CN117368789A - Power failure co-location method and system based on multi-source heterogeneous data - Google Patents
Power failure co-location method and system based on multi-source heterogeneous data Download PDFInfo
- Publication number
- CN117368789A CN117368789A CN202311424117.8A CN202311424117A CN117368789A CN 117368789 A CN117368789 A CN 117368789A CN 202311424117 A CN202311424117 A CN 202311424117A CN 117368789 A CN117368789 A CN 117368789A
- Authority
- CN
- China
- Prior art keywords
- data
- time sequence
- time
- power supply
- enhancement
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 239000013598 vector Substances 0.000 claims abstract description 68
- 238000010606 normalization Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 14
- 239000013604 expression vector Substances 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 14
- 230000002159 abnormal effect Effects 0.000 claims description 10
- 230000007246 mechanism Effects 0.000 claims description 7
- 230000015654 memory Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000007500 overflow downdraw method Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000008045 co-localization Effects 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 2
- 230000004807 localization Effects 0.000 claims 1
- 230000007787 long-term memory Effects 0.000 claims 1
- 238000004891 communication Methods 0.000 abstract description 27
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 description 7
- 238000012549 training Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 206010033799 Paralysis Diseases 0.000 description 1
- 241000540325 Prays epsilon Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000005111 flow chemistry technique Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/40—Testing power supplies
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Locating Faults (AREA)
Abstract
The invention belongs to the technical field of communication power failure positioning, and provides a power failure collaborative positioning method and a system based on multi-source heterogeneous data, which acquire power data; carrying out normalization processing on the power supply data; carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data; and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model. The method has the advantages that the time sequence of the communication power supply is considered, the power supply data under three time scales of the current time, the same time of the last week and the same time of the last year are collected, normalization processing is carried out, all indexes are in the same order of magnitude, the time sequence data enhancement technology based on a sliding window is introduced to represent the dependency relationship and the statistical characteristics among the time sequence data, the influence of different dimensions and the time sequence on prediction is solved, and the power failure positioning accuracy is improved.
Description
Technical Field
The invention belongs to the technical field of communication power failure positioning, and particularly relates to a power failure collaborative positioning method and system based on multi-source heterogeneous data.
Background
Communication power supplies are a key infrastructure of communication networks, known as the heart of the communication system. If the communication power supply fails abnormally, the communication equipment is interrupted and cannot work and operate normally, and even the whole system is paralyzed. With the development of telecommunication industry, the reliability and stability of the communication network have higher standard requirements, so that the detection of the health condition of the communication power supply is of great importance.
The health condition of the communication power supply can be often reflected by a plurality of indexes such as current, voltage, operating temperature, battery operating life and the like during operation. However, as the scale of the communication network is continuously enlarged, the data size of the related indexes of the communication power supply is large, and the relationship between different indexes is complicated. And when the fault occurs, the fault point is timely salvaged, so that the influence on the operation of the whole communication network can be reduced. Therefore, due to the large data volume, complex relationship among data indexes, real-time requirement on the rush repair of the fault point and the like, the use of a manual method for judging whether the communication power supply fails or not is not feasible. With the continuous development of artificial intelligence and deep learning technologies, these technologies are also gradually applied to solve the problems of fault location and the like in various power scenes.
The inventor finds that the difference of the dimensions of the related data of the communication system can influence the learning effect of the neural network; the power data often has strong time sequence, the network prediction model obtained by directly learning the power data does not consider the time sequence influence among the data, so that the prediction result of the network prediction model is inaccurate, and the effective power failure positioning cannot be performed.
Disclosure of Invention
In order to solve the problems, the invention provides a power failure collaborative positioning method and a system based on multi-source heterogeneous data.
In order to achieve the above object, the present invention is realized by the following technical scheme:
in a first aspect, the present invention provides a power failure co-location method based on multi-source heterogeneous data, including:
acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
carrying out normalization processing on the power supply data under three time scales;
carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
Further, during the training of the fault location network model, feature extraction is carried out to obtain low-dimensional expression vectors on three time scales; performing weighted fusion on the low-dimensional representation vectors on three time scales; inputting the weighted and fused characteristics into a network model to learn time sequence information; after learning, the distance between the abnormal sample feature representation and the normal sample feature representation is pulled away, and the distance between the normal sample feature representations is pulled up.
