A backflow cable anti-theft method and system based on power carrier noise suppression

Abstract

The application relates to a backflow cable anti-theft method and system based on power carrier noise suppression, and belongs to the technical field of power system security and protection. Through sliding window preprocessing and empirical mode decomposition of carrier communication signals collected by multiple monitoring nodes, various noise components in the signals are effectively separated. Then, a combination of classification weighting and LSTM network prediction is adopted to realize intelligent noise suppression and future trend prediction. Through the synergistic effect of a dynamic threshold mechanism and an unscented Kalman filtering algorithm, the system can adaptively adjust the noise suppression strength, maximize the elimination of interference components while ensuring signal integrity. Finally, based on the double criteria of short-time energy and zero-crossing rate characteristics, combined with a multi-node time delay difference positioning algorithm, accurate identification and precise positioning of the theft event are realized.

Description

Technical Field

[0001] This application relates to the field of power system security technology, and in particular to a method and system for preventing theft of return cables based on power line carrier noise suppression. Background Technology

[0002] With the continuous advancement of smart grid construction, power line communication (PLC), as a low-cost communication method requiring no additional wiring, has been widely applied in distribution automation, electricity consumption information collection, and fault monitoring. Among these applications, the return cable, a crucial component of the power system, not only serves the function of current return but is also frequently used to carry power line carrier communication signals. However, due to its exposure to the outdoor environment and its widespread distribution, it is highly susceptible to theft, such as illegal cutting of cables to obtain copper and other metal resources. Therefore, how to achieve real-time monitoring of the return cable status and accurate identification and location of theft incidents without affecting normal communication has become a critical issue in the safe operation and maintenance of current power systems.

[0003] Currently, existing technologies determine the presence of abnormal interference or physical damage by analyzing the amplitude changes, frequency characteristics, or time-domain features of signals transmitted on power lines. However, the power line channel itself has a complex electromagnetic environment with significant background noise and sudden pulse interference, such as transient disturbances caused by switching operations, load changes, and lightning strikes. This results in the original signal exhibiting significant non-stationarity and time-varying characteristics, making direct analysis of such signals prone to misjudgment. Furthermore, traditional fixed-threshold detection mechanisms are difficult to adapt to noise fluctuations under different operating conditions, easily leading to missed or false alarms, especially under conditions of severe noise fluctuations or periodic interference, where detection accuracy decreases significantly.

[0004] Therefore, how to effectively suppress noise interference, improve signal quality, and achieve high-precision detection and location of theft incidents in complex and ever-changing power line carrier communication environments is a technical challenge that urgently needs to be solved. Summary of the Invention

[0005] To improve the security level of the return cable system, this application provides a return cable anti-theft method and system based on power line carrier noise suppression.

[0006] Firstly, this application provides a method for preventing theft of return cables based on power line carrier noise suppression, employing the following technical solution:

[0007] A method for preventing theft of return cables based on power line carrier noise suppression, the method comprising:

[0008] The original signal of the return cable carrier communication is acquired by at least two monitoring nodes in the power line carrier communication network, the original signal is subjected to sliding window truncation and normalization, and a preprocessed signal is output.

[0009] Empirical mode decomposition is performed on the preprocessed signal, and an intrinsic mode component set is generated through envelope construction and iterative decomposition.

[0010] The intrinsic mode components are classified into impulse noise components or background noise components based on their attenuation characteristics, and a set of intrinsic mode components containing classification labels is output.

[0011] Based on the classification labels, target components are selected from the set of intrinsic modal components, and a preliminary noise reduction signal is reconstructed.

[0012] The intrinsic mode components are assigned weights according to the classification labels, and the weighted intrinsic mode components are input into a long short-term memory network for noise prediction, and the predicted noise value for future time periods is output.

[0013] Extract the current measured noise value from the preliminary noise reduction signal, calculate the absolute difference between the predicted noise value and the measured noise value, and update the historical difference sequence.

[0014] Calculate the dynamic event triggering threshold based on the updated historical difference sequence;

[0015] When the absolute difference exceeds the dynamic event trigger threshold, the preliminary noise reduction signal is input into an unscented Kalman filter algorithm for filtering, and an optimized noise reduction signal is output; otherwise, the preliminary noise reduction signal is directly output as the optimized noise reduction signal.

[0016] Short-time energy feature values ​​and zero-crossing rate feature values ​​are extracted from the optimized noise-reduced signal;

[0017] When the short-time energy characteristic value exceeds the energy threshold or the zero-crossing rate characteristic value exceeds the zero-crossing threshold, a theft event is determined to have occurred.

[0018] The coordinates of the theft location are calculated based on the time delay difference of the signals arriving at the at least two monitoring nodes that detected the theft event.

[0019] By adopting the above technical solutions, signal acquisition is performed using a power line carrier communication network. Empirical mode decomposition and autocorrelation analysis are used to effectively separate impulse noise from background noise. Adaptive noise reduction is achieved by combining LSTM neural network prediction and unscented Kalman filtering algorithms, significantly improving signal quality. Simultaneously, the introduction of a dynamic event triggering mechanism and dual feature criteria of short-time energy / zero-crossing rate enables accurate identification of theft events and effectively reduces false alarm rates. Multi-node time delay difference positioning technology can accurately calculate the location of theft, realizing fully automated anti-theft monitoring from signal acquisition, intelligent noise reduction, anomaly detection, and precise positioning. This greatly improves the security level and operational efficiency of the return cable system.

[0020] Optionally, the step of performing empirical mode decomposition on the preprocessed signal, and generating a set of intrinsic mode components through envelope construction and iterative decomposition, includes:

[0021] Identify local maxima and local minima in the preprocessed signal;

[0022] Construct the upper envelope based on all local maxima and the lower envelope based on all local minima.

[0023] Calculate the envelope mean of the upper envelope and the lower envelope;

[0024] A new signal is generated by subtracting the mean envelope value from the preprocessed signal;

[0025] Determine whether the new signal satisfies the preset intrinsic mode function condition;

[0026] If the condition is not met, the new signal is used as the iterative input signal, and the above actions of identifying local extreme points, constructing upper and lower envelopes, calculating the envelope mean, and generating a new signal are repeated until the new signal satisfies the intrinsic mode function condition.

[0027] If the conditions are met, then the new signal is defined as an intrinsic mode component;

[0028] The difference between the preprocessed signal and the intrinsic mode components is calculated to generate a residual signal;

[0029] Determine whether the residual signal satisfies the decomposition termination condition; wherein, the decomposition termination condition includes the residual signal being a monotonic function or a constant;

[0030] When the residual signal does not meet the decomposition termination condition, the residual signal is used as a new preprocessed signal, and the above actions of identifying local extreme points, constructing upper and lower envelopes, calculating the envelope mean, generating new signals, defining intrinsic mode components, and generating residual signals are repeated until the residual signal meets the decomposition termination condition.

[0031] Collect all generated intrinsic mode components to obtain the intrinsic mode component set.

[0032] By employing the above technical solution, based on the empirical mode decomposition method, the preprocessed signal is adaptively decomposed into multiple intrinsic mode components and a residual trend term. The entire process is based on the local characteristics of the signal itself, requiring no preset basis functions, and exhibits good adaptability and time-frequency locality. Through repeated screening and iterative judgment, the system can effectively extract multi-scale oscillatory components from the signal, achieving fine decomposition and analysis of complex non-stationary signals.

[0033] Optionally, the step of classifying the intrinsic mode components into impulse noise components or background noise components based on their attenuation characteristics, and outputting a set of intrinsic mode components containing classification labels, includes:

[0034] Obtain a set of intrinsic modal components; wherein the set of intrinsic modal components contains multiple intrinsic modal components sorted by frequency from high to low;

[0035] Calculate the autocorrelation function and partial autocorrelation function for each intrinsic mode component;

[0036] Extract the decay characteristics of the autocorrelation function and the truncation characteristics of the partial autocorrelation function;

[0037] If the attenuation characteristic of the intrinsic mode component is lower than the first threshold and the truncation characteristic is lower than the second threshold, then the classification label is determined to be the impulse noise component; otherwise, the classification label is determined to be the background noise component.

[0038] The output contains a set of intrinsic modal components that include the category label.

[0039] By employing the aforementioned technical solution and combining autocorrelation and partial autocorrelation functions, automatic noise type classification of intrinsic mode components after EMD decomposition is achieved. This method fully utilizes the differences in temporal correlation characteristics among different noise types, and realizes intelligent noise identification and classification through quantitative analysis, providing a reliable foundation for subsequent targeted signal processing. The entire classification process has a sound theoretical basis and practical value, effectively improving the accuracy and efficiency of signal processing.

[0040] Optionally, the step of filtering target components from the intrinsic mode component set based on the classification labels and reconstructing to generate a preliminary denoised signal includes:

[0041] Obtain the set of intrinsic mode components and classification labels, wherein the classification labels identify each intrinsic mode component as an impulse noise component or a background noise component;

[0042] Delete all intrinsic mode components marked as impulse noise components in the intrinsic mode component set, and retain all intrinsic mode components marked as background noise components in the intrinsic mode component set to obtain the target component subset;

[0043] Each background noise component in the target component subset is weighted.

[0044] The weighted background noise components are linearly superimposed to reconstruct the initial denoised signal.

[0045] By employing the above technical solution, an effective conversion from the original signal to a preliminary denoised signal is achieved. Its core technological advantage lies in combining statistical analysis, frequency domain processing, and adaptive weighting to form a complete noise suppression framework. This solution not only effectively removes sudden interference such as impulse noise but also enhances useful signal components through a frequency-related weighting mechanism, thereby significantly improving signal quality while ensuring signal integrity. This processing method is particularly suitable for signal processing applications in complex noisy environments, providing a reliable data foundation for subsequent precise signal analysis and processing.

