CN111382897A - Transformer area low-voltage trip prediction method and device, computer equipment and storage medium - Google Patents
Transformer area low-voltage trip prediction method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN111382897A CN111382897A CN201911020923.2A CN201911020923A CN111382897A CN 111382897 A CN111382897 A CN 111382897A CN 201911020923 A CN201911020923 A CN 201911020923A CN 111382897 A CN111382897 A CN 111382897A
- Authority
- CN
- China
- Prior art keywords
- data
- training data
- low
- sample
- voltage
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000012549 training Methods 0.000 claims abstract description 179
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 95
- 238000012360 testing method Methods 0.000 claims abstract description 50
- 238000012545 processing Methods 0.000 claims abstract description 37
- 238000007781 pre-processing Methods 0.000 claims abstract description 15
- 238000012937 correction Methods 0.000 claims description 19
- 238000005457 optimization Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 15
- 230000002159 abnormal effect Effects 0.000 claims description 13
- 238000003672 processing method Methods 0.000 claims description 7
- 238000000926 separation method Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 description 28
- 238000011156 evaluation Methods 0.000 description 16
- 238000006243 chemical reaction Methods 0.000 description 15
- 230000000694 effects Effects 0.000 description 13
- 238000003066 decision tree Methods 0.000 description 10
- 230000009466 transformation Effects 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 7
- 238000010276 construction Methods 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 7
- 238000012423 maintenance Methods 0.000 description 6
- 230000003068 static effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000005611 electricity Effects 0.000 description 4
- 238000005192 partition Methods 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 238000000638 solvent extraction Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000002354 daily effect Effects 0.000 description 3
- 230000000670 limiting effect Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011157 data evaluation Methods 0.000 description 2
- 230000003203 everyday effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000013506 data mapping Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000003090 exacerbative effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 230000007774 longterm Effects 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
- 238000010606 normalization Methods 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 239000011541 reaction mixture Substances 0.000 description 1
- 230000008263 repair mechanism Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a prediction method, a prediction device, computer equipment and a storage medium for low-voltage tripping in a transformer area, wherein the method comprises the following steps: acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data into training data and testing data; the training data can be synthesized into new sample training data by adopting an isolated algorithm and an SMOTE-NC combined data processing mode; obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; constructing a trip prediction model according to the optimized training data; substituting the test data into the trip prediction model to obtain the probability of the low-voltage trip fault of the transformer area so as to realize accurate prediction of the low-voltage trip fault of the transformer area; preprocessing a transformer in the transformer area according to the low-voltage tripping probability; and the hidden trouble of the distribution transformer fault is found in time, the power failure risk is prevented, and important decision support is provided for improving the power supply reliability.
Description
Technical Field
The invention relates to the technical field of power systems, in particular to a transformer area low-voltage trip early warning method and device, computer equipment and a storage medium.
Background
Most regional distribution networks have the characteristics of high load density, concentrated power consumption, high power supply requirement and the like, equipment in part of urban areas has long operating years, distribution networks are weak, and the load-to-power supply faces significant challenges. Especially in high-temperature periods in summer, the use of high-power electrical appliances causes the electrical load to greatly rise, partial distribution and transformation capacity cannot meet the increasing demand of customer electricity consumption, and the problems of unstable voltage, power failure and the like occur in a transformer area. Frequent power failure caused by low-voltage tripping faults of a distribution network area is a main reason for causing power supply customer service complaints. At present, the management and control measures for low-voltage tripping mainly comprise emergency repair mechanisms such as switch replacement, low-voltage load regulation, capacity increase transformation, public transformation, special repair and the like and long-term solution means, and the processing mode mainly comprises post-processing and lacks of pre-judging work. Therefore, the research of the low-voltage tripping early warning model of the distribution network area is developed, the hidden trouble of the distribution transformer fault can be found in time, the power failure risk is prevented, and important decision support is provided for improving the power supply reliability.
In order to meet the research requirement of early warning of distribution transformer fault hidden danger in power distribution network planning and operation, scholars at home and abroad do a great deal of work on the prediction and analysis of distribution transformer operation states. Some researches locate a risk area by taking distribution transformer overload prediction as an entry point, such as distribution transformer overload correlation analysis and advance prediction based on a statistical method, and distribution transformer overload prediction based on machine learning algorithms such as logistic regression and BP (back propagation) neural network. Although distribution transformer heavy overload is an important cause for low-voltage tripping, part of low-voltage tripping faults are caused by operation management factors such as uneven load distribution among branches and unbalanced three-phase current, and equipment factors such as switch line aging and switch setting value setting problems. Therefore, the low-voltage trip risk areas are positioned and divided based on the heavy overload prediction methods, and a great problem exists in the accuracy rate.
Disclosure of Invention
Therefore, it is necessary to provide a transformer area low-voltage trip early warning method, device, computer equipment and storage medium for the problem of low transformer area low-voltage trip prediction accuracy.
A transformer district low-voltage tripping prediction method comprises the following steps: acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data into training data and testing data; according to an isolated forest algorithm, removing a few types of sample data in the training data to obtain the removed few types of sample data; according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data; obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; constructing a tripping prediction model according to the optimized training data; substituting the test data into the tripping prediction model to obtain the low-voltage tripping fault probability of the transformer area; and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
According to the station area low-voltage trip prediction method, the characteristic variable data influencing the station area low-voltage trip fault are divided into training data and testing data, the training data are used for constructing a low-voltage trip prediction model, the training data are processed by adopting an isolated algorithm and a SMOTE-NC combined data processing mode, the synthesized new sample training data are not prone to deviating from the geometric space of a class sample set, the problem of data distribution marginalization is solved, the accuracy of the construction of the trip prediction model is improved, the testing data are used for calculating the low-voltage trip fault probability, so that the station area low-voltage trip fault can be accurately predicted, a worker can conveniently preprocess a transformer of the station area according to the prediction result, distribution transformer fault hidden dangers can be found in time, power failure risks are prevented, and important decision support is provided for improving the power supply reliability.
In one embodiment, the station area low voltage trip prediction device comprises: the acquisition module is used for acquiring characteristic variable data influencing the low-voltage trip fault of the transformer area; the dividing module is used for dividing the characteristic variable data into training data and testing data; the first forest algorithm module is used for removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data; the second algorithm module is used for performing oversampling processing on the removed minority class of sample data according to the SMOTE-NC algorithm to obtain synthesized minority class of sample data; the optimization module is used for obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; the trip prediction model building module is used for building a trip prediction model according to the optimized training data; the prediction module is used for substituting the test data into the trip prediction model to obtain the distribution room low-voltage trip fault probability; and the preprocessing module is used for preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
In one embodiment, a computer device comprises a memory storing a computer program and a processor implementing the steps of the method of any of the above embodiments when the processor executes the computer program.
In one of the embodiments, a computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of any of the above embodiments.
