CN117421699A - Electric energy meter fault fusion prediction method and system - Google Patents
Electric energy meter fault fusion prediction method and system Download PDFInfo
- Publication number
- CN117421699A CN117421699A CN202311724429.0A CN202311724429A CN117421699A CN 117421699 A CN117421699 A CN 117421699A CN 202311724429 A CN202311724429 A CN 202311724429A CN 117421699 A CN117421699 A CN 117421699A
- Authority
- CN
- China
- Prior art keywords
- similarity
- prediction result
- electric energy
- energy meter
- fault
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000004927 fusion Effects 0.000 title claims abstract description 22
- 239000013598 vector Substances 0.000 claims abstract description 143
- 230000002159 abnormal effect Effects 0.000 claims abstract description 15
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 11
- 230000002123 temporal effect Effects 0.000 claims description 38
- 238000000605 extraction Methods 0.000 claims description 13
- 101100534164 Arabidopsis thaliana SPT6 gene Proteins 0.000 claims description 12
- 101100067649 Caenorhabditis elegans gta-1 gene Proteins 0.000 claims description 12
- 101150107647 TUG1 gene Proteins 0.000 claims description 12
- 101150052307 gtb1 gene Proteins 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000004891 communication Methods 0.000 claims description 5
- 238000012217 deletion Methods 0.000 claims description 3
- 230000037430 deletion Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a system for predicting fault fusion of an electric energy meter, wherein the method comprises the following steps: collecting an original operation data sequence of the electric energy meter; performing multi-stage DWT decomposition on the normalized data sequence, inputting the normalized data sequence into a GCN model to extract each group of spatial features, and inputting a dual-channel GRU network to extract each group of time features; the method comprises the steps of respectively inputting each group of spatial features and each group of time features into a component prediction model after fusion to obtain a component prediction result, and inputting a reconstructed vector obtained after reconstruction into a knowledge graph model to predict abnormal working conditions of the electric energy meter to serve as a first fault prediction result; calculating and analyzing the similarity of the feature vectors to obtain a second fault prediction result; and finally, effectively fusing fault prediction results. The invention realizes the fault fusion prediction of the electric energy meter based on the combination of the DWT, the GCN model, the double-channel GRU network and the knowledge graph model, and improves the accuracy of the fault prediction.
Description
Technical Field
The invention relates to the field of fault prediction, in particular to a method and a system for predicting fault fusion of an electric energy meter.
Background
The electric energy meter is used as daily equipment widely used at present, effectively monitors and predicts faults, and can ensure normal life and normal working operation of various aspects of families, companies, units and the like. However, the current fault prediction for the electric energy meter is in a simple neural network prediction application aspect, and does not relate to deep mining and fusion prediction aspects of operation characteristics of the electric energy meter.
The knowledge graph is taken as a basic technology for analyzing and processing the current data, and is widely applied to various fields, but the prior art does not apply the knowledge graph model to the fault prediction aspect of the related equipment of the electric energy meter. Therefore, a method for improving the fault prediction accuracy in the electric energy meter field by fully utilizing the advantages of the knowledge graph model is needed.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a fault fusion prediction method and a fault fusion prediction system for an electric energy meter, which are based on the combination of DWT (Discrete wavelet transform) and GCN (Graph Convolutional Network) models, a double-channel GRU network (the GRU network is a gating logic unit and is also called a gating circulating unit) and a knowledge graph model to realize the fault fusion prediction of the electric energy meter, so that the depth characteristics of an operation data sequence of the electric energy meter are excavated, and the accuracy of fault prediction in the field of electric energy meter equipment is greatly improved.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
in a first aspect, the present invention provides a method for predicting fault fusion of an electric energy meter, the method comprising:
step 1, acquiring an original operation data sequence of an electric energy meter, and carrying out normalization operation;
step 2, performing multi-stage DWT decomposition on the normalized data sequence, wherein the multi-stage DWT decomposition is performed on the normalized data sequence to obtain an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
step 3, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model respectively to extract each group of spatial features;
step 4, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network respectively to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
step 5, after fusing each group of spatial features and each corresponding group of time features, respectively inputting a component prediction model to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
step 6, reconstructing the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), wherein Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
step 7, inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter, and taking the reconstruction vector Y as a first fault prediction result;
step 8, for the normalized data sequence in step 1, directly inputting a GCN model to extract space class information, and converting the space class information into a space feature vector x 1 ；
Step 9, for the normalized data sequence in step 1, directly inputting a dual-channel GRU network to extract time sequence information, and converting the time sequence information into a time feature vector x 2 ；
Step 10, the space vector set under the normal operation condition of the electric energy meter is GTA1, and the space vector set under the operation fault condition of the electric energy meter is GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarity as a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
step 11, comparing the first similarity with a first threshold after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
and step 12, effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
Further, the method further comprises the following steps: step 1, an original operation data sequence of the electric energy meter is collected, and normalization operation is carried out, specifically comprising the following steps:
the method comprises the steps of obtaining output voltage, output current and output power of a current electric energy meter, calculating electric energy, wiring modes, communication protocols and years, and performing data smoothing, outlier deletion and normalization operation.