[0046] Optionally, the step of filtering target components from the intrinsic mode component set based on the classification labels and reconstructing to generate a preliminary denoised signal includes:

[0047] Obtain the set of intrinsic mode components and classification labels, wherein the classification labels identify each intrinsic mode component as an impulse noise component or a background noise component;

[0048] Filter all impulse noise components in the intrinsic mode component set, and extract the pulse width parameter and amplitude parameter for each component.

[0049] Pulse noise components with pulse width parameters less than a preset width threshold are identified as high-frequency glitches, and the amplitude parameters of the pulse noise components are corrected using an exponential decay function.

[0050] The corrected impulse noise components are merged with the uncorrected components to generate an optimized set of intrinsic mode components.

[0051] Based on the optimized set of intrinsic mode components, a preliminary denoised signal is reconstructed.

[0052] By adopting the above technical solution, a complete impulse noise processing workflow was established, enabling differentiated processing of different types of noise. Its core innovation lies in identifying the specific characteristics of high-frequency glitches and using an exponential decay function for fine-tuning, thereby achieving more effective noise reduction while maintaining signal integrity. This solution is particularly suitable for processing real-world signals containing complex impulse interference, providing a higher-quality data foundation for subsequent signal analysis and processing.

[0053] Optionally, the steps of assigning weights to the intrinsic mode components according to the classification labels, inputting the weighted intrinsic mode components into a long short-term memory network for noise prediction, and outputting the predicted noise values ​​for future time periods include:

[0054] Obtain a set of intrinsic mode components with classification labels; the classification labels include impulse noise component identifiers and background noise component identifiers;

[0055] Each intrinsic mode component is assigned a weight coefficient based on the classification label. The intrinsic mode component with the classification label as the impulse noise component is assigned a first weight value, and the intrinsic mode component with the classification label as the background noise component is assigned a second weight value, resulting in a weighted set of intrinsic mode components. The first weight value is greater than the second weight value.

[0056] The weighted intrinsic mode component set is sorted from high to low frequency to generate a time-series input sequence;

[0057] The time-series input sequence is divided into training sample blocks of fixed length, and each training sample block contains component values ​​at M consecutive time points;

[0058] The training sample blocks are input into a pre-constructed long short-term memory network, and the hidden state is calculated through forget gate, input gate and output gate mechanisms;

[0059] Generate a sequence of predicted noise values ​​for future time periods based on the final hidden state.

[0060] By employing the above technical solutions, an LSTM prediction model based on classification label weight allocation was constructed, achieving accurate prediction of complex noise signals. By combining the multi-scale analysis capabilities of Empirical Mode Decomposition (EMD) with the temporal modeling advantages of deep learning, a differentiated weight allocation strategy enhanced the model's focus on key noise components. Simultaneously, the gating mechanism of the LSTM network effectively captured the long-term dependencies of noise.

[0061] Optionally, the step of calculating the coordinates of the theft location based on the signal arrival delay difference detected by the at least two monitoring nodes includes:

[0062] Obtain the signal arrival timestamps when at least two monitoring nodes detect the theft event, and read the preset node spatial coordinate data;

[0063] Calculate the signal arrival delay difference Δt between any two monitoring nodes. ij Generate a set of time delay differences;

[0064] Based on the aforementioned time delay difference set and the electromagnetic wave propagation speed constant in the return cable, the distance difference Δd is calculated. ij;

[0065] Construct a system of distance difference equations: ;

[0066] In the above formula, (x,y,z) are the estimated coordinates of the theft location, (x... i ,y i ,z i ) and (x j ,y j ,z j These are the node spatial coordinate data of the two monitoring nodes, respectively.

[0067] The distance difference equations are solved using the least squares method to output the estimated coordinates of the theft location.

[0068] Read the return cable path function based on the pre-constructed return cable topology space;

[0069] Calculate the vertical projection point from the estimated theft location coordinates to the return cable path function, and output the cable length coordinates corresponding to the vertical projection point as the theft location coordinates.

[0070] By employing the aforementioned technical solution, a three-dimensional positioning system based on time-of-flight measurement was constructed. Combining least-squares optimization and spatial projection transformation, precise location of return cable theft incidents was achieved. The innovation of this solution lies in the organic integration of electromagnetic wave propagation theory, spatial geometry, and optimization algorithms. Through multi-node collaborative measurement and mathematical optimization, the limitations of single-point measurement are effectively overcome, improving positioning accuracy and reliability. Simultaneously, by introducing the return cable path function and vertical projection calculation, the three-dimensional spatial coordinates are transformed into more practical cable length coordinates for engineering applications, making the positioning results more intuitive and easier for engineering use. This technical solution has significant practical value in fields such as power system security monitoring and communication line maintenance, providing an effective technical means for the timely detection and handling of theft incidents.

[0071] Optionally, after the step of calculating the coordinates of the theft location based on the signal arrival delay difference of the at least two monitoring nodes detecting the theft event, the method further includes:

[0072] Retrieve a preset power transmission network topology database, the database storing at least the start coordinates, end coordinates, and corresponding segment identifiers of the line segments;

[0073] Traverse all line segments in the topology database, determine whether the coordinates of the theft location are within the closed interval between the start and end coordinates of any line segment, and output the segment identifier of the successful match.

[0074] Based on the section identifier, query the preset section-circuit breaker mapping table to obtain the corresponding circuit breaker control address and one or more backup power supply circuit numbers.

[0075] A power transmission network topology reconfiguration instruction set is generated and sent to the power transmission network control system interface; wherein, the instruction set includes a disconnection instruction and an activation instruction, the disconnection instruction is used to trigger a circuit breaker disconnection operation according to the circuit breaker control address, and the activation instruction is used to start the backup power supply circuit according to the backup power supply circuit number.

[0076] By adopting the above technical solution and combining spatial positioning technology with network topology management, an intelligent power system security protection system has been constructed. From time delay difference positioning to network reconfiguration response, the entire process demonstrates the modern power system's ability to quickly perceive, accurately locate, and intelligently handle security incidents. This solution not only improves the detection accuracy and response speed of theft incidents, but more importantly, through an automated network reconfiguration mechanism, it minimizes the impact on system operation when security incidents occur, ensuring the continuity and reliability of power supply.

[0077] Secondly, this application provides a return cable anti-theft system based on power line carrier noise suppression, which adopts the following technical solution:

[0078] A return cable anti-theft system based on power line carrier noise suppression, the system comprising:

[0079] The signal acquisition and processing module is used to acquire the original signal of the return cable carrier communication through at least two monitoring nodes in the power line carrier communication network, perform sliding window truncation and normalization on the original signal, and output the preprocessed signal.

[0080] The empirical mode decomposition module is used to perform empirical mode decomposition on the preprocessed signal, and generate a set of intrinsic mode components through envelope construction and iterative decomposition.

[0081] The classification module is used to calculate the autocorrelation function and partial autocorrelation function for each intrinsic mode component, classify the intrinsic mode component into impulse noise component or background noise component according to the attenuation characteristics, and output a set of intrinsic mode components containing classification labels.

[0082] The preliminary noise reduction module is used to filter target components from the set of intrinsic mode components based on the classification labels and reconstruct and generate a preliminary noise reduction signal;

[0083] The noise prediction module is used to assign weights to the intrinsic mode components according to the classification labels, and input the weighted intrinsic mode components into the long short-term memory network for noise prediction, and output the predicted noise value for future time periods.

[0084] The measured noise extraction module is used to extract the current measured noise value from the preliminary noise reduction signal;

[0085] The difference calculation module is used to calculate the absolute difference between the predicted noise value and the measured noise value and update the historical difference sequence.

[0086] The threshold generation module is used to calculate dynamic event triggering thresholds based on the updated historical difference sequence.

[0087] The judgment module is used to determine whether the absolute difference exceeds the dynamic event triggering threshold. If yes, it outputs a first judgment result; if no, it outputs a second judgment result.

[0088] The Kalman filter module is used to respond to the first judgment result, input the preliminary noise reduction signal into the unscented Kalman filter algorithm for filtering, and output the optimized noise reduction signal;

[0089] The preliminary noise reduction signal output module is used to directly output the preliminary noise reduction signal as the optimized noise reduction signal in response to the second judgment result.

[0090] The feature extraction module is used to extract short-time energy feature values ​​and zero-crossing rate feature values ​​from the optimized noise reduction signal;

[0091] The theft detection module is used to determine that a theft event has occurred when the short-time energy characteristic value exceeds the energy threshold or the zero-crossing rate characteristic value exceeds the zero-crossing threshold.

[0092] The theft location module is used to calculate the theft location coordinates based on the arrival time delay difference of the signals detected by the at least two monitoring nodes that detected the theft event.

[0093] Optionally, the system further includes:

[0094] The database retrieval module is used to retrieve a preset power transmission network topology database, which stores at least the start coordinates, end coordinates, and corresponding segment identifiers of line segments.

[0095] The segment matching module is used to traverse all line segments in the topology database, determine whether the coordinates of the theft location are within the closed interval between the start and end coordinates of any line segment, and output the segment identifier of the successful match.

[0096] The query module is used to query a preset section-circuit breaker mapping table based on the section identifier to obtain the corresponding circuit breaker control address and one or more backup power supply circuit numbers.

[0097] The power transmission network control module is used to generate a power transmission network topology reconfiguration instruction set and send it to the power transmission network control system interface; wherein, the power transmission network topology reconfiguration instruction set includes a disconnection instruction and an activation instruction, the disconnection instruction is used to trigger a circuit breaker disconnection operation according to the circuit breaker control address, and the activation instruction is used to start the backup power supply circuit according to the backup power supply circuit number.

[0098] In summary, this application includes at least one of the following beneficial technical effects: It constructs a return cable anti-theft system based on power line carrier noise suppression, achieving high-precision detection and location of theft incidents. Firstly, by performing sliding window preprocessing and empirical mode decomposition on the carrier communication signals collected by multiple monitoring nodes, various noise components in the signal are effectively separated. Then, a combination of classification weighting and LSTM network prediction is used to achieve intelligent noise suppression and future trend prediction. Through the synergistic effect of a dynamic threshold mechanism and an unscented Kalman filter algorithm, the system can adaptively adjust the noise suppression intensity, maximizing the elimination of interference components while ensuring signal integrity. Finally, based on the dual criteria of short-time energy and zero-crossing rate characteristics, combined with a multi-node time delay difference positioning algorithm, accurate identification and precise location of theft incidents are achieved. The overall scheme, through multi-level signal processing and intelligent analysis, significantly improves the detection accuracy, response speed, and anti-interference capability of the return cable anti-theft system, providing reliable technical support for the safe operation of the power system. Attached Figure Description

[0099] Figure 1 This is a schematic diagram of the first process of a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application.

[0100] Figure 2 This is a second flowchart illustrating a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application.

[0101] Figure 3 This is a schematic diagram of the third process of a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application.

[0102] Figure 4 This is a schematic diagram of the fourth process of a return cable anti-theft method based on power line carrier noise suppression according to one embodiment of this application.

[0103] Figure 5 This is a schematic diagram of the fifth process of a return cable anti-theft method based on power line carrier noise suppression according to one embodiment of this application.

[0104] Figure 6 This is a schematic diagram of the sixth process of a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application.

[0105] Figure 7 This is a schematic diagram of the seventh process of a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application.

[0106] Figure 8 This is a schematic diagram of the eighth process of a return cable anti-theft method based on power line carrier noise suppression, according to one embodiment of this application. Detailed Implementation

[0107] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-8 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0108] This application discloses a method for preventing theft of return cables based on power line carrier noise suppression.

[0109] Reference Figure 1 A method for preventing theft of return cables based on power line carrier noise suppression, the specific steps of which include:

[0110] Step S101: Collect the original signal of the return cable carrier communication through at least two monitoring nodes in the power line carrier communication network, perform sliding window truncation and normalization on the original signal, and output the preprocessed signal.

[0111] In this system, multiple monitoring nodes are deployed in the power line carrier communication network to form a spatial sampling matrix. Due to the complex electromagnetic environment of the power line channel, such as impedance mismatch, load switching, and switching operations, the acquired signals exhibit significant non-stationarity and time-varying characteristics. To extract signal segments with relatively stable characteristics, a sliding window technique is used to truncate the original signal, ensuring that each signal segment possesses quasi-steady-state characteristics within a local range, thereby avoiding boundary effects caused by signal abrupt changes.

[0112] Subsequently, the truncated signal is normalized, for example using the Z-score normalization method. The purpose is to eliminate amplitude shifts caused by differences in sensor sensitivity or environmental disturbances between different times or nodes, ensuring that the signal energy is concentrated within a finite dynamic range. This provides a uniform and standardized input condition for subsequent time-frequency analysis. This preprocessing not only improves signal consistency but also provides a good initial data foundation for subsequent Empirical Mode Decomposition (EMD).

[0113] Step S102: Perform empirical mode decomposition on the preprocessed signal, and generate a set of intrinsic mode components through envelope construction and iterative decomposition.

[0114] Empirical Mode Decomposition (EMD) is a data-driven time-frequency analysis method based on the local extremum characteristics of a signal. Its core idea is to decompose a complex signal layer by layer into several intrinsic mode functions (IMFs) and a residual term. This decomposition method does not require pre-defined basis functions and can effectively capture transient characteristics in the signal, making it particularly suitable for the analysis of nonlinear and non-stationary signals. Therefore, it has significant advantages in power line noise separation.

[0115] Step S103: Classify the intrinsic mode components into impulse noise components or background noise components according to the attenuation characteristics, and output a set of intrinsic mode components containing classification labels.

[0116] Specifically, the periodicity and correlation structure of a signal can be measured by calculating the autocorrelation function (ACF), while the partial autocorrelation function (PACF) can be used to identify short-term dependencies in the signal. Impulse noise is characterized by its sudden onset and short duration; therefore, the ACF curve typically decays rapidly to near zero when the lag time τ is small, and the PACF exhibits a significant truncation characteristic. In contrast, background noise often has a longer correlation, resulting in slow ACF decay and a PACF tailing phenomenon. Based on this difference, a decay threshold θ can be set for classification to quantitatively distinguish the physical causes of noise.

[0117] Step S104: Based on the classification label, select the target component from the set of intrinsic modal components and reconstruct to generate the initial denoised signal;

[0118] Among them, the classification label identifies the noise type, the target component screening aims to retain the dominant components of the signal, the reconstruction strategy aims to remove interference components, and the core frequency band of power line carrier communication is retained to improve the signal-to-noise ratio of the signal. This step is not only a key link in noise suppression, but also the basis for subsequent event detection and localization.

[0119] Step S105: Assign weights to intrinsic mode components according to classification labels, and input the weighted intrinsic mode components into a long short-term memory network for noise prediction, and output the predicted noise value for future time periods.

[0120] Specifically, a Weighted Long Short-Term Memory Network (Weighted LSTM) can be introduced to predict and model noise. Considering that different types of noise have varying degrees of impact on the system, this application designs a weight allocation mechanism based on classification labels: the impulse noise component is given a higher weight (e.g., 1.5) to emphasize its potential impact on the system state due to its suddenness; the background noise component is given a lower weight (e.g., 0.8) to suppress its long-term steady-state interference; the remaining components retain the default weight (1.0).

[0121] Subsequently, the weighted IMF sequence is fed into the LSTM network for training and prediction. LSTM, a special type of recurrent neural network (RNN), effectively captures the temporal dependencies of noise through its internal forget gate, input gate, and output gate mechanisms, making it particularly suitable for modeling non-stationary signals with long-term memory characteristics. Finally, the LSTM outputs the predicted noise values ​​for the next k steps, providing predictive support for subsequent event triggering and adaptive filtering.

[0122] Step S106: Extract the current measured noise value from the preliminary noise reduction signal, calculate the absolute difference between the predicted noise value and the measured noise value, and update the historical difference sequence.

[0123] The measured noise value is extracted from the initial denoised signal. The background noise level is calculated using short-time Fourier transform or direct difference. The absolute difference δ, which is the absolute value of the difference between the predicted noise value and the measured noise value, is used to reflect the accuracy of the prediction model (e.g., a large δ indicates a sudden noise change or prediction failure). The historical difference sequence serves as a time series database, storing the k most recent difference values ​​(e.g., k=50). It is updated using a first-in-first-out (FIFO) strategy, overflowing the oldest value when a new absolute difference δ is added.

[0124] Step S107: Calculate the dynamic event triggering threshold based on the updated historical difference sequence;

[0125] Specifically, the standard deviation σd can be calculated based on a set of historical differential sequences:

[0126] ;

[0127] Where M is the number of differences in the current set, and μd is the arithmetic mean of all differences in the set.

[0128] Subsequently, the preset sensitivity adjustment factor is multiplied by the standard deviation σd to obtain the basic threshold parameter. Then, the basic threshold parameter is subjected to boundary constraint processing. If the basic threshold parameter is less than the preset minimum threshold, the output dynamic time trigger threshold is the preset minimum threshold. If the basic threshold parameter is greater than the preset maximum threshold, the output dynamic time trigger threshold is the preset maximum threshold. Otherwise, the output dynamic time trigger threshold is the basic threshold parameter ε(t).

[0129] Understandably, this mechanism can automatically adjust the threshold based on the current noise fluctuation level: when the noise fluctuates drastically, the threshold automatically decreases, increasing the filtering trigger frequency; conversely, it reduces the processing burden. This dynamic mechanism is more flexible than a fixed threshold and can adapt to noise changes caused by sudden interferences (such as lightning strikes) in the power system, avoiding over-filtering or under-filtering.

[0130] Step S108: Determine whether the absolute difference exceeds the dynamic event trigger threshold; if yes, proceed to step S109; if no, proceed to step S110.

[0131] Step S109: Input the initial noise reduction signal into the unscented Kalman filter algorithm for filtering, and output the optimized noise reduction signal;

[0132] Step S110: Directly output the preliminary noise reduction signal as the optimized noise reduction signal;

[0133] Unscented Kalman Filtering (UKF) is a nonlinear filtering algorithm that approximates the state distribution using Sigma points (generating 2n+1 points, where n is the state dimension). When the absolute difference δ > ε(t), it indicates large noise fluctuations (such as anomalies caused by theft). UKF is used to filter the initial denoised signal. After the Sigma points propagate, the state estimate and covariance matrix are updated, and the optimized denoised signal is output. If δ ≤ ε(t), the Unscented Kalman Filtering (UKF) is skipped, and the initial denoised signal is output directly, reducing the computational burden.

[0134] Understandably, the advantage of unscented Kalman filtering (UKF) lies in its ability to handle non-Gaussian noise and nonlinear systems (such as time-varying power line channels). This dynamic processing mechanism not only improves filtering accuracy but also avoids repeated processing of stationary noise through event-triggered mechanisms, meeting the computing power constraints of power terminal equipment.

[0135] Step S111: Extract short-time energy feature values ​​and zero-crossing rate feature values ​​from the optimized noise-reduced signal;

[0136] Short-time energy (STE) is calculated as the square of the integrated signal within a fixed window, reflecting signal strength (e.g., a sudden increase in energy caused by a theft). Zero-crossing rate (ZCR) is the number of times the signal crosses zero, indicating frequency changes (e.g., high-frequency noise causes an increase in ZCR). Feature extraction is based on time-frequency analysis principles, and STE and ZCR are key indicators for theft detection.

[0137] Step S112: When the short-term energy characteristic value exceeds the energy threshold or the zero-crossing rate characteristic value exceeds the zero-crossing rate threshold, a theft event is determined to have occurred.

