Grid-connected line electric energy quality on-line detection device

Abstract

According to the grid-connected line electric energy quality on-line detection device, the system comprises a broadband sensing module which is deployed on a grid-connected line to form an original analog signal; the conditioning module forms a standardized digital sampling signal; the edge calculation module receives the standardized digital sampling signal, analyzes a voltage effective value, a harmonic distortion rate, a sag depth and an event duration index in real time through a parallel calculation architecture, and forms an electric energy quality characteristic data set; the intelligent analysis module receives the electric energy quality characteristic data set to form an abnormal event report; and the communication interaction module receives the abnormal event report, encapsulates the abnormal event report into a standardized data frame through a configurable protocol stack, synchronously pushes the standardized data frame to the local monitoring terminal and the cloud management platform, receives an external control instruction and feeds back the external control instruction to the intelligent analysis module to optimize an analysis strategy. The grid-connected line electric energy quality on-line detection device can solve the problem that complex disturbance cannot be recognized in real time and cannot be accurately traced to the source.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy grid-connected power quality monitoring, and in particular to an online detection device for grid-connected line power quality. Background Art

[0002] With the large-scale integration of distributed energy sources such as wind power and photovoltaics, power line quality issues are becoming increasingly prominent. Compound disturbances such as harmonic distortion, voltage sags, and high-frequency oscillations are becoming frequent, posing a serious threat to grid security. Existing detection technologies suffer from three major flaws: The sensing layer is limited: Traditional electromagnetic transformers require power outages for installation, while electronic transformers have a narrow frequency band (typically <2kHz), making it difficult to capture high-frequency transient signals. Measurement accuracy is prone to drift in environments with strong electromagnetic interference, and the lack of online calibration mechanisms distorts the original signal. Second, insufficient analytical capabilities: Reliance on centralized servers to process massive amounts of data results in computational latency exceeding 100 milliseconds, making it incapable of meeting the requirements for real-time response to transient events. Disturbance identification often relies on threshold judgment and FFT analysis, resulting in an error rate exceeding 30% for superimposed disturbances (such as harmonics plus sags). Furthermore, the lack of topological correlation analysis makes it impossible to locate the node responsible for the disturbance. Third, a lack of system coordination: Monitoring units operate independently, lacking closed-loop linkage with control equipment (such as SVG and APF). Heterogeneous data protocols make it difficult for cloud platforms to integrate information from multiple sources, resulting in operational and maintenance response delays of several hours. Summary of the Invention

[0003] In view of the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an online detection device for the power quality of grid-connected lines, which is used to solve the problem that complex disturbances cannot be identified in real time and cannot be accurately traced. The present invention captures voltage and current signals through a non-invasive broadband sensing module without power outages, and quantifies them through a conditioning module for anti-interference; the edge computing module analyzes key indicators in parallel to generate a feature data set. The intelligent analysis module integrates lightweight convolutional neural networks with topological parameters to classify disturbances and locate source nodes; the communication interaction module encapsulates event reports and feeds back optimization instructions, forming a "perception-analysis-decision-making-optimization" closed loop, which overcomes the technical bottlenecks of poor real-time performance, ambiguous traceability, and delayed response of traditional systems.

[0004] The present invention provides an online detection device for grid-connected line power quality, comprising:

[0005] A broadband sensing module is deployed on the grid-connected line and synchronously captures line voltage and current signals through electric field coupling and magnetic field induction to form original analog signals;

[0006] The conditioning module receives the original analog signal and performs anti-aliasing filtering, impedance matching and range adaptive adjustment in sequence to form a standardized digital sampling signal;

[0007] The edge computing module receives standardized digital sampling signals and uses a parallel computing architecture to analyze voltage RMS, harmonic distortion rate, sag depth, and event duration indicators in real time to form a power quality feature data set.

[0008] Intelligent analysis module: The intelligent analysis module receives power quality feature data sets, classifies and identifies complex disturbances based on a lightweight convolutional neural network, and integrates line topology parameters to generate disturbance source tracing results and form abnormal event reports;

[0009] The communication interaction module receives abnormal event reports, encapsulates them into standardized data frames through a configurable protocol stack, and pushes them synchronously to the local monitoring terminal and the cloud management platform. It also receives external control instructions and feeds them back to the intelligent analysis module to optimize the analysis strategy.

