate a fourth target fault tag signature. Each tag centroid influencing factor corresponds to a reference power failure state tag, such as frequency anomalies, overheating, overload, mechanical failure, and the like.

Step S133, performing smooth extremum conversion on the fourth target fault tag feature according to a smooth extremum conversion mechanism, generating a fifth target fault tag feature, and outputting the fifth target fault tag feature as a first fault classification chart, where feature unit parameters in the fifth target fault tag feature represent a first confidence level of the power fault detection network in estimating the reference power system unit triggering corresponding to the reference power fault state tag.

For example, finally, the fourth target fault signature is processed using a smooth extremum shifting mechanism to generate a fifth target fault signature. This step may help to eliminate extrema and noise in the data, making the generated features more stable. Then, the fifth target failure tag feature is output as the first failure classification map. In this figure, each reference power fault status label corresponds to a confidence value that indicates the confidence that the power fault detection network predicts that the device will trigger the fault. For example, if the confidence of the frequency anomaly is 0.9, the confidence of the overheat is 0.2, the confidence of the overload is 0.3, and the confidence of the mechanical fault is 0.1, then it can be concluded that the equipment is most likely to be frequency anomaly.

Step S140 may include:

and step S141, determining a third similarity parameter value between the second power system running path vector and each tag centroid influencing factor.

Step S142, determining a preset parameter value based on the threshold confidence.

Step S143, calculating the difference between the maximum third similarity parameter value and the preset parameter value, and generating a third target parameter value.

Step S144, determining a second fault detection error parameter in the knowledge learning process based on the third target parameter value.

For example, the previous power system failure scenario continues to be used to explain this process:

it is assumed that after the processing of the first feature extraction unit and the second feature extraction unit, a new power system operation path vector, i.e., a second power system operation path vector, is obtained. For example, this vector may be data that has been transformed with smoothed extrema, such as [0.95, 0.27, 0.14]. Then, the similarity between this vector and each of the tag centroid influencing factors (e.g., influencing factors corresponding to frequency anomalies, overheating, overload, and mechanical failure) is calculated to obtain a third similarity parameter value.

Next, a threshold confidence level is set, e.g., 0.8, indicating that a fault condition is considered to be highly likely when the confidence level of the fault condition exceeds this value. Based on this threshold confidence level, a preset parameter value is determined, for example also 0.8.

Then, the maximum value of all third similarity parameter values, for example 0.95, is found, and then the difference between this maximum value and the preset parameter value (0.8) is calculated, resulting in a third target parameter value, for example 0.15. This third target parameter value reflects the deviation between the actual operating state of the power system and the expected fault state.

Finally, a second fault detection error parameter in the knowledge learning process is determined based on the third target parameter value (0.15). This second fault detection error parameter may be 0.15 reflecting the predictive accuracy of the power fault detection network.

The third target parameter value and the second fault detection error parameter are in a forward association relationship, namely the larger the deviation is, the larger the error is. This second fault detection error parameter may help to understand the performance of the power fault detection network and provide a basis for subsequent optimization.

In one possible implementation, step S140 may further include: and calculating the difference between the first confidence coefficient corresponding to each reference power fault state label and the threshold confidence coefficient, and generating a second target parameter value corresponding to each reference power fault state label. And if the second target parameter value greater than 0 exists, comparing the magnitude of each second target parameter value, and determining the second fault detection error parameter based on the largest second target parameter value. Or if the second target parameter values are not greater than 0, determining that the magnitude of the second fault detection error parameter is 0.

at this stage, a first confidence has been obtained for each reference power fault condition (such as frequency anomalies, overheating, overload, mechanical faults, etc.). Then, a difference between the first confidence and the threshold confidence of each fault state is calculated, and a second target parameter value corresponding to each reference power fault state label is generated. For example, if the first confidence level of the frequency anomaly is 0.9 and the threshold confidence level is 0.8, then the second target parameter value of the frequency anomaly is 0.9-0.8=0.1.

Next, it may be detected whether there is a second target parameter value greater than 0. If so, the magnitudes of all second target parameter values are compared and a second fault detection error parameter is determined based on the largest one. For example, if the second target parameter values for frequency anomalies, overheating, overload, and mechanical faults are 0.1, -0.2, 0.3, and 0.05, respectively, then the largest 0.3 is selected as the second fault detection error parameter. This means that the current power system state is most likely to result in an overload fault. And if all the second target parameter values are not greater than 0, determining that the second fault detection error parameter is 0, wherein the current power system state does not exceed the normal range, and the fault is not worried about.

