An edge-computing-based power communication data distributed processing system

Abstract

The application relates to the technical field of power system automation, and discloses a power communication data distributed processing system based on edge calculation, which comprises an edge calculation gateway arranged at each node of a power distribution network. A topology mapping module analyzes real-time switch states, constructs a weighted adjacent matrix, and dynamically generates a logical communication routing table consistent with the physical connection of the power grid; a feature extraction module extracts a transient feature vector during a fault and interacts between neighbor nodes; an anchor point election module compares electrical impedance distances, and determines a computing power anchor point that undertakes a computing task according to the minimum impedance rule; a time delay compensation module uses a power frequency voltage phase as a whole-network public reference, and realizes time sequence alignment of multi-source data in combination with a time stamp; and a fault positioning module performs differential calculation based on the aligned data. The application realizes adaptive following of the communication routing to the change of the power grid topology, and significantly improves the real-time performance and accuracy of the distributed fault positioning without relying on external high-precision satellite timing.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, specifically to a distributed processing system for power communication data based on edge computing. Background Technology

[0002] As distribution networks evolve towards active distribution systems, the complexity of line topologies and the requirements for power supply reliability are increasing. To achieve rapid detection and accurate location of distribution line faults, existing distribution automation systems typically employ high-frequency data acquisition schemes based on traveling wave ranging or transient quantity analysis. In the traditional centralized processing architecture, terminal devices distributed at various monitoring points need to upload all the massive amounts of high-frequency waveform data collected to a cloud-based master station or substation for centralized analysis. This model can trigger sudden data transmission surges at the moment a fault occurs, easily causing communication network congestion, data transmission delays, and even packet loss, making it difficult to meet the real-time requirements for millisecond-level fault handling.

[0003] To alleviate communication bandwidth pressure, the industry has begun to introduce edge computing technology, attempting to perform data preprocessing on the terminal side. However, most existing edge computing solutions are based on static network configurations and fail to fully consider the variable characteristics of the physical topology of distribution networks. During operation, distribution networks frequently undergo switching operations due to load transfer, fault isolation, or maintenance work, resulting in changes to the physical connections of primary equipment. If the logical routing of the communication network cannot be synchronized with the physical topology, the data interaction between edge nodes will fail to reflect the actual electrical connection status, leading to a disconnect between fault diagnosis logic and actual current flow, affecting the accuracy of the fault location results.

[0004] Fault location technology based on multi-terminal data differential has extremely high requirements for time synchronization accuracy. Although deploying a Global Navigation Satellite System (GNSS) timing module at each monitoring node can solve the synchronization problem, this significantly increases construction costs and installation and maintenance difficulties, especially in scenarios with weak satellite signals, such as underground utility tunnels or indoor switch stations. If only relying on Network Time Protocol (NTP) for soft synchronization, traditional time compensation algorithms are insufficient to eliminate random errors introduced by network transmission due to the asymmetric routing and queuing delay jitter issues that often exist in power distribution communication networks. This results in the inability to effectively synthesize multi-source data under a unified time reference, limiting the improvement of distributed fault location accuracy. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a power communication data distributed processing system based on edge computing. This system solves the problems of low efficiency in multi-terminal data collaborative processing and poor fault location accuracy in existing power distribution network fault monitoring systems, which suffer from communication routing failure to dynamically adjust with the physical topology and network transmission delay jitter in environments without satellite time synchronization.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a distributed processing system for power communication data based on edge computing, comprising;

[0007] An edge computing gateway deployed at various nodes of the power distribution network, wherein the edge computing gateway is configured with:

[0008] The topology mapping module is used to parse the substation configuration description file and the real-time switch status obtained through process layer messages, construct a weighted adjacency matrix representing the physical connection relationship of the power grid, and dynamically generate a logical communication routing table between edge computing gateways based on the weighted adjacency matrix.

[0009] The feature extraction module is used to start recording when the power signal exceeds the limit. It uses a signal processing algorithm to extract transient feature vectors from the recorded data and sends the transient feature vectors to neighboring nodes with electrical connections through the logical communication routing table. The transient feature vectors include the amplitude, attenuation factor, frequency and phase parameters of the dominant component.

[0010] The anchor point election module is used to receive transient feature vectors from neighboring nodes, combine them with local measurement data to calculate the electrical impedance distance of the fault point relative to the current edge computing gateway, compare the electrical impedance distance of all nodes participating in the same event, and determine the computing power anchor node to undertake the main computing task according to the minimum impedance rule.

[0011] The delay compensation module is used to start when the edge computing gateway is determined to be the computing power anchor node. It uses the phase of the local power grid frequency voltage signal as the common phase reference of the whole network, parses the phase stamp information in the received data packet, calculates the absolute delay value of the data packet in the transmission process of the dynamic routing network, and performs time sequence alignment on the waveform data from different nodes.