Furthermore, a one-dimensional convolutional neural network is adopted to extract the features, a concentration mechanism feature fusion method is adopted to distribute different concentration weights to the low-dimensional representation vectors on three time scales for fusion, and the weighted and fused features are input into a long-period memory network for learning time sequence information.
Further, learning the fused weighted fusion vector to obtain a comprehensive expression vector; calculating a contrast learning loss value by using the comprehensive representation vector to pull away the distance between the abnormal sample representation vector and the normal sample representation vector and pull back the distance between the positive sample representation vectors; performing fault discrimination by adopting a multi-layer perceptron and a softmax function, and calculating a fault discrimination loss value output by the softmax function; based on the obtained comparison loss value and fault discrimination loss value, parameters in the model are adjusted by utilizing a gradient back propagation algorithm, and the fault positioning accuracy is continuously optimized.
Further, normalization processing is performed to make all indexes in the power supply data under three time scales in the same order of magnitude.
Further, a sliding window based data enhancement module generates a timing enhancement vector including a dependency relationship between timing data and statistical features.
Furthermore, a multi-layer perceptron is introduced to complete the positioning of power failure.
In a second aspect, the present invention also provides a power failure co-location system based on multi-source heterogeneous data, including:
a data acquisition module configured to: acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
a normalization module configured to: carrying out normalization processing on the power supply data under three time scales;
a data enhancement module configured to: carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
a fault location module configured to: and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
In a third aspect, the present invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the steps of the multi-source heterogeneous data based power failure co-location method of the first aspect.
In a fourth aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the power failure co-location method based on multi-source heterogeneous data according to the first aspect when the processor executes the program.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention fully considers the time sequence of the communication power supply, collects the power supply data under three time scales of the current time, the same time of the last week and the same period of the last year, performs normalization processing to ensure that all indexes are in the same order of magnitude, introduces a time sequence data enhancement technology based on a sliding window to express the dependency relationship and the statistical characteristics among the time sequence data, solves the influence of different dimensions and the time sequence on prediction, and improves the positioning accuracy of the power supply fault;
2. based on the obtained three time scale information, the invention provides an improved CNN-LSTM network to better learn and fuse time sequence characteristic information from different time scales and generate a comprehensive representation vector; meanwhile, contrast learning is introduced, a contrast learning loss value is calculated, the distance between the abnormal sample and the normal sample representing vectors is pulled away, the distance between the positive sample representing vectors is pulled close, and the accuracy of fault positioning is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification, illustrate and explain the embodiments and together with the description serve to explain the embodiments.
FIG. 1 is an overall flow chart of embodiment 1 of the present invention;
FIG. 2 is a diagram showing a data flow processing procedure according to embodiment 1 of the present invention;
FIG. 3 is a schematic diagram of a system structure according to embodiment 1 of the present invention;
fig. 4 is a flowchart of embodiment 2 of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
Example 1:
the difference of the dimensions of the related data of the communication system can influence the learning effect of the neural network. And the power data often has strong time sequence, so that when specific problems are solved, the characteristics in the data are required to be fully extracted, and the time sequence information in the data is required to be fully learned, so that the power failure positioning can be more accurately performed. Based on this, the present embodiment provides a power failure co-location method based on multi-source heterogeneous data, including:
acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
carrying out normalization processing on the power supply data under three time scales;
carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
Specifically, by carrying out normalization processing on the power supply data, all indexes are in the same order of magnitude; the time sequence of the communication power supply is fully considered, the power supply data under the three time scales of the current time, the same time of the last week and the same time of the last year are collected, the time sequence data enhancement technology based on a sliding window is introduced to express the dependency relationship and the statistical characteristics among the time sequence data, the influence of different dimensions and the time sequence on prediction is solved, and the power supply fault positioning accuracy is improved.