[0010] In one embodiment of the present invention, a broadband sensing module includes a distributed probe array and an adaptive calibration unit. The distributed probe array uses orthogonally arranged electric field sensing plates and a Rogowski coil structure to eliminate interference from adjacent conductors through a space vector synthesis algorithm to form a multi-dimensional electromagnetic coupling signal. The adaptive calibration unit dynamically compensates for sensor drift based on changes in ambient temperature and humidity and line load, generates a calibration coefficient by comparison with a reference source, and corrects the amplitude-frequency characteristics of the original analog signal in real time. The corrected signal carries a calibration identifier when transmitted to the conditioning module, triggering initial parameter optimization for adaptive range adjustment.

[0011] In one embodiment of the present invention, the conditioning module integrates a noise suppression channel and a dynamic sampling engine. The noise suppression channel uses a common-mode differential decoupling circuit to separate the power frequency fundamental wave and high-frequency noise, and eliminates the signal group delay through a phase compensation filter. The dynamic sampling engine automatically switches the sampling rate according to the signal spectrum entropy value: when transient event characteristics are detected, mega-sampling is enabled, and it is reduced to kilo-sampling during steady-state operation. A sampling mode mark is added when the standardized digital sampling signal is output for the edge computing module to match the analysis algorithm.

[0012] In one embodiment of the present invention, the edge computing module has built-in heterogeneous computing cores and indicator fusers. The heterogeneous computing cores are composed of programmable logic units and multi-core processors. The voltage effective value and harmonic distortion rate are calculated in parallel through hardware acceleration circuits. At the same time, the general processor core analyzes the timing characteristics of transient events. The indicator fuser aligns the time scales of discrete calculation results to a unified time window, and generates a power quality feature data set with weight coefficients based on event correlation. The data set is compressed and packaged according to the timestamp and transmitted to the intelligent analysis module.

[0013] In one embodiment of the present invention, the intelligent analysis module is deployed with a disturbance decision tree and a topology tracing engine. The disturbance decision tree inputs the classification results of the lightweight convolutional neural network into the rule verification layer, eliminates misjudgments by comparing the historical event feature library, and outputs a disturbance type label with confidence. The topology tracing engine calculates the propagation time difference and attenuation gradient of the disturbance signal at multiple monitoring points based on the line impedance matrix and the node power flow direction, generates a traceability positioning result including the coordinates of the responsible node, and reports the merged label and positioning result of the abnormal event, and adds a governance strategy priority score.

[0014] In one embodiment of the present invention, the topology tracing engine is connected to the real-time topology update interface, which receives the switch change information of the power grid dispatching system and dynamically reconstructs the line connection relationship model. When a network topology change is detected, it automatically triggers the impedance matrix recalculation and traceability path re-verification to ensure that the positioning result matches the current operation mode. The reconstructed topology parameters are fed back to the iterative training process of the intelligent analysis module as an incremental data set.

[0015] In one embodiment of the present invention, the communication interaction module is provided with a protocol conversion gateway and an instruction arbitrator. The protocol conversion gateway converts the standardized data frame into an encapsulation format according to the requirements of the receiving end: the local monitoring terminal adopts a low-latency binary stream, the cloud management platform adopts a lightweight structured text, the instruction arbitrator verifies the digital signature and permission level of the external control instruction, and executes the conflicting instructions in a sequence according to the preset strategy. The instructions fed back to the intelligent analysis module carry a version identifier, triggering a rolling update of the analysis strategy.

[0016] In one embodiment of the present invention, a redundant diagnostic module is added to the system, which continuously monitors the heartbeat signals of each module. When a timeout failure occurs in the edge computing module or the intelligent analysis module, it automatically switches to the backup computing node and loads the most recent valid state snapshot. During the failure, the original analog signal of the broadband sensing module is directly passed to the communication interaction module to seal the original data. After the system recovers, the breakpoint resumption mechanism is triggered to reprocess the backlog data.