Further, the method further comprises the following steps: the DWT discrete wavelet transform uses a Mallat fast algorithm.
Further, the method further comprises the following steps: step 3, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3, and the detail parameter W4 into a GCN model to extract each set of spatial features, specifically including:
the extracted spatial features are spatial feature D11 corresponding to the approximate parameter D1, spatial feature W11 corresponding to the detail parameter W1, spatial feature W21 corresponding to the detail parameter W2, spatial feature W31 corresponding to the detail parameter W3, and spatial feature W41 corresponding to the detail parameter W4.
Further, the method further comprises the following steps: step 4, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3, and the detail parameter W4 into a dual-channel GRU network respectively to extract each set of time features, which specifically includes:
the extracted temporal features are a temporal feature D12 corresponding to the approximate parameter D1, a temporal feature W12 corresponding to the detail parameter W1, a temporal feature W22 corresponding to the detail parameter W2, a temporal feature W32 corresponding to the detail parameter W3, and a temporal feature W42 corresponding to the detail parameter W4.
Further, the method further comprises the following steps: the step of respectively inputting the fused spatial features of each group and the corresponding temporal features of each group into a component prediction model, specifically comprising the following steps:
the combination of the spatial feature D11 and the temporal feature D12, the combination of the spatial feature W11 and the temporal feature W12, the combination of the spatial feature W21 and the temporal feature W22, the combination of the spatial feature W31 and the temporal feature W32, and the combination of the spatial feature W41 and the temporal feature W42 are input into the component prediction model, respectively.
Further, the method further comprises the following steps: the component prediction model and the knowledge graph model are both completed through pre-training.
Further, the method further comprises the following steps: step 11, comparing the first similarity with a first threshold after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result, and the method specifically comprises the following steps:
the first similarity and the third similarity are weighted and summed to form a fifth similarity, and the second similarity and the fourth similarity are weighted and summed to form a sixth similarity;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is larger than the second threshold value, the fault of the current electric energy meter is indicated;
when the fifth similarity is larger than the first threshold value and the sixth similarity is smaller than the second threshold value, the current electric energy meter is indicated to have no fault;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is smaller than the second threshold value, indicating that the current electric energy meter has fault possibility;
and when the fifth similarity is larger than the first threshold value and the sixth similarity is larger than the second threshold value, indicating that the current electric energy meter has fault possibility.
Further, the method further comprises the following steps: in the step 12, after effectively fusing the first fault prediction result and the second fault prediction result to obtain a final fault prediction result of the electric energy meter, the method further includes:
and carrying out layered and hierarchical alarm according to the superimposed final fault prediction result of the electric energy meter.
In a second aspect, the present invention also provides a system for predicting fault fusion of an electric energy meter, the system comprising:
the acquisition module is used for acquiring an original operation data sequence of the electric energy meter and carrying out normalization operation;
the DWT module is used for carrying out multi-stage DWT decomposition on the normalized data sequence, and decomposing the normalized data sequence into an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
the first extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model to extract each group of spatial features;
the second extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
the component prediction module is used for respectively inputting the component prediction model after fusing each group of spatial features and each corresponding group of time features to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
a reconstruction module, configured to reconstruct the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), where Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
the first prediction module is used for inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter and taking the reconstruction vector Y as a first fault prediction result;
a third extraction module for directly inputting the GCN model to extract the space class information aiming at the normalized data sequence, and converting the space class information into the space feature vector x 1 ；
A fourth extraction module for directly inputting a dual-channel GRU network to extract time sequence information aiming at the normalized data sequence, and converting the time sequence information into a time feature vector x 2 ；
The setting calculation module is used for setting the space vector set under the normal operation condition of the electric energy meter as GTA1 and the space vector set under the operation fault condition of the electric energy meter as GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarity as a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
the second prediction module is used for comparing the first similarity with a first threshold value after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold value after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
and the prediction fusion module is used for effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
The beneficial effects are that:
1. the invention realizes the fault fusion prediction of the electric energy meter based on the combination of the DWT, the GCN model, the double-channel GRU network and the knowledge graph model, and the depth characteristics of the operation data sequence of the electric energy meter are excavated, so that the characteristics excavation and the fault prediction method can achieve the aim of high speed and high efficiency, and the accuracy of fault prediction in the field of electric energy meter equipment is greatly improved.
