Real-time power failure judgment method, device and equipment, storage medium and product

Abstract

The invention discloses a real-time power failure judgment method and device, equipment, a storage medium and a product. The real-time power failure judgment method comprises the steps of obtaining power failure characteristics and basic attribute characteristics in a mobile network and a power network; according to the power failure features and the basic attribute features, a clustering algorithm is adopted to divide the space region into a plurality of scenes, and clustering scene data is obtained; performing power failure hidden danger prediction based on the power failure characteristics, the basic attribute characteristics and the clustering scene data to obtain power failure hidden danger characteristics and a power failure judgment strategy of each scene; extracting power failure hidden danger key features from the power failure hidden danger features, and extracting real-time power failure key features from the real-time data of the power failure features; and according to a power failure judgment strategy, carrying out fusion calculation on the power failure hidden danger key features and the real-time power failure key features to obtain a power failure judgment result. According to the invention, the accuracy of real-time power failure judgment is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a real-time power outage determination method, apparatus, device, storage medium, and product. Background Art

[0002] Current real-time outage detection technologies rely on a single data source, typically analyzing and determining outages based solely on relevant power system data. This lack of user-side network data directly reflecting outages results in low accuracy. Furthermore, existing real-time outage detection solutions primarily rely on correlation analysis and statistics of power data, using simple cross-validation of business data and user complaint data to determine if a region has experienced an outage. This can lead to delays and misjudgments in outage detection results, leaving much room for improvement. Summary of the Invention

[0003] To solve the above technical problems, the present application provides a real-time power outage determination method, device, equipment, storage medium and product, which improve the accuracy of real-time power outage determination.

[0004] This application provides a real-time power outage determination method, including:

[0005] Obtain power outage characteristics and basic attribute features in mobile networks and power networks;

[0006] Based on the power outage characteristics and the basic attribute characteristics, a clustering algorithm is used to divide the spatial area into several scenes to obtain clustered scene data;

[0007] Based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data, power outage hidden danger prediction is performed to obtain power outage hidden danger characteristics and power outage determination strategies for each scenario;

[0008] Extracting key features of power outage hidden dangers from the power outage hidden danger features, and extracting real-time key features of power outages from real-time data of the power outage features;

[0009] According to the power outage determination strategy, the power outage hidden danger key features and the real-time power outage key features are fused and calculated to obtain a power outage determination result.

[0010] As an improvement to the above solution, the power outage hidden danger prediction is performed based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data to obtain the power outage hidden danger characteristics and the power outage determination strategy for each scenario, including:

[0011] Extracting power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data;

[0012] Inputting the power outage performance information, the basic attribute information, and the clustering scenario information into a preset multi-class feature encoder to obtain global fusion information; the multi-class feature encoder is composed of a plurality of first encoding modules, each of which is composed of a global pooling layer, a self-attention network, and a fully connected network;

[0013] According to the power outage performance information, the basic attribute information and the global fusion information, a first decoder is used to perform feature decoding and prediction to obtain power outage hidden danger features;

[0014] According to the global fusion information, the basic attribute information and the clustering scenario information, a second decoder is used to perform feature decoding prediction to obtain a power outage determination strategy for each scenario.

[0015] As an improvement to the above solution, the extracting of power outage performance information, basic attribute information and clustering scenario information based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data includes:

[0016] Inputting the power outage hidden danger data and real-time data in the power outage feature into corresponding sub-encoders respectively to obtain a first code corresponding to the power outage hidden danger data and a second code corresponding to the real-time data; splicing the first code and the second code and inputting them into a preset multi-category feature encoder to obtain power outage performance information;

[0017] Inputting the spatial feature variable, device feature variable, and user feature variable in the basic attribute features into corresponding sub-encoders respectively to obtain feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable respectively; splicing the feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable into a preset multi-category feature encoder to obtain a third code; inputting the climate feature variable in the basic attribute features into a corresponding sub-encoder to obtain a fourth code; splicing the third code and the fourth code into a preset multi-category feature encoder to obtain basic attribute information;

[0018] Inputting the data of each scene in the clustered scene data into the corresponding sub-encoder respectively to obtain the feature code corresponding to each scene; splicing the feature codes corresponding to each scene and inputting them into the preset multi-category feature encoder to obtain clustered scene information;

[0019] Among them, each sub-encoder is composed of several superimposed second encoding modules, and the second encoding module is composed of a self-attention network, a feedforward network and a feedforward dimensionality reduction network.

[0020] As an improvement to the above solution, the first decoder is used to perform feature decoding prediction based on the power outage performance information, the basic attribute information, and the global fusion information to obtain power outage hidden danger features, including:

[0021] The power outage performance information and the basic attribute information are spliced ​​and then subjected to feature fusion decoding to obtain a first feature;

[0022] Performing feature dimension reduction and decoding on the global fusion information to obtain a second feature;

[0023] Inputting the first feature and the second feature into an attention network to obtain a power outage hidden danger attention feature;

[0024] The power outage hidden danger attention feature is input into a feedforward network to predict the credibility of hidden danger occurrence, the power outage hidden danger area, and the power outage hidden danger network to obtain the power outage hidden danger feature.

[0025] As an improvement to the above solution, the method of using a second decoder to perform feature decoding prediction based on the global fusion information, the basic attribute information, and the clustering scenario information to obtain a power outage determination strategy for each scenario includes:

[0026] The basic attribute information and the clustering scene information are spliced ​​together and then feature fusion decoding is performed to obtain a third feature;

[0027] Performing feature dimensionality reduction and decoding on the global fusion information to obtain a fourth feature;

[0028] Inputting the third feature and the fourth feature into the attention network to obtain a scene strategy attention feature;

[0029] The scene strategy attention features are input into the feedforward network to predict the power outage determination strategy, and the power outage determination strategy for each scene is obtained.

[0030] As an improvement to the above solution, the step of extracting key features of power outage hidden dangers from the power outage hidden danger features and extracting real-time key features of power outages from real-time data of the power outage features includes:

[0031] Inputting the power outage hidden danger features into a preset first neural network for feature extraction to obtain key features of the power outage hidden danger;

[0032] The real-time data of the power outage characteristics is input into a preset second neural network for feature extraction to obtain real-time power outage key features; wherein the first neural network and the second neural network are parallel network structures.