[0017] In one embodiment of the present invention, the intelligent analysis module is connected to an offline training platform, which receives historical power quality feature data sets and manual annotation results, and uses knowledge distillation technology to compress the deep learning model. The compressed model is pushed to the embedded storage area of ​​the edge computing module after encryption and signature. During the model update, the dual-stream verification mechanism is activated to compare the output differences of the new and old models until convergence is achieved.

[0018] The present invention also includes a target-driven intelligent optimization method for energy supply and demand, comprising:

[0019] S1: Synchronously captures line voltage and current signals through electric field coupling and magnetic field induction to form original analog signals;

[0020] S2: Receives the original analog signal and performs anti-aliasing filtering, impedance matching, and range adaptive adjustment in sequence to form a standardized digital sampling signal;

[0021] S3: Receives standardized digital sampling signals and uses a parallel computing architecture to analyze voltage RMS, harmonic distortion rate, sag depth, and event duration indicators in real time to form a power quality characteristic data set.

[0022] S4: Receives power quality feature datasets, classifies and identifies complex disturbances based on a lightweight convolutional neural network, and integrates line topology parameters to generate disturbance source tracing results and form abnormal event reports.

[0023] S5: Receive abnormal event reports, encapsulate them into standardized data frames through the configurable protocol stack, and simultaneously push them to the local monitoring terminal and cloud management platform. It also receives external control instructions and feeds them back to the intelligent analysis module to optimize the analysis strategy.

[0024] The online power quality monitoring device for grid-connected lines provided by this invention uses a non-intrusive, broadband sensing module to capture voltage and current signals without power outages. These signals are then quantified through a conditioning module for interference reduction. An edge computing module analyzes key indicators in parallel to generate a feature dataset. An intelligent analysis module integrates a lightweight convolutional neural network with topological parameters to classify disturbances and locate source nodes. A communication interaction module encapsulates event reports and provides feedback for optimization instructions, forming a closed loop of "perception-analysis-decision-making-optimization." This overcomes the technical bottlenecks of traditional systems, such as poor real-time performance, ambiguous source tracing, and delayed response. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 This is the system architecture diagram of the online detection device for grid-connected power quality;

[0027] Figure 2 The present invention is a method flow chart of a target-driven intelligent optimization method for energy supply and demand. DETAILED DESCRIPTION

[0028] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0029] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0030] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0031] See Figure 1-2 , which shows the online detection device for the power quality of a grid-connected line of the present invention. The online detection device for the power quality of a grid-connected line of the present invention includes a broadband sensing module, which is deployed on the grid-connected line and synchronously captures the line voltage and current signals through electric field coupling and magnetic field induction to form an original analog signal; a conditioning module, which receives the original analog signal and performs anti-aliasing filtering, impedance matching and range adaptive adjustment in sequence to form a standardized digital sampling signal; an edge computing module, which receives the standardized digital sampling signal and analyzes the voltage effective value, harmonic distortion rate, sag depth and event duration indicators in real time through a parallel computing architecture to form a power quality feature data set; an intelligent analysis module, which receives the power quality feature data set, classifies and identifies complex disturbances based on a lightweight convolutional neural network, and integrates line topology parameters to generate a disturbance tracing and positioning result to form an abnormal event report; a communication interaction module, which receives the abnormal event report, encapsulates it into a standardized data frame through a configurable protocol stack, and synchronously pushes it to the local monitoring terminal and the cloud management platform, and receives external control instructions to feed back to the intelligent analysis module to optimize the analysis strategy.

