A power quality identification method and device, electronic equipment and storage medium

Abstract

The application provides a power quality identification method and device, electronic equipment and a storage medium, wherein the method comprises: acquiring three-phase voltage, three-phase current, power grid frequency and ambient temperature collected at a public connection point connected to a power distribution network of a zero-carbon park and each power consumption side; extracting a first preset time sequence feature based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; inputting the preset time sequence feature into a pre-trained identification model to obtain a power quality identification result at a current time and a power quality identification result at a future time. It can be applied to a complex power grid environment of a zero-carbon park containing a high proportion of renewable energy, improve the intelligent level and comprehensive management ability of the power quality management of the zero-carbon park, and help to ensure the safe, stable and efficient operation of the park energy system.

Description

Technical Field

[0001] This application relates to the field of power technology, and more specifically, to a method, apparatus, electronic device, and storage medium for identifying power quality. Background Technology

[0002] As the world continues to strive towards carbon neutrality, zero-carbon parks, as important vehicles for energy transition, are receiving increasing attention for power quality management. Modern zero-carbon parks typically employ distributed energy systems, integrating renewable energy sources such as photovoltaics and wind power, along with energy storage devices to achieve energy self-sufficiency. In this complex energy structure, power quality issues such as voltage fluctuations, frequency deviations, harmonic pollution, and three-phase imbalances occur frequently, directly impacting the operational stability and energy efficiency of precision equipment within the park.

[0003] Existing solutions typically design monitoring algorithms for specific voltage levels (such as 10kV or 380V), lacking the ability to conduct collaborative analysis across multiple voltage levels. In terms of identification dimensions, existing systems often process data individually, leading to inaccurate identification of power quality problems and failing to meet the power quality control requirements of zero-carbon industrial parks. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, electronic device and storage medium for identifying power quality, so as to solve the technical problem of poor power quality identification effect in existing zero-carbon parks.

[0005] In a first aspect, the present invention provides a method for identifying power quality, the method comprising acquiring three-phase voltage, three-phase current, grid frequency, and ambient temperature collected at the common connection endpoints connecting the power distribution network of a zero-carbon park to each power consumption side; extracting a first preset time-series feature based on the acquired three-phase voltage, three-phase current, grid frequency, and ambient temperature; and inputting the preset time-series feature into a pre-trained identification model to obtain the power quality identification result at the current moment and the power quality identification result at the future moment output by the identification model.

[0006] In an optional implementation, the fault category label is determined based on the power quality identification results at the current time and the power quality identification results at future times; Based on the fault category label, determine the corresponding compensation plan; Based on the determined compensation scheme, the corresponding compensation equipment is controlled to compensate for the power quality output at the target common connection endpoint.

[0007] In an optional implementation, the compensation scheme is used to indicate the compensation value, the compensation device, and the compensation object. The second preset time sequence features are extracted based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; The second preset temporal features are input into the pre-trained compensation inference model to obtain the predicted compensation value output by the supplementary inference model. The corresponding compensation scheme is executed based on the predicted compensation value.

[0008] In an optional implementation, the method further includes verifying the predicted compensation value based on a physical model to determine whether the predicted compensation value passes the test. If successful, proceed with the step of implementing the corresponding compensation scheme based on the predicted compensation value.

[0009] In an optional implementation, the second preset feature includes the three-phase voltage, grid frequency, voltage deviation, reference impedance, compensation value, and equipment aging factor at the common connection point connecting the distribution network of the zero-carbon park to the target power consumption side. The compensated inference model consists of an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer normalizes the second preset feature of the input; The first hidden layer performs nonlinear fusion of the normalized second preset features through 32 nodes and outputs the fused features; The second hidden layer enhances the fused features with physical constraints through 16 nodes embedded with residual connections and outputs refined features; The output layer outputs predicted compensation values ​​by linearly transforming and scaling the refined features using an activation function.