2. According to the invention, the fault prediction is analyzed through the similarity of the spatial feature vector and the time feature vector respectively calculated and the calculated combination, so that the traditional direct fusion of the spatial feature and the time feature is avoided, and the similarity calculation of the spatial feature and the time feature of an independent class is effectively utilized, thereby realizing the prediction of the fault and greatly improving the accuracy of the fault prediction.
Drawings
Fig. 1 is a schematic flow chart of a fault fusion prediction method of an electric energy meter.
Fig. 2 is a schematic diagram of DWT discrete wavelet transform.
Fig. 3 is a schematic structural diagram of a GRU network.
Fig. 4 is a schematic structural diagram of the GCN model.
Detailed Description
The present application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the invention are shown in the drawings.
As shown in fig. 1 to 4, the present embodiment provides a method for predicting fault fusion of an electric energy meter, which includes:
step 1, acquiring an original operation data sequence of an electric energy meter, and carrying out normalization operation;
specifically, preprocessing operation is performed on the acquired data, including: and obtaining various data such as output voltage, output current, output power, calculated electric energy, wiring mode, communication protocol, year and the like of the current electric energy meter, and performing preprocessing operations such as data smoothing, outlier deletion, normalization and the like. The relevant dataset tables are shown in table 1;
TABLE 1
Step 2, performing multi-stage DWT decomposition on the normalized data sequence, wherein the multi-stage DWT decomposition is performed on the normalized data sequence to obtain an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
specifically, wavelet decomposition is carried out on the normalized data sequence, so that wavelet coefficients of different scales are obtained, wherein the wavelet coefficients comprise an approximation parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4; thereby reducing the non-stationarity of the data.
Specifically, DWT decomposes into discrete wavelet transforms. As shown in fig. 2, the h low frequency filtering may remove a high frequency portion in the data signal to obtain an approximate portion (low frequency component); g, high-frequency filtering can filter out a low-frequency part in the data signal to obtain a detail component (high-frequency component); for example, the input signal x is passed through G1, H1 filters to produce an approximation parameter D1, a detail parameter W1.
Step 3, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model respectively to extract each group of spatial features;
specifically, the GCN model has obvious advantages for extracting spatial features, which are the spatial feature D11 corresponding to the approximate parameter D1, the spatial feature W11 corresponding to the detail parameter W1, the spatial feature W21 corresponding to the detail parameter W2, the spatial feature W31 corresponding to the detail parameter W3, and the spatial feature W41 corresponding to the detail parameter W4.
Specifically, the GCN model is a graph roll-up network.
Step 4, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network respectively to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
specifically, the GRU network has obvious advantages for extracting time characteristics, and has wide application in the field of time sequence data prediction. The extracted temporal features are a temporal feature D12 corresponding to the approximate parameter D1, a temporal feature W12 corresponding to the detail parameter W1, a temporal feature W22 corresponding to the detail parameter W2, a temporal feature W32 corresponding to the detail parameter W3, and a temporal feature W42 corresponding to the detail parameter W4.
The GCN model and the dual-channel GRU network are used for simultaneously extracting the characteristics, so that more obvious space-time characteristics can be mined, and the characteristic extraction efficiency is improved.
Step 5, after fusing each group of spatial features and each corresponding group of time features, respectively inputting a component prediction model to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
specifically, wavelet coefficients of each layer are predicted by a pre-trained component prediction model.