[0032] like Figure 1As shown, the online detection system for grid-connected line power quality protected by this patent consists of five core modules forming a highly coordinated working chain. The broadband sensing module serves as the physical interface of the system perception layer. It is directly deployed on the surface of the conductor of the energized grid-connected line or in the vicinity. It adopts a dual-mode synchronous capture mechanism of electric field coupling and magnetic field induction: the electric field coupling unit captures the electric field changes around the conductor through a plate made of a high dielectric constant composite material and converts it into a voltage sensing signal; the magnetic field sensing unit uses a Rogowski coil with no magnetic saturation characteristics to surround the conductor and generates a current sensing signal through the principle of electromagnetic induction. The two types of analog signals are transmitted to the signal conditioning circuit via a low-noise coaxial cable. In this process, the original analog signal representing the real-time electrical status of the line is formed. After being transmitted to the conditioning module, the signal undergoes three steps of processing: first, a steep-cutoff anti-aliasing filter removes noise components above the Nyquist frequency to prevent sampling aliasing distortion; second, impedance matching is performed, using a programmable resistor network to dynamically adapt the output impedance of the pre-sensing module to the input impedance of the post-sampling circuit to ensure lossless signal energy transmission; and finally, adaptive range adjustment is performed, automatically switching the operational amplifier gain level based on the signal peak detection results to ensure that the signal amplitude accurately matches the range of the analog-to-digital converter. The standardized digital sampled signal generated by this processing is transmitted to the edge computing module, which leverages a heterogeneous architecture of field programmable gate arrays and multi-core processors for parallel computing. The field programmable gate array hardware threads handle sliding window calculations of the voltage and current RMS values, while the multi-core processors simultaneously perform fast Fourier transforms of the harmonic distortion rate, extract the sag depth envelope, and time the event duration. Ultimately, these processes are integrated to generate a power quality characteristic dataset with a timestamp. Once this dataset is transferred to the intelligent analysis module, a dual analysis process is triggered. First, a pre-configured lightweight convolutional neural network model performs multi-level abstraction on the input features, outputting a composite disturbance classification probability matrix to identify complex events such as harmonic superposition and sags. Second, the network topology database utilizes line impedance parameters and node connectivity to construct a disturbance propagation path model. The responsible node is located by comparing the arrival time differences of events at multiple monitoring points. The analysis results are encapsulated as an abnormal event report containing the event type, location coordinates, and severity level, and transmitted to the communication interaction module. This module uses a configurable protocol stack to convert the report into different encapsulation formats: a header-compressed binary stream for local monitoring terminals to reduce transmission latency, and a lightweight text format with a self-describing structure for the cloud-based management platform. The module also continuously monitors the external control port. Upon receiving digitally signed optimization instructions (such as neural network weight update parameters), these instructions are fed back to the intelligent analysis module, triggering dynamic adjustment of the analysis strategy, forming a closed-loop control loop from data collection to strategy optimization.

[0033] Furthermore, based on the architecture of claim 1, the broadband sensing module further integrates a distributed probe array and an adaptive calibration unit to improve signal fidelity. The distributed probe array adopts a spatial orthogonal layout strategy: four sets of electric field sensing plates are evenly deployed around the conductor, and common-mode interference is offset by a differential circuit; three layers of Rogowski coils of different diameters are nested axially, covering the fundamental, harmonic, and transient frequency bands respectively. The output signals of each probe are input into a space vector synthesis processor, which uses the geometric relationship of the conductor position to establish an electromagnetic field distribution model, and uses a least squares algorithm to eliminate the coupling interference of adjacent phase conductors to generate an electromagnetic coupling signal that integrates multi-physical field information. The adaptive calibration unit constructs an environmental parameter compensation system: the temperature sensor monitors the internal junction temperature of the probe in real time, the humidity sensor detects the condensation state of the outer insulation surface, and the load current transformer collects the line working current. The above parameters are input into the drift compensation model, which is based on a three-dimensional lookup table (temperature-humidity-load as independent variables) established in the previous calibration experiment, and outputs the amplitude correction coefficient and phase compensation angle. At the same time, a high-stability reference source is set up, and standard test signals are injected regularly for self-test. When it is detected that the deviation between the actual output and the theoretical value exceeds the threshold, the calibration coefficient is triggered to iteratively update. The original analog signal after dynamic compensation is appended with an identifier containing the calibration timestamp and environmental parameters. This identifier is read first after being transmitted to the conditioning module and is used to initialize the gain prediction algorithm for adaptive range adjustment: the current signal amplitude range is predicted based on the historical load curve, and the optimal amplification factor is preset to reduce adjustment hysteresis. This design enables the system to maintain a full-scale measurement error of less than 0.5% under ambient temperature fluctuations of -40°C to 85°C, significantly improving the monitoring reliability in extremely cold / high-temperature areas.