Drawings
Fig. 1 is a schematic flow chart of a method for predicting a low-voltage trip fault in a distribution room according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of the iTree partitioning principle in the isolated forest algorithm according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a principle of an iForest-SMOTE-NC data processing method in an embodiment of the present invention;
fig. 4 is a flow chart of a prediction model for low-voltage tripping in a transformer area according to an embodiment of the present invention;
FIG. 5 is a graph comparing ROC-AUC evaluation of the prediction effects of different combinations of K and R parameters in a validation set in an embodiment of the present invention;
FIG. 6 is a graph comparing ROC curves and AUC for different algorithms in accordance with an embodiment of the present invention;
FIG. 7 is a graph comparing ROC curves and AUC for different algorithms in accordance with an embodiment of the present invention;
FIG. 8 is a diagram illustrating an importance ranking of feature variables in accordance with an embodiment of the present invention;
FIG. 9a is a graph of a confusion matrix at a risk probability threshold of 30% according to an embodiment of the invention;
FIG. 9b is a graph of a confusion matrix at a risk probability threshold of 50% according to an embodiment of the invention;
FIG. 9c is a graph of a confusion matrix at a risk probability threshold of 70% according to an embodiment of the invention;
fig. 10 is an internal structural view of a computer device in one embodiment of the present invention.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
For example, a zone low voltage trip prediction method includes: acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data into training data and testing data; according to an isolated forest algorithm, removing a few types of sample data in the training data to obtain the removed few types of sample data; according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data; obtaining optimized training data according to the synthesized few types of sample data and most types of sample data in the training data; constructing a tripping prediction model according to the optimized training data; substituting the test data into the tripping prediction model to obtain the low-voltage tripping probability of the transformer area; and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
According to the station area low-voltage trip prediction method, the characteristic variable data influencing the station area low-voltage trip fault are divided into training data and testing data, the training data are used for constructing a low-voltage trip prediction model, the training data are processed by adopting an isolated algorithm and a SMOTE-NC combined data processing mode, the synthesized new sample training data are not prone to deviating from the geometric space of a class sample set, the problem of data distribution marginalization is solved, the accuracy of the construction of the trip prediction model is improved, the testing data are used for calculating the low-voltage trip fault probability, so that the station area low-voltage trip fault can be accurately predicted, a worker can conveniently preprocess a transformer of the station area according to the prediction result, distribution transformer fault hidden dangers can be found in time, power failure risks are prevented, and important decision support is provided for improving the power supply reliability.
In one embodiment, a transformer area low-voltage trip prediction method comprises the following steps:
and S110, acquiring characteristic variable data influencing the low-voltage tripping fault of the transformer area.
Specifically, the more the characteristic variable is obtained, the more factors are considered for the trip prediction model, the more accurate the low-voltage trip fault prediction is, for example: the characteristic variable data comprise time sequence characteristic variable data and static characteristic variable data, the time sequence characteristic variable data comprise distribution transformation daily load, weather conditions, holiday information and the like, and the characteristics are updated daily and mainly used for analyzing the change trend of the low-voltage tripping event along with the change of time and the power supply environment state. S120, dividing the characteristic variable data into training data and testing data; the static characteristic variable data comprise distribution transformer configuration information, switch configuration information, line power utilization properties, regional characteristics and the like, and are mainly used for mining and dividing low-voltage tripping event distribution rules under each static characteristic dimension.
And S130, removing the few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data.
Specifically, an isolated Forest algorithm (isolated Forest t), referred to as an iFores algorithm for short, is an extension of a decision tree algorithm based on an isolated partitioning mechanism. For a dataset, i.e. a few sample data types, X ═ X1,...,xN},x∈RpAnd the isolated forest adopts an integrated learning strategy to construct T binary trees named iTree, each tree extracts subsamples in X, and randomly selects a characteristic variable and a partition threshold value in a value range to recursively partition the subsample space until the depth of the tree reaches a set limit value or a leaf node only contains one data point and can not be continuously partitioned, so that the iTree construction is completed. And eliminating the data with high abnormal degree in the few types of sample data.
And S140, performing oversampling processing on the removed few types of sample data according to the SMOTE-NC algorithm to obtain synthesized few types of sample data.
Specifically, the SMOTE-NC (Synthetic minor Over-sampling Technique-nonsinical continuous) algorithm belongs to a method for processing a data imbalance problem from a data plane, and is suitable for an imbalance data set mixed with continuous numerical characteristics and nominal characteristics. The basic idea of the algorithm is to add a synthesized sample between two adjacent minority samples by a random interpolation method, so that a data set tends to be balanced, belongs to a method for processing the data imbalance problem on a data level, and is suitable for an imbalance data set mixed with continuous numerical features and nominal features. The basic idea of the SMOTE-NC algorithm is to add a synthesized sample between two adjacent minority samples by a random interpolation method, so that a data set tends to be balanced.
Specifically, as the selectable neighbors of a few types of samples are determined by the distribution of the samples, when the samples are located at the distribution edges of the types, new samples generated by the SMOTE-NC algorithm are also located at the category edges, so that the marginalization of data is intensified, the classification boundaries are fuzzy, and the classification difficulty is increased.
And S150, obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data.
Specifically, the optimized training data is a combination of the synthesized few types of sample data and the majority of sample data in the training data, and the majority of sample data in the training data is the majority of sample data in the original training data into which the feature variable data is divided. And combining the synthesized few types of sample data and the multiple types of sample data in the training data to obtain optimized training data, wherein the few types of sample data synthesized by adopting an isolated forest algorithm and an SMOTE-NE algorithm are not easy to deviate from the geometric space of the class sample set, so that the marginalization problem of data distribution is improved, and the integrity of the data is kept by combining the multiple types of sample data, so that a trip prediction model is accurately constructed subsequently.
And S160, constructing a tripping prediction model according to the optimized training data.
Specifically, the trip prediction model, namely the transformer area low-voltage trip fault prediction model, is a model for predicting the probability of transformer area low-voltage trip faults; it should be understood that, because characteristic variable data have more or less influence on the area low-voltage trip fault, that is, the trip prediction model is constructed according to the optimized training data, that is, the trip prediction model is constructed according to the influence degree of the optimized training data on the area low-voltage trip, for example, the trip prediction model is constructed by processing the optimized training data by using an XGBoost (eXtreme Gradient) algorithm; for another example, the optimized training data is processed by using a GBDT (Gradient Boosting Decision Tree) algorithm to construct the trip prediction model.
And S170, substituting the test data into the tripping prediction model to obtain the low-voltage tripping probability of the transformer area.
Specifically, the trip prediction model adopts characteristic variables as input quantities and the transformer area low-voltage trip fault probability as output quantities, namely the transformer area low-voltage trip fault probability is obtained through calculation by obtaining test characteristic variable data. It should be understood that the test data, i.e., the characteristic variable data of the month, may also be predictive data, i.e., the probability of low-voltage tripping at a certain landing zone needs to be predicted. The trip prediction model generally takes characteristic variable data as input quantity, the zone low-voltage trip probability as output quantity, and the zone low-voltage trip probability in the time period can be calculated by substituting test data into the trip prediction model.
And S180, preprocessing the transformer in the transformer area according to the low-voltage tripping probability.
Specifically, the possibility of low-voltage tripping in the transformer area is judged according to the calculated low-voltage tripping probability, for example, when the low-voltage tripping probability exceeds a preset threshold value, that is, a low-voltage tripping fault is likely to occur in the transformer area is predicted, potential transformer faults of the transformer area can be found in time, and a worker can preprocess the transformer in the transformer area, for example, maintain and overhaul the transformer, or perform load transfer, so as to avoid stopping power supply due to the occurrence of the low-voltage tripping fault, and improve the reliability of power supply.