Step 6, reconstructing the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), wherein Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
specifically, the reconstruction aims at constructing a feature vector related to the operation data of the electric energy meter according to the predicted wavelet coefficients of each layer, and taking the feature vector as the input of a subsequent analysis model. Specifically, the reconstruction method comprises the modes of splicing, overlapping, combining and the like of the data.
Step 7, inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter, and taking the reconstruction vector Y as a first fault prediction result;
specifically, the trained knowledge-graph model includes two nodes: reconstructing the vector nodes and the fault result nodes to form a fault mapping relation table.
The specific training method of the knowledge graph model comprises the following steps: taking the reconstruction vector in the training set as the input of the knowledge-graph analysis model, taking the fault prediction grade as the output of the knowledge-graph analysis model, and carrying out iterative training on the knowledge-graph analysis model to obtain a fault-type knowledge-graph analysis model;
the knowledge graph model structure comprises two nodes, namely a reconstruction vector node and a fault level node.
Step 8, for the normalized data sequence in step 1, directly inputting a GCN model to extract space class information, and converting the space class information into a space feature vector x 1 ；
Specifically, the spatial feature vector x 1 = (x 11, x12, x13, x14, x15, x16, x 17), wherein x11, x12, x13, x14, x15, x16, x17 are output characteristics obtained by preprocessing output voltage, output current, output power, calculation electric energy, wiring mode, communication protocol and annual input GCN model respectively.
Step 9, for the normalized data sequence in step 1, directly inputting a dual-channel GRU network to extract time sequence information, and converting the time sequence information into a time feature vector x 2 ；
Specifically, a temporal feature vector x 2 = (x 21, x22, x23, x24, x25, x26, x 27), wherein x21, x22, x23, x24, x25, x26, x27 are output characteristics obtained after the preprocessed output voltage, output current, output power, calculated electric energy, wiring mode, communication protocol and age are input into the dual-channel GRU network respectively.
Step 10, the space vector set under the normal operation condition of the electric energy meter is GTA1, and the space vector set under the operation fault condition of the electric energy meter is GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarity as a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
step 11, comparing the first similarity with a first threshold after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
specifically, the weighted sum of the first similarity and the third similarity is a fifth similarity, and the weighted sum of the second similarity and the fourth similarity is a sixth similarity;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is larger than the second threshold value, the fault of the current electric energy meter is indicated;
when the fifth similarity is larger than the first threshold value and the sixth similarity is smaller than the second threshold value, the current electric energy meter is indicated to have no fault;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is smaller than the second threshold value, indicating that the current electric energy meter has fault possibility;
and when the fifth similarity is larger than the first threshold value and the sixth similarity is larger than the second threshold value, indicating that the current electric energy meter has fault possibility.
Specifically, the first threshold and the second threshold can be set according to the actual working condition and the own needs, and the first threshold and the second threshold both have set initial values, and the initial values can be changed, so that the ductility of fault prediction is improved.
And step 12, effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
Specifically, hierarchical alarm is carried out according to the final fault prediction result of the electric energy meter after superposition.
Specifically, the decision of the final result can be realized according to a decision tree mode, specifically, the first fault prediction result and the second fault prediction result are input into the decision tree to obtain the final electric energy meter fault prediction result.
In an alternative embodiment, the method further comprises: the DWT discrete wavelet transform uses a Mallat fast algorithm. The algorithm uses wavelet filters, each time filtering gets a low frequency component and a high frequency component.
In an alternative embodiment, the method further comprises: the step of respectively inputting the fused spatial features of each group and the corresponding temporal features of each group into a component prediction model, specifically comprising the following steps:
the combination of the spatial feature D11 and the temporal feature D12, the combination of the spatial feature W11 and the temporal feature W12, the combination of the spatial feature W21 and the temporal feature W22, the combination of the spatial feature W31 and the temporal feature W32, and the combination of the spatial feature W41 and the temporal feature W42 are input into the component prediction model, respectively.
In an alternative embodiment, the method further comprises: the component prediction model and the knowledge graph model are both completed through pre-training.