[0012] The fault location module is used to perform differential synthesis on multi-source data after time-series alignment, construct a fault location equation based on the traveling wave ranging principle, and calculate the specific location of the fault point.

[0013] Preferably, the topology mapping module executes the following logic when constructing the weighted adjacency matrix:

[0014] For any two nodes, if there is a direct power line connection between them, the weight values ​​of the matrix elements are determined according to the switch status on the power line. When the switch is closed, the weight value is set to the static impedance parameter of the line, and when the switch is open, the weight value is set to infinity.

[0015] When generating a logical communication routing table, the topology mapping module executes a path search algorithm to calculate the electrical distance between nodes. The electrical distance is defined as the sum of the weights of all edges on the shortest connected path between two nodes. Only when the electrical distance is less than a preset effective communication threshold, a routing entry with the adjacent node on the physical path as the next-hop gateway is established in the logical communication routing table.

[0016] Preferably, the feature extraction module uses the Prony algorithm to fit discrete time series data into a linear combination model of complex exponential functions, and separates multiple transient components by establishing a linear prediction model and solving the Vandermonde matrix equation.

[0017] When extracting transient feature vectors, the feature extraction module performs the following filtering steps: first, it removes components whose frequencies are within the tolerance range of the power frequency and its harmonic frequencies; then, it calculates the energy values ​​of the remaining components, selects the component with the largest energy value as the dominant component, and encapsulates the parameters of the dominant component into the transient feature vector.

[0018] Preferably, the anchor election module includes a consensus comparison unit, which is configured to start a receiving time window and collect election messages sent by neighboring nodes that contain electrical impedance distance estimates;

[0019] At the end of the receiving time window, the consensus comparison unit compares the locally calculated electrical impedance distance with the electrical impedance distance of all neighboring nodes. If the locally calculated electrical impedance distance is the minimum value, the current edge computing gateway's role state is switched to anchor state, and an anchor confirmation message is broadcast to the neighboring nodes.

[0020] If the locally calculated electrical impedance distance is not the minimum value, the current edge computing gateway will switch to a subordinate state and establish a connection with the winning computing power anchor node to upload the full waveform data segment.

[0021] Preferably, the anchor point selection module uses an impedance estimation method based on attenuation characteristics correction when calculating the electrical impedance distance:

[0022] First, calculate the ratio of the amplitude of the locally extracted voltage transient dominant component to the amplitude of the current transient dominant component;

[0023] Secondly, the difference between the attenuation factor of the voltage transient dominant component and the attenuation factor of the current transient dominant component is used to calculate the correction coefficient in combination with the preset impedance correction constant with time dimension.

[0024] Finally, the ratio is weighted and adjusted using the correction coefficient to obtain the electrical impedance distance, in order to compensate for the error caused by the non-uniform attenuation of voltage and current signals.

[0025] Preferably, the time delay compensation module includes a phase reference extraction unit and a synchronization alignment calculation unit;

[0026] The phase reference extraction unit is configured to track the change in the phase of the local bus voltage in real time through a software phase-locked loop and normalize the phase value to a standard range.

[0027] The computing power anchor node maintains a phase-time history circular buffer in memory. The synchronization alignment calculation unit retrieves the phase-time history circular buffer based on the sender NTP timestamp carried in the data packet, obtains the corresponding anchor local phase value, and calculates the phase deviation by the difference between the sender phase stamp carried in the packet and the anchor local phase value.

[0028] Preferably, when performing time-series alignment on waveform data from different nodes, the synchronization alignment calculation unit performs the following steps:

[0029] The coarse transmission delay is estimated by using the difference between the NTP timestamp at the transmitting end and the current system time at the receiving end, and then converted into a coarse number of cycles in the phase domain by combining the fundamental frequency of the power grid.

[0030] The coarse number of cycles is corrected based on the phase deviation, and the integer number of complete power frequency cycles that need to be compensated is determined by minimizing the error.

[0031] The phase deviation is converted into an accurate time offset by the integer number of complete power frequency cycles, and this time offset is superimposed on the transmitting end NTP timestamp to obtain an absolute alignment time unified to the receiving end clock domain.

[0032] Preferably, the fault location module includes a data resampling interpolation unit. The data resampling interpolation unit is configured to construct a target time series based on the local sampling clock of the computing power anchor node, process waveform data from neighboring nodes using a cubic spline interpolation algorithm, calculate the value of the piecewise cubic polynomial function passing through the original data points in the target time series by solving the piecewise cubic polynomial function, and generate a resampled waveform sequence that is strictly aligned with the local data of the computing power anchor node on the time axis.

[0033] Preferably, the fault location module further includes a transient wavefront calibration unit, which is configured to process the local waveform data and the resampled waveform sequence using the wavelet transform modulus maxima method, extract the modulus maxima points of the wavelet transform coefficients at each scale and form a modulus maxima line, trace along the modulus maxima line to the minimum scale to determine the convergence point of the modulus maxima line at the minimum scale, and mark the time corresponding to the convergence point as the arrival time of the fault traveling wave front.