According to the station area low-voltage trip prediction method, the characteristic variable data influencing the station area low-voltage trip fault are divided into training data and testing data, the training data are used for constructing a low-voltage trip prediction model, the training data are processed by adopting an isolated algorithm and a SMOTE-NC combined data processing mode, the synthesized new sample training data are not prone to deviating from the geometric space of a class sample set, the problem of data distribution marginalization is solved, the accuracy of the construction of the trip prediction model is improved, the testing data are used for calculating the low-voltage trip fault probability, so that the station area low-voltage trip fault can be accurately predicted, a worker can conveniently preprocess a transformer of the station area according to the prediction result, distribution transformer fault hidden dangers can be found in time, power failure risks are prevented, and important decision support is provided for improving the power supply reliability.
In one embodiment, the step of dividing the feature variable data into training data and test data includes: and dividing the characteristic variable data by adopting a time sequence-based sliding window method to obtain the training data and the test data. Specifically, the sliding Window method is a sliding Window (Moving Window) algorithm, which is similar to the Window hopping algorithm and controls the traffic volume by limiting the maximum number of cells that can be received in each time Window. In the sliding window algorithm, the time window is not a forward jump, but is slid forward every one cell time, and the length of the sliding is one cell time. Therefore, the characteristic variable data are divided by adopting a sliding window method, so that the time intervals of the divided data are equal, and the construction accuracy of the prediction model is improved.
In order to better identify the training data, in one embodiment, the step of removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data includes: classifying the training data according to the numerical value type to obtain classified training data; converting the classified training data to obtain converted training data; and according to an isolated forest algorithm, removing the few types of sample data in the converted training data to obtain the removed few types of sample data. It can be understood that the characteristic variables affecting the low-voltage trip fault of the transformer area can comprise a time characteristic variable and a static characteristic variable, the corresponding data of the characteristic variables can be numerical values and can also be characters, and therefore, the identification and calculation of the isolated forest algorithm are facilitated by converting training data in the characteristic variable data.
Further, in one embodiment, the classified training data includes: continuous numerical training data, discrete numerical training data and nominal training data; the step of converting the classified training data to obtain converted training data includes: and processing the continuous numerical training data by a standardized processing method, processing the discrete numerical training data by a box separation method, and converting the nominal training data by one-hot coding to obtain the converted training data. In one embodiment, the continuous numerical training data is processed by a standardized processing method to obtain first conversion data, the discrete numerical training data is processed by a binning method to obtain second conversion data, the nominal training data is converted by one-hot coding to obtain third conversion data, and the converted training data is obtained according to the first conversion data, the second conversion data and the third conversion data. Specifically, the continuous numerical training data is normalized by a normalization method to obtain normalized data having a mean value of 0 and a variance of 1, and the continuous numerical training data includes f in table 11To f9(ii) a Processing the discrete numerical training data by adopting a box separation method, namely integrating the discrete numerical training data by adopting the box separation method, and then coding the normalized boxes, wherein the discrete data training data comprises f in a table13To f15(ii) a Because there is no size relation between the nominal training data, the information contained in the nominal features can be reasonably represented by adopting a one-hot feature conversion mode, and the algorithm identification is convenient. Thus, different types of training data are processed by adopting corresponding conversion methodsAnd converting the data into corresponding numerical values so as to facilitate the subsequent algorithm to carry out recognition calculation on the training data.
Table 1 details and meanings of the station characteristic variables considered
Serial number | Feature name | Description of the features |
f1 | Load factor of distribution transformer | Apparent power/ |
f2 | Rate of current imbalance | ∣Iphase-Iaverage∣/Iaverage*100% |
f3 | Temperature of | Highest temperature in the daytime (. degree. C.) |
f4 | Number of distribution transformer users | Total number of low-voltage users in district |
f5 | Number of switching users | Shunt switch low-voltage user number (household) |
f6 | Operating life of distribution transformer | Distribution transformer time of operation (year) |
f7 | Historical trip times | Trip times in approximately three months |
f8 | Number of historical complaints | Number of complaints of power supply type customers in nearly three months |
f9 | Number of off-limits | One standard monthly overload number of days |
f10 | Heavy overload evaluation | 0: no heavy overload, 1: heavy overload (f 1)>80%) |
f11 | Evaluation of Current imbalance ratio | 0: qualified, 1: three-phase imbalance (f 2)>40%) |
f12 | Whether it is holiday | 0: non-holiday, 1: holiday |
f13 | Distribution capacity | 500VA, 630VA, 800VA and the like |
f14 | Shunt switch capacity | 250A, 315A, 400A, 630A, etc |
f15 | Shunt switch wire path | 95mm2、120mm2、150mm2Etc. of |
f16 | Weather type | Sunny, cloudy, rainy, rainstorm, etc |
f17 | Wind power class | No wind, light wind, strong wind, etc |
f18 | Regional characteristics | Cities and towns, urban districts, city centers, villages in cities and the like |
f19 | Electrical property of circuit | Residential, commercial, industrial, comprehensive electricity utilization, and the like |
f20 | Transformation information of distribution transformer switch | Capacity-increasing transformation, three-phase load adjustment, switch replacement and the like |
Note: the load ratio and the current unbalance ratio of the distribution transformer are maximum values in 96 points in a day, Iphase is phase current, and Iaverage is a three-phase current average value.
In one embodiment, the step of removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data includes: obtaining abnormal values of the few types of sample data according to an isolated forest algorithm; and removing the few types of sample data of which the abnormal value is greater than a preset threshold value to obtain the removed few types of sample data.
Specifically, the isolated forest algorithm is an extension of the decision tree algorithm based on an isolated partitioning mechanism. For dataset X ═ X1,...,xN},x∈RpAnd the isolated forest adopts an integrated learning strategy to construct T binary trees named iTree, each tree extracts subsamples in X, and randomly selects a characteristic variable and a partition threshold value in a value range to recursively partition the subsample space until the depth of the tree reaches a set limit value or a leaf node only contains one data point and can not be continuously partitioned, so that the iTree construction is completed. In iTree, the height of the tree between the partitioned leaf nodes to the root node of observation point x is defined as the path length h (x). The smaller the value of h (x) is, the more easily the data point x is isolated, the higher the degree of abnormality is, otherwise, the data is normal. Taking FIG. 2 as an example, the abnormal value xoIsolated after 3 divisions, and normal data xiThe point needs to be divided by 9 times, and in the corresponding iTree, the point xoPath length h (x) of split leaf nodeo) Smaller than other observational objects, earlier located and isolated by the iTree. To measure the degree of abnormality of a data point, the isolated forest algorithm defines the abnormality score of any data point x as:
in the formula (1), E (h (x)) is the average value of the path length h (x) of the data point x in the T iTrees; and c (n) is the path length average value of all data points when the sampling number of the subsamples is n.
c(n)=2H(n-1)-(2(n-1)/n) (3)
In formula (3), H (·) ═ ln (·) + γ, γ is an euler constant.
The more the iForest abnormal score s calculated by the method is close to 1, the observation point is isolated very early, and the abnormal degree is high; s is close to 0, which means that the data point is not easy to be isolated and the safety is high. In practical application, the proportion of the number of data points removed by the iForest in a data set is set, then points with higher abnormal degree are preferentially removed according to the sorting condition of abnormal values until the number of the data points reaches the corresponding number of the set proportion, and thus, the data with high abnormal degree in a small number of types of sample data is better removed by calculating the abnormal values of the small number of types of sample data.
