Power transmission line alarm method based on fault data flow dynamic analysis and lightweight compression

Abstract

The invention is suitable for the technical field of power transmission line alarm, and provides a power transmission line alarm method based on fault data flow dynamic analysis and lightweight compression, and the method comprises the steps: a terminal collects line state data in real time through a multi-mode sensor disposed on a power transmission line; the terminal carries out lightweight compression processing on the line state data, lossless compression is carried out on steady state data by using improved Huffman coding, and lossy compression is carried out on transient fault data by using combination of wavelet transform and sparse coding; the terminal uploads the compressed data to the cloud according to a preset priority scheduling rule; the cloud end automatically identifies a data format and decompresses the data format through a protocol non-inductive analysis engine, and reconstructs a transient waveform; performing space-time correlation analysis on the analyzed line data, meteorological data and a historical fault library, and calculating a fault confidence coefficient; and according to the fault diagnosis result, generating a grading alarm instruction, and transmitting the grading alarm instruction to the terminal for visualization, thereby improving the power grid fault response speed and the operation and maintenance intelligent level.

Description

Technical Field

[0001] The present invention relates to the technical field of transmission line warning, and particularly to a transmission line warning method based on dynamic analysis and lightweight compression of fault data flow. Background Art

[0002] With the continuous expansion of the scale of the power system and the improvement of the intelligent level, the real-time monitoring and fault warning of transmission lines have become the key links to ensure the safe and stable operation of the power grid. However, there are some significant problems in the existing systems in terms of fault warning. For example, some key warning data is transmitted mixed with conventional monitoring data, resulting in high-priority data being forced to queue during network congestion; due to the lack of real-time monitoring of network throughput or delay, it is impossible to trigger the suspension of low-priority data transmission or cache retransmission during congestion. And there is no full-link optimization for the characteristics of transmission line data, such as steady-state redundancy, transient burstiness, or spatio-temporal correlation, resulting in transmission delay, compression distortion, and inefficient cloud processing.

[0003] In view of this, a transmission line warning method based on dynamic analysis and lightweight compression of fault data flow is needed. Summary of the Invention

[0004] The embodiments of the present application provide a transmission line warning method based on dynamic analysis and lightweight compression of fault data flow, which is used to solve the problems of data transmission delay, compression distortion, and inefficient cloud processing.

[0005] The embodiments of the present application provide a transmission line warning method based on dynamic analysis and lightweight compression of fault data flow, including:

[0006] The terminal collects line status data in real time through multimodal sensors deployed on the transmission line, and the line status data includes current, voltage, temperature, vibration, and partial discharge signals;

[0007] The terminal performs lightweight compression processing on the line status data by using a hybrid compression framework, performs lossless compression on the steady-state data in the line status data by using an improved Huffman coding, and performs lossy compression on the transient fault data in the line status data by combining wavelet transform and sparse coding;

[0008] The terminal uploads the compressed data to the cloud according to a preset priority scheduling rule;

[0009] The cloud automatically identifies the data format and decompresses the received data stream through a protocol-insensitive parsing engine, and reconstructs the transient waveform by using an inverse wavelet transform accelerated by FPGA hardware;

[0010] Perform spatio-temporal correlation analysis on the parsed line data, meteorological data, and historical fault database, and calculate the fault confidence level through an air graph convolutional network model;

[0011] Generate hierarchical alarm instructions according to the fault diagnosis results, and transmit the hierarchical alarm instructions to the terminal for visualization.

[0012] Furthermore, the lossless compression of the steady-state data in the line state data using improved Huffman coding includes:

[0013] Based on the real-time collected steady-state data stream, statistically analyze the frequency distribution of current parameters and voltage parameters according to a preset time window;

[0014] Construct a Huffman tree according to the frequency distribution and generate a corresponding coding table;

[0015] When the statistical characteristics of the steady-state data change beyond a threshold, trigger the update of the coding table while retaining the historical coding table.

[0016] Furthermore, the lossless compression of the steady-state data in the line state data using improved Huffman coding also includes:

[0017] Divide the steady-state data into data segments according to a fixed duration, and independently generate a Huffman coding table for each segment of data;

[0018] Embed a segment identifier and a coding table index at the head of the compressed data stream to quickly match the coding table when performing segmented decompression.

