A park power grid harmonic pollution graph attention tracing method

Abstract

The application provides a park power grid harmonic pollution graph attention tracing method, relates to the technical field of power quality analysis and energy efficiency management, and comprises the following steps: synchronously collecting time sequence operation state parameters of each power utilization equipment and power quality parameters of a power grid public connection point; then, a harmonic characteristic dynamic coding vector of each equipment under the current working condition is generated, and power grid harmonic data is reconstructed in a phase space to separate different pollution modal components; subsequently, a power grid dynamic topology graph taking the equipment as a node and the electrical connection as an edge is constructed, and the coding vector is input into a graph attention tracing model as a node characteristic; finally, taking the separated harmonic pollution modal as a target, the model performs reverse attention calculation along a topology path, and the contribution probability of each equipment to the pollution modal is output, so that accurate positioning and responsibility quantification of the pollution source are realized.

Description

Technical Field

[0001] This application belongs to the field of power quality analysis and energy efficiency management technology, specifically involving a graph attention source tracing method for harmonic pollution in a power grid in a park. Background Technology

[0002] With the deepening of the global energy transition and the proposal of dual-carbon goals, industrial parks, as concentrated areas of energy consumption and carbon emissions, have made refined energy efficiency management a key link in achieving energy conservation and emission reduction. Traditional industrial park energy efficiency management systems mainly rely on metering devices such as smart meters to monitor, statistically analyze, and collect data on the total and individual consumption of energy media such as water, electricity, gas, and heat. Their focus is on the consumption of active electrical energy and the comparison of macro-level energy efficiency indicators.

[0003] At a more refined level, power quality, especially harmonic pollution, is increasingly attracting attention as a form of hidden energy waste. The power grid in industrial parks contains numerous nonlinear loads such as frequency converters, rectifiers, and uninterruptible power supplies (UPS), which inject harmonic and interharmonic currents into the grid during operation. This harmonic pollution causes additional heat and losses in equipment such as transformers, motors, and cables, triggers capacitor bank resonance or damage, and even interferes with the normal operation of precision instruments. This invisibly increases the overall energy consumption and equipment maintenance costs of the entire park, constituting a hidden waste that is difficult to measure directly with traditional electricity meters.

[0004] To address harmonic pollution, existing technological solutions primarily focus on two levels: first, at the monitoring level, power quality analyzers are installed at common junction points or critical branches to monitor and issue warnings for indicators such as total harmonic distortion (THD) and harmonic content; second, at the remediation level, centralized or localized passive / active filters are typically used for harmonic suppression. However, these existing methods have significant limitations: firstly, monitoring and remediation solutions are often reactive, taking action only after pollution levels exceed limits, lacking foresight. More importantly, existing technologies generally cannot accurately identify the specific sources of harmonic pollution. Because harmonics generated by multiple devices can vector-superimpose in the public power grid, existing spectrum analysis techniques (such as Fast Fourier Transform) can only resolve the harmonic components present in the grid, but cannot dynamically correlate a specific harmonic mode in the time domain with which specific device generated it under what operating conditions, nor can they trace its propagation path spatially. This leads to a "tragedy of the commons" dilemma for park managers: although they can perceive the existence and harm of pollution, they cannot accurately link pollution responsibility and corresponding remediation costs to specific polluting equipment and their respective companies. Therefore, existing harmonic control methods mostly adopt a "one-size-fits-all" centralized filtering approach, which is not only uneconomical but also fails to incentivize pollution source equipment to proactively optimize and fundamentally solve the problem.

[0005] Furthermore, while a few studies have attempted to combine equipment operation data with grid data for analysis, these have mostly remained at the level of simple time series comparisons or correlation coefficient calculations, failing to delve into the complex nonlinear coupling relationship between the internal dynamic operating states of the equipment (such as load rate, control commands, and mechanical vibration) and the harmonic emission characteristics of the external power grid. The lack of modeling for this dynamic coupling mechanism, and the failure to incorporate the physical topology of the industrial park's power grid into the analytical framework, results in a lack of uniqueness and accuracy in pollution source tracing, making it difficult to apply to practical energy efficiency assessments and refined governance. Summary of the Invention

[0006] This application provides a graph attention source tracing method for harmonic pollution in industrial park power grids to solve one of the aforementioned technical problems.

[0007] The technical solution adopted in this application is as follows: This application provides a graph attention source tracing method for harmonic pollution in a power grid within a park, including: Synchronously collect equipment-level time-series operating status parameters of each electrical device in the park, as well as grid-level power quality time-series parameters of the park's common connection points; For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters. Phase space reconstruction is performed on the harmonic data in the power grid-level power quality time series parameters to separate harmonic pollution mode components from different sources; A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism. Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

[0008] According to one embodiment of this application, the equipment-level timing operation status parameters include at least one of the following: speed and load rate commands of the electrical drive equipment, start / stop and process signals of the production equipment, and vibration and temperature monitoring data of the equipment.

