A risk monitoring method for operating state of distribution transformer

Abstract

The present application belongs to the technical field of data acquisition control system, and relates to a risk monitoring method for operation state of distribution transformer. An edge fusion gateway collects voltage, current, oil temperature and environmental humidity data of the distribution transformer in real time, receives a distribution network topology structure change message to generate a local topology correlation matrix, performs convolution operation on the multi-dimensional data and the matrix to extract an edge feature vector reflecting the electrical coupling state and uploads the main station, calculates the operation state risk value according to the abnormal fluctuation component in the edge feature vector, issues a control instruction when the risk value exceeds the threshold, and drives the low-voltage side switch to perform the opening operation according to the instruction. The present application introduces the network topology physical correlation into the edge side data fusion, eliminates the risk false alarm caused by topology reconstruction, realizes the risk source positioning, reduces the backbone communication link transmission load, and shortens the control response time delay under abnormal working conditions.

Description

Technical Field

[0001] This invention belongs to the field of data acquisition and control system technology, and relates to a risk monitoring method for the operating status of a distribution transformer. Background Technology

[0002] Current distribution transformer data acquisition and control systems typically employ a centralized processing architecture. Various sensors are deployed locally on the transformer to acquire operational data such as voltage, current, and oil temperature. This raw data is then directly transmitted to the central master station via a communication network. Upon receiving this raw data, the risk control assessment model within the master station performs isolated analysis on the current data thresholds or short-term historical data of individual transformer devices to determine if there is a risk of overload or insulation abnormalities. Throughout the entire monitoring process, terminal devices are only responsible for data acquisition and forwarding; they do not participate in data feature extraction or topology correlation calculations.

[0003] The aforementioned existing technologies cannot accurately identify the actual operational risks of transformers when the distribution network topology undergoes dynamic reconfiguration. Distribution network operation often results in topology changes due to line switching or load transfers. At this time, the electrical coupling state between adjacent nodes changes, and the transient fluctuations caused by power flow transfer directly affect the transformer itself. Because the central master station of the existing technology receives isolated raw data from individual devices and does not incorporate physical correlation information of the network topology, the master station's risk control assessment model may misjudge the coupling fluctuations between adjacent nodes caused by dynamic topology reconfiguration as transformer failures. This leads to frequent false alarms during normal switching operations in the distribution network, and the inability to pinpoint the specific spatial location of the fluctuation source when a real risk occurs. Summary of the Invention

[0004] The purpose of this invention is to provide a risk monitoring method for the operating status of distribution transformers, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A risk monitoring method for the operating status of a distribution transformer is applied to a data acquisition and control system, which includes an edge fusion gateway and a master station server. The method includes: the edge fusion gateway acquiring voltage data, current data, oil temperature data, and ambient humidity data of the distribution transformer in real time, and receiving distribution network topology change messages; the edge fusion gateway generating a local topology correlation matrix based on the distribution network topology change messages; the edge fusion gateway performing convolution operations on the voltage data, current data, oil temperature data, and ambient humidity data with the local topology correlation matrix to extract edge feature vectors reflecting the electrical coupling status of the distribution transformer with adjacent nodes, and uploading the edge feature vectors to the master station server; the master station server calculating the operating status risk value of the distribution transformer based on the abnormal fluctuation components in the edge feature vectors; when the operating status risk value of the distribution transformer exceeds a set threshold, the master station server issuing a control command to the edge fusion gateway; and the edge fusion gateway driving the low-voltage side switch of the distribution transformer to perform a tripping operation according to the control command.

[0006] Preferably, the edge fusion gateway generates a local topology association matrix based on the distribution network topology change message, including: the edge fusion gateway parsing the distribution network topology change message to obtain the current closed switch state and node connection relationship; the edge fusion gateway determining the first-level adjacent nodes directly connected to the distribution transformer and the second-level adjacent nodes connected through the first-level adjacent nodes based on the node connection relationship; the edge fusion gateway obtaining the historical impedance parameters of the first-level adjacent nodes and the second-level adjacent nodes, and constructing an initial adjacency matrix in combination with the current closed switch state; and the edge fusion gateway normalizing the elements in the initial adjacency matrix and using the processed matrix as the local topology association matrix.

[0007] Preferably, the convolution operation of the voltage data, current data, oil temperature data, and ambient humidity data with the local topology association matrix by the edge fusion gateway includes: the edge fusion gateway concatenating the voltage data and current data according to a preset time window to generate an electrical input matrix, and concatenating the oil temperature data and ambient humidity data to generate a non-electrical input matrix; the edge fusion gateway performing feature mapping on the electrical input matrix using a preset first convolution kernel, and performing feature mapping on the non-electrical input matrix using a preset second convolution kernel; the edge fusion gateway merging the mapped electrical feature map and non-electrical feature map, performing a dot product operation on the merged feature map and the local topology association matrix, and using the dot product operation result as the edge feature vector.

[0008] Preferably, the process by which the master station server calculates the risk value of the distribution transformer's operating status based on the abnormal fluctuation components in the edge feature vector includes: the master station server inputting the received edge feature vector into a pre-established time-series decomposition network; the time-series decomposition network decomposing the edge feature vector and outputting trend components and residual components; the master station server calculating the second derivative sequence of the residual components and marking data points in the second derivative sequence whose absolute values ​​are greater than a preset fluctuation threshold as abnormal fluctuation components; the master station server counting the cumulative occurrences and durations of the abnormal fluctuation components within the current time window, and inputting the cumulative occurrences and durations into a preset risk quantification function to calculate the risk value of the distribution transformer's operating status.

[0009] Preferably, when the risk value of the operating status of the distribution transformer exceeds a set threshold, the master station server sends a control command to the edge fusion gateway, including: the master station server compares the risk value of the operating status of the distribution transformer with multiple preset risk level intervals to determine the current target risk level interval; the master station server extracts the delay parameter and the switching action sequence corresponding to the target risk level interval, and encapsulates the delay parameter and the switching action sequence to generate the control command; the edge fusion gateway drives the low-voltage side switch of the distribution transformer to perform a tripping operation according to the control command, including: the edge fusion gateway parses the control command to obtain the delay parameter and the switching action sequence, and after waiting for the time corresponding to the delay parameter, sequentially disconnects each phase contact of the low-voltage side switch according to the switching action sequence.

