CN113837477B - Method, device and equipment for predicting power grid faults under typhoon disasters driven by data - Google Patents
Method, device and equipment for predicting power grid faults under typhoon disasters driven by data Download PDFInfo
- Publication number
- CN113837477B CN113837477B CN202111139000.6A CN202111139000A CN113837477B CN 113837477 B CN113837477 B CN 113837477B CN 202111139000 A CN202111139000 A CN 202111139000A CN 113837477 B CN113837477 B CN 113837477B
- Authority
- CN
- China
- Prior art keywords
- disaster
- data
- causing
- sample
- prediction model
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000009826 distribution Methods 0.000 claims abstract description 81
- 230000003068 static effect Effects 0.000 claims abstract description 53
- 230000007246 mechanism Effects 0.000 claims abstract description 21
- 230000015654 memory Effects 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 17
- 238000013507 mapping Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 48
- 238000012360 testing method Methods 0.000 claims description 47
- 238000004422 calculation algorithm Methods 0.000 claims description 43
- 230000006870 function Effects 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 238000011160 research Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 9
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 7
- 238000005259 measurement Methods 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 231100001261 hazardous Toxicity 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 3
- 239000000956 alloy Substances 0.000 claims 1
- 229910045601 alloy Inorganic materials 0.000 claims 1
- 230000009471 action Effects 0.000 abstract description 7
- 239000010410 layer Substances 0.000 description 22
- 238000010586 diagram Methods 0.000 description 9
- 230000006403 short-term memory Effects 0.000 description 9
- 238000004590 computer program Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 4
- 201000004569 Blindness Diseases 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000001965 increasing effect Effects 0.000 description 3
- 230000007787 long-term memory Effects 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000011229 interlayer Substances 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
- 230000003287 optical effect Effects 0.000 description 1
- 238000004540 process dynamic Methods 0.000 description 1
- 238000012958 reprocessing Methods 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000012795 verification 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
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method, a device and equipment for predicting power grid faults under a typhoon disaster driven by data, wherein the method comprises the following steps: constructing a disaster causing data set, balancing the disaster causing data set, constructing a double-channel prediction model and predicting by using the double-channel prediction model, classifying multiple influencing factors of the disaster receiving condition of the power distribution network under typhoon disasters into static data and dynamic data, extracting the characteristics of the static data by utilizing a feedforward neural network, extracting the characteristics of the dynamic data by utilizing a long-short-period memory network reinforced by a multi-head self-attention mechanism, finally fusing all the extracted characteristics by adopting a linear layer, and establishing the mapping relation between the multiple influencing factors and the disaster receiving condition of the power distribution network. Stability of the action of static data on the disaster-stricken condition of the power distribution network and time-varying and accumulating performance of the action of dynamic data on the disaster-stricken condition of the power distribution network are fully considered, and a power distribution network fault prediction model under typhoon disasters with higher accuracy and stronger interpretability is constructed.
Description
Technical Field
The invention belongs to the technical field of power grid fault prediction, and particularly relates to a method, a device and equipment for predicting power grid faults under a typhoon disaster with data double driving.
Background
Typhoons have a large influence range and a long duration, and the tropical cyclone ratio of typhoons and above strength is continuously increased along with the change of global climate in recent years, so that the normal operation of a power transmission and distribution network in coastal areas is threatened greatly. Compared with a transmission network, the power distribution network has more equipment, serious equipment aging problems and is more easily influenced by natural disasters such as typhoons. Therefore, a method for predicting the faults of the power distribution network under typhoon disasters needs to be researched aiming at the destructiveness of typhoons and the vulnerability of the power distribution network, and reliable prior information is provided for an elasticity enhancement strategy of the power distribution network.
The study on the power distribution network fault prediction method under typhoon disasters is mainly divided into a physical model based on a disaster causing mechanism and a data driving model based on historical data. The physical model research thought is to build a wind load model of a power distribution network line and a pole tower according to probability distribution of actual wind speed and design wind speed of the equipment, and correct the model by combining factors such as geographical environment where the equipment is located, service life of the equipment and the like, so that fault probability of the line and the pole tower under typhoon disasters is obtained. The research thought of the data driving model is generally to construct a data set containing disaster factors and network faults based on historical meteorological information, geographic information and power grid information, learn the data set through a machine learning model and establish corresponding mapping relations. Meanwhile, in consideration of the fact that the fault data of the distribution network under typhoon disasters contain a large number of samples with zero fault quantity, a machine learning model generates larger prediction deviation in the samples with non-zero fault quantity, the existing research generally adopts a synthetic minority over-sampling technology (SMOTE) to generate a minority sample balance data set, or adopts a cost sensitive learning method to assign different punishment coefficients to different categories, and the learning side weight of the model to the minority samples is improved.
Under the limitation of modeling complexity, the physical model is difficult to comprehensively and finely model the influence factors of the power distribution network equipment faults, so that certain prediction precision is lost. With the increasing perfection of data collection and management systems of electric power departments and meteorological departments, current research focuses on predicting power distribution network fault conditions under typhoon disasters through a data-driven model. However, the existing data driving model only considers the relation between each influence factor in each time section and the power distribution network fault, and does not consider the accumulation of the action of partial factors on the power distribution network fault. Meanwhile, the SMOTE algorithm adopted in the current research has certain blindness and randomness in the selection process of the sample synthesis object, the quality of the generated few types of samples is poor, and when penalty coefficients of all types are determined by the cost sensitive learning method, the parameters need to be repeatedly adjusted according to the model performance, and the adjustment direction is subjective.
Disclosure of Invention
The invention provides a method, a device and equipment for predicting power grid faults under a typhoon disaster by data double driving, which are used for improving the accuracy and the interpretability of the power distribution network fault prediction method under the typhoon disaster and enhancing the capability of the power distribution network for resisting the typhoon disaster.
In order to achieve the above purpose, the method for predicting the power distribution network faults under the typhoon disaster driven by data in double mode comprises the following steps:
step 2, carrying out equalization treatment on the disaster-causing data set;
step 3, extracting the characteristics of static data in disaster-causing data sets by utilizing a feedforward neural network, extracting the sequence characteristics of dynamic data in disaster-causing data sets by utilizing a long-period memory network and a multi-head self-attention mechanism, establishing a dual-channel prediction model of power grid faults under typhoon disasters, solving and optimizing model parameters based on the disaster-causing data sets after sample equalization treatment, and finally obtaining an optimized dual-channel prediction model; and evaluate its performance; if the performance meets the requirement, carrying out the step 4, otherwise, continuing to optimize;
and 4, collecting corresponding multi-element influence data of a predicted area under the future typhoon disaster, constructing a disaster-causing data set, and inputting the disaster-causing data set into the optimized double-channel prediction model in the step 3 to obtain a predicted value of the power grid fault condition of the research area under the future typhoon disaster.