[0009] According to one embodiment of this application, the phase space reconstruction of harmonic data in grid-level power quality time-series parameters specifically includes: Based on the time series of the harmonic data, a high-dimensional phase space trajectory is reconstructed using the time delay embedding method. In the high-dimensional phase space trajectory, cluster analysis is used to separate trajectory clusters that characterize different dynamic properties, and each trajectory cluster corresponds to a harmonic pollution mode component.

[0010] According to one embodiment of this application, the source tracing model is a spatiotemporal graph attention network, which is used to learn the dynamic propagation and superposition relationship of the influence of node features on the harmonic state of monitoring points under temporal changes.

[0011] According to one embodiment of this application, it also includes: Based on the contribution probability, the additional losses and carbon emissions caused by the harmonic pollution mode components are proportionally allocated to the corresponding electrical equipment to quantify their responsibility for hidden waste.

[0012] According to one embodiment of this application, the quantification of the additional loss is calculated based on the amplitude and frequency of the harmonic pollution mode component and a preset harmonic loss model.

[0013] A second aspect of this application provides a graph attention source tracing system for harmonic pollution in a power grid, comprising: The data synchronization acquisition module is used to synchronously acquire the equipment-level time-series operating status parameters of each electrical device in the park, as well as the grid-level power quality time-series parameters of the park's common connection points; The equipment status coding module is used to generate a dynamic status coding vector for each electrical device, based on its equipment-level time-series operating status parameters, reflecting the harmonic emission characteristics under its current operating conditions. The harmonic mode analysis module is used to reconstruct the phase space of the harmonic data in the power grid-level power quality time series parameters and separate the harmonic pollution mode components from different sources. The topology model building module is used to construct a dynamic topology diagram of the park's power grid, with electrical equipment as nodes and electrical connections as edges. The pollution source tracing calculation module is used to input the dynamic state encoding vector as the feature of the corresponding node into the source tracing model based on the graph attention mechanism, and to use the harmonic pollution mode component as the target. The source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode component, so as to complete the pollution source tracing.

[0014] According to one embodiment of this application, it also includes: The hidden waste quantification module is used to proportionally allocate the additional losses and carbon emissions caused by the harmonic pollution mode components to the corresponding electrical equipment based on the contribution probability, so as to quantify their responsibility for hidden waste.

[0015] A third aspect of this application provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps described in the method.

[0016] A fourth aspect of this application provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method as described.

[0017] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows: This application generates a dynamic state-coded "harmonic fingerprint" for each device by synchronously collecting equipment operating status and power grid quality data, which changes in real time according to its operating conditions. Simultaneously, it separates different harmonic pollution mode components from the mixed power grid signal. By constructing a dynamic power grid topology map and using a graph attention model for inverse calculation, it can dynamically output the contribution probability of each device to a specific pollution based on the matching degree between the device's "fingerprint" and the pollution mode, as well as their electrical connection relationships. This directly transforms the general statement "harmonic exceedance in the industrial park" into a quantitative conclusion that "device A is responsible for X% of harmonic C under operating condition B," providing a crucial basis for refined energy efficiency management.

[0018] This study reveals the implicit coupling between the internal state of the equipment and external harmonic emissions, enhancing the depth of the analysis. The source-tracing capability of this application is rooted in the nonlinear correlation modeling of equipment operating conditions and harmonic characteristics. The dynamic state encoding vector profoundly reflects the influence of the equipment's internal working mechanism on its external harmonic emission characteristics, enabling the model not only to locate the pollution source but also to explain the specific operating conditions under which pollution occurs, providing crucial insights for eradicating pollution from the operational optimization level.

[0019] This has driven a shift in harmonic control from a "passive response" to a "proactive and precise" approach. Based on accurate source tracing results, management strategies can shift from traditional centralized filtering across the entire network to targeted early warning, assessment, or treatment of specific high-pollution source equipment. This achieves a transformation from post-event remediation to in-process monitoring and pre-event prevention, enabling more economical and effective reduction of "hidden waste" at the source.

[0020] This provides core technological support for constructing enterprise-level "power quality responsibility profiles." Based on the pollution contribution probability of each enterprise's equipment, a "grid cleanliness index" can be generated, thereby fairly and accurately allocating the additional losses caused by harmonics to carbon emissions. This breaks the "tragedy of the commons" dilemma by making implicit costs explicit, incentivizing enterprises to proactively optimize electricity use, and contributing to the overall energy efficiency improvement and green transformation of the industrial park. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1A flowchart illustrating a graph attention source tracing method for harmonic pollution in a power grid within a park, provided as an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0022] Figure label: 810, Processor; 820, Communication interface; 830, Memory; 840, Communication bus. Detailed Implementation

[0023] To more clearly illustrate the overall concept of this application, a detailed explanation is provided below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below. It should be noted that, unless otherwise specified, the embodiments of this application and the features thereof can be combined with each other.

[0025] In this application, unless otherwise expressly specified and limited, the "above" or "below" of the second feature can mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0026] Example 1 like Figure 1 As shown, a graph attention-based source tracing method for harmonic pollution in a power grid within a park includes: The system synchronously collects equipment-level time-series operating status parameters of various electrical devices within the park, as well as grid-level power quality time-series parameters of the park's common connection points.