In one embodiment, the step of performing oversampling processing on the removed minority class of sample data according to a SMOTE-NC algorithm to obtain a synthesized minority class of sample data includes: calculating the median of the standard deviations of all continuous numerical training data in the removed few types of sample data; taking the median as a penalty item of distance calculation, and calculating nominal training data by adopting an Euclidean distance calculation method to obtain neighbor sample data; synthesizing the continuous data training data by a random linear interpolation method of an SMOTE algorithm to obtain synthesized continuous numerical training data; selecting a mode value in each nominal type training data in the adjacent sample data to obtain synthesized nominal type training data; and combining the synthesized continuous data training data and the synthesized nominal type training data to obtain a few types of synthesized sample data.
In particular, SMOTE-NC belongs to a method for processing data imbalance problems from a data plane, and is suitable for an imbalance data set mixed with continuous numerical characteristics and nominal characteristics. The basic idea of the algorithm is to add a synthesized sample between two adjacent minority samples by a random interpolation method, so that the data set tends to be balanced, and the quality of training data is improved.
Specifically, for a few types of sample data, X ═ { X ═ X1,x2,...,xN},xi=(xi1,xi2,...,xim,...,xin)TFor the ith (i ═ 1, 2.., N) few class sample instances, xi1,xi2,...,ximIs xiM continuous numerical training data values of (1), xi(m+1),xi(m+2),...,xinIs xiN-m nominal type characteristic attributes. SMOTE-NC Synthesis of a New sample the procedure was as follows:
and (5) calculating a median. Calculating the median of the standard deviation of all the continuous numerical features in a minority of classes, and recording the median as Med:
Med=median(σ1,σ2,...,σm) (5)
in the formula (4), mukIs the average value of kth continuous numerical type characteristics of all few sample data in the set X.
And (4) calculating nearest neighbor. On the basis of an original Euclidean distance calculation method, considering the influence of nominal feature difference, adding Med in the formula (5) as a penalty term for distance calculation, and defining any minority class xiAnd xjThe distance between:
in the formula (6), n is xiAnd xjThe number of differences d between the nominal training data. It should be noted that the difference number of the nominal training data after one-hot transcoding is doubled, and n should be d/2 in this case.
Calculating minority class sample x according to equation (6)iK neighboring samples of (a), denoted xiNeighbor sample setk defaults to 5.
The synthesized sample was added. For synthetic sample xnewContinuous numerical feature fraction x'newSynthesizing by adopting a random linear interpolation method of the traditional SMOTE algorithm:
in the formula (7), xi=(xi1,xi2,...,xim)T,rand (0,1) represents a random number of the interval (0,1),is composed ofA random point in (c).
For nominal type characteristic resultant value x ″)newThen select xiNeighbor sample setThe mode value of each nominal feature in (1). Finally, the two parts are characterized to be synthesized into a value x'newAnd x ″)newAnd merging to obtain a few types of synthesized sample data, wherein the nominal type characteristic synthetic value is synthesized nominal type training data.
Specifically, as shown in part (a) of fig. 3, from a geometric perspective, since the distribution of a few classes of samples determines the selectable neighbors thereof, when a sample is at the distribution edge of a class, a new sample generated by the SMOTE-NC algorithm is also at the class edge, thereby exacerbating the marginalization of data, blurring the classification boundary, and increasing the classification difficulty. Aiming at the problem, the iForest-SMOTE-NC data processing method firstly detects the abnormal degree of the data points, eliminates outlier samples in the data points, and then carries out oversampling, thereby avoiding generating unqualified new samples. As shown in part (b) of fig. 3, the new sample synthesized by the combined data processing method of iForest and SMOTE-NC proposed in the present application is not easy to deviate from the geometric space of the class sample set, and the marginalization problem of data distribution is improved.
In order to further improve the accuracy of building the trip prediction model, in one embodiment, the step of building the trip prediction model according to the optimized training data includes: and processing the optimized training data by adopting an XGboost algorithm, and constructing the trip prediction model. Specifically, the XGBoost (eXtreme Gradient boost) algorithm is an integrated learning algorithm based on a Gradient boost theory, has good expansibility and high operational efficiency facing a large data set, is further improved in the aspects of loss function, regularization, parallelization and the like compared with the traditional Gradient boost Decision Tree algorithm (GBDT) and the XGBoost algorithm, and has more excellent classification performance so as to further improve the accuracy of constructing the trip prediction model.
Further, in one embodiment, the step of processing the optimized training data by using an XGBoost algorithm to construct the trip prediction model includes:
for the training data that includes N samples and M-dimensional features D { (x)i,yi)},i=1,2,...,N,xi∈RM,yi∈ R, final prediction value of XGboost algorithmCalculated by an integrated model formed by adding a plurality of classification regression decision tree functions,the expression of (a) is:k is the number of decision trees; f. ofk(xi) Calculating a score for the ith sample in the dataset for the kth CART; f is allThe CART function constitutes a function space.
Combining an objective function of model learning in the XGboost algorithm with a loss function and a regular term, wherein the regular term is used for controlling the complexity of the model, and the expression of the regular term isT and w represent the tree f, respectivelykThe number of middle leaf nodes and the leaf weight, and gamma and lambda are regular term coefficients.
Note the bookNewly adding a CART decision tree function f for the predicted value of the ith sample instance in the t-th iterationtFurther reducing the target function, expanding the target function into a second-order Taylor series form, and removing a constant term of the expanded target function; wherein the expanded objective function expression is:
wherein,the first derivative of the loss function l (·),the second derivative of the loss function l (·); i isj={i|q(xi) J is the set of all sample indices mapped to the jth leaf node;
obtaining an optimal tree structure according to a greedy algorithm;
acquiring the optimal weight of each leaf node of the expanded objective function and the corresponding optimal objective function value according to the optimal tree structure;
and constructing the trip prediction model according to the optimal weight and the optimal objective function value.
To facilitate understanding of the technical solution of the present embodiment, the following detailed explanation is made, specificallyFor optimized training data containing N samples and M-dimensional features, D { (x)i,yi)},i=1,2,...,N,xi∈RM,yi∈ R, final prediction value of XGboost algorithmThe integrated model formed by adding a plurality of Classification Regression Tree (CART) functions is calculated as follows:
in the formula (8), K is the number of decision trees; f. ofk(xi) Calculating a score for the ith sample in the dataset for the kth CART; f is a function space formed by all CART functions.
The target function of model learning in the XGboost algorithm combines a loss function and a regular term, the regular term is used for controlling the complexity of the model, overfitting of the model is avoided, and the expression of the regular term is as follows:
in the formula (9), T and w represent a tree fkThe number of middle leaf nodes and the leaf weight, and gamma and lambda are regular term coefficients.
Note the bookNewly adding a CART decision tree function f for the predicted value of the ith sample instance in the t-th iterationtThe objective function is further reduced, expanded into the form of a second order taylor series, and the constant term is removed:
in the formula (10), the compound represented by the formula (10),both are the first and second derivatives of the loss function l (·), respectively; i isj={i|q(xi) J is the set of all sample indices mapped to the jth leaf node.
Based on the above derivation, each leaf node has an optimal weight for a particular tree structureAnd the corresponding optimal objective function value Obj expression:
in the formula,obj may be a scoring function that measures the quality of the tree structure, with lower scores indicating better tree structure.