[0019] Furthermore, the lossy compression of the transient fault data in the line state data using wavelet transform and sparse coding combination includes:

[0020] Select the Daubechies wavelet basis to perform multi-scale decomposition on the transient fault data;

[0021] Extract the high-frequency subband coefficients of the nth layer after decomposition, where n is the decomposition layer number and satisfies n≥3, and the high-frequency subband coefficients are the coefficient set corresponding to the highest frequency component after wavelet decomposition;

[0022] Calculate the energy value of the high-frequency subband coefficients. If the energy value of a single coefficient is lower than a preset energy threshold, set the energy value to zero to achieve preliminary dimensionality reduction.

[0023] Furthermore, the lossy compression of the transient fault data in the line state data using wavelet transform and sparse coding combination also includes:

[0024] Based on the transient fault waveform samples in the historical fault database, extract the high-frequency subband coefficients of each sample as the training set;

[0025] The K-SVD algorithm is used to perform dictionary training on the training set to generate an overcomplete dictionary, and the atoms of the overcomplete dictionary match the high-frequency subband coefficient features of the historical fault waveforms;

[0026] Orthogonal matching pursuit sparse coding is performed on the wavelet coefficients after preliminary dimensionality reduction to generate a sparse representation vector;

[0027] Orthogonal matching pursuit sparse coding is performed on the wavelet coefficients after preliminary dimensionality reduction to generate a sparse representation vector containing non-zero coefficients and their positions;

[0028] Quantize the non-zero coefficients in the sparse representation vector into 8-bit fixed-point numbers, and encode the positions of the sparse representation vector into a binary code stream to reduce the amount of compressed data.

[0029] Furthermore, the lossy compression of the transient fault data in the line state data using the combination of wavelet transform and sparse coding further includes:

[0030] Calculate the signal-to-noise ratio of the transient fault data in real time. If the signal-to-noise ratio is lower than the set threshold, adjust at least one of the following parameters according to a preset rule:

[0031] Increase the number of wavelet decomposition levels to the maximum of 8 levels;

[0032] Adjust the sparsity constraint parameter k, where k is the maximum number of non-zero coefficients in sparse coding;

[0033] Embed an optimization parameter flag at the head of the compressed data stream, and the optimization parameter flag includes the current number of wavelet decomposition levels, the sparsity constraint parameter k, and the signal-to-noise ratio threshold;

[0034] When decompressing, the cloud reads the optimization parameter flag and reconstructs the transient waveform based on the parameter values in the flag.

[0035] Furthermore, the terminal uploads the data after compression processing to the cloud according to a preset priority scheduling rule, including:

[0036] Divide the compressed data into first data and second data according to the data type and alarm criticality, and embed a binary priority label, where the priority of the first data is higher than that of the second data;

[0037] Allocate bandwidth by setting an emergency channel and a regular channel. The emergency channel is used to transmit the first data, and the regular channel is used to transmit the second data;

[0038] Monitor the network status in real time. If the delay of the first data exceeds the limit, suspend the transmission of the second data and enable a cache retransmission mechanism.

[0039] Furthermore, the cloud automatically identifies the data format of the received data stream through a protocol-insensitive parsing engine, decompresses it, and reconstructs the transient waveform using the inverse wavelet transform accelerated by FPGA hardware, including:

[0040] Extract the header identifier of the compressed data stream through the protocol-insensitive parsing engine, and match the decompression algorithm and data protocol based on the format marker in the identifier;

[0041] Use FPGA to perform the inverse wavelet transform. The wavelet basis parameters and sparse coding dictionaries corresponding to the compression stage are preset in the FPGA to accelerate the transient waveform reconstruction;

[0042] Fuse the reconstructed transient waveform with the steady-state data to generate a complete line state timing signal.

[0043] Furthermore, the spatio-temporal correlation analysis of the parsed line data with meteorological data and the historical fault library is performed, and the fault confidence is calculated through the spatio-graph convolutional network model, including:

[0044] Align the parsed line data with the meteorological data and historical fault records in the corresponding time window and geographical area in space and time;

[0045] Construct a graph structure of the transmission line based on the aligned data. The graph structure uses transmission nodes as vertices and line connection relationships as edges. The vertex features include real-time monitoring parameters, meteorological parameters, and historical fault frequencies;

[0046] Perform feature aggregation and inference on the graph structure through the spatio-graph convolutional network model, and output the fault confidence of each node.