[0010] In an optional implementation, the loss function of the compensation inference model is... for: ; ; ; ; in, For data fitting loss, Loss due to physical conservation. For normal loss, , , These are the weighting coefficients. The total number of training batches. For the first The predicted compensation value for the batch, For the first Theoretical compensation value for the batch For the first The three-phase voltage of the batch For the first Batch reference impedance, For the first Target voltage value for the batch These are the network weight parameters. For parameter space.

[0011] In an optional implementation, the initial compensation amount is calculated in the following manner: Calculate the product of the current voltage difference, voltage level coefficient, and dynamic correction coefficient as the initial compensation amount.

[0012] Secondly, the present invention provides a power quality identification device, the device comprising: The data acquisition module is used to acquire the three-phase voltage, three-phase current, grid frequency, and ambient temperature at the common connection endpoints of the power distribution network and each power consumption side in the zero-carbon park. The preprocessing module is used to extract the first preset timing features based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; The identification module is used to input the preset time-series features into a pre-trained identification model to obtain the power quality identification results at the current time and the power quality identification results at future time output by the identification model.

[0013] Thirdly, the present invention provides an electronic device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the power quality identification method as described in any of the foregoing embodiments.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the power quality identification method as described in any of the foregoing embodiments.

[0015] This application provides a method, apparatus, electronic device, and storage medium for identifying power quality. The method includes acquiring three-phase voltage, three-phase current, grid frequency, and ambient temperature data collected at the common connection points connecting the power distribution network of a zero-carbon park to each power consumption side; extracting a first preset time-series feature based on the acquired three-phase voltage, three-phase current, grid frequency, and ambient temperature; and inputting the preset time-series feature into a pre-trained identification model to obtain the power quality identification result at the current moment and the power quality identification result at future moments output by the identification model. By comprehensively monitoring the power quality of multiple voltage levels and multiple nodes within the park, and combining multi-dimensional time-series features with an intelligent model to improve identification accuracy, this method is applicable to the complex grid environment of zero-carbon parks containing a high proportion of renewable energy. It enhances the intelligent level and comprehensive governance capabilities of power quality management in zero-carbon parks, contributing to ensuring the safe, stable, and efficient operation of the park's energy system. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a power quality identification method provided in an embodiment of this application; Figure 2 A schematic diagram of the structure of a power quality identification device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0019] Example 1 Currently, the technology for monitoring and compensating for power quality in zero-carbon industrial parks typically relies on sensors to collect data such as voltage and current, which is then transmitted to a remote server for analysis. Only after a power quality issue arises can the problem be identified and a simple compensation suggestion provided. This technology is not only limited to a single voltage level and fails to fully identify common problems such as voltage instability, frequency deviation, and three-phase imbalance, but it also suffers from a complete time lag and untimely compensation.

[0020] Based on this, this application provides a method, apparatus, electronic device, and storage medium for identifying power quality.

[0021] Figure 1 A flowchart illustrating a power quality identification method provided in an embodiment of this application. Figure 1 As shown, this application provides a method for identifying power quality, comprising: S10. Collect the three-phase voltage, three-phase current, grid frequency, and ambient temperature at the common connection points of the power distribution network and each power consumption side of the zero-carbon park.

[0022] Zero-carbon industrial parks typically have multiple voltage modes, including 220V, 380V, and 400V. 220V is suitable for general electricity use, such as residential users, small office equipment, lighting, and ordinary sockets. 380V / 400V can be used for industrial power equipment, commercial central air conditioning systems, electric vehicle charging stations (fast charging), large water pumps and fans, distributed energy sources (such as the AC side of photovoltaic inverters), and loads in data centers, laboratories, and other similar locations within the park.