[0010] Preferably, the edge fusion gateway's real-time acquisition of voltage, current, oil temperature, and ambient humidity data of the distribution transformer includes: the edge fusion gateway synchronously acquires the voltage and current data at a first sampling rate via a built-in analog-to-digital converter, and acquires the oil temperature and ambient humidity data at a second sampling rate via a built-in temperature and humidity sensor; when missing sampling points are detected in the oil temperature or ambient humidity data, the edge fusion gateway extracts the data values ​​of multiple valid sampling points adjacent to the missing sampling point, and fills in the missing sampling point using a cubic spline interpolation algorithm; the edge fusion gateway downsamples the filled oil temperature and ambient humidity data to align the timestamps of the data points of the processed oil temperature and ambient humidity data with the timestamps of the data points of the voltage and current data.

[0011] Preferably, the edge fusion gateway acquires the historical impedance parameters of the first-level adjacent nodes and the second-level adjacent nodes, and constructs an initial adjacency matrix in conjunction with the current closed switch state, including: the edge fusion gateway reads the historical impedance parameters of the previous monitoring period from the local storage area; the edge fusion gateway identifies the target switch in the current closed switch state that has undergone a state flip, and acquires the location identifier of the target switch; the edge fusion gateway locates the initial impedance value of the corresponding branch in the historical impedance parameters according to the location identifier, and reads the contact resistance increment corresponding to the current open / close state of the target switch; the edge fusion gateway superimposes the initial impedance value and the contact resistance increment to generate updated branch impedance parameters; the edge fusion gateway fills the updated branch impedance parameters into the corresponding positions of the blank matrix generated by the node connection relationship to generate the initial adjacency matrix.

[0012] Preferably, the edge fusion gateway performs channel merging of the mapped electrical feature map and non-electrical feature map, and performs a dot product operation between the merged feature map and the local topology association matrix, including: the edge fusion gateway flattens the electrical feature map into a one-dimensional electrical vector and the non-electrical feature map into a one-dimensional non-electrical vector according to row priority; the one-dimensional electrical vector and the one-dimensional non-electrical vector are concatenated end-to-end to generate a concatenated vector, and the concatenated vector is reshaped into a two-dimensional joint feature matrix; the edge fusion gateway extracts the target row vector corresponding to the distribution transformer in the local topology association matrix, expands the target row vector into a target topology matrix with the same number of rows and columns as the two-dimensional joint feature matrix; the edge fusion gateway calculates the product of the elements at the same position in the two-dimensional joint feature matrix and the target topology matrix, and uses the matrix formed by all products as the result of the dot product operation.

[0013] Preferably, the temporal decomposition network decomposes the edge feature vector and outputs a trend component and a residual component, comprising: a first fully connected layer, a recurrent network layer, and a second fully connected layer connected in sequence; the first fully connected layer performs a linear transformation on the edge feature vector and outputs a hidden state vector; the recurrent network layer receives the hidden state vector and calculates the cell state and hidden layer output at the current time step through an internal forget gate, an input gate, and an output gate; the second fully connected layer receives the hidden layer output and maps the hidden layer output through two sets of parallel output neurons, one of which outputs the trend component and the other outputs the residual component; before outputting the trend component and the residual component, the temporal decomposition network performs a smoothing filter on the trend component and the residual component.

[0014] Preferably, the edge fusion gateway uses a cubic spline interpolation algorithm to fill in the missing sampling points, including: the edge fusion gateway extracts a first preset number of forward valid sampling points and a second preset number of backward valid sampling points adjacent to the missing sampling points, and constructs an interpolation node set from the extracted sampling points; the edge fusion gateway establishes a system of cubic spline interpolation equations containing cubic polynomial coefficients based on the data values ​​and timestamps of each node in the interpolation node set; the edge fusion gateway introduces the boundary first derivative of the first node in the forward valid sampling points and the boundary first derivative of the last node in the backward valid sampling points as constraints; the edge fusion gateway uses a chasing method to solve the system of cubic spline interpolation equations with the constraints, obtains the values ​​of each cubic polynomial coefficient, and substitutes the timestamp of the missing sampling point into the corresponding cubic polynomial to calculate the filled value.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces a local topology correlation matrix construction and convolution operation mechanism at the edge fusion gateway of the data acquisition and control system to jointly extract features from multi-dimensional transformer acquisition data and the physical structure of the distribution network topology. While acquiring voltage, current, oil temperature, and ambient humidity data, the edge fusion gateway generates a local topology correlation matrix containing the connection relationships between primary and secondary adjacent nodes, based on distribution network topology change messages. This matrix is ​​then mapped to the electrical and non-electrical input matrices using double convolution kernels, and channel merging is performed. Finally, a dot product operation is performed with the topology matrix. This processing method fuses the operating status of individual devices with the electrical coupling status of the entire distribution network before data upload, removing interference components caused by dynamic topology reconstruction and power flow transfer between adjacent nodes. This ensures that the edge feature vectors acquired by the master station accurately reflect the state of the transformer itself, eliminating false alarms during normal switching operations of the distribution network. Furthermore, it achieves physical location of risk sources based on features containing spatial topology information.

[0016] 2. This invention incorporates multiple physical constraints and dynamic adjustment mechanisms in the data preprocessing and risk quantification stages. During the data acquisition phase, for multidimensional data acquired at different sampling frequencies, a cubic spline interpolation algorithm based on first-order derivative constraints at the boundary is used to fill in missing sampling points, and downsampling alignment is performed to ensure the continuity and consistency of the data input to the convolution operation in the time dimension. In the risk assessment phase, the main station extracts residual components from the edge feature vectors through a time-series decomposition network, identifies abnormal fluctuation components using the second-order derivative sequence, quantifies the risk value by combining the cumulative occurrence frequency and duration, and disconnects the low-voltage side switch phase by phase based on the matched delay parameters and switching action sequence when the risk exceeds the limit. This process transforms risk monitoring from fixed threshold comparison to a comprehensive judgment based on temporal evolution trends and topological correlations, avoiding false trips caused by sudden disturbances and realizing physical constraint control of the tripping operation sequence under abnormal transformer operating conditions. Attached Figure Description

[0017] Figure 1 This is the overall flowchart of the risk monitoring of the operating status of the distribution transformer according to the present invention; Figure 2 This is a flowchart illustrating the data preprocessing and local topology association matrix construction process of the present invention. Figure 3 The flowchart of the cubic spline interpolation algorithm of the present invention for filling missing sampling points is shown below; Figure 4 This is a flowchart of the convolution operation between multidimensional data and topological correlation matrix in this invention; Figure 5 This is a flowchart of the internal processing of the time-series decomposition network of the present invention; Figure 6 This is a flowchart of the risk value calculation and hierarchical control based on time-series decomposition of the present invention. Detailed Implementation

