CN117034169A - Power grid main transformer equipment abnormal state prediction method based on time sequence causality network - Google Patents
Power grid main transformer equipment abnormal state prediction method based on time sequence causality network Download PDFInfo
- Publication number
- CN117034169A CN117034169A CN202311017786.3A CN202311017786A CN117034169A CN 117034169 A CN117034169 A CN 117034169A CN 202311017786 A CN202311017786 A CN 202311017786A CN 117034169 A CN117034169 A CN 117034169A
- Authority
- CN
- China
- Prior art keywords
- main transformer
- data
- transformer equipment
- equipment
- power grid
- 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
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000001364 causal effect Effects 0.000 claims abstract description 52
- 238000001514 detection method Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 230000005856 abnormality Effects 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 21
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 12
- 230000010339 dilation Effects 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 9
- 238000012544 monitoring process Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000004587 chromatography analysis Methods 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 7
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Optimization (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Public Health (AREA)
- Human Resources & Organizations (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Water Supply & Treatment (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
Abstract
Compared with the prior art, the method for predicting the abnormal state of the power grid main transformer equipment based on the time sequence causal relationship network solves the defect that causal relationships among abnormal variables of the main transformer equipment are not explored. The invention comprises the following steps: acquiring and preprocessing operation data of main transformer equipment; constructing a main transformer equipment data set; constructing an abnormal state detection model of the main transformer equipment of the power grid; training an abnormal state detection model of the main transformer equipment of the power grid; acquiring real-time running data of main transformer equipment; and predicting the abnormal state of the main transformer equipment of the power grid. Based on the characteristics of the relation between the variables of the dynamic equipment operation and the change with time, the causal relation between the abnormal variables of the main transformer equipment is fully explored, more hidden causal relation is obtained from the dynamic behavior of the variables on time distribution based on a time sequence causal relation network method, and the reliability of the abnormal prediction of the main transformer equipment is improved.
Description
Technical Field
The invention relates to the technical field of operation monitoring of power grid main transformer equipment, in particular to a power grid main transformer equipment abnormal state prediction method based on a time sequence causality network.
Background
The transformer substation is an important hub of the power system, and the safe and stable operation of the transformer equipment is an important foundation for ensuring the reliable power supply of the power system. Device anomaly management is also one of the important contents of secure production management. The greatly increased equipment operation abnormality consumes a great deal of operation and maintenance force, so that the establishment of a scientific prediction method is particularly important for predicting the equipment abnormality.
The power grid equipment generates a large amount of abnormal result data and equipment operation monitoring data in the operation process. However, the problems of different operation conditions and different service lives of the transformer stations, different equipment configuration, different manufacturers and different models and the like exist, and the abnormal operation of equipment is easily missed due to a plurality of information sources and impurities. In recent years, machine learning, deep learning and the like are gradually applied to power grid data analysis, and many problems which are difficult to solve by the traditional method are solved. However, since substation operation data is affected by many factors, and changes continuously with time. The traditional method lacks of mining the causal relationship affecting the abnormal operation of the equipment, and the causal relationship plays an important role in revealing the dynamic relationship among variables and explaining the operation rule of the system. The equipment abnormality prediction method based on the time sequence causality network is researched based on multi-dimensional time sequence data such as power grid equipment on-line monitoring data, working condition data, state evaluation data, historical abnormality data and the like, so that equipment operation abnormality is predicted more accurately. The prediction result can assist the operation and inspection department to pay important attention to and maintain the equipment which is possibly abnormal.
Therefore, it is an urgent technical problem to be solved to study how to use the time sequence causality network to predict the abnormality of the equipment.
Disclosure of Invention
The invention aims to solve the defect that the causal relationship between abnormal variables of main transformer equipment is not explored in the prior art, and provides a method for predicting the abnormal state of power grid main transformer equipment based on a time sequence causal relationship network to solve the problems.