[0023] Corresponding data acquisition components, such as Hall effect current sensors, voltage transformers, and temperature sensors, can be installed at the core circuit breakers on the high- and medium-voltage side of the distribution network. Corresponding hardware circuits can be configured to adapt to three voltage levels, enabling the acquisition of data such as three-phase voltage, three-phase current, grid frequency, and ambient temperature. Ambient temperature can be acquired using temperature sensors installed on the cables or inside the core circuit breaker housing.

[0024] The data acquisition element can collect data every second and transmit it to the local edge computing device in real time via RS485 / LoRa protocol. The local edge computing device is deployed in the zero-carbon park, located near the acquisition element (within 30 meters). It can be an edge terminal with an integrated ARM+NPU architecture (computing power ≥ 0.5 TOPS), which can process data quickly and integrate four major functions: "interference filtering, fault inference tag generation, compensation scheme formulation, and device linkage". It can directly access locally stored data and also store real-time processing results.

[0025] S11. Extract the first preset timing features based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature.

[0026] Before step S11, the data is preprocessed. An adaptive threshold filtering method can be used to filter out interference signals, and then the data can be smoothed by the moving average method. Finally, the min-max standardization method is used to map all the data to the [0,1] interval to form a fusion dataset with a unified format.

[0027] To address the high-frequency interference and random fluctuations caused by new energy power generation (photovoltaic and wind power), interference filtering should be performed using auxiliary parameters. A unified data format is needed to ensure that different types of features can be integrated for analysis.

[0028] An adaptive threshold filtering method can be used to construct a dynamic threshold range (such as the normal voltage fluctuation range ±3%) based on historical normal data. Data that exceeds the threshold and has no abnormal related parameters (such as only voltage fluctuation but stable current and temperature) is identified as interference signals brought by new energy output and filtered out. The filtered data is further smoothed by the moving average method to retain the true abnormal characteristics.

[0029] The feature data with different dimensions and ranges are uniformly normalized. The min-max normalization method is used to map features such as voltage deviation (unit V), frequency deviation (unit Hz), temperature (unit ℃), and correlation coefficient (dimensionless) to the [0,1] interval. The classification features (such as the number of phases and the frequency level of the problem) are converted into numerical features using the one-hot encoding method. Finally, a fusion feature dataset with unified dimensions that can be directly input into the analysis model is formed.

[0030] The first preset timing features here may include core data such as three-phase voltage, grid frequency, three-phase power balance, and voltage drop, as well as three-phase current and ambient temperature as auxiliary data.

[0031] Each time-series feature can be continuous data within 5 seconds.

[0032] Specifically, for the collected continuous time-series data such as voltage deviation, frequency deviation, three-phase voltage imbalance, voltage sag, current, and temperature, multi-dimensional feature extraction methods are needed to extract key information reflecting the essence of power quality problems. This provides support for subsequent accurate identification and prediction, while filtering out interference signals caused by fluctuations in renewable energy output to ensure the effectiveness of the features. Specific extraction steps may include, but are not limited to: numerical statistical methods, duration statistical methods, slope calculation methods, trend fitting and inflection point detection methods, correlation coefficient analysis methods, and time-series synchronicity analysis methods.

[0033] S12. Input the preset time-series features into the pre-trained recognition model to obtain the power quality recognition results at the current time and the power quality recognition results at future time output by the recognition model.

[0034] The identification model here can be a deep learning model based on time-series features, which outputs the current problem type, severity, and prediction results for the next 30 seconds, and generates fault classification labels.

[0035] The types of problems here can include voltage deviation, frequency deviation, three-phase voltage imbalance, voltage drop, etc. Among them, the acceptable range for voltage deviation is ±0.5% of the rated voltage, which can be divided into minor: 0.5%-1%; moderate: 1%-2%; severe: >2%. When the current exceeds 1.1 times the rated value, the level is upgraded by one level.

[0036] The acceptable range for frequency deviation is ±0.2Hz. The classification standard can be: slight: 0.2-0.5Hz; severe: >0.5Hz. When the temperature exceeds 85℃, the level is upgraded by one level.