[0018] The risk monitoring method for the operating status of distribution transformers disclosed in this invention is applied to a data acquisition and control system. The data acquisition and control system includes an edge fusion gateway and a master station server. The edge fusion gateway is deployed on the local side of the distribution transformer and establishes wired communication connections with the low-voltage side switches, voltage transformers, current transformers, oil temperature sensors, and ambient humidity sensors of the distribution transformer. The master station server is deployed in the distribution network dispatch center and establishes a two-way encrypted data interaction link with all edge fusion gateways in the jurisdiction through a dedicated wide area communication network for the distribution network. The communication process uses the SM2 national cryptographic algorithm to complete message encryption and authentication, and message transmission follows the IEC60870-5-104 communication protocol.

[0019] refer to Figure 1In one embodiment, the edge fusion gateway collects voltage, current, oil temperature, and ambient humidity data of the distribution transformer in real time and receives distribution network topology change messages. Specifically, the voltage data includes the instantaneous values ​​of the three-phase phase voltage and line voltage on the high-voltage and low-voltage sides of the distribution transformer; the current data includes the instantaneous values ​​of the three-phase phase current and line current on the high-voltage and low-voltage sides of the distribution transformer; the oil temperature data includes the top-layer oil temperature and winding oil temperature data of the distribution transformer; and the ambient humidity data is the relative humidity data of the environment at the installation location of the distribution transformer. The distribution network topology change message is generated by the distribution network dispatching system after performing switching operations, line switching, or load transfer operations. The message includes a switch status identifier, node connection relationship identifier, operation timestamp, and CRC32 checksum. The edge fusion gateway receives the message through a dedicated distribution network communication channel, performs CRC32 verification on the message, extracts the valid data fields in the message if the verification passes, and discards the message and sends a message anomaly notification to the main station server if the verification fails.

[0020] The edge fusion gateway generates a local topology association matrix based on the distribution network topology change message. Specifically, the edge fusion gateway parses the node connection relationship field in the verified topology change message and constructs a two-dimensional blank matrix containing the connection relationships between the target node where the distribution transformer is located and its adjacent nodes. The rows and columns of the matrix correspond to the physical node numbers in the distribution network, respectively. Based on the switch status identifier and node electrical parameters in the message, the gateway fills the corresponding positions in the two-dimensional blank matrix with electrical connection strength values ​​between nodes. After standardizing all elements in the matrix, a local topology association matrix is ​​generated. The element values ​​of the local topology association matrix range from 0 to 1, with higher element values ​​corresponding to stronger electrical coupling between two nodes.

[0021] The edge fusion gateway performs convolution operations on the voltage data, current data, oil temperature data, and ambient humidity data with the local topology association matrix to extract edge feature vectors reflecting the electrical coupling state between the distribution transformer and adjacent nodes. These edge feature vectors are then uploaded to the main station server. Specifically, the edge fusion gateway synchronizes and aligns the collected multi-dimensional operational data according to the time dimension, generating a time-consistent two-dimensional input matrix. This two-dimensional input matrix is ​​then convolved with the local topology association matrix. During the operation, the electrical coupling association information between the distribution transformer and adjacent nodes is retained, while interference components unrelated to the transformer's operating state are filtered out. After the operation, a one-dimensional edge feature vector is generated. The edge fusion gateway encapsulates the edge feature vector, along with the corresponding timestamp and the unique identifier of the distribution transformer, into an uplink data packet. This packet is then uploaded to the main station server via an encrypted communication channel. The uplink data packet transmission period is consistent with the data acquisition time window length.

[0022] The master server calculates the risk value of the distribution transformer's operating status based on the abnormal fluctuation components in the edge feature vector. Specifically, after receiving the uplink data packet, the master server decrypts and verifies the integrity of the packet, extracts the edge feature vector within the packet, performs time-series decomposition processing on the edge feature vector, separates the abnormal fluctuation components characterizing the changes in the transformer's physical state, and calculates the risk value of the distribution transformer's operating status based on the temporal distribution characteristics of the abnormal fluctuation components within the current time window. The risk value ranges from 0 to 1, with higher values ​​corresponding to a higher degree of risk in the distribution transformer's operating status.

[0023] When the risk value of the operating status of the distribution transformer exceeds a set threshold, the master station server sends a control command to the edge fusion gateway. Specifically, the master station server has a preset risk threshold in its local storage area, which corresponds to the safe operating boundary of the distribution transformer. The master station server compares the calculated operating status risk value with the set threshold in real time. When the risk value is greater than the set threshold, it generates a control command containing switch opening operation parameters, encrypts the control command using the SM2 algorithm, and sends it to the edge fusion gateway of the corresponding distribution transformer.

[0024] The edge fusion gateway drives the low-voltage side switch of the distribution transformer to perform a tripping operation according to the control command. Specifically, after receiving the control command, the edge fusion gateway decrypts and verifies the validity of the command, including the sender's identity, the command's validity period, and operating permissions. After successful verification, the gateway parses the operating parameters within the control command, generates the corresponding switch drive level signal, and outputs the drive signal to the tripping control circuit of the low-voltage side switch of the distribution transformer. This drives the operating mechanism of the low-voltage side switch to perform the tripping operation, disconnecting the low-voltage side output circuit of the distribution transformer and completing the emergency protection for abnormal operation of the distribution transformer.

[0025] Specifically, in this embodiment, the mapping relationship between the distribution network topology nodes and the elements of the local topology association matrix is ​​defined by Table 1.

[0026] Table 1. Mapping Relationship between Distribution Network Topology Nodes and Local Topology Elements in the Association Matrix Table 1 lists node types including distribution transformer nodes, line nodes, load nodes, switch nodes, and bus nodes. Matrix element positions are identified in the form of (row number, column number). Initial element values ​​correspond to the electrical connection status between nodes. When there is a direct electrical connection between two nodes, the initial element value is the positive-sequence impedance per unit of the branch between the nodes; when there is no direct electrical connection between two nodes, the initial element value is 0. The initial values ​​of the diagonal elements of the matrix are the self-impedance per unit of the corresponding node. Based on this table, the correspondence between physical nodes of the distribution network and elements of the local topology association matrix can be realized, ensuring that the topology association matrix accurately reflects the physical connection relationships of the distribution network.