[0033] The present application also provides a real-time power outage determination device, comprising:

[0034] A data acquisition module is used to obtain power outage characteristics and basic attribute characteristics in mobile networks and power networks;

[0035] A scene clustering module, configured to divide a spatial area into a number of scenes using a clustering algorithm based on the power outage characteristics and the basic attribute characteristics, thereby obtaining clustered scene data;

[0036] A power outage prediction module, configured to predict power outage hazards based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data, and obtain power outage hazard characteristics and power outage determination strategies for each scenario;

[0037] A feature extraction module is used to extract key features of power outage hidden dangers from the power outage hidden danger features, and extract real-time key features of power outages from the real-time data of the power outage features;

[0038] The power outage determination module is used to perform a fusion calculation on the power outage hidden danger key features and the real-time power outage key features according to the power outage determination strategy to obtain a power outage determination result.

[0039] The present application also provides a computer device, including a processor and a memory, wherein a computer program is stored in the memory, and the computer program is configured to be executed by the processor, and when the processor executes the computer program, any of the above-mentioned real-time power outage determination methods is implemented.

[0040] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any of the above-mentioned real-time power outage determination methods.

[0041] The present application also provides a computer program product, including a computer program, which implements any of the above-mentioned real-time power outage determination methods when executed by a processor.

[0042] Compared with the prior art, the real-time power outage determination method, device, equipment, storage medium and product provided by this application have the following beneficial effects:

[0043] By acquiring power outage characteristics and basic attribute characteristics in mobile networks and power networks, and integrating mobile network data and power network data, a more comprehensive power outage judgment can be made; a clustering algorithm is used to divide the spatial area into several scenes to obtain clustered scene data, which is conducive to power outage judgment for each type of scene, and improves the accuracy and flexibility of real-time power outage judgment; power outage hidden dangers are predicted based on the power outage characteristics, the basic attribute characteristics and the clustered scene data, and power outage hidden danger characteristics and power outage judgment strategies for each scene are obtained, which can integrate multiple data and identify power network line hidden dangers, and improve the accuracy of real-time power outage judgment; furthermore, power outage hidden danger key features are extracted from the power outage hidden danger features, and real-time power outage key features are extracted from the real-time data of the power outage features, and the power outage hidden danger key features are fused and calculated with the real-time power outage key features according to the power outage judgment strategy to obtain a power outage judgment result, which can efficiently integrate real-time power outage data and predicted power outage hidden danger data to achieve accurate real-time power outage judgment, improve the accuracy of real-time power outage judgment and reduce the misjudgment rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a real-time power outage determination method provided by an embodiment of the present application;

[0045] Figure 2 This is a schematic diagram of the structure of the power outage hidden danger prediction model provided by the embodiment of the present application;

[0046] Figure 3 This is a flowchart of obtaining power outage performance information provided by an embodiment of the present application;

[0047] Figure 4 This is a schematic diagram of the process of obtaining basic attribute information provided by an embodiment of the present application;

[0048] Figure 5 This is a schematic diagram of a process for obtaining clustering scene information provided by an embodiment of the present application;

[0049] Figure 6 is a structural diagram of the second encoding module provided in an embodiment of the present application;

[0050] Figure 7 Schematic diagram of the structure of the multi-class feature encoder provided in an embodiment of the present application;

[0051] Figure 8 is a schematic structural diagram of a first decoder provided in an embodiment of the present application;

[0052] Figure 9 is a schematic structural diagram of a second decoder provided in an embodiment of the present application;

[0053] Figure 10Schematic diagram of the structure of the dimensionality reduction decoding submodule provided in an embodiment of the present application;

[0054] Figure 11 This is a schematic diagram of the structure of the information fusion decoding module provided in an embodiment of the present application;

[0055] Figure 12 This is a structural diagram of a real-time power outage determination device provided in an embodiment of the present application;

[0056] Figure 13 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0058] See also Figure 1 , Figure 1 : is a flow chart of a real-time power outage determination method provided by an embodiment of the present application. The real-time power outage determination method includes:

[0059] S1: Obtain power outage characteristics and basic attribute characteristics in mobile networks and power networks;

[0060] S2: Divide the spatial area into several scenes using a clustering algorithm based on the power outage characteristics and the basic attribute characteristics to obtain clustered scene data;

[0061] S3: Based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data, predict power outage hidden dangers to obtain power outage hidden danger characteristics and power outage determination strategies for each scenario;

[0062] S4: extracting key features of power outage hidden dangers from the power outage hidden danger features, and extracting real-time key features of power outages from the real-time data of the power outage features;

[0063] S5: According to the power outage determination strategy, the power outage hidden danger key features and the real-time power outage key features are fused and calculated to obtain a power outage determination result.

[0064] Specifically, in step S1, the control platform acquires feature data related to power outages in the mobile network and the power network, including power outage features and basic attribute features. The power outage features include power outage potential data and real-time data, and the power outage potential data is data from the historical period.

[0065] Regarding power outage hazard data, the power outage hazard data in the mobile network includes the duration of the power outage of the mobile optical modem, the heartbeat interval of the mobile optical modem, etc.; the power outage hazard data in the power network includes the total power difference of the power equipment, the voltage difference of the power equipment, etc.

[0066] If the historical period is set to the past 10 days, the calculation formula for the duration of the mobile optical modem power outage is as follows:

[0067] H=h0+h1+…+h n

[0068] Where H is the duration of the mobile optical modem power outage, h0, h1, ..., h n They are the duration of each mobile optical modem power outage in the past 10 days.

[0069] The calculation formula for the heartbeat interval of a mobile optical modem is as follows:

[0070]

[0071] Where t is the heartbeat interval of the mobile optical modem, t0, t1, ..., t m-1 They are respectively the heartbeat intervals of each mobile optical modem in the past 10 days.

[0072] The calculation formula for the total power difference of power equipment is as follows:

[0073]

[0074] Where Q is the total power difference of the power equipment, Q0, Q1, ..., Q8, Q9 are the total power of the power equipment from the current day to the previous 9 days, Q 10 , Q 11 ,…,Q 18 , Q 19 They are the total daily electricity consumption of power equipment from the previous 10 days to the previous 19 days.