[0034] Preferably, when calculating the specific location of the fault point, the fault location module executes the following calculation logic:

[0035] Obtain the total length of the power lines between the computing power anchor node and its neighboring nodes, as well as the propagation speed of the traveling wave signal in the power lines;

[0036] Calculate the time difference between the time when the fault traveling wavefront arrives at the computing power anchor node and the time when it arrives at the neighbor node;

[0037] The product of the time difference and the propagation speed is used as the distance difference term. The total length of the power line is calculated as half of the sum of the distance difference term, and the result is used as the line length from the fault point to the computing power anchor node.

[0038] This invention provides a distributed processing system for power communication data based on edge computing. It has the following advantages:

[0039] 1. This invention constructs a weighted adjacency matrix reflecting the physical connections of the power grid by parsing the substation configuration description file and real-time switch status, and dynamically generates a logical communication routing table based on this matrix. This mechanism enables the communication path of the edge computing gateway to automatically adjust in response to switching operations or fault isolation and reconstruction in the distribution network, ensuring that transient fault data is transmitted only between nodes with direct electrical connections. This effectively solves the problem that traditional static communication configuration cannot adapt to frequent topology changes in the distribution network, avoids invalid data occupying network bandwidth, and ensures the consistency between fault handling logic and actual current flow.

[0040] 2. This invention automatically selects the node with the closest electrical distance to the fault point as the computing power anchor point through consensus comparison among neighboring nodes. This approach abandons the traditional model of relying on centralized processing at the cloud master station, and pushes the computing task down to the edge node with the best signal quality. This not only reduces the latency and packet loss risks caused by long-distance transmission of raw waveform data, but also utilizes high signal-to-noise ratio data near the fault point for calculation, significantly improving the real-time performance and accuracy of fault location.

[0041] 3. This invention utilizes the synchronization characteristics of the voltage phase across the entire distribution network to propose an asymmetric time delay compensation method based on power frequency phase reference. By combining NTP coarse time synchronization with fine voltage phase alignment, and using data resampling interpolation to eliminate asynchronous errors in the sampling clock, the system can achieve sub-cycle-level synchronization of multi-source data without relying on expensive GPS / BeiDou timing hardware deployed across the entire network. This design effectively overcomes the impact of random jitter in the communication network on the accuracy of traveling wave ranging, reducing the hardware construction cost and maintenance difficulty of the system. Attached Figure Description

[0042] Figure 1This is a schematic diagram of the system framework of the present invention;

[0043] Figure 2 This is a hardware structure block diagram of the edge computing gateway of the present invention;

[0044] Figure 3 This is a flowchart of the interactive process for electing computing power anchor points based on electrical impedance weights according to the present invention. Detailed Implementation

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see the appendix Figure 1 - Appendix Figure 3 This invention provides a distributed processing system for power communication data based on edge computing, comprising:

[0047] The topology mapping module is used to parse the substation configuration description and real-time switch status, construct the power grid physical connection diagram, and dynamically generate a logical communication routing table between edge computing gateways based on changes in the power grid physical connection relationship, and establish a virtual communication tunnel that matches the current power grid operation mode.

[0048] The feature extraction module is used to start recording when the power signal exceeds the limit, extract transient feature vectors using signal processing algorithms, and send the transient feature vectors to neighboring nodes with electrical connections through a virtual communication tunnel.

[0049] The anchor point election module is used to receive transient feature vectors from neighboring nodes, combine them with local measurement data to calculate the electrical impedance distance of the fault point relative to the current edge computing gateway, compare the electrical impedance distance of all nodes participating in the same event, and determine the computing power anchor node to undertake the main computing task according to the minimum impedance rule.

[0050] The delay compensation module is activated when the edge computing gateway is identified as a computing power anchor node. It uses the zero-crossing phase of the power grid frequency voltage signal as a reference, parses the phase stamp information in the received data packet, calculates the absolute delay value of the data packet during the transmission process of the dynamic routing network, and performs timing alignment on the waveform data from different nodes.

[0051] The fault location module is used to perform differential synthesis on multi-source data after time alignment, construct fault location equations based on the traveling wave ranging principle, and calculate the specific location and fault type of the fault point.

[0052] The specific implementation principles of each module of the present invention will be explained in detail below with reference to the accompanying drawings.

[0053] The physical communication topology mapping module specifically includes a state acquisition and parsing unit, a topology model construction unit, and a routing strategy generation unit. It is used to establish a real-time mapping relationship between the electrical connections of the power grid physical layer and the logical data transmission paths of the communication layer.

[0054] The status acquisition and parsing unit is configured to receive and parse the substation configuration description (SCD) file and real-time process layer messages through the substation communication network interface. The status acquisition and parsing unit pre-reads the SSD (System Specification Description) section of the SCD file and extracts the primary equipment connection relationships, including circuit breakers and disconnectors, as well as the static parameters of the power lines. The static parameters include at least the line length and unit impedance value.