[0047] Furthermore, the construction of the graph structure of the transmission line based on the aligned data. The graph structure uses transmission nodes as vertices and line connection relationships as edges. The vertex features include real-time monitoring parameters, meteorological parameters, and historical fault frequencies, including:

[0048] If there is a physical connection wire or they are in the same transmission section between two transmission nodes, then there is an edge between the two transmission nodes, and the weight of the edge is determined according to the electrical distance of the wire;

[0049] Input the vertex feature matrix, edge connection matrix, and weight matrix into the spatio-graph convolutional network model to generate a spatio-temporal correlation graph structure.

[0050] From the above technical solutions, it can be seen that the present invention has the following advantages:

[0051] The present invention improves the efficiency of power transmission data through priority scheduling and lightweight compression; utilizes the inverse wavelet transform accelerated by FPGA and the protocol-insensitive parsing engine to ensure the high-fidelity reconstruction of transient waveforms and improve the cloud processing efficiency; combines the fusion analysis of multi-dimensional spatio-temporal data by the empty graph convolutional network to improve the accuracy of calculating the fault confidence level, thereby reducing the false alarm rate; overall enhances the real-time response ability of the power grid to faults and the level of operation and maintenance intelligence, and ensures the safe and stable operation of the power transmission line. Brief Description of the Drawings

[0052] Figure 1 It is a schematic flowchart of an embodiment of a transmission line alarm method based on dynamic parsing and lightweight compression of fault data flow in the present invention. Detailed Embodiment

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] The transmission line alarm method based on dynamic parsing and lightweight compression of fault data flow in this embodiment is used to solve the problems of data transmission delay, compression distortion and low cloud processing efficiency. The implementation method in this embodiment can be implemented in the system, can be implemented on the server, or can be implemented on the terminal, and no specific limitation is made.

[0055] Embodiment 1

[0056] Please refer to Figure 1 , an embodiment of a transmission line alarm method based on dynamic parsing and lightweight compression of fault data flow in the present invention includes the following steps:

[0057] S11. The terminal collects line status data in real time through multi-modal sensors deployed on the transmission line. The line status data includes current, voltage, temperature, vibration and partial discharge signals;

[0058] In this embodiment, the terminal refers to intelligent data acquisition devices deployed along the transmission line or at key nodes, such as poles, insulator strings, and wire joints, which are built with a multi-modal sensor array and an edge computing module. The multi-modal sensors include current transformers (CTs), voltage transformers (PTs), fiber optic temperature sensors, three-axis accelerometers and high-frequency partial discharge (PD) sensors, which respectively collect the current, voltage, wire temperature, mechanical vibration and partial discharge signals of the line in real time. The terminal filters, normalizes and synchronizes the timestamps of the original signals to generate structured data packets. When the communication is interrupted, the built-in flash memory module of the terminal can cache data for at least 72 hours to ensure data integrity. By covering multi-dimensional states such as electricity, mechanics, and environment with multi-modal sensors, the limitations of a single data source are avoided.

[0059] S12. The terminal uses a hybrid compression framework to perform lightweight compression on the line status data. It performs lossless compression on the steady-state data in the line status data using improved Huffman coding, and performs lossy compression on the transient fault data in the line status data using the combination of wavelet transform and sparse coding;

[0060] Performing lossless compression on the steady-state data in the line status data using improved Huffman coding includes the following steps:

[0061] 1. Based on the real-time collected steady-state data stream, statistically analyze the frequency distribution of current parameters and voltage parameters according to a preset time window;

[0062] Divide the real-time collected steady-state current and voltage data streams into independent data segments at a fixed duration (e.g., every 5 minutes). Discretize the current and voltage values within each data segment, and statistically analyze the occurrence frequency of the data in each bin. If the mean deviation of the current or voltage within the data segment exceeds ±5% of the rated value, shorten the time window to 1 minute to capture rapid fluctuations.

[0063] 2. Construct a Huffman tree based on the frequency distribution and generate the corresponding coding table;

[0064] Assign short codewords to the high-frequency values in the frequency distribution, such as the current / voltage bins with an occurrence frequency > 10%, and assign long codewords to the low-frequency values to generate an optimal prefix coding tree. Convert the Huffman tree into a coding table, with the storage format being key-value pairs, and associate it with the time window identifier.