Optionally, based on the obtained three time scales of the current time, the same time of the last week and the same period of the last year, adopting a min-max normalization method to normalize the collected data so as to eliminate dimension, and enabling all indexes to be in the same order of magnitude;
generating a time sequence enhancement vector containing the dependency relationship and the statistical characteristics among time sequence data by using the normalized data of three time scales and adopting a data enhancement method based on a sliding window; the preprocessed data is input to a data enhancement module based on a sliding window, and a time sequence enhancement vector containing the dependency relationship and the statistical characteristics among time sequence data is generated;
performing feature learning by using the obtained time sequence enhancement vectors on three time scales and adopting an improved CNN-LSTM network based on multiple time scales; specifically, firstly, a one-dimensional convolutional neural network (Convolution Neural Network, CNN) is adopted to extract characteristics, redundant information is removed, and a low-dimensional representation vector only containing representative key information is obtained; then, using the obtained low-dimensional representation vectors on three time scales, and adopting an attention mechanism (Attention Mechanism) feature fusion method to allocate different attention weights to the low-dimensional representation vectors on the three time scales for fusion, so as to generate a weighted and fused representation vector; finally, constructing a multi-layer LSTM neural network by using the weighted fusion vector, and learning the fused weighted fusion vector to obtain a comprehensive representation vector, namely, inputting the weighted fusion characteristic into a Long short-term memory (LSTM) learning time sequence information; the method comprises the steps of carrying out a first treatment on the surface of the
After CNN-LSTM learning based on multiple time scales, introducing contrast learning, and pulling away the distance between the abnormal sample characteristic representation and the normal sample characteristic representation and pulling up the distance between the normal sample characteristic representations; specifically, the obtained comprehensive expression vector is utilized, the idea of contrast learning is introduced to calculate a contrast learning loss value, so that the distance between the abnormal sample (negative sample) expression vector and the normal sample (positive sample) expression vector is pulled away, the distance between the positive sample expression vectors is pulled up, and more effective feature expression vectors are learned. Meanwhile, a Multi-Layer Perceptron (MLP) is introduced to complete the positioning of power failure; specifically, performing fault discrimination by adopting a multi-layer perceptron and a softmax function, and calculating a fault discrimination loss value output by the softmax function; then, based on the obtained comparison loss value and fault discrimination loss value, utilizing a gradient back propagation algorithm to adjust parameters in the model, and continuously optimizing fault positioning accuracy;
after model training is completed, data which does not participate in the training process can be used for testing, a model test result is compared with an actual situation, and the model is optimized and adjusted, so that weight data in the model are optimized continuously, and fault positioning accuracy is improved continuously.
Corresponding to the scheme, the embodiment also provides a power failure co-location system based on multi-source heterogeneous data, which comprises a data preprocessing module, a time sequence data enhancement module, a time sequence data learning module, a comparison learning module, a failure location module and a feedback updating module; optionally:
the data preprocessing module is used for carrying out normalization processing by using a min-max normalization method by utilizing the collected three time scales of the current time, the same time of the previous week and the same period of the last year, so that all indexes are in the same order of magnitude;
the time sequence data enhancement module is used for processing the normalized three time scale data by using a data enhancement method based on a sliding window, and generating a time sequence enhancement representation vector containing the dependency relationship and the statistical characteristics among the time sequence data;
the time sequence data learning module firstly utilizes a one-dimensional convolutional neural network to extract characteristics of time sequence enhancement expression vectors on three time scales respectively, and removes redundant information to obtain low-dimensional expression vectors only containing representative key information; then, using an attention mechanism feature fusion method to distribute different weights to the key information low-dimensional representation vectors on three time scales and fusing the key information low-dimensional representation vectors to generate a weighted and fused representation vector; then, learning the weighted and fused representation vector by using a multi-layer LSTM neural network to finally obtain a comprehensive representation vector;
the contrast learning module is used for calculating contrast learning loss values by using the obtained comprehensive expression vectors and introducing the thought of contrast learning so as to pull the distance between the abnormal sample (negative sample) expression vectors and the normal sample (positive sample) expression vectors and pull the distance between the positive sample expression vectors to learn more effective feature expression vectors;
the fault positioning module is used for carrying out fault judgment by using the obtained comprehensive expression vector and adopting a Multi-Layer Perceptron (MLP) and a softmax function, and calculating a fault judgment loss value output by the softmax function; then, based on the obtained comparison loss value and fault discrimination loss value, utilizing a gradient back propagation algorithm to adjust parameters in the model, and continuously optimizing fault positioning accuracy;
and the feedback updating module is used for testing by using data which does not participate in the training process after the model is trained, comparing the model test result with the actual situation, optimizing and adjusting the model, continuously optimizing weight data in the model and continuously improving the fault positioning accuracy.