[0055] The status acquisition and parsing unit monitors GOOSE or SV messages conforming to the IEC 61850 standard in real time and extracts Boolean status bits reflecting the switch position from the message dataset. When a position status bit of any switch is detected to flip, the status acquisition and parsing unit generates a status change event containing the device ID, change time, and current status value, and transmits it to the topology model building unit.

[0056] The topology model building unit is used to maintain the mathematical model of the power grid's physical topology and updates the adjacency matrix in real time based on state change events. The topology model building unit defines the power grid's physical topology graph. ,in This represents a collection of nodes where an edge computing gateway has been deployed. This represents the set of physical power lines between nodes. The topology model building unit maintains a state vector. Used to characterize the on / off status of all switching devices within the area:

[0057] ;

[0058] in, For binary variables, when Time indicates the first When a switch is closed, Time indicates the first One switch is off. This indicates the total number of switching devices in the area. For switch index .

[0059] When receiving state change data causes the state vector to change... Become At that time, the graph update unit performs a matrix update operation. For any two nodes... If there is a direct power line connection between the two, the map update unit will update the map based on the switches on that line. Status and pre-stored line impedance parameters Determine the elements of the weighted adjacency matrix Value:

[0060] like If it is in a closed state, then ;

[0061] like If it is in the disconnected state, then .

[0062] The route generation unit reconstructs the logical topology of the communication network based on the updated weighted adjacency matrix. The route generation unit executes a path search algorithm, traversing all node pairs in the graph. The path search algorithm employs either breadth-first search or Dijkstra's algorithm. During the search process, the route generation unit calculates node... and electrical distance between This distance represents the weights of all edges on the shortest connected path between the two nodes. The sum of .

[0063] The route generation unit manages communication routes according to preset mapping rules:

[0064] Determine in physical topology Does a connected path exist between them, and what is the electrical distance of the connected path? Less than the preset effective communication threshold .

[0065] If the determination result is yes, the route generation unit adds or activates a node in the local routing table. A route entry with the destination address and the next-hop gateway being the adjacent node on the physical path;

[0066] If the result is negative, then remove or disable the route to the node in the local routing table. Directly connected route entries.

[0067] When the operation mode of the physical power grid changes and the network is reconfigured, the state resolution unit detects the change in switch state, triggers the graph update unit to update the weight values ​​of the weighted adjacency matrix, and drives the route generation unit to cancel the original logical links that no longer have electrical connection by rewriting the routing table entries, and establish new logical links, so that the communication reachability of the edge computing gateway is consistent with the current electrical connection range of the power grid.

[0068] The transient feature extraction and multicast module includes a signal monitoring and recording unit, a transient algorithm processing unit, and a feature fingerprint interaction unit.

[0069] The signal monitoring and recording unit connects to the analog input interface of the edge computing gateway to acquire voltage and current signals of the distribution network lines in real time. The signal monitoring and recording unit allocates a first-in-first-out (FIFO) circular buffer in memory to continuously store the most recent... The sampling data of each period, among which This is a configurable integer. The signal monitoring and recording unit is equipped with start-up criterion logic to calculate the instantaneous rate of change of voltage or current in real time. and When the calculated rate of change exceeds a preset threshold, or the sampling amplitude exceeds a set protection value, the signal monitoring and recording unit determines that a disturbance event has occurred. In response to this determination, the signal monitoring and recording unit stops the overwrite operation of the circular buffer and locks the data containing the events preceding the fault. Each cycle and after the fault A continuous sampling data sequence of several cycles is generated, and this full waveform data segment is transmitted to the transient algorithm processing unit.

[0070] The transient algorithm processing unit performs time-frequency domain analysis on full waveform data segments to extract key parameters characterizing fault features. The unit employs the Prony algorithm to fit discrete time-series data into a linear combination model of complex exponential functions.

[0071] Transient algorithm processing unit constructs discrete-time signal model Its mathematical expression is:

[0072] ;

[0073] in, For sampling point number ( ), For data length, The sampling interval is... The model order is... For the first The amplitude of each component, As the attenuation factor, The oscillation frequency is... For the initial phase, Represents the imaginary unit. Represented by natural constant An exponential function with base 0. This is the index for transient components.

[0074] The transient algorithm processing unit executes the following steps for solving the Prony algorithm: First, a linear prediction model is established and the prediction coefficients are solved;

[0075] Secondly, the poles are obtained by finding the roots of the polynomial of the prediction coefficients. Utilizing the extreme point Calculate the attenuation factor and frequency ;

[0076] Finally, the least squares method is used to solve the Vandermonde matrix equation, calculate the complex amplitude, and then separate the amplitude values. and phase .