[0138] Among them, the energy threshold E th and zero-crossing rate threshold Z th The preset upper limit is the normal operating condition limit (e.g., set based on historical training data statistics). The judgment logic is: if the short-time energy characteristic value E > E th This indicates strong physical interference (such as cable cutting); if the zero-crossing rate eigenvalue Z > Z th This indicates a surge in high-frequency noise (such as unauthorized touching), and either of these conditions is sufficient to determine if a theft has occurred.

[0139] Understandably, this dual-feature criterion design takes into account the detection needs of both high-amplitude destructive behavior and low-amplitude electricity theft, thereby improving the robustness and coverage of the system.

[0140] Step S113: Calculate the coordinates of the theft location based on the time delay difference of the signals that at least two monitoring nodes detected the theft event.

[0141] Among them, spatial location of theft events is achieved based on the Time Delay Difference Location (TDOA) principle. When multiple monitoring nodes detect a theft event simultaneously, the arrival timestamps of signals from any two nodes are obtained (where i is the node index), and the time delay difference Δt is calculated. ij =t i -t j Based on the propagation speed v of electromagnetic waves in the cable, a system of hyperbolic equations is established to solve for the spatial coordinates of the theft point.

[0142] In the above implementation, a power line carrier communication network is used for signal acquisition. Empirical mode decomposition and autocorrelation analysis are used to effectively separate impulse noise from background noise. Adaptive noise reduction is performed by combining LSTM neural network prediction and unscented Kalman filtering algorithm, which significantly improves signal quality. At the same time, a dynamic event triggering mechanism and short-time energy / zero-crossing rate dual feature criteria are introduced to accurately identify theft events and effectively reduce the false alarm rate. Through multi-node time delay difference positioning technology, the location of theft can be accurately calculated. This realizes fully automated anti-theft monitoring from signal acquisition, intelligent noise reduction, anomaly detection and accurate positioning, which greatly improves the security protection level and operation and maintenance efficiency of the return cable system.

[0143] Reference Figure 2 As one implementation of step S102, the step of performing empirical mode decomposition on the preprocessed signal and generating a set of intrinsic mode components through envelope construction and iterative decomposition includes:

[0144] Step S201: Identify local maxima and local minima in the preprocessed signal;

[0145] This process involves detecting extrema in the original signal to provide foundational data for constructing the upper and lower envelopes. A local maximum (Local Maximum) is a point in time where the signal amplitude is greater than its neighboring points; a local minimum (Local Minimum) is a point where the signal amplitude is less than its neighboring points. Accurate identification of extrema is fundamental to the EMD algorithm because they reflect the local fluctuation characteristics of the signal at different time scales. Inaccurate extrema identification will distort the constructed envelope, thus affecting the accuracy of the entire decomposition process.

[0146] Step S202: Construct the upper envelope based on all local maxima and the lower envelope based on all local minima.

[0147] Specifically, the envelope is typically constructed using interpolation methods, such as cubic spline interpolation, to ensure a smooth envelope that accurately encompasses all extreme points. The upper envelope reflects the upper bound of local signal fluctuations, while the lower envelope reflects the lower bound. By constructing both upper and lower envelopes, the local oscillation characteristics of the signal can be grasped holistically, providing a basis for subsequent calculation of the envelope mean.

[0148] Step S203: Calculate the envelope mean of the upper and lower envelope lines;

[0149] The envelope mean is generated by the arithmetic mean of the upper and lower envelopes. The calculation of the envelope mean aims to extract the local average trend of the signal, i.e., the central trend line of the signal at the current scale. This trend line reflects the average behavior of the signal within a local time window and can be considered as the transition between high-frequency and low-frequency components in the signal. By removing this average trend from the original signal, the high-frequency oscillation components in the signal can be gradually extracted.

[0150] Step S204: Subtract the envelope mean from the preprocessed signal to generate a new signal;

[0151] The purpose of this operation is to perform "detrending" of the signal, that is, to remove components in the signal that are correlated with the local average trend, thereby highlighting the high-frequency oscillating components in the signal. This "sieving" process is one of the core mechanisms of the EMD algorithm. Through repeated sieving, different frequency components in the signal are gradually separated. The new signal can be regarded as a candidate for the potential IMF components extracted in the current iteration.

[0152] Step S205: Determine whether the new signal satisfies the preset intrinsic mode function condition;

[0153] If satisfied, proceed to step S206;

[0154] If the condition is not met, the new signal is used as the iterative input signal, and the process returns to step S201. The above actions of identifying local extreme points, constructing upper and lower envelopes, calculating the envelope mean, and generating a new signal are repeated until the new signal satisfies the intrinsic mode function condition.

[0155] Step S206: Define the new signal as an intrinsic mode component;

[0156] Intrinsic mode components (IMFs) are a special type of signal defined in the EMD algorithm. They must satisfy two basic conditions: first, the number of extreme points and the number of zero points in the entire time domain are equal or differ by at most one; second, the average value of the upper and lower envelopes of the signal is zero. These two conditions ensure that IMFs have good local symmetry and zero-mean characteristics, thus giving them clear physical meaning in time-frequency analysis.

[0157] Understandably, determining whether the IMF condition is met is a crucial step in the EMD algorithm to ensure the validity of the decomposition results. If the condition is not met, it indicates that the current new signal still contains too many trend components or asymmetric fluctuations, requiring further screening. Through continuous iteration, the system gradually approaches the true IMF component, making each newly generated signal more consistent with the IMF definition. The convergence of this process depends on the characteristics of the signal itself and the setting of the screening stopping criterion, typically using the standard deviation criterion or the extreme point stability criterion as the stopping condition.

[0158] Once a new signal satisfies the IMF condition, it is defined as an intrinsic mode component (IMF). An IMF represents the local oscillation behavior of the original signal at a certain time scale. Its frequency and amplitude may change with time, exhibiting good adaptability and local characteristics. Each IMF can be regarded as an expression of a certain frequency component in the signal, and it has important application value in time-frequency analysis, signal denoising, and feature extraction.

[0159] Step S207: Calculate the difference between the preprocessed signal and the intrinsic mode components to generate the residual signal;

[0160] The residual signal represents the portion of the original signal remaining after removing the current IMF component; that is, the components in the signal that have not yet been decomposed. The residual signal can be considered as the low-frequency trend term or remaining component of the original signal, and it will serve as the input signal for the next round of EMD decomposition. By continuously extracting IMF components and updating the residual signal, the original signal is progressively decomposed into multiple IMF components and a final trend term.

[0161] Step S208: Determine whether the residual signal satisfies the decomposition termination condition; wherein, the decomposition termination condition includes the residual signal being a monotonic function or a constant.

[0162] If the decomposition termination condition is met, proceed to step S209;

[0163] Specifically, when the residual signal becomes a monotonic function, it indicates that it no longer contains any local fluctuation characteristics, and it is no longer possible to extract meaningful IMF components. Therefore, the decomposition process can be terminated. This termination condition ensures that the EMD algorithm does not iterate indefinitely, while also avoiding further decomposition of meaningless signals.

[0164] If the decomposition termination condition is not met, the residual signal is used as a new preprocessed signal, and the process returns to step S201. The above actions of identifying local extreme points, constructing upper and lower envelopes, calculating the envelope mean, generating new signals, defining intrinsic mode components, and generating residual signals are repeated until the residual signal meets the decomposition termination condition.

[0165] This process constitutes the "outer loop" of the EMD algorithm. Through layer-by-layer decomposition, the system decomposes the original signal into multiple IMF components and a final residual trend term. Each round of decomposition corresponds to the extraction of components at a specific frequency scale in the signal, thereby achieving multi-scale decomposition of the signal.

[0166] Step S209: Collect all generated intrinsic mode components to obtain the intrinsic mode component set.

[0167] The intrinsic mode component set (IMF) contains the local oscillation characteristics of the original signal at different time scales and is the core output of the EMD algorithm. Each IMF component has a clear physical meaning and good time-frequency locality, and can be used in subsequent signal analysis, feature extraction, denoising and other applications.

[0168] In the above implementation, based on the empirical mode decomposition method, the preprocessed signal is adaptively decomposed into multiple intrinsic mode components and a residual trend term. The entire process is based on the local characteristics of the signal itself, requiring no preset basis functions, and exhibits good adaptability and time-frequency locality. Through repeated screening and iterative judgment, the system can effectively extract multi-scale oscillatory components from the signal, achieving fine decomposition and analysis of complex non-stationary signals.

[0169] Reference Figure 3 As one implementation of step S103, the step of classifying intrinsic mode components into impulse noise components or background noise components based on attenuation characteristics and outputting a set of intrinsic mode components containing classification labels includes:

[0170] Step S301: Obtain the intrinsic modal component set; wherein, the intrinsic modal component set contains multiple intrinsic modal components sorted from high to low frequency;

[0171] Specifically, the intrinsic mode components (IMCs) are obtained by a pre-processing EMD procedure, with each IMC representing the local oscillation behavior of the original signal at a specific time scale. The frequency ordering characteristic means that high-frequency components typically correspond to rapidly changing elements in the signal, while low-frequency components correspond to slowly changing trending elements. This ordering provides a good foundation for subsequent noise classification, as different types of noise tend to be distributed across different frequency scales.

[0172] Step S302: Calculate the autocorrelation function and partial autocorrelation function for each intrinsic mode component;

[0173] The autocorrelation function (ACF) is a fundamental tool in time series analysis. It measures the self-similarity of a time series by calculating the dot product of the component time series with itself at different time lags. Specifically, for a time series {x(t)}, its ACF R(τ) is defined as R(τ) = E[x(t)x(t+τ)], where τ represents the lag order and E[·] represents the expectation operation. The ACF reflects the degree of correlation of the series at different time intervals. For periodic signals, the ACF exhibits a clear periodic decay characteristic; while for random noise, the ACF usually decays rapidly to zero.