[0034] Specifically, the function of the conditioning module of claim 1 is enhanced, focusing on solving the problems of signal extraction and sampling resource optimization in strong noise environments. The noise suppression channel adopts a three-stage processing architecture: the primary common-mode differential decoupling circuit uses the high common-mode rejection ratio characteristics of the instrumentation amplifier to eliminate the common-mode noise introduced by the distributed capacitance between the transmission line and the ground; the secondary is designed with a phase compensation filter with nonlinear phase-frequency characteristics to maintain a constant group delay within the passband to avoid waveform distortion of transient events; the final stage is configured with an adaptive notch filter, which tracks the system base frequency through a phase-locked loop and dynamically adjusts the stopband center frequency to eliminate background harmonic interference. The dynamic sampling engine constructs an analysis decision chain based on signal characteristics: first, the spectral entropy value of the standardized digital sampling signal is calculated, and the basic sampling rate is maintained when the entropy value is lower than the set threshold (indicating that the signal energy is concentrated in the base frequency); when a sudden increase in high-frequency component energy is detected or the waveform mutation index exceeds the standard, it is determined that a transient event has occurred and immediately switched to the mega-sampling mode. This mode uses oversampling technology in conjunction with a digital downsampling filter to expand the effective bandwidth while ensuring equivalent bit accuracy. The sampling pattern tag generation logic is linked to an event signature detector: if the signal change rate exceeds the limit for 100 microseconds, it is marked as a "voltage sag," and if the signal-to-noise ratio drops sharply in the high-frequency band, it is marked as a "harmonic oscillation." This tag information is encapsulated synchronously with the sampled data. Upon receiving the standardized digital sampled signal with the tag, the edge computing module initiates an algorithm matching mechanism: For basic sampled data, an optimized floating-point arithmetic library is used to calculate steady-state indicators; for megasampled data, a dedicated transient analysis pipeline is activated, using wavelet transforms to extract microsecond-level event features and a time series compression algorithm to reduce data transmission. This dynamic mechanism reduces the system's processing resource usage by 60 percent during steady-state operation and can detect voltage sags as short as five microseconds during transient event capture.

[0035] In one embodiment of the present invention, the core innovation of the edge computing module lies in the collaborative architecture of heterogeneous computing cores and indicator aggregators. The heterogeneous computing core consists of a programmable logic unit and a multi-core processor, with a physical layer division of labor: the programmable logic unit has a built-in hardened fast Fourier transform pipeline and a sliding window root mean square calculator. The former eliminates memory access conflicts of butterfly operations through address flipping addressing technology, and the latter uses the CORDIC algorithm to iteratively update the effective values ​​of voltage and current, achieving a result refresh every sampling cycle; the multi-core processor is allocated three dedicated thread pools: thread pool one performs sag depth detection, identifying the start and end points of events through an envelope extraction algorithm combined with an adaptive threshold comparator; thread pool two calculates harmonic distortion rate, suppressing spectrum leakage based on windowed interpolation Fourier transform; thread pool three monitors three-phase imbalance, using the symmetrical component method to decompose positive and negative sequence components. The indicator fusion unit, serving as the data integration hub, first adds a high-precision Beidou time stamp to each thread output and then aggregates relevant indicators according to an event-triggered mechanism. When a sag event is detected, it automatically correlates the harmonic distortion data and imbalance data within the same time window, generating a power quality characteristic dataset with confidence intervals using preset weight coefficients. This dataset is stored in a segmented circular buffer, with compression and packaging triggered every 256 data sets. A combination of dictionary and differential encoding strategies is used to reduce the amount of transmitted data to less than 40 percent of its original size. The data is then transmitted to the intelligent analysis module via a high-speed serial interface. This architecture reduces the latency of voltage RMS calculation to milliseconds and reduces the response time for sag event identification to within one-tenth of the power frequency cycle, meeting the most stringent real-time requirements.