[0027] Specifically, equipment-level time-series operating status parameters refer to time-series data that characterize the internal operating conditions of various monitored electrical devices (such as frequency converters, motors, compressors, and production lines) within the industrial park, collected directly or indirectly. These parameters typically include, but are not limited to: electrical quantities (such as three-phase current, voltage, active / reactive power, and power factor), mechanical quantities (such as speed, torque, vibration amplitude, and temperature), and control system status signals (such as load rate setpoints, operating modes, and start / stop switching quantities). These parameters directly reflect the internal working mode and load conditions of the equipment.

[0028] Power grid-level power quality time-series parameters refer to time-series data collected by power quality monitoring devices at the common connection point (i.e., the main incoming line or key bus node) of the power supply system in the industrial park. At its core are electrical quantities containing harmonic components, mainly involving waveform data of voltage and current, or time-series indicators such as the amplitude, phase, content, and total harmonic distortion rate of each harmonic (and interharmonic) obtained after analysis. These parameters macroscopically reflect the pollution disturbances experienced by the power grid.

[0029] The synchronization of the two is the basis for ensuring that the operating events of a specific device can be accurately correlated with specific harmonic disturbances in the power grid at a specific point in time.

[0030] For example, in a painting workshop of an automotive manufacturing park, this step requires the following simultaneous operations: On one hand, from a frequency converter driving a ventilation fan, its output frequency (e.g., 45Hz), load torque percentage (e.g., 80%), DC bus voltage, etc., are collected in real time to form its "equipment-level timing operation status parameters"; from an air compressor, its loading / unloading status signal, motor vibration amplitude, etc., are collected. On the other hand, at the low-voltage side outlet of the distribution transformer supplying power to the painting workshop (i.e., a key park common connection point), a power quality analyzer is used to simultaneously collect the total harmonic distortion (THD) of the three-phase voltage and current, as well as the amplitude and phase of the 5th, 7th, and 11th characteristic harmonics, etc., which constitute "grid-level power quality timing parameters". All data is tagged with a uniform high-precision time stamp, so that at the moment "13:05:30", the system knows that the frequency converter is in a high load output state and that there is a sudden increase in the content of the 5th harmonic in the power grid at the same time, which provides the possibility for establishing the causal relationship between the two.

[0031] For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters.

[0032] Specifically, this step takes the continuously changing time-varying device-level operating state parameters of each electrical device as input, and outputs a fixed-dimensional numerical vector through an encoding process. This vector is designed to characterize the tendency of the device, determined by its internal electrical characteristics, to inject harmonic current into the power grid under its current specific operating state, and the content of this vector changes dynamically with the device's operating state. For example, for a servo drive of an injection molding machine, its device-level timing operating state parameters might include: the motor's real-time speed, torque current, DC bus voltage, and drive heatsink temperature. These parameters are input into a pre-trained encoder model, which learns the intrinsic correlation between these state variables and harmonic emission modes. When the driver is in a low-speed, high-torque state, the encoder generates a specific vector based on the current state parameters (e.g., 200 rpm, 80% of rated torque current). This vector points to a characteristic region in the model space representing "enhanced low-order harmonic (e.g., 5th, 7th) emission." When the driver switches to high-speed, no-load operation, based on the new state parameters (e.g., 3000 rpm, 10% of rated torque current), the encoder generates another different vector, which may point to a characteristic region of "prominent high-order harmonics and switching frequency sideband interference." Thus, each dynamic state-coded vector acts as a "harmonic behavior summary" of the device at a given moment.

[0033] The harmonic data in the time series parameters of the power grid-level power quality are reconstructed in phase space to separate the harmonic pollution mode components from different sources.

[0034] Specifically, this step applies the phase space reconstruction method from nonlinear dynamics to analyze the power grid harmonic time series data, which appears as a single mixed waveform, measured from the point of common coupling. The aim is to decompose the seemingly mixed overall harmonic signal into several essentially independent components. Specifically, this method takes a one-dimensional harmonic amplitude (or phase) time series as input and, by selecting appropriate delay times and embedding dimensions, reconstructs this one-dimensional sequence into a trajectory in a high-dimensional space. In this reconstructed high-dimensional phase space, the inherent geometric structure of the data point distribution becomes apparent. Since harmonic components generated by different physical sources (i.e., different polluting devices) have their own unique dynamic evolution laws, these laws determine their flow direction and aggregation characteristics in the phase space trajectory. Therefore, by analyzing the reconstructed phase space trajectory, for example, by using clustering algorithms to identify different dense regions or evolution paths in the trajectory, the originally superimposed mixed signal in the time and frequency domains can be separated into several "harmonic pollution mode components" corresponding to different pollution sources. Each mode component characterizes the overall characteristic pattern of the harmonic response excited by a specific pollution source in the power grid.

