CN110321947A - A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks - Google Patents
A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks Download PDFInfo
- Publication number
- CN110321947A CN110321947A CN201910567009.3A CN201910567009A CN110321947A CN 110321947 A CN110321947 A CN 110321947A CN 201910567009 A CN201910567009 A CN 201910567009A CN 110321947 A CN110321947 A CN 110321947A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- convolutional neural
- neural networks
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The present invention relates to electric power network technique field, in particular to a kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks.The process of process and data training including data prediction;The process of data prediction includes: to carry out attribute hierarchies analysis to data by pretreatment script file;It requires to carry out attribute hierarchies classification to data according to training fault data;Classification approximation data are merged;It recalculates and generates the distance between Various types of data;It generates preliminary linear data and judges the classification accuracy of data;Data classification completes storage storage;Data preparation is completed;The process of data training includes: production abnormal data training dataset;Check whether state confirmation is adjusted into monitoring center;After completing training, data processing is carried out, abnormal sample is obtained;Using normal anomaly sample adjustment means, magnanimity negative sample is automatically generated;Normal anomaly sample is made into data set, is trained;Training is completed to test model;Training is completed.
Description
(1) technical field
The present invention relates to electric power network technique field, in particular to a kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks.
(2) background technique
With the fast development of depth learning technology, more and more applications have been all made of depth learning technology.In statistical data point
Analysis field is especially in progress rapidly, current to realize that the method for solving failure is that data are acquired and sent using existing acquisition equipment
Onto back-end server platform, server end carries out analysis storage to data, and platform is according to the threshold values of various abnormal problem data
Alarm failure, but with national grid, various enterprises and family to electricity consumption rely on, once occur abnormal problem again into
Row processing, generally causes certain loss, therefore is badly in need of using a kind of method, can be before the failure occurs in a period of time
Providing an anticipation, some place or equipment may break down in following certain time, and timely arrangement personnel mention
Before overhauled, more losses are generated after avoiding failure from occurring, in practice it has proved that use the data characteristics model based on deep learning
Analysis anticipation exception is a selection well.
It based on the algorithm research of deep learning neural network, realizes and multiplexing electric abnormality is prejudged, can effectively improve abnormal ask
Inscribe recognition speed and accuracy rate.
(3) summary of the invention
In order to compensate for the shortcomings of the prior art, the present invention provides a kind of multiplexing electric abnormality anticipation side based on convolutional neural networks
Method.
The present invention is achieved through the following technical solutions:
A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks includes the process of data prediction and the mistake of data training
Journey;The process of the data prediction is the following steps are included: S11 carries out attribute hierarchies point to data by pretreatment script file
Analysis;S12 requires to carry out attribute hierarchies classification to data according to training fault data;S13 merges classification approximation data;
S14, which is recalculated, generates the distance between Various types of data;S15 generates preliminary linear data and judges the classification accuracy of data;
S16 data classification completes storage storage;S17 data preparation is completed;The process of the data training is the following steps are included: S21 system
Make fault data training dataset;S22 enters monitoring center and checks whether state confirmation is adjusted;After S23 completes training, into
Row data processing obtains the sample of failure;S24 automatically generates magnanimity fault sample using normal and fault sample adjustment means;
Normal and fault sample is made into data set by S25, is trained;S26 training is completed to test model;S27 training is completed.
Wherein, the process of the data prediction needs to carry out the configuration of data prediction application platform in advance, specific to wrap
It includes and builds data prediction platform on source server;Configure sorting parameter and database;Whether confirmation pre-conditioning stage is disposed
Success.
The process of data prediction specifically includes the following steps:
S11 carries out attribute hierarchies analysis to data by pretreatment script file;
Wherein, attribute hierarchies disaggregated model is divided into three layers, and top layer is destination layer, and centre is rule layer, and bottom is solution layer, with
Top layer and each element of middle layer are criterion, construct next layer of attribute of an element judgment matrix related with it.
The data of acquisition specifically include electric current, voltage, power, four class data of electricity.
S12 requires to carry out attribute hierarchies classification to data according to training fault data;
Wherein, fault data includes voltage failure data, current failure data, power failure data, electricity fault data;
Wherein, voltage failure data include: under-voltage data, over-voltage data;Current failure data: current unbalance data, secondary string
Connect data;Power failure data: current loop data, secondary concatenated data;Electricity fault data: table code unequal number when total score
Data are jumped according to, table code data.
S13 merges classification approximation data;
S14, which is recalculated, generates the distance between Various types of data;
Various types of data refers to above-mentioned collected four classes data in S14;
S14 is the exceptional value based on Euclidean distance, means that exceptional value periphery does not have enough data.