[0027] In this embodiment, the local topology association matrix is ​​constructed and the features of multi-dimensional operating data are extracted through the edge fusion gateway. Before the data is uploaded to the master station server, the transformer operating data and the distribution network topology association information are fused. This ensures that the edge feature vector obtained by the master station server contains the electrical coupling status information of the distribution transformer and adjacent nodes, avoiding the interference of power flow fluctuations caused by topology changes on risk assessment. The master station server calculates the risk value based on the edge feature vector and performs a tripping operation through the edge fusion gateway when the risk exceeds the limit, thereby realizing closed-loop risk monitoring of the operating status of the distribution transformer.

[0028] refer to Figure 2 In a preferred embodiment, the edge fusion gateway synchronously acquires voltage and current data at a first sampling rate via a built-in analog-to-digital converter, and acquires oil temperature and ambient humidity data at a second sampling rate via a built-in temperature and humidity sensor. The first and second sampling rates are preset fixed values, with the first sampling rate being greater than the second sampling rate. The edge fusion gateway has a preset sampling parameter configuration file in its local storage area. This configuration file contains the first sampling rate, the second sampling rate, the sampling channel number, and the measurement range parameters. After startup, the edge fusion gateway initializes and calibrates the acquisition channels according to the parameters in the configuration file.

[0029] refer to Figure 3When missing sampling points are detected in the oil temperature data or the ambient humidity data, the edge fusion gateway extracts the data values ​​of multiple valid sampling points adjacent to the missing sampling point and fills in the missing sampling point using a cubic spline interpolation algorithm. Specifically, the edge fusion gateway performs continuity verification on the collected oil temperature data and ambient humidity data. When the time interval between the timestamp of a certain sampling point and the timestamps of its adjacent sampling points is greater than 1.5 times the sampling time interval corresponding to the second sampling rate, the sampling point is determined to be a missing sampling point. The edge fusion gateway extracts a first preset number of forward valid sampling points and a second preset number of backward valid sampling points adjacent to the missing sampling point, and constructs the extracted sampling points into an interpolation node set. Preferably, both the first preset number and the second preset number are set to 3, and the interpolation node set contains 6 valid sampling points, corresponding to the 3 consecutive valid sampling points before and after the missing sampling point, respectively.

[0030] The edge fusion gateway establishes a system of cubic spline interpolation equations containing cubic polynomial coefficients based on the data values ​​and timestamps of each node in the interpolation node set. For any two adjacent nodes in the interpolation node set... and The corresponding cubic spline polynomial is: in, The timestamp of the sampling point For the first The timestamp of each interpolation node , , , The first The constant term, linear term, quadratic term, and coefficient of the cubic term in a segmental cubic polynomial. For timestamps The corresponding interpolation calculation value.

[0031] The edge fusion gateway introduces the boundary first derivative of the first node in the forward valid sampling points and the boundary first derivative of the last node in the backward valid sampling points as constraints. Specifically, the boundary first derivative of the first node is set as the ratio of the numerical difference to the time difference between the first two nodes in the forward valid sampling points, and the boundary first derivative of the last node is set as the ratio of the numerical difference to the time difference between the last two nodes in the backward valid sampling points. The corresponding constraint expressions are: in, The timestamp of the first node in the interpolation node set. This is the timestamp of the last node in the interpolation node set. , , , These are the sampled values ​​for the corresponding nodes. Let be the boundary first derivative of the first node. Let be the boundary first derivative of the last node. This represents the total number of nodes in the interpolation node set.

[0032] The edge fusion gateway uses a catch-up method to solve the cubic spline interpolation equations with the added constraints, obtaining the values ​​of the coefficients of each cubic polynomial. The timestamps of the missing sampling points are then substituted into the corresponding cubic polynomials to calculate the missing values. Specifically, the cubic spline interpolation equations are transformed into a tridiagonal linear equation system. The catch-up method decomposes the tridiagonal matrix into the product of a lower triangular matrix and an upper triangular matrix. The equation system is solved in two steps: forward substitution and backward substitution, to obtain the coefficients of each polynomial segment, and then the missing sampling point values ​​are calculated.

[0033] The edge fusion gateway downsamples the filled-in oil temperature data and the ambient humidity data to align the timestamps of the processed oil temperature and humidity data with the timestamps of the voltage and current data. The calculation expression for downsampling is: in, For the voltage and current data, the first The timestamp of each data point To align the oil temperature or ambient humidity data to the timestamp The value at that location, To reduce the sampling rate, It equals the ratio of the first sampling rate to the second sampling rate. This is the sampling time interval corresponding to the second sampling rate. This refers to the original oil temperature or ambient humidity data after filling.

[0034] Furthermore, the edge fusion gateway parses the distribution network topology change message to obtain the current closed switch status and node connection relationship. The edge fusion gateway performs protocol parsing on the verified topology change message, extracting the switch status field from the message. This field contains the on / off status identifiers of all relevant switches in the distribution network; 0 indicates the switch is in the open state, and 1 indicates the switch is in the closed state. The gateway also extracts the node connection relationship field from the message, which contains the number, type, and physical connection relationship of each node in the distribution network.

[0035] The edge fusion gateway determines the first-level adjacent nodes directly connected to the distribution transformer and the second-level adjacent nodes connected through the first-level adjacent nodes based on the node connection relationships. Specifically, the node where the distribution transformer is located is the target node, the nodes connected to the target node through direct electrical branches are first-level adjacent nodes, and the nodes connected to the first-level adjacent nodes through direct electrical branches but not including the target node are second-level adjacent nodes. The edge fusion gateway numbers and stores the first-level and second-level adjacent nodes to construct a set of adjacent nodes for the target node. The total number of nodes in the adjacent node set determines the dimension of the subsequently generated topology association matrix.

[0036] The edge fusion gateway acquires the historical impedance parameters of the first-level and second-level adjacent nodes and constructs an initial adjacency matrix based on the current closed switch state. Specifically, the edge fusion gateway reads the historical impedance parameters from the local storage area for the previous monitoring period. These historical impedance parameters include the positive-sequence impedance per-unit values ​​of the branches between each node in the adjacent node set. The edge fusion gateway identifies the target switch in the current closed switch state that has undergone a state flip. A state flip refers to a change in the switch's open / closed state compared to the previous monitoring period. The edge fusion gateway acquires the location identifier of the target switch, which corresponds to the two connection node numbers of the branch where the switch is located.

