Intelligent power distribution network fault positioning method and system based on HPLC (High Performance Liquid Chromatography) multi-mode communication

Abstract

The invention relates to an intelligent power distribution network fault positioning method and system based on HPLC (High Performance Liquid Chromatography) multi-modal communication, and relates to the technical field of power system automation, and the method comprises the following steps: S1, constructing a multi-modal communication link through an HPLC communication network and at least two auxiliary communication networks, and dynamically selecting an optimal communication channel; s2, collecting multi-source fault data of the power distribution network monitoring terminal in real time through the selected communication channel, and performing encryption transmission; s3, performing space-time fusion processing on the transmitted fault data to generate a space-time correlation feature vector; s4, based on the space-time correlation feature vector, extracting fault space-time features through a deep learning model; and S5, in combination with the power grid topological structure and the fault spatio-temporal characteristics, performing fault positioning decision. Through a multi-modal communication link dynamic preferential mechanism and a space-time fusion processing model, the problem of time delay sensitivity of traditional single-channel transmission is effectively solved, and stable fault feature extraction capability can be maintained under the conditions of strong noise interference and partial data missing.

Description

Technical Field

[0001] The present application relates to the field of power system automation technology, and in particular to a distribution network fault intelligent location method and system based on HPLC multimodal communication. Background Art

[0002] In the field of distribution network fault location technology, existing methods face multiple challenges in complex operating environments. Traditional communication architectures often rely on a single transmission channel, which can easily lead to packet loss and timing disruption in areas with strong electromagnetic interference or densely populated nodes, compromising the integrity of fault signature information. Existing data fusion algorithms inadequately address the spatiotemporal alignment of multi-source heterogeneous data, making it difficult to effectively coordinate monitoring data with varying sampling rates and transmission delays, impacting the accuracy of subsequent signature analysis.

[0003] Current fault identification models primarily rely on steady-state electrical quantity thresholds, which inherently limit their ability to capture transient features. The weak signals generated by high-resistance ground faults are easily overwhelmed by background noise, while the non-stationary nature of intermittent arc faults renders traditional frequency-domain analysis methods ineffective. Existing feature extraction methods fail to fully incorporate grid topology information and have limited utilization of spatial correlation features, hindering the ability to identify complex fault patterns.

[0004] In terms of system security and reliability, existing technologies lack comprehensive protection mechanisms from data collection to decision-making. Monitoring data is subject to tampering during transmission, while centralized processing architectures present single points of failure. Traditional fault-tolerance strategies, often based on simple majority voting, struggle to cope with complex anomalies like coordinated attacks. They also lack trusted audit and traceability capabilities, making them unable to meet the stringent requirements for secure and reliable operations and maintenance in smart grids. Summary of the Invention

[0005] The purpose of this application is to provide a distribution network fault intelligent positioning method and system based on HPLC multimodal communication, which solves the technical problems of existing distribution network fault positioning technology in complex operating environments, such as low communication reliability, insufficient transient feature extraction, and weak safety protection mechanism, resulting in insufficient positioning accuracy and poor anti-interference ability.

[0006] In a first aspect, the present application provides a method for intelligently locating distribution network faults based on HPLC multimodal communication, comprising the following steps: S1. Construct a multimodal communication link through the HPLC communication network and at least two auxiliary communication networks, and dynamically select the optimal communication channel; S2. Collect multi-source fault data from the distribution network monitoring terminal in real time through the selected communication channel and transmit it in encrypted form; S3. Perform spatiotemporal fusion processing on the transmitted fault data to generate a spatiotemporal correlation feature vector; S4. Extract fault spatiotemporal features through deep learning model based on spatiotemporal correlation feature vectors; S5. Combine the grid topology and the temporal and spatial characteristics of the fault to make a fault location decision: Preferably, the algorithm for dynamically selecting the optimal communication channel in S1 satisfies: Among them, P i is the power consumption of the ith channel, R i is the transmission rate of the i-th channel, x i ∈{0,1} is the channel selection state, R th is the minimum rate threshold, N is the total number of available channels; The auxiliary communication network includes a combination of 5G slicing network, 230MHz wireless private network and low-orbit satellite communication, and satellite communication is automatically enabled when the main channel signal strength is lower than -110dBm.