Further, in step 1, the static data includes forest coverage, land type, maintenance degree of the power grid and population density, and the dynamic data includes distance between a typhoon center and an area center, center lowest air pressure of typhoons, near center maximum wind speed of typhoons, moving direction angle of typhoons, seven-level wind circle radius, average wind speed of a predicted area and precipitation of the predicted area.
Further, the process of step 2 is: dividing a minority sample set according to the distribution of disaster-causing data sets in a high-dimensional space by using a Borderline-SMOTE1 algorithm, and generating samples aiming at minority samples at decision boundaries after division; and then checking the difference of the data distribution of the training set and the testing set through a discrimination model, performing parameter tuning on the Borderline-SMOTE1 algorithm according to the difference, and finally applying the Borderline-SMOTE1 algorithm after parameter optimization to balance the disaster-causing data set.
Further, step 2 includes the steps of:
step 2.1, calculating m nearest neighbor samples of each mild failure sample by using a K nearest neighbor algorithm;
step 2.2, classifying the m nearest neighbor samples of the mild fault type sample into a safety type sample, a dangerous type sample and a noise type sample according to the duty ratio of the mild fault sample in the m nearest neighbor samples;
Step 2.3, for each hazardous class sample x i Selecting a required number of light fault samples from K nearest neighbor samples;
step 2.4, for each selected neighbor sample x j ' generating a slightly faulty new sample x using linear interpolation i,j ;
Step 2.5, adding the generated light fault new sample into an original disaster training set to obtain an updated disaster data set;
and 2.6, checking the updated disaster causing data set, if the updated disaster causing data set meets the requirements, performing step 3, and if the updated disaster causing data set does not meet the requirements, performing parameter adjustment on the Borderline-SMOTE1 algorithm until the disaster causing data set meets the requirements.
Further, step 2.6 includes the steps of:
step 2.6.1, randomly sampling the disaster causing training set to make the number of samples of the sampled training set equal to the number of samples of the disaster causing test set; then respectively setting labels of the training set sample and the test set sample to be 0 and 1, mixing to form a discrimination data set, and dividing the discrimination data set into a new training set and a new test set according to a proportion;
step 2.6.2, based on a new training set and a new testing set, using a cross entropy function as a loss function, obtaining the gradient of each parameter value of the judging model through an error back propagation method, and updating all parameters of the judging model through an Adam gradient descent algorithm to obtain the judging accuracy of the disaster-causing training set and the disaster-causing testing set;
2.6.3, using a discrimination model to distinguish sample distribution differences of a disaster causing training set and a disaster causing test set, and when the discrimination accuracy is higher than an accuracy threshold, adjusting parameters such as nearest neighbor sample number of a Borderline-SMOTE1 algorithm; and when the judging accuracy is lower than the accuracy threshold, executing the step 3.
Further, step 3 includes the steps of:
step 3.1, extracting static characteristics from static data based on a feedforward neural network; extracting dynamic characteristics from dynamic data based on a long-short-period memory network and a multi-head attention mechanism;
step 3.2, splicing the static features and the dynamic features, mapping the static features and the dynamic features into the prediction probability of each fault condition type of the power grid through a linear layer, and taking the prediction fault condition type with the maximum probability value corresponding to the disaster type as a sample to obtain a prediction model; using a cross entropy function as a loss function, and measuring the difference degree between the predicted value and the actual value; then obtaining a gradient value of each parameter in the cross entropy function pair model through an error back propagation algorithm; finally, updating the prediction model parameters by using a small batch Adam algorithm in combination with the learning rate, the batch size and the number of neurons of each layer;
And 3.3, taking the precision rate and the recall rate as basic index systems, introducing a macro average mechanism to comprehensively consider the performances of the prediction model in different types of sample sets in the disaster-causing test set, and evaluating the prediction model.
Further, step 3.3 includes the steps of:
step 3.3.1, counting whether each sample in the disaster-causing test set belongs to an actual value and a predicted value of the disaster-causing type according to the predicted value obtained after the disaster-causing test set is input into the prediction model, and forming three classification confusion moments;
step 3.3.2, obtaining a group of true positive TP corresponding to each confusion matrix according to matrix elements i False positive FP i True negative TN i And false negative FN i Thereby obtaining the corresponding precision P i Sum recall R i ;
Step 3.3.3 according to the precision P i Recall ratio R i And F1 measurement to obtain macro precision macro-P, macro recall ratio macro-R and macro F1 value macro-F1;
and 3.3.4, evaluating the performance of the power grid fault condition prediction model under typhoon disasters according to four indexes of a macro Cha Zhun rate, a macro recall rate, a macro F1 and an accuracy rate.
A grid fault prediction device under typhoon disasters, comprising:
the acquisition module is used for acquiring data and transmitting the acquired data to the calculation output module; the data comprise multi-element influence data of power grid faults under typhoon disasters, the sum of permanent tripping times of the power grid in the predicted area and real-time typhoon data;
And the calculation output module is used for training a prediction model according to the collected data set and outputting a power grid fault prediction value according to the prediction model and real-time typhoon data.
A computer device comprising a memory and a processor electrically connected, the memory having stored thereon a computing program executable on the processor, the processor implementing the steps of the method of any of claims 1-8 when the computing program is executed.
Compared with the prior art, the invention has at least the following beneficial technical effects:
according to the invention, multiple influencing factors of the power distribution network fault condition under typhoon disasters are classified into static data and dynamic data, the characteristics of the static data are extracted by utilizing a feedforward neural network, the characteristics of the dynamic data are extracted by utilizing a long-short-period memory network after strengthening of a multi-head self-attention mechanism, and finally, all the extracted characteristics are fused by adopting a linear layer, so that the mapping relation between the multiple influencing factors and the power distribution network fault condition is established. The two-channel prediction model constructed by the method fully considers the stability of static data on the disaster condition of the power distribution network and the time-varying and cumulative property of dynamic data on the fault condition of the power distribution network, and constructs the power distribution network fault prediction model under typhoon disasters with higher accuracy and stronger interpretability.
The Borderline-SMOTE1 algorithm used in the invention identifies samples at the decision boundary based on the K nearest neighbor algorithm, and uses the random linear interpolation to synthesize new samples, so that the defects of blindness and randomness in the sample generation process, subjectivity and complexity of a punishment coefficient determination mode and the like in the existing sample imbalance processing mode are overcome, the imbalance degree of a disaster-causing data set is effectively reduced, a good data basis is laid for training a power grid fault condition prediction model, the accuracy of a power distribution network fault prediction method under typhoon disasters is improved, and the capability of the power distribution network for resisting typhoon disasters is further enhanced.