[0174] The partial autocorrelation function (PAC) is a further processing of the autocorrelation function. It eliminates the influence of intermediate lags through recursive least squares regression, thereby extracting the autocorrelation characteristics of the pure lag τ in the sequence. The PAC φ(τ) represents the conditional correlation of the time series at lag τ after controlling for the influence of intermediate lags 1 to τ-1. This characteristic allows the PAC to more accurately reflect the intrinsic structural features of the sequence.

[0175] Step S303: Extract the decay features of the autocorrelation function and the truncation features of the partial autocorrelation function;

[0176] The decay characteristic is defined as the lag order at which the autocorrelation function value first crosses zero. This indicator reflects the duration of the sequence correlation. For impulse noise, due to its suddenness and brevity, the autocorrelation function usually decays rapidly to zero within a short lag order. However, for background noise, especially persistent random noise, the decay process of its autocorrelation function is relatively slow, and the lag order at which it first crosses zero is larger.

[0177] The truncation feature is defined as the maximum lag order at which the partial autocorrelation function (PAC) value consistently falls within a confidence interval after a preset lag order. The confidence interval is typically set based on statistical principles. When the PAC value falls within this interval, it indicates that the correlation at that lag order is not statistically significant. The truncation feature reflects the range of lag orders in which the sequence is significantly correlated. For impulse noise, due to its transient and localized nature, the PAC function usually exhibits truncation characteristics after a relatively small lag order; while the truncation feature of the PAC function for background noise is often more significant and stable.

[0178] Step S304: Determine whether the attenuation characteristic of the intrinsic mode component is lower than the first threshold and the truncation characteristic is lower than the second threshold; if yes, proceed to step S305; if no, proceed to step S306.

[0179] Step S305: Determine the classification label as impulse noise component;

[0180] Step S306: Determine the classification label as the background noise component;

[0181] The logical basis of this classification criterion is that impulse noise typically exhibits characteristics of strong suddenness, short duration, and rapid correlation decay; therefore, its autocorrelation function decay characteristic and partial autocorrelation function truncation characteristic are relatively small. Background noise, on the other hand, tends to be characterized by strong persistence and long correlation duration, resulting in relatively large values ​​for these two characteristics. By setting appropriate thresholds, the system can effectively distinguish between these two types of noise components.

[0182] For example, the first threshold can be set to 5, based on the typical hysteresis order of the impulse noise ACF crossing zero; the second threshold can be set to 2, based on the statistical boundary of the truncated background noise PACF.

[0183] Step S307: Output the set of intrinsic modal components containing the classification labels.

[0184] In the above implementation, the joint analysis of the autocorrelation function and the partial autocorrelation function enables automatic classification of noise types in the intrinsic mode components after EMD decomposition. This method fully utilizes the differences in temporal correlation characteristics among different types of noise, and achieves intelligent noise identification and classification through quantitative analysis, providing a reliable foundation for subsequent targeted signal processing. The entire classification process has a sound theoretical basis and practical value, and can effectively improve the accuracy and efficiency of signal processing.

[0185] Reference Figure 4 As one implementation of step S104, the step of filtering target components from the intrinsic mode component set based on classification labels and reconstructing to generate a preliminary denoised signal includes:

[0186] Step S401: Obtain the set of intrinsic mode components and classification labels; the classification labels identify each intrinsic mode component as either an impulse noise component or a background noise component.

[0187] The accuracy of the classification labels directly affects the subsequent reconstruction results, because different types of noise require different processing strategies. Impulse noise components typically manifest as short-lived interference with strong bursts and large amplitudes. If these components are retained in the reconstructed signal, they will severely affect the signal quality. While background noise components also fall under the category of noise, they often have a certain degree of continuity and regularity and may contain some useful signal information, thus requiring different treatment.

[0188] Step S402: Delete all intrinsic mode components marked as impulse noise components in the intrinsic mode component set, and retain all intrinsic mode components marked as background noise components in the intrinsic mode component set to obtain the target component subset;

[0189] The theoretical basis of this screening strategy lies in the fundamental difference between impulse noise and background noise: impulse noise is usually generated by sudden external interference or system transient response, exhibiting a sharp pulse shape in the time domain and broadband characteristics in the frequency domain; while background noise reflects the inherent characteristics of the system and the level of environmental noise, and often has a stronger correlation with the useful signal. Through this "removing the bad and keeping the good" screening mechanism, the system can effectively remove most sudden interference, laying a good foundation for subsequent signal reconstruction.

[0190] Step S403: Perform weighted processing on each background noise component in the target component subset;

[0191] The weighting weights are calculated based on the correlation between the center frequency of the target component and the preset signal frequency band. The weighting weight w... i The calculation formula is:

[0192] ;

[0193] In the above formula, f i f is the center frequency of the i-th component. c Let denoted as the center frequency of the carrier signal, and Δf as the permissible frequency offset threshold. The introduction of the permissible frequency offset threshold Δf provides a scale reference for weighted attenuation, and its value directly affects the selectivity of the weighting function. A smaller Δf value will produce a steeper attenuation characteristic, enhancing frequency selectivity but potentially losing edge band information; a larger Δf value will produce a smoother attenuation characteristic, improving band coverage but potentially introducing more noise components.

[0194] Understandably, in practical applications, useful signals are typically concentrated within a specific frequency band, while noise components may be distributed across a wider spectrum. By calculating the proximity of the center frequency of each component to the center frequency of the carrier signal, the system can identify which components are more likely to contain useful signal information. This frequency-correlation-based weighting strategy achieves differentiated processing of components from different frequency bands, enhancing components close to the carrier frequency while suppressing those far from it. Furthermore, this strategy avoids simple binarization (i.e., complete retention or complete deletion) and instead employs continuous weight adjustments, better preserving signal integrity. Finally, the Gaussian weighting function exhibits good frequency domain localization characteristics, effectively controlling sidelobe leakage while ensuring the main lobe response.

[0195] Step S404: The weighted background noise components are linearly superimposed to reconstruct the initial noise reduction signal.

[0196] Through the aforementioned weighting process, the system can finely adjust the filtered subset of background noise components, enabling the reconstructed signal to achieve a better signal-to-noise ratio while maintaining its original structural characteristics. The final signal reconstruction process linearly superimposes all weighted target components to generate a preliminary denoised signal. This reconstruction method fully utilizes the adaptive and frequency-sorting characteristics of EMD decomposition, combined with intelligent classification and weighting mechanisms, to achieve efficient noise suppression and signal enhancement.

[0197] The above implementation achieves an effective conversion from the original signal to a preliminary denoised signal. Its core technological advantage lies in combining statistical analysis, frequency domain processing, and adaptive weighting to form a complete noise suppression framework. This scheme can not only effectively remove sudden interference such as impulse noise, but also enhance useful signal components through a frequency-related weighting mechanism, thereby significantly improving signal quality while ensuring signal integrity. This processing method is particularly suitable for signal processing applications in complex noisy environments, providing a reliable data foundation for subsequent precise signal analysis and processing.

[0198] Reference Figure 5 As another implementation of step S104, the step of filtering target components from the intrinsic mode component set based on classification labels and reconstructing to generate a preliminary denoised signal includes:

[0199] Step S501: Obtain the set of intrinsic mode components and classification labels; the classification labels identify each intrinsic mode component as an impulse noise component or a background noise component.

[0200] Step S502: Filter all impulse noise components in the intrinsic mode component set, and extract the pulse width parameter and amplitude parameter one by one;

[0201] The pulse width parameter reflects the duration of the pulse event on the time axis and is typically determined by calculating the time interval during which the pulse envelope exceeds a certain threshold. The amplitude parameter characterizes the intensity of the pulse and is generally taken as the maximum or effective value of the pulse envelope. This parameter extraction process requires the integration of envelope detection techniques from signal processing; commonly used methods include Hilbert transform, wavelet transform, or moving average filtering. By quantifying these characteristic parameters, the system can distinguish and classify different types of impulse noise.

[0202] Step S503: The pulse noise component with a pulse width parameter less than the preset width threshold is identified as high-frequency glitches, and the amplitude parameter of the pulse noise component is corrected by an exponential decay function.

[0203] Specifically, high-frequency glitches are a special type of impulse noise characterized by extremely short duration but large amplitude, typically generated by transient processes such as circuit switching, digital signal switching, or electromagnetic interference. This type of noise manifests as sharp spikes in the time domain and exhibits broadband characteristics in the frequency domain, severely impacting signal quality. By setting a preset width threshold, the system can automatically identify these high-frequency glitches and process them using specialized correction strategies.

[0204] The impulse noise component after correction by the exponential decay function is:

[0205] ;

[0206] In the above formula, I i (t) represents the original impulse noise component, t p Let k be the peak pulse value and k be a constant attenuation coefficient. This correction method is based on the energy distribution characteristics of pulse signals: real useful signals typically have relatively smooth variation characteristics, while high-frequency glitches manifest as sharp pulses with highly concentrated energy. Through exponential attenuation processing, the system can effectively reduce the peak intensity of glitches while maintaining their basic temporal position and morphological characteristics, avoiding excessive influence on adjacent useful signal components. The choice of attenuation coefficient k needs to be optimized according to the specific application scenario. A larger k value will produce a faster attenuation rate, which is suitable for processing glitches with extremely high amplitude but extremely short duration.

[0207] Step S504: The corrected impulse noise component is merged with the uncorrected component to generate an optimized intrinsic mode component set.

[0208] Not all impulse noise requires correction; only those components identified as high-frequency glitches require special handling. Other types of impulse noise may contain useful transient information or their characteristics may not meet the definition criteria for high-frequency glitches. This selective correction strategy allows the system to remove harmful noise while preserving potentially useful signal components. The merging process needs to ensure precise alignment of the components on the time axis, typically using interpolation or resampling techniques to guarantee time synchronization.

[0209] Step S505: Based on the optimized set of intrinsic mode components, a preliminary denoised signal is reconstructed.