Through the steps, the operation state of the power system can be effectively evaluated, and the possible fault type can be predicted.

In one possible implementation, step S150 may include:

step S151, determining the first cluster number of the first example power system operation data clusters and the second cluster number of the second example power system operation data clusters in the sample power system operation data cluster sequence.

in the foregoing sample power system operation data cluster sequence, two types of data are included: the first example power system operation data cluster and the second example power system operation data cluster. It is first necessary to determine how many clusters of data, i.e. the first cluster number and the second cluster number, each of the two types of data contains.

Step S152, determining a first importance coefficient corresponding to the first fault detection error parameter based on the first cluster number, and determining a second importance coefficient corresponding to the second fault detection error parameter based on the second cluster number.

For example, the importance coefficients of the first and second fault detection error parameters may be determined according to the first and second cluster numbers, respectively. For example, if the number of first clusters is large, the first importance coefficient corresponding to the first failure detection error parameter is high, and conversely, low.

Step S153, performing weight fusion on the first fault detection error parameter and the second fault detection error parameter based on the first importance coefficient and the second importance coefficient, so as to generate a global fault detection error parameter.

For example, the first fault detection error parameter and the second fault detection error parameter may be weighted and fused according to the first importance coefficient and the second importance coefficient to obtain the global fault detection error parameter. This process can be understood as giving different weights to the respective error parameters according to the number of data of each class, and then fusing them together.

And step S154, optimizing and adjusting the power failure detection network based on the global failure detection error parameter to generate an optimized and adjusted power failure detection network.

Finally, the power failure detection network is optimally adjusted according to the global failure detection error parameter, for example. This may include modifying the network structure, adjusting the learning rate, etc. After optimization, the optimized and adjusted power failure detection network is obtained, and the failure state of the power system can be predicted more accurately.

In a possible implementation manner, in step S150, the step of performing power failure detection on the input target system state monitoring data based on the target power failure detection network includes:

Step S155, obtaining basic system state monitoring data, performing recursive feature elimination on each basic system state monitoring data, generating initial traversal selection features corresponding to each basic system state monitoring data, adding the obtained basic system state monitoring data into a data sequence to be processed, detecting the number of sequence members in the current data sequence to be processed, and if the number of the sequence members is not smaller than the set number, extracting a first number of basic system state monitoring data from the data sequence to be processed according to a preset number interval parameter, so as to generate a first system state monitoring data sequence.

For example, basic system state monitoring data of the electric power system, such as voltage, current, rotational speed, and temperature, are first acquired. To simplify the problem, these four parameters can be considered as four features. Then, each piece of basic system state monitoring data is subjected to recursive feature elimination to screen out the most important features, and the screened out features form initial traversal selection features. Illustratively, recursive feature elimination (Recursive Feature Elimination, RFE) is a feature selection method that eliminates or retains the best or worst performing feature by iteratively modeling and selecting it, and then repeating this process with the remaining features until all features have been traversed. In an example, recursive feature elimination may be used to determine which of the four features voltage, current, speed, and temperature are most important for power failure detection. The specific operation may be as follows: first, an initial fault detection model is constructed using all four features, and then the importance of each feature is evaluated. Assuming that in the first iteration the voltage is found to be the least important feature, it is removed from the feature list. Next, feature importance evaluation and selection is performed again among the remaining three features (current, rotation speed, and temperature), and so on until the most important feature is found. Thus, for each piece of underlying system state monitoring data, a corresponding initial traversal selection feature is generated, i.e., the feature that is considered most important after recursive feature elimination. These features will be used in the subsequent power failure detection process.

On this basis, all collected basic system state monitoring data may be added to the data sequence to be processed. If the number of data in the data sequence to be processed reaches a set number, e.g. 1000, data is extracted therefrom according to a predetermined number interval parameter (e.g. one group of every 200 data), a first system state monitoring data sequence is generated.

Step S156, outputting the initial traversal selection feature corresponding to the basic system state monitoring data in the current data sequence to be processed as a recursive feature elimination node, and dividing each basic system state monitoring data in the first system state monitoring data sequence based on the recursive feature elimination node to generate the target system state monitoring data.