[0077] The transient algorithm processing unit selects from preset filtering rules. The dominant component is determined from the components. The specific filtering rules are as follows:

[0078] First, eliminate frequency. Components within the power frequency (50Hz or 60Hz) and its harmonic frequency tolerance range, retain transient high-frequency components;

[0079] Then the energy values ​​of the remaining components were calculated. Select energy value The largest component is taken as the dominant component. The transient algorithm processing unit constructs a feature fingerprint vector based on this dominant component. Its data structure is defined as follows:

[0080] ;

[0081] ;

[0082] In the formula, This is the MAC address or unique device identifier of the current edge computing gateway. This is the absolute timestamp of the moment the fault was triggered. This is a feature parameter vector containing the magnitudes of the dominant components. Attenuation factor ,frequency and initial phase .

[0083] The feature fingerprint interaction unit is used to perform data interaction. It retrieves the IP address list of all logical neighbor nodes of the current node by accessing the logical communication routing table generated by the physical communication topology mapping module. It is encapsulated as an application layer message and uses an application layer multicast mechanism, that is, for each neighboring node address in the IP address list, a copy containing the feature fingerprint vector is copied and sent separately. UDP packets.

[0084] The feature fingerprint interaction unit only sends the feature fingerprint vector during the initial interaction phase. Instead of sending the original full waveform data segment stored in the circular buffer, the feature fingerprint interaction unit reads the original full waveform data segment from the buffer and transmits it only when a waveform request message from a specific neighbor node is subsequently received.

[0085] The distributed election module for computing power anchors specifically includes an impedance distance estimation unit, a consensus comparison unit, and a role state management unit.

[0086] The impedance distance estimation unit is configured to calculate physical indicators characterizing the electrical distance of a fault based on locally extracted transient features.

[0087] The impedance distance estimation unit reads the amplitude of the dominant voltage component output by the transient algorithm processing unit. and the amplitude of the dominant current component Calculate the equivalent fault impedance magnitude .

[0088] The impedance calculation formula performed by the impedance distance estimation unit is as follows:

[0089] ;

[0090] In the formula, This represents the estimated equivalent electrical impedance from the local measurement point to the fault source. This represents the amplitude of the locally extracted dominant voltage transient component. This represents the amplitude of the locally extracted dominant transient current component. This represents a tiny positive number that prevents the denominator from being zero, and takes the value of... , This represents the impedance correction factor, used to compensate for calculation errors caused by inconsistencies in voltage and current attenuation characteristics. This factor is a preset dimensionless constant. The attenuation factor represents the dominant voltage transient component. The attenuation factor represents the decay factor of the dominant transient component of the current.

[0091] The impedance distance estimation unit will calculate the... The value is encapsulated in the election message, multicasted to logical neighbor nodes through the feature fingerprint interaction unit, and then passed to the consensus comparison unit.

[0092] The consensus comparison unit is used to collect and process election messages from logical neighbor nodes. An election list is built in memory. When the signal monitoring and recording unit determines that a disturbance event has occurred, the consensus comparison unit initiates a period of [duration to be specified]. The consensus comparison unit continuously listens to and parses the election messages sent by neighboring nodes within the reception time window, and extracts the neighboring node identifiers. and the corresponding impedance estimate Save to campaign list .

[0093] Consensus comparison unit during reception time window At the end, the minimum value filtering logic is triggered. The locally calculated minimum value is then used. With campaign list The impedance values ​​of all neighboring nodes recorded in the data are compared numerically to determine the node with the lowest impedance value. The filtering logic follows the formula below:

[0094] ;

[0095] In the formula, The operator represents the operation that takes the value of the function as the minimum. This represents a set that includes the local node itself. For nodes The corresponding estimated equivalent fault impedance value, The node is indexed. If multiple nodes have the same impedance value and are all the minimum during the comparison process, the consensus comparison unit compares the MAC address values ​​of these nodes and selects the node with the smallest MAC address value as the index. .

[0096] The role state management unit controls the operating mode of the edge computing gateway based on the consensus comparison results. The role state management unit maintains a finite state machine containing candidate states, anchor states, and subordinate states. Initially, during the fault triggering phase, the role state management unit defaults to the candidate state. When the consensus comparison unit outputs... When the unique identifier matches that of the local node, the role state management unit switches the state machine to the anchor state.

[0097] Nodes in anchor state execute server-side logic: they bind to and listen on a specified TCP port in the local operating system, and initialize a transient data aggregation service thread. Simultaneously, the node generates an anchor confirmation message, the data payload of which includes the anchor confirmation flag, the node's IP address, and the aforementioned TCP listening port number. The role state management unit broadcasts this anchor confirmation message to all neighboring nodes through the feature fingerprint interaction unit.

[0098] When the consensus comparison unit outputs When the unique identifier of the local node is inconsistent, the role state management unit switches the state machine to a subordinate state. The node in the subordinate state executes client logic: it enters a waiting-to-receive mode until it receives an anchor acknowledgment message. It then parses the message to extract the anchor node's IP address and port number, establishes a network socket connection with the anchor node, and uploads the full waveform data segment locked by the local signal monitoring and recording unit to the anchor node through this connection.

