Power grid monitoring system front channel adjusting method and system based on risk prediction

Through real-time abnormality identification and evaluation of the grid pre-channel abnormality monitoring model, the channel and server abnormalities in the grid monitoring system are automatically handled, which solves the problem of low robustness and intelligence of the scheduling method in the existing technology, and realizes intelligence dynamic adjustment and data transmission reliability improvement.

CN118972423APending Publication Date: 2024-11-15杨磊杰
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410964073.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing power grid monitoring system has poor robustness and low intelligence, and cannot solve various problems automatically, resulting in data loss, data errors and excessive server load during transmission.

Method used

By obtaining channel monitoring data and server status record data, sorting and subsetting, using the power grid pre-channel abnormality monitoring model for real-time abnormality identification and evaluation, and automatically switching devices to achieve intelligent dynamic adjustments. The model includes monitoring data input, feature extraction, exception evaluation and output units, using GRU and Attention mechanisms to extract timing and morphological features, and combining cluster analysis to determine exception types and scheduling strategies.

Benefits of technology

Real-time monitoring and automated exception handling are realized, which improves the intelligence and robustness of the pre-service, can accurately identify abnormalities and take targeted regulatory measures, and improves the reliability and efficiency of data transmission.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118972423A_ABST
    Figure CN118972423A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of digital power grids, in particular to a power grid monitoring system front channel adjusting method and system based on risk prediction. According to the invention, the monitoring data of each channel and the state record data of each server are obtained, sorting and subset division are carried out, and the corresponding numbers are obtained according to the connection relationship and real-time monitoring is carried out; abnormal information of the channel monitoring data subset and the state recording data constant subset is obtained in the monitoring process, and if the channel monitoring data subset or the state recording data constant subset exceeds a corresponding threshold value, corresponding equipment without abnormal information is scheduled according to the number to be switched; if the channel monitoring data subset and the state recording data subset both exceed the corresponding threshold values, attribution analysis is carried out to obtain front channel abnormal information, so that a corresponding scheduling strategy is obtained to carry out equipment switching; by means of the method, the system can achieve intelligent and automatic dynamic adjustment, so that various problems occurring in the transmission process are avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of digital power grids, and in particular to a front channel adjustment method and system for a power grid monitoring system based on risk prediction. Background Art

[0002] SCADA (Supervisory Control and Data Acquisition) services and front-end channel services are important components of the current power grid monitoring system. SCADA services mainly use the large amount of actual data obtained by the front-end services for subsequent conversion, calculation, storage and alarm to achieve real-time monitoring of the actual data. In order to achieve efficient and timely power grid data monitoring services, the data transmission quality of the front-end channel plays a decisive role.

[0003] The front channel is mainly responsible for real-time data collection through the RTU (Remote Terminal Unit) unit. After collecting data, the actual data is transmitted to the server for subsequent processing. As the amount of actual collected data gradually increases, in order to ensure the quality of the collected data, a collaborative mode of multiple servers, main channels and backup channels is adopted to collect and transmit data.

[0004] However, in actual work, due to the uncertainty of data transmission and scheduling strategy issues, the following problems are likely to occur: first, the transmission volume of the main channel or backup channel is too large due to the scheduling strategy; second, data errors and data loss occur during the scheduling process; third, a single server causes anomalies due to the large amount of collected data due to the scheduling strategy, making it impossible to collect data normally afterwards.

[0005] Based on the above problems, CN202110644541.8 proposed a front-end service scheduling method for a power grid monitoring system based on load balancing distribution. By sorting the servers in sequence, setting the first target server to establish a main channel, setting the second target server to establish a backup channel, and adjusting the first target server and the second target server through the sequence relationship to solve the server scheduling problem; CN202110946680.6 proposed a front-end service fault handling method for a power grid monitoring system based on clustering algorithm technology. Based on the clustering algorithm, the service anomaly is analyzed to obtain abnormal hang-up messages on a daily basis. The abnormal hang-up messages characterize the abnormal servers. During the transmission process, the main channel and the backup channel are switched through the abnormal hang-up messages to improve the efficiency of the traditional scheduling method; CN202010989326.7 proposed a power grid monitoring system that supports dynamic expansion. By establishing a bus connection and decoupling the front-end service and the SCADA service, multiple front-end services are combined to avoid that when a single front-end service is restarted, it will not affect the overall work, and it can be switched to other front-end services for processing, thereby improving work efficiency.

[0006] The existing methods for solving pre-service problems are mostly based on solving the current problem with a single technology, which has poor robustness; and the current methods have a low level of intelligence and cannot automatically solve the various problems that currently arise in pre-services.

[0007] In order to improve the intelligence level of the solution method for the front-end service problem and realize automatic dynamic adjustment, the present invention proposes a front-end channel adjustment method and system for a power grid monitoring system based on risk prediction. Summary of the invention

[0008] The purpose of the present invention is to provide a front-end channel adjustment method and system for a power grid monitoring system based on risk prediction, by acquiring each channel monitoring data and each server's status record data, sorting and subset division, acquiring corresponding numbers according to connection relationships and performing real-time monitoring; obtaining abnormal information of the channel monitoring data subset and the status record data subset during the monitoring process, if the channel monitoring data subset or the status record data constant subset exceeds the corresponding threshold value, the corresponding equipment without abnormal information is switched according to the number scheduling; if both the channel monitoring data subset and the status record data constant subset exceed the corresponding threshold value, attribution analysis is performed to obtain the front-end channel abnormal information, so as to obtain the corresponding scheduling strategy for equipment switching; the above method allows the system to realize intelligent and automated dynamic adjustment to avoid various problems that occur during the transmission process.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] The front channel adjustment method of the power grid monitoring system based on risk prediction includes:

[0011] Channel monitoring data are numbered according to each channel type, and each channel monitoring data is obtained; the channel monitoring data are sorted and divided into multiple subsets to generate channel monitoring data subsets; the channel monitoring data are sorted according to usage status, usage duration, start time of use and channel attributes; multiple channel monitoring data subsets divide the channel monitoring data according to the division time T;

[0012] According to the connection relationship between each channel and the server, the state record data is associated with the channel number and a server number is generated, the state record data of each server is obtained, and the state record data is divided according to the division condition of the channel monitoring data subset to generate a state record data subset;

[0013] Identify abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time;

[0014] Further, the identification of the abnormal information is obtained based on the power grid front channel abnormality monitoring model;

[0015] The power grid front channel abnormality monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit;

[0016] The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data; wherein the preprocessing includes denoising and data standardization;

[0017] The monitoring data feature extraction unit includes a monitoring data time series feature extraction unit and a monitoring data morphology feature extraction unit; the monitoring data time series feature extraction unit includes 3 layers of GRU units and 1 layer of Attention mechanism; each layer of GRU unit includes a reset gate calculation module and an update gate calculation module, the reset gate calculation module is used to capture long-term time series features, and the update gate calculation module is used to capture short-term time series features. The specific calculation is as follows:

[0018] re=σ[W re ·(input t-1 ,feature t )+b re ];

[0019] update = σ[W up ·(input t-1 ,feature t )+bup ];

[0020]

[0021] Among them, re is the result of the reset gate calculation module, σ and tanh are activation functions, W re To reset the gate calculation module weight, b re To reset the gate calculation module bias function, input t-1 is the input data at time t-1, feature t is the feature at time t, update is the result of updating the gate calculation module, W up To reset the gate calculation module weights, b up is the reset gate calculation module bias function, output is the output result, W out is the output weight, b out is the output bias function;

[0022] The expression of the monitoring data time series extraction feature unit is as follows:

[0023]

[0024] Among them, Output is the output data, input is the input data, Attention is the Attention mechanism, and GRU is the GRU unit;

[0025] The monitoring data morphology feature extraction unit includes 1 1*1 convolution layer, 2 1*3 convolution layers, 1 1*2 average pooling layer, 2 relu activation layers and 1 fully connected layer. The expression of the monitoring data morphology feature extraction unit is as follows:

[0026] IN = relu(input);

[0027] ConvA = Conv1 × 1 (IN);

[0028] ConvB=Conv1*3{pool[Conv1*3(IN)]};

[0029]

[0030] Where, IN is the feature input, relu is the relu activation layer, input is the input data, ConvA and ConvB are two convolution calculation methods, Conv1×1 is the 1*1 convolution layer, Conv1*3 is the 1*3 convolution layer, pool is the 1*2 average pooling layer, fc is the fully connected layer, and Output is the output data;

[0031] The monitoring data anomaly assessment unit performs an anomaly assessment on the monitoring model feature data to obtain monitoring model assessment data; the monitoring data anomaly output unit outputs an anomaly type according to the score of the monitoring model assessment data;

[0032] The abnormality evaluation calculation of the monitoring data abnormality evaluation unit is:

[0033] AE i =α1·S 1i +α2·S 2i ;

[0034] α1+α2=1;

[0035]

[0036] Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, β li is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of

[0037] If the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormality threshold, the abnormal channel number is recorded, the channel number without the abnormal information is obtained, and the abnormal channel number is switched;

[0038] If the abnormal information of the status record data subset exceeds the status record abnormal threshold, the abnormal server number is recorded, the server number without the abnormal information is obtained, and the server number is switched to the abnormal server number;

[0039] Further, the channel monitoring abnormality threshold and the status record abnormality threshold are obtained by obtaining historical abnormality data corresponding to each type of abnormal information based on historical channel monitoring data and historical status record data; dividing the historical abnormality data according to a division time T, and obtaining a duration S according to the division time T;

[0040] Count the duration S of various historical abnormal data and obtain the maximum value S of the abnormal duration S max and abnormal persistent minimum S min , with S max and S min Establish an abnormal interval and obtain its corresponding threshold;

[0041] If both the channel monitoring data subset and the status record data subset are abnormal, attribution analysis is performed based on the channel abnormal information and the server abnormal information, the front channel abnormal information is obtained, the corresponding solution strategy is obtained through the front channel abnormal information, and switching is performed according to the channel number, the server number and the scheduling strategy;

[0042] Furthermore, the front channel abnormal information acquisition is to obtain the monitoring abnormal information y of the channel monitoring data subset c and the state abnormality information y of the state record data subset s ; Based on the monitoring abnormal information and the state abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s );

[0043] Performing front channel anomaly clustering on the abnormal information coordinates to obtain clustering similarity, wherein the cluster with the largest similarity is the abnormal type of the abnormal information coordinates, and the abnormal type is the front channel abnormal information;

[0044] The cluster similarity is calculated as follows:

[0045]

[0046] Among them, P y is the front channel abnormal information, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula;

[0047] The present invention also provides a front channel adjustment system of a power grid monitoring system based on risk prediction, including a channel monitoring data division module, a state recording data division module, an abnormal information acquisition module, a channel monitoring data abnormality processing module, a state recording data abnormality processing module and a multi-abnormality processing module, specifically:

[0048] A channel monitoring data division module is used to number the channel monitoring data according to each channel type and obtain each channel monitoring data; sort and divide the channel monitoring data to generate a channel monitoring data subset;

[0049] A status record data division module, according to the connection relationship between each channel and the server, associates the status record data with the channel number and generates a server number, obtains the status record data of each server, divides the status record data according to the division condition of the channel monitoring data subset, and generates a status record data subset;

[0050] An abnormal information acquisition module, which identifies abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time;

[0051] Further, the identification of the abnormal information is obtained based on the power grid front channel abnormality monitoring model;

[0052] The power grid front channel abnormality monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit;

[0053] The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data;

[0054] The monitoring data anomaly assessment unit performs an anomaly assessment on the monitoring model feature data to obtain monitoring model assessment data; the monitoring data anomaly output unit outputs an anomaly type according to the score of the monitoring model assessment data;

[0055] Furthermore, the abnormality evaluation calculation of the monitoring data abnormality evaluation unit is:

[0056] AE i =α1·S 1i +α2·S 2i ;

[0057] α1+α2=1;

[0058]

[0059] Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, β li is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of

[0060] A channel monitoring data exception processing module records the abnormal channel number, obtains the channel number without the abnormal information and switches with the abnormal channel number if the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormal threshold;

[0061] A status record data exception processing module, if the exception information of the status record data subset exceeds the status record exception threshold, records the abnormal server number, obtains and switches to the server number without the abnormal information;

[0062] A multi-exception processing module, if both the channel monitoring data subset and the status record data subset are abnormal, performs attribution analysis based on the channel abnormal information and the server abnormal information, obtains the front channel abnormal information, obtains the corresponding scheduling strategy through the front channel abnormal information, and switches according to the channel number, the server number and the scheduling strategy;

[0063] Further, the front channel abnormal information acquisition of the multi-abnormal processing module is to obtain the monitoring abnormal information y of the channel monitoring data subset c and the state abnormality information y of the state record data subsets ; Based on the monitoring abnormal information and the state abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s );

[0064] Performing front channel anomaly clustering on the abnormal information coordinates to obtain clustering similarity, wherein the cluster with the largest similarity is the abnormal type of the abnormal information coordinates, and the abnormal type is the front channel abnormal information;

[0065] The cluster similarity is calculated as follows:

[0066]

[0067] Among them, P y is the front channel abnormal information, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. In order to be able to analyze abnormal data in real time for regulation, the collected channel monitoring data and status record data are divided and input into the power grid front-end channel abnormal monitoring model. After the data is preprocessed, the time series characteristics and morphological characteristics of the real-time data are obtained, and the characteristics are evaluated and classified according to the evaluation indicators to determine whether the data is abnormal and output the corresponding abnormal type. Through this model, the abnormal indicators of the server and channel can be automatically obtained during the real-time monitoring process, and intelligent processing is performed through the abnormal indicators, which helps to improve the intelligence level of the front-end service regulation.

[0070] 2. In order to accurately obtain the anomaly type to facilitate the judgment, classification and dynamic adjustment of the subsequent data, based on the temporal characteristics and morphological characteristics recognized by the model, the similarity is calculated with the various anomaly data stored in the model, and then a comprehensive evaluation is performed based on the weighted results of each feature. This helps to improve the accuracy of the model in automatic anomaly classification. Through higher recognition accuracy, the intelligence level of the front-end service can be further improved, laying a good data foundation for subsequent automatic dynamic adjustment.

[0071] 3. Based on the abnormal classification results of the power grid front-end channel abnormal monitoring model, the corresponding abnormal information coordinates are generated by the data formed by the abnormal information of the classification results, and the abnormal clustering calculation of the front-end channel is performed according to the abnormal information coordinates. Combined with the clustering similarity calculation results of each clustering result, when abnormalities occur in both the channel and the server, the main cause of the abnormality can be obtained and targeted control methods can be taken; clustering calculation is used to facilitate the system to perform automatic analysis and adjustment after identifying the abnormality, thereby improving the intelligence of scheduling and realizing dynamic automatic scheduling strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0073] Figure 2 It is a schematic diagram of the flow chart of the power grid front channel abnormality monitoring model of the present invention;

[0074] Figure 3 It is a schematic diagram of clustering similarity calculation of the present invention;

[0075] Figure 4 It is a schematic diagram of the system flow of the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] As a data acquisition component of the entire power grid data monitoring service, the quality of the collected data of the front-end service channel plays a key role in the subsequent process; however, as the amount of data gradually increases, traditional methods can no longer solve the problems of excessive channel transmission volume, data errors or data loss in the scheduling process, and excessive server data collection during the transmission process; in order to solve the related problems of the front-end service channel, the current method mainly adopts a single technology to solve the problems that arise. Although it can solve the current problems, the robustness is poor, and the intelligence level of the related methods is low, which cannot solve the various problems that have arisen.

[0078] In order to improve the intelligence level of the solution to the front-end service problem and realize automatic dynamic adjustment, the present invention provides a front-end channel adjustment method of a power grid monitoring system based on risk prediction. The technical solution is as follows:

[0079] Reference Figure 1As shown, the channel monitoring data is numbered according to each channel type to obtain each channel monitoring data; the channel monitoring data is sorted and divided into multiple subsets to generate channel monitoring data subsets;

[0080] According to the connection relationship between each channel and the server, the state record data is associated with the channel number and a server number is generated, the state record data of each server is obtained, and the state record data is divided according to the division condition of the channel monitoring data subset to generate a state record data subset;

[0081] Identify abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time;

[0082] If the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormality threshold, the abnormal channel number is recorded, the channel number without the abnormal information is obtained, and the abnormal channel number is switched;

[0083] If the abnormal information of the status record data subset exceeds the status record abnormal threshold, the abnormal server number is recorded, the server number without the abnormal information is obtained, and the server number is switched to the abnormal server number;

[0084] If abnormalities occur in both the channel monitoring data subset and the status record data subset, attribution analysis is performed based on the channel abnormality information and the server abnormality information, the front channel abnormality information is obtained, the corresponding solution strategy is obtained through the front channel abnormality information, and switching is performed according to the channel number, the server number and the scheduling strategy.

[0085] In order to realize automatic intelligent dynamic adjustment, the monitoring data in the monitoring channel is implemented, and the corresponding channel number is set for each channel to facilitate subsequent scheduling; the server connected to the channel is associated with the channel number to generate the corresponding server number, so as to facilitate the subsequent adjustment of the server, and at the same time, the server status record data and channel monitoring data are divided to generate multiple subsets for the subsequent analysis process;

[0086] During the process, the abnormal information is analyzed and identified in real time based on the segmented subsets. If abnormal information appears in the channel, it is switched to a channel without abnormal information; if abnormal information appears in the server, it is switched to a server without abnormal information.

[0087] If both the server and the channel have exceptions, in order to determine the specific cause of the exception so as to take corresponding processing methods, attribution analysis is performed on the channel exception information and the server exception information. Through comprehensive feature analysis of the two parts, specific problem features are obtained to determine the specific cause of the exception. Corresponding processing solutions are taken according to the cause, which can improve the accuracy of problem solving during automated adjustments, and can also improve the intelligence and robustness of the entire processing solution.

[0088] Embodiment 1

[0089] For specific explanation, the present invention is described in conjunction with the following embodiments:

[0090] The front channel adjustment method of the power grid monitoring system based on risk prediction includes:

[0091] Channel monitoring data are numbered according to each channel type, and each channel monitoring data is obtained; the channel monitoring data are sorted and divided into multiple subsets to generate channel monitoring data subsets; the channel monitoring data are sorted according to usage status, usage duration, start time and channel attributes; multiple channel monitoring data subsets divide the channel monitoring data according to the division time T; channel attributes include main channels and backup channels; the channel monitoring data is mainly based on the data volume of power data collected in each time period, and a time series is formed in the order of time; the default duration of the division time T is 15 minutes, which can be adjusted according to actual conditions in the specific implementation plan;

[0092] The channels are numbered according to the main channel and the backup channel, for example, the main channel is A, the backup channel is B, and the total number is 4. The numbering sequence is A1, A2, B1, B2. In the specific implementation, other rules such as direction sequence, time sequence, etc. can be added for numbering;

[0093] In actual use, data transmission channels are generally divided into main channels and backup channels, and the priority of the main channel is higher than that of the backup channel. Therefore, marking the main channel and the backup channel separately when numbering will help improve the accuracy of subsequent scheduling work. At the same time, in order to facilitate subsequent analysis work, the accuracy of data anomaly identification can be improved while reducing the amount of calculation. Therefore, the data usage of the channel and the attributes of the main channel and the backup channel are sorted accordingly, which can significantly reduce the calculation pressure of the model analysis process and improve the accuracy of analysis.

[0094] According to the connection relationship between each channel and the server, the status record data is associated with the channel number and the server number is generated, the status record data of each server is obtained, and the status record data is divided according to the division conditions of the channel monitoring data subset to generate the status record data subset; for example, two servers are connected to the main channel A1, and the server numbers are A11 and A12. If the backup channel B2 is connected, the server numbers are B21 and B22. In the specific implementation plan, it can be adjusted according to the specific situation; Among them, the status record data is mainly the data volume of the power data stored by the server based on each time period. In the specific implementation plan, other data can be added according to the situation;

[0095] Real-time identification of abnormal information of channel monitoring data subsets and corresponding status record data subsets;

[0096] Furthermore, the identification of abnormal information is obtained based on the abnormal monitoring model of the power grid front channel;

[0097] The power grid front channel abnormal monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit. Figure 2 As shown;

[0098] The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data;

[0099] The monitoring data feature extraction unit includes a monitoring data time series extraction feature unit and a monitoring data morphology extraction feature unit; the monitoring data time series extraction feature unit includes 3 layers of GRU units and 1 layer of Attention mechanism; each layer of GRU unit includes a reset gate calculation module and an update gate calculation module. The reset gate calculation module is used to capture long-term time series features, and the update gate calculation module is used to capture short-term time series features. The specific calculation is as follows:

[0100] re=σ[W re ·(input t-1 ,feature t )+b re ];

[0101] update = σ[W up ·(input t-1 ,feature t )+b up ];

[0102]

[0103] Among them, re is the result of the reset gate calculation module, σ and tanh are activation functions, W re To reset the gate calculation module weights, b re To reset the gate calculation module bias function, input t-1 is the input data at time t-1, feature t is the feature at time t, update is the result of updating the gate calculation module, W up To reset the gate calculation module weights, b up is the reset gate calculation module bias function, output is the output result, W out is the output weight, b out is the output bias function;

[0104] The expression of the feature extraction unit of monitoring data time series is as follows:

[0105]

[0106] Among them, Output is the output data, input is the input data, Attention is the Attention mechanism, and GRU is the GRU unit;

[0107] The monitoring data morphology feature extraction unit includes 1 1*1 convolution layer, 2 1*3 convolution layers, 1 1*2 average pooling layer, 2 relu activation layers and 1 fully connected layer. The expression of the monitoring data morphology feature extraction unit is as follows:

[0108] IN = relu(input);

[0109] ConvA = Conv1 × 1 (IN);

[0110] ConvB=Conv1*3{pool[Conv1*3(IN)]};

[0111]

[0112] Among them, IN is the feature input, relu is the relu activation layer, input is the input data, ConvA and ConvB are two convolution calculation methods, Conv1×1 is a 1*1 convolution layer, Conv1*3 is a 1*3 convolution layer, pool is a 1*2 average pooling layer, fc is a fully connected layer, and Output is the output data;

[0113] The monitoring data anomaly evaluation unit performs an anomaly evaluation on the monitoring model feature data to obtain monitoring model evaluation data; the monitoring data anomaly output unit outputs anomaly types according to the scores of the monitoring model evaluation data;

[0114] In order to achieve dynamic regulation of the entire process, the power grid front channel abnormality monitoring model is used to automatically acquire and analyze real-time data, extract the time series and morphological characteristics of the data, conduct targeted classification evaluation based on their respective characteristics, and obtain the specific abnormality type based on the evaluation results; by adding a deep learning analysis model to the analysis process, an automated and intelligent analysis process can be achieved, and data analysis can be performed in real time when data is collected, thus improving the intelligence of the entire solution;

[0115] Among them, the abnormal evaluation calculation of the monitoring data abnormal evaluation unit is:

[0116] AE i =α1·S 1i +α2·S 2i ;

[0117] α1+α2=1;

[0118]

[0119] Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, β li is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of α1 and α2 is set to 0.5 and 0.5 by default, and can be adjusted within the allowable range of the formula according to actual conditions; β li and β lkIt can be obtained based on historical data, or it can be set according to historical data within the allowable range of the formula;

[0120] In order to improve the accuracy of anomaly classification in the process of automated analysis, by weighted evaluation of the extracted morphological features and time series features, the feature set of each abnormal data obtained based on a large amount of historical data, and by calculating the similarity of morphological features and time series features between the current data and the historical abnormal data, the accuracy of the abnormal classification evaluation process can be improved. Through multi-faceted evaluation and analysis, it can help the intelligent analysis process and improve the reliability of automated adjustment.

[0121] In order to determine the data division time T, the present invention calculates the anomaly recognition accuracy of historical data with different division times T. The selected historical data is the same data, including 5 months of data, which includes abnormal data and non-abnormal data. The present invention counts the abnormal recognition results, as shown in Table 1:

[0122] Table 1 Anomaly recognition accuracy at different partitioning times T

[0123] Divide time T Anomaly recognition accuracy 5 minutes 85.7% 8 minutes 89.6% 10 minutes 92.4% 15 minutes 94.3% 30 minutes 93.8% 45 minutes 91.6% 1 hour 90.4%

[0124] From Table 1, we can see that the 15-minute division time has the highest recognition accuracy;

[0125] If the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormal threshold, the abnormal channel number is recorded, the channel number without abnormal information is obtained, and the abnormal channel number is switched;

[0126] If the abnormal information of the status record data subset exceeds the status record abnormal threshold, the abnormal server number is recorded, the server number without abnormal information is obtained, and the server number is switched to the abnormal server number;

[0127] Further, the channel monitoring abnormality threshold and the status record abnormality threshold are obtained by obtaining historical abnormality data corresponding to various abnormal information based on historical channel monitoring data and historical status record data; the historical abnormality data is divided according to the division time T, and the duration S is obtained according to the division time T; the division time T is the same as the division time T of the channel monitoring data subset and the status record data itself;

[0128] Count the duration S of various historical abnormal data and obtain the maximum value S of the abnormal duration S max and abnormal persistent minimum S min , with S max and S min Establish an abnormal interval to obtain its corresponding threshold; wherein, the duration S is specifically obtained, for example, the division time T is 5 minutes, and the historical abnormal data occupies 3 division times T, so the duration S is 15 minutes;

[0129] Since historical abnormal data is uncertain and has no clear regularity, if it is set manually, some special situations may be missed or the range may be too large, resulting in some sudden fluctuation data being set as historical abnormal data; and in the process of transmission, data abnormalities are inevitable. In order to automatically determine when the automatic dynamic adjustment of channels and servers is required, based on a large amount of historical abnormal data, the historical data is divided according to the current division time T, so as to unify the time dimension with the time dimension of model analysis, and determine the abnormal interval as the threshold of dynamic adjustment in combination with the time division result. Through the corresponding data quantization strategy, this method can have a more accurate data reference in automatic adjustment, effectively improving the intelligence level of scheduling;

[0130] If both the channel monitoring data subset and the status record data subset are abnormal, perform attribution analysis based on the channel abnormal information and server abnormal information, obtain the front channel abnormal information, obtain the corresponding solution strategy through the front channel abnormal information, and switch according to the channel number, server number and scheduling strategy;

[0131] Furthermore, the front channel abnormal information acquisition is to obtain the monitoring abnormal information y of the channel monitoring data subset c and the state abnormal information y of the state record data subset s ; Based on the monitoring abnormal information and the status abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s );

[0132] The abnormal information coordinates are clustered into front channel abnormalities to obtain cluster similarity. The cluster with the largest similarity is the abnormal type of the abnormal information coordinates, and the abnormal type is the front channel abnormal information.

[0133] The cluster similarity is calculated as follows:

[0134]

[0135] Among them, P y is the anomaly information of the front channel, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula; the fuzzy clustering parameter is defined based on historical data, and can also be defined by summarizing artificial rules. The value range of the fuzzy clustering parameter is θ>1;

[0136] In the clustering process, it is very likely that an abnormal point is included in multiple abnormal ranges. Therefore, it is necessary to calculate based on cluster similarity in order to accurately obtain the cause of the abnormality; refer to Figure 3 As shown in the figure, A and B are two cluster centers, C is an outlier, and C is included in both the cluster range of A and the cluster range of B during the clustering process. In order to determine which type of anomaly C belongs to, a cluster similarity calculation is performed. The calculation results are shown in the figure. Figure 3 As shown, B is 0.87 and A is 0.52, so point C belongs to the cluster range of B;

[0137] When both the channel and the server have problems, the cause of the exception may be a problem in the scheduling process, or it may be a unilateral problem or even a problem in both the channel and the server. Therefore, after obtaining the abnormal data, in order to accurately and dynamically adjust the complex fault problems, the coordinates generated by the current problem label are clustered with the historical data through the clustering results of the historical data. The specific cause of the fault can be accurately analyzed through the clustering similarity calculation results. Based on the cause, a more accurate solution can be achieved to improve the intelligence level of the entire scheduling process;

[0138] Embodiment 2

[0139] The present invention also provides a front channel adjustment system for a power grid monitoring system based on risk prediction. Through this system, when channel problems and / or server problems occur, the system can automatically analyze the collected data according to the provided algorithms and solutions, and can intelligently handle the problems according to the analysis results and corresponding solutions, which can greatly improve the automation and intelligence of the solution. The system includes a channel monitoring data division module, a state recording data division module, an abnormal information acquisition module, a channel monitoring data abnormality processing module, a state recording data abnormality processing module and a multi-abnormality processing module, refer to Figure 4 As shown, specifically:

[0140] The channel monitoring data division module assigns channel numbers to the channel monitoring data according to the channel types and obtains the monitoring data of each channel; sorts and divides the channel monitoring data to generate a subset of the channel monitoring data;

[0141] The status record data division module associates the status record data with the channel number and generates a server number according to the connection relationship between each channel and the server, obtains the status record data of each server, divides the status record data according to the division conditions of the channel monitoring data subset, and generates a status record data subset;

[0142] The abnormal information acquisition module identifies the abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time;

[0143] Furthermore, the identification of abnormal information is obtained based on the abnormal monitoring model of the power grid front channel;

[0144] The power grid front channel abnormality monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit;

[0145] The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data;

[0146] The monitoring data anomaly evaluation unit performs an anomaly evaluation on the monitoring model feature data to obtain monitoring model evaluation data; the monitoring data anomaly output unit outputs anomaly types according to the scores of the monitoring model evaluation data;

[0147] By introducing the power grid front channel abnormality monitoring model into the system, the system can automatically identify data anomalies after a large amount of historical data training. Combined with the feature recognition and classification capabilities of deep learning, the system can quickly perform intelligent recognition on real-time collected data, while ensuring that it does not affect the channel and server data collection, and improve the overall intelligence level of the system;

[0148] Furthermore, the abnormality evaluation calculation of the monitoring data abnormality evaluation unit is:

[0149] AE i =α1·S 1i +α2·S 2i ;

[0150] α1+α2=1;

[0151]

[0152] Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, βli is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of

[0153] In order to ensure the accuracy of system recognition, the time series features and morphological features extracted by the monitoring data feature extraction unit of the model are combined with historical data to obtain the similarity of various types of abnormal data and obtain the abnormal features existing in the data; the data calculation results obtained based on big data analysis can help improve the accuracy of the system's evaluation and judgment, and further improve the intelligence of the system;

[0154] The channel monitoring data exception processing module records the abnormal channel number, obtains the channel number without abnormal information and switches with the abnormal channel number if the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormal threshold;

[0155] The status record data exception processing module records the abnormal server number if the abnormal information of the status record data subset exceeds the status record abnormal threshold, obtains and switches to the server number without abnormal information;

[0156] The multi-exception processing module, if both the channel monitoring data subset and the status record data subset are abnormal, performs attribution analysis based on the channel abnormal information and server abnormal information, obtains the front channel abnormal information, obtains the corresponding scheduling strategy through the front channel abnormal information, and switches according to the channel number, server number and scheduling strategy;

[0157] Furthermore, the front channel abnormal information acquisition of the multi-abnormal processing module is to obtain the monitoring abnormal information y of the channel monitoring data subset c and the state abnormal information y of the state record data subset s ; Based on the monitoring abnormal information and the status abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s );

[0158] The abnormal information coordinates are clustered for front channel abnormalities to obtain cluster similarity. The abnormal type of the abnormal information coordinates with the largest cluster similarity is the abnormal type of the front channel abnormal information. In order to ensure the accuracy and efficiency of the corresponding measures taken by the system when complex problems occur, clustering features are obtained through a large amount of historical data. When multiple abnormal problems occur, the system can accurately obtain the abnormal causes based on the feature similarity of the abnormal problems, and automatically adopt corresponding solutions to deal with them. This method can effectively improve the intelligence of the system when dealing with complex problems.

[0159] The cluster similarity is calculated as follows:

[0160]

[0161] Among them, P y is the anomaly information of the front channel, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula.

[0162] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A front channel adjustment method for a power grid monitoring system based on risk prediction, characterized in that: include: Channel monitoring data are numbered according to the channel type, and the monitoring data of each channel is obtained; Sorting the channel monitoring data and dividing it into a plurality of subsets to generate channel monitoring data subsets; According to the connection relationship between each channel and the server, the status record data is associated with the channel number and a server number is generated, the status record data of each server is obtained, and the status record data is divided according to the division condition of the channel monitoring data subset to generate a status record data subset; Identify abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time; If the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormality threshold, the abnormal channel number is recorded, the channel number without the abnormal information is obtained, and the abnormal channel number is switched; If the abnormal information of the status record data subset exceeds the status record abnormal threshold, the abnormal server number is recorded, the server number without the abnormal information is obtained, and the server number is switched to the abnormal server number; If abnormalities occur in both the channel monitoring data subset and the status record data subset, attribution analysis is performed based on the channel abnormality information and the server abnormality information, the front channel abnormality information is obtained, the corresponding solution strategy is obtained through the front channel abnormality information, and switching is performed according to the channel number, the server number and the scheduling strategy.

2. The method for adjusting the front channel of a power grid monitoring system based on risk prediction according to claim 1 is characterized in that The channel monitoring data is sorted according to usage status, usage duration, start time of use and channel attributes; The plurality of channel monitoring data subsets divide the channel monitoring data according to the division time T.

3. The method for adjusting the front channel of a power grid monitoring system based on risk prediction according to claim 1 is characterized in that The method for identifying abnormal information includes: The identification of the abnormal information is obtained based on the power grid front channel abnormality monitoring model; The power grid front channel abnormality monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit; The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data; The monitoring data anomaly assessment unit performs an anomaly assessment on the monitoring model feature data to obtain monitoring model assessment data; the monitoring data anomaly output unit outputs an anomaly type according to a score of the monitoring model assessment data.

4. The method for adjusting the front channel of a power grid monitoring system based on risk prediction according to claim 3 is characterized in that The abnormality evaluation calculation of the monitoring data abnormality evaluation unit includes: AE i =α1·S 1i +α2·S 2i ; α1+α2=1; Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, β li is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of .

5. The method for adjusting the front channel of a power grid monitoring system based on risk prediction according to claim 1 is characterized in that The acquisition of the channel monitoring abnormality threshold and the status record abnormality threshold includes: Obtain historical abnormal data corresponding to each type of abnormal information according to historical channel monitoring data and historical status record data; divide the historical abnormal data according to a division time T, and obtain a duration S according to the division time T; Count the duration S of various historical abnormal data and obtain the maximum value S of the abnormal duration S max and abnormal persistent minimum S min , with S max and S min Establish an abnormal interval and obtain its corresponding threshold.

6. The method for adjusting the front channel of a power grid monitoring system based on risk prediction according to claim 1 is characterized in that The front channel abnormal information acquisition includes: Get the monitoring abnormal information y of the channel monitoring data subset c and the state abnormality information y of the state record data subset s ; Based on the monitoring abnormal information and the state abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s ); Performing front channel anomaly clustering on the abnormal information coordinates to obtain clustering similarity, wherein the cluster with the largest similarity is the abnormal type of the abnormal information coordinates, and the abnormal type is the front channel abnormal information; The cluster similarity is calculated as follows: Among them, P y is the front channel abnormal information, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula.

7. The front channel adjustment system of the power grid monitoring system based on risk prediction is characterized by: include: A channel monitoring data division module is used to assign channel numbers to the channel monitoring data according to the channel types and obtain the channel monitoring data; Sorting and dividing the channel monitoring data to generate channel monitoring data subsets; A status record data division module, according to the connection relationship between each channel and the server, associates the status record data with the channel number and generates a server number, obtains the status record data of each server, divides the status record data according to the division condition of the channel monitoring data subset, and generates a status record data subset; An abnormal information acquisition module, which identifies abnormal information of the channel monitoring data subset and the corresponding status record data subset in real time; A channel monitoring data exception processing module records the abnormal channel number, obtains the channel number without the abnormal information and switches with the abnormal channel number if the abnormal information of the channel monitoring data subset exceeds the channel monitoring abnormal threshold; A status record data exception processing module, if the exception information of the status record data subset exceeds the status record exception threshold, records the abnormal server number, obtains and switches to the server number without the abnormal information; The multi-exception processing module performs attribution analysis based on the channel exception information and the server exception information if exceptions occur in both the channel monitoring data subset and the status record data subset, obtains the front channel exception information, obtains the corresponding scheduling strategy through the front channel exception information, and switches according to the channel number, the server number and the scheduling strategy.

8. The front channel adjustment system of the power grid monitoring system based on risk prediction according to claim 7 is characterized in that The abnormal information acquisition module includes: The identification of the abnormal information is obtained based on the power grid front channel abnormality monitoring model; The power grid front channel abnormality monitoring model includes a monitoring data input unit, a monitoring data feature extraction unit, a monitoring data abnormality evaluation unit and a monitoring data abnormality output unit; The monitoring data input unit preprocesses the channel monitoring data subset or the corresponding status record data subset to generate monitoring model input data; the monitoring data feature extraction unit is used to extract the time series features and morphological features of the monitoring model input data to generate monitoring model feature data; The monitoring data anomaly assessment unit performs an anomaly assessment on the monitoring model feature data to obtain monitoring model assessment data; the monitoring data anomaly output unit outputs an anomaly type according to a score of the monitoring model assessment data.

9. The front channel adjustment system of the power grid monitoring system based on risk prediction according to claim 7 is characterized in that The abnormality evaluation calculation of the monitoring data abnormality evaluation unit includes: AE i =α1·S 1i +α2·S 2i ; α1+α2=1; Among them, AE i is the abnormal evaluation score of the i-th data, S 1i is the abnormal morphology score of the i-th data, S 2i is the abnormal time series score of the i-th data, α1 and α2 are evaluation weights, S li For S 1i or S 2i , κ l (x i ,x i ) is the abnormal feature x of the i-th data i The kernel function, n l is the number of data similarity calculations, y li is the normal label of the i-th data, β li is the normal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the similarity center x between the abnormal data and the normal data of the i-th data lj The kernel function, y lk is the abnormal label of the i-th data, β lk is the abnormal penalty coefficient of the i-th data, κ l (x i ,x lj ) is the center of the normal similarity of the i-th data and the abnormal similarity x lk The kernel function of .

10. The front channel adjustment system of the power grid monitoring system based on risk prediction according to claim 7, characterized in that The front channel exception information acquisition of the multi-exception processing module includes: Get the monitoring abnormal information y of the channel monitoring data subset c and the state abnormality information y of the state record data subset s ; Based on the monitoring abnormal information and the state abnormal information, the abnormal information coordinate y is formed unm =(y c ,y s ); Performing front channel anomaly clustering on the abnormal information coordinates to obtain clustering similarity, wherein the cluster with the largest similarity is the abnormal type of the abnormal information coordinates, and the abnormal type is the front channel abnormal information; The cluster similarity is calculated as follows: Among them, P y is the front channel abnormal information, PN is the number of cluster centers, y m is the coordinate of the mth cluster center point, y m+v is the coordinate of the m+vth cluster center point, θ is the fuzzy clustering parameter, max{} is the maximum value, and D() is the distance calculation formula.