S15 generates preliminary linear data and judges the classification accuracy of data;
S16 data classification completes storage storage;
S17 data preparation is completed;
The process of data prediction enhances method using data prediction-fault data:
Since the scarcity of fault data causes corresponding sample less, simultaneously because lack must for the preservation and arrangement of Primary Stage Data
The means wanted, therefore power failure related data lacks sufficient sample at this stage, is needed using the recognition methods of convolutional neural networks
A large amount of sample is wanted to carry out feature learning, the method enhanced using a kind of fault data is as follows:
Based on close sample multiple in distance metric selection classification, a sample is selected, and is randomly choosed a certain number of close
Adjacent sample, and reasonable noise increased to an attribute dimensions of that sample of selection, an attribute dimensions per treatment, in this way
More training datas can be generated.The data of construction are newly-generated rational samples, and are not present in original data set
's.Confrontation network is generated by condition GAN, close sample is automatically generated by finite sample, solves the few awkward situation of sample.
Data training process specifically includes the following steps:
Wherein, the process of data training needs to carry out the configuration of data training platform in advance, specifically includes and is taking on the server
Fault data prediction model platform is built on business device;Configure training initial parameter and model data;Confirm that data set and operation are flat
Platform deployment success.
Data training process specifically includes the following steps:
S21 makes fault data training dataset;
S22 enters monitoring center and checks state, is confirmed whether to be adjusted;
After carrying out initial training to abnormal data, state is checked into data storage designated position, is confirmed whether to be adjusted;Tool
Body are as follows: manual inspection training effect if training result and does not adjust initiation parameter if expected be consistent;If training result is not inconsistent
It closes expected results and then adjusts initiation parameter accordingly, until training result is consistent with expected results conjunction;
The training parameter initial value that the parameter of adjustment specifically refers to neural network meets Gaussian Profile mode.
After S23 completes training, data processing is carried out, abnormal sample is obtained;
Data processing is wherein carried out to specifically refer to handle original normal data using trained model.
S24 utilizes normal anomaly sample adjustment means, automatically generates magnanimity exceptional sample;
Data enhancing and expansion are carried out to data by relative program script in S24, so that generating magnanimity exceptional sample.
Normal anomaly sample is made into data set by S25, is trained;
S26 training is completed to test model;
S27 training is completed.
The process of data training uses shot and long term memory models LSTM(Long-Short Term Memory) carry out feature
It extracts, carries out tagsort using support vector machines (Support Vector Machine).Using TensorFlow training
Frame.
The electric power data multiplexing electric abnormality prediction model building neural network based of the process of data training, specifically:
Neural network is a kind of method of traditional supervised learning, it simulates human nerve's meta structure, can be approached arbitrary
The linear relationship of input and output, achieves good effect in practical application scene, and the model of deep learning has deeper
Structure and more complicated unit, neural network are substantially the processes for extracting data characteristics, and each layer of output can be
The character representation of initial data, its feature of more shallow layer is more specific and initial data is more similar, and the attribute of more deep layer is more abstract, tool
There is more obvious feature.This achievement utilizes depth network model, excavates the attribute of power failure data, constructs reflecting for input and output
Relationship is penetrated, can accurately predict the type of power failure.
Whole network includes two parts of feature extraction network and sorter network, and feature extraction can be just with recursive convolution mind
Through Chief Web Officer short-term memory network, traditional svm support vector machines is can be used in sorter network.Feature extraction forms a feature
Remove the classification layer of the last layer, only obtain corresponding characteristic information, and form abnormal data feature database in space.Loss function
Selection center loss function (Center Loss), one center of each Category Learning, and by all feature vectors of each classification
Corresponding class center is pulled to, joint Softmax is used together.The method of network optimization loss function uses Adam optimizer
(adaptive moment estimation), learning rate use 0.0001.
The beneficial effects of the present invention are: having been carried out whole the present invention is based on newest deep learning recursive neural network technology
The optimization of body is promoted, and is not only had greatly improved to linear data analysis discrimination, in performance and is trained to present aspect and has
Apparent advantage, classification, the analysis of data, the optimization of model greatly improve fault data identification prediction efficiency.Improve data
Analysis ability and failure predication accuracy.
(4) attached drawing
Fig. 1: the process flow diagram flow chart in the system platform of preprocessing server is applied;
Fig. 2: the process flow diagram flow chart in the system platform of training server is applied.