[0065] 3. When the statistical characteristics of the steady-state data change by more than the threshold, trigger the update of the coding table, and at the same time retain the historical coding table.

[0066] Calculate the KL divergence of the frequency distributions of adjacent time windows. If the KL divergence > 0.1, it is determined that the statistical characteristics have changed significantly. Reconstruct the Huffman tree based on the frequency distribution of the new time window, generate a new coding table, and increment the version number. The old coding table is saved to the local cache or cloud database, and the retention duration is ≥ 30 days.

[0067] In addition, it also includes the following steps:

[0068] 1. Divide the steady-state data into data segments at a fixed duration, and independently generate a Huffman coding table for each segment of data;

[0069] Divide the steady-state data stream at a fixed duration. Each segment contains N sampling points. Statistically analyze the frequency distribution for each segment of data separately and construct a Huffman tree. The coding table is bound to the data segment, which can solve the problem of the attenuation of compression efficiency caused by the gradual change of the distribution of long-term steady-state data.

[0070] 2. Embed the segmentation identifier and the coding table index in the header of the compressed data stream to quickly match the coding table during segmented decompression.

[0071] Reserve a 128-byte field in the header of the compressed data stream, and write the segmentation identifier and the coding table version number. The index storage path can point to the local cache or the cloud database. This step can achieve quick matching of the coding table during decompression and avoid global search.

[0072] Use the combination of wavelet transform and sparse coding for lossy compression of the transient fault data in the line status data, including the following steps:

[0073] 1. Select the Daubechies wavelet basis to perform multi-scale decomposition on the transient fault data;

[0074] 2. Extract the high-frequency subband coefficients of the nth layer after decomposition, where n is the decomposition layer number and satisfies n≥3. The high-frequency subband coefficients are the coefficient set corresponding to the highest frequency component after wavelet decomposition;

[0075] 3. Calculate the energy value of the high-frequency subband coefficients. If the energy value of a single coefficient is lower than the preset energy threshold, set the energy value to zero to achieve preliminary dimensionality reduction.

[0076] Specifically, use the Daubechies wavelet basis as the decomposition tool to perform 5-layer wavelet decomposition on the partial discharge signal in the transient fault data to ensure effective separation of the low-frequency baseline and the high-frequency details. Multi-scale decomposition stratifies the signal by frequency and can retain high-frequency details, such as key fault features of pulses and oscillations. Extract the high-frequency subband coefficients from the 3rd layer and above of the wavelet decomposition, corresponding to the highest frequency component in the signal, such as partial discharge pulses >10MHz. Here, the preset energy threshold is 10% of the average energy of all coefficients in this layer, that is, if the energy of a single coefficient is lower than this threshold, it is determined as noise and set to zero.

[0077] Using the combination of wavelet transform and sparse coding for lossy compression of the transient fault data in the line status data also includes the following steps:

[0078] 1. Based on the transient fault waveform samples in the historical fault database, extract the high-frequency subband coefficients of each sample as the training set;

[0079] 2. Use the K-SVD algorithm to perform dictionary training on the training set to generate an overcomplete dictionary, and the atoms of the overcomplete dictionary match the high-frequency subband coefficient features of the historical fault waveforms;

[0080] 3. Perform orthogonal matching pursuit sparse coding on the wavelet coefficients after preliminary dimensionality reduction to generate a sparse representation vector;

[0081] 4. Perform orthogonal matching pursuit sparse coding on the wavelet coefficients after preliminary dimensionality reduction to generate a sparse representation vector containing non-zero coefficients and their positions;

[0082] 5. Quantize the non-zero coefficients in the sparse representation vector into 8-bit fixed-point numbers, and encode the positions of the sparse representation vector into a binary code stream to reduce the amount of compressed data.

[0083] Specifically, extract transient waveform samples from the historical fault database, such as lightning strike and short-circuit waveforms, perform 5-layer wavelet decomposition on them, and extract the 5th-layer high-frequency coefficients as the training set. Set the number of dictionary atoms to 128 and the sparsity to 8, and iteratively train to generate an over-complete dictionary so that each atom matches a typical high-frequency fault feature. Perform the OMP algorithm on the wavelet coefficients after dimensionality reduction, select 8 atoms from the dictionary (sparsity k = 8), and generate a sparse vector containing non-zero coefficient values and their positions. Quantize the coefficient values into 8-bit fixed-point numbers (range -128 to 127, precision 0.01) to reduce the storage space; encode the atom indices into a 7-bit binary code stream. Reduce the data volume to 15% - 20% of the original size through sparse representation while retaining the key features of the waveform.