Drawings
FIG. 1 is a schematic diagram of a disaster causing dataset;
FIG. 2 is a schematic diagram of classification of a light failure class sample of the Borderline-SMOTE1 algorithm;
FIG. 3 is a schematic diagram of a test sample distribution of a discriminant model;
FIG. 4 is a block diagram of the LSTM cell;
FIG. 5 is a network architecture diagram of a two-channel prediction model;
fig. 6 is a schematic block diagram of a power grid fault prediction device provided by the invention;
fig. 7 is a schematic structural diagram of a computer device according to the present invention.
Detailed Description
In order to make the purpose and technical scheme of the invention clearer and easier to understand. The present invention will now be described in further detail with reference to the drawings and examples, which are given for the purpose of illustration only and are not intended to limit the invention thereto.
In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, unless otherwise indicated, the meaning of "a plurality" is two or more. In the description of the present invention, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
Example 1
Referring to fig. 1, a method for predicting faults of a power distribution network under typhoon disasters based on static and dynamic data dual driving comprises four major parts, namely, constructing a disaster-causing data set, balancing the disaster-causing data set, constructing a dual-channel prediction model and predicting fault conditions of a regional power distribution network under future typhoon disasters.
step 2, aiming at the phenomenon of sample imbalance in the disaster causing data set, dividing a minority sample set according to the distribution of the disaster causing data set in a high-dimensional space by using a Borderline-SMOTE1 algorithm, and generating samples aiming at minority samples at a decision boundary after division; thirdly, checking the difference of data distribution of the training set and the testing set through a discrimination model, performing parameter tuning on the Borderline-SMOTE1 algorithm according to the difference, and finally applying the Borderline-SMOTE1 algorithm after parameter optimization to balance disaster-causing data sets;
And 3, extracting the characteristics of static data in the disaster-causing data set by utilizing a feedforward neural network, extracting the sequence characteristics of dynamic data in the disaster-causing data set by utilizing a long-short-term memory network (LSTM) and a multi-head self-attention mechanism, establishing a dual-channel prediction model of power distribution network faults under typhoon disasters, solving and optimizing model parameters based on the disaster-causing data set after sample equalization processing by combining a cross entropy loss function, an error back propagation method and the like, finally obtaining an optimized dual-channel prediction model, and evaluating the performance of the model. And (4) if the performance meets the requirement, carrying out step 4, otherwise, continuing to optimize.
And 4, collecting corresponding data of a certain research area under the future typhoon disaster, constructing a disaster-causing data set, and inputting the disaster-causing data set into the optimized double-channel prediction model in the step 3 to obtain a predicted value of the power distribution network fault condition of the research area under the future typhoon disaster.
The specific process of each step is as follows:
1. constructing disaster causing data sets
The disaster-causing data are selected from four angles of weather information, geographic information, power grid information and population information, are divided into static data and dynamic data according to time domain change conditions of the data during typhoon passing, jointly form a sample of a disaster-causing data set, and finally are combined with fault condition types of a power distribution network under typhoon disasters to form the disaster-causing data set.
In the time scale of typhoon passing, partial disaster-causing data basically does not change, and the influence on the fault condition of the power distribution network is stable. Therefore, the disaster causing data are classified into static data, including forest coverage rate, land type, maintenance degree of a power grid and population density. And partial disaster-causing data has larger change along with time, and has time variability and accumulation on the influence of the fault condition of the power distribution network. Such disaster causing data are classified into dynamic data including eight data of distance between typhoon center and regional center, center lowest air pressure of typhoon, near center maximum wind speed of typhoon, moving direction angle of typhoon, seven-level wind circle radius, average wind speed of regional and precipitation of regional. It should be noted that the static data is constituted by data of a single time section, and the dynamic data is constituted by sequential data of 48 consecutive hours.
Considering that typhoon disasters are accompanied by strong winds and heavy rainfall, power elements such as overhead lines, underground cables, towers and the like of a power distribution network can be damaged to a certain extent, therefore, the invention sums up permanent tripping times of the power distribution network for 24 hours under the typhoon disasters, considers that the regional power distribution network normally operates when the tripping times sum is 0, considers that the regional power distribution network slightly breaks down when the tripping times sum is 1 to 9, considers that the regional power distribution network severely breaks down when the tripping times sum is greater than 9, and takes three disaster-causing condition types of power distribution network fault conditions as labels of disaster-causing data sets.
2. In summary, the invention combines static and dynamic data to form a disaster-causing data set sample, takes the type of the fault condition of the power distribution network as a data set sample label, forms a final disaster-causing data set together, and divides the final disaster-causing data set into a disaster-causing training set and a disaster-causing test set according to the proportion of eight to two. A schematic diagram of the sample and sample tag is shown in FIG. 1, where f 1 、f 2 、f 3 And f 4 For static data, f ,i,j For dynamic data, i=5, 6, … … 12; j=1, 2, … …; the dynamic data of the jth hour of the ith item, n LO,k For the number of trips of the kth hour, k=1, 2, … … 24; equalization disaster-causing datasets.
Typhoons are one of extreme natural disasters, and have low occurrence probability and limited coverage area. Therefore, the disaster-causing data set has the most samples of normal operation classes, the second samples of light failure classes and the least samples of heavy failure classes, namely the light failure class samples and the heavy failure class samples are minority class samples.
The phenomenon of sample imbalance of disaster-causing data sets causes that a power distribution network fault condition prediction model lacks learning for few types of samples in the training process, and finally causes that the prediction accuracy of the power distribution network fault condition prediction model for the few types of samples is low. Considering that the cost-sensitive learning method has certain subjectivity and the parameter adjustment process is complicated, the method reduces the unbalance degree of the disaster-causing data set based on the Borderline-SMOTE1 algorithm and generates the quality of a few samples through the detection of a discrimination model.
1) Borderline-SMOTE1 sample generation algorithm
The SMOTE algorithm commonly used in the over-sampling technique has greater blindness and randomness when selecting the target sample for sample generation, and is easy to generate new samples that are meaningless or have interference to define decision boundaries. Therefore, the method is based on a Borderline-SMOTE1 algorithm, divides the minority class samples according to the type distribution characteristics around the minority class samples, selects the minority class samples close to the decision boundary for sample generation, and reduces the unbalance of the disaster-causing data set. The algorithm steps of the Borderline-SMOTE1 are described by taking the generation of a light fault class sample as an example, and the generation process of a heavy fault class sample is the same. It should be noted that the sample generation algorithm is only applied to the disaster causing training set.
Step1: calculating m nearest neighbor samples of each mild failure sample by using a K nearest neighbor algorithm;
step2: the light failure samples are classified into the following three classes according to the duty ratio of the light failure samples in m nearest neighbor samples of the light failure class samples, and a classification schematic diagram is shown in fig. 2.