[0035] For example, the total harmonic current data measured at the public connection point in the park is actually a vector superposition of harmonics emitted by various loads of different natures, such as the inverter unit in Workshop A, the air conditioning group control system in Building B, and the precision instruments in Laboratory C. Traditional spectrum analysis can only show the presence of the 5th, 7th, and 11th harmonics, but cannot distinguish which type of load these harmonics originate from. Through phase space reconstruction in this step, after mapping this total harmonic current time series data to a high-dimensional space, its trajectory may show three geometrically distinguishable clusters or evolution paths. After separation, three independent modal components can be obtained: Modal component one may exhibit the characteristic of the 5th harmonic being dominant and its amplitude fluctuating slowly, which corresponds to the load change pattern of the inverter unit in Workshop A during continuous production; Modal component two may exhibit the intermittent pulse characteristics accompanied by the 7th and 11th harmonics, which corresponds to the periodic start-stop pattern of the air conditioning compressor group in Building B; Modal component three may exhibit the characteristic of a wide-bandwidth, low-amplitude continuous spectrum, which corresponds to the comprehensive interference characteristics of various precision instruments in Laboratory C. In this way, the macroscopic mixed signal is decomposed into several independent components associated with potential pollution sources.

[0036] A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism.

[0037] Specifically, this step first abstracts each electrical device as a graph node based on the actual physical connection structure of the park's power supply network, and abstracts the electrical connections between devices (such as connections via the same busbar, feeder, or transformer) as edges connecting the nodes, thereby constructing a graph structure representing the park's power grid topology. This topology graph is dynamic, meaning it can contain attributes reflecting real-time electrical states; for example, the on / off state of a switch affects the connectivity of edges. Then, the dynamic state encoding vector generated for each electrical device in the previous step is assigned as a feature attribute to the corresponding graph node. Finally, this complete graph structure, containing node features and edge connections, is input into a source tracing model based on a graph attention mechanism. The model works by calculating the mutual influence weights between any two nodes in the graph through an attention mechanism. In particular, it learns to start from the node representing the pollution monitoring results (such as the common connection point) and trace back along the edge direction to assess the degree of influence of upstream device nodes on the harmonic state of downstream monitoring points.

[0038] For example, in a small power distribution system containing an air compressor, a fan inverter, and lighting circuits, the constructed topology graph contains three device nodes. The "air compressor" node and the "fan inverter" node are connected to the "bus M" node via the edge "feeder A". The "lighting circuit" node is also connected to "bus M" via the edge "feeder B". "Bus M" is then connected to the monitoring node representing the "point of common connection" via an edge. At a certain moment, the air compressor is under load, the fan inverter is operating at 40Hz, and the lighting circuit is fully on. Their respective state encoding vectors (e.g., representing "start-up impulse harmonic characteristics", "intermediate frequency operation harmonic characteristics", and "linear load characteristics") are assigned to the corresponding three device nodes. The graph attention source tracing model receives this graph and targets the 5th harmonic mode detected by the "point of common connection". The model might calculate that the path from the "common connection point" node to the "air compressor" node has the highest attention weight. Combining this with the feature vector of the "air compressor" node, it could determine that the 5th harmonic at the current moment is mainly contributed by the air compressor, which is currently in the loading and startup phase. If the air compressor unloads and the fan inverter accelerates to 50Hz in the next moment, the feature vectors of each node will be updated, and the model will recalculate. The determination of the dominant pollution source node may then change.

[0039] Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

[0040] Specifically, this step uses a specific harmonic pollution mode component separated from the mixed harmonics of the power grid in the previous step as the analysis target, instructing the source tracing model to perform calculations on the constructed dynamic topology graph. The calculation direction is reversed, that is, starting from the node representing the harmonic monitoring location (where the target is located), it moves against the path of electrical connection towards the upstream electrical equipment nodes to transmit information and perform calculations. The core calculation of the model is attention weight allocation, which evaluates the correlation between the characteristics (i.e., its dynamic state encoding vector) of each upstream electrical equipment node in the graph and the current target harmonic pollution mode component through an algorithm, and simultaneously considers the path and impedance relationship of the electrical connection between nodes, comprehensively calculating the probability value of each upstream equipment node's influence on the specific harmonic pollution mode component. This probability value is the contribution probability. Finally, the contribution probability set of all relevant electrical equipment is output, thereby completing the source tracing process from the observed pollution phenomenon to the potential pollution source and the quantification of its responsibility.

[0041] For example, a target harmonic pollution mode component is identified as "250Hz (5th harmonic) with a phase characteristic of Φ_A". The source tracing model starts from the common connection point node monitoring this component and traces backward in the topology graph. It traverses upstream paths: one path leads to the "air compressor" and "fan inverter" nodes via "bus M", and the other leads to the "lighting circuit" node. The model performs calculations: First, it compares the target mode (250Hz, Φ_A) with the harmonic emission modes implied by the feature vectors of each node. It finds that the typical harmonic mode associated with the current dynamic state encoding vector of the "air compressor" node (representing "high load start-up state") contains a strong 5th harmonic component with a phase characteristic highly similar to Φ_A; meanwhile, the 5th harmonic component is weak in the mode associated with the current feature vector of the "fan inverter" node (representing "medium-speed stable operation"); and the feature vector of the "lighting circuit" node shows that it generates almost no 5th harmonics. Secondly, the model incorporates the properties of electrical paths (connection edges) for evaluation. After calculation, the model outputs a set of contribution probabilities: the air compressor contributes 85%, the fan inverter contributes 12%, and the lighting circuit contributes 3%. This result clearly indicates that the 5th harmonic pollution characterized by Φ_A currently appearing in the power grid mainly originates from air compressors that are currently in a high-load startup state.