In order to achieve the above object, the technical scheme of the present invention is as follows:
a power grid main transformer equipment abnormal state prediction method based on a time sequence causality network comprises the following steps:
11 Main transformer equipment operation data acquisition and preprocessing: acquiring operation data of main transformer equipment, wherein the operation data comprises external data and data acquired by a sensor when the main transformer equipment operates, and performing data error correction preprocessing;
12 Main transformer device data set construction: constructing a two-dimensional matrix data set of a variable with a time sequence characteristic and a target variable based on the preprocessed main transformer equipment operation data, and taking the two-dimensional matrix data set as a main transformer equipment data set;
13 Building an abnormal state detection model of the power grid main transformer equipment: constructing an abnormal state detection model of the main transformer equipment of the power grid based on a causal convolution network;
14 Training of an abnormal state detection model of power grid main transformer equipment: inputting a main transformer equipment data set into a power grid main transformer equipment abnormal state detection model for training;
15 Main transformer equipment operation real-time data acquisition: acquiring external real-time data and real-time data acquired by a sensor when the main transformer equipment operates;
16 Prediction of abnormal state of power grid main transformer equipment: and inputting the real-time running data of the main transformer equipment into a trained abnormal state detection model of the power grid main transformer equipment to obtain a prediction result of the abnormal state of the power grid main transformer equipment.
The construction of the main transformer equipment data set comprises the following steps:
21 Main transformer equipment operation data acquisition comprises equipment ID, equipment name, equipment type, transformer substation voltage, equipment model, manufacturer, production date, operation date, weather temperature, weather description, state evaluation result, load degree index, iron core grounding current index and oil chromatography index, and whether abnormal data occur or not is taken as prediction target data;
22 Vectorizing the preprocessed main transformer equipment operation data, wherein numerical data are directly used, text variables are used after being quantized by adopting a TF-IDF representation method, and date data are used after calculating the difference value between the date data and the month of the input operation time;
23 Fusing the vectorized data into unified structured data to construct a two-dimensional matrix data set of variables with time sequence characteristics and target variables, wherein each row represents a multidimensional variable and target variable at a certain time, and each column represents a variable sequence;
24 For a time period in which no abnormality occurs, filling a row in the two-dimensional matrix with the month as a time window, wherein the variable data is an average value of month monitoring data, and the target variable is that no abnormality occurs.
The construction of the abnormal state detection model of the power grid main transformer equipment comprises the following steps:
31 Setting an abnormal state detection model of the main transformer equipment of the power network based on the causal convolution network,
setting the input of a causal convolution network as two-dimensional matrix data operated by main transformer equipment, and outputting a causal relation feature mapping layer after the capture view of the expanded historical information iterated to the time t after the encoder network is constructed by a causal expansion convolution structure;
32 For a certain device operation variable x= (X) of the input 1 ，x 2 ，...x T ) Y= (Y) 1 ，y 2 ，...y T ) For the device state value, x before time t is used 1 ，x 2 ，...x t-1 Previous device state y 1 ，y 2 ，...y t-1 Data input, constructing causal relation between network layers;
y t ＝C_Net(x 1 ，x 2 ，...x t ，y 1 ，y 2 ，...y t-1 )，
wherein c_net is a causal convolutional network;
33 Using a causal convolution in combination with a causal convolution to form a causal dilation convolution structure,
for a certain input data characteristic variable s (s epsilon (1, 2,..N), wherein N is the total number of characteristic variables), operating time sequence data of equipment with the time length of T, obtaining a characteristic mapping layer T(s) of s after the convolution filtering action of a convolution kernel with the size of k and the expansion coefficient of d in a network layer, and calculating the T(s) by a characteristic sequence element s and a convolution filter f: {0, 1..k-1 } performs a dilation convolution operation with the formula:
wherein X is an input long time sequence, s is different operation data characteristic variables in the input long time sequence, and F d S-d.i represents a sequence corresponding to an element in the convolution kernel as a convolution filter function of the expansion coefficient d;
34 Setting different expansion coefficients to obtain multi-scale information of data, wherein the expansion coefficient d increases exponentially with the depth of the network, and for an m-th layer network: d=2 m-1 ；
35 33) for N feature variables, a matrix of t×n constituted by the device operation time sequence data of length T, obtaining an output T being N feature mapping layers, i.e. the output of the encoder network: causal relationship characteristics between the N variables.