[0037] The edge fusion gateway locates the initial impedance value of the corresponding branch in the historical impedance parameters based on the location identifier, reads the contact resistance increment corresponding to the current opening and closing state of the target switch, and the update expression for the branch impedance parameter is: in, For the updated branch impedance parameters, This represents the initial impedance value of the corresponding branch. This represents the increase in contact resistance corresponding to the current open / closed state of the target switch. When the switch is in the closed state... This is the rated contact resistance per unit value of the switch when the switch is in the open position. Since the value is infinity, the corresponding branch impedance is set to infinity.

[0038] The edge fusion gateway fills the updated branch impedance parameters into the corresponding positions of the blank matrix generated by the node connection relationship, thus generating the initial adjacency matrix. Specifically, the number of rows and columns of the blank matrix are equal to the total number of nodes in the adjacent node set, and the rows and columns of the matrix correspond to the node numbers, respectively. For any element in the matrix... When node With nodes When there is a direct electrical connection, The value is the updated impedance parameter for the corresponding branch, when the node With nodes When there is no direct electrical connection between them, The value is 0, and the diagonal elements of the matrix are... The value is the self-impedance value of the corresponding node.

[0039] The edge fusion gateway normalizes the elements in the initial adjacency matrix and uses the processed matrix as the local topology association matrix. The expression for the normalization process is: in, The first in the local topological correlation matrix Line 1 The element values ​​of the column, This represents the element value at the corresponding position in the initial adjacency matrix. The maximum value of all non-zero elements in the initial adjacency matrix. The minimum value of all non-zero elements in the initial adjacency matrix, when When it is 0, The value remains 0.

[0040] Specifically, in this embodiment, the node parameters and polynomial coefficients of the cubic spline interpolation are recorded in Table 2.

[0041] Table 2. Results of Solving Nodal Parameters and Polynomial Coefficients for Cubic Spline Interpolation In Table 2, the interpolation node number corresponds to the sequence number of the valid sampling point before and after the missing sampling point, the timestamp is the acquisition time corresponding to the sampling point, the sampling value is the valid data value of oil temperature or ambient humidity collected by the sensor, the boundary first derivative value is the constraint condition of the interpolation equation system, and the coefficients a, b, c, and d are the corresponding coefficients of the cubic spline polynomials obtained by solving each segment. Based on the parameters in this table, the filling value of any missing sampling point can be directly calculated, ensuring the reproducibility of the interpolation calculation process.

[0042] In this embodiment, missing temperature and humidity data are filled in using a cubic spline interpolation algorithm, and timestamp alignment of multi-dimensional collected data is achieved through downsampling processing, ensuring the continuity and consistency of input data in the time dimension. First- and second-level adjacent nodes are determined by parsing topology change messages, and an initial adjacency matrix is ​​constructed by updating branch impedance parameters in combination with switch status. After normalization processing, a local topology correlation matrix is ​​generated, which can accurately reflect the electrical connection relationship of nodes after the change of distribution network topology structure, providing accurate topological physical constraints for subsequent feature fusion calculations.

[0043] refer to Figure 4In a preferred embodiment, the edge fusion gateway concatenates the voltage data and current data according to a preset time window to generate an electrical input matrix, and concatenates the oil temperature data and ambient humidity data to generate a non-electrical input matrix. The preset time window is a fixed-length time interval containing a preset number of time-aligned data points. Preferably, the time window contains 128 consecutive time-aligned data points, and the sliding step size of the time window is consistent with the data acquisition cycle. Specifically, the electrical input matrix generation process is as follows: the edge fusion gateway arranges the three-phase voltage data and three-phase current data within the time window according to a time dimension, with each row corresponding to a timestamp and each column corresponding to an electrical parameter. The generation dimension is... The electrical input matrix, where The number of data points within the time window. The total number of channels for electrical parameters, in this embodiment The value equals 6, corresponding to the three-phase phase voltage and three-phase phase current, respectively. The generation process of the non-electrical input matrix is ​​as follows: the edge fusion gateway arranges the oil temperature data and ambient humidity data within the time window according to the same time dimension, with each row corresponding to a timestamp and each column corresponding to a non-electrical parameter. The generation dimension is... The non-electrical input matrix, where The total number of channels is a non-electrical parameter, as described in this embodiment. The value is 2, corresponding to oil temperature and ambient humidity, respectively.

[0044] The edge fusion gateway performs feature mapping on the electrical input matrix using a preset first convolutional kernel and on the non-electrical input matrix using a preset second convolutional kernel. Both the first and second convolutional kernels are one-dimensional, with a stride of 1 and identical padding, ensuring the length of the output feature map in the time dimension matches that of the input matrix. The weight matrices and bias terms of both the first and second convolutional kernels are pre-stored in the local storage of the edge fusion gateway. After each fixed monitoring cycle, the edge fusion gateway receives a convolutional kernel parameter update command from the master server and updates the parameters of both kernels to adapt to changes in the distribution network topology and operating conditions.

[0045] The feature mapping expression of the first convolution kernel to the electrical input matrix is: in, The resulting electrical feature map is shown below. For electrical input matrix, The weight matrix of the first convolution kernel. This is the bias term for the first convolution kernel. This is a one-dimensional convolution operator. Let be the linear rectified activation function. The expression for the linear rectified activation function is: .

[0046] The feature mapping expression of the second convolution kernel to the non-electrical input matrix is: in, The non-electrical feature map obtained by mapping, It is a non-electrical input matrix. The weight matrix of the second convolution kernel. This is the bias term for the second convolution kernel.

[0047] Preferably, the first convolutional kernel has a kernel size of 3 and 16 output channels, and the second convolutional kernel has a kernel size of 3 and 8 output channels, such that the dimension of the electrical feature map is [missing value]. The dimension of the non-electrical feature map is .

