AI prediction type low-voltage communication network fault self-recovery method

Abstract

The invention is suitable for the technical field of electric power Internet of Things, and relates to an AI prediction type low-voltage communication network fault self-recovery method, which comprises the following steps: S10, collecting quality parameters of each communication link in a low-voltage transformer area in real time through a PLC / RF dual-mode communication module; s20, constructing a dynamic link quality thermodynamic diagram based on the data acquired in the step S10 in combination with transformer-branch-box-household GIS geographic information of the transformer area; s30, collecting a plurality of months of quality parameters and corresponding fault records of each communication link in the S10, training the LSTM time sequence prediction model after preprocessing, and outputting a fault type and a fault probability; and S40, calling a corresponding scheme from a preset strategy library according to the fault type in the step S30, and executing a self-healing strategy through a fault self-healing scheduling component of the electric red operating system. The method is simple in process and convenient to operate, and solves the problems that fault detection of a traditional low-voltage communication network is lagged, self-healing is passive, and visualization is insufficient.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power Internet of Things, and particularly relates to an AI prediction type low-voltage communication network fault self-healing method. BACKGROUND

[0002] Under the promotion of new power system construction, large-scale access of "source network load storage" equipment in low-voltage transformer area puts forward urgent needs for high reliability, low latency and self-healing ability of communication network.

[0003] However, the existing low-voltage communication network fault processing has significant shortcomings: the traditional method relies on manual inspection or passive reporting after the fault occurs, resulting in delayed fault detection, inability to predict problems such as PLC open-phase or RF interference in advance, causing business interruption such as meter reading data loss and charging pile scheduling delay, and difficulty in meeting the low-latency requirement of "500 node collection ≤1 minute"; link state visualization is insufficient, only single-point parameter viewing can be achieved without global understanding of the link distribution state of the transformer area, and the average time consumption for fault location is more than 30 minutes, which is inconsistent with the visualization requirement of GIS-based link quality heat map; the self-healing ability is passive and inefficient, manual intervention is required to switch the link after the fault, the self-healing delay is up to hundreds of milliseconds, and the AI dynamic resource adjustment ability is lacking, which cannot meet the N-1 / N-2 redundancy and high-reliability communication requirements. In addition, the existing method is not adapted to the unified Internet of Things management architecture of the electric Hong operation system, and it is difficult to integrate into the "low-voltage transformer area one network" construction system. The patent with publication number CN117938204B provides a low-voltage power line carrier communication network self-repairing method, which monitors the power line carrier communication network transmission nodes, judges the fault immediately when an abnormal transmission is found, selects other nodes without fault to maintain the normal operation of the original carrier communication network, and simultaneously predicts the fault of the power line carrier communication network in advance. In this patent, the global understanding of the link distribution state of the transformer area is insufficient, and the visualization is insufficient, which has the same disadvantages as the existing technology.

[0004] Therefore, how to solve the problems of delayed fault detection, passive self-healing and insufficient visualization of the traditional low-voltage communication network to improve the reliability and self-healing ability of the low-voltage communication network is urgent. SUMMARY

[0005] In view of the shortcomings of the prior art, the purpose of the present application is to provide an AI prediction type low-voltage communication network fault self-healing method to solve the problems of delayed fault detection, passive self-healing and insufficient visualization of the low-voltage communication network in the prior art.

[0006] In order to solve the above technical problems, the present application adopts the following technical scheme:

[0007] The present application provides an AI prediction type low-voltage communication network fault self-healing method, comprising the following steps:

[0008] S10. The quality parameters of each communication link in the low-voltage distribution area are collected in real time through the PLC / RF dual-mode communication module.

[0009] S20. Based on the data collected in S10, and combined with the GIS geographic information of the transformer-branch-box-household area, a dynamic link quality heat map is constructed.

[0010] S30. Collect the quality parameters and corresponding fault records of each communication link in S10 for several months, train the LSTM time series prediction model after preprocessing, and output the fault type and fault probability.