[0075] The calculation formula for the voltage difference of power equipment is as follows:

[0076]

[0077] Where U is the voltage difference of the power equipment, and U0, U1, U2, …, U8, and U9 are the daily average voltages of the power equipment of the same equipment from the current day to the previous 9 days.

[0078] For real-time data, the real-time data in the mobile network includes the standard deviation of the time when the mobile optical modem power outage alarm occurs; the real-time data in the power network includes the voltage and current of power equipment.

[0079] The calculation formula for the standard deviation of the mobile optical modem power outage alarm occurrence time is as follows:

[0080]

[0081] Where w is the standard deviation of the time when the mobile optical modem power outage alarm occurs, w0, w1, ..., w s-1 They are respectively the time when the power outage alarm of each mobile optical modem occurs within 1 minute.

[0082] The calculation formula for the voltage of power equipment is as follows:

[0083]

[0084] Where u is the voltage of the power equipment, u0, u1, ..., u k-1 They are the voltage of each electrical device detected.

[0085] The calculation formula for the current of power equipment is as follows:

[0086]

[0087] Where, I is the voltage of the power equipment, I0, I1, ..., I k-1 They are the current of each electrical device detected.

[0088] Furthermore, depending on the data domain, the basic attribute characteristics include: spatial characteristic variables, device characteristic variables, user characteristic variables, climate characteristic variables, etc.

[0089] Among them, spatial characteristic variables include: spatial type, longitude, latitude and other data;

[0090] Device characteristic variables include: device name, space, device location and other data;

[0091] User characteristic variables include: user age, gender, occupation and other data;

[0092] Climate characteristic variables include: rainfall, snowfall and other data.

[0093] It should be noted that non-numeric features within the basic attribute features, such as space type, device name, space affiliation, gender, and occupation, are represented using one-hot encoding. For example, if the user's occupations are doctor, teacher, and programmer, these three items are selected as feature variables: 1 for being a doctor and 0 for not being a doctor; 1 for being a teacher and 0 for not being a teacher; and 1 for being a programmer and 0 for not being a programmer. Numerical features are represented using their original values.

[0094] As one of the optional embodiments, after acquiring the power outage characteristics and basic attribute characteristics in the mobile network and the power network, the method further includes:

[0095] Performing data cleaning on the power outage features and the basic attribute features;

[0096] By evaluating the correlation between each feature and the target variable, the features whose correlation is less than a preset correlation threshold are screened out;

[0097] The importance of each feature is obtained by evaluating the importance of each feature, and the features whose importance is less than a preset importance threshold are screened out;

[0098] The power outage characteristics and the basic attribute characteristics are standardized.

[0099] Specifically, after obtaining raw data from the mobile network and the power grid in step S1, the data needs to be preprocessed to ensure that the processed data conforms to the format of subsequent model input. Data cleaning is first performed, specifically identifying missing values, outliers, and duplicates in the data. Missing values ​​are filled using the mean method, and outliers and duplicates are removed. Feature selection is then performed to select the features that are most relevant and important for power outage risk prediction and real-time power outage prediction. This specifically includes two steps: first, using a statistical model to evaluate the correlation between each feature and the target variable (the target variable is the power outage risk and the real-time power outage probability), filtering out features with correlations less than a preset correlation threshold, that is, eliminating features with no obvious correlation with the target variable to reduce the impact of noise and redundant information; second, using a feature importance assessment method to evaluate and filter features, filtering out features with feature importance less than a preset importance threshold, that is, eliminating features with low contribution to subsequent prediction and real-time power outage determination. Furthermore, the filtered features are normalized to eliminate the influence of different dimensions and value ranges. Specifically, the Min-Max method can be used for data normalization.

[0100] Furthermore, step S2 clusters scenarios based on the filtered outage features and basic attribute features. Specifically, a quasi-hierarchical clustering algorithm is used to divide the spatial area into several types of scenarios, generating clustered scenario data to improve the accuracy of subsequent outage determination strategy predictions. Each scenario considers factors such as the space type, weather type, number of rooms / users in the space, number of power outage alarms, power outage alarm frequency, number of optical modem heartbeats, optical modem heartbeat frequency, total power difference, flow difference, and voltage difference. These factors constitute the clustered scenario data.

[0101] Specifically, in a spatial area where power outage detection is required, the spatial area is divided into multiple sub-areas, each of which has corresponding power outage characteristics and basic attribute characteristics. This embodiment hierarchically clusters the sub-areas based on the power outage characteristics and basic attribute characteristics, and divides the spatial area into multiple scenarios. Each type of scenario includes several sub-areas with the same or similar characteristics. Furthermore, in subsequent strategy prediction, this application aggregates the rules or strategies corresponding to all sub-areas under each type of scenario, and obtains a representative strategy as the power outage determination strategy for that scenario. Therefore, when the power outage determination is finally performed, the corresponding power outage determination strategy can be selected according to different scenarios, thereby improving the accuracy of real-time power outage determination.

[0102] As one of the optional embodiments, the power outage hidden danger prediction is performed based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data to obtain the power outage hidden danger characteristics and the power outage determination strategy for each scenario, including:

[0103] Extracting power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data;

[0104] Inputting the power outage performance information, the basic attribute information, and the clustering scenario information into a preset multi-class feature encoder to obtain global fusion information; the multi-class feature encoder is composed of a plurality of first encoding modules, each of which is composed of a global pooling layer, a self-attention network, and a fully connected network;

[0105] According to the power outage performance information, the basic attribute information and the global fusion information, a first decoder is used to perform feature decoding and prediction to obtain power outage hidden danger features;

[0106] According to the global fusion information, the basic attribute information and the clustering scenario information, a second decoder is used to perform feature decoding prediction to obtain a power outage determination strategy for each scenario.

[0107] Specifically, in step S3, a power outage hidden danger prediction model is constructed, such as Figure 2 As shown in the figure, the power outage hazard prediction model extracts the coded fusion information of the input features through the feature encoding fusion module, and then outputs the prediction results through the feature decoding prediction module. The input features of the power outage hazard prediction model include power outage characteristics, basic attribute characteristics, and clustering scenario data. The coded fusion information includes power outage manifestation information, basic attribute information, clustering scenario information, and global fusion information. The prediction results include power outage hazard characteristics and power outage determination strategies for each scenario. The power outage hazard characteristics include the power outage hazard area, the power outage hazard network (line), and the credibility of the hazard occurrence.