For evaluating the performance of the method provided by the application, average absolute error MAE and average absolute percentage error MAPE are selected as evaluation indexes of a model, and are shown as follows:and selecting CNN convolutional neural network and the method provided by the application to conduct prediction result comparison analysis, wherein the comparison result is shown in table 2,
TABLE 2
Based on the same inventive concept, the present embodiment provides an electric energy meter fault fusion prediction system, which includes:
the acquisition module is used for acquiring an original operation data sequence of the electric energy meter and carrying out normalization operation;
the DWT module is used for carrying out multi-stage DWT decomposition on the normalized data sequence, and decomposing the normalized data sequence into an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
the first extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model to extract each group of spatial features;
the second extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
the component prediction module is used for respectively inputting the component prediction model after fusing each group of spatial features and each corresponding group of time features to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
a reconstruction module, configured to reconstruct the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), where Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
the first prediction module is used for inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter and taking the reconstruction vector Y as a first fault prediction result;
a third extraction module for directly inputting the GCN model to extract the space class information aiming at the normalized data sequence, and converting the space class information into the space feature vector x 1 ；
A fourth extraction module for directly inputting a dual-channel GRU network to extract time sequence information aiming at the normalized data sequence, and converting the time sequence information into a time feature vector x 2 ；
The setting calculation module is used for setting the space vector set under the normal operation condition of the electric energy meter as GTA1 and the space vector set under the operation fault condition of the electric energy meter as GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarityAs a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
the second prediction module is used for comparing the first similarity with a first threshold value after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold value after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
and the prediction fusion module is used for effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
Claims (10)
1. The utility model provides an electric energy meter fault fusion prediction method which is characterized in that the method comprises the following steps:
step 1, acquiring an original operation data sequence of an electric energy meter, and carrying out normalization operation;
step 2, performing multi-stage DWT decomposition on the normalized data sequence, wherein the multi-stage DWT decomposition is performed on the normalized data sequence to obtain an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
step 3, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model respectively to extract each group of spatial features;
step 4, inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network respectively to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
step 5, after fusing each group of spatial features and each corresponding group of time features, respectively inputting a component prediction model to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
step 6, reconstructing the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), wherein Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
step 7, inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter, and taking the reconstruction vector Y as a first fault prediction result;
step 8, for the normalized data sequence in step 1, directly inputting a GCN model to extract space class information, and converting the space class information into a space feature vector x 1 ；
Step 9, for the normalized data sequence in step 1, directly inputting a dual-channel GRU network to extract time sequence information, and converting the time sequence information into a time feature vector x 2 ；
Step 10, the space vector set under the normal operation condition of the electric energy meter is GTA1, and the space vector set under the operation fault condition of the electric energy meter is GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarity as a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
step 11, comparing the first similarity with a first threshold after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
and step 12, effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
2. The method according to claim 1, wherein the step 1 is to collect an original operation data sequence of the electric energy meter and perform a normalization operation, and specifically includes:
the method comprises the steps of obtaining output voltage, output current and output power of a current electric energy meter, calculating electric energy, wiring modes, communication protocols and years, and performing data smoothing, outlier deletion and normalization operation.
3. The method as recited in claim 1, further comprising:
the DWT discrete wavelet transform uses a Mallat fast algorithm.
4. The method according to claim 1, wherein the step 3 of inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model to extract each set of spatial features comprises:
the extracted spatial features are spatial feature D11 corresponding to the approximate parameter D1, spatial feature W11 corresponding to the detail parameter W1, spatial feature W21 corresponding to the detail parameter W2, spatial feature W31 corresponding to the detail parameter W3, and spatial feature W41 corresponding to the detail parameter W4.
5. The method according to claim 4, wherein the step 4 of inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network to extract each set of time features comprises:
the extracted temporal features are a temporal feature D12 corresponding to the approximate parameter D1, a temporal feature W12 corresponding to the detail parameter W1, a temporal feature W22 corresponding to the detail parameter W2, a temporal feature W32 corresponding to the detail parameter W3, and a temporal feature W42 corresponding to the detail parameter W4.
6. The method according to claim 5, wherein the merging each set of spatial features and the corresponding each set of temporal features and then inputting the merged spatial features and the corresponding sets of temporal features into the component prediction model respectively, specifically comprises:
the combination of the spatial feature D11 and the temporal feature D12, the combination of the spatial feature W11 and the temporal feature W12, the combination of the spatial feature W21 and the temporal feature W22, the combination of the spatial feature W31 and the temporal feature W32, and the combination of the spatial feature W41 and the temporal feature W42 are input into the component prediction model, respectively.
7. The method as recited in claim 1, further comprising: the component prediction model and the knowledge graph model are both completed through pre-training.