[0048] The edge fusion gateway performs channel merging on the mapped electrical feature map and non-electrical feature map, performs a dot product operation on the merged feature map and the local topology association matrix, and uses the result of the dot product operation as the edge feature vector. Specifically, the edge fusion gateway flattens the electrical feature map into a one-dimensional electrical vector and the non-electrical feature map into a one-dimensional non-electrical vector in row-major order. The one-dimensional electrical vector and the one-dimensional non-electrical vector are then concatenated end-to-end to generate a concatenated vector, which is then reshaped into a two-dimensional joint feature matrix. The row-major flattening process involves sequentially concatenating the elements of each row of the electrical feature map, generating a dimension of... A one-dimensional electrical vector is generated by concatenating the elements of each row of the non-electrical feature map end-to-end, resulting in a vector of dimension 1. A one-dimensional non-electrical vector is concatenated to the end of a one-dimensional electrical vector to generate a vector of dimension 1. The concatenated vector is reshaped into a vector with dimension [missing information]. The two-dimensional joint feature matrix has each row corresponding to a timestamp and each column corresponding to a fused feature channel.

[0049] The edge fusion gateway extracts the target row vector corresponding to the distribution transformer from the local topology association matrix and expands the target row vector into a target topology matrix with the same number of rows and columns as the two-dimensional joint feature matrix. Specifically, the target row vector is the row vector of the node where the distribution transformer is located in the local topology association matrix. The elements in the row vector contain the electrical connection strength values ​​between the distribution transformer and all first- and second-level adjacent nodes. The edge fusion gateway repeatedly expands the target row vector, and the number of rows after expansion is equal to the number of rows in the two-dimensional joint feature matrix. Each row's elements are identical to the target row vector, generating a dimension of... The target topology matrix is ​​expanded by repeatedly filling the elements of the target row vectors so that the number of columns in the target topology matrix is ​​consistent with the number of columns in the two-dimensional joint feature matrix.

[0050] The edge fusion gateway calculates the product of the elements at the same position in the two-dimensional joint feature matrix and the target topology matrix, and takes the matrix formed by all products as the result of the dot product operation. The expression for the dot product operation is: in, This is the result of the dot product operation. It is a two-dimensional joint characteristic matrix. The target topology matrix, This is the Hadamard product operator, which multiplies the elements at the same position in two matrices. The edge fusion gateway flattens the result of the dot product operation into a one-dimensional vector in row-major order, and uses this one-dimensional vector as the edge feature vector. The dimension of the edge feature vector is... It includes the multi-dimensional operating characteristics of distribution transformers and the topological characteristics of distribution networks.

[0051] Specifically, in this embodiment, the correspondence between the input and output dimensions and parameters of each stage of the convolution operation is clarified in Table 3.

[0052] Table 3. Correspondence between input / output dimensions and parameters for convolution operations. Table 3 shows that the computational process includes two stages: electrical feature mapping and non-electrical feature mapping. The dimension of the input matrix corresponds to the number of rows and columns of the input data in each stage. The kernel size is the length of a one-dimensional convolution kernel. The stride is the sliding step size of the convolution operation. The number of output channels is the number of feature channels generated after the convolution operation. The dimension of the output feature map corresponds to the number of rows and columns of the feature map generated after the convolution operation. The activation function type corresponds to the type of non-linear activation function used after the convolution operation. Based on this table, the computational process of convolutional feature mapping can be completely reproduced, ensuring that the feature extraction process is fully disclosed.

[0053] In this embodiment, feature mapping is performed on the electrical input matrix and the non-electrical input matrix using dual convolution kernels to extract deep features of different types of operating data. A joint feature matrix is ​​generated by channel merging. The joint feature matrix is ​​then multiplied with the expanded topological correlation matrix, achieving deep fusion of transformer operating features and distribution network topological electrical coupling features. The generated edge feature vector can accurately reflect the operating status of the distribution transformer in the distribution network topology and remove interference components caused by power flow fluctuations of adjacent nodes.

[0054] refer to Figure 6 In a preferred embodiment, the master station server inputs the received edge feature vectors into a pre-established time-series decomposition network. The time-series decomposition network is pre-trained using historical operating data of the distribution network. The training dataset contains edge feature vector samples under different topologies and operating conditions, along with corresponding trend and residual component labels. During training, the mean squared error loss function is used to optimize the network parameters, and the Adam optimizer is used to update the parameters. The learning rate is set to 0.001, the batch size is 32, and the training iterations are 1000. When the loss function value of the validation set no longer decreases after 50 consecutive iterations, training is terminated early, and the optimal network parameters are saved. The trained time-series decomposition network is deployed in the local computing unit of the master station server.

[0055] refer to Figure 5 The temporal decomposition network decomposes the edge feature vector, outputting trend and residual components. The temporal decomposition network comprises a first fully connected layer, a recurrent network layer, and a second fully connected layer connected sequentially. The recurrent network layer employs a long short-term memory (LSTM) network structure, effectively capturing the long-term dependency features of temporal data. The first fully connected layer performs a linear transformation on the edge feature vector, outputting a hidden state vector. The input dimension of the first fully connected layer is the same as the dimension of the edge feature vector, and the output dimension is 64. The expression for the linear transformation is: in, This is the hidden state vector output by the first fully connected layer. This is the weight matrix of the first fully connected layer. The input edge feature vector, This is the bias term for the first fully connected layer.

[0056] The recurrent network layer receives the hidden state vector and calculates the cell state and hidden layer output at the current time step through internal forget gates, input gates, and output gates. The gating calculation process for each time step is as follows: in, For the current time step, For the output of the forget gate, The output of the input gate, For the output of the output gate, This represents the candidate cell state at the current time step. This represents the cell state at the current time step. This represents the cell state at the previous time step. This is the output of the hidden layer at the current time step. This is the output of the hidden layer from the previous time step. The input is the hidden state vector at the current time step. , , , These are the weight matrices for the corresponding gating. , , , These are the bias terms for the corresponding gating. The sigmoid activation function is expressed as follows: , It is the hyperbolic tangent activation function.

[0057] The second fully connected layer receives the output of the hidden layer and maps it to the hidden layer output through two sets of parallel output neurons. One output neuron outputs the trend component, and the other output neuron outputs the residual component. The second fully connected layer contains two sets of parallel fully connected neurons. The first set of neurons has an output dimension of 1 and outputs the trend component, while the second set of neurons has an output dimension of 1 and outputs the residual component. The mapping expression is as follows: in, For the first The trend component at each time step For the first The residual components at each time step , These are the weight matrices for the two groups of neurons. , These are the bias terms for the two groups of neurons. Before outputting the trend component and the residual component, the temporal decomposition network performs a smoothing filter on the trend component and the residual component. Preferably, the smoothing filter uses a moving average filter with a sliding window length of 5. The filtered trend component and residual component can remove high-frequency noise interference and retain effective temporal variation characteristics.