[0108] In the feature encoding fusion module, the power outage features, basic attribute features, and clustering scenario data are first respectively passed through the corresponding information extraction module (encode module) to obtain power outage performance information, basic attribute information, and clustering scenario information. The three types of extracted information are then concatenated (contact) and input into the multi-class feature encoder (multiple encoder) for fusion encoding to obtain global fusion information.

[0109] As one of the optional embodiments, the extracting of power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data includes:

[0110] Inputting the power outage hidden danger data and real-time data in the power outage feature into corresponding sub-encoders respectively to obtain a first code corresponding to the power outage hidden danger data and a second code corresponding to the real-time data; splicing the first code and the second code and inputting them into a preset multi-category feature encoder to obtain power outage performance information;

[0111] Inputting the spatial feature variable, device feature variable, and user feature variable in the basic attribute features into corresponding sub-encoders respectively to obtain feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable respectively; splicing the feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable into a preset multi-category feature encoder to obtain a third code; inputting the climate feature variable in the basic attribute features into a corresponding sub-encoder to obtain a fourth code; splicing the third code and the fourth code into a preset multi-category feature encoder to obtain basic attribute information;

[0112] Inputting the data of each scene in the clustered scene data into the corresponding sub-encoder respectively to obtain the feature code corresponding to each scene; splicing the feature codes corresponding to each scene and inputting them into the preset multi-category feature encoder to obtain clustered scene information;

[0113] Among them, each sub-encoder is composed of several superimposed second encoding modules, and the second encoding module is composed of a self-attention network, a feedforward network and a feedforward dimensionality reduction network.

[0114] See also Figure 3The power outage features include power outage potential data and real-time data. The power outage potential data is input into the superimposed subcode encoder (subcode encoder×3) to extract the first code corresponding to the power outage potential data. The real-time data is input into the superimposed subcode encoder (subcode encoder×3) to extract the second code corresponding to the real-time data. The first and second codes are then concatenated (contact) and input into the multi-class feature encoder (multiple encoder) to mine power outage performance information containing the correlation relationship between the two types of feature data.

[0115] See also Figure 4 The basic attribute features include spatial, device, user, and climate variables. First, each basic attribute feature is fed into the corresponding superimposed subcoder (subcode encoder × 3) to extract the feature code. Because spatial, device, and user variables have a certain degree of stability, the feature codes of these three basic attribute features are concatenated (concatenated) and fed into a multiple feature encoder (multiple encoder). This further explores the correlations between the three features to obtain a third code. Furthermore, the feature code corresponding to the climate variable (i.e., the fourth code) is concatenated (concatenated) with the third code and fed into a multiple feature encoder (multiple encoder) to extract the basic attribute information.

[0116] See also Figure 5 The clustered scene data contains data from various scenes. First, the data of different scenes are input into the corresponding superimposed subcoder (subcode encoder×3) to extract feature codes. Then, the feature codes of each scene are concatenated (contact) and input into the multi-class feature encoder (multiple encoder) for fusion coding to explore the intrinsic relationship between the scene data and finally obtain the clustered scene information.

[0117] Furthermore, the obtained power outage performance information, basic attribute information and clustering scene information are spliced ​​(contact) and input into a multiple feature encoder (multiple encoder) for fusion encoding to obtain global fusion information.

[0118] Specifically, the sub-encoders used above have the same basic structure. Each sub-encoder is composed of three superimposed second encoding modules (subcode encoders). The specific structure of the second encoding module (subcode encoder) is as follows: Figure 6As shown in the figure, it consists of a self-attention network, a feedforward network (fully-connected layers), and a feedforward dimensionality reduction network (fully-connected layers). Among them, the feedforward dimensionality reduction network can compress feature dimensions, suppress noise, and reduce information redundancy.

[0119] Furthermore, the basic structure of the multiple feature encoders used in this application is the same, such as Figure 7 As shown, each multi-class feature encoder is composed of several identical first encoding modules (encoder) ( Figure 7 (The figure is composed of three stacked encoders). Each first encoder module consists of a global pooling layer, a self-attention network, and a fully connected network. The global pooling layer extracts global information about each feature class, learns the relationship between each feature class through the connected fully connected network, and then weights the input features. The self-attention network can explore the correlation between features and extract richer comprehensive feature information.

[0120] Specifically, in each first encoding module (encoder), the input features pass through the global pooling layer and the fully connected network in sequence before entering the weighted layer, then through the self-attention network and the residual connection, and finally through the fully connected layer and the residual connection to obtain the feature output. Among them, the calculation formula of the global pooling layer (global avgpool) for each type of feature is as follows:

[0121]

[0122] Among them, z c is the calculation result of the c-th type of feature, H is the number of features of each type of feature, W is the dimension of each feature, F c It is the c-type feature.

[0123] The first fully-connected network consists of two fully-connected layers and the activation function after them. The activation function after the first fully-connected layer is the ReLU function, and the activation function after the second fully-connected layer is the sigmoid function. The specific calculation formula is as follows:

[0124] a=sigmoid(W2ReLU(W1z))

[0125] Among them, a is the weighted value of each type of feature, z is the calculation result of each type of feature output by the global pooling layer, and W1 and W2 are the weight parameters of the two fully connected layers.

[0126] The calculation formula of the weighted layer is as follows:

[0127]

[0128] in, is the weighted feature of the c-th category, a c is the weighted value of the c-th feature output by the excitation layer.

[0129] After further feature extraction through the self-attention network, we get:

[0130]

[0131] Among them, F att is the output feature of the self-attention network, W Q 、W K 、W V is the weight matrix corresponding to the self-attention network parameters, Q, K, V are the self-attention network parameters, d k is the parameter dimension.

[0132] According to F att Perform residual connection to obtain feature F r :

[0133] F r =F att +F c

[0134] Finally, F r Through the full connection layer and residual connection, the feature output F is obtained output :

[0135] F output =F r +sigmoid(W4ReLU(W3F r ))

[0136] Among them, F output is the output of the first encoding module, W3 and W4 are the weight parameters in the fully connected layer.