[0210] The final denoised output signal is obtained by linearly superimposing the processed IMF components. The reconstruction process fully utilizes the completeness and orthogonality of EMD decomposition to ensure that the reconstructed signal accurately reflects the main features of the original signal. Compared with the traditional direct deletion strategy, this scheme processes high-frequency glitches by correction rather than deletion, which can better maintain the integrity of the signal and avoid signal distortion caused by over-processing.

[0211] The above implementation establishes a complete impulse noise processing workflow, enabling differentiated processing of different types of noise. Its core innovation lies in identifying the unique characteristics of high-frequency glitches and employing an exponential decay function for refined correction, thereby achieving more effective noise reduction while maintaining signal integrity. This scheme is particularly suitable for processing real-world signals containing complex impulse interference, providing a higher-quality data foundation for subsequent signal analysis and processing.

[0212] Reference Figure 6 As one implementation of step S105, the steps of assigning weights to intrinsic mode components according to classification labels, inputting the weighted intrinsic mode components into a long short-term memory network for noise prediction, and outputting the predicted noise value for future time periods include:

[0213] Step S601: Obtain the set of intrinsic mode components with classification labels; the classification labels include impulse noise component identifiers and background noise component identifiers.

[0214] Step S602: Assign weight coefficients to each intrinsic mode component according to the classification label. Assign a first weight value to the intrinsic mode component whose classification label is the impulse noise component, and assign a second weight value to the intrinsic mode component whose classification label is the background noise component, to obtain the weighted intrinsic mode component set.

[0215] In one embodiment of this application, the first weight value is greater than the second weight value.

[0216] Specifically, the theoretical basis for weight allocation stems from the differential analysis of the impact of noise on the system: impulse noise, due to its suddenness and high amplitude characteristics, often has a more severe impact on signal quality, thus requiring a higher weight to enhance its contribution to the prediction model. Background noise, although persistent, has a relatively mild impact, therefore it is assigned a lower weight. This weight allocation strategy not only aligns with the actual assessment of noise hazard but also provides more targeted input features for subsequent deep learning models. The specific setting of weight values ​​requires extensive experimental verification and statistical analysis, typically employing methods such as cross-validation to optimize weight parameters, ensuring that the prediction model accurately captures the evolution of noise.

[0217] Step S603: Sort the weighted intrinsic mode component set in descending order of frequency to generate a time-series input sequence;

[0218] The IMF components generated by Empirical Mode Decomposition (EMD) possess a clear frequency hierarchy. High-frequency components typically correspond to the detailed features and transient changes of the signal, while low-frequency components reflect the overall trend of the signal. The strategy of sorting by frequency from high to low is based on the following considerations: high-frequency noise often exhibits greater randomness and unpredictability, but its variation patterns are of significant reference value for subsequent predictions; simultaneously, this sorting method aligns with human cognitive habits and the conventional process of signal processing, helping deep learning models better learn and memorize the temporal evolution of noise. The sorting process requires accurate identification of the center frequency of each IMF component, typically using methods such as Hilbert transform or power spectral analysis to determine frequency characteristics.

[0219] Step S604: Divide the time-series input sequence into training sample blocks of fixed length, each training sample block containing component values ​​of M consecutive time points;

[0220] Wherein, the fixed length M satisfies:

[0221] ;

[0222] In the above formula, f S f is the sampling rate. min The minimum effective signal frequency is defined by this formula. This design ensures that each training sample block contains sufficient information to characterize the temporal features of the noise, while avoiding increased computational complexity and overfitting risks caused by excessively long sample lengths. The sample segmentation process needs to maintain temporal continuity, ensuring appropriate temporal overlap between adjacent sample blocks so that the model can learn the continuous evolution characteristics of the noise.

[0223] Step S605: Input the training sample block into the pre-constructed long short-term memory network and calculate the hidden state through the forget gate, input gate and output gate mechanism;

[0224] Specifically, Long Short-Term Memory (LSTM) networks, as a special type of recurrent neural network structure, effectively solve the gradient vanishing and gradient exploding problems of traditional RNNs when processing long sequences through their unique gating mechanism. The forget gate determines what information to discard from the cell state, controlling the degree of information retention through a sigmoid function outputting a value between 0 and 1; the input gate determines what new information to store in the cell state, updating the cell state in combination with the sigmoid gate and the tanh layer; and the output gate determines the output value based on the cell state. This mechanism enables LSTM to effectively capture long-term dependencies in noisy sequences, showing significant advantages for predicting noise with periodic or trending characteristics.

[0225] Step S606: Generate a sequence of predicted noise values ​​for future time periods based on the final hidden state.

[0226] In LSTM networks, the hidden state, as the core memory unit, contains complete representation information of the input sequence. Through output structures such as fully connected layers or softmax layers, the hidden state can be mapped to specific predicted values. This prediction mechanism not only considers the instantaneous characteristics of noise, but more importantly, captures the temporal evolution of noise, providing forward-looking information support for noise suppression and signal processing.

[0227] In the above implementation, an LSTM prediction model based on classification label weight allocation is constructed, achieving accurate prediction of complex noise signals. By combining the multi-scale analysis capability of empirical mode decomposition with the temporal modeling advantages of deep learning, a differentiated weight allocation strategy enhances the model's attention to key noise components, while the gating mechanism of the LSTM network effectively captures the long-term dependencies of noise.

[0228] Reference Figure 7 As one implementation of step S113, the step of calculating the coordinates of the theft location based on the signal arrival delay difference of at least two monitoring nodes detecting the theft event includes:

[0229] Step S701: Obtain the signal arrival timestamps when at least two monitoring nodes detect the theft event, and read the preset node spatial coordinate data;

[0230] Monitoring nodes are typically deployed at key locations on the return cable. When a theft occurs, they generate specific electromagnetic or mechanical vibration signals on the cable, which propagate as electromagnetic waves. Obtaining timestamps requires a high-precision clock synchronization mechanism, typically using GPS synchronization or Network Time Protocol (NTP) to ensure time consistency between nodes. The accuracy of time synchronization directly affects the accuracy of subsequent delay difference calculations. Acquiring node spatial coordinate data requires a pre-measurement and calibration process, usually using the Global Positioning System (GPS) or precision measuring instruments to determine the three-dimensional spatial coordinates of each monitoring node. This coordinate data serves as known parameters in subsequent positioning calculations.

[0231] Step S702: Calculate the signal arrival delay difference Δt between any two monitoring nodes. ij Generate a set of time delay differences;

[0232] Among them, the time delay difference Δt ij The calculation formula is Δt ij =t i -t j , where t i and t j These represent the timestamps of the signal arriving at the i-th and j-th monitoring nodes, respectively. Since the propagation speed of electromagnetic waves in the return cable is constant, the time delay difference directly reflects the difference in distance between the location of the theft incident and the two monitoring nodes. Generating the time delay difference set requires considering all possible node pair combinations; for N monitoring nodes, C(N,2) independent time delay difference values ​​can be generated. The key to this step lies in noise suppression and outlier handling. In practical applications, this may be affected by environmental interference, equipment errors, and other factors, requiring the use of filtering algorithms or statistical methods to improve the reliability of the time delay difference calculation.

[0233] Step S703: Calculate the distance difference Δd based on the time delay difference set and the constant propagation speed of electromagnetic waves in the return cable. ij ;

[0234] The formula for calculating the distance difference is Δd. ij =v·Δt ij Where v is the propagation speed constant of electromagnetic waves in the return cable. This speed constant needs to be obtained through experimental calibration and is usually affected by environmental factors such as cable material, structure, and temperature. Therefore, in practical applications, it is necessary to establish a mapping relationship between speed and environmental parameters. Distance difference Δd ij It has a clear geometric meaning; it represents the difference in distance between the location of the theft and the two monitoring nodes. This difference constitutes the basic constraint for subsequent spatial positioning calculations.

[0235] Step S704, construct the distance difference equation system: ;

[0236] In the above formula, (x,y,z) are the estimated coordinates of the theft location, (x... i ,y i ,z i ) and (x j ,y j ,z j These are the node spatial coordinate data of the two monitoring nodes, respectively.

[0237] Specifically, for any two monitoring nodes i and j, the distance difference constraint equation can be expressed as described above. Essentially, it is a nonlinear system of equations, the existence and uniqueness of which depend on the geometric distribution of the monitoring nodes and the quality of the measurement data. When the number of monitoring nodes is greater than or equal to four, an overdetermined system of equations can theoretically be obtained, thereby improving positioning accuracy and robustness.

[0238] Step S705: Solve the distance difference equation system using the least squares method and output the estimated coordinates of the theft location;

[0239] Since noise and errors are inevitable in actual measurements, directly solving the nonlinear equations often fails to yield an exact solution. Therefore, optimization algorithms such as the least squares method are needed to find the optimal solution that minimizes the sum of squared errors of all constraints.

[0240] Step S706: Read the return cable path function based on the pre-built return cable topology space;

[0241] Step S707: Calculate the vertical projection point of the estimated theft location coordinates onto the return cable path function, and output the cable length coordinates corresponding to the vertical projection point as the theft location coordinates.

[0242] Specifically, the above steps demonstrate the transformation process from three-dimensional spatial coordinates to actual engineering coordinates, based on the principle of projection transformation in spatial geometry. The return cable path function is typically represented by a parametric curve or a piecewise linear function, accurately describing the cable's trajectory in three-dimensional space. Calculating the vertical projection requires finding the shortest distance from the estimated location point to the cable path; this point represents the actual location of the theft. Calculating the cable length coordinates requires integrating the path function; the arc length from the cable's starting point to the projection point is the final coordinate of the theft location. This coordinate representation better meets the practical needs of engineering applications.