(1) Security class sample: more than half of the nearest neighbor samples are light fault samples, such as sample A in FIG. 2;
(2) Dangerous class sample: less than half of the nearest neighbor samples are light failure samples, such as sample B in fig. 2;
(3) Noise-like samples: the nearest neighbor sample has no light failure sample, such as sample C in fig. 2;
step3: for each hazard class sample x i Selecting a required number of light fault samples from K nearest neighbor samples;
step4: for each selected neighbor sample x' j Generating a light failure class new sample x using linear interpolation i,j The calculation formula is as follows:
x i,j =x i +γ(x′ j -x i ) (1)
wherein γ is a random number ranging from 0 to 1.
Step5: and adding the generated light fault new sample into the original disaster training set.
2) Verification of disaster causing datasets
Considering that the data distribution of the disaster-causing training set is artificially changed by adding a few types of generated samples, when the quality of the generated samples is low, the sample distribution difference of the disaster-causing training set and the disaster-causing test set can be increased, and further the generalization capability of the prediction model on the disaster-causing test set is reduced. Therefore, the invention designs a discrimination model, performs sample distribution test on a disaster-causing training set and a disaster-causing test set after adding a generated sample, and adjusts the parameter setting of a sample generation method according to the test result, wherein the specific principle is shown in figure 3. The following describes a specific procedure for discriminating differences in sample distribution by the model.
(1) Construction of a discrimination data set: the discrimination data set is based on the self-supervision learning idea, and the sample division condition of the training set and the test set is used as the label source of the discrimination data set. Considering that the number of samples of the disaster-causing training set is generally several times that of the disaster-causing testing set, random sampling is carried out on the disaster-causing training set, and the number of samples of the sampled training set is ensured to be equal to that of the disaster-causing testing set. And then respectively setting labels of the training set sample and the test set sample to be 0 and 1, mixing to form a discrimination data set, and dividing the discrimination data set into a new training set and a new test set according to the ratio of 8:2.
(2) Training process of the discrimination model: based on a new training set and a new testing set, a cross entropy function is used as a loss function, the gradient of each parameter value of the judging model is obtained through an error back propagation method, and all parameters of the judging model are updated through an Adam gradient descent algorithm, so that the judging accuracy of the disaster-causing training set and the disaster-causing testing set is obtained.
(3) And (3) analyzing a test result of the discrimination model: and distinguishing sample distribution differences of the disaster causing training set and the ability measurement of the disaster causing test set by using a distinguishing model. When the judging accuracy is higher than the accuracy threshold, the fact that the sample distribution difference between the disaster causing training set and the disaster causing test set is large is indicated, and the disaster causing training set needs to be subjected to reconstruction processing, namely, parameters such as the nearest neighbor sample number of a Borderline-SMOTE1 algorithm are adjusted; when the judging accuracy is lower than the accuracy threshold, the sample distribution difference of the judging accuracy and the sample distribution difference of the judging accuracy are small, and the judging accuracy and the sample distribution difference can be directly used for training and testing of a prediction model. The accuracy threshold of the discriminant model is typically set to 70%.
3. Construction of a two-channel predictive model
In order to consider the stability of static data action and the time-varying and accumulating property of dynamic data action, the invention provides an interpretable neural network architecture which is used for respectively extracting characteristics of static data and dynamic data so as to establish a mapping relation between the static data and the dynamic data and the type of power distribution network fault condition under typhoon disasters. The following describes the feature extraction process of static and dynamic data and the training method of the dual-channel prediction model in detail.
3.1 static feature extraction based on feedforward neural network
The feedforward neural network is composed of an input layer, a hidden layer and an output layer, all the layers of neurons are connected, and an intra-layer connecting structure and a cross-layer connecting structure do not exist, so that the information transmission process of the feedforward neural network is unidirectional. Considering the stability of the effect of static data of typhoon disasters on the fault condition of the power distribution network, the static characteristics of the static data are extracted layer by adopting a multi-layer feedforward neural network aiming at the static data which are kept unchanged within 48 hours.
3.2 dynamic feature extraction based on Long-short term memory network and Multi-head attention mechanism
Unlike feed forward neural networks, long Short Term Memory (LSTM) networks not only pass information from layer to layer, but also pass information within the same layer. By the addition of such intra-layer connection structures, LSTM has "memory" and "transitivity" for the processing of data. Meanwhile, the LSTM unit structure comprises a plurality of gate structures, so that the gradient disappearance and gradient explosion problems caused by an interlayer connection structure can be effectively solved.
Each unit of LSTM includes three gate structures of a forgetting gate, an input gate and an output gate, and long-term memory and short-term memory after processing of the gate structures are transferred simultaneously when information is transferred between the same layers, and the unit structure of LSTM is shown in fig. 4. LSTM inputs information x according to current time t And short-term memory h of the previous time t-1 Respectively calculating forgetting door gating signals f t Input gate control signal i t And outputs a gate control signal o t :
f t =σ(U f x t +W f h t-1 +b f ) (2)
i t =σ(U i x t +W i h t-1 +b i ) (3)
o t =σ(U o x t +W o h t-1 +b o ) (4)
In the formula, sigma refers to Sigmoid activation function, U f For the current input x t Connection weight with forgetting door structure, U i For the current input x t Connection weight with input gate structure, U O For the current input x t Connection weight with output gate structure, W f Short-term memory h for the last moment t-1 Connection weight with forgetting door structure, W i Short-term memory h for the last moment t-1 Connection weight with input gate structure, W O Short-term memory h for the last moment t-1 Connection weight to output gate structure, b f Bias for forgetting door structure b i B for inputting the bias of the gate structure o To output the bias of the gate structure.
LSTM inputs information x to the current time based on three gating signals t And short-term memory h of the previous time t-1 Reprocessing to update the long-term memory c t And short-term memory h t The specific calculation formula is as follows:
in the method, in the process of the invention,u as candidate long-term memory c For inputting information and candidate long-term memory->Connection weight, W of c For short-term memory and candidate long-term memory->Connection weight of b) c Is candidate long-term memory->Is provided.
In order to further enhance the feature extraction capability of the network on dynamic data, the invention adopts a multi-head attention mechanism, utilizes a plurality of mapping subspaces to extract key components in the known data in an omnibearing and multi-angle way, and maximizes the utilization of the known data information. The multi-head Attention mechanism first maps the data Q to a plurality of subspaces and calculates the relevance and dependency between the data using the self-Attention formula Attention (Q). Self-attention value head corresponding to ith head i The specific calculation formula of (Q) is as follows:
head i (Q)=Attention(QW i Q ,QW i K ,QW i V ) (9)
wherein d Q I=1, 2, h is the number of heads of the attention mechanism, W, for the dimension of the input data Q i Q 、W i K 、W i V Each is a subspace transform matrix corresponding to the i-th header.