[0042] According to one embodiment of this application, the equipment-level timing operation status parameters include at least one of the following: speed and load rate commands of the electrical drive equipment, start / stop and process signals of the production equipment, and vibration and temperature monitoring data of the equipment.

[0043] Specifically, the device-level time-series operating status parameters refer to a series of data collected directly or indirectly from each electrical device within the park, which change continuously over time and characterize the real-time operating status and load of the device. These parameters include, but are not limited to, the following categories: The first category consists of speed and load rate commands for electrically driven equipment. This refers to the control signals issued by the frequency converter or controller of rotating equipment such as motors, fans, and pumps, including setpoints for the target motor speed and percentage commands for output torque. These commands directly determine the output level and operating mode of the electrically driven equipment and are key factors affecting its current waveform and harmonic emission characteristics.

[0044] The second category consists of start / stop signals and process signals from production equipment. This refers to discrete status signals generated during the operation of production equipment such as injection molding machines, machine tools, and production lines. Examples include switch signals indicating the start or stop of the equipment, and process code signals identifying the specific processing step or stage the equipment is currently in. These signals reflect the macroscopic operating stage and load switching status of the equipment.

[0045] The third category is vibration and temperature monitoring data of the equipment. This refers to physical quantities measured by sensors on the equipment itself, such as the vibration amplitude and frequency spectrum of motors or bearings, and the temperature values ​​of windings, radiators, or key components. These data can indirectly reflect the mechanical health status, load, and thermal effects of the equipment. These states are related to the electrical parameters and efficiency of the equipment, and may thus affect its harmonic emission characteristics.

[0046] The above parameters are all state information generated during equipment operation that is related to grid friendliness. Their temporal changes together depict the complete working profile of the equipment under different operating conditions.

[0047] According to one embodiment of this application, the phase space reconstruction of harmonic data in grid-level power quality time-series parameters specifically includes: Based on the time series of the harmonic data, a high-dimensional phase space trajectory is reconstructed using the time delay embedding method. In the high-dimensional phase space trajectory, cluster analysis is used to separate trajectory clusters that characterize different dynamic properties, and each trajectory cluster corresponds to a harmonic pollution mode component.

[0048] Specifically, the steps described are methods for processing and analyzing the harmonic components (usually a sequence of amplitudes, phases, etc. of each harmonic that change over time) in the power quality time-series parameters collected from power grid monitoring points.

[0049] First, based on the one-dimensional time series of these harmonic data, phase space reconstruction is performed using the time-delay embedding method. Specifically, from a single-variable harmonic time series, multiple historical data points are selected with a fixed time delay to form a multi-dimensional data point. This multi-dimensional point represents a complete state of the system at a certain moment. After transforming the entire time series according to this rule, a series of such multi-dimensional data points are obtained. The path formed by these points arranged in chronological order in multi-dimensional space is the reconstructed high-dimensional phase space trajectory. This reconstruction process aims to more completely reveal the multi-dimensional information implicit in the time series data that reflects the dynamic characteristics of the harmonic source.

[0050] Then, in the reconstructed high-dimensional phase space trajectory, a clustering analysis algorithm is used to analyze all state points (i.e., the aforementioned multidimensional data points). This algorithm divides the state points into several different groups, i.e., trajectory clusters, based on their position and distribution characteristics (e.g., distance, density) in the high-dimensional space. The state points within each trajectory cluster have similar dynamic characteristics, meaning they are likely to originate from the same or similar harmonic pollution sources with stable dynamic behavior. Therefore, each identified trajectory cluster corresponds to a physically independent harmonic pollution mode component, thus achieving the goal of separating pollution components from different sources from the mixed observation signal.

[0051] According to one embodiment of this application, the source tracing model is a spatiotemporal graph attention network, which is used to learn the dynamic propagation and superposition relationship of the influence of node features on the harmonic state of monitoring points under temporal changes.

[0052] Specifically, the source tracing model employs a machine learning model called a spatiotemporal graph attention network. This model is designed to simultaneously process two types of key information: one is the spatial topology of the power grid composed of nodes and edges, and the other is the feature data of each node that changes over time.

[0053] In the spatial dimension, the model uses a graph attention mechanism to dynamically calculate the mutual influence weights between any two nodes based on the connectivity relationships between nodes in the power grid topology. This enables it to characterize the spatial propagation path of harmonic pollution in electrical networks.

[0054] In the time dimension, the model can receive and process the feature sequence arranged in chronological order at each node (electrical equipment), that is, the change of the dynamic state encoding vector over time. This enables it to capture the trend of the evolution of the equipment's operating state over time.