[0037] The acceptable range for three-phase voltage imbalance is ≤1%. The classification standards are: slight: 1%-2%; moderate: 2%-3%; severe: >3%. When the phase difference exceeds 3°, the level is upgraded by one level.

[0038] The acceptable range for voltage drop is ≥90% of the rated voltage. The classification standards are: minor: 80%-90%; moderate: 70%-80%; emergency: <70%. If it lasts for more than 0.5 seconds, it is considered a problem. If it lasts for more than 3 seconds, the level is upgraded.

[0039] In a specific implementation, the model can be trained based on historical fault-related data from the project park. The data covers three voltage scenarios, fluctuations in new energy output, and various power quality issues and their severity levels. Each data point is labeled with "key features within 5 seconds - problem type - severity level - problem change after 30 seconds," and is divided into training data, validation data, and test data in a 7:2:1 ratio to ensure the model's accuracy.

[0040] The trained model can do two things at the same time: first, determine what kind of power quality problem exists and how serious it is, and output the probability of each problem (e.g., the probability of voltage deviation is 92%, and the probability of other problems is extremely low); second, predict whether the problem will worsen in the next 30 seconds (e.g., the voltage deviation increases from 1.2% to 1.8%).

[0041] Input the first preset timing feature within 5 seconds. After rapid analysis, the model outputs the probability of the problem and the prediction result for the next 30 seconds. The problem with a probability ≥ 90% is selected as the current core problem. If multiple problems exist simultaneously, the core problem is determined according to the priority order of "voltage drop > frequency deviation > three-phase imbalance > voltage deviation".

[0042] The labels here can be formatted as "problem type - severity level - trend prediction" to ensure that subsequent compensation is based on evidence.

[0043] Specifically, the label format can be divided into full labels and simplified labels. Full labels are easy to trace (e.g., "Voltage deviation - moderate exceedance (1.2%) - rise to 1.8% in the next 30 seconds (continuously worsening)"), while simplified labels are used to control equipment (e.g., "Three-phase imbalance - slight - no worsening trend").

[0044] The basic rules for labeling can be as follows: the problem type is determined by the high probability result output by the model; the severity level is matched with the judgment criteria in the first step; and the trend prediction is determined by the result output by the model in the next 30 seconds (maintaining the status quo, continuing to worsen, or gradually recovering).

[0045] Furthermore, the severity level can be adjusted by combining auxiliary data (e.g., if the voltage deviation is moderate and the current exceeds the rated value by 1.1 times, it can be upgraded to a severe level). Duration requirements can be set for different levels (minor issues require five consecutive sampling points (0.1 seconds) to exceed the limit before being flagged; urgent issues require only one sampling point to exceed the limit, to avoid misjudging instantaneous fluctuations).

[0046] For example, if the input features are a voltage deviation of 1.2% and a current of 1.05 times the rated value for 6 seconds, the model outputs a voltage deviation probability of 92%, predicting it will rise to 1.8% within 30 seconds; since the current does not exceed 1.1 times the rated value, no correction level is needed, and the final label is "Voltage Deviation - Moderate Exceeding Standard (1.2%) - Rising to 1.8% in the Next 30 Seconds (Continuing to Worsen)".

[0047] The power quality identification method provided in this application provides a method for overall monitoring of power quality at multiple voltage levels and nodes within a park, and improves identification accuracy by combining multi-dimensional time-series characteristics and intelligent models. This method is applicable to the complex power grid environment of zero-carbon parks with a high proportion of renewable energy, improves the intelligent level and comprehensive governance capabilities of power quality management in zero-carbon parks, and helps to ensure the safe, stable, and efficient operation of the park's energy system.