[0084] For lossy compression of transient fault data in line state data using the combination of wavelet transform and sparse coding, the following steps are also included:

[0085] 1. Calculate the signal-to-noise ratio of the transient fault data in real time. If the signal-to-noise ratio is lower than the set threshold, adjust at least one of the following parameters according to the preset rules:

[0086] 2. Increase the number of wavelet decomposition layers to the maximum of 8 layers;

[0087] 3. Adjust the sparsity constraint parameter k, where k is the maximum number of non-zero coefficients in sparse coding;

[0088] 4. Embed an optimization parameter marker at the head of the compressed data stream. The optimization parameter marker includes the current number of wavelet decomposition layers, the sparsity constraint parameter k, and the signal-to-noise ratio threshold;

[0089] 5. The cloud reads the optimization parameter marker during decompression and reconstructs the transient waveform based on the parameter values in the marker.

[0090] Specifically, the signal-to-noise ratio of the current transient data is calculated in real time. If the signal-to-noise ratio is less than 20 dB, it is determined as a low-quality signal. The wavelet decomposition level is increased from 5 levels to 8 levels to enhance the high-frequency detail extraction ability; k is increased from 8 to 12, allowing more atoms to participate in the sparse representation and improving the signal fidelity. A 32-byte field is embedded in the head of the compressed data stream, including the current decomposition level (3-bit binary), the k value (4-bit binary), and the signal-to-noise ratio threshold (8-bit floating point number). When decompressing at the cloud, the marked parameters are read, the corresponding wavelet basis and dictionary are called, and the inverse wavelet transform and sparse reconstruction are performed according to the marked values. This embodiment adapts to complex environments by dynamically adjusting the decomposition level and sparsity parameters, ensuring high-fidelity reconstruction in low signal-to-noise ratio scenarios.

[0091] S13. The terminal uploads the compressed data to the cloud according to the preset priority scheduling rules;

[0092] Step S13 also includes the following:

[0093] 1. The compressed data is divided into first data and second data according to the data type and alarm criticality, and a binary priority label is embedded. The priority of the first data is higher than that of the second data;

[0094] 2. Bandwidth is allocated by setting up an emergency channel and a regular channel. The emergency channel is used to transmit the first data, and the regular channel is used to transmit the second data;

[0095] 3. The network status is monitored in real time. If the delay of the first data exceeds the limit, the transmission of the second data is suspended and a cache retransmission mechanism is enabled.

[0096] Specifically, the first data is high-priority data, including transient fault data, such as partial discharge signals, lightning strike waveforms, alarm data with a signal-to-noise ratio lower than the threshold, and abnormal data with current and voltage sudden changes exceeding ±10% of the rated value. The second data is low-priority data, including periodic monitoring data such as steady-state current, voltage, and temperature, and normal data with a signal-to-noise ratio higher than the threshold. Two bits of binary code are reserved in the head field of the compressed data stream. The coding rule is 01 for the first data (emergency alarm) and 10 for the second data (regular monitoring). The label is bound to the data segment, and the label value is preferentially extracted during cloud parsing.

[0097] The emergency channel exclusively occupies 70% of the total bandwidth, with a minimum guaranteed bandwidth of 50 Mbps, and transmits the first data in real time; the regular channel occupies the remaining 30% of the bandwidth, with a maximum rate limit of 20 Mbps, and transmits the second data. The throughput of the emergency channel is monitored in real time. If its actual usage rate is less than 50%, the redundant bandwidth is released to the regular channel; if the total network bandwidth fluctuates (such as from 100 Mbps to 60 Mbps), the minimum guaranteed bandwidth of the emergency channel is synchronously adjusted to 60 Mbps × 70% = 42 Mbps.

[0098] Calculate the end-to-end transmission delay of the first data in real time, from the terminal sending to the cloud receiving. If the delay exceeds 50 ms continuously for 3 times, it is determined as network congestion. The trigger condition is that the delay threshold = 50 ms and the detection window = 1 second. Immediately pause all data transmissions on the regular channel, release the bandwidth resources to the emergency channel, and the terminal enables the local flash memory to cache the second data, and the cache capacity supports at least 24 hours of data storage. After the network is restored, the cached second data is retransmitted in the order of timestamps, and the retransmission rate is limited to 50% of the maximum rate of the regular channel to avoid secondary congestion.

[0099] S14. The cloud automatically identifies the data format and decompresses the received data stream through the protocol-insensitive parsing engine, and uses the inverse wavelet transform accelerated by FPGA hardware to reconstruct the transient waveform;

[0100] Step S14 includes the following steps:

[0101] 1. Extract the header identifier of the compressed data stream through the protocol-insensitive parsing engine, and match the decompression algorithm and data protocol based on the format marker in the identifier;

[0102] 2. Use FPGA to perform the inverse wavelet transform, and the wavelet basis parameters and sparse coding dictionaries corresponding to the compression stage are preset in the FPGA to accelerate the transient waveform reconstruction;

[0103] 3. Fuse the reconstructed transient waveform with the steady-state data to generate a complete line state timing signal.

[0104] Specifically, the cloud reads the header identifier of the compressed data stream through the protocol-insensitive parsing engine, automatically matches the corresponding decompression algorithm according to the identifier, and calls the preset data protocol template to parse the data format; the FPGA loads the wavelet basis parameters consistent with the compression stage and the sparse coding dictionary generated by historical training, and performs the inverse wavelet transform through parallel pipelining calculation to reconstruct the transient waveform, such as the partial discharge pulse waveform; align the reconstructed waveform with the decompressed steady-state data according to the millisecond-level timestamp, and generate a complete timing signal, the synchronization sequence of current, voltage, and waveform details, through interpolation fusion for subsequent fault confidence analysis by the empty graph convolutional network.

[0105] S15. Perform spatio-temporal correlation analysis on the parsed line data, meteorological data, and historical fault library, and calculate the fault confidence through the empty graph convolutional network model;

[0106] Step S15 includes the following:

[0107] 1. Align the parsed line data with the meteorological data and historical fault records in the corresponding time window and geographical area in space and time;

[0108] 2. Construct a graph structure of the transmission line based on the aligned data. The graph structure uses transmission nodes as vertices and line connection relationships as edges. The vertex features include real-time monitoring parameters, meteorological parameters, and historical fault frequencies.

[0109] This step also includes:

[0110] If there is a physical connection wire between two transmission nodes or they are in the same transmission section, then there is an edge between the two transmission nodes, and the weight of the edge is determined according to the electrical distance of the wire.

[0111] Input the vertex feature matrix, edge connection matrix, and weight matrix into the empty graph convolutional network model to generate a spatio-temporal correlation graph structure.

[0112] 3. Through the empty graph convolutional network model, perform feature aggregation and inference on the graph structure, and output the fault confidence of each node.

[0113] Specifically, accurately match the parsed line data with the meteorological data and historical fault records corresponding to the time and geographical area to ensure that the data is aligned in the same time and space dimensions. For example, match the local meteorological information according to the GPS coordinates of the transmission node and associate the historical fault cases in the same area. Then, construct a graph structure of the transmission line. Each transmission node is used as a vertex in the graph, and the vertex features include parameters such as current and voltage monitored in real time, meteorological data such as current wind speed and rainfall, and the number of faults of this node in the past year. The edges between vertices represent physical connections or associations in the same transmission section. Finally, through the empty graph convolutional network model, perform multi-layer feature aggregation on the graph structure, and output the fault probability of each node after synthesizing the neighborhood information. When the probability value exceeds the set threshold, an alarm is triggered.

[0114] When constructing the graph structure, determine whether there is a physical wire connection between two transmission nodes or whether they belong to the same transmission section to determine whether to establish an edge. The weight of the edge is calculated according to the physical distance and electrical impedance between the nodes. The closer the distance and the smaller the impedance, the higher the weight, reflecting a stronger electrical coupling relationship. Input the matrix containing all vertex features, the edge connection relationship matrix, and the weight matrix into the empty graph convolutional network model. The model analyzes the vertex features and the weights of the edges, learns the power grid topology and fault propagation rules, and finally generates a graph structure that fuses spatio-temporal correlations, providing a basis for accurately calculating the fault confidence.

[0115] S16. Generate a hierarchical alarm instruction according to the fault diagnosis result, and transmit the hierarchical alarm instruction to the terminal for visualization.

[0116] In this embodiment, according to the fault confidence level output by the empty graph convolutional network model, for example, a probability value ≥ 0.7 is a high-level alarm, 0.5 - 0.7 is a medium-level alarm, and < 0.5 is a low-level alarm. Combining with the fault type, which includes short circuit, partial discharge, and conductor galloping, and the influence range, such as single-node or multi-node linkage, a hierarchical alarm instruction is generated. The instruction content includes the alarm level, fault location, confidence value, timestamp, and recommended handling measures. The alarm instruction is pushed to the terminal in real time through the MQTT protocol. The terminal marks the fault points in different colors on the visualization interface, red - high level, orange - medium level, yellow - low level, and overlays meteorological data in the map view. At the same time, it triggers an audible and visual alarm and automatically dispatches work orders to ensure that the operation and maintenance personnel respond immediately.

[0117] The above embodiment reduces data transmission delay and bandwidth occupancy through dynamic priority scheduling and lightweight compression; combines FPGA acceleration reconstruction and the empty graph convolutional network to improve the accuracy of fault confidence, and at the same time integrates multi-dimensional spatio-temporal data to form a full-link closed loop from data acquisition to intelligent alarm, improving the power grid fault response speed and the level of operation and maintenance intelligence.

[0118] Those of ordinary skill in the art can realize that the units of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0119] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc. In addition, the functional units in each embodiment of the present invention can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0120] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0121] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A transmission line warning method based on dynamic analysis and lightweight compression of fault data flow, characterized in that Including: The terminal collects line status data in real time through multimodal sensors deployed on the transmission line. The line status data includes current, voltage, temperature, vibration, and partial discharge signals. The terminal uses a hybrid compression framework to perform lightweight compression processing on the line status data. It performs lossless compression on the steady-state data in the line status data using an improved Huffman coding, and performs lossy compression on the transient fault data in the line status data using a combination of wavelet transform and sparse coding. The terminal uploads the compressed data to the cloud according to a preset priority scheduling rule. The cloud automatically identifies the data format and decompresses the received data stream through a protocol-insensitive parsing engine, and reconstructs the transient waveform using wavelet inverse transform accelerated by FPGA hardware. Perform spatio-temporal correlation analysis on the parsed line data, meteorological data, and historical fault database, and calculate the fault confidence through an air graph convolutional network model. Generate a hierarchical alarm instruction according to the fault diagnosis result, and transmit the hierarchical alarm instruction to the terminal for visualization.

2. The transmission line warning method based on dynamic parsing and lightweight compression of fault data according to claim 1, wherein The lossless compression of the steady-state data in the line status data using the improved Huffman coding includes: Based on the real-time collected steady-state data stream, statistically analyze the frequency distribution of current parameters and voltage parameters in a preset time window. Construct a Huffman tree according to the frequency distribution and generate a corresponding coding table. When the statistical characteristics of the steady-state data change beyond the threshold, trigger the update of the coding table, and at the same time retain the historical coding table.

3. The transmission line warning method based on dynamic analysis and lightweight compression of fault data flow according to claim 1 or 2, characterized in that, The lossless compression of the steady-state data in the line status data using the improved Huffman coding also includes: Divide the steady-state data into data segments according to a fixed duration, and independently generate a Huffman coding table for each segment of data. Embed a segment identifier and a coding table index at the head of the compressed data stream to quickly match the coding table when performing segmented decompression.

4. The transmission line warning method based on dynamic parsing and lightweight compression of fault data according to claim 1, characterized in that The lossy compression of the transient fault data in the line status data using a combination of wavelet transform and sparse coding includes: Select the Daubechies wavelet basis to perform multi-scale decomposition on the transient fault data. Extract the high-frequency subband coefficients of the nth layer after decomposition, where n is the decomposition layer number and satisfies n≥3. The high-frequency subband coefficients are a set of coefficients corresponding to the highest frequency component after wavelet decomposition. Calculate the energy value of the high-frequency subband coefficients. If the energy value of a single coefficient is lower than the preset energy threshold, set the energy value to zero to achieve preliminary dimensionality reduction.

5. The transmission line warning method based on dynamic analysis and lightweight compression of fault data according to claim 4, characterized in that The lossy compression of the transient fault data in the line status data using a combination of wavelet transform and sparse coding also includes: Based on the transient fault waveform samples in the historical fault database, extract the high-frequency subband coefficients of each sample as the training set. Use the K-SVD algorithm to perform dictionary training on the training set to generate an overcomplete dictionary. The atoms of the overcomplete dictionary match the high-frequency subband coefficient characteristics of the historical fault waveform. Perform orthogonal matching pursuit sparse coding on the preliminarily dimensionally reduced wavelet coefficients to generate a sparse representation vector. Perform orthogonal matching pursuit sparse coding on the preliminarily dimensionally reduced wavelet coefficients to generate a sparse representation vector including non-zero coefficients and positions. Quantize the non-zero coefficients in the sparse representation vector into 8-bit fixed-point numbers, and encode the positions of the sparse representation vector into a binary code stream to reduce the amount of compressed data.

6. The transmission line warning method based on dynamic parsing and lightweight compression of fault data according to claim 5, characterized in that The lossy compression of the transient fault data in the line state data using the combination of wavelet transform and sparse coding further includes: Calculating the signal-to-noise ratio of the transient fault data in real time. If the signal-to-noise ratio is lower than the set threshold, adjust at least one of the following parameters according to a preset rule: Increase the number of wavelet decomposition layers to the maximum of 8 layers; Adjust the sparsity constraint parameter k, where k is the maximum number of non-zero coefficients in sparse coding; Embed an optimization parameter flag at the head of the compressed data stream, and the optimization parameter flag includes the current number of wavelet decomposition layers, the sparsity constraint parameter k, and the signal-to-noise ratio threshold; When decompressing, the cloud reads the optimization parameter flag and reconstructs the transient waveform based on the parameter values in the flag.

7. The transmission line warning method based on dynamic parsing and lightweight compression of fault data according to claim 1, characterized in that The terminal uploads the compressed data to the cloud according to a preset priority scheduling rule, including: Dividing the compressed data into first data and second data according to the data type and alarm criticality, and embedding a binary priority label, where the priority of the first data is higher than that of the second data; Allocate bandwidth by setting up an emergency channel and a regular channel. The emergency channel is used to transmit the first data, and the regular channel is used to transmit the second data; Monitor the network status in real time. If the transmission of the first data is delayed beyond the limit, suspend the transmission of the second data and enable the cache retransmission mechanism.

8. The transmission line warning method based on dynamic analysis and lightweight compression of fault data flow according to claim 1, characterized in that, The cloud automatically identifies the data format and decompresses the received data stream through a protocol-insensitive parsing engine, and reconstructs the transient waveform using the inverse wavelet transform accelerated by FPGA hardware, including: Extract the header identifier of the compressed data stream through a protocol-insensitive parsing engine, and match the decompression algorithm and data protocol based on the format flag in the identifier; Use FPGA to perform the inverse wavelet transform. The wavelet basis parameters and sparse coding dictionary corresponding to the compression stage are preset in the FPGA to accelerate the reconstruction of the transient waveform; Fuse the reconstructed transient waveform with the steady-state data to generate a complete line state time series signal.

9. The transmission line warning method based on dynamic parsing and lightweight compression of fault data according to claim 1, characterized in that The spatio-temporal correlation analysis of the parsed line data with meteorological data and historical fault libraries, and calculating the fault confidence through an empty graph convolutional network model, including: Perform spatio-temporal alignment of the parsed line data with meteorological data and historical fault records in the corresponding time window and geographical area; Construct a graph structure of the transmission line based on the aligned data. The graph structure has transmission nodes as vertices and line connection relationships as edges, and the vertex features include real-time monitoring parameters, meteorological parameters, and historical fault frequencies; Perform feature aggregation and inference on the graph structure through an empty graph convolutional network model, and output the fault confidence of each node.

10. The transmission line warning method based on dynamic analysis and lightweight compression of fault data according to claim 9, characterized in that, Construct a graph structure of the transmission line based on the aligned data. The graph structure has transmission nodes as vertices and line connection relationships as edges, and the vertex features include real-time monitoring parameters, meteorological parameters, and historical fault frequencies, including: If there is a physical connection wire or they are in the same transmission section between two transmission nodes, then there is an edge between the two transmission nodes, and the weight of the edge is determined according to the electrical distance of the wire; Input the vertex feature matrix, edge connection matrix, and weight matrix into the empty graph convolutional network model to generate a spatio-temporal correlation graph structure.

Images