Data channel state sensing method based on multi-source time series data fusion and topological structure

Abstract

The invention relates to a data channel state sensing method based on multi-source time sequence data fusion and a topological structure. The method comprises the following steps: collecting multi-source time sequence data; cleaning is conducted; obtaining a multi-source fusion time sequence; constructing a knowledge graph structure, wherein the knowledge graph structure comprises data and operation conditions of an equipment link topological structure, an equipment body and topological association equipment; obtaining a trained long short-term memory network model; and inputting real-time data into the trained long-short-term memory network model, predicting the real-time data by the trained model, and dynamically sensing the running state of the equipment. According to the method, a knowledge graph structure is adopted, modeling is carried out on data and operation conditions of an equipment link topological structure, an equipment body and associated equipment, the complex association relation between the equipment can be expressed in the mode, the sensing capacity of the equipment state is improved in combination with the actual operation data, and the real-time performance of the equipment is improved. Therefore, the monitoring system can perform more comprehensive state analysis by using the existing professional knowledge and the association between the devices.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to a data channel state perception method based on multi-source time series data fusion and topological structure. Background Art

[0002] With the continuous development of power systems, substations, as key nodes for power transmission and distribution, have a direct impact on the safety and stability of the power grid. At present, substation equipment monitoring and status perception mainly rely on traditional monitoring methods, which have the following main problems:

[0003] First, traditional monitoring systems usually rely on a single type of data source, such as using only sensor data or historical operation data. The limitation of a single data source is that it cannot fully reflect the operating status of the equipment, which can easily lead to incomplete monitoring information and wrong decisions.

[0004] Second, in the existing technology, the data update frequency of many monitoring systems is low, and they cannot reflect the status changes of the equipment in real time. Especially in the face of sudden failures, the lack of timely data support may delay the fault handling and maintenance time, increasing the risk of system operation;

[0005] Third, as a tool that can represent complex relationships and structured information, knowledge graphs have been applied in many fields. However, the application of knowledge graph technology in power system monitoring is relatively rare.

[0006] Fourth, existing monitoring systems lack comprehensive modeling of equipment and their relationships, and are unable to fully utilize existing expertise and experience to gain a comprehensive understanding of equipment status. Summary of the invention

[0007] In order to solve the problem of incomplete monitoring information and inability to fully utilize the correlation between complex devices and structured information, the purpose of the present invention is to provide a data channel state perception method based on multi-source time series data fusion and topological structure, which can realize the comprehensive utilization of multi-source time series data, ensure the integrity and accuracy of the data, comprehensively reflect the operating status of the equipment, express the complex correlation between devices, and enhance the perception ability of the equipment status.

[0008] To achieve the above object, the present invention adopts the following technical solution: a data channel state perception method based on multi-source time series data fusion and topological structure, the method comprising the following steps in order:

[0009] (1) Collect multi-source time series data from various sensors, equipment logs, historical operation data, and environmental monitoring data in the substation;

[0010] (2) Cleaning the collected multi-source time series data, including data deduplication, data completion and data standardization;

[0011] (3) Using the K-means clustering algorithm to group the cleaned data according to similarity, the grouped data is mined and identified to obtain a multi-source fusion time series;

[0012] (4) constructing a knowledge graph structure, wherein the knowledge graph structure includes a device link topology structure, data and operating conditions of the device body and topology-related devices, wherein the operating conditions include operating performance, historical data, and fault status;

[0013] (5) constructing a long short-term memory network model, inputting the multi-source fusion time series and knowledge graph structure into the long short-term memory network model for training, and obtaining a trained long short-term memory network model;

[0014] (6) The real-time data is input into the trained long short-term memory network model. The trained model predicts the real-time data and dynamically senses the operating status of the equipment. During the real-time monitoring process, the input data is updated through the sliding window, and the long short-term memory network model outputs the equipment status at the current moment.

[0015] In step (1), the sensors include voltage sensors, current sensors and temperature sensors; the device log includes log information generated during the operation of the device; the historical operation data includes the past operation records and performance parameters of the device; and the environmental monitoring data includes temperature, humidity and air pressure.

[0016] In step (2), the data deduplication refers to the use of a density-based spatial clustering algorithm to perform anomaly detection on multi-source time series data. The density-based spatial clustering algorithm analyzes the density distribution of data points, identifies and marks abnormal data points, removes the data points marked as abnormal, and obtains a data set after data deduplication. For the data set after data deduplication, if there are missing values, the Lagrange interpolation method is used to complete the data; finally, data standardization is performed to convert the data into a unified scale and standardize the data.

[0017] The step (3) specifically refers to: grouping the cleaned data using the K-means clustering algorithm:

[0018] C = KMeans(D clean , k)

[0019] Among them, k is the number of clusters, C is the clustering result; D clean is the cleaned data set;

[0020] Apriori algorithm is used to mine frequent phase association rules from the cleaned data set:

[0021] The support calculation formula for frequent itemsets is:

[0022]

[0023] Among them, A is the item set and N is the total number of records;

[0024] The confidence calculation formula of the association rule is:

[0025]

[0026] In the formula, B is the item set;

[0027] Set the support threshold to 0.2 and the confidence threshold to 0.8; when the support of a frequent item set is higher than 0.2 and the confidence is higher than 0.8, the frequent item set is considered to be representative enough and enters the next step of pattern recognition:

[0028] The pattern recognition refers to identifying the time series pattern in the frequent itemset data, and using the DWT dynamic time warping algorithm to calculate the similarity:

[0029]

[0030] Among them, x and y are time series, d(x i ,y i ) is the distance metric function; n is the length of the time series;

[0031] The similarity threshold is set to 0.8. When the similarity of a pair of sequences is higher than 0.8, they are considered to belong to similar patterns, and the time series data is used as the basis for multi-source data fusion, and finally a multi-source fusion time series is obtained.

[0032] The step (4) specifically refers to: constructing a knowledge graph structure including device entity nodes, attribute nodes and relationship edges, the knowledge graph structure is used to represent the device link topology and the relationship between the device body and the associated devices; the device entity nodes include various devices in the substation; the attribute nodes refer to the performance parameters and status information of the equipment; the relationship edges represent the association relationship between the devices; the expression for constructing the knowledge graph KG is:

[0033] KG={(E i , R ij , E j )|i,j∈V}

[0034] Among them, E i and E j is the device entity node, R ij is the relationship edge, V is the node set, i and j represent the node indexes respectively.

[0035] The step (5) specifically refers to:

[0036] The formula of the long short-term memory network model is:

[0037] y t =σ(W h ·h t-1 +W x ·x t +b)

[0038] Among them, y t is the output at the current moment, σ is the activation function, W h and W x are the weight matrices of the hidden layer and the input layer, respectively, h t-1 is the hidden state of the previous moment, x t is the input at the current moment, b is the bias;

[0039] Use 70% of the historical data of the knowledge graph structure as the training set to train the long short-term memory network model, and use the back propagation algorithm for parameter optimization; the loss function of the long short-term memory network model uses the mean square error:

[0040]

[0041] Among them, y j is the true value, is the predicted value, and K is the number of samples.

[0042] It can be seen from the above technical scheme that the beneficial effects of the present invention are as follows: first, the traditional monitoring system only relies on a single type of data source, while the present invention realizes the comprehensive utilization of multi-source time series data by fusing sensor data, equipment logs, historical operation data and environmental monitoring data; through data cleaning and standardization, the integrity and accuracy of the data are ensured, the operation status of the equipment is fully reflected, and the problem of incomplete information from a single data source is solved; second, the knowledge graph structure is adopted to model the equipment link topology structure, the data of the equipment body and the associated equipment and the operating conditions. This method can express the complex association relationship between the equipment, and combined with the actual operation data, it improves the perception ability of the equipment status, so that the monitoring system can use the existing professional knowledge and the association between the equipment to perform a more comprehensive state analysis; third, a long short-term memory network model is constructed, and the historical data of the knowledge graph is combined for training, so that the system can dynamically perceive the operation status of the equipment. The long short-term memory network model can process the input data in real time, update the data through the sliding window mechanism, and timely reflect the changes in the equipment status, especially when a fault occurs, it can provide timely data support and reduce the fault processing time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the method of the present invention;

[0044] Figure 2 It is a schematic diagram of the knowledge graph structure construction;

[0045] Figure 3 It is a structural diagram of the long short-term memory network model of dynamic perception. DETAILED DESCRIPTION

[0046] like Figure 1 As shown, a data channel state perception method based on multi-source time series data fusion and topological structure includes the following steps in order:

[0047] (1) Collect multi-source time series data from various sensors, equipment logs, historical operation data, and environmental monitoring data in the substation;

[0048] (2) Cleaning the collected multi-source time series data, including data deduplication, data completion and data standardization;

[0049] (3) Using the K-means clustering algorithm to group the cleaned data according to similarity, the grouped data is mined and identified to obtain a multi-source fusion time series;

[0050] (4) constructing a knowledge graph structure, wherein the knowledge graph structure includes a device link topology structure, data and operating conditions of the device body and topology-related devices, wherein the operating conditions include operating performance, historical data, and fault status;

[0051] (5) constructing a long short-term memory network model, inputting the multi-source fusion time series and knowledge graph structure into the long short-term memory network model for training, and obtaining a trained long short-term memory network model;

[0052] (6) The real-time data is input into the trained long short-term memory network model. The trained model predicts the real-time data and dynamically senses the operating status of the equipment. During the real-time monitoring process, the input data is updated through the sliding window, and the long short-term memory network model outputs the equipment status at the current moment.

[0053] In step (1), the sensors include voltage sensors, current sensors and temperature sensors; the device log includes log information generated during the operation of the device; the historical operation data includes the past operation records and performance parameters of the device; and the environmental monitoring data includes temperature, humidity and air pressure.

[0054] In step (2), the data deduplication refers to the use of a density-based spatial clustering algorithm to perform anomaly detection on multi-source time series data. The density-based spatial clustering algorithm analyzes the density distribution of data points, identifies and marks abnormal data points, removes the data points marked as abnormal, and obtains a data set after data deduplication. For the data set after data deduplication, if there are missing values, the Lagrange interpolation method is used to complete the data; finally, data standardization is performed to convert the data into a unified scale and standardize the data.

[0055] The step (3) specifically refers to: grouping the cleaned data using the K-means clustering algorithm:

[0056] C = KMeans(D clean , k)

[0057] Among them, k is the number of clusters, C is the clustering result; D clean is the cleaned data set;

[0058] Apriori algorithm is used to mine frequent phase association rules from the cleaned data set:

[0059] The support calculation formula for frequent itemsets is:

[0060]

[0061] Among them, A is the item set and N is the total number of records;

[0062] The confidence calculation formula of the association rule is:

[0063]

[0064] In the formula, B is the item set;

[0065] Set the support threshold to 0.2 and the confidence threshold to 0.8; when the support of a frequent item set is higher than 0.2 and the confidence is higher than 0.8, the frequent item set is considered to be representative enough and enters the next step of pattern recognition:

[0066] The pattern recognition refers to identifying the time series pattern in the frequent itemset data, and using the DWT dynamic time warping algorithm to calculate the similarity:

[0067]

[0068] Among them, x and y are time series, d(x i ,y i ) is the distance metric function; n is the length of the time series;

[0069] By calculating the similarity of different item sets, we can evaluate the similarity of time series between different data sources and identify those sequences with high similarity. The similarity threshold is set to 0.8. When the similarity of a pair of sequences is higher than 0.8, they are considered to belong to similar patterns, and the time series data is used as the basis for multi-source data fusion, and finally a multi-source fusion time series is obtained.

[0070] like Figure 2 As shown, the step (4) specifically refers to: constructing a knowledge graph structure including device entity nodes, attribute nodes and relationship edges, the knowledge graph structure is used to represent the device link topology and the relationship between the device body and the associated devices; the device entity nodes include each device in the substation; the attribute nodes refer to the performance parameters and status information of the equipment; the relationship edges represent the association relationship between the devices; the expression for constructing the knowledge graph KG is:

[0071] KG={(E i , R ij , E j )|i,j∈V}

[0072] Among them, E i and E j is the device entity node, R ij is the relationship edge, V is the node set, i and j represent the node indexes respectively.

[0073] like Figure 3 As shown, the step (5) specifically refers to:

[0074] The formula of the long short-term memory network model is:

[0075] y t =σ(W h ·h t-1 +W x ·x t +b)

[0076] Among them, y t is the output at the current moment, σ is the activation function, W h and W x are the weight matrices of the hidden layer and the input layer, respectively, h t-1 is the hidden state of the previous moment, x t is the input at the current moment, b is the bias;

[0077] Use 70% of the historical data of the knowledge graph structure as the training set to train the long short-term memory network model, and use the back propagation algorithm for parameter optimization; the loss function of the long short-term memory network model uses the mean square error:

[0078]

[0079] Among them, y j is the true value, is the predicted value, and K is the number of samples.

[0080] In summary, the present invention realizes the comprehensive utilization of multi-source time series data by fusing sensor data, equipment logs, historical operation data and environmental monitoring data; through data cleaning and standardization, the integrity and accuracy of the data are ensured, the operating status of the equipment is fully reflected, and the problem of incomplete information from a single data source is solved; the knowledge graph structure is adopted to model the equipment link topology structure, the data of the equipment body and associated equipment, and the operating conditions. This method can express the complex association relationship between devices, and combined with the actual operation data, it improves the perception ability of the equipment status, so that the monitoring system can use the existing professional knowledge and the relationship between devices to perform a more comprehensive status analysis; construct a long short-term memory network model, and train it with the historical data of the knowledge graph, so that the system can dynamically perceive the operating status of the equipment. The long short-term memory network model can process input data in real time, update data through a sliding window mechanism, and promptly reflect changes in equipment status, especially when a fault occurs, it can provide timely data support and reduce fault handling time.

Claims

1. A data channel state perception method based on multi-source time series data fusion and topological structure, characterized by: The method comprises the following steps in order: (1) Collect multi-source time series data from various sensors, equipment logs, historical operation data, and environmental monitoring data in the substation; (2) Cleaning the collected multi-source time series data, including data deduplication, data completion and data standardization; (3) Using the K-means clustering algorithm to group the cleaned data according to similarity, the grouped data is mined and identified to obtain a multi-source fusion time series; (4) constructing a knowledge graph structure, wherein the knowledge graph structure includes a device link topology structure, data and operating conditions of the device body and topology-related devices, wherein the operating conditions include operating performance, historical data, and fault status; (5) constructing a long short-term memory network model, inputting the multi-source fusion time series and knowledge graph structure into the long short-term memory network model for training, and obtaining a trained long short-term memory network model; (6) The real-time data is input into the trained long short-term memory network model. The trained model predicts the real-time data and dynamically senses the operating status of the equipment. During the real-time monitoring process, the input data is updated through the sliding window, and the long short-term memory network model outputs the equipment status at the current moment.

2. The data channel state perception method based on multi-source time series data fusion and topological structure according to claim 1 is characterized by: In step (1), the sensor includes a voltage sensor, a current sensor and a temperature sensor; the device log includes log information generated during the operation of the device; The historical operation data includes the past operation records and performance parameters of the equipment; the environmental monitoring data includes temperature, humidity and air pressure.

3. The data channel state perception method based on multi-source time series data fusion and topological structure according to claim 1 is characterized by: In step (2), the data deduplication refers to the use of a density-based spatial clustering algorithm to perform anomaly detection on multi-source time series data. The density-based spatial clustering algorithm analyzes the density distribution of data points, identifies and marks abnormal data points, removes the data points marked as abnormal, and obtains a data set after data deduplication. For the data set after data deduplication, if there are missing values, the Lagrange interpolation method is used to complete the data; finally, data standardization is performed to convert the data into a unified scale and standardize the data.

4. The data channel state perception method based on multi-source time series data fusion and topological structure according to claim 1 is characterized by: The step (3) specifically refers to: grouping the cleaned data using the K-means clustering algorithm: C=KMeans(D clean ,k) Among them, k is the number of clusters, C is the clustering result; D clean is the cleaned data set; Apriori algorithm is used to mine frequent phase association rules from the cleaned data set: The support calculation formula for frequent itemsets is: Among them, A is the item set and N is the total number of records; The confidence calculation formula of the association rule is: In the formula, B is the item set; Set the support threshold to 0.2 and the confidence threshold to 0.8; when the support of a frequent item set is higher than 0.2 and the confidence is higher than 0.8, the frequent item set is considered to be representative enough and enters the next step of pattern recognition: The pattern recognition refers to identifying the time series pattern in the frequent itemset data, and using the DWT dynamic time warping algorithm to calculate the similarity: Among them, x and y are time series, d(x i ,y i ) is the distance metric function; n is the length of the time series; The similarity threshold is set to 0.

8. When the similarity of a pair of sequences is higher than 0.8, they are considered to belong to similar patterns, and the time series data is used as the basis for multi-source data fusion, and finally a multi-source fusion time series is obtained.

5. The data channel state perception method based on multi-source time series data fusion and topological structure according to claim 1 is characterized by: The step (4) specifically refers to: constructing a knowledge graph structure including device entity nodes, attribute nodes and relationship edges, the knowledge graph structure is used to represent the device link topology and the relationship between the device body and the associated devices; the device entity nodes include various devices in the substation; the attribute nodes refer to the performance parameters and status information of the equipment; the relationship edges represent the association relationship between the devices; the expression for constructing the knowledge graph KG is: KG={(E i , R ij ,E j )|i,j∈V} Among them, E i and E j is the device entity node, R ij is the relationship edge, V is the node set, i and j represent the node indexes respectively.

6. The data channel state perception method based on multi-source time series data fusion and topological structure according to claim 1 is characterized by: The step (5) specifically refers to: The formula of the long short-term memory network model is: y t =σ(W h ·h t-1 +W x ·x t +b) Among them, y t is the output at the current moment, σ is the activation function, W h and W x are the weight matrices of the hidden layer and the input layer, respectively, h t-1 is the hidden state of the previous moment, x t is the input at the current moment, b is the bias; Use 70% of the historical data of the knowledge graph structure as the training set to train the long short-term memory network model, and use the back propagation algorithm for parameter optimization; the loss function of the long short-term memory network model uses the mean square error: Among them, y j is the true value, is the predicted value, and K is the number of samples.

Images