[0099] The time delay compensation module specifically includes a phase reference extraction unit, a dual-mode time marking unit, and a synchronization alignment calculation unit.

[0100] The phase reference extraction unit is deployed in each edge computing gateway and configured to track the phase changes of the local power grid voltage signal in real time. The phase reference extraction unit executes software phase-locked loop (SPLL) logic, using the fundamental component of the locally acquired bus voltage signal as input, and calculates and outputs a synchronized power frequency phase value in real time through an orthogonal signal generator and coordinate transformation. The phase reference extraction unit normalizes the output phase value and maps it to... The standard range. Due to the physical characteristic of network-wide frequency synchronization under steady-state operation, the phase reference extraction unit uses the power frequency voltage phase as a network-wide common phase reference independent of the communication network transmission path.

[0101] The dual-mode time-marking unit is configured to perform dual time and phase marking on the data transmitted from the transmitter. When a node in slave mode needs to upload a full waveform data segment to the anchor node, the dual-mode time-marking unit locks the sampling point index corresponding to the fault trigger moment. Simultaneously, the dual-mode time-marking unit reads the local NTP system time at that moment. and the instantaneous phase value output by the phase reference extraction unit. The dual-mode time-marking unit combines these two parameters to generate a time synchronization tag. The tag is then written into the header of the data message and sent along with the waveform data.

[0102] The synchronization alignment calculation unit is configured to perform phase-based fine-grained time alignment at the receiving end. Anchor nodes maintain a phase-time history circular buffer in memory, indexed by microsecond-level timestamps, cyclically recording the local system time for a preset time period in the past. With local phase The mapping relationship. When a data packet is received from a subordinate node, the synchronization alignment calculation unit parses out the time synchronization tag. Perform the following compensation steps to eliminate delay jitter errors caused by asymmetric routing.

[0103] The synchronization alignment calculation unit first calculates the phase deviation. This unit is based on the message... Retrieve the closest time point from the local phase-time history circular buffer and obtain the corresponding anchor point local phase value. And calculate its relationship with the message carried. The difference between them:

[0104] ;

[0105] In the formula, This indicates the phase deviation between the transmitting and receiving ends at the moment of rough synchronization. This represents the instantaneous phase value of the sender at the moment the fault was triggered, carried in the message. This indicates that the local time of the receiving end (anchor point) is... The instantaneous phase value recorded at time.

[0106] Synchronous alignment of computing units to calculate precise time offset compensation. To address the multiple solutions problem (i.e., periodic ambiguity) caused by phase periodicity, the synchronous alignment computing unit introduces a periodic cycle number in conjunction with the coarse time reference provided by NTP. The synchronization alignment calculation unit first solves for the integer number of cycles by minimizing the deviation between the coarse time difference and the phase estimation time difference. The calculation formula is:

[0107] ;

[0108] In the formula, This indicates the number of complete power frequency cycles that need to be compensated during phase alignment. This represents the rounding function. This represents a rough estimate of the time difference, calculated by subtracting the transmission time in the message from the receiver's current reception time. This includes network transmission latency and NTP time synchronization error. Indicates the fundamental frequency of the power grid. This represents the constant value of pi.

[0109] In determining After the value is calculated, the synchronization alignment calculation unit calculates the final absolute alignment time. The formula is:

[0110] ;

[0111] in, This indicates the high-precision fault occurrence time after phase correction. This indicates the NTP system time at the time the sender's fault was triggered, carried in the message. This indicates the phase deviation between the transmitting and receiving ends. This indicates the number of complete power frequency cycles that need to be compensated during phase alignment. This indicates the fundamental frequency of the power grid.

[0112] The synchronization alignment calculation unit uses the calculated The time axis of the received waveform data segments is recalibrated, thereby unifying the data of subordinate nodes and the local data of anchor nodes under the same high-precision phase reference system, achieving sub-cycle level waveform synchronization.

[0113] The fault location module specifically includes a data resampling interpolation unit, a transient wavefront calibration unit, and a ranging equation solving unit.

[0114] The data resampling interpolation unit is configured to eliminate discretization errors caused by clock asynchrony between the transmitter and receiver. Although the delay compensation module has corrected the absolute start time of the data packet... However, the sampling points of subordinate nodes may not fall on the sampling time grid of the anchor node. The data resampling interpolation unit constructs the target time series based on the sampling clock of the local anchor node. ,in For local sampling interval, Indicates the sequence index of the sampling points. Indicates the sampling interval of the local anchor node. This indicates the start time of the resampling sequence. The data resampling interpolation unit uses a cubic spline interpolation algorithm to process the waveform data of the subordinate nodes by solving a piecewise cubic polynomial function passing through all the original data points. Calculate the function in the target time series. The values ​​at the specified locations are used to generate a resampled waveform sequence that is strictly aligned with the local data on the time axis.