Example 2:
to further explain the method in embodiment 1, the embodiment provides a power failure co-location method based on multi-source heterogeneous data, which specifically includes:
s1, based on the fact investigation, whether the communication power supply at the current position fails or not, and not only the data index of the current power supply needs to be considered. The data index of the communication power supply is different from the normal state and does not belong to the fault state in different seasons or important time sections.
Therefore, the data index of the mass communication power supply is acquired, including the current time t 1 At the same time t of the previous week 2 Three times t in the same period of the last year 3 The method comprises the steps of preprocessing acquired data according to the characteristics of the scale such as alternating current voltage, direct current voltage, total load current and machine room temperature, including data cleaning, missing value complementation and data normalization.
Specifically, based on the data from 28 in 2020, 6 and 28 in 2023 provided by a certain electric company, the data includes information such as ac voltage, dc output voltage, total load current, machine room temperature, and battery operation time of the communication power supply at a certain time point.
Because the coverage range of the power supply data is wide, different dimensions and dimension units exist among the data indexes, and the data analysis result is influenced. In order to eliminate the influence of dimensions among different data indexes, a Min-Max normalization method is firstly applied to each collected dimension characteristic data x to perform normalization processing, so that all indexes are in the same order of magnitude:
wherein, max () is a maximum value method in a certain dimension characteristic value; min () is the minimum method of taking the feature value of this dimension.
S2, based on each dimension of the feature data z' preprocessed in the step S1, a feature expression vector expression V of the communication power supply at a certain moment can be obtained.
S2.1, based on the communication power supply historical data, feature expression vectors between different time points are interdependent. Therefore, the present embodiment uses a sliding window based data enhancement method to generate timing enhancement vectors including dependency relationships and statistical features between timing data for three time scales of timing data, respectively. Where the feature space of the time-sequential enhancement representation vector is much larger than the original time-sequential feature space, this will facilitate the recognition of the most representative feature in the smaller space at a later feature extraction.
Taking the current time scale as an example, for a time series with a length of C, a sliding window Win is set first i The size of b (b)>1) Then a time series is obtainedThe windows are as follows:
T t1 ＝<Win 1 ，Win 2 ，…，Win C >
Win i ＝<V i ，V i+1 ，…，V i+b-1 >
this practice isIn an embodiment, the sliding window size is optionally set to 4. For a coverage size between two consecutive sliding windows of
For each dimension of the time series of each sliding window, 2 derived features, namely a canonical (NOR) feature and a canonical difference feature (difference of norm, DON) are calculated, and the calculation formula of the j-th dimension feature is shown as the formula:
the new time series is obtained by calculation through the formula as follows:
T′ t1 ＝<G 1 ，G 2 ，…，G C′ >
wherein,is a matrix; k represents the k-dimensional features, the first row represents the computed canonical (NOR) features and the canonical interpolation features (difference of norm, DON).
S2.2, selecting a sliding window with the size f based on the new time sequence T' obtained in the S2.1, and covering the sliding window with the size fThe window is obtained as follows: :
Win i ＝<G i ，G i+1 ，…，G i+f-1 >
and 8 statistical features of mean, minimum, maximum, first quartile, second quartile, third quartile, standard deviation, and peak-to-peak were calculated for each derivative feature (NOR and DON), respectively. The time sequence enhancement sequence is obtained through calculation as follows:
T″ t1 ＝<H 1 ，H 2 ，…，H C″ >
wherein,is a matrix.