Determining the best tree structure is a core problem of XGBoost, however, enumerating all possible tree structures to find the optimal Obj score is difficult to achieve, which requires a large computational effort. Aiming at the problem, the XGboost adopts a greedy algorithm to search an optimal tree structure, a division point with the maximum gain is selected each time for splitting, and the gain expression is as follows:
in the formula (13), the reaction mixture is,a left sub-tree score representing a partitioning scheme,the right sub-tree score is represented,representing a non-split score and gamma representing a complexity penalty factor. When the tree reaches the depth limit or all nodes split scheme Gain < 0, the tree stops splitting.
In one embodiment, the step of dividing the feature variable data into training data and test data includes: dividing the characteristic variable data into training data, testing data and correcting data; after the step of constructing the trip prediction model according to the optimized training data, the method further comprises the following steps: and updating the prediction model according to the correction data. In one embodiment, the step of dividing the feature variable data into training data, test data, and correction data includes dividing the feature variable data by a time-sequential sliding window method to obtain the training data, the test data, and the correction data. In one embodiment, the step of updating the prediction model according to the correction data includes: and correcting the trip prediction model by adopting a Bayesian optimization algorithm according to the correction data. Specifically, the actually measured data of a certain area public transformer station in spring and summer season month, namely 5 months in 2018 are used as test data to verify the validity of the extracted model, the data of 3 months and 4 months in 2018 are used as correction data to evaluate the effect of the model, and the history data of the calendar 1 before the correction data is used as training data. Parameter optimization is a key link for improving the performance of the model, and the divided correction data mainly acts on model parameter tuning and optimization to ensure the generalization capability of the model. Due to the time sequence characteristics of the power load, irregular operation and maintenance transformation of the distribution transformer and other factors, on the data level, the time sequence characteristics and static characteristic data of the distribution transformer between adjacent months show higher similarity, and the low-voltage tripping event occurrence rules under all characteristic dimensions are closer. Therefore, in order to improve the prediction effect of the test data, the sliding window method uses the data of the last two months as the correction data for model parameter optimization, and the average value of the results of the two correction data evaluation indexes is taken during result evaluation.
It should be understood that it is a small probability event for a public transformer station area that a low voltage trip occurs, and that the daily electricity consumption data of the station area have a high similarity in a short period, in which case it is difficult to predict the specific occurrence date of the low voltage trip of the station area. For the particularity, in one embodiment of the application, a standard month is taken as a period, the trip probability of the transformer area is updated every day, the latest observation date is cut off, and the maximum value of the trip probability of the day in the standard month is selected as the current-month trip probability of the transformer area. When the low-voltage tripping probability is larger than a set threshold value, the district is considered to have a low-voltage tripping fault in the month.
The following is a specific embodiment of the present application, a zone low voltage trip prediction method, including:
acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data by adopting a time sequence-based sliding window method to obtain training data, test data and correction data; classifying the training data according to the numerical type to obtain classified training data, wherein the classified training data comprises continuous numerical training data, discrete numerical training data and nominal training data; carrying out standardized processing method processing on the continuous numerical training data to obtain first conversion data; processing the discrete numerical training data by adopting a box separation method to obtain second conversion data; converting the nominal type training data by adopting one-hot coding to obtain third conversion data; obtaining the converted training data according to the first conversion data, the second conversion data and the third conversion data; according to an isolated forest algorithm, removing few types of sample data in the converted training data to obtain the removed few types of sample data; according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data; obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; processing the optimized training data by adopting an XGboost algorithm, and constructing the trip prediction model; according to the correction data, correcting the trip prediction model by adopting a Bayesian optimization algorithm; substituting the test data into the tripping prediction model to obtain the low-voltage tripping fault probability of the transformer area; and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
In order to make those skilled in the art better understand the beneficial effects of the present application, the zone low voltage trip prediction method of the present application is incorporated into the practical application below.
Summer is the season with the highest low-voltage tripping fault occurrence frequency, and if distribution transformer tripping hidden dangers can be found in time in spring and summer alternation, operation and maintenance personnel can be helped to better prevent power failure risks in summer. In view of this, the present embodiment uses actual measurement data of a public transformer station area in a certain area in the month of the spring and summer season of 2018, that is, in the month of 5 of 2018, as test data to verify the effectiveness of the constructed trip prediction model. The flow of the low-voltage trip early warning model of the public transformer area is shown in fig. 4.
And carrying out data division by adopting a time sequence-based sliding window method. Taking fig. 4 as an example, when the data of 5 months is test data, the data of 3 months and 4 months are taken as correction data to evaluate the model effect, and the data of 1 year calendar history before the correction data is taken as training data. Parameter optimization is a key link for improving the performance of the model, and the divided correction data mainly acts on model parameter tuning and optimization to ensure the generalization capability of the model. Due to the time sequence characteristics of the power load, irregular operation and maintenance transformation of the distribution transformer and other factors, on the data level, the time sequence characteristics and static characteristic data of the distribution transformer between adjacent months show higher similarity, and the low-voltage tripping event occurrence rules under all characteristic dimensions are closer. Therefore, in order to improve the prediction effect of the test data, the sliding window method uses the data of two months as the correction data for model parameter optimization. In the result evaluation, the average of the two correction data evaluation index results is taken.
The dynamic outlier rejection and sample oversampling are the premise for improving the data quality of the training set. In order to further improve the model prediction effect, in this embodiment, the parameters of the iForest and SMOTE-NC algorithm in the data processing process are selected by using a grid search method. Meanwhile, the over-parameters of the trip prediction model, namely the XGBoost model, are optimized by using a bayesian optimization algorithm (bayesian optimization).
An ROC (Receiver Operating Characteristic) Curve and a PR (precision Recall) Curve are common model effect evaluation methods in classification problems, and both curves can be obtained through the same confusion matrix. In addition, the coverage area of the ROC curve is AUC (area Under the dark), the larger the AUC is, the better the classification performance of the model is, and the evaluation index is a more comprehensive evaluation index for measuring the quality of the classifier; the PR curve is quite sensitive to the data unbalance degree, the unbalanced data set classification result is more intuitively reflected, and similar to the ROC curve, the larger the area AUC covered by the PR curve is, the better the overall prediction performance of the model is. Both ROC-AUC and PR-AUC were chosen as measures of the predictive performance of the models herein.
TABLE 2 two-class confusion matrix
As shown in table 2, a confusion matrix of the two-class problem is listed for prediction visualization. In the threshold selecting and analyzing process, Recall (Recall), Precision (Precision) and F1 measurement (F1-measure) integrating the Recall and Precision are adopted as analysis indexes:
the data of the embodiment is derived from actually measured data of 581 test public transformer areas in a certain area, one data point is obtained every day, historical data of 15 months from 3 months in 2017 to 5 months in 2018 are collected, the total amount of the data is 265517, 1690 low-voltage tripping events occur in total, and the unbalance proportion of a data set is about 156: 1. as shown in fig. 4, the example takes 2018 month 5 as a sample of the test set, where 142 public transformer stations have had low voltage trip events.
In order to select the optimal parameters of the data processing link, firstly, an experiment is designed, all the parameters are combined for data processing, and the prediction result is compared and analyzed. The ratio of the minority sample to the majority sample after SMOTE-NC oversampling processing is R, and the ratio of the number of outliers expected to be removed by iForest in the minority set is K. The value ranges of the parameters R and K are set to be 0.1-1 and 0-0.3 respectively. The evaluation conditions of the check data prediction results under each parameter combination are shown in fig. 5 by taking the ROC-AUC value as an evaluation index and the XGBoost as a basic classifier.