[0058] The main server calculates the second derivative sequence of the residual components and marks data points in the second derivative sequence whose absolute values ​​are greater than a preset fluctuation threshold as the abnormal fluctuation components. For discrete residual component sequences... The expression for calculating its second derivative sequence is: in, For the residual component in the th The second derivative value at each time step , , The first , , The residual component values ​​at each time step. The main server has a preset fluctuation threshold, which is determined based on the standard deviation of the second derivative sequence of the residual components under historical normal operating conditions. At that time, the data point at that time step is marked as an abnormal fluctuation component, where This is the preset fluctuation threshold.

[0059] The main server counts the cumulative occurrences and durations of the abnormal fluctuation components within the current time window, and inputs these counts and durations into a preset risk quantification function to calculate the operational risk value of the distribution transformer. in, This represents the risk value of the operating status of the distribution transformer. This represents the cumulative number of occurrences of the abnormal fluctuation component within the current time window. The longest duration of the abnormal fluctuation component within the current time window. and The preset weighting coefficients, and The values ​​are all greater than 0, and are determined based on the statistical analysis results of historical fault data.

[0060] The master server compares the risk value of the operating status of the distribution transformer with multiple preset risk level ranges to determine the current target risk level range. The master server's local storage area has multiple consecutive risk level ranges preset, each corresponding to a different risk level and control strategy. Preferably, there are four risk level ranges: low risk, medium risk, high risk, and emergency risk. The low risk range is 0 to 0.3, the medium risk range is 0.3 to 0.6, the high risk range is 0.6 to 0.9, and the emergency risk range is 0.9 to 1.

[0061] The main server extracts the delay parameters and switch action sequences corresponding to the target risk level intervals, and encapsulates the delay parameters and switch action sequences to generate the control command. Each risk level interval corresponds to a unique delay parameter and switch action sequence. The delay parameter is the waiting time from the issuance of the control command to the execution of the switch opening operation. The switch action sequence is the opening order of each phase contact of the low-voltage side switch. The delay parameter corresponding to the low-risk interval and the medium-risk interval is infinite, that is, no opening operation is triggered, only a warning signal is output; the delay parameter corresponding to the high-risk interval is 30 seconds, and the switch action sequence is that phases A, B, and C open in sequence; the delay parameter corresponding to the emergency risk interval is 0 seconds, and the switch action sequence is that all three phases open simultaneously.

[0062] The edge fusion gateway parses the control command to obtain the delay parameter and the switching action sequence. After waiting for the time corresponding to the delay parameter, it sequentially disconnects the contacts of each phase of the low-voltage side switch according to the switching action sequence. Specifically, after parsing the control command, the edge fusion gateway starts a hardware timer with a duration equal to the delay parameter. When the timer expires, it outputs a tripping level signal to the contact drive circuit of the corresponding phase in the order specified in the switching action sequence, completing the tripping operation of the low-voltage side switch. After the tripping operation is completed, the edge fusion gateway sends an operation completion feedback message to the main station server.

[0063] Specifically, the mapping relationship between risk level ranges and corresponding control parameters in this embodiment is clearly defined in Table 4.

[0064] Table 4. Mapping Table of Risk Level Ranges and Corresponding Control Parameters In Table 4, the risk level corresponds to the risk value classification level, the risk value range is the range of risk values ​​for the corresponding level, the delay parameter is the waiting time from the issuance of the control command to the tripping operation, the switch action sequence is the tripping sequence of each phase contact of the low-voltage side switch, and the operation type is the operation type triggered under the corresponding risk level. Based on this table, the risk level and control strategy can be accurately matched to ensure that the tripping operation is hierarchically controllable.

[0065] In this embodiment, the edge feature vector is decomposed by a temporal decomposition network containing a long short-term memory network structure, separating the trend component and the residual component. The abnormal fluctuation component is identified based on the second derivative sequence of the residual component. The risk value is quantified by combining the cumulative occurrence and duration of the abnormal fluctuation component, thus realizing risk judgment based on temporal evolution characteristics. By matching the corresponding delay parameters and switching action sequences with the risk level interval, hierarchical control of the tripping operation is realized, avoiding switch maloperation caused by sudden disturbances and ensuring the safe and stable operation of the distribution transformer.

Claims

1. A risk monitoring method for the operating status of a distribution transformer, applied to a data acquisition and control system, characterized in that, The data acquisition and control system includes an edge fusion gateway and a main station server. The method includes: the edge fusion gateway collects voltage data, current data, oil temperature data and ambient humidity data of the distribution transformer in real time, and receives distribution network topology change messages. The edge fusion gateway generates a local topology association matrix based on the power distribution network topology change message; The edge fusion gateway performs convolution operations on the voltage data, current data, oil temperature data, and ambient humidity data with the local topology association matrix to extract edge feature vectors that reflect the electrical coupling state between the distribution transformer and adjacent nodes, and uploads the edge feature vectors to the main station server. The main station server calculates the operating status risk value of the distribution transformer based on the abnormal fluctuation component in the edge feature vector. When the risk value of the operating status of the distribution transformer exceeds the set threshold, the master station server sends a control command to the edge fusion gateway; The edge fusion gateway drives the low-voltage side switch of the distribution transformer to perform a tripping operation according to the control command.

2. The risk monitoring method for the operating status of a distribution transformer according to claim 1, characterized in that, The edge fusion gateway generates a local topology association matrix based on the power distribution network topology change message, including: the edge fusion gateway parses the power distribution network topology change message to obtain the current closed switch status and node connection relationship; The edge fusion gateway determines the primary adjacent nodes directly connected to the distribution transformer and the secondary adjacent nodes connected through the primary adjacent nodes based on the node connection relationship. The edge fusion gateway obtains the historical impedance parameters of the first-level adjacent nodes and the second-level adjacent nodes, and constructs an initial adjacency matrix by combining the current closed switch state; The edge fusion gateway normalizes the elements in the initial adjacency matrix and uses the processed matrix as the local topology association matrix.

3. The risk monitoring method for the operating status of a distribution transformer according to claim 1, characterized in that, The edge fusion gateway performs convolution operations on the voltage data, current data, oil temperature data, and ambient humidity data with the local topology association matrix, including: the edge fusion gateway concatenates the voltage data and current data according to a preset time window to generate an electrical input matrix, and concatenates the oil temperature data and ambient humidity data to generate a non-electrical input matrix; The edge fusion gateway uses a preset first convolution kernel to perform feature mapping on the electrical input matrix and a preset second convolution kernel to perform feature mapping on the non-electrical input matrix. The edge fusion gateway merges the mapped electrical feature map with the non-electrical feature map, performs a dot product operation on the merged feature map and the local topology association matrix, and uses the result of the dot product operation as the edge feature vector.