[0137] As one of the optional embodiments, the first decoder is used to perform feature decoding prediction based on the power outage performance information, the basic attribute information, and the global fusion information to obtain power outage hidden danger features, including:

[0138] The power outage performance information and the basic attribute information are spliced ​​and then subjected to feature fusion decoding to obtain a first feature;

[0139] Performing feature dimension reduction and decoding on the global fusion information to obtain a second feature;

[0140] Inputting the first feature and the second feature into an attention network to obtain a power outage hidden danger attention feature;

[0141] The power outage hidden danger attention feature is input into a feedforward network to predict the credibility of hidden danger occurrence, the power outage hidden danger area, and the power outage hidden danger network to obtain the power outage hidden danger feature.

[0142] Specifically, the feature data (power outage features, basic attribute features and clustering scenario data) enters the feature decoding prediction module after passing through the feature coding fusion module. The feature coding fusion module includes a first decoder and a second decoder, where the first decoder is used to extract power outage hidden danger features, and the second decoder is used to extract power outage judgment strategies.

[0143] See also Figure 8 , Figure 8 This is a schematic diagram of the structure of the first decoder. After the first decoder concatenates the power outage performance information and basic attribute information, it is input into several stacked information fusion decoding modules for special fusion decoding and outputting the first feature, which serves as the query feature of the attention network. On the other hand, the global fusion information is input into several stacked dimensionality reduction decoding submodules for feature dimensionality reduction and decoding, and the dimension of the output feature is made the same as the dimension of the output feature of the information fusion decoding module to obtain the second feature. The features obtained after feature dimensionality reduction and decoding are respectively used as the key feature and value feature of the attention network. Furthermore, the attention network extracts the power outage hidden danger attention feature, which is then passed through a feedforward network (fully-connected layers) to predict the power outage hidden danger area, power outage hidden danger network (line), and hidden danger occurrence credibility to obtain the power outage hidden danger feature.

[0144] As one of the optional embodiments, the method of using a second decoder to perform feature decoding prediction based on the global fusion information, the basic attribute information, and the clustering scenario information to obtain a power outage determination strategy for each scenario includes:

[0145] The basic attribute information and the clustering scene information are spliced ​​together and then feature fusion decoding is performed to obtain a third feature;

[0146] Performing feature dimensionality reduction and decoding on the global fusion information to obtain a fourth feature;

[0147] Inputting the third feature and the fourth feature into the attention network to obtain a scene strategy attention feature;

[0148] The scene strategy attention features are input into the feedforward network to predict the power outage determination strategy, and the power outage determination strategy for each scene is obtained.

[0149] Specifically, see Figure 9 , Figure 9 This is a schematic diagram of the second decoder. The second decoder has the same basic structure as the first decoder. After concatenating the clustered scene information and basic attribute information, the second decoder inputs the information into a stack of information fusion decoding modules for feature fusion decoding, outputting a third feature that serves as the query feature for the attention network. The global fusion information is input into a stack of dimensionality reduction decoding submodules for feature dimensionality reduction and decoding. The output feature dimensions are aligned with those of the information fusion decoding module, yielding a fourth feature. The features obtained after dimensionality reduction and decoding serve as the key and value features of the attention network, respectively. Furthermore, the attention network extracts the scene strategy attention features, which are then predicted through a feedforward network (fully-connected layers) to determine the power outage determination strategy for each scenario.

[0150] Furthermore, the structure of the dimensionality reduction decoding submodule (subcode decoder) in the first decoder (first decoder) and the second decoder (second decoder) is as follows Figure 10 As shown in Figure 2, the dimensionality reduction decoding submodule (subcode decoder) consists of a self-attention network, a feedforward network (fully-connected layers) and a feedforward dimensionality reduction network (fully-connected layers), which can decode the features and reduce the global features to the same dimension as the output features of the information fusion decoding module. The structure of the information fusion decoding module (decoder) is shown in Figure 2. Figure 11 As shown in the figure, the information fusion decoding module (decoder) consists of a self-attention network and a feedforward network (fully-connected layers), which can fuse and decode the input features.

[0151] The embodiment of the present application utilizes a squeeze excitation network, an attention network, and a feedforward network to mine the associated information of data features, thereby achieving accurate prediction of power outage hazards and power outage determination rules for each scenario, which is conducive to achieving accurate real-time power outage determination.

[0152] As one of the optional embodiments, extracting the key features of the power outage hidden danger from the power outage hidden danger features, and extracting the real-time key features of the power outage from the real-time data of the power outage features, includes:

[0153] Inputting the power outage hidden danger features into a preset first neural network for feature extraction to obtain key features of the power outage hidden danger;

[0154] The real-time data of the power outage characteristics is input into a preset second neural network for feature extraction to obtain real-time power outage key features; wherein the first neural network and the second neural network are parallel network structures.

[0155] Specifically, the real-time data of power outage characteristics include: standard deviation of the time of occurrence of mobile optical modem power outage alarm, power equipment voltage, power equipment current and other real-time data in the mobile network and the power network.

[0156] Since the feature data types of power outage hidden danger features and real-time data of power outage features are different, power outage hidden danger features involve historical power outage business data, while real-time data of power outage features are real-time heterogeneous information (different data sources). Therefore, this application constructs a parallel network structure for power outage hidden danger features and real-time power outage features. The use of a parallel network structure can better model and fuse the two, make full use of the information of data from different periods, and improve the expression ability and prediction accuracy of the model. On the one hand, power outage hidden dangers are identified based on historical data to obtain key features of power outage hidden dangers (including extraction of power outage hidden danger features and extraction of key features from power outage hidden danger features). On the other hand, real-time data is screened and processed to obtain real-time power outage key features, and then the power outage hidden danger key features and real-time power outage key features are combined to perform accurate real-time power outage judgment. Among them, the first neural network and the second neural network are trained based on historical power outage hidden danger features and historical data of power outage features, respectively.

[0157] The embodiment of the present application constructs and trains a first neural network to extract key features of power outage hazards. The construction process of the first neural network includes: using MLP (multi-layer perceptron) as the main neural network architecture, defining the input layer, hidden layer and output layer, using ReLU function as the activation function, and performing nonlinear mapping on the linear combination of the hidden layer output to enhance the expression and generalization capabilities of the model; then introducing Dropout regularization technology to ensure that each neuron in the hidden layer and the output layer is connected to the neurons in other layers through weights, and the weights are continuously optimized through training to fit the relationship between input and output; finally, the power outage hazard features are input into the defined neural network, and the mean square error loss function is used to measure the difference between the predicted results and the true labels, and the Backpropagation optimization algorithm is used to minimize the loss function as the goal to update the neural network parameters, and finally obtain the trained first neural network. The first neural network can effectively learn and represent the hidden danger features of power outages, and provide accurate input for subsequent precise and real-time power outage judgments.