[0115] The transient wavefront calibration unit is configured to extract the precise arrival times of the fault traveling wave at both ends under a unified time reference. The transient wavefront calibration unit simultaneously receives the raw waveform data from the local anchor node and the resampled waveform data from the subordinate node output by the data resampling interpolation unit. The transient wavefront calibration unit processes the two signals using the Wavelet Transform Modulus Maximal Method (WTMM), selecting a Daubechies wavelet (e.g., db4) with compact support characteristics as the mother wavelet for multi-scale decomposition. The transient wavefront calibration unit extracts the modulus maxima points of the wavelet transform coefficients at each scale and connects these modulus maxima points at different scales to form a modulus maxima line. The transient wavefront calibration unit traces along the modulus maxima line towards the finest scale (i.e., the smallest scale), determining the convergence point of the modulus maxima line at the finest scale as the signal singularity point, i.e., the arrival time of the fault traveling wavefront. Through the above process, the transient wavefront calibration unit identifies the arrival times of the fault traveling wavefront at the anchor node. and the time of arrival at the subordinate node .

[0116] The ranging equation solving unit is configured to calculate the specific fault location based on the time difference of arrival of the traveling waves at both ends. The ranging equation solving unit reads pre-stored line topology parameters, including the total line length. and traveling wave propagation speed Among them, the propagation speed of traveling waves This is a preset value, determined by the inductance per unit length of the line. and capacitor Through formula The distance is calculated or obtained through periodic online signal injection calibration. The ranging equation solving unit substitutes the extracted time parameters into the two-end traveling wave ranging formula for calculation.

[0117] The fault distance calculation formula executed by the ranging equation solving unit is as follows:

[0118] ;

[0119] In the formula, This indicates the line length from the fault point to the anchor node. This represents the total length of the power line between the anchor node and its subordinate nodes. This indicates the propagation speed of a traveling wave signal in a power distribution line. This indicates the absolute moment when the faulty traveling wavefront reaches the anchor node. This indicates the absolute moment when the fault traveling wave front arrives at the slave node.

[0120] The distance measurement equation solving unit calculates... Then, execute the validity verification logic: if the calculation result satisfies... The ranging equation solving unit determines that the fault is located within the current monitoring section and then... The data is mapped to latitude and longitude coordinates and output. If the calculation result exceeds the above range, the distance measurement equation solving unit determines that the fault is located outside the area and does not generate fault location results within the area.

Claims

1. A distributed processing system for power communication data based on edge computing, characterized in that, include; An edge computing gateway deployed at various nodes of the power distribution network, wherein the edge computing gateway is configured with: The topology mapping module is used to parse the substation configuration description file and the real-time switch status obtained through process layer messages, construct a weighted adjacency matrix representing the physical connection relationship of the power grid, and dynamically generate a logical communication routing table between edge computing gateways based on the weighted adjacency matrix. The feature extraction module is used to start recording when the power signal exceeds the limit, extract transient feature vectors from the recorded data using signal processing algorithms, and send the transient feature vectors to neighboring nodes with electrical connection through the logical communication routing table. The anchor point election module is used to receive transient feature vectors from neighboring nodes, combine them with local measurement data to calculate the electrical impedance distance of the fault point relative to the current edge computing gateway, compare the electrical impedance distance of all nodes participating in the same event, and determine the computing power anchor node to undertake the main computing task according to the minimum impedance rule. The delay compensation module is used to start when the edge computing gateway is determined to be the computing power anchor node. It uses the phase of the local power grid frequency voltage signal as the common phase reference benchmark for the whole network, parses the phase stamp information in the received data packet, calculates the absolute delay value of the data packet in the transmission process of the dynamic routing network, and performs time sequence alignment on the waveform data from different nodes. The fault location module is used to perform differential synthesis on multi-source data after time alignment, construct fault location equations based on the traveling wave ranging principle, and calculate the specific location of the fault point.

2. The power communication data distributed processing system based on edge computing according to claim 1, characterized in that, When constructing the weighted adjacency matrix, the topology mapping module executes the following logic: For any two nodes, if there is a direct power line connection between them, the weight values ​​of the matrix elements are determined according to the switch status on the power line. When the switch is closed, the weight value is set to the static impedance parameter of the line, and when the switch is open, the weight value is set to infinity. When generating a logical communication routing table, the topology mapping module executes a path search algorithm to calculate the electrical distance between nodes. The electrical distance is defined as the sum of the weights of all edges on the shortest connected path between two nodes. Only when the electrical distance is less than a preset effective communication threshold, a routing entry with the adjacent node on the physical path as the next-hop gateway is established in the logical communication routing table.