[0055] By coupling and jointly training the aforementioned spatiotemporal information, the network can learn and simulate the following complex relationship: how the characteristics of each electrical equipment node (whose state is constantly changing) propagate, interact, and ultimately superimpose through fixed electrical connection paths at continuous time points, jointly influencing the dynamic change process of the harmonic state (such as the amplitude and phase of specific mode components) at the monitoring point. In short, this model aims to understand and quantify the spatiotemporal control effect of each dynamic source point in the network on the observed target point.

[0056] According to one embodiment of this application, it also includes: Based on the contribution probability, the additional losses and carbon emissions caused by the harmonic pollution mode components are proportionally allocated to the corresponding electrical equipment to quantify their responsibility for hidden waste.

[0057] Specifically, first, it is necessary to calculate the additional losses caused by the harmonic pollution mode components themselves. This is usually done based on the electrical characteristics of the mode component (such as frequency and amplitude) and a preset harmonic loss calculation model. This model can estimate the additional heat loss generated by the specific harmonic current when it flows through the network impedance and electrical equipment, based on circuit principles, and convert this loss into equivalent active power consumption.

[0058] Secondly, based on the contribution probability of each electrical device to the harmonic pollution mode component calculated in the preceding steps, the total additional loss calculated above is allocated proportionally. For example, if the contribution probability of device A is X%, then the additional loss allocated to device A is X% of the total additional loss. This allocation reflects the principle of responsibility attribution.

[0059] Next, based on the additional losses allocated to each device, the resulting additional carbon emissions can be calculated using a standard energy-carbon emission conversion factor. These carbon emissions are also attributed to the corresponding electrical equipment with the same contribution probability.

[0060] Ultimately, through this series of calculations, the previously difficult-to-detect "hidden" energy waste and carbon emissions caused by power quality pollution are quantitatively allocated and recorded to specific responsible equipment, thereby quantifying the responsibility for this hidden waste. This provides a precise data foundation for refined energy efficiency management, carbon footprint accounting, and differentiated energy-saving incentives.

[0061] According to one embodiment of this application, the quantification of the additional loss is calculated based on the amplitude and frequency of the harmonic pollution mode component and a preset harmonic loss model.

[0062] Specifically, the quantification process of the additional losses is accomplished based on the specific electrical characteristic parameters of the separated harmonic pollution mode components and a pre-established physical or mathematical model.

[0063] Specifically, the calculation relies on two core inputs: first, the amplitude of the harmonic pollution mode component, usually expressed as the effective value of the current or the content rate, which directly determines the strength of the harmonic current; second, its frequency, which identifies which harmonic the component is. When harmonic currents of different frequencies flow through the power grid and equipment, the additional losses caused by their different skin effect and core loss characteristics are also different.

[0064] These parameters are input into a pre-defined harmonic loss model for calculation. This model is typically built upon circuit theory, electromagnetic field principles, or empirical formulas, and internally defines the mathematical relationship between harmonic current and additional losses. The model comprehensively considers the amplitude and frequency of the harmonic current, as well as other parameters that may be included, such as the equivalent resistance of the power grid at the corresponding frequency and the loss coefficient of related electrical equipment at that frequency. By calculating, it determines the additional active power loss (i.e., additional loss) generated by the specific harmonic pollution mode component circulating through the system once. This calculation result serves as the quantitative benchmark for subsequent responsibility allocation.

[0065] A second aspect of this application provides a graph attention source tracing system for harmonic pollution in a power grid, comprising: The data synchronization acquisition module is used to synchronously acquire the equipment-level time-series operating status parameters of each electrical device in the park, as well as the grid-level power quality time-series parameters of the park's common connection points; The equipment status coding module is used to generate a dynamic status coding vector for each electrical device, based on its equipment-level time-series operating status parameters, reflecting the harmonic emission characteristics under its current operating conditions. The harmonic mode analysis module is used to reconstruct the phase space of the harmonic data in the power grid-level power quality time series parameters and separate the harmonic pollution mode components from different sources. The topology model building module is used to construct a dynamic topology diagram of the park's power grid, with electrical equipment as nodes and electrical connections as edges. The pollution source tracing calculation module is used to input the dynamic state encoding vector as the feature of the corresponding node into the source tracing model based on the graph attention mechanism, and to use the harmonic pollution mode component as the target. The source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode component, so as to complete the pollution source tracing.

[0066] According to one embodiment of this application, it also includes: The hidden waste quantification module is used to proportionally allocate the additional losses and carbon emissions caused by the harmonic pollution mode components to the corresponding electrical equipment based on the contribution probability, so as to quantify their responsibility for hidden waste.

[0067] A second aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any of the embodiments of the first aspect above.