[0158] The embodiment of the present application constructs and trains a second neural network to extract key features of real-time power outages. The construction process of the second neural network includes: using LSTM (long short-term memory network) as the main neural network architecture, defining the input processing of LSTM, receiving input data (usually an element in the sequence) and the hidden state of the previous time step as input at each time step, and updating the memory; then inputting the clustering scene data obtained in step S2 into the defined LSTM network, using the mean square error loss function to measure the difference between the predicted result and the true label, and using the Backpropagation optimization algorithm to minimize the loss function as the goal, update the neural network parameters, and finally obtain a trained second neural network. The second neural network can effectively extract the key feature representation of real-time power outages from the heterogeneous real-time power outage data of mobile networks and power networks, and provide effective input for accurate real-time power outage judgment.

[0159] Furthermore, in step S5, the power outage determination strategy for each scenario obtained in step S3 is used to fuse the key features of power outage hazards and the real-time key features of power outages extracted in step S4, thereby obtaining an accurate comprehensive determination result of real-time power outages. In real-time power outage determination, the key features of power outage hazards represent the hidden dangers and credibility of the power network, the real-time key features of power outages represent real-time power outage events, and the power outage determination strategy specifies the determination rules. By integrating this data, this embodiment can provide a more comprehensive and accurate portrait of power outage events, thereby more promptly and accurately determining whether a power outage exists in a certain area or network (line), greatly reducing the misjudgment rate.

[0160] Specifically, the determination process in step S5 is calculated using the Bayesian formula. The Bayesian formula states that when the analysis sample is large enough to approach the population, the probability of an event occurring in the sample will be close to the probability of the event occurring in the population. For the case where there are more than two variables, the specific formula is as follows:

[0161]

[0162] Where P(A) and P(B) are prior probabilities, P(A|B,C) represents the conditional probability of A given the occurrence of B and C, P(B|A) represents the conditional probability of B given the occurrence of A, P(C|A,B) represents the conditional probability of C given the occurrence of A and B, and P(C|B) represents the conditional probability of C given the occurrence of B. In this embodiment, A represents a real power outage event, and B, C, etc. represent factors that cause power outages when a mobile optical modem heartbeat or power outage alarm occurs, or when the total power consumption, total current, or total power consumption deteriorates.

[0163] Furthermore, step S5 specifically constructs a power outage determination model. This model is used to fuse key features of power outage hazards with real-time key features based on the power outage determination strategy to obtain a power outage determination result. Through iterative training, the power outage determination model continuously learns how to rationally fuse key features of power outage hazards with real-time key features of power outages, and performs calculations based on the power outage determination strategy for the corresponding scenario, thereby achieving more accurate and timely power outage determination and root cause analysis.

[0164] It should be noted that in the present application, the spatial area (such as the power network / line) is divided into several scenes, and a corresponding power outage determination strategy is assigned to each scene. The power outage determination rules include but are not limited to feature selection rules, fusion weights, and judgment thresholds, etc., and then, when making a power outage determination in a certain area, the key features of the power outage hidden dangers in the area and the real-time power outage key features can be fused and calculated and the power outage determination can be made according to the power outage determination strategy of the corresponding scene in the area, so as to obtain an accurate power outage determination result for the area, avoid a one-size-fits-all power outage determination, and select a more matching power outage determination strategy for the different data features of each area, thereby improving the accuracy and flexibility of the power outage determination.

[0165] Furthermore, in practical applications, after obtaining the outage determination results, emergency repair tasks are generated and assigned based on the results. This provides highly accurate outage events, guides emergency repairs, promotes socioeconomic development, and improves the motivation of repair personnel and customer experience. Operators can also more precisely understand which areas or time periods are at risk of power outages, thereby optimizing resource allocation, such as pre-deploying emergency generators or increasing maintenance personnel to mitigate the potential impact of outages.

[0166] The embodiment of the present application obtains power outage characteristics and basic attribute characteristics in the mobile network and the power network, integrates the mobile network data and the power network data, and can make a more comprehensive power outage judgment; a clustering algorithm is used to divide the spatial area into several scenes to obtain clustered scene data, which is conducive to making power outage judgments for each type of scene, thereby improving the accuracy and flexibility of real-time power outage judgments; power outage hidden dangers are predicted based on the power outage characteristics, the basic attribute characteristics and the clustered scene data to obtain power outage hidden danger characteristics and power outage judgment strategies for each scene, which can integrate multiple data and identify power network line hidden dangers, thereby improving the accuracy of real-time power outage judgments; furthermore, power outage hidden danger key features are extracted from the power outage hidden danger characteristics, and real-time power outage key features are extracted from the real-time data of the power outage characteristics, and the power outage hidden danger key features are fused and calculated with the real-time power outage key features according to the power outage judgment strategy to obtain a power outage judgment result, which can efficiently integrate real-time power outage data and predicted power outage hidden danger data to achieve accurate real-time power outage judgment, improve the accuracy of real-time power outage judgment and reduce the error rate.

[0167] Accordingly, the present application also provides a real-time power outage determination device that can implement all processes of the real-time power outage determination method in the above embodiment.

[0168] See also Figure 12 , Figure 12 : This is a schematic diagram of the structure of a real-time power outage determination device provided in an embodiment of the present application. The real-time power outage determination device includes:

[0169] Data acquisition module 201, used to acquire power outage characteristics and basic attribute characteristics in mobile networks and power networks;

[0170] A scene clustering module 202 is configured to divide a spatial region into a number of scenes using a clustering algorithm based on the power outage characteristics and the basic attribute characteristics to obtain clustered scene data;

[0171] The power outage prediction module 203 is configured to predict power outage hazards based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data, and obtain power outage hazard characteristics and power outage determination strategies for each scenario;

[0172] A feature extraction module 204 is configured to extract key features of power outage hazards from the power outage hazard features, and extract real-time key features of power outages from the real-time data of the power outage features;

[0173] The power outage determination module 205 is configured to perform a fusion calculation on the power outage potential key features and the real-time power outage key features according to the power outage determination strategy to obtain a power outage determination result.