3. The power communication data distributed processing system based on edge computing according to claim 1, characterized in that, The feature extraction module uses the Prony algorithm to fit discrete time series data into a linear combination model of complex exponential functions. By establishing a linear prediction model and solving the Vandermonde matrix equation, multiple transient components are separated. When extracting transient feature vectors, the feature extraction module performs the following filtering steps: First, it removes components whose frequencies are within the tolerance range of power frequency and its harmonic frequencies. Then, it calculates the energy values ​​of the remaining components, selects the component with the largest energy value as the dominant component, and encapsulates the parameters of the dominant component into the transient feature vector. The transient eigenvector includes the amplitude, attenuation factor, frequency, and phase parameters of the dominant component.

4. The power communication data distributed processing system based on edge computing according to claim 1, characterized in that, The anchor election module includes a consensus comparison unit, which is configured to start a receiving time window and collect election messages sent by neighboring nodes that contain electrical impedance distance estimates. At the end of the receiving time window, the consensus comparison unit compares the locally calculated electrical impedance distance with the electrical impedance distance of all neighboring nodes. If the locally calculated electrical impedance distance is the minimum value, the current edge computing gateway's role state is switched to anchor state, and an anchor confirmation message is broadcast to the neighboring nodes. If the locally calculated electrical impedance distance is not the minimum value, the current edge computing gateway will switch to a subordinate state and establish a connection with the winning computing power anchor node to upload the full waveform data segment.

5. A power communication data distributed processing system based on edge computing according to claim 4, characterized in that, The anchor point selection module uses an impedance estimation method based on attenuation characteristics correction when calculating electrical impedance distance. First, calculate the ratio of the amplitude of the locally extracted voltage transient dominant component to the amplitude of the current transient dominant component; Secondly, the difference between the attenuation factor of the voltage transient dominant component and the attenuation factor of the current transient dominant component is used to calculate the correction coefficient in combination with the preset impedance correction constant with time dimension. Finally, the ratio is weighted and adjusted using the correction coefficient to obtain the electrical impedance distance, in order to compensate for the error caused by the non-uniform attenuation of voltage and current signals.

6. The power communication data distributed processing system based on edge computing according to claim 1, characterized in that, The time delay compensation module includes a phase reference extraction unit and a synchronization alignment calculation unit; The phase reference extraction unit is configured to track the change in the phase of the local bus voltage in real time through a software phase-locked loop and normalize the phase value to a standard range. The computing power anchor node maintains a phase-time history circular buffer in memory. The synchronization alignment calculation unit retrieves the phase-time history circular buffer based on the sender NTP timestamp carried in the data packet, obtains the corresponding anchor local phase value, and calculates the phase deviation by the difference between the sender phase stamp carried in the packet and the anchor local phase value.

7. A power communication data distributed processing system based on edge computing according to claim 6, characterized in that, When performing time-series alignment of waveform data from different nodes, the synchronization alignment calculation unit performs the following steps: The coarse transmission delay is estimated by using the difference between the NTP timestamp at the transmitting end and the current system time at the receiving end, and then converted into a coarse number of cycles in the phase domain by combining the fundamental frequency of the power grid. The coarse number of cycles is corrected based on the phase deviation, and the integer number of complete power frequency cycles that need to be compensated is determined by minimizing the error. The phase deviation is converted into an accurate time offset by the integer number of complete power frequency cycles, and this time offset is superimposed on the transmitting end NTP timestamp to obtain an absolute alignment time unified to the receiving end clock domain.

8. The power communication data distributed processing system based on edge computing according to claim 1, characterized in that, The fault location module includes a data resampling interpolation unit. The data resampling interpolation unit is configured to construct a target time series based on the local sampling clock of the computing power anchor node, process waveform data from neighboring nodes using a cubic spline interpolation algorithm, calculate the value of the piecewise cubic polynomial function passing through the original data points, and generate a resampled waveform sequence that is strictly aligned with the local data of the computing power anchor node on the time axis.

9. A power communication data distributed processing system based on edge computing according to claim 8, characterized in that, The fault location module further includes a transient wavefront calibration unit. The transient wavefront calibration unit is configured to process the local waveform data and the resampled waveform sequence using the wavelet transform modulus maxima method, extract the modulus maxima points of the wavelet transform coefficients at each scale and form a modulus maxima line, trace along the modulus maxima line to the minimum scale to determine the convergence point of the modulus maxima line at the minimum scale, and mark the time corresponding to the convergence point as the arrival time of the fault traveling wave front.

10. A power communication data distributed processing system based on edge computing according to claim 9, characterized in that, When calculating the specific location of the fault point, the fault location module executes the following calculation logic: Obtain the total length of the power lines between the computing power anchor node and its neighboring nodes, as well as the propagation speed of the traveling wave signal in the power lines; Calculate the time difference between the time when the fault traveling wavefront arrives at the computing power anchor node and the time when it arrives at the neighbor node; The product of the time difference and the propagation speed is used as the distance difference term. The total length of the power line is calculated as half of the sum of the distance difference term, and the result is used as the line length from the fault point to the computing power anchor node.

Images