[0068] Figure 2 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 2As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call logical instructions in the memory 830 to execute the method in any of the embodiments of the first aspect described above, the method including: Synchronously collect equipment-level time-series operating status parameters of each electrical device in the park, as well as grid-level power quality time-series parameters of the park's common connection points; For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters. Phase space reconstruction is performed on the harmonic data in the power grid-level power quality time series parameters to separate harmonic pollution mode components from different sources; A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism. Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

[0069] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0070] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer being able to perform the methods provided by the above methods, the method comprising: Synchronously collect equipment-level time-series operating status parameters of each electrical device in the park, as well as grid-level power quality time-series parameters of the park's common connection points; For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters. Phase space reconstruction is performed on the harmonic data in the power grid-level power quality time series parameters to separate harmonic pollution mode components from different sources; A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism. Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

[0071] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided by the above methods, the method comprising: Synchronously collect equipment-level time-series operating status parameters of each electrical device in the park, as well as grid-level power quality time-series parameters of the park's common connection points; For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters. Phase space reconstruction is performed on the harmonic data in the power grid-level power quality time series parameters to separate harmonic pollution mode components from different sources; A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism. Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

[0072] Example 2 The technical solution of this application will be described in detail below with reference to a specific example of an automobile manufacturing park. The power distribution network of this park includes multiple workshops, and the key loads include the variable frequency drive fans in the painting workshop, the robot servo system in the welding workshop, the production line in the final assembly workshop, and the air conditioning units of public facilities.

[0073] Step S1: Synchronous Acquisition A power quality monitoring device is deployed at the low-voltage side busbar (point of common coupling) of the 10kV / 400V distribution transformer in the park. The device simultaneously collects the three-phase voltage and current waveforms at a sampling rate of 12,800 points per second and calculates the amplitude and phase timing sequence of each harmonic (up to the 50th) in real time.

[0074] At the same time, status parameters are collected from various key devices through IoT gateways: Paint shop fan frequency converter (equipment D1): Collects its real-time output frequency (Unit: Hz) and percentage of torque current (unit:%).

[0075] The welding workshop robot servo driver (device D2) collects its DC bus voltage. (Unit: V) Instantaneous speed fed back by the joint motor encoder (Unit: rpm).

[0076] Main air compressor of the air compressor station (equipment D3): Collect its loading / unloading status signals. (Boolean value, 1 for loading) and effective value of host vibration acceleration (Unit: m / s²).

[0077] All data is synchronized at the microsecond level via the NTP protocol, forming a time-aligned time-series dataset. .

[0078] Step S2: Generate dynamic state coding vector A lightweight encoder based on physical mechanisms is constructed for each device. Taking the frequency converter (D1) as an example, its harmonic emission characteristics are strongly correlated with the modulation process and load conditions. Its state coding vector is defined. The generation algorithm is as follows: First, calculate the estimated value of the fundamental current: ,in For coefficients, This is the rated frequency.

[0079] Then, according to the inverter harmonic generation model, its first Amplitude of secondary characteristic harmonics (such as the 5th and 7th harmonics) There is an approximate relationship between the fundamental current and the modulation ratio (related to output voltage / frequency). A characteristic mapping can be constructed: .

[0080] Finally, the encoding is performed through a trainable fully connected layer: .in, and The encoder parameters are obtained through training on historical data, such that... In vector space, different harmonic emission modes (such as higher harmonic dominance and lower harmonic dominance) can be clustered and distinguished.

[0081] A similar approach is used for the servo drive (D2) and the air compressor (D3), but their encoder inputs and mapping functions are designed according to their respective working principles (e.g., the PWM frequency of D2 is related to the DC voltage, and the loading shock of D3 is related to the vibration).

[0082] Step S3: Harmonic mode separation (phase space reconstruction and decomposition) The time series of the amplitude of the 5th harmonic current measured at the PCC point Perform phase space reconstruction.

[0083] Time Delay Embedding: Determining the Optimal Delay Steps Using Mutual Information Selecting the embedding dimension Reconstructing a one-dimensional sequence into trajectory points in a high-dimensional phase space: .

[0084] Mode decomposition: Nonnegative matrix factorization (NMF) is used to decompose the trajectory point matrix. Let all trajectory points form a matrix. NMF decomposes it into: .in, Each column represents the fundamental mode of a "harmonic contamination mode component" in phase space (i.e., the centroid direction of the "trace cluster" in the claims). The rows represent the time activation coefficients of each modal component. Through decomposition, for example, three modal components can be separated: , , Each component is a time-series signal that represents the dynamics of an independent pollution source.

[0085] Steps S4 and S5: Construct a dynamic topology graph and perform graph attention source tracing. Constructing the topology diagram: Based on the primary wiring diagram, construct a tree-like topology diagram with PCC as the root node and each device as a child node. The edge weights are initialized to the approximate impedance per unit value estimated based on the short-circuit capacity. The node feature is the state encoding vector of the corresponding device at time t. .

[0086] Spatiotemporal Graph Attention Network (ST-GAT) model: Spatial attention: At each layer, for nodes (such as PCC) and its neighboring nodes (e.g., D1), calculate the attention coefficient: .in, and For learnable parameters, Indicates splicing, It's an edge feature. Then, for... Attention weights are obtained by performing softmax normalization. .