The outputs of all heads are then stitched and mapped by the linear layer to the final attention weighted value, i.e., multiHead (Q):
MultiHead(Q)=Concat(head 1 ,...,head h )W O (10)
in the formula, concat is splicing operation, W o To output a mapping matrix.
The invention firstly utilizes the LSTM network to extract dynamic data characteristics, then adds a multi-head attention mechanism layer to the LSTM network, and further extracts deep dynamic data characteristics in dynamic data, thereby laying a foundation for the establishment of a final mapping relation.
3.3 network structure of double-channel prediction model and training method
According to the invention, a feedforward neural network is adopted to process static data, a multi-head self-attention mechanism enhanced LSTM network is adopted to process dynamic data, finally deep features extracted by the static data and the dynamic data are spliced, the deep features are mapped into the prediction probability of each fault condition type of the power distribution network through a linear layer, the prediction fault condition type with the maximum probability value corresponding to the disaster type as a sample is taken, and the network structure of the prediction model is shown in figure 5. The convergence of the prediction model is improved by adding a corresponding batch normalization layer and a nonlinear activation function after the first linear layer of the feedforward neural network.
Because the prediction of the power distribution network fault condition type under typhoon disasters belongs to the classification problem, the invention uses a cross entropy function as a loss function on the basis of a disaster-causing training set to measure the difference degree between a predicted value and an actual value. And then obtaining the gradient value of the cross entropy function to each parameter in the model through an error back propagation algorithm. And finally, updating the parameters of the prediction model by using a small batch Adam algorithm in combination with super parameters such as the learning rate, the batch size and the number of neurons of each layer.
3.4 evaluation method of double-channel prediction model
Because the prediction of the power distribution network fault condition type is a three-classification problem, and the number of samples of each class of the disaster-causing test set is not equal. In order to relieve the leading effect of most sample evaluation results on the prediction accuracy and comprehensively consider the performance of a prediction model in various categories, the invention takes the precision, recall ratio and F1 measurement as a basic index system, and introduces a macro average mechanism to comprehensively consider the performance of the prediction model in different sample sets in a disaster-causing test set, and the specific process is as follows.
Firstly, according to the predicted value obtained after the disaster-causing test set is input into the prediction model, counting each disaster-causing test setWhether each sample belongs to the actual value and the predicted value of the disaster type can form three classification confusion moments altogether. After the matrix is formed, a group of true positive TP corresponding to each confusion matrix is obtained according to the matrix elements i False positive FP i True negative TN i And false negative FN i Thereby obtaining the corresponding precision P i Sum recall R i . Finally, obtaining three indexes of macro precision macro-P, macro recall ratio macro-R and macro F1 value macro-F1 according to a macro average mechanism, and comprehensively measuring the performance of the prediction model, wherein a specific calculation formula is as follows.
In consideration of the prediction accuracy, the performance of the model can be visually highlighted, so that four indexes of macro Cha Zhun rate, macro recall rate, macro F1 and accuracy are selected to evaluate the performance of the power distribution network fault condition prediction model under typhoon disasters.
4. Prediction using a two-channel prediction model
And acquiring weather forecast data issued by a weather department before typhoons pass through a border, geographic data, population data and power grid data of each research area, constructing a corresponding disaster-causing data set, and inputting the disaster-causing data set into a double-channel forecast model after parameter optimization to obtain forecast values of the fault condition types of the distribution network of each research area under future typhoons.
Example 2
As shown in fig. 6, the power grid fault prediction device under typhoon disasters provided by the invention comprises an acquisition module and a calculation output module;
the system comprises an acquisition module, a calculation output module and a calculation output module, wherein the acquisition module is used for acquiring data and transmitting the acquired data to the calculation output module; the data comprises historical dynamic data, static data and real-time typhoon data, wherein the real-time typhoon data comprises dynamic data and static data.
And the calculation output module is used for training a prediction model according to the historical dynamic data, the static data and the sum of the permanent tripping times of the predicted regional power grid, and then outputting a power grid fault prediction value according to the prediction model, the real-time dynamic data and the static data.
Example 3
As shown in fig. 7, the computer device provided by the present invention includes a memory and a processor electrically connected, where the memory stores a computing program that can be run on the processor, and when the processor executes the computing program, the steps of the prediction method are implemented.
Example 4
The prediction means, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the present invention may implement all or part of the flow of the method of the above embodiment, or may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a Read-only memory (ROM), a random access memory (RAM, randomAccessMemory), an electrical carrier signal, a telecommunication signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
The method is a universal power distribution network fault condition prediction model under typhoon disasters, and can be used for predicting, so that the inherent problem of unbalanced data sets can be effectively reduced, and the quality of generated samples is improved. Meanwhile, the prediction method considers the stability of static data action and the accumulative property of dynamic data action, further improves the accuracy and the interpretability of the prediction model, and provides more accurate prediction information for the power distribution network to cope with typhoon disasters.