For example, the dominant underlying system state monitoring data (e.g., up-to-date or abnormally frequent) in the data sequence to be processed may be extracted, with its corresponding initial traversal selection feature as a recursive feature elimination node. The first system state monitoring data sequence may then be partitioned using the recursive feature elimination node to obtain target system state monitoring data.

Illustratively, first, it is necessary to determine which underlying system state monitoring data is dominant, i.e., most important for fault detection. For example, if it is found through analysis that voltage abnormality is the main cause of the fault, then the voltage data is the dominant basic system state monitoring data. Then, a portion of the features from the voltage data sequence are selected as initial traversal selection features, such as a maximum, minimum, and average of the voltages, which may be selected, and output as a recursive feature elimination node. The other underlying system state monitoring data sequences (such as current and frequency) are then partitioned according to the recursive feature elimination node. For example, it is possible to find out the change in current and frequency when the voltage maximum exceeds a threshold. In this way, more valuable information is extracted from the original system state monitoring data, generating target system state monitoring data. Through the steps, the most important features can be screened from massive system state monitoring data, and more accurate data support is provided for subsequent fault detection.

Step S157, loading the target system state monitoring data into the target power failure detection network, classifying the power failure state label triggered by the target power system in the target system state monitoring data according to the target power failure detection network, and generating a third failure classification chart. The third fault classification map includes a third confidence that the target power fault detection network estimates that the target power system triggered each of the reference power fault status tags.

For example, target system status monitoring data may be entered into the target power failure detection network. The network classifies the power failure states that may be triggered based on the input data and generates a third failure classification map. The third fault classification map includes a third confidence level for each of the reference power fault conditions (e.g., frequency anomalies, overheating, etc.), reflecting the likelihood that the power system may trigger such faults.

And step S158, comparing the third confidence coefficient corresponding to each reference power fault state label with a threshold confidence coefficient, and if the third confidence coefficient larger than the threshold confidence coefficient exists, comparing the third confidence coefficient, and determining a target power fault state label based on the reference power fault state label corresponding to the largest third confidence coefficient. Or if the third confidence coefficient is not greater than the threshold confidence coefficient, determining that the target power failure state label is an unknown power failure state label.

For example, the third confidence for each reference power fault condition may be compared to a set threshold confidence (e.g., 0.8). If any fault state has the third confidence coefficient greater than the threshold confidence coefficient, comparing all the third confidence coefficient, and taking the largest corresponding fault state as a target power fault state label. For example, if the third confidence of the frequency anomaly is 0.9 and the third confidence of the other faults are all less than 0.8, then the target power fault status tag is considered to be a frequency anomaly. If all of the third confidences are not greater than the threshold confidence, then the target power failure state label is determined to be an unknown power failure state label, indicating that the current system state does not exceed the threshold of any of the failure states.

Fig. 2 schematically illustrates a big data analysis based power failure detection system 100 that may be used to implement various embodiments described herein.

For one embodiment, FIG. 2 illustrates a big data analysis based power failure detection system 100, the big data analysis based power failure detection system 100 having a plurality of processors 102, a control module (chipset) 104 coupled to one or more of the processor(s) 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM)/storage device 108 coupled to the control module 104, a plurality of input/output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 104.

Processor 102 may include a plurality of single-core or multi-core processors, and processor 102 may include any combination of general-purpose or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In some alternative implementations, the big data analysis based power failure detection system 100 can function as a server device such as a gateway as described in the embodiments of the present application.

In some alternative embodiments, the big data analysis based power failure detection system 100 may include a plurality of computer readable media (e.g., memory 106 or NVM/storage 108) having instructions 114 and a plurality of processors 102 combined with the plurality of computer readable media configured to execute the instructions 114 to implement the modules to perform the actions described in this disclosure.

For one embodiment, the control module 104 may include any suitable interface controller to provide any suitable interface to one or more of the processor(s) 102 and/or any suitable device or component in communication with the control module 104.

The control module 104 may include a memory controller module to provide an interface to the memory 106. The memory controller modules may be hardware modules, software modules, and/or firmware modules.

Memory 106 may be used to load and store data and/or instructions 114 for power failure detection system 100 based on big data analysis, for example. For one embodiment, memory 106 may comprise any suitable volatile memory, such as, for example, a suitable DRAM. In some alternative embodiments, memory 106 may comprise a double data rate type four synchronous dynamic random access memory.

For one embodiment, the control module 104 may include a plurality of input/output controllers to provide interfaces to the NVM/storage 108 and the input/output device(s) 110.