[0087] Temporal gating: To capture temporal dependencies, the concept of a gated recurrent unit (GRU) is introduced when updating node features. The hidden state of the node in the previous time step and the neighbor information after attention aggregation are input into the update gate to determine the degree of update of the current state.

[0088] Reverse source tracing calculation: using a separated modal component As the target (true value). In model training and inference, the information flow is reversed: [the information is then transferred to the target (true value)]. The characteristics of the PCC node are assigned as "target features". The ST-GAT model calculates the "target features" of the PCC node for each upstream device node through multi-layer attention propagation. The "attribution gradient" may directly affect the score. Finally, the softmax function is used to... Converted to contribution probability: The output is the probability distribution of each device's contribution to this modal component, for example: .

[0089] Step S6: Quantify responsibility for hidden waste Calculate additional losses: for modal components The corresponding effective value of the harmonic current is The frequency is The simplified model for harmonic additional losses recommended by IEEE Std 1459 is used to calculate the total additional losses on the campus transformer and representative cables. : .in, and This is the equivalent resistance at the corresponding frequency.

[0090] Allocation and calculation: based on contribution probability The additional losses borne by equipment D1 are allocated as follows: Furthermore, based on the local power grid carbon emission factor... Calculate the corresponding implicit carbon emission responsibility for (kg CO2 / kWh): .in This is for the duration of the statistical analysis. The results can be integrated into the enterprise's energy efficiency digital profile.

[0091] For any parts not mentioned in this application, existing technologies may be used or referenced.

[0092] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0093] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A graph attention-based source tracing method for harmonic pollution in industrial park power grids, characterized in that, Includes the following steps: Synchronously collect equipment-level time-series operating status parameters of each electrical device in the park, as well as grid-level power quality time-series parameters of the park's common connection points; For each electrical device, a dynamic state coding vector reflecting the harmonic emission characteristics under its current operating conditions is generated based on its device-level time-series operating state parameters. Phase space reconstruction is performed on the harmonic data in the power grid-level power quality time series parameters to separate harmonic pollution mode components from different sources; A dynamic topology graph of the park power grid is constructed with electrical equipment as nodes and electrical connection relationships as edges. The dynamic state encoding vector is used as the feature of the corresponding node and input into the source tracing model based on graph attention mechanism. Using the harmonic pollution mode components as targets, the source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode components, so as to complete the pollution source tracing.

2. The method according to claim 1, characterized in that, The equipment-level timing operation status parameters include at least one of the following: speed and load rate commands of electrical drive equipment, start / stop and process signals of production equipment, and vibration and temperature monitoring data of equipment.

3. The method according to claim 1, characterized in that, The phase space reconstruction of harmonic data in the time series parameters of power quality at the grid level specifically includes: Based on the time series of the harmonic data, a high-dimensional phase space trajectory is reconstructed using the time delay embedding method. In the high-dimensional phase space trajectory, cluster analysis is used to separate trajectory clusters that characterize different dynamic properties, and each trajectory cluster corresponds to a harmonic pollution mode component.

4. The method according to claim 1, characterized in that, The source tracing model is a spatiotemporal graph attention network, which is used to learn the dynamic propagation and superposition relationship of the influence of node features on the harmonic state of the monitoring point under temporal changes.

5. The method according to claim 1, characterized in that, Also includes: Based on the contribution probability, the additional losses and carbon emissions caused by the harmonic pollution mode components are proportionally allocated to the corresponding electrical equipment to quantify their responsibility for hidden waste.

6. The method according to claim 5, characterized in that, The quantification of the additional loss is calculated based on the amplitude and frequency of the harmonic pollution mode component and a preset harmonic loss model.

7. A graph-based source tracing system for harmonic pollution in a power grid within a park, characterized in that, include: The data synchronization acquisition module is used to synchronously acquire the equipment-level time-series operating status parameters of each electrical device in the park, as well as the grid-level power quality time-series parameters of the park's common connection points; The equipment status coding module is used to generate a dynamic status coding vector for each electrical device, based on its equipment-level time-series operating status parameters, reflecting the harmonic emission characteristics under its current operating conditions. The harmonic mode analysis module is used to reconstruct the phase space of the harmonic data in the power grid-level power quality time series parameters and separate the harmonic pollution mode components from different sources. The topology model building module is used to construct a dynamic topology diagram of the park's power grid, with electrical equipment as nodes and electrical connections as edges. The pollution source tracing calculation module is used to input the dynamic state encoding vector as the feature of the corresponding node into the source tracing model based on the graph attention mechanism, and to use the harmonic pollution mode component as the target. The source tracing model performs reverse attention weight calculation along the path of the dynamic topology graph, and outputs the contribution probability of each electrical device to the harmonic pollution mode component, so as to complete the pollution source tracing.

8. The system according to claim 7, characterized in that, Also includes: The hidden waste quantification module is used to proportionally allocate the additional losses and carbon emissions caused by the harmonic pollution mode components to the corresponding electrical equipment based on the contribution probability, so as to quantify their responsibility for hidden waste.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images