[0048] Example 2 In one embodiment of this application, to further ensure the stable operation of the distribution network, a power compensation method is also provided. Based on the generated tags, a targeted compensation plan is formulated (clearly defining what problem to compensate for, when to compensate, and the amount of compensation). Nearby compensation equipment is directly controlled without going through a remote server, enabling rapid response. During and after the compensation process, power quality is continuously monitored to check the effectiveness of the compensation; if the effect is unsatisfactory, the plan is adjusted promptly, and all data is stored locally to optimize subsequent prediction and compensation strategies.

[0049] In one feasible implementation, a fault category label can be determined based on the power quality identification results at the current moment and the power quality identification results at future moments. Based on the fault category label, a corresponding compensation scheme is determined. According to the determined compensation scheme, the corresponding compensation equipment is controlled to compensate for the power quality output at the target common connection endpoint.

[0050] The compensation scheme here indicates the compensation value, compensation equipment, and compensation object. Compensation equipment includes, but is not limited to, voltage stabilization devices (SVG Static Var Generator), three-phase power balance adjustment devices (STATCOM Dynamic Var Compensator), and filtering devices (APF Active Power Filter), etc., which can receive instructions from local edge computing devices via the Modbus-RTU protocol and directly execute the compensation operation.

[0051] Specifically, a second preset time-series feature can be extracted based on the collected three-phase voltage, three-phase current, grid frequency, and ambient temperature. The second preset feature includes the three-phase voltage, grid frequency, voltage deviation, reference impedance, compensation value, and equipment aging factor at the common connection point connecting the distribution network of the zero-carbon park to the target power consumption side.

[0052] The second preset temporal feature is input into the pre-trained compensation inference model to obtain the predicted compensation value output by the supplementary inference model. The corresponding compensation scheme is then executed based on the predicted compensation value.

[0053] The compensation inference model here can be trained and generated based on the PINN network structure.

[0054] The compensation inference model comprises an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer normalizes the input second preset feature (a seven-dimensional vector). The first hidden layer performs nonlinear fusion of the normalized second preset feature using 32 nodes (ReLU) and outputs the fused feature. The second hidden layer enhances the fused feature with physical constraints using 16 nodes (ReLU) embedded with residual connections and outputs refined features. The output layer outputs the predicted compensation value through linear transformation and activation function scaling of the refined feature.

[0055] The input layer undergoes no parameter transformation, only dimensional alignment and normalization. The seven heterogeneous dimensional variables (voltage V, frequency Hz, temperature ℃, etc.) are normalized to the range of [-1,1] to eliminate dimensional differences and signal baseline drift, such as voltage rise during low-load nighttime conditions.

[0056] The weights of the first hidden layer are quantized using INT8 and then fused non-linearly. ; in, These are the normalized input features.

[0057] The second hidden layer introduces residual connections, which are achieved through the loss function. In this layer, the neuron weights automatically learn an approximate analytical solution to the Thevenin equation, ensuring that gradients are directly backpropagated to the first layer, preventing gradient vanishing due to insufficient training data at the edges. Considering the unique "intermittent + volatile" nature of new energy scenarios, the output of this layer is more sensitive to changes than traditional DNNs.

[0058] The second hidden layer can be represented as: = +ReLU + .

[0059] The output of the output layer can be represented as: ; ; The range [-300, +300] is adapted to the compensation device and dequantized after being scaled by Sigmoid.

[0060] In feasible implementations, the loss function of the compensation inference model is... for: ; ; ; ; in, For data fitting loss, Loss due to physical conservation. For normal loss, , , These are the weighting coefficients. The total number of training batches. For the first The predicted compensation value for the batch, For the first Theoretical compensation value for the batch For the first The three-phase voltage of the batch For the first Batch reference impedance, For the first Target voltage value for the batch These are the network weight parameters. For parameter space, including , , , , .

[0061] The weighting coefficient can be determined by looking up a table based on temperature. It can be 1, It can be 50, It can be 0.001.

[0062] Here, the compensation schemes can be prioritized; specifically, three priorities, P1 to P3, can be set.