4. The risk monitoring method for the operating status of a distribution transformer according to claim 1, characterized in that, The main station server calculates the risk value of the operating status of the distribution transformer based on the abnormal fluctuation components in the edge feature vector, including: the main station server inputs the received edge feature vector into a pre-established time-series decomposition network; The temporal decomposition network decomposes the edge feature vector and outputs trend components and residual components; The main station server calculates the second derivative sequence of the residual components and marks the data points in the second derivative sequence whose absolute values ​​are greater than a preset fluctuation threshold as the abnormal fluctuation components. The main station server counts the cumulative number of occurrences and duration of the abnormal fluctuation component within the current time window, and inputs the cumulative number of occurrences and the duration into a preset risk quantification function to calculate the risk value of the operating status of the distribution transformer.

5. The risk monitoring method for the operating status of a distribution transformer according to claim 1, characterized in that, When the risk value of the operating status of the distribution transformer exceeds a set threshold, the master station server sends a control command to the edge fusion gateway, including: the master station server compares the risk value of the operating status of the distribution transformer with multiple preset risk level ranges to determine the current target risk level range; The main station server extracts the delay parameters and switching action sequences corresponding to the target risk level range, and encapsulates the delay parameters and switching action sequences to generate the control command; The edge fusion gateway drives the low-voltage side switch of the distribution transformer to perform a tripping operation according to the control command, including: the edge fusion gateway parses the control command to obtain the delay parameter and the switch action sequence, and after waiting for the time corresponding to the delay parameter, disconnects each phase contact of the low-voltage side switch in sequence according to the switch action sequence.

6. The risk monitoring method for the operating status of a distribution transformer according to claim 1, characterized in that, The edge fusion gateway collects voltage data, current data, oil temperature data, and ambient humidity data of the distribution transformer in real time, including: the edge fusion gateway synchronously collects the voltage data and the current data at a first sampling rate through a built-in analog-to-digital converter, and collects the oil temperature data and the ambient humidity data at a second sampling rate through a built-in temperature and humidity sensor. When missing sampling points are detected in the oil temperature data or the ambient humidity data, the edge fusion gateway extracts the data values ​​of multiple valid sampling points adjacent to the missing sampling point and uses a cubic spline interpolation algorithm to fill in the missing sampling points numerically. The edge fusion gateway downsamples the filled oil temperature data and the ambient humidity data to align the data point timestamps of the processed oil temperature data and ambient humidity data with the data point timestamps of the voltage data and the current data.

7. The risk monitoring method for the operating status of a distribution transformer according to claim 2, characterized in that, The edge fusion gateway obtains the historical impedance parameters of the first-level adjacent nodes and the second-level adjacent nodes, and constructs an initial adjacency matrix in combination with the current closed switch state, including: the edge fusion gateway reads the historical impedance parameters of the previous monitoring period from the local storage area; The edge fusion gateway identifies the target switch that has undergone a state flip in the current closed switch state and obtains the location identifier of the target switch; The edge fusion gateway locates the initial impedance value of the corresponding branch in the historical impedance parameters based on the location identifier, and reads the contact resistance increment corresponding to the current opening and closing state of the target switch. The edge fusion gateway superimposes the initial impedance value with the contact resistance increment to generate updated branch impedance parameters. The edge fusion gateway fills the updated branch impedance parameters into the corresponding positions of the blank matrix generated by the node connection relationship to generate the initial adjacency matrix.

8. The risk monitoring method for the operating status of a distribution transformer according to claim 3, characterized in that, The edge fusion gateway performs channel merging of the mapped electrical feature map and non-electrical feature map, and performs dot multiplication operation between the merged feature map and the local topology association matrix. This includes: the edge fusion gateway flattens the electrical feature map into a one-dimensional electrical vector in row priority order, flattens the non-electrical feature map into a one-dimensional non-electrical vector, concatenates the one-dimensional electrical vector and the one-dimensional non-electrical vector end to end to generate a concatenated vector, and reshapes the concatenated vector into a two-dimensional joint feature matrix. The edge fusion gateway extracts the target row vector corresponding to the distribution transformer from the local topology association matrix, and expands the target row vector into a target topology matrix with the same number of rows and columns as the two-dimensional joint feature matrix. The edge fusion gateway calculates the product of the elements at the same position in the two-dimensional joint feature matrix and the target topology matrix, and uses the matrix formed by all the products as the result of the dot product operation.

9. The risk monitoring method for the operating status of a distribution transformer according to claim 4, characterized in that, The temporal decomposition network decomposes the edge feature vector and outputs trend components and residual components, including: the temporal decomposition network comprises a first fully connected layer, a recurrent network layer and a second fully connected layer connected in sequence; The first fully connected layer performs a linear transformation on the edge feature vector and outputs a hidden state vector; The recurrent network layer receives the hidden state vector and calculates the cell state and hidden layer output at the current time step through the internal forget gate, input gate and output gate; The second fully connected layer receives the output of the hidden layer and maps the output of the hidden layer through two sets of parallel output neurons. One output neuron outputs the trend component, and the other output neuron outputs the residual component. Before outputting the trend component and the residual component, the time series decomposition network performs smoothing filtering on the trend component and the residual component.

10. The risk monitoring method for the operating status of a distribution transformer according to claim 6, characterized in that, The edge fusion gateway uses a cubic spline interpolation algorithm to fill in the missing sampling points by extracting a first preset number of forward valid sampling points and a second preset number of backward valid sampling points adjacent to the missing sampling points, and constructing the extracted sampling points into an interpolation node set. The edge fusion gateway establishes a set of cubic spline interpolation equations containing cubic polynomial coefficients based on the data values ​​and timestamps of each node in the interpolation node set. The edge fusion gateway introduces the boundary first derivative of the first node in the forward effective sampling points and the boundary first derivative of the last node in the backward effective sampling points as constraints. The edge fusion gateway uses the catch-up method to solve the cubic spline interpolation equation system with the added constraints, obtains the values ​​of the coefficients of each cubic polynomial, and substitutes the timestamps of the missing sampling points into the corresponding cubic polynomials to calculate the filling values.

Images