S3, based on time sequence enhancement vectors on three time scales, learning is conducted through an improved CNN-LSTM network based on multiple time scales.
S3.1, obtaining time sequence enhancement vectors T' on three time scales based on the step S2 t1 、T″ t2 And T' t3 For H therein i And splicing the rows of the matrix to obtain vectors with the sizes of 2 x 8 x k respectively. And then, extracting the obtained features by using a one-dimensional convolutional neural network, removing redundant information to obtain a low-dimensional representation vector only containing representative key information, and in the example, optionally setting the size of a convolutional kernel to be 3, wherein the process is calculated as shown in the following formula:
L t1 ＝Conv 1D (Concat(T″ t1 ))
L t2 ＝Conv 1D (Concat(T″ t2 ))
L t3 ＝Conv 1D (Concat(T″ t3 ))
s3.2, based on the low-dimensional representation vectors on the three time scales generated in the step S3.1, considering that the information obtained on different time scales has different influences on the final result, the attention mechanism (Attention Mechanism) feature fusion method is adopted to fuse the distributed different attention weights in the embodiment, and the fusion calculation formula is as follows:
wherein L is w High-level vector representation of what is a vector containing important information, considered as a fixed query, which is a training processRandom initialization and learning; norm () represents a layer normalization operation; d, d embedding Representing the dimensions of a low-dimensional representation vector, L t Is a vector representation obtained after fusion.
And S3.3, based on the fused weighted fusion vector obtained in the step S3.2, inputting the weighted fusion vector into a multi-layer LSTM network to learn time sequence characteristics, wherein the LSTM can selectively update and forget the time sequence information. The LSTM network calculation formula is as follows:
f t ＝σ(W f ·[h t-1 ，L t ]+b f )
i t ＝σ(W i ·[h t-1 ，L t ]+b i )
o t ＝σ(W o ·[h t-1 ，L t ]+b o )
h t ＝o t *tanh(C t )
wherein W is f ，W i ，W C And W is o All represent weight parameter matrix, b f ，b i ，b C And b o Are offset values; σ () represents a sigmoid function.
S4, based on the comprehensive representation vector Out which can be obtained after LSTM network learning in the step S3, in the embodiment, a supervised contrast learning method is used, firstly, samples of normal data and abnormal data samples are respectively divided into a positive sample set and a negative sample set according to sample labels, then, contrast learning loss values are calculated by utilizing the comprehensive representation vector Out, distances between the abnormal samples and the normal sample representation vectors are pulled, and distances between the positive sample representation vectors are shortened so as to learn more effective feature representation vectors, and a contrast learning calculation formula is as follows:
w＝Proj(out)
where Proj (·) represents a single linear network, called a mapping network. The network maps out to w, and at the end of this period the normalization is applied, i.e. n=w-i w i.
Wherein L is sup Representing the loss value calculated by contrast learning, i≡i {1 … N } representing the index value of the sample; out is provided with i Representing the output value of the ith sample in the mapping network; representing a dot product operation; τ ε R + A real number parameter is set up>0). In this embodiment it is set to τ=0.1. For a (I) ≡i\ { I }, the index value I is called the anchor, and p is the index value of all positive samples. It should be noted that for any one anchor i, it may have multiple positive samples. P (i) ≡ { p.epsilon.A { i } p ＝y i The index set of all positive samples except i, where y i The label corresponding to the sample with index i is represented, and the I P (i) I represents the cardinal number of the sample.
S5, based on the LSTM network, the comprehensive representation vector Out can be obtained after learning, in the embodiment, a multi-layer perceptron is used as a classification layer, whether the power supply at the current position fails or not is judged, and the calculation process is as follows:
y′＝Softmax(MLP(Out))
and F, acquiring a final judging result of whether the communication power supply fails or not by adopting an iterative minimization objective function mode based on historical training sample set data, and turning to the step F.