The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (5)
1. The method for predicting the power grid faults under the typhoon disaster driven by the data is characterized by comprising the following steps of:
step 1, collecting multi-element influence data of power grid faults under typhoon disasters and the sum of permanent tripping times of a predicted regional power grid, dividing the data into static data and dynamic data according to time domain change attributes of the data, and constructing a disaster-causing data set by utilizing the static data, the dynamic data and the sum of the permanent tripping times of the predicted regional power grid; dividing the total tripping times into slight faults of the regional distribution network, severe faults of the regional distribution network and normal operation of the regional distribution network, and taking the three disaster-stricken condition types as labels of disaster-stricken data sets;
Step 2, carrying out equalization treatment on the disaster-causing data set;
step 3, extracting the characteristics of static data in disaster-causing data sets by utilizing a feedforward neural network, extracting the sequence characteristics of dynamic data in disaster-causing data sets by utilizing a long-period memory network and a multi-head self-attention mechanism, establishing a dual-channel prediction model of power grid faults under typhoon disasters, solving and optimizing model parameters based on the disaster-causing data sets after sample equalization treatment, and finally obtaining an optimized dual-channel prediction model; and evaluate its performance; if the performance meets the requirement, carrying out the step 4, otherwise, continuing to optimize;
step 4, collecting corresponding multi-element influence data of a prediction area under future typhoon disasters, constructing a disaster-causing data set, inputting the disaster-causing data set into the optimized double-channel prediction model in the step 3, and obtaining a prediction value of the power grid fault condition of the research area under the future typhoon disasters;
the process of the step 2 is as follows: dividing a minority sample set according to the distribution of disaster-causing data sets in a high-dimensional space by using a Borderline-SMOTE1 algorithm, and generating samples aiming at minority samples at decision boundaries after division; thirdly, checking the difference of data distribution of the training set and the testing set through a discrimination model, performing parameter tuning on the Borderline-SMOTE1 algorithm according to the difference, and finally applying the Borderline-SMOTE1 algorithm after parameter optimization to balance disaster-causing data sets;
The step 2 comprises the following steps:
step 2.1, useKThe neighbor algorithm calculates each light fault class samplemNearest neighbor samples;
step 2.2 based on the light failure class samplemThe duty ratio of the light fault sample in the nearest neighbor sample is divided into a safety sample, a dangerous sample and a noise sample;
step 2.3 for each hazardous class sampleIn itKSelecting a required number of light fault samples from the nearest neighbor samples;
step 2.4, for each selected neighbor sampleGenerating a light failure class new sample using linear interpolation +.>;
Step 2.5, adding the generated light fault new sample into an original disaster training set to obtain an updated disaster data set;
step 2.6, checking the updated disaster causing data set, if the updated disaster causing data set meets the requirements, performing step 3, and if the updated disaster causing data set does not meet the requirements, performing parameter adjustment on the Borderline-SMOTE1 algorithm until the disaster causing data set meets the requirements;
the step 3 comprises the following steps:
step 3.1, extracting static characteristics from static data based on a feedforward neural network; extracting dynamic characteristics from dynamic data based on a long-short-period memory network and a multi-head attention mechanism;
step 3.2, splicing the static features and the dynamic features, mapping the static features and the dynamic features into the prediction probability of each fault condition type of the power grid through a linear layer, and taking the prediction fault condition type with the maximum probability value corresponding to the disaster type as a sample to obtain a prediction model; using a cross entropy function as a loss function, and measuring the difference degree between the predicted value and the actual value; then obtaining a gradient value of each parameter in the cross entropy function pair model through an error back propagation algorithm; finally, updating the prediction model parameters by using a small batch Adam algorithm in combination with the learning rate, the batch size and the number of neurons of each layer;
Step 3.3, taking the precision rate and the recall rate as basic index systems, introducing a macro average mechanism to comprehensively consider the performances of the prediction model in different types of sample sets in the disaster-causing test set, and evaluating the prediction model;
the step 3.3 comprises the following steps:
step 3.3.1, counting whether each sample in the disaster-causing test set belongs to an actual value and a predicted value of the disaster-causing type according to the predicted value obtained after the disaster-causing test set is input into the prediction model, and forming three classification confusion moments;
step 3.3.2, obtaining a group of true positives corresponding to each confusion matrix according to the matrix elementsFalse positives->True negative->And false negative +.>Thereby obtaining the corresponding precision +.>Sum recall->;
Step 3.3.3 according to the precisionRecall->AndF1measurement results in macro precision->Macro recall ratioSum macroF1Value->;
Step 3.3.4, according to macro Cha Zhun rate, macro recall, macroF1And the accuracy rate is four indexes in total, and the performance of the power grid fault condition prediction model under typhoon disasters is evaluated.
2. The method according to claim 1, wherein in the step 1, the static data includes forest coverage, land type, maintenance level of the power grid and population density, and the dynamic data includes distance between center of typhoon and center of area, lowest air pressure of center of typhoon, maximum wind speed of near center of typhoon, moving speed of typhoon, moving direction angle of typhoon, seven-level wind circle radius, average wind speed of predicted area and precipitation of predicted area.
3. The method for predicting grid faults in a typhoon disaster driven by data as claimed in claim 1, wherein the step 2.6 comprises the following steps:
step 2.6.1, randomly sampling the disaster causing training set to make the number of samples of the sampled training set equal to the number of samples of the disaster causing test set; then respectively setting labels of the training set sample and the test set sample to be 0 and 1, mixing to form a discrimination data set, and dividing the discrimination data set into a new training set and a new test set according to a proportion;
step 2.6.2, based on a new training set and a new testing set, using a cross entropy function as a loss function, obtaining the gradient of each parameter value of the judging model through an error back propagation method, and updating all parameters of the judging model through an Adam gradient descent algorithm to obtain the judging accuracy of the disaster-causing training set and the disaster-causing testing set;
2.6.3, using a discrimination model to distinguish sample distribution differences of a disaster causing training set and a disaster causing test set, and when the discrimination accuracy is higher than an accuracy threshold, adjusting parameters such as nearest neighbor sample number of a Borderline-SMOTE1 algorithm; and when the judging accuracy is lower than the accuracy threshold, executing the step 3.
4. A grid fault prediction device under typhoon disasters, comprising:
The acquisition module is used for acquiring data and transmitting the acquired data to the calculation output module; the data comprise multi-element influence data of power grid faults under typhoon disasters, the sum of permanent tripping times of the power grid in the predicted area and real-time typhoon data;