[0036] like Figure 1As shown, the intelligent analysis module builds a two-layer analysis engine based on the data set output by claim 4. The disturbance decision tree engine first receives the preliminary classification results of the lightweight convolutional neural network - the network uses depthwise separable convolution to replace the standard convolution layer, reducing the number of parameters to one-fifth of the traditional model. The input layer receives the time series matrix of the standardized power quality characteristic data set, extracts spatial features through four layers of convolution, and outputs the disturbance probability vector through the global pooling layer. The probability vector is input into the rule verification layer for credibility verification: the rule base presets 320 expert experience rules (for example, "If the third harmonic increase is greater than 50% of the fundamental decrease during the voltage sag, it is determined to be accompanied by harmonic oscillation"). When the probability output by the neural network is lower than the threshold or conflicts with the rule base, it triggers historical event matching based on the dynamic time warping algorithm, and retrieves similar waveforms from the 100,000-level case library to re-label the type. The topology tracing engine operates simultaneously: it acquires line impedance matrix and node injection power data in real time from the grid energy management system, building a full-network admittance model that includes distributed generation access points. Upon receiving a disturbance event notification, it extracts the disturbance waveform characteristics of multiple monitoring points, calculates the signal propagation time difference using a cross-correlation function, and inverts the disturbance source location using the line propagation velocity constant. For complex network structures, it uses an improved particle swarm optimization algorithm to solve the attenuated gradient equations, keeping the positioning error within the substation interval level. The resulting abnormal event report contains four dimensions: a verified disturbance type label, the coordinates of the responsible node, a confidence score, and a recommended priority for the remediation strategy (e.g., "Prioritize adjustment of the active output of the No. 3 photovoltaic inverter").

[0037] Furthermore, the enhanced topology tracing engine solution focuses on dynamic grid reconstruction scenarios. The real-time topology update interface establishes a data subscription channel with the dispatching master station, continuously monitoring circuit breaker opening and closing signals and disconnector position change information. When a network structure change event is received (such as "connector switch S112 changes from open to closed"), a three-link mechanism is triggered: Mechanism 1 calls the network topology analyzer to automatically update the node connection relationship diagram based on the switch state change, delete the disconnected branch circuits, and add new connection paths; Mechanism 2 starts the impedance matrix online calculator, calls the unit length resistance reactance values ​​in the line parameter library based on the latest topology, and reconstructs the node admittance matrix of the entire network; Mechanism 3 activates the traceability path validator, selects typical disturbance cases within the past three hours for backtracking simulation, compares the deviation of the positioning results of the new and old models, and sends a model calibration request to the intelligent analysis module if the deviation exceeds the safety margin. To ensure data continuity during the transition period, a topology version management area is set up to save the last ten valid topology snapshots and corresponding timestamps. When analyzing historical events, it automatically matches the valid network structure at the time of the event. The reconstructed parameters are pushed to the intelligent analysis module in the form of an incremental update package, triggering the online learning of its internal prediction model: a transfer learning strategy is used to freeze the weights of the convolutional layer, and only the fully connected layer is fine-tuned to adapt to the new topological features. Adversarial samples are introduced into the training process to enhance robustness. Shadow mode parallel verification is enabled during model updates, and the new model is switched after the accuracy rate meets the requirements.

[0038] like Figure 2 The figure shows a target-driven intelligent optimization method for energy supply and demand of the present invention. S1: receiving input high-level target instructions; S2: receiving the target instructions, parsing the received high-level target instructions into quantitative target parameters, and generating dynamic priority weights and constraints according to real-time environmental data to form a target quantization signal; S3: receiving the target quantization signal, constructing a reward function guided by target priority based on the signal, generating an optimization strategy through interactive learning with the environment, and outputting a strategy execution signal; S4: converting the strategy execution signal into a control instruction for energy equipment; S5: forming an environmental state signal and synchronously transmitting it to the dynamic multi-target quantization module and the reinforcement learning intelligent decision-making module to drive the adaptive adjustment of the target weight and the iterative update of the strategy.