As can be seen from FIG. 5, the data processing link has a significant effect on the model prediction effect. When the iForest parameter K is larger than 0, for example, the value is 0.05-0.1, the AUC value of the model is better than that when K is equal to 0 (the outlier is not cleared), and the effectiveness of the iForest algorithm in locating the outlier is verified. However, as K increases, the AUC evaluation value shows a trend of decreasing, which indicates that only a small number of outliers exist in a small number of samples, and the value of the parameter K is not too large, as can be seen from fig. 4, the effect is better when K is 0.05. Meanwhile, the sampling ratio R of the SMOTE-NC is increased from 0.1 to 0.4, the AUC evaluation is gradually improved, and the AUC is 0.757 when R is equal to 0.4, so that the optimal result is achieved. When R is close to 1, the number of the minority classes and the majority classes in the data set is close, the data set tends to be balanced, but the AUC evaluation is reduced to some extent, which shows that excessive synthetic samples are added in oversampling, so that the problems of information redundancy and overfitting are caused, and the model training effect is influenced. Therefore, according to the experimental comparison of the verification set, an iForest parameter K is selected to be 0.05, and an SMOTE-NC parameter R is selected to be 0.4.
In order to verify the effectiveness and the universality of the method on the low-voltage trip prediction problem, different classifiers are trained by an original training set and a training set optimized by iForest and SMOTE-NC respectively, and classification probability prediction is carried out. Fig. 6 and 7 show a prediction result ROC curve and a PR curve of a Random Forest (RF), a gradient boosting decision tree, and an XGBoost algorithm, respectively.
The evaluation results of ROC-AUC and PR-AUC of each model in FIG. 6 and FIG. 7 are analyzed, and it is found that:
(1) the prediction performance of the original RF, GBDT and XGboost algorithms is affected by unbalanced data, and the identification capability of the low-voltage tripping event is poor. Through the optimization of the iForest-SMOTE-NC algorithm on the training set, the prediction performance of each classifier is improved by different amplitudes, wherein the ROC-AUC of the XGboost algorithm is improved to 0.8 from 0.67, the PR-AUC is improved to 0.62 from 0.43, and the improvement amplitude is the maximum. The method shows that the iForest-SMOTE-NC method provided by the application can well process the unbalance problem of the low-voltage trip fault data set, and the effectiveness and the universality of the combined method are verified.
(2) Compared with an SMOTE-NC-XGboost model only subjected to oversampling processing, the iForest-SMOTE-NC-XGboost model is better in evaluation. The main reason is that the latter obtains a representative fault sample and then carries out oversampling on the basis of realizing outlier detection and separation, and can effectively avoid generating unqualified new samples. The training set optimized by the method is more beneficial to the XGboost model to learn the data mapping relation between the input characteristics and the low-voltage tripping event, and the prediction effect is improved.
(3) In the original training set and the optimized training set, the ROC-AUC and PR-AUC of the XGboost algorithm are superior to the prediction effect of the traditional RF and GBDT algorithms. The XGboost improves the loss function and the regularization term of the objective function, so that the model has good generalization capability while being fully trained. Meanwhile, the XGboost model is optimized in terms of the super-parameters through a Bayesian optimization algorithm, and the superiority of the prediction performance is further ensured.
The nodes of the trees in the XGboost model select the characteristic attribute with the largest gain each time to split, so that the times of taking a certain characteristic variable as the division attribute in all the trees can be used for judging the importance of the characteristic, and the more times of taking the characteristic variable as the node division attribute, the higher the importance of the characteristic on low-voltage trip classification prediction is. The relative importance of each characteristic variable to low voltage trip prediction can be derived, wherein the ordering of the most significant 15 characteristic variables is shown in fig. 8.
As can be seen from fig. 8, the distribution transformer load rate, the number of low-voltage users, and the weather temperature are the characteristics with higher contribution degree, because the distribution transformer and the switch heavy overload are the main reasons for triggering the low-voltage trip, and the above characteristics have linear correlation with the heavy overload. In addition, the current imbalance rate, the transformation information and other operation management factors and equipment factors also have higher characteristic importance. Considering that the transformation information, the electricity utilization property, the regional characteristics and other category type characteristic variables adopt a one-hot coding form, the characteristic values are sparsely dispersed, the contribution degree is lower than that of numerical type characteristics, but the model training is performed to a certain extent, the effectiveness of the characteristic conversion mode is verified on the side, and a plurality of influence factors of low-voltage tripping events are also shown.
In order to further clarify the predictive significance of the model, the risk level definition is performed on the low-voltage trip prediction probability, and each risk level is shown in table 3. Table 4 shows the low voltage trip prediction results of the 581 test stand zones on the 2018 month 5 test set. The confusion matrix pair at each risk level threshold is shown in fig. 8, and the recall, precision and F1 metric ratings corresponding to each threshold can be seen in table 5.
As can be seen from the analysis results of fig. 9a, 9b, 9c and table 5, as the threshold value increases, the recall ratio of the model gradually decreases and the precision ratio increases. For example, when the probability threshold is 70%, because the threshold is set to be high, when 109 low-voltage tripping faults are missed in a month, the recall ratio is only 23.24%, but 39 extremely high-risk areas are positioned by the model, 33 of the areas actually have the low-voltage tripping faults, the precision ratio reaches 91.76%, and the prediction result can provide important decision support for operation and maintenance; the situation is opposite when the threshold is 30%, the model recall ratio is relatively high due to the lower threshold setting, but the precision ratio is as low as 40.48%, and the accuracy rate cannot meet the service requirement easily; when the threshold is 50%, the precision and recall are between the two, and the F1 metric is the highest.
TABLE 3 Low Voltage trip Risk level definitions
TABLE 4 prediction probability of trip and risk class for each district
TABLE 5 comparison of the evaluation of the prediction effectiveness of the respective Risk level thresholds
In actual business, it is often difficult to consider precision and recall ratio, and for such a situation, a policy of "preferentially ensuring precision and raising recall ratio" is usually adopted, that is, the station area is technically modified according to a priority sequence from high risk level to low risk level. Therefore, the reasonable allocation of the rush-repair resources is realized, the operation and maintenance strategy is optimized, and the problem of excessive operation and maintenance is avoided.
In one embodiment, the area low-voltage trip prediction device is implemented by using the area low-voltage trip prediction method in any one of the above embodiments. In one embodiment, the platform area low-voltage trip prediction device comprises corresponding modules for realizing the steps of the platform area low-voltage trip prediction method. In one embodiment, the station area low voltage trip prediction device comprises: the trip prediction system comprises an acquisition module, a division module, a first algorithm module, a second algorithm module, an optimization module, a trip prediction model construction module, a prediction module and a preprocessing module; the acquisition module is used for acquiring characteristic variable data influencing the low-voltage tripping fault of the transformer area; the dividing module is used for dividing the characteristic variable data into training data and testing data; the isolated forest algorithm module is used for removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data; the SMOTE-NC algorithm module is used for performing oversampling processing on the removed minority class of sample data according to the SMOTE-NC algorithm to obtain synthesized minority class of sample data; the optimization module is used for obtaining optimized training data according to the synthesized few types of sample data and most types of sample data in the training data; the trip prediction model building module is used for building a trip prediction model according to the optimized training data; the prediction module is used for substituting the test data into the trip prediction model to obtain the distribution room low-voltage trip fault probability; and the preprocessing module is used for preprocessing the transformer of the transformer area according to the low-voltage tripping probability. The rest of the examples are analogized.