The fault detection is performed on the communication power supply sample to be tested, and the result is compared with the actual situation, in this embodiment, the Accuracy (Accuracy) and the F1 score (F1-score) are used as evaluation indexes, and the comparison result is shown in table 1:
table 1 fault detection contrast cases
| Accuracy | F1-score | |
| Linear Regression | 0.9229 | 0.3018 |
| LSTM | 0.9177 | 0.0028 |
| BiLSTM | 0.9188 | 0.2591 |
| ConvLSTM | 0.9182 | 0.0327 |
| ConvBiLSTM | 0.9225 | 0.1551 |
| This embodiment | 0.9559 | 0.4912 |
Based on the experimental results in table 1, the performance of the power failure positioning method provided in this embodiment is better than that of other methods.
S6, adopting cross entropy as a loss function for model discrimination, and if y is true category distribution, defining the loss function as follows:
wherein N represents the total number of samples; l (L) ce Representing the calculated cross entropy loss value.
S7, through the step S4 and the step S6, the loss function of the model overall is defined as follows:
L＝L sup +L ce
where L is the overall loss function value of the model. And (3) continuously minimizing L through the iterative solution of the step S6 and the step S7, so as to obtain the learning parameters of the CNN-LSTM network model based on multiple time scales.
Example 3:
the embodiment provides a power failure co-location system based on multi-source heterogeneous data, which comprises:
a data acquisition module configured to: acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
a normalization module configured to: carrying out normalization processing on the power supply data under three time scales;
a data enhancement module configured to: carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
a fault location module configured to: and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
The working method of the system is the same as the power failure co-location method based on multi-source heterogeneous data in embodiment 1, and will not be described here again.
Example 4:
the present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the multi-source heterogeneous data based power failure co-localization method described in embodiment 1.
Example 5:
the present embodiment provides an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the steps of the multi-source heterogeneous data based power failure co-location method of embodiment 1 are implemented when the processor executes the program.
The above description is only a preferred embodiment of the present embodiment, and is not intended to limit the present embodiment, and various modifications and variations can be made to the present embodiment by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment should be included in the protection scope of the present embodiment.
Claims (10)
1. A power failure co-location method based on multi-source heterogeneous data is characterized by comprising the following steps:
acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
carrying out normalization processing on the power supply data under three time scales;
carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
2. The power failure co-location method based on multi-source heterogeneous data according to claim 1, wherein when a failure location network model is trained, feature extraction is performed to obtain low-dimensional representation vectors on three time scales; performing weighted fusion on the low-dimensional representation vectors on three time scales; inputting the weighted and fused characteristics into a network model to learn time sequence information; after learning, the distance between the abnormal sample feature representation and the normal sample feature representation is pulled away, and the distance between the normal sample feature representations is pulled up.
3. The power failure collaborative positioning method based on multi-source heterogeneous data according to claim 2, wherein a one-dimensional convolutional neural network is adopted for feature extraction, a attention mechanism feature fusion method is adopted for fusion of low-dimensional expression vectors on three time scales by distributing different attention weights, and the weighted and fused features are input into a long-term and short-term memory network for learning time sequence information.
4. The method for collaborative positioning of power failures based on multi-source heterogeneous data according to claim 2, wherein the fused weighted fusion vectors are learned to obtain comprehensive expression vectors; calculating a contrast learning loss value by using the comprehensive representation vector to pull away the distance between the abnormal sample representation vector and the normal sample representation vector and pull back the distance between the positive sample representation vectors; performing fault discrimination by adopting a multi-layer perceptron and a softmax function, and calculating a fault discrimination loss value output by the softmax function; based on the obtained comparison loss value and fault discrimination loss value, parameters in the model are adjusted by utilizing a gradient back propagation algorithm, and the fault positioning accuracy is continuously optimized.
5. The method for collaborative localization of power failures based on multi-source heterogeneous data according to claim 1, wherein the normalization process is performed such that the indices of the power data at three time scales are in the same order of magnitude.