[0039] Furthermore, the innovation of the communication interaction module lies in protocol adaptive conversion and command security management. The protocol conversion gateway establishes a dual-channel processing pipeline: the local channel connects to the monitoring terminal and converts abnormal event reports into a streamlined binary stream. This stream removes the text description field, uses a predefined code to represent the event type (such as 0x1A for "three-phase voltage sag"), quantizes the coordinate information into a 16-bit integer, and ensures data integrity through CRC32 checksum. The cloud channel adopts a layered encapsulation strategy: the first layer is a lightweight structured text framework that defines 16 key fields (including event time, location hash value, confidence level, etc.); the middle layer embeds a natural language description template of the governance policy recommendation; the outer layer adds a digital signature and timestamp. The instruction arbitrator is designed as a three-level filtering architecture: the first level verifies the legitimacy of the digital certificate of the external control instruction, allowing only devices signed by an authorized CA organization to access; the second level parses the instruction semantics, identifies conflicting operations (such as receiving the "increase sampling rate" and "reduce power consumption" instructions at the same time), and sorts the execution priority according to the preset policy table (real-time instructions are prioritized over energy-saving instructions); the third level is pre-execution simulation, which evaluates changes in system status after instruction execution and blocks requests that may cause functional abnormalities. All adopted instructions are fed back to the intelligent analysis module with attached version serial numbers, triggering a rolling update of the analysis strategy: the policy engine maintains a version repository and retains the five most recent valid versions for rapid rollback; the update process adopts a canary release model, first diverting 10% of the data to the new strategy to verify the effect, and then fully deploying it after confirmation. This mechanism enables the system to maintain a 99.9% instruction execution reliability rate even in complex industrial electromagnetic environments.

[0040] The present invention's online power quality monitoring device for grid-connected lines uses a non-invasive, broadband sensing module to capture voltage and current signals without power outages. These signals are then quantified through a conditioning module for interference resistance. An edge computing module analyzes key indicators in parallel to generate a feature dataset. An intelligent analysis module integrates a lightweight convolutional neural network with topological parameters to classify disturbances and locate source nodes. A communication interaction module encapsulates event reports and provides feedback on optimization instructions, forming a closed "perception-analysis-decision-making-optimization" loop. This overcomes the technical bottlenecks of traditional systems, such as poor real-time performance, ambiguous source tracing, and delayed response.

[0041] Therefore, the problem of complex disturbances being unable to be identified in real time and accurately traced is solved by the online detection device for grid-connected line power quality of the present invention.

[0042] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. The online detection device for power quality of grid-connected lines is characterized by: include: A broadband sensing module is deployed on the grid-connected line and synchronously captures line voltage and current signals through electric field coupling and magnetic field induction to form original analog signals; A conditioning module receives the original analog signal and sequentially performs anti-aliasing filtering, impedance matching, and range adaptive adjustment to form a standardized digital sampling signal; An edge computing module receives the standardized digital sampling signal and analyzes voltage RMS, harmonic distortion rate, sag depth, and event duration indicators in real time through a parallel computing architecture to form a power quality feature data set; An intelligent analysis module receives the power quality characteristic data set, classifies and identifies the composite disturbance based on a lightweight convolutional neural network, and integrates the line topology parameters to generate a disturbance source tracing and location result, thereby forming an abnormal event report; The communication interaction module receives the abnormal event report, encapsulates it into a standardized data frame through a configurable protocol stack, pushes it synchronously to the local monitoring terminal and the cloud management platform, and receives external control instructions to feed back to the intelligent analysis module to optimize the analysis strategy.

2. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The broadband sensing module includes a distributed probe array and an adaptive calibration unit. The distributed probe array adopts an orthogonally arranged electric field sensing plate and Rogowski coil structure, eliminates interference from adjacent conductors through a space vector synthesis algorithm, and forms a multi-dimensional electromagnetic coupling signal. The adaptive calibration unit dynamically compensates for sensor drift based on changes in ambient temperature and humidity and line load, generates a calibration coefficient by comparison with a reference source, and corrects the amplitude-frequency characteristics of the original analog signal in real time. The corrected signal carries a calibration identifier when transmitted to the conditioning module, triggering initial parameter optimization for adaptive range adjustment.

3. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The conditioning module integrates a noise suppression channel and a dynamic sampling engine. The noise suppression channel uses a common-mode differential decoupling circuit to separate the power frequency fundamental wave from high-frequency noise, and eliminates signal group delay through a phase compensation filter. The dynamic sampling engine automatically switches the sampling rate based on the signal spectrum entropy value: when transient event characteristics are detected, mega-sampling is enabled, and it is reduced to kilo-sampling during steady-state operation. A sampling mode mark is added when the standardized digital sampling signal is output for the edge computing module to match the analysis algorithm.

4. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The edge computing module has built-in heterogeneous computing cores and indicator fusers. The heterogeneous computing cores are composed of programmable logic units and multi-core processors. The voltage effective value and harmonic distortion rate are calculated in parallel through hardware acceleration circuits. At the same time, the general processor core analyzes the timing characteristics of transient events. The indicator fuser aligns the time scales of discrete calculation results to a unified time window and generates a power quality feature data set with weight coefficients based on event correlation. The data set is compressed and packaged according to the timestamp and transmitted to the intelligent analysis module.

5. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The intelligent analysis module is deployed with a disturbance decision tree and a topology tracing engine. The disturbance decision tree inputs the classification results of the lightweight convolutional neural network into the rule verification layer, eliminates misjudgments by comparing the historical event feature library, and outputs a disturbance type label with confidence. The topology tracing engine calculates the propagation time difference and attenuation gradient of the disturbance signal at multiple monitoring points based on the line impedance matrix and the node power flow direction, and generates a traceability positioning result including the coordinates of the responsible node. The abnormal event report merges the label and positioning result, and adds a governance strategy priority score.

6. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The topology tracing engine is connected to a real-time topology update interface that receives switch position change information from the power grid dispatching system and dynamically reconstructs the line connection relationship model. When a network topology change is detected, it automatically triggers impedance matrix recalculation and traceability path reverification to ensure that the positioning result matches the current operating mode. The reconstructed topology parameters are fed back as an incremental data set to the iterative training process of the intelligent analysis module.

7. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The communication interaction module is equipped with a protocol conversion gateway and a command arbitrator. The protocol conversion gateway converts the standardized data frame into an encapsulation format according to the requirements of the receiving end: the local monitoring terminal uses a low-latency binary stream, and the cloud management platform uses lightweight structured text. The command arbitrator verifies the digital signature and permission level of the external control instruction, and executes the conflicting instructions in order according to the preset strategy. The instructions fed back to the intelligent analysis module carry a version identifier, triggering a rolling update of the analysis strategy.

8. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The system adds a redundant diagnostic module, which continuously monitors the heartbeat signals of each module. When a timeout occurs in the edge computing module or the intelligent analysis module, it automatically switches to the backup computing node and loads the most recent valid status snapshot. During the failure, the original analog signal of the broadband sensing module is directly passed to the communication interaction module for raw data sealing. After the system recovers, the breakpoint resumption mechanism is triggered to reprocess the backlog data.

9. The online detection device for power quality of grid-connected lines according to claim 1, characterized in that: The intelligent analysis module is connected to an offline training platform, which receives historical power quality feature data sets and manual annotation results, and uses knowledge distillation technology to compress the deep learning model. The compressed model is pushed to the embedded storage area of ​​the edge computing module after encryption and signature. During the model update, the dual-stream verification mechanism is activated to compare the output differences of the new and old models until convergence is achieved.

10. A target-driven intelligent optimization method for energy supply and demand according to any one of claims 1 to 9, characterized in that: include: S1: Synchronously captures line voltage and current signals through electric field coupling and magnetic field induction to form original analog signals; S2: Receive the original analog signal, perform anti-aliasing filtering, impedance matching and range adaptive adjustment in sequence to form a standardized digital sampling signal; S3: receiving the standardized digital sampling signal, and analyzing the voltage effective value, harmonic distortion rate, sag depth, and event duration indicators in real time through a parallel computing architecture to form a power quality characteristic data set; S4: receiving the power quality feature dataset, classifying and identifying the composite disturbance based on a lightweight convolutional neural network, and fusing the line topology parameters to generate a disturbance source tracing and location result, thereby forming an abnormal event report; S5: Receive the abnormal event report, encapsulate it into a standardized data frame through a configurable protocol stack, and simultaneously push it to the local monitoring terminal and the cloud management platform, and receive external control instructions to feed back to the intelligent analysis module to optimize the analysis strategy.

Images