[0243] In the above implementation, a three-dimensional positioning system based on time difference measurement is constructed. Combining least squares optimization and spatial projection transformation, precise location of return cable theft incidents is achieved. The innovation of this scheme lies in the organic integration of electromagnetic wave propagation theory, spatial geometry, and optimization algorithms. Through multi-node collaborative measurement and mathematical optimization, the limitations of single-point measurement are effectively overcome, improving positioning accuracy and reliability. Simultaneously, by introducing the return cable path function and vertical projection calculation, the three-dimensional spatial coordinates are converted into more practical cable length coordinates for engineering applications, making the positioning results more intuitive and easier for engineering use. This technical solution has significant practical value in fields such as power system security monitoring and communication line maintenance, providing an effective technical means for the timely detection and handling of theft incidents.

[0244] Reference Figure 8 As a further implementation of the return cable anti-theft method, after the step of calculating the coordinates of the theft location based on the signal arrival delay difference of at least two monitoring nodes detecting the theft event, the method further includes:

[0245] Step S801: Retrieve the preset transmission network topology database; the database stores at least the start coordinates, end coordinates and corresponding segment identifiers of the line segment;

[0246] The power transmission network topology database stores the spatial layout information of the power system in a structured manner. Line segments, as the basic building blocks of the network, have their starting and ending coordinates defining their location in three-dimensional space. Segment identifiers are unique identifiers for each segment within the network, typically using a hierarchical coding method to reflect the segment's topological relationships and importance. The database design must consider data consistency, integrity, and scalability, usually employing relational or graph databases to store this complex topological data. By retrieving this database, the system obtains the reference data needed for spatial matching, providing reliable data support for subsequent segment location.

[0247] Step S802: Traverse all line segments in the topology database, determine whether the coordinates of the theft location are within the closed interval between the start and end coordinates of any line segment, and output the identifier of the successfully matched segment.

[0248] In three-dimensional space, determining whether a point is near a line segment requires considering geometric distance and projection relationships, typically using the shortest distance from the point to the line segment as the criterion. When the shortest distance between the theft location coordinates and a certain line segment is less than a preset threshold, the point can be determined to be within the influence range of that line segment. The concept of a closed interval has a special meaning here; it includes not only the strictly defined interval of the line segment's endpoints but also a strip-shaped area with a certain width as its center. This definition takes into account positioning errors and practical engineering requirements. The efficiency of the traversal algorithm is crucial for the system's real-time performance; spatial indexing techniques such as R-trees or quadtrees are typically used to accelerate the query process and reduce unnecessary computational overhead.

[0249] Step S803: Based on the section identifier, query the preset section-circuit breaker mapping table to obtain the corresponding circuit breaker control address and one or more backup power supply circuit numbers.

[0250] The section-circuit breaker mapping table is the core data structure for power system protection configuration, reflecting the logical connections and control relationships between devices in the network. Circuit breaker control addresses typically use Uniform Resource Identifiers (URIs) or device addresses to uniquely identify circuit breaker devices in the network, providing a target address for subsequent control command transmission. The backup power supply circuit number embodies the redundancy design principle of the power system, ensuring continuous power supply capability during main circuit failures or isolation through pre-configured backup circuits. The construction of the mapping table requires consideration of the system's protection configuration scheme and network topology, and is usually determined by system designers based on safety regulations and operational experience. Its accuracy and completeness directly affect the correctness of the system response.

[0251] Step S804: Generate a set of transmission network topology reconfiguration instructions and send them to the transmission network control system interface.

[0252] The instruction set includes disconnect instructions and activation instructions. The disconnect instructions are used to trigger the circuit breaker disconnection operation based on the circuit breaker control address, and the activation instructions are used to start the backup power supply circuit based on the backup power supply circuit number.

[0253] It should be noted that the generation of the instruction set must comply with the operating procedures and safety constraints of the power system to ensure the rationality and safety of control actions. The execution of disconnection instructions must consider the operating sequence and coordination of circuit breakers to avoid system impacts or protection malfunctions caused by improper operation. Activation instructions require verification of the status and capacity of the backup circuit to ensure that the backup circuit can handle the load demand of the isolated section. The control system interface typically uses standardized communication protocols such as IEC 61850 or DNP3 to ensure the reliability and real-time performance of instruction transmission.

[0254] In the above implementation, spatial positioning technology is combined with network topology management to construct an intelligent power system security protection system. From time delay difference positioning to network reconfiguration response, the entire process demonstrates the modern power system's ability to quickly perceive, accurately locate, and intelligently handle security incidents. This solution not only improves the detection accuracy and response speed of theft incidents, but more importantly, through an automated network reconfiguration mechanism, it minimizes the impact on system operation when security incidents occur, ensuring the continuity and reliability of power supply.

[0255] This application also discloses a return cable anti-theft system based on power line carrier noise suppression.

[0256] A return cable anti-theft system based on power line carrier noise suppression, the return cable anti-theft system comprising:

[0257] The signal acquisition and processing module is used to acquire the original signal of the return cable carrier communication through at least two monitoring nodes in the power line carrier communication network, perform sliding window truncation and normalization on the original signal, and output the preprocessed signal.

[0258] The Empirical Mode Decomposition (EMD) module is used to perform EMD on the preprocessed signal. It generates a set of intrinsic mode components through envelope construction and iterative decomposition.

[0259] The classification module is used to calculate the autocorrelation function and partial autocorrelation function for each intrinsic mode component, classify the intrinsic mode components into impulse noise components or background noise components according to the attenuation characteristics, and output a set of intrinsic mode components containing classification labels.

[0260] The preliminary noise reduction module is used to filter target components from the set of intrinsic mode components based on classification labels and reconstruct and generate a preliminary noise reduction signal;

[0261] The noise prediction module is used to assign weights to intrinsic mode components based on classification labels, and input the weighted intrinsic mode components into a long short-term memory network for noise prediction, outputting the predicted noise value for future time periods.

[0262] The measured noise extraction module is used to extract the current measured noise value from the preliminary noise reduction signal;

[0263] The difference calculation module is used to calculate the absolute difference between the predicted noise value and the measured noise value and to update the historical difference sequence.

[0264] The threshold generation module is used to calculate dynamic event triggering thresholds based on the updated historical difference sequence.

[0265] The judgment module is used to determine whether the absolute difference exceeds the dynamic event trigger threshold. If it does, the first judgment result is output; if not, the second judgment result is output.

[0266] The Kalman filter module is used to respond to the first judgment result, input the preliminary noise reduction signal into the unscented Kalman filter algorithm for filtering, and output the optimized noise reduction signal.

[0267] The preliminary noise reduction signal output module is used to directly output the preliminary noise reduction signal as the optimized noise reduction signal in response to the second judgment result;

[0268] The feature extraction module is used to extract short-time energy feature values ​​and zero-crossing rate feature values ​​from the optimized noise reduction signal;

[0269] The theft detection module is used to determine that a theft event has occurred when the short-term energy characteristic value exceeds the energy threshold or the zero-crossing rate characteristic value exceeds the zero-crossing threshold.

[0270] The theft location module is used to calculate the coordinates of the theft location based on the time delay difference of the signals that detect the theft event at at least two monitoring nodes.

[0271] As a further implementation of the return cable anti-theft system, it also includes:

[0272] The database retrieval module is used to retrieve a preset transmission network topology database; the database stores at least the start coordinates, end coordinates, and corresponding segment identifiers of the line segments.

[0273] The segment matching module is used to traverse all line segments in the topology database, determine whether the coordinates of the theft location are within the closed interval between the start and end coordinates of any line segment, and output the identifier of the successfully matched segment.

[0274] The query module is used to query the preset section-circuit breaker mapping table based on the section identifier to obtain the corresponding circuit breaker control address and one or more backup power supply circuit numbers.

[0275] The power transmission network control module is used to generate a power transmission network topology reconfiguration instruction set and send it to the power transmission network control system interface. The power transmission network topology reconfiguration instruction set includes a disconnection instruction and an activation instruction. The disconnection instruction is used to trigger the circuit breaker disconnection operation according to the circuit breaker control address, and the activation instruction is used to start the backup power supply circuit according to the backup power supply circuit number.

[0276] The return cable anti-theft system based on power line carrier noise suppression according to the present application embodiment can implement any of the above-mentioned return cable anti-theft methods, and the specific working process of each module in the return cable anti-theft system can refer to the corresponding process in the above-mentioned method embodiments.

[0277] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0278] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0279] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0280] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A backflow cable theft prevention method based on power carrier noise suppression, characterized by, The method comprises: Collecting a return cable carrier communication original signal through at least two monitoring nodes in a power carrier communication network, performing sliding window interception and normalization processing on the original signal, and outputting a pretreatment signal; Performing empirical mode decomposition on the pretreatment signal, generating an intrinsic modal component set through envelope line construction and iterative decomposition; Classifying the intrinsic modal components into impulse noise components or background noise components according to the attenuation characteristics, and outputting an intrinsic modal component set containing a classification label; Based on the classification label, screening target components from the intrinsic modal component set to reconstruct a preliminary noise reduction signal; According to the classification label, assigning weights to the intrinsic modal components, and inputting the weighted intrinsic modal components into a long short-term memory network for noise prediction, and outputting predicted noise values in a future period; Extracting a current measured noise value from the preliminary noise reduction signal, calculating the absolute difference between the predicted noise value and the measured noise value, and updating a historical difference sequence; Based on the updated historical difference sequence, calculating a dynamic event trigger threshold; When the absolute difference exceeds the dynamic event trigger threshold, inputting the preliminary noise reduction signal into an unscented Kalman filter algorithm for filtering, and outputting an optimized noise reduction signal; otherwise, directly outputting the preliminary noise reduction signal as the optimized noise reduction signal; Extracting short-time energy feature values and zero-crossing rate feature values from the optimized noise reduction signal; When the short-time energy feature values exceed an energy threshold or the zero-crossing rate feature values exceed a zero-crossing rate threshold, determining that a theft event has occurred; Based on the signal arrival time delay difference of the at least two monitoring nodes detecting a theft event, calculating a theft location coordinate.