[0007] Preferably, the spatiotemporal fusion processing in S3 includes: Using the improved Kalman filter algorithm, the gain coefficient calculation satisfies: in: H k is the Kalman gain matrix at the kth moment, P k|k-1 is the state prediction covariance matrix, H k is the observation matrix, R k is the observation noise covariance matrix, which is defined as: The delay variance of the i-th communication channel.

[0008] Preferably, the fault location decision in S5 includes: The fault confidence function is defined as: in: C i is the fault confidence of node i, I 0,i is the zero-sequence current amplitude of node i, U i is the voltage at node i, is the voltage change rate, f NN is the neural network feature extraction function, X i : The multidimensional feature vector of node i.

[0009] Preferably, the key generation for the encrypted transmission satisfies: Among them: K dev Unique identification key for the device, PUFresp is the response value of the physical unclonable function, Challenge is the random challenge code input from the outside, is the XOR operation, and Hash is the cryptographic hash function.

[0010] Preferably, the deep learning model includes: a temporal convolutional network layer with a dilation factor of 2 l The hollow convolution structure is constructed, where l is the number of network layers; Graph Attention Network Layer: It aggregates the features of neighboring nodes through a multi-head attention mechanism, with the number of attention heads M ≥ 3; Feature Fusion Layer: It uses a gating mechanism to dynamically fuse spatiotemporal features.

[0011] Preferably, the multi-source fault data includes: Electrical quantity data: zero-sequence current, line voltage, phase angle; Environmental data: temperature, humidity, electromagnetic intensity; Communication quality data: bit error rate, signal strength, delay jitter.

[0012] Preferably, the method further includes a step of verifying the credibility of the fault location result: Verify decision consistency through blockchain smart contracts; When the regional confidence difference exceeds 15%, the manual review mechanism is triggered; Generates an immutable audit log with timestamp and location parameters.

[0013] Preferably, the spatiotemporal correlation feature vector includes: Time dimension characteristics: current mutation rate, voltage fluctuation period, and transient process duration; Spatial topological characteristics: node electrical distance, branch impedance correlation, regional power balance In a second aspect, the present application provides a distribution network fault intelligent location system based on HPLC multimodal communication, comprising: A communication link establishment module is used to build a multimodal communication link through the HPLC communication network and the auxiliary communication network, and dynamically select the optimal communication channel; Data acquisition and transmission module, used to collect multi-source fault data of distribution network in real time and transmit it to data processing module through encrypted channel; The time-space fusion processing module is used to perform time alignment compensation and spatial topology association on the received fault data to generate a time-space fusion feature vector; The deep feature extraction module is used to extract the spatiotemporal characteristics of faults through a hybrid model of temporal convolutional networks and graph neural networks. The intelligent decision-making module is used to combine the power grid topology and deep features to output fault location results and confidence assessments. The security protection module is used to implement device authentication and dynamic data encryption based on physically unclonable functions.

[0014] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention effectively overcomes the delay sensitivity of traditional single-channel transmission through a dynamic optimization mechanism for multimodal communication links and a spatiotemporal fusion processing model. It can maintain stable fault feature extraction capabilities under strong noise interference and partial data loss, thus resolving the existing technology's excessive dependence on communication quality. 2. This paper uses a multi-scale feature representation system built on a deep spatiotemporal feature extraction network. By leveraging the synergy of a graph attention mechanism and a temporal convolutional network, it significantly improves the ability to distinguish complex faults such as high-resistance grounding and intermittent arcing, overcoming the limitations of traditional methods that rely too heavily on steady-state features. 3. This invention uses dynamic quantization processing and a layered encryption mechanism to achieve intelligent adaptation of computing resources and communication bandwidth while ensuring positioning accuracy. This solves the deployment challenge in scenarios where edge computing nodes have limited resources and offers superior engineering applicability compared to traditional centralized processing models. 4. The distributed audit log and dual fault-tolerant verification mechanism based on blockchain technology have built a complete operation traceability chain and exception handling system, effectively preventing the risks of malicious data injection and equipment malfunction, and meeting the strict requirements of smart grid for safety and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the application method; Figure 2 It is the system architecture of this application; DETAILED DESCRIPTION

[0016] The following is combined with Figure 1-2 , further details of this application are given.

[0017] The intelligent location method for distribution network faults based on HPLC multimodal communication is characterized by comprising the following steps: S1, constructing a multimodal communication link through the HPLC communication network and at least two auxiliary communication networks, and dynamically selecting the optimal communication channel; The construction and optimization process of the multimodal communication link in the present invention is specifically implemented as follows: The communication network architecture utilizes an HPLC power line carrier communication network as the backbone, combined with 5G slicing networks, 230MHz wireless private networks, and low-orbit satellite communications to form a heterogeneous communication matrix. Each communication terminal device is equipped with a multi-mode baseband processing chipset that supports dynamic protocol stack loading and channel state awareness, enabling autonomous adaptation of communication modes.

[0018] The channel quality assessment model is built based on a multi-dimensional parameter fusion mechanism, which includes the joint evaluation of bit error rate parameters, signal strength parameters, and delay parameters. By establishing a normalized processing function: Among them, Q i : Comprehensive quality index of the i-th channel (dimensionless) BER i : Bit error rate of the i-th channel (value range 0 to 1) RSSI i : Received signal strength indicator of the i-th channel (unit: dBm) RSSI max : The system's maximum acceptable signal strength threshold (in dBm) Latency i : End-to-end transmission delay of the i-th channel (unit: ms) α, β, γ: weight coefficients (satisfying α + β + γ = 1) λ: Delay attenuation factor (preferred value is 0.05-0.2).

[0019] The dynamic channel selection algorithm is implemented using constrained optimization theory to construct the objective function: in: P i : Power consumption of the i-th channel (unit: W) R i : The effective transmission rate of the i-th channel (in Mbps) x i : Channel activation state variable (x i ∈{0,1}, 0 means disabled, 1 means enabled) R th : System minimum transmission rate requirement threshold (unit: Mbps) N: Total number of available channels.

[0020] The optimization problem is solved using a branch-and-bound algorithm, minimizing energy consumption while meeting the minimum communication quality threshold. The communication channel switching mechanism sets dual trigger conditions: when the RSSI of the primary channel remains below -110dBm for 500ms, or the end-to-end delay exceeds 200ms, the satellite communication emergency channel is automatically activated. A dual-transmission mechanism is used during the switching process to ensure the integrity of critical data packets through redundant transmission. The communication security authentication mechanism is based on physically unclonable function (PUF) technology, and the terminal device has a built-in PUF chip to generate a hardware fingerprint: Where: ID dev: Unique device identification code (128-bit hash value) PUF resp : Physical unclonable function response value (256-bit binary sequence) Challenge: The challenge code (128-bit random number) issued by the authentication server Bitwise XOR operator Hash: cryptographic hash function (preferably SM3 algorithm) This mechanism effectively prevents device counterfeiting attacks and ensures the entity credibility of the communication link. The channel status monitoring module uses a sliding window method to perform quality parameter statistics, and the window length is preferably 30 sampling cycles. Dynamic variance is calculated for the delay parameter: in: The delay variance of the i-th channel (unit: ms) 2 ) The delay value of the kth sampling of the i-th channel (unit: ms) Mean delay of the i-th channel (in ms) N: Sliding window length (preferred value 20 to 50) This variance value is used to dynamically adjust the noise covariance matrix of the Kalman filter algorithm to improve data fusion accuracy.

[0021] The multi-mode baseband chipset integrates an HPLC modem, a 5G NR physical layer processing unit, and a satellite communication codec unit, supporting parallel processing of multi-standard signals. Preferably, a shared memory area is set up between each processing unit, enabling cross-protocol data exchange through a DMA controller to reduce bus transmission latency.

[0022] The communication protocol stack utilizes a layered abstraction design, with modular encapsulation of the physical and MAC layers, supporting dynamic loading of protocol specifications for different communication formats. This design allows adding new communication modes by simply expanding the protocol module library without modifying the core system architecture.

[0023] S2. Real-time collection of multi-source fault data from the distribution network monitoring terminal through the selected communication channel and encrypted transmission. In this embodiment, the multi-source data collection and secure transmission process is specifically implemented as follows: The data acquisition system is equipped with multi-dimensional sensing units, including an electrical quantity monitoring module, an environmental parameter detection module, and a communication quality perception module. Each module utilizes a distributed data cache architecture, aligning multi-source information through a time-stamped data packet structure, with optimal time-stamp accuracy reaching microseconds. The data acquisition terminal incorporates a built-in high-precision ADC conversion unit and supports multi-channel parallel sampling, ensuring complete capture of transient processes.

[0024] Electrical quantity data collection targets zero-sequence current, line voltage, and phase angle parameters. The sampling rate is adaptively adjusted according to the system frequency to meet the following requirements: f s ≥10·f max ; f s : Actual sampling frequency (unit Hz) f max : Preset highest harmonic frequency component (unit Hz) The acquired data is pre-processed by an FIR anti-aliasing filter and then compressed using an improved μ-law compression algorithm. The compression ratio can be configured from 4:1 to 8:1.

[0025] Environmental parameter detection includes temperature, humidity and electromagnetic field strength measurement. Sensor arrays are deployed at key nodes of power distribution equipment. Spatial interpolation algorithms are used for data fusion: in Where d ij represents the electrical distance between nodes i and j, ∈ is a small constant to prevent division by zero, and its preferred value is 0.01.

[0026] in: K root : 256-bit root key PUF resp : Physical unclonable function response value (512 bits) Challenge: 128-bit random challenge code issued by the authentication server Counter: 32-bit incrementing counter to prevent replay attacks.

[0027] ∣∣: data concatenation operator.

[0028] The session key is derived using the HKDF algorithm: K session =HKDF-Expand(K root ,Nonce,L); in: K session : 128-bit session key Nonce: 64-bit random salt value (including GPS timestamp hash) L: Key length parameter (value is 128) Data encryption uses a block dynamic encryption strategy to divide the data stream into fixed-length data blocks D i , each data block is encrypted independently: C i: the i-th ciphertext block D i : The i-th plaintext block (length 128 bits) Nonce i : Block-level random number (generated by concatenating the main Nonce and the block number i). After encryption, a security verification tag is added to the data: Tag = SM3 (C1 ‖ C2 ‖ ... ‖ C n ‖Timestamp); in: Tag: 256-bit checksum tag Timestamp: 64 A bit-accurate timestamp (UTC format).

[0029] The verification tag contains dual verification information of data integrity and timeliness.

[0030] The time synchronization mechanism uses a hybrid clock source solution, with the primary clock source being the IEEE 1588 Precision Time Protocol and the backup clock source being the GPS1PPS signal. The time scale generation algorithm compensates for transmission delays: where τ up , τ down are the uplink and downlink delay measurements, respectively, obtained through two-way message exchange. sync is the precise time after synchronization, t local is the local clock reading, Δt offset The master-slave clock offset. Data preprocessing includes an outlier detection and repair module, using an improved Z-score algorithm: Where μ=median(X),σ=1.4826·MAD; Among them, z i is the standardized deviation value of the i-th data point, x i is the original data point, μ is the median of the data set, MAD is the median absolute deviation, and X is the sliding window data set (window length N = 30).

[0031] When |z i When |>3, the data repair mechanism is triggered, and interpolation compensation is performed using the weighted average of adjacent node data.

[0032] The communication quality data acquisition module monitors channel status parameters in real time, including bit error rate, signal strength, and latency jitter. This monitoring data is encapsulated into a separate Management Information Base (MIB) and transmitted via an in-band management channel, taking precedence over service data scheduling.

[0033] S3. Perform spatiotemporal fusion processing on the transmitted fault data to generate a spatiotemporal correlation feature vector; In this embodiment, the spatiotemporal fusion process achieves accurate fault feature extraction through multi-stage data processing. First, a delay compensation model is established to address the asynchrony problem of multi-channel transmission and construct a time-varying compensation matrix: Where: τ i (t): The transmission delay measurement value of the i-th communication channel at time t (unit: ms), i∈{1,2,3} corresponds to HPLC, 5G, and satellite channels, respectively.

[0034] α i : Delay attenuation factor (dimensionless), preferably dynamically configured according to the channel type, satisfying

[0035] Delay compensation coefficient, used to correct the arrival time deviation of data packets.

[0036] Preferably, it is dynamically configured according to the channel type. The compensation matrix acts on the original data stream to eliminate the time asynchronous deviation of the multi-source data. Then the improved Kalman filter algorithm is implemented to define the state space model: in, is the system state vector at time k, is the observation vector, is the state transition matrix, is the observation matrix, w k and v k are process noise and observation noise respectively. Noise covariance matrix is dynamically updated: Q k =diag(q1,q2,q3), q i : Process noise variance component, preset according to the dynamic characteristics of the system.

[0037] The observation noise variance component is calculated as follows: N: Sliding window length (preferably 20-50 sampling periods) The delay value of the kth sampling of the i-th channel.

[0038] The Kalman gain matrix calculation formula is: K k =P k|k-1 H T (HP k|k-1 H T +R k ) -1 ; Where, P k|k-1 is the state prediction covariance matrix, through the recursive formula P k|k-1 =AP k-1 A T +Q k-1 Update, Q k-1 is the process noise covariance matrix. This design enables adaptive tracking of noise characteristics. Furthermore, the generation of spatiotemporal correlation feature vectors includes the following processing steps: performing wavelet packet decomposition on the compensated time series data to extract the energy characteristics of each frequency band: Where: W i,j (t) is the wavelet packet coefficient of the jth node in the i-th decomposition layer, T is the analysis window length (preferably 5-10 power frequency cycles), E i,j is the frequency band energy feature, reflecting the spectral characteristics of the transient process. The spatial features are extracted through the graph convolutional network: Where: Add the self-connected adjacency matrix (A is the power grid topology adjacency matrix, l is the identity matrix).

[0039] is a degree matrix that satisfies

[0040] The l-th layer node feature matrix (N is the number of nodes, d is the feature dimension) Trainable parameter matrix.

[0041] σ: Activation function (preferably ReLU). This process can capture the deep correlation characteristics of the power grid topology. The resulting spatiotemporal correlation feature vector contains the following dimensions: transient process duration, current mutation rate, and voltage fluctuation period in the time dimension; and node electrical distance, branch impedance correlation, and regional power balance in the spatial dimension. The feature vector is normalized: Among them, F i is the original eigenvector, μ F is the feature mean vector, dynamically calculated by the online statistics module, σ F is the characteristic standard deviation vector, updated by the sliding window method, This is the normalized feature vector used as input for subsequent deep learning models. This process ensures the compatibility of features of different dimensions.

[0042] During the processing, a data verification mechanism is set up between modules. When the eigenvector dimension does not match or the value exceeds the reasonable range, a data retransmission request is triggered. The verification threshold is preferably set to the ±3σ interval of the theoretical range of the eigenvalue.

[0043] S4. Extract fault spatiotemporal features through deep learning model based on spatiotemporal correlation feature vectors; In this embodiment, the deep feature extraction process achieves hierarchical abstraction of multi-dimensional features through a spatiotemporal hybrid neural network architecture. The network structure includes a temporal dimension feature extraction unit and a spatial topology feature extraction unit, and a gated fusion mechanism is used to achieve cross-domain feature interaction.

[0044] The temporal convolutional network (TCN) adopts an expanded causal convolution structure, which is mathematically expressed as: Where: The output feature vector (dimension D) of layer l at time t.

[0045] The weight matrix of the k-th convolution kernel in the l-th layer.

[0046] Represents the input features of layer l-1 at time t. d = 2 l is the expansion factor, and K is the convolution kernel size (preferably 5-9). This structure captures long-range temporal dependencies by exponentially expanding the field while maintaining temporal causality constraints. The Graph Attention Network (GAT) constructs a power grid topology graph G = (V, E, W), where the node set V corresponds to the monitoring point and the edge weight W ij Obtained by electrical distance and impedance correlation calculation. The feature aggregation process is defined as: h i ′ =σ(∑ j∈N(i) α ij Wh j ); Where: The updated feature vector of node i.

[0047] σ: non-linear activation function (ELU is preferred).

[0048] N(i): The set of neighboring nodes of node i.

[0049] α ij ∈{0,1}: the attention weight of node j to i.

[0050] Trainable linear transformation matrix.

[0051] The raw feature vector of node j.

[0052] The attention coefficient calculation adopts a multi-head mechanism: Where: Normalized weight of the mth attention head.

[0053] The trainable parameter vector of the mth attention mechanism.

[0054] The feature transformation matrix of the m-th head.

[0055] ∣∣: Feature vector concatenation operation.

[0056] h i , Input features of nodes i, j.

[0057] m=1,...,M represents the number of attention heads (preferably M≥3), a m is a trainable parameter vector. This design can adaptively focus on key topologically related nodes. The feature fusion layer adopts a dynamic gating mechanism to define the spatiotemporal feature fusion weight: g = σ(U g [h tcn ‖h gat ]+b g ); Where h tcn and h gat are the output features of the temporal convolutional network and the graph attention network, U g is the weight matrix, σ is the sigmoid activation function. The final fusion feature is calculated as: h fusion =g⊙h tcn +(1-g)⊙h gat ; This mechanism dynamically adjusts the contribution of spatiotemporal features according to feature importance. Network training adopts a multi-task learning strategy and defines a composite loss function: The classification loss Using the cross entropy function, reconstruction loss Feature decoupling and regularization terms are achieved through the autoencoder structure Contains weight decay and topology constraints, λ1, λ2, and λ3 are loss weight coefficients.

[0058] Feature decoupling module design orthogonal projection matrix satisfy: P TP = I; The Schmidt orthogonalization process is used to decouple the feature space and extract the main components of the fault features. This process can suppress redundant information interference and improve feature discrimination. The network optimization adopts an adaptive learning rate strategy and defines the parameter update rules: in and are the bias-corrected estimators of the first and second moments of the gradient, ∈ is a small constant to prevent division by zero, The model parameters of the tth iteration. The optimizer can automatically adapt to the characteristic scale of different parameters. Dynamic quantization is implemented in the model deployment phase to convert 32-bit floating-point weights into 8-bit fixed-point representation: Where μ is the weight mean, s is the scaling factor, and b = 8 is the number of quantization bits. This process significantly reduces computing resource consumption while maintaining model accuracy.

[0059] S5. Combine the grid topology and the temporal and spatial characteristics of the fault to make fault location decisions.

[0060] In this embodiment, the fault location decision process achieves accurate positioning through multi-dimensional criterion fusion and topology constraint optimization. First, a fault confidence evaluation function is constructed, integrating electrical quantity characteristics and depth characteristics: Where, I 0,i represents the zero-sequence current amplitude of node i (unit: A), and the fundamental component is extracted through the FIR filter; is the voltage change rate (unit: kV / ms), calculated using the central difference method; f NN (X i ) represents the feature weights output by the deep neural network, is the spatiotemporal fusion feature vector, and n is the total number of power grid nodes.

[0061] Furthermore, a topology constraint optimization model is established to encode the grid structure information into a graph Laplacian matrix: L = DW; Where D is the degree matrix (D ii =∑ j W ij ), W is the adjacency matrix, and the edge weight is determined by the inverse of the electrical distance between nodes. The Laplacian matrix encodes the grid topology. The optimization objective function is defined as: Where, is the node feature matrix, λ is the regularization coefficient (preferably 0.3-0.7), and tr(·) represents the matrix trace operation. The model is solved using the alternating direction multiplier method (ADMM), balancing the original confidence and topological consistency. The decision verification module implements a multi-level verification mechanism. First, a temporal continuity test is performed: When three consecutive sampling cycles (T=3) Exceeding the threshold θ c When the value is greater than 0.15 (preferably 0.15), the data review process is triggered. The review process calls the blockchain smart contract to compare the consistency of the decision results of adjacent nodes.

[0062] The audit log generation module uses the Merkle tree structure to store positioning parameters and build tamper-proof records: m =Hash(H m-1 ||Hash(t||C i ||coord i )); Where H m is the hash value of the mth block, t is the exact timestamp (UTC format), coord i is the fault coordinate (WGS-84 coordinate system), || represents the data splicing operation, and the log storage uses distributed ledger technology to ensure data traceability.

[0063] The result output module defines a standardized interface protocol, including: Fault coordinates: (lat, lon) = WGS84 (bus i d); Confidence: Disposal recommendation code: coded according to IEC 61850-7-4 standard.

[0064] Where μ C and σ C is a historical confidence statistical parameter, dynamically updated using a sliding window method (the window length is preferably 200-500 samples), and e is the natural logarithm base. The output data is encapsulated in ASN.1 format and supports IEC 60870-5-104 protocol transmission.

[0065] The exception handling mechanism sets a dual fault tolerance strategy. When the confidence of the main positioning algorithm output is lower than the threshold θ l (preferred 0.65), automatically switch to the backup algorithm: Where deg(i)∈N is the degree of node i (number of adjacent nodes) N(i) is the set of neighboring nodes of node i.

[0066] Cj ∈[0,1] is the original confidence of the neighborhood node j This strategy improves the positioning reliability of edge nodes by weighted averaging the confidence scores of neighboring nodes. The switching process is recorded in the audit log and triggers an operation and maintenance alarm signal.

[0067] Another embodiment of the present invention provides a distribution network fault intelligent location system based on HPLC multimodal communication, comprising: A communication link establishment module is used to build a multimodal communication link through the HPLC communication network and the auxiliary communication network, and dynamically select the optimal communication channel; Data acquisition and transmission module, used to collect multi-source fault data of distribution network in real time and transmit it to data processing module through encrypted channel; The time-space fusion processing module is used to perform time alignment compensation and spatial topology association on the received fault data to generate a time-space fusion feature vector; The deep feature extraction module is used to extract the spatiotemporal characteristics of faults through a hybrid model of temporal convolutional networks and graph neural networks. The intelligent decision-making module is used to combine the power grid topology and deep features to output fault location results and confidence assessments. The security protection module is used to implement device authentication and dynamic data encryption based on physically unclonable functions.

[0068] The examples of this specific embodiment are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, any equivalent changes made based on the structure, shape, and principle of this application should be included in the scope of protection of this application.

Claims

1. A distribution network fault intelligent location method based on HPLC multimodal communication, characterized in that: The following steps are involved: S1. Construct a multimodal communication link through the HPLC communication network and at least two auxiliary communication networks, and dynamically select the optimal communication channel; S2. Collect multi-source fault data from the distribution network monitoring terminal in real time through the selected communication channel and transmit it in encrypted form; S3. Perform spatiotemporal fusion processing on the transmitted fault data to generate a spatiotemporal correlation feature vector; S4. Extract fault spatiotemporal features through deep learning model based on spatiotemporal correlation feature vectors; S5. Combine the grid topology and the temporal and spatial characteristics of the fault to make fault location decisions.

2. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The algorithm for dynamically selecting the optimal communication channel in S1 satisfies: Among them, P i is the power consumption of the ith channel, R i is the transmission rate of the i-th channel, x i ∈{0,1} is the channel selection state, R th is the minimum rate threshold, N is the total number of available channels; The auxiliary communication network includes a combination of 5G slicing network, 230MHz wireless private network and low-orbit satellite communication, and satellite communication is automatically enabled when the main channel signal strength is lower than -110dBm.

3. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The spatiotemporal fusion processing in S3 includes: Using the improved Kalman filter algorithm, the gain coefficient calculation satisfies: in: H k is the Kalman gain matrix at the kth moment, P k|k-1 is the state prediction covariance matrix, H k is the observation matrix, R k is the observation noise covariance matrix, which is defined as: The delay variance of the i-th communication channel.

4. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The fault location decision in S5 includes: The fault confidence function is defined as: in: C i is the fault confidence of node i, I 0,i is the zero-sequence current amplitude of node i, U i is the voltage at node i, is the voltage change rate, f NN is the neural network feature extraction function, X i : The multidimensional feature vector of node i.

5. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The key generation for the encrypted transmission satisfies: K dev =Hash(PUF resp ⊕Challenge); Among them: K dev Unique identification key for the device, PUF resp is the response value of the physical unclonable function, Challenge is the random challenge code input from the outside, ⊕ is the XOR operation, and Hash is the cryptographic hash function.

6. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The deep learning model includes: a temporal convolutional network layer with a dilation factor of 2 l The hollow convolution structure is constructed, where l is the number of network layers; the graph attention network layer aggregates the features of neighboring nodes through a multi-head attention mechanism, and the number of attention heads M ≥ 3; the feature fusion layer uses a gating mechanism to dynamically fuse spatiotemporal features.

7. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: The multi-source fault data includes: Electrical quantity data: zero-sequence current, line voltage, phase angle; Environmental data: temperature, humidity, electromagnetic intensity; Communication quality data: bit error rate, signal strength, delay jitter.

8. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, wherein: It also includes the following steps to verify the credibility of the fault location results: Verify decision consistency through blockchain smart contracts; When the regional confidence difference exceeds 15%, the manual review mechanism is triggered; Generates an immutable audit log with timestamp and location parameters.

9. The method for intelligently locating distribution network faults based on HPLC multimodal communication according to claim 1, characterized in that: The spatiotemporal correlation feature vector includes: Time dimension characteristics: current mutation rate, voltage fluctuation period, and transient process duration; Spatial topological characteristics: node electrical distance, branch impedance correlation, and regional power balance.

10. A distribution network fault intelligent positioning system based on HPLC multimodal communication, according to any one of claims 1 to 9, characterized in that: include: A communication link establishment module is used to build a multimodal communication link through the HPLC communication network and the auxiliary communication network, and dynamically select the optimal communication channel; Data acquisition and transmission module, used to collect multi-source fault data of distribution network in real time and transmit it to data processing module through encrypted channel; The time-space fusion processing module is used to perform time alignment compensation and spatial topology association on the received fault data to generate a time-space fusion feature vector; A deep feature extraction module, which is used to extract fault spatiotemporal features through a hybrid model of temporal convolutional networks and graph neural networks; Intelligent decision-making module, which combines the grid topology and deep features to output fault location results and confidence assessment; The security protection module is used to implement device authentication and dynamic data encryption based on physically unclonable functions.

Images