For example, NVM/storage 108 may be used to store data and/or instructions 114. NVM/storage 108 may include any suitable non-volatile memory (e.g., flash memory) and/or may include any suitable non-volatile storage(s).

NVM/storage 108 may include storage resources that are physically part of the device on which power failure detection system 100 is installed based on big data analysis, or which may be accessible by the device, but may not be necessary as part of the device. For example, NVM/storage 108 may be accessed via input/output device(s) 110 in connection with a network.

Input/output device(s) 110 may provide an interface for power failure detection system 100 based on big data analysis to communicate with any other suitable device, input/output device 110 may include a communication component, pinyin component, sensor component, and the like. The network interface 112 may provide an interface for the big data analysis based power failure detection system 100 to communicate in accordance with a plurality of networks, the big data analysis based power failure detection system 100 may communicate wirelessly with a plurality of components of a wireless network in accordance with any of a plurality of wireless network standards and/or protocols, such as accessing a wireless network in accordance with a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof.

For one embodiment, one or more of the processor(s) 102 may be packaged together with logic of a plurality of controllers (e.g., memory controller modules) of the control module 104. For one embodiment, one or more of the processor(s) 102 may be packaged together with logic of multiple controllers of the control module 104 to form a system in package. For one embodiment, one or more of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104. For one embodiment, one or more of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104 to form a system-on-chip.

In various embodiments, the big data analysis based power failure detection system 100 may be, but is not limited to being: a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet, a netbook, etc.), and the like. In various embodiments, the big data analysis based power failure detection system 100 may have more or fewer components and/or different architectures. For example, in some alternative embodiments, the big data analysis based power failure detection system 100 includes a plurality of cameras, a keyboard, a liquid crystal display screen (including a touch screen display), a non-volatile memory port, a plurality of antennas, a graphics chip, an application specific integrated circuit, and a speaker.

The foregoing has outlined rather broadly the more detailed description of the present application, wherein specific examples have been provided to illustrate the principles and embodiments of the present application, the description of the examples being provided solely to assist in the understanding of the method of the present application and the core concepts thereof; meanwhile, as those skilled in the art will have modifications in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.

Claims

1. A power failure detection method based on big data analysis, characterized by being applied to a power failure detection system based on big data analysis, the method comprising:

2. The method for detecting a power failure based on big data analysis according to claim 1, wherein the sequence of sample power system operation data clusters further includes failure marking data corresponding to the first sample power system operation data cluster, and the failure marking data reflects an actual failure state label of a power failure state label triggered by the reference power system unit in the first sample power system operation data cluster;

3. The power failure detection method based on big data analysis according to claim 2, wherein the power failure detection network includes a first feature extraction unit and a second feature extraction unit; the loading the first sample power system operation data cluster into the power failure detection network, classifying the power failure state label triggered by the reference power system unit in the first sample power system operation data cluster according to the power failure detection network, and generating a second failure classification diagram, including:

4. The method for detecting a power failure based on big data analysis according to claim 3, wherein the determining a first failure detection error parameter in a knowledge learning process based on the failure annotation data and the second failure classification map includes:

5. The method for detecting a power failure based on big data analysis according to claim 3, wherein the performing smooth extremum conversion on the first target failure tag feature according to a smooth extremum conversion mechanism, generating a second target failure tag feature, includes:

6. The power failure detection method based on big data analysis according to claim 1, wherein the power failure detection network includes a first feature extraction unit and a second feature extraction unit; the loading the second sample power system operation data cluster into the power failure detection network, classifying the power failure state label triggered by the reference power system unit in the second sample power system operation data cluster according to the power failure detection network, and generating a first failure classification diagram, including:

determining a preset parameter value based on the threshold confidence;

7. The method for detecting a power failure based on big data analysis according to claim 1, wherein the determining a second failure detection error parameter in a knowledge learning process based on the first confidence level and a threshold confidence level includes:

8. The method for detecting a power failure based on big data analysis according to claim 1, wherein the power failure detection network is optimally adjusted based on the first failure detection error parameter and the second failure detection error parameter, and the generating an optimally adjusted target power failure detection network includes:

9. The power failure detection method based on big data analysis according to any one of claims 1 to 8, wherein the step of performing power failure detection on the input target system state monitoring data based on the target power failure detection network includes:

10. A big data analysis based power failure detection system, comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the big data analysis based power failure detection method of any of claims 1-9.