(5) specific embodiment
It is a kind of application of concrete condition of this method below:
The process of the data prediction the following steps are included:
S11 carries out attribute hierarchies analysis by data of the preprocessor script file to voltage, electricity, electric current, power, according to
The weight of each attribute carries out comprehensive assessment;
S12 requires to carry out attribute hierarchies classification to the data of four classes according to training fault data;
S13 merges collected all kinds of approximate datas;
S14 recalculates the Euclidean distance generated between Various types of data;
S15 generates preliminary linear classification data and judges the classification accuracy of data;
S16 data classification is completed and stores storage;
S17 data preparation is completed;
Data training process the following steps are included:
S21 makes the failure training dataset of four classes acquisition data;
S22 enters monitoring center and checks whether state confirmation is adjusted;
After S23 completes training, data processing is carried out, abnormal sample is obtained;
S24 utilizes normal anomaly sample adjustment means, automatically generates all kinds of magnanimity negative samples using procedure script;
Normal and fault sample is made into data set by S25, carries out network parameter training;
S26 carries out interim test to the model that training process generates by script, need to adjust, repeat the above steps;
S27 training is completed.
The present invention is described by way of example above, but the present invention is not limited to above-mentioned specific embodiment, all to be based on
Any changes or modifications that the present invention is done are fallen within the scope of the claimed invention.
Claims (7)
1. a kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks, it is characterised in that: the process including data prediction
With the process of data training;The process of the data prediction is the following steps are included: S11 passes through pretreatment script file to data
Carry out attribute hierarchies analysis;S12 requires to carry out attribute hierarchies classification to data according to training fault data;S13 is to classification approximation
Data merge;S14, which is recalculated, generates the distance between Various types of data;S15 generates preliminary linear data and judges data
Classification accuracy;S16 data classification completes storage storage;S17 data preparation is completed;The process of data training include with
Lower step: S21 makes fault data training dataset;S22 enters monitoring center and checks whether state confirmation is adjusted;S23
After completing training, data processing is carried out, abnormal sample is obtained;S24 utilizes normal anomaly sample adjustment means, automatically generates magnanimity
Negative sample;Normal anomaly sample is made into data set by S25, is trained;S26 training is completed to test model;S27 has been trained
At.
2. the multiplexing electric abnormality pre-judging method according to claim 1 based on convolutional neural networks, it is characterised in that: the number
The process of Data preprocess needs to be pre-configured with application platform, comprising the following steps: it is flat that data prediction is built on source server
Platform;Configure sorting parameter and database;Confirm pre-conditioning stage whether deployment success.
3. the multiplexing electric abnormality pre-judging method according to claim 1 based on convolutional neural networks, it is characterised in that: the number
It needs to be pre-configured with application platform according to trained process, comprising the following steps: build number of faults on the server
It is predicted that model platform;Configure training initial parameter and model data;Confirm data set and operation platform deployment success.
4. the multiplexing electric abnormality pre-judging method according to claim 1 based on convolutional neural networks, it is characterised in that: the category
Property hierarchy model, be divided into three layers, top layer is destination layer, centre be rule layer, bottom is solution layer, with top layer and middle layer
Each element be criterion, construct related with it next layer of attribute of an element judgment matrix.
5. the multiplexing electric abnormality pre-judging method according to claim 1 based on convolutional neural networks, it is characterised in that: acquisition
The data specifically include electric current, voltage, power, four class data of electricity.
6. the multiplexing electric abnormality pre-judging method according to claim 1 based on convolutional neural networks, it is characterised in that: number of faults
According to including voltage failure data, current failure data, power failure data, electricity fault data.
7. the multiplexing electric abnormality pre-judging method according to claim 6 based on convolutional neural networks, it is characterised in that: the electricity
Pressing fault data includes: under-voltage data, over-voltage data;Current failure data: current unbalance data, secondary concatenated data;Power
Fault data: current loop data, secondary concatenated data;Electricity fault data: table code unequal number evidence, table code data are jumped when total score
Parameter evidence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910567009.3A CN110321947A (en) | 2019-06-27 | 2019-06-27 | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910567009.3A CN110321947A (en) | 2019-06-27 | 2019-06-27 | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110321947A true CN110321947A (en) | 2019-10-11 |
Family
ID=68120478
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910567009.3A Pending CN110321947A (en) | 2019-06-27 | 2019-06-27 | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110321947A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110737685A (en) * | 2019-10-24 | 2020-01-31 | 南方电网科学研究院有限责任公司 | Data anomaly judgment method for electric power big data |