The training of the abnormal state detection model of the power grid main transformer equipment comprises the following steps:
41 Inputting two-dimensional matrix data of main transformer equipment operation except for a target variable at the current moment into an encoder network in a step of constructing an abnormal state detection model of the main transformer equipment of the power grid, and obtaining causal relationship characteristics among N variables;
42 Setting an abnormal state training model of the main transformer equipment of the power grid, adopting a time convolution network TCN combined with a residual block, and outputting the prediction probability of abnormality of the equipment at a certain moment;
421 Setting the convolutional layer of the TCN to use a causal dilation convolutional structure;
422 Setting a residual block of TCN, setting two layers of convolution and a ReLU nonlinear function in the residual block, normalizing weights of convolution kernels by weights, and adding Dropout after each convolution in the residual block to realize regularization;
43 The TCN carries out regression prediction on the causal relationship characteristics generated in the step 41) to obtain a predicted value, and the predicted value is compared with a true value of equipment abnormality occurrence;
model training is carried out in a supervised mode, the attenuation trend of error loss tends to be stable in the training process to serve as a mark for stopping model training, mean square error is adopted in calculation of error loss L, and the expression is as follows:
wherein n is the number of predicted data, y j For the true value of the occurrence of the j-th device anomaly,a predicted abnormal probability value of the j-th data output by the model;
44 Using Adam and back propagation algorithm to minimize the loss function, returning to the trained predictive model, status (t) being the abnormal condition of the model output at time t,
wherein,the state abnormal predicted value at the moment t is the state abnormal value, and m is the equipment abnormal state offline scoring;
when (when)The model output device state is abnormal when +.>And when the model output equipment state is normal.
Advantageous effects
Compared with the prior art, the method for predicting the abnormal state of the power grid main transformer equipment based on the time sequence causality network fully explores the causality among the abnormal variables of the main transformer equipment based on the characteristics of the relation among the variables which are changed along with time and are operated by the dynamic equipment, obtains more hidden causality from the dynamic behaviors of the variables on time distribution based on the method of the time sequence causality network, and improves the reliability of the abnormal prediction of the main transformer equipment.
The invention combines the expansion convolution and the causal convolution to form a causal expansion convolution structure, which can enlarge the receptive field of the model, so that the receptive field can capture the correlation characteristics between non-adjacent variables in the running data of the equipment, and the multi-scale information of the data can be obtained by setting different expansion coefficients, thereby being beneficial to enlarging the receptive field without losing the prediction precision.
The equipment abnormality prediction method based on the time sequence causality network can predict the future operation abnormality of the equipment in a certain time window. The prediction result can assist the operation and inspection department to pay attention to and maintain equipment which is possibly abnormal, so that loss caused by abnormal operation of the equipment is reduced.
Drawings
FIG. 1 is a process sequence diagram of the present invention;
FIG. 2 is a graph showing the comparison of the results of the state of a part of the equipment predicted by the model according to the present invention.
Detailed Description
For a further understanding and appreciation of the structural features and advantages achieved by the present invention, the following description is provided in connection with the accompanying drawings, which are presently preferred embodiments and are incorporated in the accompanying drawings, in which:
as shown in fig. 1, the method for predicting the abnormal state of the power grid main transformer equipment based on the time sequence causality network comprises the following steps:
firstly, acquiring and preprocessing operation data of main transformer equipment: and acquiring operation data of the main transformer equipment, wherein the operation data comprises external data and data acquired by a sensor when the main transformer equipment operates, and performing data error correction preprocessing.
Secondly, constructing a main transformer device data set: and constructing a two-dimensional matrix data set of the variable with the time sequence characteristic and the target variable based on the preprocessed main transformer equipment operation data, and taking the two-dimensional matrix data set as a main transformer equipment data set.
Variable data table for predicting main transformer equipment state obtained in table 1
(1) The main transformer equipment operation data acquisition comprises equipment ID, equipment name, equipment type, transformer substation voltage, equipment model, manufacturer, production date, operation date, weather temperature, weather description, state evaluation result, load degree index, iron core grounding current index and oil chromatography index, and whether abnormal data occur or not is taken as prediction target data.
(2) And carrying out vectorization processing on the preprocessed main transformer equipment operation data, wherein numerical data are directly used, text variables are used after being quantized by adopting a TF-IDF representation method, and the difference between the numerical data and the input operation time month is calculated for the date data.