[0063] P1 is an emergency level, or a severe level, predicted to become emergency within 10 seconds, requiring immediate compensation action (command delay ≤ 500 milliseconds), such as when the voltage suddenly drops to 70% of the rated voltage. P2 is a moderate / severe level, or a minor level, predicted to worsen within 30 seconds, requiring 10 seconds of advance equipment preparation (starting the internal IGBT module of the compensation equipment to preheat, adjusting the DC-side capacitor voltage to a preset value, such as 560V in a 400V scenario), and execution at the designated time, such as when the three-phase imbalance is 2.5% and predicted to rise to 3.2%. P3 is a minor level with no worsening trend; it can be observed for 6 seconds, and execution only occurs if there is no worsening, such as when the voltage deviation is 0.8% and predicted to remain unchanged. If the problem affects critical production equipment in the park, the priority is directly increased by one level to ensure the safety of critical equipment.

[0064] The compensation equipment here can be selected based on the type of problem, prioritizing equipment that is close by and in good condition. Specifically, when there is a voltage deviation, the compensation equipment can be an SVG static var generator. The preferred equipment is one with a load not exceeding 80%, no faults, and normal communication. In the event of a main equipment failure, a backup SVG static var generator within 50 meters should be selected as the second choice.

[0065] When the three voltage phases are unbalanced, the compensation device can be a STATCOM dynamic reactive power compensation device or a phase adjustment module. Devices with adjustable phase and fast response are preferred, followed by the nearest available backup STATCOM dynamic reactive power compensation device.

[0066] When voltage drops / flickers occur, compensation devices can be dynamic reactive power compensation devices or active power filters (APF). Devices that can filter interference and have instantaneous response are preferred. If no backup device is available, a composite compensation mode (SVG+APF working together) should be activated.

[0067] When multiple problems exist simultaneously, a multi-device collaborative approach using SVG+STATCOM+APF can be adopted. Ideally, all devices should meet the status requirements; if some devices fail, secondary solutions should compensate for the core problem by activating corresponding backup devices.

[0068] The compensation value here can be calculated dynamically. The initial calculation scheme can calculate the product of the current voltage difference, the voltage level coefficient, and the dynamic correction coefficient as the initial compensation amount. That is: Compensation value = (Current excess value - Standard acceptable upper limit) × Voltage level coefficient × Dynamic correction coefficient.

[0069] Among them, the voltage level coefficient is 1.0 for 220V, 1.1 for 380V, and 1.2 for 400V; the dynamic correction coefficient is 1.0-1.1.

[0070] For example, when the voltage deviation in the 400V area is moderate (1.2%), the acceptable range is ±0.5%, the compensation amount = (1.2% - 0.5%) × 400V × 1.2 × 1.05 ≈ 3.53V (rounded to 4V) to ensure that the voltage deviation is reduced to the acceptable range after compensation.

[0071] Subsequently, the compensation status can be monitored in real time based on the compensation inference model, and the compensation value can be adjusted in a timely manner. The basic compensation duration is no less than 5 seconds. If it is predicted that the problem will continue to worsen, the duration is extended to cover the entire process of the problem worsening (e.g., if it is predicted that the problem will worsen within 30 seconds, the duration is set to 45 seconds).

[0072] Furthermore, before each compensation value is executed, the predicted compensation value can be verified based on the physical model to determine whether the predicted compensation value passes the verification. If it passes, the step of executing the corresponding compensation scheme based on the predicted compensation value is executed. The physical model here can be the calculation model of the initial compensation value. If the predicted compensation value and the compensation value calculated by the physical model are within a specified range, compensation can be executed; otherwise, a new predicted compensation value is generated iteratively.

[0073] In another embodiment, the compensation can be verified by a local simulation model (built based on the parameters of the park's power supply system) before compensation to confirm that there are no new problems after compensation (such as the current not exceeding the rated value and the harmonic distortion rate meeting the standard). If there are problems, the parameters are fine-tuned, up to 3 times, and the linkage equipment is then used to perform compensation after ensuring that the solution is feasible.

[0074] Data is monitored every 0.5 seconds during the compensation process to assess its effectiveness. If the compensation is ineffective (problem improvement less than 80%), the compensation intensity is adjusted immediately. If the compensation is moderate (improvement 80%-90%), the compensation period is extended. If the compensation is effective (improvement ≥90%), the plan is maintained, and compensation is gradually discontinued once the data returns to satisfactory levels.

[0075] Furthermore, by combining historical compensation results with real-time operating condition adjustments, if historical compensation results for similar problems are unsatisfactory, the compensation intensity can be appropriately increased. Simultaneously, fluctuations in current, temperature, and renewable energy output can be considered to fine-tune the compensation intensity (e.g., slightly increase the compensation intensity if the current is too high).

[0076] The power quality compensation scheme provided in this application improves the accuracy and physical rationality of compensation value prediction by introducing a compensation inference model based on PINN networks and embedding physical laws as constraints into the loss function. The model employs residual connections and quantization design to enhance training stability and inference efficiency in scenarios with limited data at the edge. Through priority partitioning, multi-device collaboration, and dynamic compensation value calculation, it achieves rapid and accurate responses to various problems such as voltage deviation and imbalance. Local decision-making and real-time monitoring and adjustment mechanisms effectively guarantee the compensation effect and significantly improve the power quality and operational stability of the distribution network.

[0077] Example 3 Figure 2 This is a schematic diagram of the structure of a power quality identification device provided in an embodiment of this application. Figure 2 As shown, based on the same inventive concept, this application also provides a power quality identification device 20, which includes: The data acquisition module 210 is used to acquire the three-phase voltage, three-phase current, grid frequency and ambient temperature collected at the common connection endpoints of the power distribution network and each power consumption side in the zero-carbon park. The preprocessing module 220 is used to extract the first preset timing features based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; The identification module 230 is used to input the preset time-series features into a pre-trained identification model to obtain the power quality identification results at the current time and the power quality identification results at future time output by the identification model.

[0078] In a preferred embodiment, a compensation module (not shown in the figure) is further included, which is used to determine the fault category label based on the power quality identification result at the current time and the power quality identification result at future time. Based on the fault category label, determine the corresponding compensation plan; Based on the determined compensation scheme, the corresponding compensation equipment is controlled to compensate for the power quality output at the target common connection endpoint.

[0079] In a preferred embodiment, the compensation scheme is used to indicate the compensation value, the compensation device, and the compensation object. The compensation module is also used to extract a second preset timing feature based on the collected three-phase voltage, three-phase current, grid frequency and ambient temperature; The second preset temporal features are input into the pre-trained compensation inference model to obtain the predicted compensation value output by the supplementary inference model. The corresponding compensation scheme is executed based on the predicted compensation value.

[0080] In a preferred embodiment, the method further includes verifying the predicted compensation value based on a physical model to determine whether the predicted compensation value passes the test. If successful, proceed with the step of implementing the corresponding compensation scheme based on the predicted compensation value.

[0081] In a preferred embodiment, the second preset features include the three-phase voltage, grid frequency, voltage deviation, reference impedance, compensation value, and equipment aging factor at the common connection endpoint connecting the distribution network of the zero-carbon park to the target power consumption side. The compensated inference model consists of an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer normalizes the second preset feature of the input; The first hidden layer performs nonlinear fusion of the normalized second preset features through 32 nodes and outputs the fused features; The second hidden layer enhances the fused features with physical constraints through 16 nodes embedded with residual connections and outputs refined features; The output layer outputs predicted compensation values ​​by linearly transforming and scaling the refined features using an activation function.

