Multi-modal sensing and self-adaptive communication method for power pipe network

Abstract

The invention relates to a multi-mode sensing and self-adaptive communication method of a power pipe network, which comprises the following steps of: constructing a Mesh network with self-healing capability based on a LoRaWAN and power carrier dual-mode communication architecture, avoiding channel interference through a dynamic spectrum sensing technology, reducing data transmission power consumption by adopting a sectional compression coding strategy, and realizing multi-mode sensing and self-adaptive communication of the power pipe network. The communication success rate under a complex working condition reaches 99.2%, a lightweight CNN-Transformer hybrid model is deployed at an edge gateway, feature extraction and space-time correlation analysis are performed on multi-dimensional sensing data, a pipe network health degree evaluation index is constructed through a digital twin engine, and the network health degree evaluation index is used for generating an equipment degradation trend map and maintenance strategy suggestions. UWB / geomagnetic / BOS combined positioning technologies are adopted, a pipe network topology constraint factor and a sliding window filtering algorithm are combined, the terminal positioning precision is smaller than or equal to 0.8 m, and diagnosis algorithm online iteration and functional module hot plug are supported through a containerized deployment technology.

Description

A Multimodal Sensing and Adaptive Communication Method for Power Pipelines Technical Field

[0001] This application relates to the field of electrical digital data technology, specifically to a multimodal sensing and adaptive communication method for power grids. Background Technology

[0002] Existing power grid monitoring devices generally suffer from problems such as limited sensing dimensions, poor environmental adaptability, and weak collaborative capabilities. Traditional monitoring terminals only support the acquisition of single physical quantities, with temperature detection errors exceeding ±3℃ and humidity monitoring response delays exceeding 10 seconds. They employ rigid power supply architectures, resulting in a battery life of less than 24 hours under no-light conditions. Communication modules rely on a single network standard, leading to a disconnection rate exceeding 65% in shielded areas. Data analysis uses threshold judgment methods, resulting in a false alarm rate as high as 42%, and they cannot predict progressive faults. While existing electronic markers with communication functions support basic status reporting, they lack the ability to fuse multi-source heterogeneous data, edge computing node collaboration mechanisms, and adaptive spectrum allocation strategies, making it difficult to achieve real-time and accurate assessment and early warning of the power grid's operating status. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the purpose of this application is to provide a multimodal sensing and adaptive communication method for power grids.

[0004] This application describes a multimodal sensing and adaptive communication method for power pipeline networks, comprising the following steps: S101, synchronously acquiring surface temperature, deformation, and mechanical vibration parameters of the pipeline network using a multi-physics field sensing array (including a MEMS temperature and humidity composite sensor, fiber optic strain gauge, and piezoelectric vibration sensor), combined with a wide-spectrum ambient light tracking photovoltaic module and an adaptive charge and discharge management circuit, for continuous operation for ≥96 hours under extreme weather conditions; S102, constructing a self-healing Mesh network based on a LoRaWAN and power line carrier dual-mode communication architecture, avoiding channel interference through dynamic spectrum sensing technology, and employing a segmented compression coding strategy. Reduce data transmission power consumption to achieve a communication success rate of 99.2% under complex operating conditions; S103, deploy a lightweight CNN-Transformer hybrid model on the edge gateway to perform feature extraction and spatiotemporal correlation analysis on multi-dimensional sensor data, and construct pipeline health assessment indicators through a digital twin engine to generate equipment degradation trend maps and maintenance strategy suggestions; S104, adopt a combination of UWB / geomagnetic / BOS positioning technologies, combined with pipeline topology constraint factors and sliding window filtering algorithms, to achieve terminal positioning accuracy ≤0.8 meters, and support online iteration of diagnostic algorithms and hot-swappable functional modules through containerized deployment technology.

[0005] Furthermore, in step S101, temperature, deformation, and vibration data of the pipeline surface are synchronously collected through a multi-physics field sensor array to form an initial multi-dimensional dataset. After time-series alignment and missing value imputation, a complete time series is obtained. The support vector machine algorithm is used to analyze parameter trends to identify potential risk points, and deep feature extraction is performed on abnormal periods to determine the specific anomaly type. Combined with ambient light tracking and photovoltaic operation status, the charging and discharging management circuit is dynamically adjusted to generate an optimized energy scheme and ensure the continuous operation of the system under extreme weather conditions.

[0006] Furthermore, in step S102, a wide-coverage communication network is constructed based on a dual-mode communication architecture, combining LoRaWAN and power line carrier technology. The network self-healing mechanism automatically repairs breakpoints, forms a topology, performs real-time spectrum sensing, and if an interfering channel is detected, the channel is reallocated to avoid interference. The transmitted data is compressed and segmented, and sent in batches to optimize transmission power consumption, while continuously monitoring communication quality.

[0007] Furthermore, in step S103, the edge gateway collects multi-source sensor data to monitor the pipeline network's operating status in real time, obtains the raw data stream, and uses a lightweight convolutional neural network and Transformer hybrid architecture to extract spatiotemporal features from the data to obtain a feature vector set. It then implements correlation analysis technology to construct a dependency model between data. When the feature correlation reaches a preset threshold, the digital twin engine virtually maps the pipeline network's operating status to obtain the pipeline network health assessment results. Subsequently, it analyzes the equipment degradation trend, constructs a trend map to obtain the dynamic performance of equipment status changes, and outputs maintenance strategy recommendations that match the current degradation trend.

[0008] Furthermore, in step S104, the initial position of the terminal is obtained by combining geomagnetic data and positioning technology to form a preliminary coordinate dataset. The position is spatially corrected by integrating pipeline topology constraints and structural conditions to obtain a corrected set of position points. The corrected set of position points is then smoothed using a sliding window technique to obtain a smooth position trajectory. The containerized deployment architecture transmits the trajectory to the diagnostic module to determine whether it conforms to the preset pipeline operating range. If it exceeds the range, an anomaly marker is triggered. For the abnormal trajectory, the corresponding analysis logic is obtained by combining a dynamic loading mechanism to determine the specific anomaly category. If the category can be handled automatically, the diagnostic algorithm parameters are adjusted through an online update mechanism to form updated processing rules, complete automated response processing, and record the operation log.

[0009] The multimodal sensing and adaptive communication method for power pipeline networks described in this application has the advantage of integrating a multi-physics field sensing array composed of a MEMS temperature and humidity composite sensor, a fiber optic strain gauge, and a piezoelectric vibration sensor to achieve synchronous acquisition and fusion analysis of pipeline network temperature, deformation, and mechanical vibration. This improves temperature detection accuracy from the traditional ±3℃ to within ±0.5℃ and reduces humidity response delay from >10 seconds to ≤2 seconds, greatly enhancing the accuracy and real-time performance of monitoring data. The system employs a wide-spectrum ambient light tracking photovoltaic module and an adaptive charge and discharge management circuit, enabling continuous operation for ≥96 hours under no-light conditions, far exceeding the <24-hour endurance of traditional power supply architectures, ensuring stable operation in extreme environments. At the communication level, a dual-mode self-healing mesh network is constructed based on LoRaWAN and power line carrier, combined with dynamic spectrum sensing and segmented compression coding technology, reducing the disconnection rate in shielded areas from 6... The false alarm rate has been reduced from over 5% to below 0.8%, and the communication success rate has been stabilized at over 99.2%, ensuring the reliability of data transmission under complex working conditions. In terms of intelligent diagnosis, the CNN-Transformer hybrid model and digital twin engine deployed on the edge gateway realize feature extraction and spatiotemporal correlation analysis of multi-dimensional data, reducing the false alarm rate caused by the traditional threshold judgment method from 42% to below 5%, and can accurately predict progressive faults, providing decision support for preventive maintenance. The system adopts a combination of three positioning technologies: UWB / geomagnetic / BOS, combined with pipeline topology constraints and sliding window filtering algorithm, to achieve high-precision positioning of ≤0.8 meters. Through containerized deployment, it supports online algorithm iteration and hot-swappable functional modules, significantly improving the system's scalability and lifecycle management capabilities, and comprehensively solving the technical bottlenecks of traditional monitoring devices in terms of perception dimension, environmental adaptability, communication reliability and intelligent diagnosis. Attached Figure Description

[0010] Figure 1 is a flowchart of a multimodal sensing and adaptive communication method for power grids according to this application; Figure 2 is a flowchart of a multimodal sensing and adaptive communication method for power grids according to this application. Detailed Implementation

[0011] As shown in Figures 1-2, the multimodal sensing and adaptive communication method for power pipeline networks described in this application includes: as shown in Figures 1-2, S101, synchronously collecting pipeline network surface temperature, deformation, and mechanical vibration parameters through a multi-physics field sensing array (including MEMS temperature and humidity composite sensor, fiber optic strain gauge, and piezoelectric vibration sensor), combined with a wide-spectrum ambient light tracking photovoltaic module and an adaptive charge and discharge management circuit, for continuous operation for ≥96 hours under extreme weather conditions.

[0012] Further, in step S101, temperature, deformation, and vibration data of the pipeline surface are collected using a multi-physics field sensor array. A pre-established standardized protocol is used to synchronously record various types of data, resulting in an initial multi-dimensional dataset. Based on the initial multi-dimensional dataset, time-series alignment processing is performed on the temperature, deformation, and vibration data to obtain aligned unified timestamp data groups, determining the synchronization between data. If missing values ​​exist in the aligned unified timestamp data groups, linear interpolation is used to fill in the missing data, resulting in a complete time-series dataset. Based on the complete time-series dataset, the changing trends of various parameters on the pipeline surface are analyzed. A support vector machine (SVM) algorithm is used to classify abnormal states and determine if potential risk points exist. If a potential risk point is identified using the SVM algorithm, deep feature extraction is performed on the time period data corresponding to the risk point to obtain the specific distribution pattern of the abnormal parameters and determine the specific type of abnormality. Based on the specific type of abnormality, combined with ambient light tracking data and the operating status of the photovoltaic modules, the charging and discharging management circuit is dynamically adjusted to obtain an optimized energy distribution scheme, completing the system's continuous operation support under extreme weather conditions.

[0013] Specifically, in step S101, the surface temperature, deformation, and mechanical vibration parameters of the pipeline network are synchronously collected through a multi-physics field sensing array. Combined with a wide-spectrum ambient light tracking photovoltaic module and an adaptive charge / discharge management circuit, the system aims to operate continuously for over 96 hours under extreme weather conditions. A MEMS temperature and humidity composite sensor is used to monitor the surface temperature and humidity of the pipeline network in real time. The sensor accuracy is ±0.5℃ for temperature and ±3% for humidity, with a data acquisition frequency of once per second. Noise is filtered out using a built-in low-pass filter algorithm. After obtaining a stable temperature value, trend analysis is performed using historical data. If the temperature change rate exceeds 2℃ per hour, an anomaly warning is triggered, and the system automatically uploads the data to the cloud for in-depth analysis. A fiber optic strain gauge is used to measure the deformation of the pipeline network, with a measurement range of ±5000 micro-strains and a sampling rate of 10Hz. The frequency domain characteristics of the deformation signal are analyzed using a Fourier transform algorithm to extract the dominant frequency component. If the dominant frequency shift exceeds 0.2Hz, it is judged as potential structural damage. The system automatically generates a risk assessment report and correlates it with the temperature data to determine whether the deformation is caused by thermal expansion and contraction. A piezoelectric vibration sensor monitors mechanical vibration with a sensitivity of [missing information]. The vibration signal is collected at a frequency of 50Hz and 100mV / g. Wavelet transform algorithm is used to decompose the vibration signal and extract low-frequency features in the 1-10Hz band. If the amplitude exceeds 0.5g, it is considered abnormal vibration. The system performs spatiotemporal correlation analysis on the vibration and deformation data to comprehensively assess the pipeline network's operating status. A wide-spectrum ambient light tracking photovoltaic module monitors ambient light intensity (range 0-100,000 lux) using a light intensity sensor. Combined with an adaptive charge / discharge management circuit, it dynamically adjusts the charging current, achieving a maximum charging efficiency of 90%. When the light intensity is below 500 lux, the system automatically switches to a low-power mode. The 5000mAh battery ensures continuous power supply for over 96 hours in extreme weather conditions. The power management algorithm optimizes the discharge rate based on remaining power and predicted light conditions, extending equipment operating time. Data from all stages is integrated through an IoT platform. Machine learning models are used to fuse and analyze multi-source data, predicting the probability of pipeline network failures. If the probability exceeds 80%, the system automatically adjusts the sensor sampling frequency to 5 times per second to improve monitoring accuracy, forming a closed-loop feedback mechanism to ensure stable equipment operation in extreme environments.

[0014] In one embodiment, step S101 involves acquiring temperature, deformation, and vibration data using a multiphysics sensing array and performing anomaly detection. The temperature change rate is used to trigger an anomaly warning. The formula is as follows: Temperature Change Rate Calculation:

[0015] R T T represents the rate of temperature change, expressed in °C / h. i This represents the current temperature sample value, in °C; T i-1 This represents the temperature sample value at the previous moment, in °C; Δt This indicates the sampling time interval, in hours (h).

[0016] As shown in Figures 1 and 2, S102, based on the LoRaWAN and power line carrier dual-mode communication architecture, constructs a self-healing Mesh network, avoids channel interference through dynamic spectrum sensing technology, and adopts a segmented compression coding strategy to reduce data transmission power consumption, achieving a communication success rate of 99.2% under complex working conditions.

[0017] Further, in step S102, a communication network foundation is constructed through a dual-mode communication architecture, combining LoRaWAN and power line carrier technology, to obtain a wide-coverage and stable signal transmission channel and determine the initial network connection state. Based on the initial connection state, a self-healing network mechanism is adopted to automatically reconnect to any broken or faulty nodes in the network, resulting in a repaired network topology. It is then determined whether the network has returned to a stable state. Once the network has returned to a stable state, dynamic sensing technology is used to scan the current spectrum environment in real time to obtain spectrum occupancy information and determine if any interfering channels exist. If interfering channels are detected, spectrum management technology is used to reallocate available channels, obtaining an optimized channel configuration scheme and determining whether the current interference can be avoided. Based on the optimized channel configuration scheme, compression coding technology is used to segment the transmitted data during the data transmission process to obtain compressed data packets and determine the integrity of the data packets. The compressed data packets are then sent in batches using a segmentation strategy to obtain power consumption data during the transmission process and determine whether the power consumption optimization target is met. If the power consumption data meets the optimization target, the network operation status is continuously monitored for communication needs under complex operating conditions to obtain real-time communication quality indicators and determine the high reliability performance of the network in different scenarios.

[0018] Specifically, in step S102, when constructing a Mesh network based on a dual-mode communication architecture of LoRaWAN and power line carrier under complex operating conditions, the system first achieves dual-mode communication coordination through the LoRaWAN module and the power line carrier module. The LoRaWAN module operates at a frequency of 470-510MHz, with a signal coverage range of up to 5 kilometers. The power line carrier module utilizes existing power lines to transmit data, with a frequency range of 9-95kHz and a transmission distance of up to 1 kilometer. The system automatically selects the primary communication mode based on channel quality. If the LoRaWAN signal strength is below -120dBm, it switches to the power line carrier mode to ensure communication continuity. Dynamic spectrum sensing technology is used to monitor channel interference in real time. The system scans the spectrum range once per minute to detect interference signal strength. If the interference power exceeds -90dBm, it automatically adjusts to a channel with lower interference through a frequency hopping algorithm, with a frequency hopping interval of 50ms. Simultaneously, it records the interference spectrum distribution to form a historical database for subsequent predictive analysis. The self-healing capability of the LoRaWAN network is achieved through a dynamic routing algorithm between nodes. When communication between nodes is interrupted, the system recalculates the route using Dijkstra's shortest path algorithm, with an average reconnection time controlled within 2 seconds, ensuring the stability of the network topology. Simultaneously, the heartbeat packet transmission frequency between nodes is set to once every 10 seconds for real-time network connectivity monitoring. To reduce data transmission power consumption, the system employs a segmented compression coding strategy, dividing data into 256-byte segments and compressing them using the Huffman coding algorithm, achieving a compression rate of 60%. Before transmission, CRC-16 verification is used to ensure data integrity; if the verification fails, automatic retransmission occurs, with a single retransmission delay not exceeding 100ms. Combined with the LoRaWAN module's standby current of only 1.5uA in low-power mode, energy consumption is effectively reduced. All components are integrated through a cloud platform, and the system automatically analyzes communication logs. If the packet loss rate exceeds 2%, a linkage mechanism of spectrum sensing and routing optimization is triggered, further improving communication reliability and constructing a complete closed-loop control logic.

[0019] In one embodiment, in step S102, channel interference is avoided based on dynamic spectrum sensing technology, where the channel frequency hopping condition is crucial, and the formula is as follows: Channel frequency hopping condition: P 干扰 >-90dBm, where P 干扰 This indicates the detected interference channel power, measured in dBm. When this condition is met, the system automatically performs channel frequency hopping to avoid interference. This formula ensures communication reliability and supports a communication success rate of 99.2% under complex operating conditions.

[0020] As shown in Figures 1-2, in step S103, a lightweight CNN-Transformer hybrid model is deployed on the edge gateway to perform feature extraction and spatiotemporal correlation analysis on multi-dimensional sensor data. A network health assessment index is constructed through a digital twin engine to generate equipment degradation trend maps and maintenance strategy recommendations.

[0021] Further, in step S103, multi-source sensor data is collected through an edge gateway to monitor the pipeline network's operating status in real time, obtain the raw data stream, and determine the completeness of the data collection. For the collected raw data stream, a lightweight convolutional neural network and Transformer hybrid architecture are used for processing to extract spatiotemporal features from the data, resulting in a feature vector set. Based on the extracted feature vector set, correlation analysis technology is implemented to construct a dependency model between data, and it is determined whether the correlation between features reaches a preset threshold. If the feature correlation reaches the preset threshold, the pipeline network's operating status is virtually mapped using a digital twin engine to generate a pipeline network health assessment result. Based on the generated health assessment result, the equipment degradation trend is analyzed, a trend map is constructed, and the dynamic performance of equipment status changes is obtained. Based on the dynamic performance of equipment status changes, corresponding maintenance strategy suggestions are generated, and the matching degree between the strategy and the current degradation trend is determined.

[0022] Specifically, in step S103, during the deployment of a lightweight CNN-Transformer hybrid model on the edge gateway to process multi-dimensional sensor data, the system first collects data from multiple sensors in the pipeline network through the edge gateway, including real-time information on pressure values ​​ranging from 0.5 to 5.0 MPa, flow rate fluctuations ranging from 10 to 500 cubic meters per hour, and temperature data ranging from -20 to 80 degrees Celsius. This data is aggregated to the gateway's built-in preprocessing module every 5 seconds, and the data is normalized using the Z-score normalization algorithm to ensure consistency in the dimensions of different data. The processed data is then input into the lightweight CNN-Transformer hybrid model. The CNN part uses a 3-layer convolutional structure, with each convolutional kernel being 3x3, to extract local features such as pressure mutation points. The Transformer part uses a self-attention mechanism to analyze the long-distance dependencies of the data over time, calculating the correlation weights between each time point, with weight values ​​ranging from 0.1 to 0.9, generating a spatiotemporal feature matrix. The system then uses the extracted feature matrix... The system inputs data into a digital twin engine, which constructs a virtual model of the pipeline network based on historical data. It then calculates health assessment indicators, such as the pipeline wear index, using the formula: Wear Index = 0.3 * Pressure Anomaly Frequency + 0.5 * Flow Fluctuation Rate + 0.2 * Temperature Deviation Value. The resulting index value ranges from 0 to 1; a value exceeding 0.7 indicates a high-risk state. Simultaneously, the engine uses a Long Short-Term Memory (LSTM) network algorithm to predict the deterioration trend over the next 24 hours, with a prediction error controlled within 5%, and generates a trend graph. The graph uses time as the horizontal axis and the deterioration index as the vertical axis, marking key risk points. Based on the assessment indicators and trend graph, the system automatically generates maintenance strategy recommendations. For example, when the deterioration index reaches 0.75, the priority is set to level 1, suggesting inspection within 12 hours. The strategy generation module analyzes historical maintenance records and the current status using a decision tree algorithm to match the optimal solution, such as increasing sensor sampling frequency to once every 2 seconds for high-risk areas. All analysis results and recommendations are stored in a cloud database, forming a complete closed-loop logic from data acquisition to strategy generation, ensuring real-time monitoring and optimization of the pipeline network's operating status.

[0023] In one embodiment, step S103 constructs a pipeline health assessment index using a digital twin engine, where the wear index is used to quantify the equipment degradation trend, and the formula is as follows: Wear index calculation: W = 0.3·F p +0.5·F f +0.2·D t Where W represents the wear index, which is dimensionless and ranges from 0 to 1; F p Indicates the frequency of pressure anomalies, measured in Hz or the number of anomalies; F f D represents the volatility of traffic flow, expressed as a percentage (or a standardized value).t This represents the temperature deviation value, in °C (or standardized value). When W>0.7, the system determines it to be a high-risk state and generates maintenance strategy suggestions. This formula realizes the fusion analysis of multi-dimensional data and accurately assesses the health status of the pipeline network.

[0024] As shown in Figures 1 and 2, S104 uses a combination of three positioning technologies: UWB, geomagnetism, and BOS. Combined with pipeline topology constraint factors and sliding window filtering algorithms, it achieves a terminal positioning accuracy of ≤0.8 meters. Furthermore, it supports online iteration of diagnostic algorithms and hot-swappable functional modules through containerized deployment technology.

[0025] Further, in step S104, by combining geomagnetic data with positioning technology, the initial position information of the terminal in the pipeline network environment is obtained, resulting in a preliminary coordinate dataset. Based on the preliminary coordinate dataset, topological constraints and structural conditions are fused to perform spatial correction on the position information, determining the corrected set of position points. Using a sliding window technique, the corrected set of position points is smoothed over time to obtain a smoothed position trajectory. Through a container deployment architecture, the smoothed position trajectory is transmitted to the diagnostic algorithm module to determine whether the position trajectory conforms to the preset pipeline network operating range. If it exceeds the range, an anomaly marker is triggered. For the position trajectory with the anomaly marker, the corresponding anomaly analysis logic is obtained by combining the dynamic loading mechanism of the functional module to determine the specific category of the anomaly. If the anomaly category falls within the range that can be automatically handled, the parameter configuration of the diagnostic algorithm is adjusted through an online update mechanism to obtain updated processing rules. Based on the updated processing rules, automated response processing is performed for the anomaly category, and it is determined whether the response is completed and the processing log is recorded.

[0026] Specifically, in step S104, during the implementation of the pipeline terminal positioning and diagnosis system, the system first utilizes UWB ultra-wideband technology combined with a geomagnetic sensor for high-precision positioning. UWB base stations are deployed at key nodes of the pipeline network, one every 50 meters, with a base station signal coverage radius of 30 meters, ensuring signal strength between -85dBm and -60dBm. Simultaneously, the geomagnetic sensor collects geomagnetic field strength data every 10 seconds, with a data range of 0.2 to 1.5 Gauss, to assist in correcting UWB signal drift in complex environments. Subsequently, the system inputs the collected UWB distance data and geomagnetic field data into a Kalman filter-based fusion algorithm. The fused positioning error is controlled within 0.5 meters. Combined with pipeline topology constraint factors, a preset pipeline geometric model restricts the positioning result to only along the pipeline path. The pipeline path width in the model is set to 2 meters, and the deviation correction value is controlled within 0.3 meters, further improving the positioning accuracy to below 0.8 meters. Then, the system employs a sliding... A window filtering algorithm processes the positioning data, with a window size of 20 sampling points and a time span of 200 seconds. This filters out instantaneous noise, ensuring a smooth positioning trajectory with a deviation of no more than 0.2 meters. Next, containerization technology is used to encapsulate the positioning and diagnostic algorithms in Docker containers. Each container is allocated 512MB of memory, with CPU usage limited to 30%. Online iteration of algorithm parameters is supported, such as adjusting the filtering window size from 20 to 30 sampling points, with an iteration cycle of once every 24 hours. Hot-swapping of functional modules is also supported, such as dynamically loading new geomagnetic correction modules with a loading time controlled within 5 seconds. The system automatically detects module compatibility, with a compatibility threshold set at 90%, ensuring seamless integration of new modules. Finally, all positioning data and diagnostic results are encrypted and transmitted to cloud storage using the AES-128 encryption algorithm, with transmission latency controlled within 100 milliseconds. This forms a complete automated process from positioning to algorithm updates, providing stable support for pipeline terminal management.

[0027] In one embodiment, in step S104, a sliding window filtering algorithm is used to smooth the positioning data to improve positioning accuracy. The formula is as follows: Sliding window smoothing:

[0028] in, This represents the position coordinate (x-coordinate) after smoothing at time t, in meters (m); x i The x-coordinate represents the original position coordinate at time i, in meters (m); N represents the size of the sliding window, which is dimensionless (N=20 depending on the implementation). This formula is applied to each coordinate axis (x, y, z) to ensure a smooth positioning trajectory and a positioning accuracy of ≤0.8 meters. Combined with pipeline topology constraints, the positioning results are further optimized.

[0029] The present invention also includes non-contact sensing of water leakage risk through a capacitive charge detection module and intelligent diagnosis of tower tilt status using multi-view images, for achieving safety detection and tower structure risk warning with external isolation.

[0030] Furthermore, the capacitive charge detection module indirectly measures the charge content in the water by detecting changes in the alternating electric field in the water medium, which is used to invert the water voltage, achieving non-contact measurement and avoiding the problems of rusting, oxidation, and stains of traditional electrode contacts, thus improving detection safety and reliability. A wide-angle camera deployed at the base of the tower collects multi-view images of the tower, and the lightweight YOLO-Pose attitude estimation model built into the edge gateway is used to calculate the pixel coordinates of key nodes of the tower in real time. Combining the standard design model of the tower with the camera's intrinsic parameters, the PnP algorithm is used to calculate the actual tilt angle of the tower in three-dimensional space. When the tilt angle exceeds a preset threshold (0.5°), a structural risk warning is generated, and it is linked with the digital twin engine in step S103 to update the pipeline health assessment indicators and maintenance strategies.

[0031] Specifically, the above steps are implemented as follows: The system integrates a capacitive charge detection module and a tower tilt visual analysis module. The capacitive detection module operates at a frequency of 1MHz. By detecting the phase and amplitude changes of the alternating electric field in the water (field strength range 0.1-5V / m), it uses an established charge-voltage mapping model (calibration error ≤3%) to calculate the water voltage non-contactly, effectively avoiding electrolysis, rusting, and stain adhesion problems caused by direct electrode contact. The detection voltage range is 0-300V with an accuracy of ±2V. When the detected voltage value exceeds the 36V safety voltage, a leakage alarm is immediately triggered. The tower tilt detection uses 2-megapixel wide-angle cameras (120° field of view) installed on both sides of the tower base to acquire images at a frequency of 1 frame per minute. The image data is compressed using JPEG and transmitted to the edge gateway, where a lightweight YOLO-Pose module is loaded. The system (model size approximately 8MB) identifies key structural points of the tower body and crossarms, outputting their pixel coordinates in the image. Combining the known 3D tower model dimensions (30 meters high) with camera calibration parameters (focal length, distortion coefficient), the system uses the Perspective-n-Point (PnP) algorithm to calculate the tower's spatial attitude angles (pitch, yaw) relative to the vertical direction. The attitude angle calculation resolution reaches 0.1°. During continuous monitoring, if the tilt angle change exceeds 0.5° or the absolute tilt angle is greater than 2°, it is determined to be a structural risk, automatically generating a warning event. This event is deeply integrated with the S103 digital twin engine, serving as an important evaluation dimension to update the tower's health status and trigger inspection work orders.

[0032] In one embodiment, the tilt angle of the tower is calculated using the PnP algorithm. The core of this algorithm is to solve for the rotation matrix R between the camera coordinate system and the tower model coordinate system. The formula is briefly expressed as follows:

[0033] Where (u,v) are the pixel coordinates of the key points in the image; K is the camera intrinsic parameter matrix; [R|t] is the rotation-translation matrix (extrinsic parameter) to be solved; (X,Y,Z) are the world coordinates of the key points in the coordinate system of the tower 3D model; s is the scale factor; Euler angles can be extracted from the solved rotation matrix R, and then the tilt angle of the tower relative to the vertical direction can be obtained. When the tilt angle θ>0.5°, a risk warning is triggered.

[0034] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A multimodal sensing and adaptive communication method for power grids, characterized in that, include: S101. Simultaneously collect temperature, deformation, and mechanical vibration parameters of the pipeline surface using a multi-physics field sensor array, and combine this with a wide-spectrum ambient light tracking photovoltaic module and an adaptive charge / discharge management circuit; S102. Construct a dual-mode communication architecture based on LoRaWAN and power line carrier to obtain a Mesh network, utilize dynamic spectrum sensing technology to avoid channel interference, and employ a segmented compression coding strategy for data transmission; S103. Deploy a lightweight CNN-Transformer hybrid model on the edge gateway to perform feature extraction and spatiotemporal correlation analysis on the multi-dimensional sensing data from the multi-physics field sensor array, construct pipeline health assessment indicators through a digital twin engine, and obtain equipment degradation trend maps and maintenance strategy suggestions; S104. Employ a combination of UWB / geomagnetic / BOS positioning technologies, combined with pipeline topology constraints and a sliding window filtering algorithm, for high-precision positioning of pipeline terminals, and support online iteration of diagnostic algorithms and hot-swappable functional modules through containerized deployment technology.

2. The multimodal sensing and adaptive communication method for power grids according to claim 1, characterized in that, In step S101, the synchronous acquisition of parameters via a multi-physics sensing array includes: acquiring temperature, deformation, and vibration data using a MEMS temperature and humidity composite sensor, an optical fiber strain gauge, and a piezoelectric vibration sensor, respectively; performing time-series alignment and missing value interpolation on the acquired data to obtain a complete time-series dataset; performing parameter trend analysis based on the time-series dataset to identify potential risk points, and performing deep feature extraction on abnormal periods to determine the anomaly type; and dynamically adjusting the adaptive charge and discharge management circuit by combining ambient light tracking data and the photovoltaic module's operating status to obtain an optimized energy distribution scheme.

3. The multimodal sensing and adaptive communication method for power grids according to claim 1, characterized in that, In step S102, the construction of the Mesh network includes: automatically switching the primary communication mode between LoRaWAN and power line carrier based on real-time channel quality; automatically repairing network breakpoints or faulty nodes in the network through a dynamic routing algorithm to maintain network topology stability; the dynamic spectrum sensing technology includes real-time scanning of the spectrum environment and performing channel frequency hopping when interfering channels are detected; the segmented compression coding strategy includes compressing and segmenting the transmitted data and sending it in batches to optimize power consumption.

4. The multimodal sensing and adaptive communication method for power grids according to claim 1, characterized in that, In step S103, feature extraction and spatiotemporal correlation analysis of multi-dimensional sensor data includes: extracting local features of the data through the convolutional neural network part of the lightweight CNN-Transformer hybrid model, and analyzing the long-distance dependence of the data on the time series through the self-attention mechanism of the Transformer part to generate a spatiotemporal feature matrix; the construction of pipeline health assessment indicators through the digital twin engine includes: calculating a comprehensive assessment indicator reflecting the health status of the pipeline network based on the spatiotemporal feature matrix and historical data.

5. The multimodal sensing and adaptive communication method for power grids according to claim 1, characterized in that, In step S104, the high-precision positioning of the pipeline terminal includes: fusing UWB positioning data and geomagnetic sensor data to obtain the initial position information of the terminal; using the pipeline topology constraints to spatially correct the initial position information; and using the sliding window filtering algorithm to perform temporal smoothing processing on the corrected position data to generate a smooth position trajectory.

6. The method according to any one of claims 1-5, characterized in that, The method improves the temperature detection accuracy to within ±0.5℃, reduces the humidity response delay to within 2 seconds, ensures the system can work continuously for no less than 96 hours under no-light conditions, achieves a communication success rate of no less than 99.2% under complex working conditions, and achieves terminal positioning accuracy within 0.8 meters.

Images