6. The method of claim 1, wherein the sliding window based data enhancement module generates a timing enhancement vector comprising dependencies and statistics between the timing data.
7. The method for collaborative positioning of power failure based on multi-source heterogeneous data according to claim 1, wherein a multi-layer perceptron is introduced to complete power failure positioning.
8. A multi-source heterogeneous data based power failure co-location system, comprising:
a data acquisition module configured to: acquiring power supply data under three time scales of the current time, the same time of the last week and the same period of the last year;
a normalization module configured to: carrying out normalization processing on the power supply data under three time scales;
a data enhancement module configured to: carrying out data enhancement on the normalized power supply data to obtain a time sequence enhancement vector comprising the dependency relationship and the statistical characteristics among time sequence data;
a fault location module configured to: and obtaining a positioning result based on the time sequence enhancement vector and a preset fault positioning network model.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the multi-source heterogeneous data based power failure co-localization method as claimed in any one of claims 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the multi-source heterogeneous data based power failure co-localization method of any of claims 1-7 when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311424117.8A CN117368789A (en) | 2023-10-30 | 2023-10-30 | Power failure co-location method and system based on multi-source heterogeneous data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311424117.8A CN117368789A (en) | 2023-10-30 | 2023-10-30 | Power failure co-location method and system based on multi-source heterogeneous data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117368789A true CN117368789A (en) | 2024-01-09 |
Family
ID=89392679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311424117.8A Pending CN117368789A (en) | 2023-10-30 | 2023-10-30 | Power failure co-location method and system based on multi-source heterogeneous data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117368789A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117930954A (en) * | 2024-03-22 | 2024-04-26 | 深圳威尔视觉科技有限公司 | Cloud server outage monitoring method and system |
-
2023
- 2023-10-30 CN CN202311424117.8A patent/CN117368789A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117930954A (en) * | 2024-03-22 | 2024-04-26 | 深圳威尔视觉科技有限公司 | Cloud server outage monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106251059B (en) | Cable state evaluation method based on probabilistic neural network algorithm | |
| CN111813084B (en) | Mechanical equipment fault diagnosis method based on deep learning | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| JP2022037241A (en) | Abnormality detection system, abnormality detection method, abnormality detection program, and method for generating learned model | |
| JP2023042527A (en) | Data drive active power distribution network abnormal state sensing method and system | |
| CN112859822B (en) | Equipment health analysis and fault diagnosis method and system based on artificial intelligence | |
| CN111860658A (en) | Transformer fault diagnosis method based on cost sensitivity and integrated learning | |
| CN106709820A (en) | Power system load prediction method and device based on deep belief network | |
| CN117368789A (en) | Power failure co-location method and system based on multi-source heterogeneous data | |
| CN116679211B (en) | Lithium battery health state prediction method | |
| CN116186633A (en) | Power consumption abnormality diagnosis method and system based on small sample learning | |
| CN110766313A (en) | Cable tunnel comprehensive state evaluation method based on operation and maintenance system | |
| CN112884570A (en) | Method, device and equipment for determining model security | |
| CN115221233A (en) | Transformer substation multi-class live detection data anomaly detection method based on deep learning | |
| CN117150409A (en) | Power consumption abnormality detection method | |
| CN114781875A (en) | Micro-grid economic operation state evaluation method based on deep convolutional network | |
| Huang et al. | A hybrid bayesian deep learning model for remaining useful life prognostics and uncertainty quantification | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN113158448A (en) | Self-adaptive calculation method for equipment threshold of ship system | |
| CN110717577A (en) | Time series prediction model construction method for noting regional information similarity | |
| RU135434U1 (en) | DEVICE FOR AUTOMATED SIGNAL IDENTIFICATION BASED ON NEURAL NETWORK | |
| CN113610229A (en) | Active learning method based on norm uncertainty index | |
| CN112819067A (en) | Method, device, equipment and storage medium for processing bad data of power distribution network | |
| CN117952592B (en) | Intelligent management method for charging pile | |
| CN114781526B (en) | Depth semi-supervised image classification method based on discriminant feature learning and entropy |
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 |