2. The power line carrier noise suppression based backflow cable theft prevention method of claim 1, wherein, The step of performing empirical mode decomposition on the pretreatment signal to generate an intrinsic modal component set comprises: Identifying local maximum points and local minimum points in the pretreatment signal; Constructing an upper envelope line based on all local maximum points and a lower envelope line based on all local minimum points; Calculating the envelope mean of the upper envelope line and the lower envelope line; Subtracting the envelope mean from the pretreatment signal to generate a new signal; Determining whether the new signal meets a preset intrinsic modal function condition; If not, taking the new signal as an iterative input signal, and returning to repeat the actions of identifying local extreme points, constructing upper and lower envelope lines, calculating envelope mean, and generating a new signal until the new signal meets the intrinsic modal function condition; If yes, defining the new signal as an intrinsic modal component; Calculating the difference between the pretreatment signal and the intrinsic modal component to generate a residual signal; Determining whether the residual signal meets a decomposition termination condition; wherein the decomposition termination condition includes that the residual signal is a monotonic function or a constant; When the residual signal does not meet the decomposition termination condition, taking the residual signal as a new pretreatment signal, and returning to repeat the actions of identifying local extreme points, constructing upper and lower envelope lines, calculating envelope mean, generating a new signal, defining an intrinsic modal component, and generating a residual signal until the residual signal meets the decomposition termination condition; Collecting all generated intrinsic modal components to obtain an intrinsic modal component set.

3. The power line carrier noise suppression based backflow cable theft prevention method of claim 2, wherein, The step of classifying the intrinsic modal components into impulse noise components or background noise components according to the decay characteristics comprises: obtaining a set of intrinsic modal components; wherein the set of intrinsic modal components comprises a plurality of intrinsic modal components sorted in descending order of frequency; calculating an autocorrelation function and a partial autocorrelation function for each intrinsic modal component; extracting a decay feature of the autocorrelation function and a truncation feature of the partial autocorrelation function; determining a classification label as an impulse noise component if the decay feature of the intrinsic modal component is lower than a first threshold value and the truncation feature is lower than a second threshold value; otherwise, determining the classification label as a background noise component; outputting the set of intrinsic modal components comprising the classification label.

4. The power line carrier noise suppression based backflow cable theft prevention method of claim 3, wherein, The step of screening target components from the set of intrinsic modal components based on the classification label and reconstructing a preliminary noise-reduced signal comprises: obtaining the set of intrinsic modal components and the classification label, the classification label identifying each intrinsic modal component as an impulse noise component or a background noise component; deleting all intrinsic modal components marked as impulse noise components in the set of intrinsic modal components and retaining all intrinsic modal components marked as background noise components in the set of intrinsic modal components to obtain a subset of target components; performing weighting processing on each background noise component in the subset of target components; linearly superimposing the background noise components after weighting processing to reconstruct a preliminary noise-reduced signal.

5. The power line carrier noise suppression based backflow cable theft prevention method of claim 3, wherein, The step of screening target components from the set of intrinsic modal components based on the classification label and reconstructing a preliminary noise-reduced signal comprises: obtaining the set of intrinsic modal components and the classification label, the classification label identifying each intrinsic modal component as an impulse noise component or a background noise component; screening all impulse noise components in the set of intrinsic modal components and extracting impulse width parameters and amplitude parameters one by one; determining an impulse noise component with an impulse width parameter less than a preset width threshold value as a high-frequency burr noise and modifying the amplitude parameter of the impulse noise component using an exponential decay function; merging the modified impulse noise components with unmodified components to generate an optimized set of intrinsic modal components; reconstructing a preliminary noise-reduced signal based on the optimized set of intrinsic modal components.

6. The power line carrier noise suppression based backflow cable theft prevention method of claim 3, wherein, The step of assigning weights to the intrinsic modal components according to the classification label and inputting the weighted intrinsic modal components into a long short-term memory network for noise prediction to output predicted noise values in a future period comprises: obtaining a set of intrinsic modal components with classification labels; the classification labels include impulse noise component labels and background noise component labels; assigning a weight coefficient to each intrinsic modal component according to the classification label, the intrinsic modal component with a classification label as an impulse noise component being assigned a first weight value, and the intrinsic modal component with a classification label as a background noise component being assigned a second weight value, to obtain a set of weighted intrinsic modal components; wherein the first weight value is greater than the second weight value; sorting the set of weighted intrinsic modal components in descending order of frequency to generate a time-series input sequence; splitting the time series input sequence into fixed-length training sample blocks, each containing component values of consecutive M time points; inputting the training sample blocks into a pre-constructed long short-term memory network to calculate hidden states through forget gate, input gate and output gate mechanisms; generating a predicted noise value sequence of a future time period according to the final hidden state.

7. The power line carrier noise suppression based theft prevention method for a return cable according to claim 1, wherein, The step of calculating the theft location coordinates based on the signal arrival time delay difference detected by the at least two monitoring nodes includes: obtaining signal arrival time stamps when the at least two monitoring nodes detect the theft event, and reading preset node spatial coordinate data; calculating the signal arrival time delay difference Δt between any two monitoring nodes ij generating a set of delay differences; calculating a distance difference Δd based on the set of time delay differences and a constant of propagation speed of electromagnetic waves in the backflow cable ij ; Constructing the distance difference equation set: ; In the above formula, (x, y, z) is the coordinate estimation value of the theft location, (x i ,y i ,z i ) and (x j ,y j ,z j ) are the node space coordinate data of the two monitoring nodes, respectively; solving the distance difference equation set by using the least square method, and outputting a theft location coordinate estimation value; reading a return cable path function based on a pre-constructed return cable topology space; calculating a vertical projection point of the theft location coordinate estimation value to the return cable path function, and outputting a cable length coordinate corresponding to the vertical projection point as the theft location coordinate.

8. A power line carrier noise suppression based backflow cable theft prevention method according to any one of claims 1 to 7, characterized in that, After the step of calculating the theft location coordinates based on the signal arrival time delay difference detected by the at least two monitoring nodes, the method further includes: retrieving a preset power transmission network topology database, the database storing at least line section start coordinates, end coordinates and corresponding section identifiers; traversing all line sections in the topology database to determine whether the theft location coordinates are located within the closed interval of the start coordinates and the end coordinates of any line section, and outputting a section identifier that matches successfully; based on the section identifier, querying a preset section-circuit breaker mapping table to obtain a corresponding circuit breaker control address and one or more backup power supply loop numbers; generating a power transmission network topology reconstruction instruction set and sending it to a power transmission network control system interface; wherein the instruction set includes a disconnect instruction and an activation instruction, the disconnect instruction is used to trigger a circuit breaker disconnect operation according to the circuit breaker control address, and the activation instruction is used to start backup loop power supply according to the backup power supply loop number.

9. A backflow cable anti-theft system based on power carrier noise suppression, characterized by, The system comprises: a signal acquisition and processing module for acquiring return cable carrier communication original signals through at least two monitoring nodes in a power carrier communication network, performing sliding window interception and normalization processing on the original signals, and outputting preprocessed signals; an empirical mode decomposition module for performing empirical mode decomposition on the preprocessed signals, generating a set of intrinsic mode components through envelope line construction and iterative decomposition; a classification module for calculating autocorrelation functions and partial autocorrelation functions for each intrinsic mode component, classifying the intrinsic mode components into impulse noise components or background noise components according to the attenuation characteristics, and outputting a set of intrinsic mode components containing classification labels; a preliminary noise reduction module for filtering target components from the set of intrinsic mode components based on the classification labels, and reconstructing a preliminary noise reduction signal; a noise prediction module for assigning weights to the intrinsic mode components according to the classification labels, and inputting the weighted intrinsic mode components into a long short-term memory network for noise prediction, and outputting predicted noise values of a future period; a measured noise extraction module for extracting current measured noise values from the preliminary noise reduction signal; a difference calculation module configured to calculate an absolute difference between the predicted noise value and the measured noise value and update a history difference sequence; a threshold generation module configured to calculate a dynamic event trigger threshold based on the updated history difference sequence; a judgment module configured to judge whether the absolute difference exceeds the dynamic event trigger threshold, and output a first judgment result if yes, or a second judgment result if no; a Kalman filtering module configured to input the preliminary noise-reduced signal into an unscented Kalman filtering algorithm for filtering and output an optimized noise-reduced signal in response to the first judgment result; a preliminary noise-reduced signal output module configured to directly output the preliminary noise-reduced signal as the optimized noise-reduced signal in response to the second judgment result; an eigenvalue extraction module configured to extract a short-time energy eigenvalue and a zero-crossing rate eigenvalue from the optimized noise-reduced signal; a theft judgment module configured to judge a theft event when the short-time energy eigenvalue exceeds an energy threshold or the zero-crossing rate eigenvalue exceeds a zero-crossing rate threshold; a theft positioning module configured to calculate a theft position coordinate based on a signal arrival time delay difference of the theft event detected by the at least two monitoring nodes.

10. A power line carrier noise suppressed backflow cable anti-theft system according to claim 9, wherein, The system further comprises: a database retrieval module configured to retrieve a preset power transmission network topology database, wherein the database stores at least line section start coordinates, end coordinates, and corresponding section identifiers; a section matching module configured to traverse all line sections in the topology database, determine whether the theft position coordinate is located within a closed interval of the start coordinates and the end coordinates of any line section, and output a section identifier of a matching success; a query module configured to query a preset section-circuit breaker mapping table based on the section identifier, and obtain a corresponding circuit breaker control address and one or more backup power supply loop numbers; a power transmission network control module configured to generate a power transmission network topology reconstruction instruction set and send it to a power transmission network control system interface, wherein the power transmission network topology reconstruction instruction set comprises a disconnection instruction and an activation instruction, the disconnection instruction is used to trigger a circuit breaker disconnection operation according to the circuit breaker control address, and the activation instruction is used to start backup loop power supply according to the backup power supply loop number.

Images