In one embodiment, a computer device is provided, the internal structure of which may be as shown in FIG. 10. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a zone low voltage trip prediction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 10 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device includes a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program: acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data into training data and testing data; according to an isolated forest algorithm, removing a few types of sample data in the training data to obtain the removed few types of sample data; according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data; obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; constructing a tripping prediction model according to the optimized training data; substituting the test data into the tripping prediction model to obtain the low-voltage tripping fault probability of the transformer area; and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
In one embodiment, the processor, when executing the computer program, implements the steps of the zone low voltage trip prediction method in any of the above embodiments.
In one embodiment, a computer readable storage medium having a computer program stored thereon, the computer program when executed by a processor implementing the steps of:
acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area; dividing the characteristic variable data into training data and testing data; according to an isolated forest algorithm, removing a few types of sample data in the training data to obtain the removed few types of sample data; according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data; obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data; constructing a tripping prediction model according to the optimized training data; substituting the test data into the tripping prediction model to obtain the low-voltage tripping fault probability of the transformer area; and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
In one embodiment, the computer program is executed by a processor to implement the steps of the zone low voltage trip prediction method described in any of the above embodiments.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A method for predicting low-voltage tripping in a transformer area is characterized by comprising the following steps:
acquiring characteristic variable data influencing a low-voltage tripping fault of a transformer area;
dividing the characteristic variable data into training data and testing data;
according to an isolated forest algorithm, removing a few types of sample data in the training data to obtain the removed few types of sample data;
according to the SMOTE-NC algorithm, conducting oversampling processing on the removed few types of sample data to obtain synthesized few types of sample data;
obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data;
constructing a tripping prediction model according to the optimized training data;
substituting the test data into the tripping prediction model to obtain the low-voltage tripping fault probability of the transformer area;
and preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
2. The prediction method for low-voltage trip in transformer area according to claim 1, wherein the step of removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data comprises:
classifying the training data according to the numerical value type to obtain classified training data;
converting the classified training data to obtain converted training data;
and according to an isolated forest algorithm, removing the few types of sample data in the converted training data to obtain the removed few types of sample data.
3. The zone low voltage trip prediction method of claim 2 wherein the classified training data comprises: continuous numerical training data, discrete numerical training data and nominal training data;
the step of converting the classified training data to obtain converted training data includes:
and processing the continuous numerical training data by a standardized processing method, processing the discrete numerical training data by a box separation method, and converting the nominal training data by one-hot coding to obtain the converted training data.
4. The prediction method for low-voltage trip in transformer area according to claim 1, wherein the step of removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data comprises:
obtaining abnormal values of the few types of sample data according to an isolated forest algorithm;
and removing the few types of sample data of which the abnormal value is greater than a preset threshold value to obtain the removed few types of sample data.
5. The zone low voltage trip prediction method according to claim 1, wherein the step of constructing a trip prediction model based on the optimized training data comprises:
and processing the optimized training data by adopting an XGboost algorithm, and constructing the trip prediction model.
6. The method of claim 1, wherein the step of dividing the characteristic variable data into training data and test data comprises:
dividing the characteristic variable data into training data, testing data and correcting data;
after the step of constructing the trip prediction model according to the optimized training data, the method further comprises the following steps:
and correcting the prediction model according to the correction data.
7. The method of claim 6, wherein the step of dividing the characteristic variable data into training data, test data and calibration data comprises:
and dividing the characteristic variable data by adopting a time sequence-based sliding window method to obtain the training data, the test data and the correction data.
8. A block low voltage trip prediction device, comprising:
the acquisition module is used for acquiring characteristic variable data influencing the low-voltage trip fault of the transformer area;
the dividing module is used for dividing the characteristic variable data into training data and testing data;
the first algorithm module is used for removing a few types of sample data in the training data according to an isolated forest algorithm to obtain the removed few types of sample data;
the second algorithm module is used for performing oversampling processing on the removed minority class of sample data according to the SMOTE-NC algorithm to obtain synthesized minority class of sample data;
the optimization module is used for obtaining optimized training data according to the synthesized few types of sample data and the most types of sample data in the training data;
the trip prediction model building module is used for building a trip prediction model according to the optimized training data;
the prediction module is used for substituting the test data into the trip prediction model to obtain the distribution room low-voltage trip fault probability;
and the preprocessing module is used for preprocessing the transformer of the transformer area according to the low-voltage tripping probability.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911020923.2A CN111382897A (en) | 2019-10-25 | 2019-10-25 | Transformer area low-voltage trip prediction method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911020923.2A CN111382897A (en) | 2019-10-25 | 2019-10-25 | Transformer area low-voltage trip prediction method and device, computer equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111382897A true CN111382897A (en) | 2020-07-07 |
Family
ID=71216946
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911020923.2A Pending CN111382897A (en) | 2019-10-25 | 2019-10-25 | Transformer area low-voltage trip prediction method and device, computer equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111382897A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112098733A (en) * | 2020-09-22 | 2020-12-18 | 北京环境特性研究所 | Electromagnetic scattering characteristic data interpolation generation method and device |
CN112101614A (en) * | 2020-08-07 | 2020-12-18 | 贵州电网有限责任公司 | Resampling-based distribution transformer overload prediction method |
CN112307003A (en) * | 2020-11-02 | 2021-02-02 | 合肥优尔电子科技有限公司 | Power grid data multidimensional auxiliary analysis method, system, terminal and readable storage medium |
CN112734115A (en) * | 2021-01-13 | 2021-04-30 | 国网山东省电力公司日照供电公司 | Big data-based client side trip risk pre-control method and system |
CN112966879A (en) * | 2021-04-02 | 2021-06-15 | 阳光电源股份有限公司 | Environmental test chamber fault prediction method and device, computer equipment and storage medium |
CN113205125A (en) * | 2021-04-27 | 2021-08-03 | 河海大学 | XGboost-based extra-high voltage converter valve operation state evaluation method |
US11099219B2 (en) * | 2018-03-26 | 2021-08-24 | Oracle International Corporation | Estimating the remaining useful life of a power transformer based on real-time sensor data and periodic dissolved gas analyses |
CN113673575A (en) * | 2021-07-26 | 2021-11-19 | 浙江大华技术股份有限公司 | Data synthesis method, training method of image processing model and related device |
CN115102211A (en) * | 2022-05-16 | 2022-09-23 | 国网浙江省电力有限公司金华供电公司 | Energy transmission node positioning method for converting station area electric energy into power |
CN115238779A (en) * | 2022-07-12 | 2022-10-25 | 中移互联网有限公司 | Anomaly detection method, device, equipment and medium for cloud disk |
CN115564577A (en) * | 2022-12-02 | 2023-01-03 | 成都新希望金融信息有限公司 | Abnormal user identification method and device, electronic equipment and storage medium |
CN115765135A (en) * | 2022-11-10 | 2023-03-07 | 大庆恒驰电气有限公司 | Intelligent UPS energy storage system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108228978A (en) * | 2017-12-15 | 2018-06-29 | 四川金网通电子科技有限公司 | With reference to the Xgboost Time Series Forecasting Methods of complementary set empirical mode decomposition |
CN109410089A (en) * | 2018-12-29 | 2019-03-01 | 广州供电局有限公司 | Low-voltage tripping and customer complaint prediction technique, device and storage medium |
CN110135614A (en) * | 2019-03-26 | 2019-08-16 | 广东工业大学 | It is a kind of to be tripped prediction technique based on rejecting outliers and the 10kV distribution low-voltage of sampling techniques |
CN209342847U (en) * | 2018-11-13 | 2019-09-03 | 国网湖北省电力有限公司 | Distribution line fault outage prediction meanss and system based on weather and XBoost algorithm |
-
2019
- 2019-10-25 CN CN201911020923.2A patent/CN111382897A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108228978A (en) * | 2017-12-15 | 2018-06-29 | 四川金网通电子科技有限公司 | With reference to the Xgboost Time Series Forecasting Methods of complementary set empirical mode decomposition |
CN209342847U (en) * | 2018-11-13 | 2019-09-03 | 国网湖北省电力有限公司 | Distribution line fault outage prediction meanss and system based on weather and XBoost algorithm |
CN109410089A (en) * | 2018-12-29 | 2019-03-01 | 广州供电局有限公司 | Low-voltage tripping and customer complaint prediction technique, device and storage medium |
CN110135614A (en) * | 2019-03-26 | 2019-08-16 | 广东工业大学 | It is a kind of to be tripped prediction technique based on rejecting outliers and the 10kV distribution low-voltage of sampling techniques |
Non-Patent Citations (1)
Title |
---|
赵洪山,闫西慧,王桂兰,尹相龙: "应用深度自编码网络和XGBoost的风电机组发电机故障诊断", 《电力系统自动化》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11099219B2 (en) * | 2018-03-26 | 2021-08-24 | Oracle International Corporation | Estimating the remaining useful life of a power transformer based on real-time sensor data and periodic dissolved gas analyses |
CN112101614A (en) * | 2020-08-07 | 2020-12-18 | 贵州电网有限责任公司 | Resampling-based distribution transformer overload prediction method |
CN112098733A (en) * | 2020-09-22 | 2020-12-18 | 北京环境特性研究所 | Electromagnetic scattering characteristic data interpolation generation method and device |
CN112307003A (en) * | 2020-11-02 | 2021-02-02 | 合肥优尔电子科技有限公司 | Power grid data multidimensional auxiliary analysis method, system, terminal and readable storage medium |
CN112307003B (en) * | 2020-11-02 | 2022-09-09 | 合肥优尔电子科技有限公司 | Power grid data multidimensional auxiliary analysis method, system, terminal and readable storage medium |
CN112734115A (en) * | 2021-01-13 | 2021-04-30 | 国网山东省电力公司日照供电公司 | Big data-based client side trip risk pre-control method and system |
CN112966879A (en) * | 2021-04-02 | 2021-06-15 | 阳光电源股份有限公司 | Environmental test chamber fault prediction method and device, computer equipment and storage medium |
CN113205125A (en) * | 2021-04-27 | 2021-08-03 | 河海大学 | XGboost-based extra-high voltage converter valve operation state evaluation method |
CN113673575A (en) * | 2021-07-26 | 2021-11-19 | 浙江大华技术股份有限公司 | Data synthesis method, training method of image processing model and related device |
CN115102211A (en) * | 2022-05-16 | 2022-09-23 | 国网浙江省电力有限公司金华供电公司 | Energy transmission node positioning method for converting station area electric energy into power |
CN115102211B (en) * | 2022-05-16 | 2024-08-06 | 国网浙江省电力有限公司金华供电公司 | Energy transmission node positioning method for electric energy transfer in transformer area |
CN115238779A (en) * | 2022-07-12 | 2022-10-25 | 中移互联网有限公司 | Anomaly detection method, device, equipment and medium for cloud disk |
CN115238779B (en) * | 2022-07-12 | 2023-09-19 | 中移互联网有限公司 | Cloud disk abnormality detection method, device, equipment and medium |
CN115765135A (en) * | 2022-11-10 | 2023-03-07 | 大庆恒驰电气有限公司 | Intelligent UPS energy storage system |
CN115765135B (en) * | 2022-11-10 | 2023-05-05 | 大庆恒驰电气有限公司 | Intelligent UPS energy storage system |
CN115564577A (en) * | 2022-12-02 | 2023-01-03 | 成都新希望金融信息有限公司 | Abnormal user identification method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111382897A (en) | Transformer area low-voltage trip prediction method and device, computer equipment and storage medium | |
Wang et al. | Research and application of a hybrid forecasting framework based on multi-objective optimization for electrical power system | |
WO2022135265A1 (en) | Failure warning and analysis method for reservoir dispatching rules under effects of climate change | |
CN107516170B (en) | Difference self-healing control method based on equipment failure probability and power grid operation risk | |
Sheng et al. | Short-term solar power forecasting based on weighted Gaussian process regression | |
CN110097297A (en) | A kind of various dimensions stealing situation Intellisense method, system, equipment and medium | |
US9727036B2 (en) | Operation plan creating method, computer product, and operation plan creating apparatus | |
CN110503256A (en) | Short-term load forecasting method and system based on big data technology | |
CN110097220B (en) | Method for predicting monthly electric quantity of wind power generation | |
CN111339491A (en) | Evaluation method for urban power distribution network transformation scheme | |
Bastos et al. | Machine learning-based prediction of distribution network voltage and sensor allocation | |
Sun et al. | An ensemble system to predict the spatiotemporal distribution of energy security weaknesses in transmission networks | |
Fonseca et al. | Unsupervised load shape clustering for urban building performance assessment | |
Hou et al. | Spatial distribution assessment of power outage under typhoon disasters | |
CN117674119A (en) | Power grid operation risk assessment method, device, computer equipment and storage medium | |
Broderick et al. | Accuracy of clustering as a method to group distribution feeders by PV hosting capacity | |
Sun et al. | A multi-model-integration-based prediction methodology for the spatiotemporal distribution of vulnerabilities in integrated energy systems under the multi-type, imbalanced, and dependent input data scenarios | |
CN107834563B (en) | Method and system for processing voltage sag | |
Huang et al. | An integrated risk assessment model for the multi-perspective vulnerability of distribution networks under multi-source heterogeneous data distributions | |
CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
CN116796906A (en) | Electric power distribution network investment prediction analysis system and method based on data fusion | |
Liu et al. | Neyman-Pearson Umbrella Algorithm-Based Static Voltage Stability Assessment With Misclassification Restriction: An Integrated Data-Driven Scheme | |
CN113837486B (en) | RNN-RBM-based distribution network feeder long-term load prediction method | |
Li et al. | Distribution transformer mid-term heavy load and overload pre-warning based on logistic regression | |
Dou et al. | Day-Ahead Correction of Numerical Weather Prediction Solar Irradiance Forecasts Based on Similar Day Analysis |
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 | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20200923 Address after: 510620 Tianhe District, Guangzhou, Tianhe South Road, No. two, No. 2, No. Applicant after: Guangzhou Power Supply Bureau of Guangdong Power Grid Co.,Ltd. Address before: 510620 Tianhe District, Guangzhou, Tianhe South Road, No. two, No. 2, No. Applicant before: GUANGZHOU POWER SUPPLY Co.,Ltd. |
|
TA01 | Transfer of patent application right | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200707 |
|
RJ01 | Rejection of invention patent application after publication |