[0174] Preferably, the power outage hidden danger prediction is performed based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data to obtain the power outage hidden danger characteristics and the power outage determination strategy for each scenario, including:

[0175] Extracting power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data;

[0176] Inputting the power outage performance information, the basic attribute information, and the clustering scenario information into a preset multi-class feature encoder to obtain global fusion information; the multi-class feature encoder is composed of a plurality of first encoding modules, each of which is composed of a global pooling layer, a self-attention network, and a fully connected network;

[0177] According to the power outage performance information, the basic attribute information and the global fusion information, a first decoder is used to perform feature decoding and prediction to obtain power outage hidden danger features;

[0178] According to the global fusion information, the basic attribute information and the clustering scenario information, a second decoder is used to perform feature decoding prediction to obtain a power outage determination strategy for each scenario.

[0179] Preferably, the extracting of power outage performance information, basic attribute information and clustering scenario information based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data includes:

[0180] Inputting the power outage hidden danger data and real-time data in the power outage feature into corresponding sub-encoders respectively to obtain a first code corresponding to the power outage hidden danger data and a second code corresponding to the real-time data; splicing the first code and the second code and inputting them into a preset multi-category feature encoder to obtain power outage performance information;

[0181] Inputting the spatial feature variable, device feature variable, and user feature variable in the basic attribute features into corresponding sub-encoders respectively to obtain feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable respectively; splicing the feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable into a preset multi-category feature encoder to obtain a third code; inputting the climate feature variable in the basic attribute features into a corresponding sub-encoder to obtain a fourth code; splicing the third code and the fourth code into a preset multi-category feature encoder to obtain basic attribute information;

[0182] Inputting the data of each scene in the clustered scene data into the corresponding sub-encoder respectively to obtain the feature code corresponding to each scene; splicing the feature codes corresponding to each scene and inputting them into the preset multi-category feature encoder to obtain clustered scene information;

[0183] Among them, each sub-encoder is composed of several superimposed second encoding modules, and the second encoding module is composed of a self-attention network, a feedforward network and a feedforward dimensionality reduction network.

[0184] Preferably, the step of performing feature decoding and prediction using a first decoder based on the power outage performance information, the basic attribute information, and the global fusion information to obtain power outage hidden danger features includes:

[0185] The power outage performance information and the basic attribute information are spliced ​​and then subjected to feature fusion decoding to obtain a first feature;

[0186] Performing feature dimension reduction and decoding on the global fusion information to obtain a second feature;

[0187] Inputting the first feature and the second feature into an attention network to obtain a power outage hidden danger attention feature;

[0188] The power outage hidden danger attention feature is input into a feedforward network to predict the credibility of hidden danger occurrence, the power outage hidden danger area, and the power outage hidden danger network to obtain the power outage hidden danger feature.

[0189] Preferably, the step of performing feature decoding prediction using a second decoder based on the global fusion information, the basic attribute information, and the clustering scenario information to obtain a power outage determination strategy for each scenario includes:

[0190] The basic attribute information and the clustering scene information are spliced ​​together and then feature fusion decoding is performed to obtain a third feature;

[0191] Performing feature dimensionality reduction and decoding on the global fusion information to obtain a fourth feature;

[0192] Inputting the third feature and the fourth feature into the attention network to obtain a scene strategy attention feature;

[0193] The scene strategy attention features are input into the feedforward network to predict the power outage determination strategy, and the power outage determination strategy for each scene is obtained.

[0194] Preferably, the extracting the key features of the power outage hidden danger from the power outage hidden danger features and extracting the real-time key features of the power outage from the real-time data of the power outage features includes:

[0195] Inputting the power outage hidden danger features into a preset first neural network for feature extraction to obtain key features of the power outage hidden danger;

[0196] The real-time data of the power outage characteristics is input into a preset second neural network for feature extraction to obtain real-time power outage key features; wherein the first neural network and the second neural network are parallel network structures.

[0197] In specific implementation, the working principle, control process and technical effects achieved by the real-time power outage determination device provided in the embodiment of the present application are the same as those of the real-time power outage determination method in the above embodiment, and will not be repeated here.

[0198] See also Figure 13 , Figure 13 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device includes: a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program, the steps of the above-described real-time power outage determination method embodiment are implemented. Alternatively, when the processor 301 executes the computer program, the functions of the modules / units in the above-described device embodiments are implemented.

[0199] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 302 and executed by the processor 301 to implement the present application. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0200] The computer device may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that the schematic diagram is merely an example of a computer device and does not limit the computer device. The computer device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, and the like.

[0201] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor 301 is the control center of the computer device, connecting various parts of the entire computer device using various interfaces and lines.

[0202] The memory 302 can be used to store the computer programs and / or modules. The processor 301 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 302 and calling the data stored in the memory 302. The memory 302 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory 302 can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0203] Wherein, if the module / unit integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor 301, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form.

[0204] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the real-time power outage determination method described in any of the above embodiments.

[0205] An embodiment of the present application further provides a computer program product, including a computer program / instruction, which, when executed by a processor of the computer program / instruction, implements the real-time power outage determination method described in any of the above embodiments.

[0206] The embodiments of the present application provide a real-time power outage determination method, apparatus, device, storage medium and product, the beneficial effects of which are as follows: by acquiring the power outage characteristics and basic attribute characteristics in the mobile network and the power network, the mobile network data and the power network data are integrated to make a more comprehensive power outage determination; a clustering algorithm is used to divide the spatial area into several scenes to obtain clustered scene data, which is conducive to making power outage determinations for each type of scene and improves the accuracy and flexibility of real-time power outage determinations; based on the power outage characteristics, the basic attribute characteristics and the clustered scene data, a power outage hidden danger is predicted to obtain a power outage. The hidden danger characteristics and the power outage judgment strategy for each scenario can integrate multiple data and identify hidden dangers of power network lines, thereby improving the accuracy of real-time power outage judgment; furthermore, the key features of power outage hidden dangers are extracted from the power outage hidden danger characteristics, and the real-time power outage key features are extracted from the real-time data of the power outage characteristics, and the power outage hidden danger key features and the real-time power outage key features are fused and calculated according to the power outage judgment strategy to obtain the power outage judgment result, which can efficiently integrate the real-time power outage data and the predicted power outage hidden danger data to achieve accurate real-time power outage judgment, improve the accuracy of real-time power outage judgment and reduce the misjudgment rate.