8. The method according to claim 1, wherein the step 11 is performed by comparing the first similarity with a first threshold value after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold value after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result, and the method specifically comprises the following steps:
the first similarity and the third similarity are weighted and summed to form a fifth similarity, and the second similarity and the fourth similarity are weighted and summed to form a sixth similarity;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is larger than the second threshold value, the fault of the current electric energy meter is indicated;
when the fifth similarity is larger than the first threshold value and the sixth similarity is smaller than the second threshold value, the current electric energy meter is indicated to have no fault;
when the fifth similarity is smaller than the first threshold value and the sixth similarity is smaller than the second threshold value, indicating that the current electric energy meter has fault possibility;
and when the fifth similarity is larger than the first threshold value and the sixth similarity is larger than the second threshold value, indicating that the current electric energy meter has fault possibility.
9. The method according to claim 1, wherein, after the step 12 of effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result, further comprises:
and carrying out layered and hierarchical alarm according to the superimposed final fault prediction result of the electric energy meter.
10. An electric energy meter fault fusion prediction system, characterized in that the system comprises:
the acquisition module is used for acquiring an original operation data sequence of the electric energy meter and carrying out normalization operation;
the DWT module is used for carrying out multi-stage DWT decomposition on the normalized data sequence, and decomposing the normalized data sequence into an approximate parameter D1, a detail parameter W2, a detail parameter W3 and a detail parameter W4;
the first extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a GCN model to extract each group of spatial features;
the second extraction module is used for respectively inputting the approximate parameter D1, the detail parameter W2, the detail parameter W3 and the detail parameter W4 into a dual-channel GRU network to extract each group of time characteristics; the GCN model and the dual-channel GRU network are used for extracting features in a parallel mode;
the component prediction module is used for respectively inputting the component prediction model after fusing each group of spatial features and each corresponding group of time features to obtain a D1 component prediction result, a W2 component prediction result, a W3 component prediction result and a W4 component prediction result;
a reconstruction module, configured to reconstruct the D1 component prediction result, the W2 component prediction result, the W3 component prediction result, and the W4 component prediction result to obtain a reconstruction vector y= (Y1, Y2, Y3, Y4, Y5), where Y1 represents the D1 component prediction result, Y2 represents the W1 component prediction result, Y3 represents the W2 component prediction result, Y4 represents the W3 component prediction result, and Y5 represents the W4 component prediction result;
the first prediction module is used for inputting the reconstruction vector Y into a knowledge graph model to predict the abnormal working condition of the current electric energy meter and taking the reconstruction vector Y as a first fault prediction result;
a third extraction module for directly inputting the GCN model to extract the space class information aiming at the normalized data sequence, and converting the space class information into the space feature vector x 1 ；
A fourth extraction module for directly inputting a dual-channel GRU network to extract time sequence information aiming at the normalized data sequence, and converting the time sequence information into a time feature vector x 2 ；
The setting calculation module is used for setting the space vector set under the normal operation condition of the electric energy meter as GTA1 and the space vector set under the operation fault condition of the electric energy meter as GTA2; the time vector set under the condition of normal operation of the electric energy meter is GTB1, and the time vector set under the condition of operation failure of the electric energy meter is GTB2;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA1 i Similarity of (3):taking the maximum similarity as a first similarity; />Representing the 2 norms of the vectors;
calculating the spatial feature vector x 1 And each feature vector x in vector set GTA2 j Similarity of (3):taking the maximum similarity as a second similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB1 h Similarity of (3):taking the maximum similarity as a third similarity;
calculating the time feature vector x 2 And each feature vector x in vector set GTB2 a Similarity of (3):taking the maximum similarity as a fourth similarity;
the second prediction module is used for comparing the first similarity with a first threshold value after weighted summation of the first similarity and a third similarity, and comparing the second similarity with a second threshold value after weighted summation of the second similarity and a fourth similarity; the two comparison results are synthesized to obtain the abnormal working condition of the electric energy meter and serve as a second fault prediction result;