[0011] S40. Based on the fault type in S30, call the corresponding solution from the preset strategy library and execute the self-healing strategy through the fault self-healing scheduling component of the Elec-Tech operating system.

[0012] Furthermore, in S10, the quality parameters include the signal-to-interference-plus-noise ratio, bit error rate, three-way connectivity, received signal strength indication, transmit power, receive sensitivity, and packet loss rate of the PLC link; the acquisition cycle is 1-5 minutes, and the data is actively reported to the network management system through the link data acquisition component of the Elec-Tech operating system.

[0013] Furthermore, in S20, the heatmap uses color to indicate the link quality level, where green represents excellent quality (SNR ≥ 80dB and bit error rate ≤ 10⁻⁵); yellow represents medium quality (60dB ≤ SNR < 80dB and 10⁻⁵ < bit error rate ≤ 10⁻³); and red represents poor quality (SNR < 60dB or bit error rate > 10⁻³). The heatmap is dynamically updated hourly, synchronously storing historical heatmap data for the past 30 days, and supports trend backtracking.

[0014] Furthermore, the specific steps in S30 are as follows:

[0015] S301. Collect at least 3 months of link quality data and corresponding fault records, including PLC phase loss, RF channel interference, and link attenuation;

[0016] S302. Data collected is preprocessed using Kalman filtering for noise reduction and Min-Max normalization.

[0017] S303. Using the preprocessed link quality time-series features and topology features as inputs, and the fault type and fault probability as outputs, the training model achieves a prediction accuracy of ≥95% on the test set and an advance prediction time of ≥3 minutes.

[0018] S304: Input the latest link data into the model every 10 seconds and output the failure probability and type of each link in the next 3-5 minutes; when the failure probability is ≥85%, trigger a fault warning and push it to the network management system.

[0019] Furthermore, in S40, if the PLC N-1 / N-2 phase is lost: if it is N-1, then switch to the remaining two phases of PLC+RF, and the subcarrier is allocated as 3MHz for each of the two phases of PLC and 4MHz for RF; if it is N-2, then switch to the remaining one phase of PLC+RF and enable RS(127,119) forward error correction.

[0020] If it is RF interference, switch to the backup RF band, or prioritize the use of the PLC link to carry high-priority services.

[0021] If it is link attenuation, dynamically increase the PLC carrier power or RF transmit power and optimize the coupling circuit parameters.

[0022] Further, self-healing execution and verification: The self-healing strategy is executed through the fault self-healing scheduling component of the Elec-Tech operating system. The link quality is checked within 10ms. If the SNR is ≥60dB and the bit error rate is ≤10 to the power of -4, the self-healing is considered successful. Otherwise, a second self-healing is triggered. The self-healing delay is ≤50ms.

[0023] Logs and alarms: The self-healing process is reported to the network management system in real time, generating logs that include fault type, prediction accuracy, and self-healing time, and is synchronized to the security log module of the Elec-Tech operating system.

[0024] Furthermore, in S10, the dual-mode communication module supports the IPv6 protocol stack, collects data carrying the device IP address and topology location information, realizes real-time positioning of meterless archives, and conforms to the IP-based networking standard of smart IoT.

[0025] Furthermore, the LSTM timing prediction model in S30 introduces an attention mechanism, focusing on fault-sensitive parameters, including PLC three-way connectivity and RF packet loss rate, and fine-tunes it using historical fault data from the power grid area to adapt to the characteristics of low-voltage communication scenarios.

[0026] Furthermore, in S40, the self-healing strategy follows service priority: high-priority services are allocated links with better quality first, including load scheduling and fault alarms; low-priority services temporarily use links with medium quality, including historical data retransmission, to ensure concurrent transmission of multiple services from source network to load storage.

[0027] Furthermore, adaptation is completed through the Elec-Hong SDK, including uboot / kernel driver adaptation, Elec-Hong API integration of the link acquisition-AI prediction-self-healing execution module, and compliance, integrity and security testing of the Elec-Hong operating system certified version.

[0028] Compared with existing technologies, the AI-predictive low-voltage communication network fault self-healing method provided by this invention has at least the following advantages:

[0029] Traditional low-voltage communication networks suffer from lagging fault detection, passive self-healing, and insufficient visualization. This invention features a simple and convenient process. It collects multi-dimensional link quality data (SNR, bit error rate, RSSI, etc.) through a dual-mode communication module (PLC / RF), and constructs a dynamic link quality heatmap by combining it with the area's GIS topology. An LSTM time-series prediction AI model is used to predict link faults in advance. Based on the prediction results, a self-healing strategy (PLC / RF switching, N-1 / N-2 redundancy adjustment, resource allocation optimization) is automatically matched and executed through the Elec-Tech operating system, achieving a fault self-healing latency of ≤50ms. This invention achieves a closed loop of "prediction-visualization-self-healing," adapts to the needs of "low-voltage distribution areas," supports stable transmission of multiple services across "source, network, load, and storage," and ensures data acquisition latency of ≤1 minute for a 500-node distribution area, thus improving the reliability and self-healing capability of low-voltage communication networks. Attached Figure Description

[0030] To more clearly illustrate the solution of the present invention, a brief introduction will be given to the drawings used in the description of the embodiments below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0031] Figure 1 A flowchart of an AI-predictive low-voltage communication network fault self-healing method provided in an embodiment of the present invention;

[0032] Figure 2 This is a diagram illustrating the multi-dimensional data acquisition mechanism for link quality in an AI-predictive low-voltage communication network fault self-healing method provided in an embodiment of the present invention.

[0033] Figure 3 This diagram illustrates a method for constructing a heatmap of link quality in a substation area, as provided in an embodiment of the present invention for an AI-predictive self-healing method for low-voltage communication network faults.

[0034] Figure 4 This is a flowchart illustrating the design of an AI fault prediction model in an AI-predictive low-voltage communication network fault self-healing method provided in an embodiment of the present invention.

[0035] Figure 5 This is a schematic diagram of the fault self-healing strategy system in an AI predictive low-voltage communication network fault self-healing method provided in an embodiment of the present invention;

[0036] Figure 6This is a hardware and software configuration diagram of an AI predictive low-voltage communication network fault self-healing method provided in an embodiment of the present invention;

[0037] Figure 7 This diagram illustrates the entire process of implementing an AI-predictive low-voltage communication network fault self-healing method according to an embodiment of the present invention. Detailed Implementation

[0038] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0040] This invention provides an AI-based method for predicting and self-healing faults in low-voltage communication networks. This method is applied to low-voltage distribution PLC / RF dual-mode communication networks in power grids to support highly reliable, low-latency transmission of multiple services related to "source-grid-load-storage" (electricity meters, photovoltaic systems, and charging piles). The AI-based method for predicting and self-healing faults in low-voltage communication networks includes the following steps:

[0041] S10. Real-time acquisition of quality parameters of each communication link within the low-voltage distribution area via PLC / RF dual-mode communication module; S20. Construction of dynamic link quality heat map based on data acquired in S10 and combined with GIS geographic information of transformer-branch-box-customer distribution area; S30. Collection of quality parameters and corresponding fault records of each communication link in S10 for several months, training of LSTM time series prediction model after preprocessing, and output of fault type and fault probability; S40. Calling corresponding scheme from preset strategy library according to fault type in S30, and executing self-healing strategy through fault self-healing scheduling component of Dianhong operating system.

[0042] This invention has a simple process and is easy to operate, solving the problems of lagging fault detection, passive self-healing, and insufficient visualization in traditional low-voltage communication networks.

[0043] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0044] This invention provides an AI-based method for predicting and self-healing faults in low-voltage communication networks. This method is applied to low-voltage distribution PLC / RF dual-mode communication networks in power grids to support highly reliable, low-latency transmission of multiple services related to "source-grid-load-storage" (meters, photovoltaics, charging piles). Figures 1 to 7 In this embodiment, the AI-predicted low-voltage communication network fault self-healing method includes the following steps:

[0045] S10. Multi-dimensional data acquisition of link quality: Through the PLC / RF dual-mode communication module, the quality parameters of each communication link in the low-voltage distribution area are collected in real time. The quality parameters include SNR (signal-to-interference-plus-noise ratio), bit error rate, and three-way connectivity of the PLC link, as well as RSSI (received signal strength indication), TxPower (transmit power), RxSensitivity (receiver sensitivity), and packet loss rate of the RF link. The acquisition cycle is 1-5 minutes, and the data is actively reported to the network management system through the "link data acquisition component" of the Elec-Tech operating system.

[0046] In this embodiment, the dual-mode communication module supports the IPv6 protocol stack, collects data carrying the device IP address and topology location information, and realizes real-time positioning of meter-free records, which conforms to the IP-based networking standard of smart IoT.

[0047] S20. Construction of a heat map of transformer substation link quality: Based on the data collected in S10, and combined with the GIS geographic information of the "transformer-branch-box-household" area, a dynamic heat map of link quality is constructed. The heat map uses color to indicate the link quality level (green: excellent quality, SNR≥80dB and bit error rate≤10⁻⁵; yellow: medium quality, 60dB≤SNR<80dB and 10⁻⁵<bit error rate≤10⁻³; red: poor quality, SNR<60dB or bit error rate>10⁻³). The heat map is dynamically updated hourly, synchronously storing historical heat map data for the past 30 days, and supports trend backtracking.

[0048] In this embodiment, the heatmap supports three-level topology drilling: "transformer area - branch box - user". It can view the real-time parameters (SNR, RSSI) and historical trend curves (last 24 hours) of a single link, and is integrated into the web and mobile terminals of the Dianhong network management software.

[0049] S30, AI fault prediction model training and fault early warning, the specific steps are as follows:

[0050] S301. Construct a historical dataset: Collect at least 3 months of link quality data and corresponding fault records (PLC phase loss, RF channel interference, link attenuation) as described in S1.

[0051] S302, Kalman filtering is used for noise reduction and min-max normalization preprocessing.

[0052] S303. Training the LSTM time series prediction model: The model takes the preprocessed link quality time series features (5-minute sliding window mean, fluctuation variance) and topology features (branch box number to which the link belongs) as input, and the fault type (classification) and fault probability (regression) as output. The prediction accuracy of the training model on the test set is ≥95%, and the prediction time is ≥3 minutes in advance.

[0053] S304 Real-time fault prediction: The latest link data is input into the model every 10 seconds, and the fault probability (0-100%) and type of each link in the next 3-5 minutes are output; when the fault probability is ≥85%, a fault warning is triggered and pushed to the network management system.

[0054] In this embodiment, the LSTM model introduces an attention mechanism, focusing on fault-sensitive parameters (PLC three-way connectivity, RF packet loss rate), and is fine-tuned using historical fault data (≥1000 records) from the Southern Power Grid area to adapt to the characteristics of low-voltage communication scenarios.

[0055] S40. Automatic execution of fault self-healing strategy: Self-healing strategy matching: Based on the fault type in S30, the corresponding scheme is called from the preset strategy library: If it is a phase loss in PLC N-1 / N-2: N-1 (one phase loss) switches to the remaining two phases of PLC+RF, and the subcarrier is allocated as 3MHz for each of the two phases of PLC and RF. 4MHz; N-2 (two phases disconnected) switch to the remaining one phase PLC (4MHz) + RF (6MHz), enable RS (127,119) forward error correction; if it is RF interference: switch to the backup RF band (within 2.405-2.435GHz), or prioritize enabling the PLC link to carry high-priority services (meter reading, fault alarm); if it is link attenuation: dynamically increase the PLC carrier power (maximum 2dBm) or RF transmit power (maximum 3dBm), optimize the coupling circuit parameters; self-healing execution and verification: execute the strategy through the "fault self-healing scheduling component" of Elec-Tech, detect the link quality within 10ms, if SNR≥60dB and bit error rate≤10 to the power of -4, the self-healing is determined to be successful; otherwise, trigger secondary self-healing (such as replacing the backup CCO agent); self-healing delay≤50ms; logs and alarms: the self-healing process is reported to the network management system in real time, generating a log containing fault type, prediction accuracy, self-healing time, and synchronized to the Elec-Tech operating system security log module.

[0056] In this embodiment, the self-healing strategy follows the service priority: high-priority services (load scheduling, fault alarm) are given priority to high-quality links (RF or PLC high-quality links), while low-priority services (historical data retransmission) temporarily use medium-quality links to ensure concurrent transmission of multiple services in the "source network load storage" system.

[0057] Furthermore, in this embodiment, adaptation is completed through the Elec-Tech SDK, including uboot / kernel driver adaptation, Elec-Tech API integration of the "link acquisition-AI prediction-self-healing execution" module, and certification by the Elec-Tech operating system (version compliance, integrity, and security testing).

[0058] This invention provides an AI-predictive self-healing method for low-voltage communication networks. Through a closed-loop design of "data acquisition - heatmap visualization - AI prediction - self-healing execution," it addresses the pain points of traditional low-voltage communication network fault handling, including:

[0059] Multi-dimensional data collection mechanism for link quality

[0060] This system covers key quality indicators for PLC and RF dual-mode links: For the PLC link, it includes the SNR (0-100dB, target ≥60dB) of each of the three phases, bit error rate (≤10^-5 is normal), three-phase connectivity (0 = phase loss, 1 = normal), and communication latency (≤50ms); for the RF link, it includes RSSI (-105dBm to -30dBm, ≥-70dBm is excellent), TxPower (20dBm±2dBm), RxSensitivity (≤-105dBm), and packet loss rate (≤1% is normal). Data collection employs a combination of "active reporting + timed querying," reporting every minute through the "link data acquisition component" of the Elec-Tech operating system, effectively avoiding the resource waste of traditional polling mechanisms. Simultaneously, the collected data carries the device's IPv6 address and topology location information, achieving "real-time location without meter archives."

[0061] Method for constructing heatmap of network link quality

[0062] The construction of the heatmap for transformer substation link quality first involves preprocessing the raw collected data. Kalman filtering is used to eliminate power line noise interference, and min-max normalization maps multi-dimensional parameters to a unified quality scoring system of 0-100 points. The preprocessed data is then mapped to the physical GIS topology of the transformer substation ("transformer-branch box-user") to accurately locate the specific position of each link. During the visualization phase, color coding is used based on the quality score: green (80-100 points, excellent quality), yellow (60-79 points, medium quality), and red (0-59 points, poor quality), dynamically generating a GIS heatmap. This heatmap supports hourly updates and stores nearly 30 days of historical data for trend analysis and backtracking (such as quality fluctuations in links of specific branch boxes). Finally, the heatmap is integrated into the Dianhong network management software, supporting large-screen display on the web and on-site maintenance viewing on mobile devices, and allowing flexible filtering by transformer substation, time dimension, and PLC / RF link type.

[0063] AI Fault Prediction Model Design

[0064] To address the significant temporal characteristics of low-voltage communication links, this invention employs an LSTM (Long Short-Term Memory) network to construct a core fault prediction model. Its design fully considers the temporal dependence and pattern complexity of this type of network data. The model first undergoes refined feature engineering, extracting two core features from the raw data: firstly, temporal features, including the mean SNR, maximum bit error rate, and RSSI fluctuation variance within a 5-minute sliding window, used to quantify the instantaneous state and stability of the link; secondly, topological features, such as the branch box number to which the link belongs and its physical distance from the Chief Node (CCO), these features profoundly affect signal transmission and attenuation characteristics. During the model training phase, over three months of historical data from a distribution area of ​​the Southern Power Grid (containing over 1000 detailed fault records) is used as the foundation, and the data is divided into training, validation, and test sets in a 7:2:1 ratio to ensure the fairness of the evaluation. The training process utilizes the Adam optimizer with a learning rate of 0.001, employing a composite loss function combining cross-entropy (for fault classification) and mean squared error (MSE) (for fault probability regression) to simultaneously optimize both classification accuracy and probability prediction accuracy. After thorough training, the model performs exceptionally well on the test set, maintaining a fault prediction accuracy consistently above 95% and providing an early warning time exceeding 3 minutes. To meet the stringent requirements of field deployment, the trained model underwent lightweight processing (compressed to under 5MB) and integrated into the ARM Cortex-M4 core AI inference unit of the CCO device. This deployed model possesses real-time prediction capabilities, receiving and processing the latest link status data every 10 seconds, outputting the probability (0-100%) of specific types of faults (such as PLC phase loss, RF interference, and link attenuation) occurring on each link within a future period. Once the probability value exceeds a preset threshold of 85%, an early warning signal is immediately triggered to the network management system.

[0065] Fault self-healing strategy system

[0066] This system performs precise self-healing operations for different types of faults: For PLC phase loss, in scenario N-1 (one phase loss), it automatically switches to the remaining two phases of the PLC and RF link, adjusting the MIMO-OFDMA subcarrier allocation ratio to 3MHz for each of the two PLC phases and 4MHz for the RF, ensuring a total bandwidth of no less than 10Mbps; in scenario N-2 (two phase loss), it switches to the remaining one phase of the PLC (4MHz) and RF (6MHz) combination, and enables RS(127,119) forward error correction coding, achieving automatic error correction even with a link error rate ≤10^-3, ensuring a bandwidth of no less than 5Mbps. Regarding RF interference, when channel interference is detected (RSSI fluctuation > 10dBm), the system automatically switches to a backup frequency band (e.g., switching from 2.405GHz to 2.435GHz); if the interference persists, the PLC link is enabled to carry high-priority services (such as load scheduling commands and fault alarms), while low-priority services (such as historical data retransmission) are temporarily cached. For link attenuation, the impedance matching parameters of the coupling circuit are optimized by dynamically increasing the PLC carrier power (maximum 2dBm) or RF transmit power (maximum 3dBm). If a single link experiences severe attenuation, an alternative relay route is automatically selected (e.g., forwarding through an adjacent PCO). After self-healing is executed, the system verifies the process within 10ms. If the link quality reaches SNR ≥ 60dB and the bit error rate ≤ 10^-4, the self-healing is considered successful; otherwise, a secondary self-healing is triggered (e.g., replacing the backup CCO agent), ensuring that the overall self-healing latency is controlled within an extremely short range. All self-healing logic is integrated into the "Fault Self-Healing Scheduling Component" of the Elec-Tech OS, calling the underlying interface of the dual-mode communication module through the Elec-Tech API, fully complying with the security and compliance requirements of the Elec-Tech OS.

[0067] Example 1

[0068] Hardware and software configuration

[0069] The core of the dual-mode communication module is an integrated sensing and computing control chip. This module supports two communication modes: PLC (operating frequency 1.8 / 2.8 / 3.8MHz) and RF (2.4GHz band). Its built-in ARM Cortex-M4 core (150MHz, 64KB RAM) provides computing power for complex communication protocol stacks and data processing, and has been deeply adapted to the Dianhong OS V2.0. The network management system is deployed on the Southern Power Grid Cloud and built using a microservice architecture. It integrates key functional modules such as "data acquisition service," "heat map generation service," "AI prediction service," and "self-healing scheduling service" to achieve access and management of tens of millions of nodes. To verify system performance, a typical low-voltage distribution area test environment with 500 nodes was built, including 400 smart meters, 50 photovoltaic inverters, and 50 charging piles. The network topology included 1 CCO (master node) and 200 STA (slave node), ensuring full three-phase coverage of PLC communication and RF signal strength of no less than -75dBm across the entire range, providing a solid foundation for simulating real-world scenarios.

[0070] Implementation steps

[0071] First, the link quality acquisition function was integrated into the STA module, configured to collect full-dimensional quality parameters (including SNR, bit error rate, RSSI, etc.) of the PLC and RF links at a 1-minute interval, and report the data to the CCO (master node) via the IPv6 protocol. The CCO then aggregates and forwards the data to the network management system deployed on the Southern Power Grid Cloud. To ensure data accuracy, the acquisition precision was rigorously verified, confirming that the SNR measurement error of the PLC link is ≤2dB and the RSSI measurement error of the RF link is ≤3dB, which fully meets the requirements of "link data acquisition precision". After receiving the data, the network management system completes preprocessing steps such as Kalman filtering for noise reduction and min-max normalization within 30 seconds, and generates a dynamic link quality heatmap by combining the GIS topology mapping of the distribution area. This heatmap supports real-time display on a large web screen and has hourly dynamic updates and the ability to backtrack historical data for the past 30 days. In the AI ​​prediction stage, the pre-trained LSTM model is deployed on the CCO's ARM Cortex-M4 AI inference unit, and the model's single inference time is controlled to ≤5ms. In the fault prediction test, when simulating a phase loss in PLC A phase, the model predicts the fault 4.2 minutes in advance with a probability of 92% and a type recognition accuracy of 100%. When simulating RF interference, it predicts the fault 3.5 minutes in advance with a probability of 88% and a type recognition accuracy of 98%. The fault self-healing test covers multiple scenarios: in PLC In the N-2 phase loss test, after manually disconnecting phases A / B, the model triggered an early warning. The system automatically switched to phase C PLC (4MHz) + RF (6MHz). The subcarrier allocation adjustment took 8ms and RS forward error correction coding was enabled. After self-healing, the link SNR was 78dB, the bit error rate was 6×10⁻⁶, the bandwidth was 8.5Mbps, and the total self-healing delay was 38ms. In the RF interference test, 2.415GHz interference was generated in the transformer area, causing the RSSI to fluctuate to -85dBm. The system automatically switched to the 2.435GHz band and increased the RF transmit power by 2dBm. After the switch, the RSSI was -62dBm, the packet loss rate was 0.3%, and the self-healing delay was 25ms. Finally, through the Elec-Hong OS certification, it was confirmed that the version hash value was consistent with the Southern Power Grid Unified IoT OS, and all power components were loaded normally. In the integration test, the self-healing component called the link switching interface of the dual-mode module through the Elec-Hong API, with a response time ≤10ms, which fully met the functional and performance requirements of the Elec-Hong OS.

[0072] The AI-predictive low-voltage communication network fault self-healing method described in the above embodiments, compared with existing technologies, suffers from lagging fault detection, passive self-healing, and insufficient visualization in traditional low-voltage communication networks. This invention features a simple process and convenient operation. It collects multi-dimensional link quality data (SNR, bit error rate, RSSI, etc.) through a dual-mode communication module (PLC / RF), constructs a dynamic link quality heatmap by combining it with the GIS topology of the distribution area, uses an LSTM time-series prediction AI model to predict link faults in advance, automatically matches self-healing strategies (PLC / RF switching, N-1 / N-2 redundancy adjustment, resource allocation optimization) based on the prediction results, and schedules and executes them through the Elec-Tech operating system, achieving a fault self-healing latency of ≤50ms. This invention achieves a closed loop of "prediction-visualization-self-healing," adapts to the needs of "low-voltage distribution areas," supports stable transmission of multiple services across "source, network, load, and storage," ensures data acquisition latency of ≤1 minute for a 500-node distribution area, and improves the reliability and self-healing capability of low-voltage communication networks.

[0073] Obviously, the embodiments described above are merely preferred embodiments of the present invention, and not all embodiments. The accompanying drawings illustrate preferred embodiments of the present invention, but do not limit the scope of the patent. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this invention.

Claims

1. An AI-predictive self-healing method for low-voltage communication network faults, characterized in that, Includes the following steps: S10. The quality parameters of each communication link in the low-voltage distribution area are collected in real time through the PLC / RF dual-mode communication module. S20. Based on the data collected in S10, and combined with the GIS geographic information of the transformer-branch-box-household area, a dynamic link quality heat map is constructed. S30. Collect the quality parameters and corresponding fault records of each communication link in S10 for several months, train the LSTM time series prediction model after preprocessing, and output the fault type and fault probability. S40. Based on the fault type in S30, call the corresponding solution from the preset strategy library and execute the self-healing strategy through the fault self-healing scheduling component of the Elec-Tech operating system.

2. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, In S10, the quality parameters include the signal-to-interference-plus-noise ratio, bit error rate, three-way connectivity, received signal strength indication, transmit power, receive sensitivity, and packet loss rate of the PLC link; the acquisition period is 1-5 minutes, and the data is actively reported to the network management system through the link data acquisition component of the Elec-Tech operating system.

3. The AI-predictive low-voltage communication network fault self-healing method according to claim 2, characterized in that, In step S20, the heatmap uses color to indicate the link quality level, where green represents excellent quality (SNR ≥ 80dB and bit error rate ≤ 10⁻⁵), yellow represents medium quality (60dB ≤ SNR < 80dB and 10⁻⁵ < bit error rate ≤ 10⁻³), and red represents poor quality (SNR < 60dB or bit error rate > 10⁻³). The heatmap is dynamically updated hourly, synchronously storing historical heatmap data for the past 30 days, and supports trend backtracking.

4. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, The specific steps in S30 are as follows: S301. Collect at least 3 months of link quality data and corresponding fault records, including PLC phase loss, RF channel interference, and link attenuation; S302. Data collected is preprocessed using Kalman filtering for noise reduction and Min-Max normalization. S303. Using the preprocessed link quality time-series features and topology features as inputs, and the fault type and fault probability as outputs, the training model achieves a prediction accuracy of ≥95% on the test set and an advance prediction time of ≥3 minutes. S304: Input the latest link data into the model every 10 seconds and output the failure probability and type of each link in the next 3-5 minutes; when the failure probability is ≥85%, trigger a fault warning and push it to the network management system.

5. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, In S40, if the PLC N-1 / N-2 phase is lost: if it is N-1, then switch to the remaining two phases of PLC+RF, and the subcarrier is allocated as 3MHz for each of the two phases of PLC and 4MHz for RF; if it is N-2, then switch to the remaining one phase of PLC+RF and enable RS (127,119) forward error correction. If it is RF interference, switch to the backup RF band, or prioritize the use of the PLC link to carry high-priority services. If it is link attenuation, dynamically increase the PLC carrier power or RF transmit power and optimize the coupling circuit parameters.

6. The AI-predictive low-voltage communication network fault self-healing method according to claim 5, characterized in that, Self-healing execution and verification: The self-healing strategy is executed through the fault self-healing scheduling component of the Elec-Tech operating system. The link quality is checked within 10ms. If the SNR is ≥60dB and the bit error rate is ≤10^-4, the self-healing is considered successful. Otherwise, a second self-healing is triggered. The self-healing delay is ≤50ms. Logs and alarms: The self-healing process is reported to the network management system in real time, generating logs that include fault type, prediction accuracy, and self-healing time, and is synchronized to the security log module of the Elec-Tech operating system.

7. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, In S10, the dual-mode communication module supports the IPv6 protocol stack, collects data carrying the device IP address and topology location information, realizes real-time positioning of meter-free records, and conforms to the IP-based networking standard of smart IoT.

8. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, The LSTM timing prediction model in S30 introduces an attention mechanism, focusing on fault-sensitive parameters, including PLC three-way connectivity and RF packet loss rate, and fine-tunes it using historical fault data from the power grid area to adapt to the characteristics of low-voltage communication scenarios.

9. The AI-predictive low-voltage communication network fault self-healing method according to claim 1, characterized in that, In S40, the self-healing strategy follows service priority: high-priority services are allocated links with better quality first, including load scheduling and fault alarms; low-priority services temporarily use links with medium quality, including historical data retransmission, to ensure concurrent transmission of multiple services from source network to load storage.

10. A method for self-healing faults in AI-predictive low-voltage communication networks according to any one of claims 1 to 9, characterized in that, Adaptation was completed using the eHong SDK, including uboot / kernel driver adaptation, eHong API integration of the link acquisition-AI prediction-self-healing execution module, and compliance, integrity and security testing of the eHong operating system certified version.

Images