the calculation output module is used for training a prediction model according to the collected data set and outputting a power grid fault prediction value according to the prediction model and real-time typhoon data;
the calculation output module comprises a disaster-causing data set construction module, an equalization processing module, a prediction model construction module and an output module;
the disaster-causing data set construction module is used for dividing the data into static data and dynamic data according to the time domain change attribute of the data, and constructing a disaster-causing data set by utilizing the static data, the dynamic data and the sum of the permanent tripping times of the predicted regional power grid; dividing the total tripping times into slight faults of the regional distribution network, severe faults of the regional distribution network and normal operation of the regional distribution network, and taking the three disaster-stricken condition types as labels of disaster-stricken data sets;
the equalization processing module is used for carrying out equalization processing on the disaster-causing data set;
the prediction model construction module is used for extracting the characteristics of static data in the disaster-causing data set by utilizing the feedforward neural network, extracting the sequence characteristics of dynamic data in the disaster-causing data set by utilizing the long-short-period memory network and the multi-head self-attention mechanism, establishing a dual-channel prediction model of power grid faults under typhoon disasters, solving and optimizing model parameters based on the disaster-causing data set after sample equalization treatment, and finally obtaining an optimized dual-channel prediction model; and optimizing the performance of the alloy to meet the requirements;
The output module is used for collecting corresponding multi-element influence data of a prediction area under the future typhoon disaster, constructing a disaster-causing data set, inputting the disaster-causing data set into the optimized double-channel prediction model, and obtaining a prediction value of the power grid fault condition of the research area under the future typhoon disaster;
the equalization processing of the disaster-causing data set comprises the following steps: dividing a minority sample set according to the distribution of disaster-causing data sets in a high-dimensional space by using a Borderline-SMOTE1 algorithm, and generating samples aiming at minority samples at decision boundaries after division; thirdly, checking the difference of data distribution of the training set and the testing set through a discrimination model, performing parameter tuning on the Borderline-SMOTE1 algorithm according to the difference, and finally applying the Borderline-SMOTE1 algorithm after parameter optimization to balance disaster-causing data sets;
the equalization processing of the disaster-causing data set comprises the following steps:
step 2.1, useKThe neighbor algorithm calculates eachOf light failure typemNearest neighbor samples;
step 2.2 based on the light failure class samplemThe duty ratio of the light fault sample in the nearest neighbor sample is divided into a safety sample, a dangerous sample and a noise sample;
Step 2.3 for each hazardous class sampleIn itKSelecting a required number of light fault samples from the nearest neighbor samples;
step 2.4, for each selected neighbor sampleGenerating a light failure class new sample using linear interpolation +.>;
Step 2.5, adding the generated light fault new sample into an original disaster training set to obtain an updated disaster data set;
step 2.6, checking the updated disaster causing data set, if the updated disaster causing data set meets the requirements, performing step 3, and if the updated disaster causing data set does not meet the requirements, performing parameter adjustment on the Borderline-SMOTE1 algorithm until the disaster causing data set meets the requirements;
the prediction model construction module operates the following steps:
step 3.1, extracting static characteristics from static data based on a feedforward neural network; extracting dynamic characteristics from dynamic data based on a long-short-period memory network and a multi-head attention mechanism;
step 3.2, splicing the static features and the dynamic features, mapping the static features and the dynamic features into the prediction probability of each fault condition type of the power grid through a linear layer, and taking the prediction fault condition type with the maximum probability value corresponding to the disaster type as a sample to obtain a prediction model; using a cross entropy function as a loss function, and measuring the difference degree between the predicted value and the actual value; then obtaining a gradient value of each parameter in the cross entropy function pair model through an error back propagation algorithm; finally, updating the prediction model parameters by using a small batch Adam algorithm in combination with the learning rate, the batch size and the number of neurons of each layer;
3.3, taking the precision rate and the recall rate as basic index systems, introducing a macro average mechanism to comprehensively consider the performances of the prediction model in different types of sample sets in the disaster-causing test set, and evaluating the prediction model;
the step 3.3 comprises the following steps:
step 3.3.1, counting whether each sample in the disaster-causing test set belongs to an actual value and a predicted value of the disaster-causing type according to the predicted value obtained after the disaster-causing test set is input into the prediction model, and forming three classification confusion moments;
step 3.3.2, obtaining a group of true positives corresponding to each confusion matrix according to the matrix elementsFalse positives->True negative->And false negative +.>Thereby obtaining the corresponding precision +.>Sum recall->;
Step 3.3.3 according to the precisionRecall->AndF1measurement results in macro precision->Macro recall ratioSum macroF1Value->;
Step 3.3.4, according to macro Cha Zhun rate, macro recall, macroF1And the accuracy rate is four indexes in total, and the performance of the power grid fault condition prediction model under typhoon disasters is evaluated.
5. A computer device, comprising: a memory and a processor electrically connected, said memory having stored thereon a computing program executable on the processor, said processor, when executing said computing program, performing the steps of the method according to any of claims 1-3.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111139000.6A CN113837477B (en) | 2021-09-27 | 2021-09-27 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
PCT/CN2022/081314 WO2023045278A1 (en) | 2021-09-27 | 2022-03-17 | Data dual-drive method, apparatus, and device for predicting power grid failure during typhoon |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111139000.6A CN113837477B (en) | 2021-09-27 | 2021-09-27 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113837477A CN113837477A (en) | 2021-12-24 |
CN113837477B true CN113837477B (en) | 2023-06-27 |
Family
ID=78970684
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111139000.6A Active CN113837477B (en) | 2021-09-27 | 2021-09-27 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN113837477B (en) |
WO (1) | WO2023045278A1 (en) |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113537552A (en) * | 2021-04-28 | 2021-10-22 | 桂林电子科技大学 | Student performance prediction method based on mixed deep learning and attention mechanism |
CN113837477B (en) * | 2021-09-27 | 2023-06-27 | 西安交通大学 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
CN116016219A (en) * | 2022-12-20 | 2023-04-25 | 缀初网络技术(上海)有限公司 | Edge cloud server loss prediction method and device |
CN116522764B (en) * | 2023-04-17 | 2023-12-19 | 华中科技大学 | Hot wave-flood composite disaster assessment method considering influence of climate change |
CN116629465B (en) * | 2023-07-26 | 2024-01-12 | 李波 | Smart power grids video monitoring and risk prediction response system |
CN117034755B (en) * | 2023-08-07 | 2024-06-18 | 兰州理工大学 | Cold-rolled steel mechanical property prediction method integrating multi-head attention mechanism |
CN117390948A (en) * | 2023-08-10 | 2024-01-12 | 苏州黑盾环境股份有限公司 | Multi-head attention long-short-term memory neural network based water chilling unit monitoring method |
CN116910493B (en) * | 2023-09-12 | 2023-12-08 | 山东能源数智云科技有限公司 | Construction method and device of equipment fault diagnosis model based on multi-source feature extraction |
CN116956197B (en) * | 2023-09-14 | 2024-01-19 | 山东理工昊明新能源有限公司 | Deep learning-based energy facility fault prediction method and device and electronic equipment |
CN116933981B (en) * | 2023-09-15 | 2023-12-08 | 安徽方能电气技术有限公司 | Regional power stability analysis and evaluation method based on power outage and restoration event data monitoring |
CN117422438B (en) * | 2023-10-07 | 2024-03-29 | 国家电网有限公司华东分部 | Method and device for determining reinforcement scheme of power transmission line |
CN117056402B (en) * | 2023-10-12 | 2024-04-02 | 国网浙江省电力有限公司余姚市供电公司 | Motor diagnosis method and device based on multi-source signals and storage medium |
CN117056734B (en) * | 2023-10-12 | 2024-02-06 | 山东能源数智云科技有限公司 | Method and device for constructing equipment fault diagnosis model based on data driving |
CN117347803A (en) * | 2023-10-25 | 2024-01-05 | 爱科特科技(海南)有限公司 | Partial discharge detection method, system, equipment and medium |
CN117150243B (en) * | 2023-10-27 | 2024-01-30 | 湘江实验室 | Fault isolation and estimation method based on fault influence decoupling network |
CN117216485B (en) * | 2023-11-09 | 2024-01-30 | 国网山东省电力公司电力科学研究院 | Objective weighting-based power transmission wave-recording bird damage fault judging method and system |
CN117495147B (en) * | 2023-12-22 | 2024-03-22 | 中国石油大学(华东) | Crack network expansion intelligent prediction method considering intersegmental fracturing process difference |
CN117522177B (en) * | 2024-01-08 | 2024-03-12 | 国网江苏省电力有限公司信息通信分公司 | Smart power grid stability prediction method |
CN117828490B (en) * | 2024-03-06 | 2024-05-17 | 南京信息工程大学 | Typhoon disaster forecasting method and system based on ensemble learning |
CN117909719B (en) * | 2024-03-19 | 2024-05-24 | 勤源(江苏)科技有限公司 | Terminal internet of things equipment monitoring and early warning method and system based on autonomous learning algorithm |
CN117973440B (en) * | 2024-04-02 | 2024-06-14 | 长江三峡集团实业发展(北京)有限公司 | Regional ionosphere delay prediction method based on LSTM-transducer model |
CN117992861B (en) * | 2024-04-04 | 2024-06-21 | 国网湖北省电力有限公司 | Electric power data accuracy checking method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101354757A (en) * | 2008-09-08 | 2009-01-28 | 中国科学院地理科学与资源研究所 | Method for predicting dynamic risk and vulnerability under fine dimension |
CN109814527A (en) * | 2019-01-11 | 2019-05-28 | 清华大学 | Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device |
JP2020042705A (en) * | 2018-09-13 | 2020-03-19 | いすゞ自動車株式会社 | Fault prediction apparatus, fault prediction method, and program |
CN111191832A (en) * | 2019-12-25 | 2020-05-22 | 国电南瑞科技股份有限公司 | Typhoon disaster power distribution network tower fault prediction method and system |
CN113191585A (en) * | 2021-03-23 | 2021-07-30 | 广东电网有限责任公司东莞供电局 | Typhoon disaster risk assessment method for power transmission line |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9262124B2 (en) * | 2011-11-21 | 2016-02-16 | International Business Machines Corporation | Natural disaster forecasting |
US10050990B2 (en) * | 2014-12-29 | 2018-08-14 | Guidewire Software, Inc. | Disaster scenario based inferential analysis using feedback for extracting and combining cyber risk information |
CN109933881A (en) * | 2019-03-06 | 2019-06-25 | 武汉大学 | A kind of Fault Diagnosis of Power Electronic Circuits method based on optimization deepness belief network |
CN113837477B (en) * | 2021-09-27 | 2023-06-27 | 西安交通大学 | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data |
-
2021
- 2021-09-27 CN CN202111139000.6A patent/CN113837477B/en active Active
-
2022
- 2022-03-17 WO PCT/CN2022/081314 patent/WO2023045278A1/en unknown
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101354757A (en) * | 2008-09-08 | 2009-01-28 | 中国科学院地理科学与资源研究所 | Method for predicting dynamic risk and vulnerability under fine dimension |
JP2020042705A (en) * | 2018-09-13 | 2020-03-19 | いすゞ自動車株式会社 | Fault prediction apparatus, fault prediction method, and program |
CN109814527A (en) * | 2019-01-11 | 2019-05-28 | 清华大学 | Based on LSTM Recognition with Recurrent Neural Network industrial equipment failure prediction method and device |
CN111191832A (en) * | 2019-12-25 | 2020-05-22 | 国电南瑞科技股份有限公司 | Typhoon disaster power distribution network tower fault prediction method and system |
CN113191585A (en) * | 2021-03-23 | 2021-07-30 | 广东电网有限责任公司东莞供电局 | Typhoon disaster risk assessment method for power transmission line |
Non-Patent Citations (5)
Title |
---|
A Stacking Ensemble Prediction Model for the Occurrences of Major Adverse Cardiovascular Events in Patients With Acute Coronary Syndrome on Imbalanced Data;Huilin Zheng等;IEEE Access;全文 * |
基于T-S模糊神经网络的模型在台风灾情预测中的应用――以海南为例;张广平;张晨晓;谢忠;;灾害学(第02期);全文 * |
应用机器学习技术预测强雨雪天气过程中的积雪;张远汀;龚伟伟;叶钰;徐希源;徐勋建;蔡泽林;陆佳政;韩俊浩;叶飞;许婧;;科学技术与工程(15);全文 * |
强台风环境下基于格点化和支持向量机的10 kV杆塔受损量预测方法;郑凌铭;舒胜文;陈彬;吴涵;黄建业;钱健;;高电压技术(第01期);全文 * |
风电机组智能状态评估与故障预测研究进展;李重桂;李录平;刘瑞;杨波;陈茜;邓子豪;;电站***工程(第04期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
WO2023045278A1 (en) | 2023-03-30 |
CN113837477A (en) | 2021-12-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113837477B (en) | Method, device and equipment for predicting power grid faults under typhoon disasters driven by data | |
CN115270965B (en) | Power distribution network line fault prediction method and device | |
CN106707099A (en) | Monitoring and locating method based on abnormal electricity consumption detection module | |
Yao et al. | A support vector machine with the tabu search algorithm for freeway incident detection | |
CN114548637A (en) | AHP-RST-based power communication backbone data network security comprehensive evaluation method | |
CN113807570B (en) | XGBoost-based reservoir dam risk level assessment method and system | |
CN113327022B (en) | Lightning protection safety risk management system and method | |
CN111680875A (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
CN109492756B (en) | Multi-element wire galloping early warning method based on deep learning and related device | |
CN114186641B (en) | Landslide susceptibility evaluation method based on deep learning | |
CN115130375A (en) | Rock burst intensity prediction method | |
CN106656357A (en) | System and method of evaluating state of power frequency communication channel | |
Manninen et al. | Health index prediction of overhead transmission lines: a machine learning approach | |
CN114169445A (en) | Day-ahead photovoltaic power prediction method, device and system based on CAE and GAN hybrid network | |
CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
CN114692692B (en) | Snowfall recognition method based on microwave attenuation signal fusion kernel extreme learning machine | |
CN116738192A (en) | Digital twinning-based security data evaluation method and system | |
CN117238126A (en) | Traffic accident risk assessment method under continuous flow road scene | |
CN113689053B (en) | Strong convection weather overhead line power failure prediction method based on random forest | |
CN116976318A (en) | Intelligent auditing system for switching operation ticket of power grid based on deep learning and model reasoning | |
CN113780094B (en) | Perimeter intrusion identification method based on Gaussian mixture model and hidden Markov model | |
CN113884807A (en) | Power distribution network fault prediction method based on random forest and multi-layer architecture clustering | |
CN116029220B (en) | Voltage transformer operation error assessment method, system, equipment and medium | |
CN117009923A (en) | Tunnel advanced geological disaster intelligent judging and identifying method based on multi-source information fusion | |
CN112232329A (en) | Multi-core SVM training and alarming method, device and system for intrusion signal recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |