Complex pipe network delivery difference hierarchical monitoring method based on topological nodes

Abstract

The present application relates to the technical field of monitoring, and particularly relates to a complex pipe network leakage differential hierarchical monitoring method based on a topological node, which comprises the following steps: generating an orthogonal coded aperture basis function based on a three-dimensional topological structure of a gas pipe network, constructing a first RIS phased matrix sequence for wide-area polling scanning, using the first RIS phased matrix sequence to control a RIS reflection array to perform beamforming on each sensing node along the pipe network, and polling and collecting multi-dimensional sensing data of each sensing node. In the beam locking state, constraint independent component analysis based on typical time sequence characteristics is input, high-frequency sensing data is deeply analyzed, environmental interference is stripped, and source signal characteristics related to leakage are extracted. This kind of hierarchical differential monitoring mechanism from wide-area planar scanning to local point-like deep detection reduces the false alarm rate, improves the accuracy and timeliness of gas pipe network leakage event determination, and helps to ensure the safe operation of the pipe network.

Description

Technical Field

[0001] This application belongs to the field of monitoring technology, and in particular relates to a hierarchical monitoring method for complex pipeline network transmission differentials based on topology nodes. Background Technology

[0002] As a crucial energy transmission infrastructure, the safe and stable operation of urban gas pipeline networks is vital to public safety. Typically, numerous sensor nodes are deployed along the pipeline network to construct an Internet of Things (IoT) monitoring system, aiming to achieve real-time monitoring of the network's operational status. However, limited by the nodes' battery capacity and power consumption, sensor nodes cannot maintain high-frequency sampling and omnidirectional broadcast communication for extended periods. In the early stages of minor leaks or anomalies within the pipeline network, poor communication link quality can easily lead to delays and packet loss in multi-dimensional sensor data transmission. Traditional static monitoring methods struggle to balance low-power polling across the entire network with accurate detection of localized sudden anomalies, resulting in significant energy waste and channel congestion. This makes it difficult to meet the dual requirements of high reliability and high real-time performance for pipeline safety monitoring.

[0003] In related technologies, for example, Chinese patent document with publication number CN116633423B discloses a low-orbit satellite-assisted communication method based on a reconfigurable smart surface. It establishes an expression for the misalignment fading parameter of the signal based on the effective receiving area of ​​the RIS and the coverage area of ​​the ground communication equipment's transmitting antenna beam at the RIS, and defines the misalignment fading coefficient when the effective receiving area of ​​the RIS and the coverage area of ​​the ground communication equipment's transmitting antenna beam at the RIS are misaligned.

[0004] RIS technology can reshape the radio electromagnetic propagation environment by programming and adjusting the phase shift and amplitude of a large reflective array unit. It can also enhance the signal strength of target nodes and expand network coverage through directional beamforming technology. However, existing RIS scheduling strategies mostly adopt fixed beam or uniform sweep modes, which cannot be deeply integrated with the complex three-dimensional topology of gas pipeline networks and the multi-dimensional spatiotemporal anomalies within the network. This results in low wide-area scanning efficiency and difficulty in effectively adjusting the monitoring focus. When a suspected local leak is detected, existing systems usually lack beam optimization and locking mechanisms based on communication quality feedback, and have limited ability to maintain stable communication in time-varying obstruction environments. Existing technologies mostly utilize the raw signals returned by nodes or use simple threshold judgment methods, and lack the means to deeply separate and extract features from multi-node high-frequency mixed sensor data. They cannot effectively separate the source signal features related to the leak from complex background noise and multipath interference, which can easily lead to false alarms or missed alarms of gas leak events. Summary of the Invention

[0005] To address the problems of poor matching between wide-area scanning strategies and complex topologies and abnormal states in existing gas pipeline monitoring systems, difficulty in stably locking local communication links in dynamic environments, and low accuracy in extracting leakage features in complex background noise, this invention proposes a hierarchical monitoring method for complex pipeline transmission differentials based on topology nodes.

[0006] This invention provides a method for hierarchical monitoring of differential transmission in complex pipeline networks based on topology nodes, comprising: S1: generating orthogonal encoded aperture basis functions based on the three-dimensional topology of the gas pipeline network, constructing a first RIS phased array matrix sequence for wide-area polling scanning, using the first RIS phased array matrix sequence to control the RIS reflection array to perform beamforming on each sensing node along the pipeline network, polling and collecting multidimensional sensing data from each sensing node, calculating the Mahalanobis distance between the multidimensional sensing data and the baseline to obtain the differential spatiotemporal anomaly score, and adjusting the RIS wide-area polling scanning strategy for the next cycle based on the differential spatiotemporal anomaly scores of all sensing nodes; S2: when the differential spatiotemporal anomaly score of any sensing node exceeds the first-level threshold, the sensing node position is... The solution is converted into a second RIS phased matrix. The second RIS phased matrix is ​​used to perform position-oriented beamforming and wake up the sensor node cluster in the location neighborhood to enter the high-frequency sampling mode. The communication quality of the sensor node cluster is used as an indicator to apply perturbation to the second RIS phased matrix and iterative optimization is performed based on feedback until the communication quality converges. The iteratively optimized second RIS phased matrix is ​​locked to achieve beam locking. S3: In the beam-locked state, a constraint separation matrix is ​​constructed based on the preset typical time series characteristics of gas leakage. Constrained independent component analysis is performed on the collected multidimensional sensor data to extract the leakage source signal characteristics. When the signal power value of the leakage source signal characteristics exceeds the second-level threshold, it is determined to be a gas leakage event. First, by dynamically allocating wide-area polling resources through differential anomaly scores, the contradiction between low power consumption across the entire network and real-time anomaly detection is resolved. Second, when a suspected primary anomaly is detected, a directional beam is used to precisely wake up a local neighborhood cluster for high-frequency sampling, and a unique perturbation iteration mechanism is used to lock the beam, overcoming the pain points of unstable communication links and easy data loss during sudden anomalies in complex pipeline environments. Finally, in the beam-locked state, constrained independent component analysis is used to perform deep feature stripping on high-frequency concurrent data, effectively eliminating the influence of complex industrial background noise and multimodal crosstalk. This method significantly reduces the false alarm rate of the system and significantly improves the accuracy and response time of determining minor leak events in gas pipeline networks.

[0007] Furthermore, the differential spatiotemporal anomaly score is obtained by calculating the Mahalanobis distance between the multidimensional sensor data and the baseline. This includes: collecting multidimensional sensor data from each sensor node under normal operating conditions for consecutive historical periods; calculating the arithmetic mean of each feature dimension to construct a mean baseline vector; calculating the covariance matrix using the multidimensional sensor data from the historical periods, and inverting the covariance matrix to obtain the accuracy matrix; subtracting the mean baseline vector from the currently collected multidimensional sensor data to obtain the corresponding difference vector; calculating the transpose of the difference vector, the product of the accuracy matrix and the difference vector, and performing a square root operation on the product result, using the obtained value as the differential spatiotemporal anomaly score of the sensor node at the current sampling time. This method introduces the Mahalanobis distance algorithm and combines the mean baseline and covariance accuracy matrix constructed from consecutive historical periods for anomaly assessment, effectively eliminating the dimensional differences between multidimensional sensor parameters, while deeply considering the inherent coupling correlation between multiple variables such as temperature, pressure, and concentration. It can more sensitively and objectively identify extremely weak operating condition anomalies in the early stages of pipeline network operation, reducing the false alarm rate caused by single environmental sudden changes.

[0008] Furthermore, the RIS wide-area polling scan strategy for the next cycle is adjusted based on the differential spatiotemporal anomaly scores of all sensor nodes. This includes: extracting the differential spatiotemporal anomaly scores of each sensor node within a preset historical period; constructing a time series of the differential spatiotemporal anomaly scores for each sensor node; calculating the average value of the time series composed of these differential spatiotemporal anomaly scores to obtain the corresponding anomaly evaluation score; sorting all sensor nodes in descending order according to the anomaly evaluation scores; generating a polling scan queue for the next cycle; setting the areas of the top-ranked sensor nodes in the scan queue as high-priority scan areas, increasing beam dwell time and polling frequency; setting the areas of the bottom-ranked sensor nodes in the scan queue as low-priority scan areas, reducing beam dwell time and polling frequency; and generating underlying control commands to update the scan task list of the RIS reflective array. This method divides monitoring nodes into descending queues based on historical anomaly assessment scores, dynamically tilting beam dwell time and polling frequency towards high-priority areas. Without increasing or even reducing the overall communication power consumption of the system, it enables limited monitoring resources to focus on potentially high-risk areas, significantly improving the early warning response speed and wide-area scanning efficiency for suspected leaking pipe sections.

[0009] Furthermore, the wake-up process involves activating a cluster of sensor nodes within the neighborhood to enter a high-frequency sampling mode. This includes: defining a spatial neighborhood with a preset distance as the radius, centered on the location of the sensor node whose differential spatiotemporal anomaly score exceeds the first-level threshold; extracting the hardware communication address set of all sensor nodes within the neighborhood; and broadcasting a wake-up control signal to the corresponding sensor nodes via a directional beam. Upon receiving the wake-up control signal, the sensor node switches from a low-power sleep / standby state to an active transmission state, increasing the sampling frequency and uplink transmission frequency of the multidimensional sensor data to a preset high-frequency sampling threshold, and transmitting the collected multidimensional sensor data using an orthogonal multiple access mode. This method, by defining the neighborhood based on spatial distance and broadcasting the wake-up signal in a directional manner, only switches the cluster around the anomaly source to active high-frequency transmission. Combined with the orthogonal multiple access mode, this ensures both the acquisition density of multidimensional data under local anomaly conditions and avoids channel collisions and multiple access interference caused by concurrent transmission of dense high-frequency signals, thus guaranteeing data integrity under sudden situations.

[0010] Furthermore, the sensor node position is calculated into a second RIS phased array matrix, including: acquiring the three-dimensional coordinates of the abnormal sensor node whose differential spatiotemporal anomaly score exceeds the first-level threshold, the three-dimensional coordinates of the base station receiving antenna, and the three-dimensional coordinates and normal vector of the RIS array; calculating the incident angle from the abnormal sensor node to the RIS array and the reflection angle from the RIS array to the base station receiving antenna based on the three-dimensional coordinates; calculating the initial phase compensation value required for directional beamforming of each reflection element of the RIS array based on the generalized Snell's law; and combining the initial phase compensation values ​​of all reflection elements into a two-dimensional phase control array to generate a second RIS phased array matrix for controlling the beam pointing of the RIS reflection array. This method directly combines the three-dimensional coordinate spatial relationship between the abnormal node, the receiving base station, and the reflection array, and uses the generalized Snell's law to quickly and accurately calculate the initial phase compensation value required for all reflection elements. This allows for the extremely rapid construction of the initial matrix for directional beamforming at the first moment of detecting a primary anomaly, greatly shortening the reaction time for monitoring center shift and high-gain communication link reconstruction.

[0011] Furthermore, a perturbation is applied to the second RIS phased array matrix, including: superimposing a random phase perturbation following a preset distribution onto the phase values ​​of each reflection element of the currently used second RIS phased array matrix, generating a candidate second RIS phased array matrix with the perturbation applied, and sending it to the RIS reflection array. This method, by superimposing a random phase perturbation following a preset distribution onto the base phase, provides rich and reasonable exploration samples for subsequent closed-loop channel optimization, avoids local deadlock in the beam optimization process, and enhances the system's adaptive adjustment capability in scenarios involving moving obstructions and time-varying electromagnetic environments.

[0012] Furthermore, iterative optimization is performed based on feedback until communication quality converges, and beamlocking is achieved by locking the first RIS phased matrix sequence. This includes: synchronously acquiring the uplink transmission link signal-to-noise ratio (SNR) of each sensor node in the sensor node cluster with applied perturbations for the candidate second RIS phased matrix; calculating the average SNR and extracting the lowest node SNR as the communication quality feedback value; calculating the difference between the current average SNR and the average SNR of the previous iteration cycle; if the difference is greater than 0, retaining the random phase perturbation and updating the candidate second RIS phased matrix to the currently used matrix; if the difference is not greater than 0, discarding the random phase perturbation and restoring the matrix of the previous cycle; repeating the perturbation and feedback process until the improvement in the average SNR is lower than the preset convergence threshold or the maximum number of iterations is reached, and locking the first RIS phased matrix sequence. This method uses the average signal-to-noise ratio and the minimum signal-to-noise ratio of the uplink of the cluster nodes as dual constraint indicators for feedback evaluation. By filtering and retaining positive gain perturbations for closed-loop iteration, it ensures that the system can always focus the highest quality electromagnetic energy on high-risk areas in complex pipeline field environments until the communication quality converges, thereby establishing and maintaining a highly reliable and delay-free critical data transmission channel in harsh environments.

[0013] Furthermore, the leakage source signal features are extracted, including: extracting preset typical time-series features of gas leakage as prior reference signals; constructing a contrast function containing an independence metric and a reference signal similarity constraint; inputting the prior reference signals into the constraints of the contrast function; maximizing the contrast function using Newton's iteration method to calculate the constraint separation matrix; and using the constraint separation matrix to perform linear transformation and demixing on the mixed multidimensional sensor data to separate the target signal component with the highest correlation to the prior reference signal, which is then used as the leakage source signal feature. This method introduces the prior reference signal reflecting typical gas leakage characteristics as a similarity constraint into the objective function, and uses Newton's iteration-optimized constraint independent component analysis for linear mapping. This enables precise and targeted extraction of pure source leakage acoustic and operating condition features from mixed high-frequency sensor data filled with environmental mechanical noise and fluid multipath interference, fundamentally improving the purity and reliability of feature extraction in complex backgrounds.

[0014] Furthermore, based on the three-dimensional topology of the gas pipeline network, an orthogonal encoded aperture basis function is generated, including: acquiring spatial data of the gas pipeline network containing latitude, longitude, and elevation information, converting it into a unified metric three-dimensional rectangular coordinate system, and establishing a three-dimensional directed graph model of the pipeline network; using the least second power of the number of pipeline network nodes as the encoding order. ,generate The method utilizes a two-dimensional extended Hadamard matrix, extracting its row vectors as orthogonal encoded aperture basis functions. It leverages real 3D topological data of gas pipeline networks, including latitude, longitude, and elevation, to establish a directed graph model. The method then uses a two-dimensional extended Hadamard orthogonal matrix to extract basis functions, enabling RIS wide-area beamforming to perfectly match the complex 3D physical spatial distribution of specific pipeline networks. This significantly improves coverage matching and hardware instruction execution efficiency during the planar polling scanning phase.

[0015] Furthermore, the signal power value of the leakage source signal characteristics is calculated, including: performing a Discrete Fourier Transform on the extracted one-dimensional array of leakage source signal characteristics to calculate the power spectral density, and numerically integrating and summing the power spectral density array along a preset analysis frequency band to obtain the normalized signal power value. This method converts the extracted leakage source signal to the frequency domain through Discrete Fourier Transform and performs numerical integration of the power spectrum for a specific analysis frequency band, effectively filtering out broadband background noise outside the target frequency band. This makes the final signal power value used for leakage identification more stable and objective, improving the scientific rigor and anti-interference capability of the system when making the final cut-off decision.

[0016] Beneficial effects: First, by calculating the Mahalanobis distance of multidimensional data to obtain differential spatiotemporal anomaly scores, early weak anomalies can be identified, and subsequent polling scanning strategies can be updated accordingly to optimize system resource allocation. After detecting primary anomalies, the beam is directed to the anomaly node region, waking up the neighboring cluster to carry out high-frequency sampling, and beam locking is achieved using a perturbation iteration mechanism, establishing a high-quality communication link in local high-risk areas. Second, under beam-locked conditions, constrained independent component analysis based on typical time series characteristics is input to perform in-depth analysis of high-frequency sensor data, removing environmental interference and extracting source signal features related to leaks. This hierarchical differential monitoring mechanism, from wide-area area scanning to local point-like in-depth detection, reduces the false alarm rate, improves the accuracy and timeliness of determining gas pipeline network leak events, and helps ensure the safe operation of the pipeline network. Attached Figure Description

[0017] Figure 1 This is a flowchart of the monitoring method in an embodiment of the present invention; Figure 2 This is a schematic diagram of differential evaluation and tracking of abnormal nodes in an embodiment of the present invention; Figure 3 This is a schematic diagram of polling scan queue scheduling in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the performance comparison of comprehensive scenario testing in an embodiment of the present invention. Detailed Implementation

[0018] This disclosure provides a hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes, such as... Figure 1 As shown, it includes the following steps: S1. Based on the three-dimensional topology of the gas pipeline network, an orthogonal encoded aperture basis function is generated. A first RIS phased array matrix sequence for wide-area polling scanning is constructed. The first RIS phased array matrix sequence is used to control the RIS reflection array to perform beamforming on each sensing node along the pipeline network. Multidimensional sensing data of each sensing node is collected in polling. The Mahalanobis distance between the multidimensional sensing data and the baseline is calculated to obtain the differential spatiotemporal anomaly score. The RIS wide-area polling scanning strategy for the next cycle is adjusted based on the differential spatiotemporal anomaly scores of all sensing nodes.

[0019] Spatial data of the gas pipeline network geographic information system, containing latitude, longitude, and elevation information, was read using the Open3D library. The latitude, longitude, and elevation information were then converted into a unified metric three-dimensional rectangular coordinate system using a preset map projection or a local east-north-sky coordinate system transformation method. A three-dimensional directed graph model of the pipeline network was then established, with the encoding order being a power of at least two that is not less than the total number of pipeline nodes. ,generate The Hadamard matrix is ​​of order 1, and its row vectors are extracted as orthogonal coding aperture basis functions.

[0020] To enable the one-dimensional encoding of the Hadamard matrix to be mapped to the RIS two-dimensional reflection hardware array, the 256×256 physical RIS reflection units are divided into rows and columns in spatial order. A logical aperture subarray, each logical aperture subarray containing several adjacent physical reflection units; when When =128, it can be divided into 16×8 logic aperture subarrays, when When the value is 256, it can be divided into 16×16 logical aperture subarrays. For any Hadamard row vector, the k-th encoded element is mapped to the k-th logical aperture subarray, and all physical reflection units within the logical aperture subarray are controlled to adopt the same basic phase state. Thus, the one-dimensional orthogonal encoded row vector can be expanded to generate a two-dimensional orthogonal encoded phase mask matrix of size 256×256, thereby completing the feasible mapping between the orthogonal encoded aperture basis function and the physical RIS reflection array. In wide-area polling scanning, the orthogonal encoded aperture basis function is not used as a beam pointing matrix alone, but is superimposed as an orthogonal encoded phase mask on the directional linear phase gradient matrix of each node, which is pre-calculated based on the azimuth, elevation, and operating wavelength of each sensing node relative to the RIS array. After modulo 2π processing and discretization, a first RIS phased matrix sequence corresponding one-to-one with the spatial position of each sensing node is generated.

[0021] Elements 1 and -1 in the basis function are mapped to the 0-phase and π-phase states of the corresponding logical aperture subarray and its internal physical reflection unit, respectively, serving as orthogonal coded phase masks. These orthogonal coded phase masks are superimposed with the directional linear phase gradient matrices corresponding to each sensing node. After modulo 2π processing and discrete quantization, multiple two-dimensional first RIS phased array matrix sequences containing node pointing information are generated. These sequences are arranged according to the spatial distribution of nodes in the pipeline network model. The main control unit sends the phased array matrix sequences to the RIS field-programmable gate array controller via high-speed SPI, parallel bus, or industrial Ethernet interface. Commonly used phased array matrices can be pre-cached in the controller's local memory and quickly retrieved through matrix indexing or incremental phase configuration. The controller uses a lookup table method to control the switching on and off of diodes, enabling the incident electromagnetic wave to form a high-gain beam at different spatial angles, which is then sequentially aligned with each node along the pipeline network using a time-slice round-robin scheduling algorithm.

[0022] During the regular wide-area polling phase, the multidimensional sensing data uploaded by the sensor nodes is uniformly defined as a multidimensional operating condition feature vector, which includes at least methane concentration, pipeline pressure, ambient temperature, and gas velocity. For pipe sections with upstream and downstream metering nodes, the difference or difference rate between the input mass flow rate and the output mass flow rate is calculated based on the gas velocity, pressure, temperature, and pipe diameter parameters of the inlet and outlet nodes of adjacent pipe sections, according to a preset gas state conversion model. This difference or difference rate is then incorporated into the multidimensional sensing data as a pipe section output difference feature dimension. The current multidimensional operating condition feature vector and the multidimensional mean baseline vector calculated by arithmetic mean under historical normal conditions, along with the inverse matrix of the corresponding dimension covariance matrix, are input to calculate a scalar value as the differential spatiotemporal anomaly score. The differential spatiotemporal anomaly scores of all nodes are input into the rule-based sorting scanning strategy update unit. The nodes are sorted in descending order according to the anomaly evaluation scores within a preset historical period, and the beam dwell time and polling frequency configuration of each node in the next period are output accordingly, completing the scanning strategy adjustment.

[0023] In one possible implementation, multidimensional sensing data from each sensing node is collected in a polling manner. This multidimensional sensing data includes node operating characteristics and pipeline transport differential characteristics converted from upstream and downstream nodes of adjacent pipeline segments. The Mahalanobis distance between the multidimensional sensing data and the baseline is calculated to obtain a differential spatiotemporal anomaly score, including: Multidimensional sensing data from each sensor node under normal operating conditions for consecutive historical periods is collected. The arithmetic mean of each feature dimension is calculated to construct the mean baseline vector. The covariance matrix is ​​calculated using the multidimensional sensing data from the historical periods, and the inverse of the covariance matrix is ​​used to obtain the precision matrix. The mean baseline vector is subtracted from the currently collected multidimensional sensing data to obtain the corresponding difference vector. The transpose of the difference vector, the product of the precision matrix and the difference vector are calculated, and the square root of the product is performed. The obtained value is used as the differential spatiotemporal anomaly score of the sensor node at the current sampling time.

[0024] A historical reference period is set, for example, a continuous 14-day period, as the baseline for normal operation. Multidimensional sensor datasets generated by each sensor node are collected at 15-minute sampling intervals. Each multidimensional sensor data vector... It includes at least four key parameters: methane concentration, pipeline pressure, ambient temperature, and gas flow rate. For pipeline segments with upstream and downstream metering nodes, the segment's transport differential characteristic dimension can also be further incorporated. For example, by extracting the arithmetic mean of the characteristic dimensions from 1344 samples collected over the aforementioned 14 days, a similar model can be constructed when only four operating condition characteristics—methane concentration, pipeline pressure, ambient temperature, and gas flow rate—are used. 4-dimensional mean baseline vector Based on 1344 historical samples, a 4×4 covariance matrix was calculated. When the pipeline segment transport difference characteristics were further incorporated, the dimensions of the mean baseline vector and the covariance matrix expanded synchronously with the feature dimensions. The covariance matrix was then inverted, with optional values ​​added during the process. The diagonal regularization term is used to prevent singularities, thereby obtaining the precision matrix S of the corresponding dimension.

[0025] During the real-time monitoring phase, when multi-dimensional sensor data at the current moment is collected... For example, if the methane concentration suddenly rises to 0.08% and the pressure drops slightly to 0.38 MPa, then... The subtraction operation generates the corresponding difference vector d. Matrix multiplication is then performed to calculate the transpose of the difference vector. The product of the precision matrix S and the difference vector d, and the square root of the result scalar, is calculated using the formula: The obtained value is 3.45, which is taken as the differential spatiotemporal anomaly score of the current sampling time of the sensing node. This calculation process eliminates the dimensional differences of multi-source parameters and takes into account the intrinsic correlation between variables, such as... Figure 2As shown, the curves of the differential spatiotemporal anomaly score of a typical sensor node change with the sliding time window are illustrated. When a potential leak occurs in the pipeline network, the score will rise rapidly and exceed the preset first-level threshold. Specifically, a typical sensor node refers to a sensor node whose calculated differential spatiotemporal anomaly score exceeds the preset first-level threshold scalar. When an abnormal operating condition or minor leak occurs at a certain sensor node in the pipeline network, key parameters such as the methane concentration and pipeline pressure of that node will change abruptly, causing the calculated differential spatiotemporal anomaly score to rise rapidly and exceed the first-level threshold. As a result, the system will identify and determine it as a typical sensor node that needs to be focused on directional beam locking.

[0026] In one possible implementation, the RIS wide-area polling strategy for the next cycle is adjusted based on the differential spatiotemporal anomaly scores of all sensing nodes, including: Extract the differential spatiotemporal anomaly scores of each sensor node within a preset historical period, construct a time series of the differential spatiotemporal anomaly scores for each sensor node, calculate the average value of the time series composed of these differential spatiotemporal anomaly scores, and obtain the corresponding anomaly evaluation score. Sort all sensor nodes in descending order according to the anomaly evaluation scores, generate the polling scan queue for the next period, set the area where the top-ranked sensor nodes in the scan queue are located as high-priority scan areas, increase the beam dwell time and increase the polling frequency, set the area where the bottom-ranked sensor nodes in the scan queue are located as low-priority scan areas, reduce the beam dwell time and decrease the polling frequency, and generate low-level control commands to update the scan task list of the RIS reflector array.

[0027] Set a preset historical period as a sliding time window, for example, the duration of the sliding time window is the past 2 hours, containing about 12 sampling points. Extract the time series of differential spatiotemporal anomaly scores of 100 sensing nodes along the pipeline within this sliding time window. Calculate the comprehensive anomaly assessment score of each node by calculating the arithmetic mean or distance-weighted average. Perform quick sort or merge sort operations to sort all 100 nodes in descending order of anomaly assessment scores, and construct the polling scan queue for the next period.

[0028] Based on the sorting results, the top 10% of sensor nodes in the polling scan queue are assigned to the high-priority scan area. At the strategy parameter level, the RIS beam dwell time is extended from the default 50ms to 150ms, and the polling sampling frequency is increased from once every 10 minutes to once every 2 minutes to obtain denser monitoring data in abnormal areas. For the low-priority security scan area containing the bottom 50% of sensor nodes, the beam dwell time is shortened to 20ms, and the polling frequency is reduced to once every 30 minutes. These dwell time and frequency parameters are converted into timing scheduling matrices and beam switching cycle configuration parameters in the underlying control instructions, and written and updated to the RIS reflective array controller's register task list via the communication bus, realizing the reconstruction of the wide-area polling task, such as... Figure 3 As shown, high-priority abnormal nodes have a longer dwell time to enhance monitoring data collection, while low-priority nodes have a shorter dwell time to reduce system power consumption, thus achieving optimized allocation of communication resources.

[0029] S2, when the differential spatiotemporal anomaly score of any sensor node exceeds the first-level threshold, the sensor node position is solved into a second RIS phased matrix. The second RIS phased matrix is ​​used to perform position-oriented beamforming, and the sensor node cluster in the location neighborhood is awakened to enter the high-frequency sampling mode. The communication quality of the sensor node cluster is used as an indicator to apply perturbation to the second RIS phased matrix, and iterative optimization is performed based on feedback until the communication quality converges. The iteratively optimized second RIS phased matrix is ​​locked to achieve beam locking.

[0030] In the main control server memory, the differential spatiotemporal anomaly scores of each sensor node are compared with the first-level threshold scalar. When an out-of-limit target node is detected, the azimuth and elevation angles of the line connecting the three-dimensional coordinates of the target node to the center of the RIS array are calculated using a spatial geometric ray tracing algorithm. The azimuth and elevation angles are then substituted into the phased array pattern function to calculate the required target phase gradient. The target phase gradient is represented as a discrete phase two-dimensional matrix supported by the array using a least-squares fitting algorithm, and is sent to the underlying hardware as the second RIS phased array matrix for directional beamforming. The KDTree algorithm is used to target the node. Centered on the coordinates, a neighborhood search is performed within a preset spatial radius to obtain a list of neighboring nodes. A media access control layer wake-up frame containing a high-frequency interrupt command is multicast to the nodes in the list via the narrowband IoT downlink control channel, which raises the sampling level of the operating condition channels such as methane concentration, pipeline pressure, ambient temperature, and gas flow rate to 10 Hz. Simultaneously, the acoustic emission acquisition channel is enabled in the sensor nodes configured with the acoustic emission leakage identification channel. The sampling frequency of the acoustic emission acquisition channel is set to no less than 200 kHz to meet the Nyquist sampling requirements of the acoustic emission characteristic frequency band from 20 kHz to 80 kHz.

[0031] The uplink signal-to-noise ratio (SNR) after demodulation by the baseband chip at the hardware layer is used as a communication quality evaluation metric. A random perturbation feedback optimization algorithm is employed to iteratively optimize the second RIS phased array matrix. In each iteration, a random phase perturbation matrix following a preset distribution is generated, resulting in a perturbed candidate second RIS phased array matrix. This perturbed candidate second RIS phased array matrix is ​​then sent to the controller, and the corresponding average SNR feedback value is collected. When the average SNR corresponding to the candidate second RIS phased array matrix is ​​higher than the average SNR of the previous cycle, the random phase perturbation is retained, and the current phased array matrix is ​​updated. When the average SNR corresponding to the candidate second RIS phased array matrix is ​​not higher than the average SNR of the previous cycle, the random phase perturbation is discarded, and the second RIS phased array matrix used in the previous cycle is restored. When the average SNR improvement for several consecutive iterations is less than a preset convergence threshold, or when the maximum number of iterations is reached, optimization is stopped, and the second RIS phased array matrix optimized in the current iteration is locked, achieving beam locking.

[0032] In one possible implementation, waking up the cluster of sensor nodes in the location neighborhood to enter a high-frequency sampling mode includes: Centered on the location of the sensor node whose differential spatiotemporal anomaly score exceeds the first-level threshold, a spatial neighborhood is defined with a preset distance as the radius. The hardware communication address set of all sensor nodes within the range is extracted. A wake-up control signal is broadcast to the sensor node corresponding to the hardware communication address set through a directional beam. After receiving the wake-up control signal, the sensor node switches from a low-power sleep standby state to an active transmission state, and increases the sampling frequency and uplink transmission frequency of the multidimensional sensor data to a preset high-frequency sampling threshold, transmitting the collected multidimensional sensor data in orthogonal multiple access mode.

[0033] When the anomaly score calculated by a certain sensing node exceeds the set first-level threshold, the edge computing unit immediately uses the three-dimensional coordinates of the anomaly node. Centered on a sphere, with a preset spatial distance as the search radius, the Kd-tree spatial indexing algorithm is used to retrieve the neighboring node groups within this spherical neighborhood, and extract their MAC address sets registered at the network layer, for example, containing 8 surrounding homogeneous nodes. The base station generates a specific hardware layer broadcast frame, which is combined with a shaped RIS gain narrow beam and directionally projected onto this three-dimensional neighborhood, and sends an 8-byte wake-up control signaling containing a specific feature code.

[0034] After the built-in wake-up receiver (WUR) of the sensor nodes within the aforementioned neighborhood detects and verifies the signal, the main control MCU triggers a hardware interrupt. The node circuit is awakened from its sleep state with a power consumption of approximately 10μA and switches to an active transmission mode with a power consumption of tens of milliamps. During the transmission parameter reconfiguration phase, the sampling frequency of the sensor's operating channel is increased from the usual once every 15 minutes to a preset high-frequency sampling threshold, such as 10Hz, or 10 times per second. For the acoustic emission channel used to identify high-frequency leakage acoustic emission characteristics, the sampling frequency is configured to be no less than 200kHz. All eight nodes in this neighborhood cluster automatically transmit high-frequency data streams concurrently to the base station using orthogonal frequency division multiple access or orthogonal code division multiple access mechanisms, based on pre-allocated pseudo-random orthogonal sequences or orthogonal frequency division resource blocks, thereby reducing multiple access interference and channel collision problems caused by dense high-frequency transmission.

[0035] In one possible implementation, when the differential spatiotemporal anomaly score of any sensing node exceeds the first-level threshold, the sensing node position is solved into a second RIS phased matrix, including: The three-dimensional coordinates of the abnormal sensing nodes whose differential spatiotemporal anomaly scores exceed the first-level threshold, the three-dimensional coordinates of the base station receiving antenna, and the three-dimensional coordinates and normal vector of the RIS array are obtained. The incident angle from the abnormal sensing node to the RIS array and the reflection angle from the RIS array to the base station receiving antenna are calculated based on the three-dimensional coordinates. The initial phase compensation value required for each reflection element of the RIS array to achieve directional beamforming is calculated based on the generalized Snell's law. The initial phase compensation values ​​of all reflection elements are combined into a two-dimensional phase control array to generate a second RIS phase control matrix for controlling the beam pointing of the RIS reflection array.

[0036] The control center's microprocessor extracts the three-dimensional coordinates of the abnormal sensor nodes that triggered the Level 1 early warning through a GIS database interface. Three-dimensional coordinates of the base station receiving antenna and the three-dimensional reference coordinates of the center of the fixedly deployed RIS panel. Including the three-dimensional unit normal vector of the array orientation Using vector algebra operations in spatial analytic geometry, the spatial incident wave direction vector originating from the anomalous sensing node and reaching the center of the RIS array is calculated, and compared with the normal vector. The included angle is calculated to obtain the three-dimensional incident angle including azimuth and elevation angles, such as azimuth 45° and elevation 30°; using the principle of equivalent geometric projection, the three-dimensional reflection angle, azimuth and elevation angles from the center of the RIS array to the base station receiving antenna are calculated.

[0037] After locking the three-dimensional spatial angle, the core algorithm engine, based on the generalized Snell's law, combines the incident direction, target reflection direction, and operating wavelength to construct a set of spatial phase gradient constraint equations for the RIS array. It then calculates the theoretical phase shift for each of the 256×256 independent reflection units on the RIS array. For a specific reflection unit located in the m-th row and n-th column of the array, a mapping function between the spatial coordinate difference and the current operating frequency wavelength is used; for example, for a 28GHz millimeter wave, the wavelength... ≈10.7mm, the initial phase compensation value necessary to compensate for the optical path difference of electromagnetic wave propagation and generate far-field directional coherence enhancement was calculated. Typically, the control state is discretized into 2 bits, such as a value of 0. , , Firstly, after the calculation is completed, the controller maps the discrete phase compensation values ​​of the 65,536 units into a two-dimensional phase digital control array according to the spatial row and column topology combination. This array is then converted into a high and low level instruction set for the FPGA underlying driver, which is used as the output deployment of the second RIS phase control matrix to regulate the beam pointing energy.

[0038] In one possible implementation, the communication quality of the sensor node cluster is used as an indicator to apply a perturbation to the second RIS phased matrix, and iterative optimization is performed based on feedback until the communication quality converges, thereby locking the first RIS phased matrix sequence to achieve beam locking, including: On the phase values ​​of each reflection unit of the currently used second RIS phased matrix, a random phase perturbation following a preset distribution is superimposed to generate a candidate second RIS phased matrix and sent to the RIS reflection array; the uplink transmission link signal-to-noise ratio of each sensor node in the sensor node cluster with the applied perturbation candidate second RIS phased matrix is ​​simultaneously collected, the average signal-to-noise ratio is calculated and the lowest node signal-to-noise ratio is extracted as the communication quality feedback value; the difference between the current average signal-to-noise ratio and the average signal-to-noise ratio of the previous iteration cycle is calculated. If the difference is greater than 0, the random phase perturbation is retained and the candidate second RIS phased matrix is ​​updated to the currently used matrix; if the difference is not greater than 0, the random phase perturbation is discarded and the matrix of the previous cycle is restored; the perturbation and feedback process is repeated until the improvement of the average signal-to-noise ratio is lower than the preset convergence threshold or the maximum number of iterations is reached, and the iteratively optimized second RIS phased matrix is ​​locked.

[0039] Employing a beam tracking mechanism based on random hill-climbing or genetic principles, the controller superimposes, at a frequency of, for example, 100 times per second, a random phase perturbation component with a mean of 0 and a small-angle uniform distribution, for example, with a variation range set to [-15°, +15°], onto the intrinsic phase value of each reflection unit of the currently executing second RIS phased array matrix. Execution mode The system performs discretization processing to generate a candidate second RIS phased array matrix carrying a perturbation factor, and sends it to the RIS reflector array backplane driver chip via the SPI interface within milliseconds to perform state refresh.

[0040] After the array state switch, the base station baseband processing unit synchronously collects the instantaneous received signal-to-noise ratio (SNR) of all nodes in the cluster within the abnormal neighborhood on the uplink shared channel within a millisecond time window, and calculates the arithmetic mean SNR of all nodes at this moment. For example, if the measured cluster average SNR improves to 15.5 dB, this is used as the communication quality feedback value for this array phase state fine-tuning. In the communication quality feedback judgment process, the improvement in average SNR is used as the main optimization index, and the lowest node SNR is used as the constraint index. When the lowest node SNR is lower than the preset reliable communication lower limit, even if the average SNR is higher than the previous cycle, the candidate second RIS phased array matrix is ​​determined not to meet the communication quality improvement conditions and is rejected. The feedback value is subtracted from the historical average signal-to-noise ratio (SNR) saved in the previous iteration cycle, for example, 15.1 dB, to obtain the evaluation difference. Since 0.4 dB > 0, the decision logic determines that the current random phase perturbation has produced positive gain optimization. Therefore, the perturbation value is retained, and the current working matrix memory pointer is updated to point to the candidate second RIS phased array matrix. Conversely, if the calculated difference is ≤ 0, it is determined to be negative attenuation, triggering a discard instruction and instructing the hardware state machine to restore the phase matrix cached in the previous cycle. The above closed-loop iterative process of applying perturbation-collecting feedback-decision evolution is executed repeatedly until it is detected that the absolute increase in average SNR is continuously lower than the preset convergence threshold (e.g., 0.05 dB) for 20 consecutive iteration cycles, or the cumulative number of iterations reaches the maximum threshold (e.g., 500 times). At this point, the optimization algorithm terminates, thereby locking the first RIS phased array matrix sequence of the system at this time, which helps to achieve beam energy locking in abnormal node regions.

[0041] S3, In beam-locked state, a constraint separation matrix is ​​constructed based on the preset typical time series characteristics of gas leaks. Constrained independent component analysis is performed on the collected multidimensional sensor data to extract the signal characteristics of the leak source. When the signal power value of the leak source signal characteristics exceeds the second-level threshold, it is determined to be a gas leak event.

[0042] For channels with different sampling rates, alignment is first performed according to a unified analysis time window. Specifically, for low-frequency operating condition channels such as methane concentration, pipeline pressure, ambient temperature, and gas flow rate, statistical feature sequences such as window mean, slope, range, and rate of change are extracted as auxiliary discriminant features for constrained independent component analysis. The acoustic emission channel maintains an original sampling rate of no less than 200kHz for power spectrum analysis within the 20kHz to 80kHz frequency band. The low-frequency operating condition channels are not directly spliced ​​with the acoustic emission channel on a sample-by-sample basis. With the channel connectivity maintained by beamlocked phased registers, the system receives the statistical feature sequences of methane concentration, pipeline pressure, ambient temperature, and gas flow rate generated by the sensor node cluster within the unified analysis time window, as well as the high-frequency acoustic sequences output from the acoustic emission acquisition channel. These statistical feature sequences and high-frequency acoustic sequences together constitute the high-frequency sensor data in the subsequent data processing flow. A pre-loaded time-series feature model is constructed in memory, consisting of a pressure exponential decay falling edge model array, a concentration step rising edge model array, and a leak acoustic emission broadband pulse envelope model array at the occurrence of a standard gas leak. The cross-correlation function between the pre-loaded time-series feature model and the received high-frequency sensor data is calculated. Specifically, the pressure exponential decay falling edge model array is cross-correlated with the received pipeline pressure change feature sequence; the concentration step rising edge model array is cross-correlated with the received methane concentration change feature sequence; and the leak acoustic emission broadband pulse envelope model array is cross-correlated with the received high-frequency acoustic sequence. The cross-correlation coefficient is used as a reference signal similarity constraint term, combined with a conventional unmixing matrix. A separation matrix model with prior reference signal constraints is constructed. Based on the FastICA negative entropy maximization iterative framework, a custom-constrained independent component analysis program is built. A reference signal similarity constraint term is added to the objective function, and the unmixing matrix is ​​updated using Newton's iteration method. A window-level observation matrix composed of acoustic emission high-frequency features and operating condition auxiliary discrimination features is input into the unmixing matrix for linear projection mapping. The independent component vector with the highest Pearson correlation coefficient with the preset leakage features is separated and used as a one-dimensional array of leakage source signal features. The power spectral density is calculated by performing a discrete Fourier transform on the extracted one-dimensional array of leakage source signal features. The power spectral density array is numerically integrated and summed along the preset analysis frequency band to obtain the normalized signal power value. An if conditional branch statement is used to determine whether the signal power value is greater than the second-level threshold configured in memory. If so, the alarm rendering thread is called on the system's graphical user interface to mark and confirm a gas leak event.

[0043] In one possible implementation, a constraint separation matrix is ​​constructed based on preset typical time series characteristics of gas leaks. Constrained independent component analysis is performed on the collected multidimensional sensor data to extract leak source signal features. When the signal power value of the leak source signal features exceeds a second-level threshold, it is determined to be a gas leak event, including: Typical time-series features of gas leaks are extracted as prior reference signals. A contrast function containing independence metric and reference signal similarity constraint is constructed. The prior reference signal is input into the constraint conditions of the contrast function. The contrast function is maximized using the Newton-Raphson iteration method to calculate the constraint separation matrix. The constraint separation matrix is ​​used to perform linear transformation and demixing on the mixed multidimensional sensor data to separate the target signal component with the highest correlation to the prior reference signal, which is used as the leak source signal feature. The power spectral density integral of the leak source signal feature is calculated to obtain the signal power value, which is used to compare and determine with the second-level threshold.

[0044] It should be noted that the typical time series characteristics of gas leakage in this embodiment are specifically embodied in the time series feature model pre-stored in memory; the time series feature model is composed of the pressure exponential decay falling edge model array, the concentration step rising edge model array, and the leakage acoustic emission broadband pulse envelope model array when a standard gas leakage occurs.

[0045] The pipeline data processing center has pre-built a time-domain waveform dictionary of standard gas rupture and leakage acoustic emission or concentration transient jump signals obtained experimentally into its flash memory. It retrieves and extracts a priori typical time-series waveform vectors exhibiting exponential energy decay superimposed with sudden broadband pulses, such as acoustic envelope curve data with a center frequency focused in the range of 20kHz to 80kHz, as a priori reference signal guiding the constraint independent component analysis. The acoustic emission acquisition channel used for this acoustic emission radio frequency band analysis has a sampling frequency of no less than 200kHz, therefore its Nyquist frequency is no less than 100kHz, capable of covering the target analysis frequency band of 20kHz to 80kHz. A composite contrast function is constructed. The first term of this function is a signal independence metric operator based on the nonlinear negative entropy criterion, used to make the components after blind separation tend to a non-Gaussian distribution. The second term is a similarity constraint function between the reference signal and the separation output, constructed, for example, using the Pearson correlation coefficient or negative standardized mean square error, to improve the correlation between the separation output and the prior reference signal, or to reduce the standardized error between them.

[0046] During the function extremum solution stage, the microprocessor calls a custom Newton iterative solver built with reference to the FastICA negative entropy maximization idea to perform a maximization approximation operation on the composite contrast function, and solves for a global constraint separation matrix of size 8×8 at the steady-state stationary point where the function derivative is 0. The dimension depends on the number of parallel fusion sensing channels, and this matrix model is used. For the hybrid high-frequency multidimensional time-domain observation array that converges to the cloud and contains environmental background noise and multimodal correlation crosstalk, A linear spatial projection mapping is applied to demix and separate an independent one-dimensional target component that has a high correlation coefficient with the preset reference feature r(t) under the Pearson correlation coefficient evaluation, such as a correlation coefficient requirement of r≥0.90. The component output is confirmed as the intrinsic signal feature of the leakage source after filtering out artifacts. A high-resolution fast Fourier transform with a length of 4096 points is performed on the intrinsic signal feature of the leakage source. In the high frequency band of the feature, such as within a specific analysis window of 35kHz-45kHz, the one-sided power spectral density function is calculated. A definite integral operation is performed on the area under the frequency domain envelope to obtain the normalized signal power value representing the intensity of the leakage feature. For example, calculations yielded... =0.72, while the second-level threshold is set to 0.45. If the signal characteristics of the leak source exceed the second-level threshold, it is determined to be a gas leak event and the explosion-proof shut-off valve is activated.

[0047] The experimental conditions were set up to construct a natural gas pipeline network simulation test field covering an area of ​​20,000 square meters. 150 multi-dimensional sensing nodes were densely deployed along the pipeline, as well as a microwave segment reconfigurable smart surface array composed of 256×256 reflective units. The test environment was set with an average temperature of 25℃ and 65dB of background industrial white noise. During the test period, 50 micro-flow methane leaks and 30 sudden pipeline rupture leaks were randomly injected. The basic sampling frequency was fixed at once every 15 minutes. The wireless network uplink transmission was based on orthogonal frequency division multiple access technology. The experiment configured three ablation verification models, all of which ran continuously for 120 hours under the same hardware environment and communication load.

[0048] Three control combinations were set up in the experiment. Model 1 was a basic architecture using fixed time window polling and a single absolute threshold alarm. Model 2 was a basic architecture superimposed with a wide-area scanning and neighborhood cluster high-frequency wake-up module based on anomaly scores. Model 3 was the comprehensive architecture adopted in this invention, which further incorporated phased matrix feedback iterative optimization and constrained independent component analysis algorithms on the basis of Model 2. Data results showed that Model 1 detected a total of 35 leaks with 14 false alarms, an average uplink signal-to-noise ratio of 10.2 dB, and an average response delay of 12 minutes. Model 2 detected 65 leaks with false alarms reduced to 6, an uplink signal-to-noise ratio of 10.5 dB, and an average response delay reduced to 3 minutes. Model 3 detected 80 leak events under the same simulation test conditions, with no false alarms recorded during the test period. The average uplink signal-to-noise ratio of the node cluster increased to 16.8 dB, and the average warning response delay was compressed to 1.5 minutes. Figure 4 As shown, with the gradual integration of the core modules of this invention, the response latency is reduced and the uplink communication signal-to-noise ratio is improved, verifying the enhancement effect of each key technology on monitoring performance.

[0049] The polling scanning queue and beam dwell time adjustment mechanism adopted in this invention improves the critical early warning response speed by 87.5% relative to the basic architecture. Model 3 achieves a 6.3dB gain in the uplink signal-to-noise ratio dimension compared to Model 2. This verifies that the method of applying random phase perturbation and performing closed-loop optimization based on signal-to-noise ratio feedback can improve the spatial electromagnetic energy distribution and stabilize the communication beam of abnormal nodes. The experimental data compared with the false alarm records also show that applying constrained independent component analysis to the blind demixing processing of high-frequency multidimensional concurrent signals can reduce artifact interference caused by environmental multimodal coupling crosstalk and improve the accuracy and reliability of leakage identification in complex industrial environments.

Claims

1. A hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes, characterized in that, include: S1: Generate orthogonal coded aperture basis functions based on the three-dimensional topology of the gas pipeline network, including: acquiring spatial data of the gas pipeline network containing latitude, longitude, and elevation information, converting it into a unified metric three-dimensional rectangular coordinate system, and establishing a three-dimensional directed graph model of the pipeline network; using the smallest power of two, not less than the total number of pipeline nodes, as the coding order. ,generate The Hadamard matrix is ​​of order 1, and the row vectors of the Hadamard matrix are extracted as orthogonal coding aperture basis functions; A first RIS phased matrix sequence for wide-area polling scanning is constructed. The first RIS phased matrix sequence is used to control the RIS reflection array to perform beamforming on each sensing node along the pipeline. Multidimensional sensing data of each sensing node is collected in polling. The Mahalanobis distance between the multidimensional sensing data and the baseline is calculated to obtain the differential spatiotemporal anomaly score. The RIS wide-area polling scanning strategy for the next cycle is adjusted based on the differential spatiotemporal anomaly scores of all sensing nodes. S2: When the differential spatiotemporal anomaly score of any sensor node exceeds the first-level threshold, the sensor node position is solved into a second RIS phased matrix. The second RIS phased matrix is ​​used to perform position-oriented beamforming, and the sensor node cluster in the location neighborhood is awakened to enter the high-frequency sampling mode. The communication quality of the sensor node cluster is used as an indicator to apply perturbation to the second RIS phased matrix, and iterative optimization is performed based on feedback until the communication quality converges. The iteratively optimized second RIS phased matrix is ​​locked to achieve beam locking. S3: In beam-locked state, a constraint separation matrix is ​​constructed based on the preset typical time series characteristics of gas leaks. Constrained independent component analysis is performed on the collected multidimensional sensor data to extract the signal characteristics of the leak source. When the signal power value of the leak source signal characteristics exceeds the second-level threshold, it is determined to be a gas leak event.

2. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, The differential spatiotemporal anomaly score is obtained by calculating the Mahalanobis distance between the multidimensional sensor data and the baseline, including: Collect multidimensional sensing data from each sensing node under normal operating conditions for consecutive historical periods, calculate the arithmetic mean of each feature dimension to construct the mean baseline vector; use the multidimensional sensing data from historical periods to calculate the covariance matrix, and invert the covariance matrix to obtain the accuracy matrix. Subtract the mean baseline vector from the currently acquired multidimensional sensing data to obtain the corresponding difference vector; calculate the product of the transpose of the difference vector, the precision matrix, and the difference vector, and perform a square root operation on the product result. Use the obtained value as the differential spatiotemporal anomaly score of the sensing node at the current sampling time.

3. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, The RIS wide-area polling scan strategy for the next cycle is adjusted based on the differential spatiotemporal anomaly scores of all sensor nodes, including: Extract the differential spatiotemporal anomaly scores of each sensor node within a preset historical period, construct a time series of the differential spatiotemporal anomaly scores for each sensor node, calculate the average value of the time series composed of these differential spatiotemporal anomaly scores, and obtain the corresponding anomaly evaluation score. Sort all sensor nodes in descending order according to the anomaly evaluation scores, generate the polling scan queue for the next period, set the area where the top-ranked sensor nodes in the scan queue are located as high-priority scan areas, increase the beam dwell time and increase the polling frequency, set the area where the bottom-ranked sensor nodes in the scan queue are located as low-priority scan areas, reduce the beam dwell time and decrease the polling frequency, and generate low-level control commands to update the scan task list of the RIS reflector array.

4. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, Wake up the cluster of sensor nodes in the location neighborhood to enter high-frequency sampling mode, including: Centered on the location of the sensor node whose differential spatiotemporal anomaly score exceeds the first-level threshold, a spatial neighborhood is defined with a preset distance as the radius. The hardware communication address set of all sensor nodes within the range is extracted, and a wake-up control signal is broadcast to the sensor node corresponding to the hardware communication address set through a directional beam. After receiving the wake-up control signal, the sensor node switches from a low-power sleep standby state to an active transmission state, and increases the sampling frequency and uplink transmission frequency of the multidimensional sensor data to a preset high-frequency sampling threshold, transmitting the collected multidimensional sensor data in orthogonal multiple access mode.

5. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1 or 2, characterized in that, The sensor node positions are calculated into a second RIS phased matrix, including: The three-dimensional coordinates of the abnormal sensing nodes whose differential spatiotemporal anomaly scores exceed the first-level threshold, the three-dimensional coordinates of the base station receiving antenna, and the three-dimensional coordinates and normal vector of the RIS array are obtained. The incident angle from the abnormal sensing node to the RIS array and the reflection angle from the RIS array to the base station receiving antenna are calculated based on the three-dimensional coordinates. The initial phase compensation value required for each reflection element of the RIS array to achieve directional beamforming is calculated based on the generalized Snell's law. The initial phase compensation values ​​of all reflection elements are combined into a two-dimensional phase control array to generate a second RIS phase control matrix for controlling the beam pointing of the RIS reflection array.

6. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, Applying a perturbation to the second RIS phased matrix includes: On the phase values ​​of each reflection unit of the currently used second RIS phased matrix, random phase perturbations following a preset distribution are superimposed to generate a candidate second RIS phased matrix with applied perturbations and then sent to the RIS reflection array.

7. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, Iterative optimization based on feedback is performed until communication quality converges. The optimized second RIS phased array matrix is ​​then locked to achieve beam locking, including: The signal-to-noise ratio (SNR) of the uplink transmission link of each sensor node in the sensor node cluster with the candidate second RIS phased array matrix subjected to perturbation is collected synchronously. The average SNR is calculated and the lowest node SNR is extracted as the communication quality feedback value. The difference between the current average SNR and the average SNR of the previous iteration cycle is calculated. If the difference is greater than 0, the random phase perturbation is retained and the candidate second RIS phased array matrix is ​​updated to the currently used matrix. If the difference is not greater than 0, the random phase perturbation is discarded and the matrix of the previous cycle is restored. The perturbation and feedback process is repeated until the improvement of the average SNR is lower than the preset convergence threshold or the maximum number of iterations is reached. The iteratively optimized second RIS phased array matrix is ​​locked to achieve beam locking.

8. The method for hierarchical monitoring of differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, Extract the signal characteristics of the leakage source, including: The typical time series features of gas leaks are extracted as a priori reference signals. A contrast function containing an independence metric and a reference signal similarity constraint is constructed. The priori reference signals are input into the constraints of the contrast function. The contrast function is maximized using the Newton-Raphson iteration method to calculate the constraint separation matrix. The constraint separation matrix is ​​used to perform a linear transformation on the mixed multidimensional sensor data to demix and separate the target signal component with the highest correlation to the priori reference signal, which is used as the leak source signal feature.

9. The hierarchical monitoring method for differential transmission in complex pipeline networks based on topology nodes according to claim 1, characterized in that, Calculate the signal power values ​​characteristic of the leakage source signal, including: The power spectral density is calculated by performing a discrete Fourier transform on the extracted one-dimensional array of leakage source signal features. The power spectral density array is then numerically integrated and summed along a preset analysis frequency band to obtain the normalized signal power value.

Images