and the prediction fusion module is used for effectively fusing the first fault prediction result and the second fault prediction result to obtain a final electric energy meter fault prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311724429.0A CN117421699B (en) | 2023-12-15 | 2023-12-15 | Electric energy meter fault fusion prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311724429.0A CN117421699B (en) | 2023-12-15 | 2023-12-15 | Electric energy meter fault fusion prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117421699A true CN117421699A (en) | 2024-01-19 |
| CN117421699B CN117421699B (en) | 2024-03-01 |
Family
ID=89523405
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311724429.0A Active CN117421699B (en) | 2023-12-15 | 2023-12-15 | Electric energy meter fault fusion prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117421699B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170308427A1 (en) * | 2016-04-26 | 2017-10-26 | Nec Laboratories America, Inc. | System Fault Diagnosis via Efficient Temporal and Dynamic Historical Fingerprint Retrieval |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| CN113627190A (en) * | 2021-08-27 | 2021-11-09 | 上海复深蓝软件股份有限公司 | Visualized data conversion method and device, computer equipment and storage medium |
| CN116483789A (en) * | 2023-04-28 | 2023-07-25 | 广东电网有限责任公司 | Distributed photovoltaic data double compression method and device based on carrier communication |
| CN116522761A (en) * | 2023-04-12 | 2023-08-01 | 广西大学 | Multi-network combined multi-region time-frequency air-wind power output prediction method |
-
2023
- 2023-12-15 CN CN202311724429.0A patent/CN117421699B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170308427A1 (en) * | 2016-04-26 | 2017-10-26 | Nec Laboratories America, Inc. | System Fault Diagnosis via Efficient Temporal and Dynamic Historical Fingerprint Retrieval |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| CN113627190A (en) * | 2021-08-27 | 2021-11-09 | 上海复深蓝软件股份有限公司 | Visualized data conversion method and device, computer equipment and storage medium |
| CN116522761A (en) * | 2023-04-12 | 2023-08-01 | 广西大学 | Multi-network combined multi-region time-frequency air-wind power output prediction method |
| CN116483789A (en) * | 2023-04-28 | 2023-07-25 | 广东电网有限责任公司 | Distributed photovoltaic data double compression method and device based on carrier communication |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117421699B (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
| CN110995475A (en) | Power communication network fault detection method based on transfer learning | |
| CN108022014B (en) | Power system load prediction method and system | |
| Gao et al. | Power quality disturbance classification under noisy conditions using adaptive wavelet threshold and DBN-ELM hybrid model | |
| CN112729381B (en) | Fault diagnosis method of high-voltage circuit breaker based on neural network | |
| CN111753968A (en) | Non-invasive load monitoring intelligent ammeter and electricity decomposition method | |
| CN112364291B (en) | Empirical mode decomposition method and device for pre-filtering extreme point optimization set | |
| CN111426905B (en) | Power distribution network common bus transformation relation abnormity diagnosis method, device and system | |
| CN112668611B (en) | Kmeans and CEEMD-PE-LSTM-based short-term photovoltaic power generation power prediction method | |
| CN112215405A (en) | Non-invasive type residential electricity load decomposition method based on DANN domain adaptive learning | |
| Kezunovic | Translational knowledge: From collecting data to making decisions in a smart grid | |
| CN113949069B (en) | Method and system for determining transient voltage stability of high-proportion new energy power system | |
| CN117421699B (en) | Electric energy meter fault fusion prediction method and system | |
| CN113092899B (en) | Transformer electrical fault identification method, system, terminal and readable storage medium | |
| CN115564075B (en) | Main and auxiliary integrated fault collaborative diagnosis method and system for urban power grid | |
| CN111521905A (en) | Power grid fault diagnosis method based on intelligent fusion of multi-dimensional data fault degrees | |
| CN116027158A (en) | High-voltage cable partial discharge fault prediction method and system | |
| CN115456013A (en) | Wind turbine generator converter power module fault diagnosis method based on operation data | |
| CN114186590A (en) | Power distribution network single-phase earth fault identification method based on wavelet and deep learning | |
| CN111611747A (en) | Online state estimation method and device for hybrid energy storage system | |
| Ding et al. | A Novel Partial Discharge Detection Algorithm in Power Transmission Lines Based on Deep Learning | |
| Wang et al. | An Automatic Identification Framework for Complex Power Quality Disturbances Based on Ensemble CNN | |
| CN116975611B (en) | High-frequency load data generation method and system based on diffusion model ODE form | |
| Kim et al. | Discrete wavelet transform-based characteristic analysis and SOH diagnosis for a Li-Ion cell | |
| CN116776228B (en) | Power grid time sequence data decoupling self-supervision pre-training method and system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |