Methods and devices for predicting the risk of power outages from overhead lines during severe convective weather

Abstract

This invention discloses a method and apparatus for predicting overhead power outage risks during severe convective weather, belonging to the field of power distribution network big data technology. The method includes: acquiring a training dataset, which includes feature data and its corresponding data labels; the training data includes severe convective meteorological feature data, geographical feature data, and power grid state feature data; initializing a decision tree; using the training data and its corresponding data labels as input to the decision tree, calculating its objective function and the output confidence of each leaf in the decision tree; selecting leaves that can generate a new tree based on the output confidence of each leaf, and generating a new decision tree; until all feature data in the training dataset has been traversed, a sparse decision tree for calculating the probability of overhead power outages is obtained. This invention can efficiently construct decision trees and has good prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of big data technology for power distribution networks, specifically to a method and device for predicting the risk of power outages on overhead lines during severe convective weather. Background Technology

[0002] With economic and social development, the coverage of power grid substations and distribution networks continues to expand. As a result, the impact of severe weather events such as strong winds, lightning, and smog on the safe operation of the power grid increases. Under the influence of extreme weather, 10kV overhead distribution lines are more prone to tripping, causing regional power grid line failures and outages, which affect the normal production and life of residents.

[0003] Currently, while there is considerable research on risk assessment and prevention of meteorological disasters affecting power grids, it primarily focuses on the impact of weather events such as snow and strong winds on distribution networks. With the increasing frequency of severe weather events like smog and icing, the normal operation of power grid equipment has been significantly affected. Many power research institutions have conducted in-depth analyses of meteorological disasters affecting power grids and are seeking preventative measures. The China Electric Power Research Institute initiated a broad survey on "The Current Status of Technology for China's Power Grid in Coping with Extreme Natural Disasters," clarifying the types and characteristics of meteorological disasters faced by my country's power grid, and studying the spatiotemporal distribution characteristics of meteorological influencing factors of power grid disasters in various regions, laying the foundation for meteorological disaster early warning and defense strategies.

[0004] Current research and analysis on overhead line outage prediction under severe convective weather is limited, mainly focusing on overhead line outage prediction based on neural networks. For example, Chinese invention patent application CN108596449A discloses a method for predicting the reliability of distribution networks that considers the impact of weather on the probability of distribution network failures. The implementation steps are: (1) analyzing the correlation between the probability of failure of distribution network components and weather variables; (2) selecting the main weather variables while considering the applicability and accuracy of the model; (3) forming a method for predicting the failure rate of distribution network components based on a learnable sparse decision tree; (4) combining the prediction results of the probability of failure of distribution network components and the method for calculating the reliability of distribution networks to predict the reliability of distribution networks. This invention can predict the reliability of distribution networks based on the impact of different weather variables on the probability of failure of distribution network components.

[0005] The neural network-based method for predicting overhead power line outages involves extracting features from a set of factors and inputting them into the neural network to obtain the final prediction confidence level. However, due to the numerous and highly variable influencing factors in overhead power line outage prediction, such as weather, geographical, and historical fault factors, the numerical and physical meanings of these factors differ significantly. The numerical differences in the input features make it difficult for the neural network to converge during training, hindering the achievement of high-performance overhead power line outage prediction accuracy. Furthermore, due to the uninterpretable nature of neural networks, the prediction results are difficult to interpret with reasonable physical meaning. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to achieve high-precision prediction of overhead power line outages during severe convective weather.

[0007] The present invention solves the above-mentioned technical problems through the following technical means:

[0008] On the one hand, this invention proposes a method for predicting the risk of overhead power outages during severe convective weather, the method comprising:

[0009] Obtain a training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective meteorological feature data, geographical feature data, and power grid status feature data.

[0010] Initialize the decision tree;

[0011] The training data and its corresponding data labels are used as input to the decision tree to calculate its objective function R(d,x,y) and the output confidence of each leaf in the decision tree.

[0012] Based on the output confidence of each leaf, select the leaves that can generate a new tree and generate a new decision tree;

[0013] The process continues until all feature data in the training dataset has been traversed to obtain a sparse decision tree.

[0014] The probability of overhead line power outage in the target area is calculated using the sparse decision tree.

[0015] This invention designs and utilizes severe convective weather characteristic data, geographical characteristic data, and power grid state characteristic data to train an optimizeable sparse decision tree, enabling it to predict the probability of overhead line power outages under severe convective weather based on input information. The decision tree algorithm itself has good handling capabilities for the differences in the dimensions of input data, and can well cope with the numerical differences of weather factors, geographical factors, etc. Furthermore, when adding leaves to form subtrees at each step, the generation of each new tree comes from the leaves that can be further divided in the old tree, rather than from the leaves that cannot be further divided in the old tree. Therefore, the decision tree in this embodiment is an asymmetric sparse optimizeable decision tree, which can efficiently construct decision trees and has a good ability to prevent overfitting, and can achieve good prediction accuracy in actual deployment.

[0016] Furthermore, the severe convective meteorological characteristic data includes the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning strikes, wherein the maximum intensity of severe convective radar is the maximum value within a set time period, the magnitude of lightning current is the average value within a set time period, and the number of radar strikes is the number of lightning strikes within a set time period.

[0017] The geographic feature data includes local geographic features and global geographic features;

[0018] The power grid status characteristic data includes the number of load cycles, the number of overload cycles, and the number of defects.

[0019] Furthermore, obtaining the training set includes:

[0020] The corresponding feature data is obtained from a period of time before the power outage signal;

[0021] Based on the label value range, the pairing of feature data and label information within the time period is obtained, wherein the label value range is:

[0022]

[0023] Furthermore, the initialized decision tree is as follows:

[0024] d=(d un δ un d split δ split (K, H)

[0025] Where, d un For the leaves that do not change in d, δ un To avoid changing the prediction result corresponding to the leaf, d split For the leaves in d that need to be further divided, δ split The features corresponding to the leaves to be further subdivided are H, where H is the total number of leaves and K is the delimiter.

[0026] Furthermore, the objective function R(d, x, y) is:

[0027] R(d,x,y)=l(d,x,y)+λH d

[0028] Where d represents the parameters of the decision tree, x represents the input feature data, y represents the label corresponding to the feature data, l(d, x, y) represents optimizing the decision tree parameters directly based on the label, and λH d This represents the regularization term.

[0029] Further, the step of selecting leaves that can generate a new tree based on the output confidence of each leaf, and generating a new decision tree, includes:

[0030] The status of the overhead line is determined based on the output confidence level of each leaf.

[0031] If the overhead line is in a power outage state, then the leaf is determined to be a leaf that can generate a new tree;

[0032] If the overhead line is in a normal state, then the leaf is determined to be a leaf that cannot generate a new tree.

[0033] Furthermore, this invention also proposes a device for predicting the risk of overhead power outages during severe convective weather, the device comprising:

[0034] The acquisition module is used to acquire a training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective meteorological feature data, geographical feature data, and power grid status feature data.

[0035] The initialization module is used to initialize the decision tree;

[0036] The training module is used to take the training data and its corresponding data labels as input to the decision tree, calculate its objective function R(d, x, y) and the output confidence of each leaf in the decision tree;

[0037] The decision tree update module is used to select leaves that can generate a new tree based on the output confidence of each leaf, and generate a new decision tree.

[0038] The output module is used to obtain a sparse decision tree until all feature data in the training dataset has been traversed.

[0039] The prediction module is used to calculate the probability of overhead line power outages in the target area using the sparse decision tree.

[0040] Furthermore, the severe convective meteorological characteristic data includes the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning strikes, wherein the maximum intensity of severe convective radar is the maximum value within a set time period, the magnitude of lightning current is the average value within a set time period, and the number of radar strikes is the number of lightning strikes within a set time period.

[0041] The geographic feature data includes local geographic features and global geographic features;

[0042] The power grid status characteristic data includes the number of load cycles, the number of overload cycles, and the number of defects.

[0043] Furthermore, the initialized decision tree is as follows:

[0044] d=(d un δ un d split δ split (K, H)

[0045] Where, d un For the leaves that do not change in d, δ un To avoid changing the prediction result corresponding to the leaf, d splitFor the leaves in d that need to be further divided, δ split The features corresponding to the leaves to be further subdivided are H, where H is the total number of leaves and K is the delimiter.

[0046] Furthermore, the objective function R(d, x, y) is:

[0047] R(d,x,y)=l(d,x,y)+λH d

[0048] Where d represents the parameters of the decision tree, x represents the input feature data, y represents the label corresponding to the feature data, l(d, x, y) represents optimizing the decision tree parameters directly based on the label, and λH d This represents the regularization term.

[0049] The advantages of this invention are:

[0050] (1) This invention designs and uses strong convective meteorological feature data, geographical feature data and power grid status feature data to train an optimizable sparse decision tree, which enables it to predict the probability of overhead line power outages under strong convective weather based on input information. The decision tree algorithm itself has good processing ability for the differences in the dimensions of input data, and can well cope with the numerical differences of weather factors, geographical factors and other factors. Furthermore, when the decision tree adds leaves to form a subtree at each step, the generation of each new tree comes from the leaves that can be further divided in the old tree, rather than from the leaves that cannot be further divided in the old tree. Therefore, the decision tree in this embodiment is an asymmetric sparse optimizable decision tree, which can efficiently construct decision trees and has a good ability to prevent overfitting, and can achieve good prediction accuracy in actual deployment.

[0051] (2) When adding leaves to form subtrees at each step, the decision tree is optimized based on a set loss function, which further improves the performance of the constructed decision tree.

[0052] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the method for predicting the risk of power outages from overhead lines during severe convective weather in the first embodiment of the present invention.

[0054] Figure 2 This is a principle block diagram of the prediction of overhead power outage risk during severe convective weather in the first embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the sparse decision tree model structure in the first embodiment of the present invention;

[0056] Figure 4 This is a comparative diagram of the sparse decision tree model and the traditional decision tree model in the first embodiment of the present invention. (a) is a structural diagram of the traditional decision tree model, and (b) is a structural diagram of the sparse decision tree model.

[0057] Figure 5 This is a schematic diagram of the structure of the overhead power outage risk prediction system for severe convective weather in the second embodiment of the present invention;

[0058] Figure 6 This is a structural block diagram of the risk prediction of overhead power outages during severe convective weather in the second embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] like Figures 1 to 2 As shown, the first embodiment of the present invention proposes a method for predicting the risk of overhead power outages during severe convective weather. The method includes the following steps:

[0061] S10. Obtain the training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective weather feature data, geographical feature data, and power grid status feature data.

[0062] S20. Initialize the decision tree.

[0063] S30. Using the training data and its corresponding data labels as input to the decision tree, calculate its objective function R(d, x, y) and the output confidence of each leaf in the decision tree.

[0064] S40. Based on the output confidence of each leaf, select the leaves that can generate a new tree and generate a new decision tree.

[0065] S50. Continue until all feature data in the training dataset has been traversed to obtain a sparse decision tree.

[0066] S60. Using the sparse decision tree, calculate the probability of power outage of overhead lines in the target area.

[0067] It should be noted that, as Figures 3 to 4As shown, unlike traditional decision tree models, the decision tree in this embodiment is a sparse optimizable decision tree, an iterative decision tree algorithm. This algorithm consists of multiple decision trees, and the conclusions of all trees are summed to obtain the final answer. The construction process of this sparse optimizable decision tree is based on the input data of each different feature, and the correlation between features is reflected in the construction of new trees, not in the leaves of old trees. Therefore, each new tree is generated from the divisible leaves of the old trees, while the leaves that cannot be further divided, i.e., d... un It can be considered that it is "dead". Based on this analysis, sparse decision trees use an asymmetric structure, which is also the manifestation of its sparsity. By constructing decision trees in this way, a large number of redundant calculations can be eliminated, which can greatly improve the efficiency of decision tree construction, while theoretically it will not affect the performance of decision trees.

[0068] This embodiment designs and utilizes severe convective weather characteristic data, geographical characteristic data, and power grid state characteristic data to train an optimizable sparse decision tree, enabling it to predict the probability of overhead line power outages under severe convective weather based on input information. The decision tree algorithm itself has good handling capabilities for the differences in the dimensions of input data, and can well cope with the numerical differences of weather factors, geographical factors, etc. Furthermore, when adding leaves to form subtrees at each step, the generation of each new tree comes from the leaves that can be further divided in the old tree, rather than from the leaves that cannot be further divided in the old tree. Therefore, the decision tree in this embodiment is an asymmetric sparse optimizable decision tree, which can efficiently construct decision trees and has a good ability to prevent overfitting, and can achieve good prediction accuracy in actual deployment.

[0069] In one embodiment, in step S10, the severe convective meteorological characteristic data includes the maximum value of severe convective radar intensity f1, the magnitude of lightning current f2, and the number of lightning return strokes f3, wherein the maximum value of severe convective radar intensity f1 is the maximum value within a set time period, the magnitude of lightning current f2 is the average value within a set time period, and the number of radar return strokes f3 is the number of lightning return strokes within a set time period.

[0070] The geographic feature data includes local geographic features and global geographic features. Local geographic features include elevation l1, slope l2, and aspect l3; global geographic features include longitude m1 and latitude m2.

[0071] The power grid status characteristic data includes the number of load cycles n1, the number of overload cycles n2, and the number of defects n3.

[0072] It should be noted that, according to the meteorological information described in this embodiment, the meteorological characteristics of severe convective weather can be covered by the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning return strokes, covering most of the information.

[0073] When selecting geographic feature information, both local geographic features, including altitude, slope, and aspect, and global geographic location (latitude and longitude) are considered. Given that overhead power lines in each region are affected differently by historical factors, latitude and longitude information needs to be included in the feature information to differentiate overhead power lines in different regions.

[0074] When selecting power grid condition characteristic data, the stability of overhead lines is considered, including information such as the number of overloads, heavy loads, and defects, which can largely describe the stability of the power grid. For example, a higher number of overloads leads to greater grid losses and a relatively higher probability of power outages. These characteristics are correlated with overhead line power outages.

[0075] It should be noted that, since this embodiment uses an optimizable sparse decision tree for overhead line outage prediction, the algorithm model implicitly has the ability to distinguish the degree of influence of different feature dimensions on the overhead line outage status, so there is no need to normalize the feature data.

[0076] Furthermore, among the feature information, meteorological information changes in real time and has a crucial impact on the overhead line outage prediction results. This embodiment designs a long-term data feature extraction algorithm for meteorological information over a one-hour interval for a specific overhead line, and performs comprehensive processing to obtain the feature information. Specifically:

[0077] (1) Maximum intensity of strong convective radar

[0078] f1 = max{f11, f12, ..., f1} n}

[0079] Among them, f11, f12...f1 n This is the sampled data of the maximum intensity of strong convection radar over the overhead line within one hour. Considering that the "maximum intensity of strong convection radar" has a certain breakdown effect, the maximum value over a period of time is used to represent the maximum intensity of strong convection radar over one hour.

[0080] (2) Magnitude of lightning current

[0081] f2 = sum{f21, f22...f2} n} / n

[0082] Among them, f21, f22...f2 n This is a sample of the lightning current magnitude of the overhead line within one hour. Considering that the "lightning current magnitude" has a certain cumulative effect, the average value over a period of time is used to represent the lightning current magnitude within one hour.

[0083] (3) Number of lightning strikes

[0084] The number of lightning strikes is already a cumulative data over a period of time. This patent uses the number of lightning strikes within 1 hour as the f3 value.

[0085] Furthermore, for overhead lines in a specific area, this embodiment uses feature = (f1, f2, f3, l1, l2, l3, m1, m2, n1, n2, n3) to represent the data. To obtain the feature information needed for overhead line outage prediction, this patent collects big data information from tens of thousands of overhead lines in various regions over the past few years, including meteorological information, geographical information, and power grid status information, from which the feature information and corresponding labels required for training and validation are extracted.

[0086] The following describes the label information in detail. Considering that the overhead line power outage status is a time-series information, this embodiment designs the following label extraction algorithm to obtain the pairing of feature information and label information within a certain period of time.

[0087] Specifically, step S10 includes the following steps:

[0088] The corresponding feature data is obtained from a period of time before the power outage signal.

[0089] Based on the label value range, the pairing of feature data and label information within the time period is obtained, wherein the label value range is:

[0090]

[0091] It should be noted that the power grid outage status is monitored in real time. Once the power grid fails, the central control station can obtain the outage signal. Therefore, the outage status in the big data information is a real-time signal. In this embodiment, the corresponding feature information is obtained from the outage signal y within one hour before the outage to generate corresponding feature data, thus completing the matching between y and the feature.

[0092] Furthermore, through the above data acquisition process, 100,000 data samples were generated for the training and verification of the algorithm. In this embodiment, a 9:1 ratio of training to verification data was used, with a training set of 90,000 samples and a test set of 10,000 samples, which were used to train and verify the sparse optimizable decision tree, respectively.

[0093] In one embodiment, in step S20, the initialized decision tree is:

[0094] d=(d un δ un d split δ split (K, H)

[0095] Where, dun For the leaves that do not change in d, δ un To avoid changing the prediction result corresponding to the leaf, d split For the leaves in d that need to be further divided, δ split The features corresponding to the leaves to be further subdivided are H, where H is the total number of leaves and K is the delimiter.

[0096] It should be noted that this embodiment assumes a tree with H leaves, d = (p1, p2, p3..., p...). K p K+1 , ..., p H The expression ) indicates that the tree is further divided when the first K leaves cannot be used to generate a subtree, and the remaining HK leaves can be further divided.

[0097] Furthermore, it can be rewritten as d = (d un δ un d split δ split ,K,H),d un For the leaves that do not change in d, δ un To avoid changing the prediction result corresponding to the leaf, d split For the leaves in d that need to be further divided, δ split The features corresponding to the leaves to be further subdivided are H, where H is the total number of leaves and K is the delimiter.

[0098] Based on this, after further dividing the leaves, a new tree d′=(d′) is obtained. un ,δ′ un ,d′ split ,δ′ split Since the decision tree is constructed hierarchically based on the input data of each different feature, the correlation between features is reflected in the construction of the new tree, but not in the leaves of the old tree. Therefore, the generation of each new tree comes from the leaves of the old tree that can be further divided.

[0099] In one embodiment, the objective function R(d, x, y) is:

[0100] R(d,x,y)=l(d,x,y)+λH d

[0101] Where d represents the parameters of the decision tree, x represents the input feature data, y represents the label corresponding to the feature data, l(d, x, y) represents optimizing the decision tree parameters directly based on the label, and λH d This represents the regularization term.

[0102] It should be noted that for each specific decision tree construction, d′=(d′ un ,δ′un ,d′ split ,δ′ split Given K′, H′, its objective function can be defined as R(d, x, y) = l(d, x, y) + λH. d In the decision tree model, the parameters are in d, x represents the data, y represents its corresponding label, l(d, x, y) represents optimizing the model parameters directly based on the labels, and λH d The regularization term prevents overfitting during optimization iterations. Based on this loss function, the model can be iteratively optimized. Each step of the decision tree, adding leaves to form a subtree, utilizes this predetermined loss function for optimization, further improving the performance of the constructed decision tree.

[0103] In one embodiment, step S40 includes the following steps:

[0104] S41. Determine the status of the overhead line based on the output confidence level of each leaf;

[0105] S42. If the overhead line is in a power outage state, then the leaf is determined to be a leaf that can generate a new tree;

[0106] S43. If the overhead line is in a normal state, then the leaf is determined to be a leaf that cannot generate a new tree.

[0107] Specifically, considering the classification issue, a confidence level greater than 0.5 is considered a normal state, while a confidence level less than 0.5 is considered a power outage state.

[0108] It should be noted that by training the sparse decision tree using the above training dataset, a power outage prediction algorithm for overhead lines during severe convective weather based on an optimizable sparse decision tree was obtained, achieving a precision of 90% and a recall of 70% on the validation set.

[0109] Specifically, the decision tree training and optimization process in this embodiment is as follows:

[0110] Input: Training dataset for overhead power lines D-90000

[0111] Output: Optimizable sparse decision tree

[0112] The training process is as follows:

[0113] (1) Initialize a decision tree d = (d un δ un d split δ split (K, H);

[0114] (2) Based on the characteristics of the data, feature = (f1, f2, f3, 11, l2, 13, m1, m2, n1, n2, n3), and the depth of the decision tree is 11;

[0115] (3) Based on the input features f i ={x i y i}, calculate R(d, x, y) and the output confidence of each leaf in d;

[0116] (4) After multiple iterations, select the leaves that can generate a new tree based on the output confidence of each leaf in d;

[0117] (5) Based on the new leaves, generate a new tree d′=(d′ un ,δ′ un ,d′ split ,δ′ split (, K′, H′);

[0118] (6) Proceed to step (3) and repeat the process until all features have been traversed, finally obtaining the sparse decision tree d. final_model =(d final H).

[0119] In addition, such as Figure 5 As shown, the second embodiment of the present invention proposes a device for predicting the risk of power outages from overhead lines during severe convective weather. The device includes:

[0120] The acquisition module 10 is used to acquire a training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective weather feature data, geographical feature data, and power grid status feature data.

[0121] Initialization module 20 is used to initialize the decision tree;

[0122] The training module 30 is used to take the training data and its corresponding data labels as input to the decision tree, calculate its objective function R(d, x, y) and the output confidence of each leaf in the decision tree;

[0123] The decision tree update module 40 is used to select leaves that can generate a new tree based on the output confidence of each leaf, and generate a new decision tree.

[0124] Output module 50 is used to obtain a sparse decision tree until all feature data in the training dataset has been traversed;

[0125] The prediction module 60 is used to calculate the probability of overhead line power outage in the target area using the sparse decision tree.

[0126] In one embodiment, the severe convective weather characteristic data includes the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning returns, wherein the maximum intensity of severe convective radar is the maximum value within a set time period, the magnitude of lightning current is the average value within a set time period, and the number of radar returns is the number of lightning returns within a set time period.

[0127] The geographic feature data includes local geographic features and global geographic features;

[0128] The power grid status characteristic data includes the number of load cycles, the number of overload cycles, and the number of defects.

[0129] In one embodiment, the initialized decision tree is:

[0130] d=(d un δ un d split δ split (K, H)

[0131] Where, d un For the leaves that do not change in d, δ un To avoid changing the prediction result corresponding to the leaf, d split For the leaves in d that need to be further divided, δ split The features corresponding to the leaves to be further subdivided are H, where H is the total number of leaves and K is the delimiter.

[0132] In one embodiment, the objective function R(d, x, y) is:

[0133] R(d,x,y)=l(d,x,y)+λH d

[0134] Where d represents the parameters of the decision tree, x represents the input feature data, y represents the label corresponding to the feature data, l(d, x, y) represents optimizing the decision tree parameters directly based on the label, and λH d This represents the regularization term.

[0135] This embodiment designs and utilizes severe convective weather characteristic data, geographical characteristic data, and power grid state characteristic data to train an optimizable sparse decision tree, enabling it to predict the probability of overhead line power outages under severe convective weather based on input information. The decision tree algorithm itself has good handling capabilities for the differences in the dimensions of input data, and can well cope with the numerical differences of weather factors, geographical factors, etc. Furthermore, when adding leaves to form subtrees at each step, the generation of each new tree comes from the leaves that can be further divided in the old tree, rather than from the leaves that cannot be further divided in the old tree. Therefore, the decision tree in this embodiment is an asymmetric sparse optimizable decision tree, which can efficiently construct decision trees and has a good ability to prevent overfitting, and can achieve good prediction accuracy in actual deployment.

[0136] It should be noted that other embodiments or implementation methods of the overhead power outage risk prediction device for severe convective weather described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0137] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0138] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0139] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0140] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0141] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the risk of overhead power outages during severe convective weather, characterized in that, The method includes: Obtain a training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective meteorological feature data, geographical feature data, and power grid status feature data. Initialize the decision tree as follows ,in, The leaves that remain unchanged in d To avoid changing the prediction result corresponding to the leaf, The leaves in d that need to be further divided. The features corresponding to the leaves to be further subdivided are defined by H, where H is the total number of leaves and K is the delimiter. The training data and its corresponding data labels are used as input to the decision tree to calculate its objective function. and the output confidence of each leaf in the decision tree, the objective function for: in, The parameters of the decision tree are represented. This represents the input feature data. The label represents the feature data. This means optimizing the decision tree parameters directly based on the labels. Represents the regularization term; Based on the output confidence of each leaf, select leaves that can generate a new tree and generate a new decision tree, including determining the status of the overhead line based on the output confidence of each leaf; if the status of the overhead line is a power outage, then determine that the leaf is a leaf that can generate a new tree; if the status of the overhead line is a normal status, then determine that the leaf is a leaf that cannot generate a new tree. The process continues until all feature data in the training dataset has been traversed to obtain a sparse decision tree. The probability of overhead line power outage in the target area is calculated using the sparse decision tree.

2. The method for predicting the risk of power outages from overhead lines during severe convective weather as described in claim 1, characterized in that, The severe convective meteorological characteristic data includes the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning returns. The maximum intensity of severe convective radar is the maximum value within a set time period, the magnitude of lightning current is the average value within a set time period, and the number of lightning returns is the number of lightning returns within a set time period. The geographic feature data includes local geographic features and global geographic features; The power grid status characteristic data includes the number of load cycles, the number of overload cycles, and the number of defects.

3. The method for predicting the risk of power outages from overhead lines during severe convective weather as described in claim 2, characterized in that, The acquisition of the training dataset includes: The corresponding feature data is obtained from a period of time before the power outage signal; Based on the label value range, the pairing of feature data and label information within the time period is obtained, wherein the label value range is: 。 4. A device for predicting the risk of power outages from overhead lines during severe convective weather, characterized in that, The device includes: The acquisition module is used to acquire a training dataset, which includes feature data and its corresponding data labels. The training data includes severe convective meteorological feature data, geographical feature data, and power grid status feature data. The initialization module is used to initialize the decision tree. ,in, The leaves that remain unchanged in d To avoid changing the prediction result corresponding to the leaf, The leaves in d that need to be further divided. The features corresponding to the leaves to be further subdivided are defined by H, where H is the total number of leaves and K is the delimiter. The training module is used to take the training data and its corresponding data labels as input to the decision tree and calculate its objective function. and the output confidence of each leaf in the decision tree, the objective function for: in, The parameters of the decision tree are represented. This represents the input feature data. The label represents the feature data. This means optimizing the decision tree parameters directly based on the labels. Represents the regularization term; The decision tree update module is used to select leaves that can generate a new tree based on the output confidence of each leaf, and generate a new decision tree. This includes determining the status of the overhead line based on the output confidence of each leaf; if the status of the overhead line is a power outage, then the leaf is determined to be a leaf that can generate a new tree; if the status of the overhead line is a normal status, then the leaf is determined to be a leaf that cannot generate a new tree. The output module is used to obtain a sparse decision tree until all feature data in the training dataset has been traversed. The prediction module is used to calculate the probability of overhead line power outages in the target area using the sparse decision tree.

5. The overhead power outage risk prediction device for severe convective weather as described in claim 4, characterized in that, The severe convective meteorological characteristic data includes the maximum intensity of severe convective radar, the magnitude of lightning current, and the number of lightning returns. The maximum intensity of severe convective radar is the maximum value within a set time period, the magnitude of lightning current is the average value within a set time period, and the number of lightning returns is the number of lightning returns within a set time period. The geographic feature data includes local geographic features and global geographic features; The power grid status characteristic data includes the number of load cycles, the number of overload cycles, and the number of defects.

Images