(3) And fusing the vectorized data into unified structured data to construct a two-dimensional matrix data set of variables with time sequence characteristics and target variables, wherein each row represents a multidimensional variable and target variable at a certain time, and each column represents a variable sequence.
(4) And filling one row in the two-dimensional matrix by taking the month as a time window for a time period in which no abnormality occurs, wherein the variable data is an average value of month monitoring data, and the target variable is that no abnormality occurs.
Thirdly, constructing an abnormal state detection model of the power grid main transformer equipment: and constructing an abnormal state detection model of the main transformer equipment of the power network based on the causal convolution network. And constructing a main transformer equipment abnormality prediction model (a power grid main transformer equipment abnormal state detection model) based on a time sequence causality network, and improving the processing efficiency of the long-time span memory unit by applying the expansion causality convolution to realize the efficient characteristic extraction of equipment abnormal variable parameters. Multiple expansion convolutions are stacked along with the increase of convolution kernels or expansion coefficients, so that a network obtains a larger receptive field, the network can capture correlation characteristics between non-adjacent variables in equipment operation data, and multi-scale information of the data can be obtained by setting different expansion coefficients, thereby being beneficial to expanding the receptive field without losing prediction accuracy.
The method for constructing the abnormal state detection model of the power grid main transformer equipment comprises the following steps of:
(1) Setting an abnormal state detection model of the main transformer equipment of the power network based on a causal convolution network,
setting the input of a causal convolution network as two-dimensional matrix data operated by main transformer equipment, and outputting a causal relation feature mapping layer after the capture field of the expanded historical information iterated to the time t after the encoder network is constructed by a causal expansion convolution structure.
(2) For a certain device operation variable x= (X) of input 1 ，x 2 ，...x T ) Y= (Y) 1 ，y 2 ，...y T ) For the device state value, x before time t is used 1 ，x 2 ，...x t-1 Previous device state y 1 ，y 2 ，...y t-1 Data input, constructing causal relation between network layers;
y t ＝C_Net(x 1 ，x 2 ，...x t ，y 1 ，y 2 ，...y t-1 )，
where c_net is a causal convolutional network.
(3) A causal dilation convolution structure is formed using a combination of dilation convolutions and causal convolutions,
for a certain input data characteristic variable s (s epsilon (1, 2,..N), wherein N is the total number of characteristic variables), operating time sequence data of equipment with the time length of T, obtaining a characteristic mapping layer T(s) of s after the convolution filtering action of a convolution kernel with the size of k and the expansion coefficient of d in a network layer, and calculating the T(s) by a characteristic sequence element s and a convolution filter f: {0, 1..k-1 } performs a dilation convolution operation with the formula:
wherein X is an input long time sequence, s is different operation data characteristic variables in the input long time sequence, and F d S-d.i represents the sequence corresponding to the elements in the convolution kernel as a convolution filter function of the expansion coefficient d.
(4) Setting different expansion coefficients to obtain multi-scale information of data, wherein the expansion coefficient d increases exponentially with the depth of the network, and for an m-th layer network: d=2 m-1 。
(5) Performing the step (3) on the N feature variables, and obtaining a matrix of t×n composed of the device operation time sequence data with the time length of T, where the output T is N feature mapping layers, that is, the output of the encoder network: causal relationship characteristics between the N variables.
Fourthly, training an abnormal state detection model of the power grid main transformer equipment: and inputting the main transformer equipment data set into a power grid main transformer equipment abnormal state detection model for training. In order to solve the problem that the equipment state information is distorted after the multi-layer convolution operation, a residual block is introduced to enable a network to transmit the dependent information expression of the original data in a cross-layer mode, in addition, a ReLU activation function is added to better extract nonlinear characteristics, and a Dropout layer is added to prevent the model from being overfitted.
(1) And inputting the two-dimensional matrix data of the main transformer equipment operation except for the target variable at the current moment into an encoder network in the step of constructing the abnormal state detection model of the main transformer equipment of the power network, so as to obtain causal relationship characteristics among N variables.
(2) Setting an abnormal state training model of the main transformer equipment of the power grid, adopting a time convolution network TCN combined with a residual block, and outputting the prediction probability of abnormality of the equipment at a certain moment;
a1 Setting the convolutional layer of the TCN to use a causal dilation convolutional structure;
a2 Setting a residual block of TCN, setting two layers of convolution and a ReLU nonlinear function, normalizing weights of convolution kernels, and adding Dropout after each convolution in the residual block to realize regularization.
(3) The TCN carries out regression prediction on the causal relationship characteristics generated in the step (1) to obtain a predicted value, and the predicted value is compared with a true value of equipment abnormality;
model training is carried out in a supervised mode, the attenuation trend of error loss tends to be stable in the training process to serve as a mark for stopping model training, mean square error is adopted in calculation of error loss L, and the expression is as follows:
wherein n is the number of predicted data, y j For the true value of the occurrence of the j-th device anomaly,and predicting the abnormal probability value of the j-th data output by the model.
(4) Minimizing the loss function by using Adam and a back propagation algorithm, returning to a trained prediction model, wherein Status (t) is the abnormal state of the model output at the moment t,
wherein,the state abnormal predicted value at the moment t is the state abnormal value, and m is the equipment abnormal state offline scoring;
when (when)The model output device state is abnormal when +.>And when the model output equipment state is normal.
Fifthly, acquiring real-time data of operation of the main transformer equipment: external real-time data and real-time data acquired by a sensor when the main transformer equipment operates are acquired.
Sixth, predicting abnormal states of power grid main transformer equipment: and inputting the real-time running data of the main transformer equipment into a trained abnormal state detection model of the power grid main transformer equipment to obtain a prediction result of the abnormal state of the power grid main transformer equipment.
As shown in fig. 2, the model prediction of the present invention is a partial device state result. Predicting whether the equipment has defects in a certain time in the future has important significance for management and maintenance of the substation equipment and improvement of reliability of a substation system.
The effect of device anomaly prediction is performed in order to compare different methods. The experiment adopts a logistic regression method, a decision tree method, a random forest method and the method for predicting the equipment state. The experiment uses 80% of the data set as a training set, the rest 20% of the data set as a test set, and the training set and the test set adopted by all the methods are consistent. The results of the comparison of the accuracy, recall, and F1-score of the different methods are shown in Table 2.
Table 2 comparison of the method of the present invention with other methods
| Method | Accuracy rate of | Recall rate of recall | F1-score |
| Logistic regression | 0.85 | 0.44 | 0.58 |
| Decision tree | 0.89 | 0.63 | 0.73 |
| Random forest | 0.9 | 0.48 | 0.63 |
| The method of the invention | 0.93 | 0.62 | 0.74 |
As can be seen by comparison, the method of the invention achieves the best results in terms of both accuracy and F1-score, and the recall rate is only 0.01 less than the best 0.63.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made therein without departing from the spirit and scope of the invention, which is defined by the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (4)
1. The method for predicting the abnormal state of the power grid main transformer equipment based on the time sequence causality network is characterized by comprising the following steps of:
11 Main transformer equipment operation data acquisition and preprocessing: acquiring operation data of main transformer equipment, wherein the operation data comprises external data and data acquired by a sensor when the main transformer equipment operates, and performing data error correction preprocessing;
12 Main transformer device data set construction: constructing a two-dimensional matrix data set of a variable with a time sequence characteristic and a target variable based on the preprocessed main transformer equipment operation data, and taking the two-dimensional matrix data set as a main transformer equipment data set;
13 Building an abnormal state detection model of the power grid main transformer equipment: constructing an abnormal state detection model of the main transformer equipment of the power grid based on a causal convolution network;
14 Training of an abnormal state detection model of power grid main transformer equipment: inputting a main transformer equipment data set into a power grid main transformer equipment abnormal state detection model for training;
15 Main transformer equipment operation real-time data acquisition: acquiring external real-time data and real-time data acquired by a sensor when the main transformer equipment operates;
16 Prediction of abnormal state of power grid main transformer equipment: and inputting the real-time running data of the main transformer equipment into a trained abnormal state detection model of the power grid main transformer equipment to obtain a prediction result of the abnormal state of the power grid main transformer equipment.
2. The method for predicting abnormal states of power grid main transformer equipment based on time sequence causality network according to claim 1, wherein the construction of the main transformer equipment data set comprises the following steps:
21 Main transformer equipment operation data acquisition comprises equipment ID, equipment name, equipment type, transformer substation voltage, equipment model, manufacturer, production date, operation date, weather temperature, weather description, state evaluation result, load degree index, iron core grounding current index and oil chromatography index, and whether abnormal data occur or not is taken as prediction target data;
22 Vectorizing the preprocessed main transformer equipment operation data, wherein numerical data are directly used, text variables are used after being quantized by adopting a TF-IDF representation method, and date data are used after calculating the difference value between the date data and the month of the input operation time;
23 Fusing the vectorized data into unified structured data to construct a two-dimensional matrix data set of variables with time sequence characteristics and target variables, wherein each row represents a multidimensional variable and target variable at a certain time, and each column represents a variable sequence;
24 For a time period in which no abnormality occurs, filling a row in the two-dimensional matrix with the month as a time window, wherein the variable data is an average value of month monitoring data, and the target variable is that no abnormality occurs.
3. The method for predicting abnormal states of power grid main transformer equipment based on time sequence causality network according to claim 1, wherein the constructing the power grid main transformer equipment abnormal state detection model comprises the following steps:
31 Setting an abnormal state detection model of the main transformer equipment of the power network based on the causal convolution network,
setting the input of a causal convolution network as two-dimensional matrix data operated by main transformer equipment, and outputting a causal relation feature mapping layer after the capture view of the expanded historical information iterated to the time t after the encoder network is constructed by a causal expansion convolution structure;
32 For a certain device operation variable x= (X) of the input 1 ，x 2 ，...x T ) Y= (Y) 1 ，y 2 ，...y T ) For the device state value, x before time t is used 1 ，x 2 ，...x t-1 Previous device state y 1 ，y 2 ，...y t-1 Data input, constructing causal relation between network layers;
y t ＝C_Net(x 1 ，x 2 ，...x t ，y 1 ，y 2 ，...y t-1 )，
wherein c_net is a causal convolutional network;
33 Using a causal convolution in combination with a causal convolution to form a causal dilation convolution structure,
for a certain input data characteristic variable s (s epsilon (1, 2,..N), wherein N is the total number of characteristic variables), operating time sequence data of equipment with the time length of T, obtaining a characteristic mapping layer T(s) of s after the convolution filtering action of a convolution kernel with the size of k and the expansion coefficient of d in a network layer, and calculating the T(s) by a characteristic sequence element s and a convolution filter f: {0, 1..k-1 } performs a dilation convolution operation with the formula:
wherein X is an input long time sequence, s is different operation data characteristic variables in the input long time sequence, and F d S-d.i represents a sequence corresponding to an element in the convolution kernel as a convolution filter function of the expansion coefficient d;
34 Setting different expansion coefficients to obtain multi-scale information of data, wherein the expansion coefficient d increases exponentially with the depth of the network, and for an m-th layer network: d=2 m-1 ；
35 33) for N feature variables, a matrix of t×n constituted by the device operation time sequence data of length T, obtaining an output T being N feature mapping layers, i.e. the output of the encoder network: causal relationship characteristics between the N variables.
4. The method for predicting abnormal states of power grid main transformer equipment based on time sequence causality network according to claim 1, wherein the training of the power grid main transformer equipment abnormal state detection model comprises the following steps:
41 Inputting two-dimensional matrix data of main transformer equipment operation except for a target variable at the current moment into an encoder network in a step of constructing an abnormal state detection model of the main transformer equipment of the power grid, and obtaining causal relationship characteristics among N variables;
42 Setting an abnormal state training model of the main transformer equipment of the power grid, adopting a time convolution network TCN combined with a residual block, and outputting the prediction probability of abnormality of the equipment at a certain moment;
421 Setting the convolutional layer of the TCN to use a causal dilation convolutional structure;
422 Setting a residual block of TCN, setting two layers of convolution and a ReLU nonlinear function in the residual block, normalizing weights of convolution kernels by weights, and adding Dropout after each convolution in the residual block to realize regularization;
43 The TCN carries out regression prediction on the causal relationship characteristics generated in the step 41) to obtain a predicted value, and the predicted value is compared with a true value of equipment abnormality occurrence;
model training is carried out in a supervised mode, the attenuation trend of error loss tends to be stable in the training process to serve as a mark for stopping model training, mean square error is adopted in calculation of error loss L, and the expression is as follows:
wherein n is the number of predicted data, y j For the true value of the occurrence of the j-th device anomaly,a predicted abnormal probability value of the j-th data output by the model;
44 Using Adam and back propagation algorithm to minimize the loss function, returning to the trained predictive model, status (t) being the abnormal condition of the model output at time t,
wherein,the state abnormal predicted value at the moment t is the state abnormal value, and m is the equipment abnormal state offline scoring; when->The model output device state is abnormal when +.>And when the model output equipment state is normal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311017786.3A CN117034169A (en) | 2023-08-14 | 2023-08-14 | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311017786.3A CN117034169A (en) | 2023-08-14 | 2023-08-14 | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117034169A true CN117034169A (en) | 2023-11-10 |
Family
ID=88627628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311017786.3A Pending CN117034169A (en) | 2023-08-14 | 2023-08-14 | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117034169A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117472898A (en) * | 2023-12-26 | 2024-01-30 | 国网江西省电力有限公司电力科学研究院 | Fusion-based power distribution network abnormal data error correction method and system |
-
2023
- 2023-08-14 CN CN202311017786.3A patent/CN117034169A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117472898A (en) * | 2023-12-26 | 2024-01-30 | 国网江西省电力有限公司电力科学研究院 | Fusion-based power distribution network abnormal data error correction method and system |
| CN117472898B (en) * | 2023-12-26 | 2024-04-02 | 国网江西省电力有限公司电力科学研究院 | Fusion-based power distribution network abnormal data error correction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Deep-convolution-based LSTM network for remaining useful life prediction | |
| Chen et al. | Anomaly detection and critical SCADA parameters identification for wind turbines based on LSTM-AE neural network | |
| CN109726524B (en) | CNN and LSTM-based rolling bearing residual service life prediction method | |
| Qian et al. | Integrated data‐driven<? show [AQ ID= Q1]?> model‐based approach to condition monitoring of the wind turbine gearbox | |
| CN111340282B (en) | DA-TCN-based method and system for estimating residual service life of equipment | |
| He et al. | RUL prediction of wind turbine gearbox bearings based on self-calibration temporal convolutional network | |
| Caceres et al. | A probabilistic Bayesian recurrent neural network for remaining useful life prognostics considering epistemic and aleatory uncertainties | |
| Yan et al. | Big-data-driven based intelligent prognostics scheme in industry 4.0 environment | |
| CN113570138A (en) | Method and device for predicting residual service life of equipment of time convolution network | |
| CN113343581B (en) | Transformer fault diagnosis method based on graph Markov neural network | |
| CN117034169A (en) | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network | |
| CN114819315A (en) | Bearing degradation trend prediction method based on multi-parameter fusion health factor and time convolution neural network | |
| Cheng et al. | Predicting the remaining useful life of rolling element bearings using locally linear fusion regression | |
| Gong et al. | Remaining useful life prediction based on multisensor fusion and attention TCN-BiGRU model | |
| CN116737510A (en) | Data analysis-based intelligent keyboard monitoring method and system | |
| Zhu et al. | Res-HSA: Residual hybrid network with self-attention mechanism for RUL prediction of rotating machinery | |
| Huang et al. | Bayesian neural network based method of remaining useful life prediction and uncertainty quantification for aircraft engine | |
| Wen et al. | A new multi-sensor fusion with hybrid Convolutional Neural Network with Wiener model for remaining useful life estimation | |
| Wang et al. | Wind turbine fault detection and identification through self-attention-based mechanism embedded with a multivariable query pattern | |
| Jiang et al. | Measurement of health evolution tendency for aircraft engine using a data-driven method based on multi-scale series reconstruction and adaptive hybrid model | |
| Wu et al. | Real-time monitoring and diagnosis scheme for IoT-enabled devices using multivariate SPC techniques | |
| Wu et al. | Intelligent fault diagnosis of rolling bearings based on clustering algorithm of fast search and find of density peaks | |
| Xu et al. | A composite quantile regression long short-term memory network with group lasso for wind turbine anomaly detection | |
| Mateus et al. | Improved GRU prediction of paper pulp press variables using different pre-processing methods | |
| CN116677570A (en) | Fault early warning method and system based on cabin temperature monitoring of offshore wind turbine |
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 |