[0207] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications are also considered to be within the scope of protection of the present application.

Claims

1. A real-time power outage determination method, characterized in that: include: Obtain power outage characteristics and basic attribute features in mobile networks and power networks; Based on the power outage characteristics and the basic attribute characteristics, a clustering algorithm is used to divide the spatial area into several scenes to obtain clustered scene data; Based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data, power outage hidden danger prediction is performed to obtain power outage hidden danger characteristics and power outage determination strategies for each scenario; Extracting key features of power outage hidden dangers from the power outage hidden danger features, and extracting real-time key features of power outages from real-time data of the power outage features; According to the power outage determination strategy, the power outage hidden danger key features and the real-time power outage key features are fused and calculated to obtain a power outage determination result.

2. The real-time power outage determination method according to claim 1, wherein: The power outage hidden danger prediction is performed based on the power outage characteristics, the basic attribute characteristics and the clustering scenario data to obtain the power outage hidden danger characteristics and the power outage determination strategy for each scenario, including: Extracting power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data; Inputting the power outage performance information, the basic attribute information, and the clustering scenario information into a preset multi-class feature encoder to obtain global fusion information; the multi-class feature encoder is composed of a plurality of first encoding modules, each of which is composed of a global pooling layer, a self-attention network, and a fully connected network; According to the power outage performance information, the basic attribute information and the global fusion information, a first decoder is used to perform feature decoding and prediction to obtain power outage hidden danger features; According to the global fusion information, the basic attribute information and the clustering scenario information, a second decoder is used to perform feature decoding prediction to obtain a power outage determination strategy for each scenario.

3. The real-time power outage determination method according to claim 2, wherein: The extracting of power outage performance information, basic attribute information, and clustering scenario information based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data includes: Inputting the power outage hidden danger data and real-time data in the power outage feature into corresponding sub-encoders respectively to obtain a first code corresponding to the power outage hidden danger data and a second code corresponding to the real-time data; splicing the first code and the second code and inputting them into a preset multi-category feature encoder to obtain power outage performance information; Inputting the spatial feature variable, device feature variable, and user feature variable in the basic attribute features into corresponding sub-encoders respectively to obtain feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable respectively; splicing the feature codes corresponding to the spatial feature variable, the device feature variable, and the user feature variable into a preset multi-category feature encoder to obtain a third code; inputting the climate feature variable in the basic attribute features into a corresponding sub-encoder to obtain a fourth code; splicing the third code and the fourth code into a preset multi-category feature encoder to obtain basic attribute information; Inputting the data of each scene in the clustered scene data into the corresponding sub-encoder respectively to obtain the feature code corresponding to each scene; splicing the feature codes corresponding to each scene and inputting them into the preset multi-category feature encoder to obtain clustered scene information; Among them, each sub-encoder is composed of several superimposed second encoding modules, and the second encoding module is composed of a self-attention network, a feedforward network and a feedforward dimensionality reduction network.

4. The real-time power outage determination method according to claim 2, wherein: The method of performing feature decoding and prediction using a first decoder based on the power outage performance information, the basic attribute information, and the global fusion information to obtain power outage hidden danger features includes: The power outage performance information and the basic attribute information are spliced ​​and then subjected to feature fusion decoding to obtain a first feature; Performing feature dimension reduction and decoding on the global fusion information to obtain a second feature; Inputting the first feature and the second feature into an attention network to obtain a power outage hidden danger attention feature; The power outage hidden danger attention feature is input into a feedforward network to predict the credibility of hidden danger occurrence, the power outage hidden danger area, and the power outage hidden danger network to obtain the power outage hidden danger feature.

5. The real-time power outage determination method according to claim 2, wherein: The method of using a second decoder to perform feature decoding prediction based on the global fusion information, the basic attribute information, and the clustering scenario information to obtain a power outage determination strategy for each scenario includes: The basic attribute information and the clustering scene information are spliced ​​together and then feature fusion decoding is performed to obtain a third feature; Performing feature dimensionality reduction and decoding on the global fusion information to obtain a fourth feature; Inputting the third feature and the fourth feature into the attention network to obtain a scene strategy attention feature; The scene strategy attention features are input into the feedforward network to predict the power outage determination strategy, and the power outage determination strategy for each scene is obtained.

6. The real-time power outage determination method according to claim 1, wherein: The step of extracting key features of power outage hidden dangers from the power outage hidden danger features and extracting real-time key features of power outages from real-time data of the power outage features includes: Inputting the power outage hidden danger features into a preset first neural network for feature extraction to obtain key features of the power outage hidden danger; The real-time data of the power outage characteristics is input into a preset second neural network for feature extraction to obtain real-time power outage key features; wherein the first neural network and the second neural network are parallel network structures.

7. A real-time power outage determination device, characterized in that: include: A data acquisition module is used to obtain power outage characteristics and basic attribute characteristics in mobile networks and power networks; A scene clustering module, configured to divide a spatial area into a number of scenes using a clustering algorithm based on the power outage characteristics and the basic attribute characteristics, thereby obtaining clustered scene data; A power outage prediction module, configured to predict power outage hazards based on the power outage characteristics, the basic attribute characteristics, and the clustering scenario data, and obtain power outage hazard characteristics and power outage determination strategies for each scenario; A feature extraction module is used to extract key features of power outage hidden dangers from the power outage hidden danger features, and extract real-time key features of power outages from the real-time data of the power outage features; The power outage determination module is used to perform a fusion calculation on the power outage hidden danger key features and the real-time power outage key features according to the power outage determination strategy to obtain a power outage determination result.

8. A computer device, characterized in that: The system comprises a processor and a memory, wherein a computer program is stored in the memory and the computer program is configured to be executed by the processor, and when the processor executes the computer program, the real-time power outage determination method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the real-time power outage determination method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the real-time power outage determination method according to any one of claims 1 to 6 is implemented.

Images