CN111597182A (en) * | 2020-05-20 | 2020-08-28 | 中国石油化工股份有限公司 | Convolutional neural network-based fault anomaly identification method for oil pumping unit driving motor |
CN113032917A (en) * | 2021-03-03 | 2021-06-25 | 安徽大学 | Electromechanical bearing fault detection method based on generation countermeasure and convolution cyclic neural network and application system |
CN113447820A (en) * | 2021-06-29 | 2021-09-28 | 国网北京市电力公司 | Electric quantity monitoring method and device, intelligent ammeter and processor |
CN113641726A (en) * | 2021-08-06 | 2021-11-12 | 国网北京市电力公司 | Unsupervised sheath current data mining system based on generation countermeasure network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105938609A (en) * | 2016-04-28 | 2016-09-14 | 国家电网公司 | Power grid operation assessment method for realizing multilayer indicator system |
CN106779069A (en) * | 2016-12-08 | 2017-05-31 | 国家电网公司 | A kind of abnormal electricity consumption detection method based on neutral net |
CN108416693A (en) * | 2018-02-01 | 2018-08-17 | 东华大学 | A kind of multiplexing electric abnormality Activity recognition method based on fuzzy neural network |
US20180262525A1 (en) * | 2017-03-09 | 2018-09-13 | General Electric Company | Multi-modal, multi-disciplinary feature discovery to detect cyber threats in electric power grid |
-
2019
- 2019-06-27 CN CN201910567009.3A patent/CN110321947A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105938609A (en) * | 2016-04-28 | 2016-09-14 | 国家电网公司 | Power grid operation assessment method for realizing multilayer indicator system |
CN106779069A (en) * | 2016-12-08 | 2017-05-31 | 国家电网公司 | A kind of abnormal electricity consumption detection method based on neutral net |
US20180262525A1 (en) * | 2017-03-09 | 2018-09-13 | General Electric Company | Multi-modal, multi-disciplinary feature discovery to detect cyber threats in electric power grid |
CN108416693A (en) * | 2018-02-01 | 2018-08-17 | 东华大学 | A kind of multiplexing electric abnormality Activity recognition method based on fuzzy neural network |
Non-Patent Citations (1)
Title |
---|
陈文娴: "基于深度学习的小样本异常用电数据检测技术研究", 《万方学位论文数据库》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110737685A (en) * | 2019-10-24 | 2020-01-31 | 南方电网科学研究院有限责任公司 | Data anomaly judgment method for electric power big data |
CN111597182A (en) * | 2020-05-20 | 2020-08-28 | 中国石油化工股份有限公司 | Convolutional neural network-based fault anomaly identification method for oil pumping unit driving motor |
CN113032917A (en) * | 2021-03-03 | 2021-06-25 | 安徽大学 | Electromechanical bearing fault detection method based on generation countermeasure and convolution cyclic neural network and application system |
CN113447820A (en) * | 2021-06-29 | 2021-09-28 | 国网北京市电力公司 | Electric quantity monitoring method and device, intelligent ammeter and processor |
CN113641726A (en) * | 2021-08-06 | 2021-11-12 | 国网北京市电力公司 | Unsupervised sheath current data mining system based on generation countermeasure network |
CN113641726B (en) * | 2021-08-06 | 2024-01-30 | 国网北京市电力公司 | Unsupervised sheath current data mining system based on generation of countermeasure network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110321947A (en) | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks | |
CN112383052B (en) | Power grid fault repairing method and device based on power internet of things | |
CN108985632A (en) | A kind of electricity consumption data abnormality detection model based on isolated forest algorithm | |
CN110995475A (en) | Power communication network fault detection method based on transfer learning | |
CN104052612B (en) | A kind of Fault Identification of telecommunication service and the method and system of positioning | |
CN109597401A (en) | A kind of equipment fault diagnosis method based on data-driven | |
CN110969194B (en) | Cable early fault positioning method based on improved convolutional neural network | |
CN108537259A (en) | Train control on board equipment failure modes and recognition methods based on Rough Sets Neural Networks model | |
CN111126820A (en) | Electricity stealing prevention method and system | |
CN117421684B (en) | Abnormal data monitoring and analyzing method based on data mining and neural network | |
CN109376061A (en) | A kind of information processing method and system | |
CN114386537A (en) | Lithium battery fault diagnosis method and device based on Catboost and electronic equipment | |
CN112668105A (en) | Helicopter transmission shaft abnormity judgment method based on SAE and Mahalanobis distance | |
CN113268370A (en) | Root cause alarm analysis method, system, equipment and storage medium | |
CN114328106A (en) | Log data processing method, device, equipment and storage medium | |
CN110059938B (en) | Power distribution network planning method based on association rule driving | |
CN114151293B (en) | Fault early warning method, system, equipment and storage medium of fan variable pitch system | |
CN111522705A (en) | Intelligent operation and maintenance solution method for industrial big data | |
CN117666546B (en) | Distributed control system fault diagnosis method and device | |
CN114021758A (en) | Operation and maintenance personnel intelligent recommendation method and device based on fusion of gradient lifting decision tree and logistic regression | |
CN117131425B (en) | Numerical control machine tool processing state monitoring method and system based on feedback data | |
CN115984158A (en) | Defect analysis method and device, electronic equipment and computer readable storage medium | |
CN110362828B (en) | Network information risk identification method and system | |
CN111209955A (en) | Airplane power supply system fault identification method based on deep neural network and random forest | |
CN115659258B (en) | Power distribution network fault detection method based on multi-scale graph roll-up twin network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191011 |
|
RJ01 | Rejection of invention patent application after publication |