[0082] In a preferred embodiment, the loss function of the compensation inference model for: ; ; ; ; in, For data fitting loss, Loss due to physical conservation. For normal loss, , , These are the weighting coefficients. The total number of training batches. For the first The predicted compensation value for the batch, For the first Theoretical compensation value for the batch For the first The three-phase voltage of the batch For the first Batch reference impedance, For the first Target voltage value for the batch These are the network weight parameters. For parameter space.

[0083] In a preferred embodiment, the compensation module calculates the initial compensation amount in the following manner: Calculate the product of the current voltage difference, voltage level coefficient, and dynamic correction coefficient as the initial compensation amount.

[0084] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 As shown, the electronic device 300 includes a processor 310, a memory 320, and a bus 330.

[0085] The memory 320 stores machine-readable instructions that can be executed by the processor 310. When the electronic device 300 is running, the processor 310 and the memory 320 communicate via the bus 330. When the machine-readable instructions are executed by the processor 310, the steps of a power quality identification method as described in the above method embodiment can be performed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0086] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of a power quality identification method as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0087] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0089] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0091] It should be noted that if the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or 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 this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0093] The above description is merely an embodiment of this application and is not intended to limit the scope of protection 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 protection of this application.

Claims

1. A method for identifying power quality, characterized in that, The method includes: The three-phase voltage, three-phase current, grid frequency, and ambient temperature were collected at the common connection points of the power distribution network and each power consumption side in the zero-carbon park. The first preset time-series features are extracted based on the collected three-phase voltage, three-phase current, power grid frequency, and ambient temperature. The preset time-series features are input into a pre-trained recognition model to obtain the power quality recognition results at the current time and the power quality recognition results at future time.

2. The method according to claim 1, characterized in that, Also includes: Based on the power quality identification results at the current moment and the power quality identification results at future moments, the fault category label is determined; Based on the fault category label, determine the corresponding compensation plan; Based on the determined compensation scheme, the corresponding compensation equipment is controlled to compensate for the power quality output at the target common connection endpoint.

3. The method according to claim 2, characterized in that, The compensation plan is used to indicate the compensation value, the compensation equipment, and the compensation object. The second preset time sequence features are extracted based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; The second preset temporal features are input into the pre-trained compensation inference model to obtain the predicted compensation value output by the supplementary inference model. The corresponding compensation scheme is executed based on the predicted compensation value.

4. The method according to claim 3, characterized in that, It also includes verifying the predicted compensation value based on the physical model to determine whether the predicted compensation value passes the test; If successful, proceed with the step of implementing the corresponding compensation scheme based on the predicted compensation value.

5. The method according to claim 4, characterized in that, The second preset features include the three-phase voltage, grid frequency, voltage deviation, reference impedance, compensation value, and equipment aging factor at the common connection point connecting the distribution network of the zero-carbon park to the target power consumption side. The compensated inference model consists of an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer normalizes the second preset feature of the input; The first hidden layer performs nonlinear fusion of the normalized second preset features through 32 nodes and outputs the fused features; The second hidden layer enhances the fused features with physical constraints through 16 nodes embedded with residual connections and outputs refined features; The output layer outputs predicted compensation values ​​by linearly transforming and scaling the refined features using an activation function.

6. The method according to claim 5, characterized in that, Loss function of the compensatory inference model for: ； ； ; ； in, For data fitting loss, Loss due to physical conservation. For normal loss, , , These are the weighting coefficients. The total number of training batches. For the first The predicted compensation value for the batch, For the first Theoretical compensation value for the batch For the first The three-phase voltage of the batch For the first Batch reference impedance, For the first Target voltage value for the batch These are the network weight parameters. For parameter space.

7. The method according to claim 6, characterized in that, The initial compensation amount is calculated using the following method: Calculate the product of the current voltage difference, voltage level coefficient, and dynamic correction coefficient as the initial compensation amount.

8. A power quality identification device, characterized in that, The device includes: The data acquisition module is used to acquire the three-phase voltage, three-phase current, grid frequency, and ambient temperature at the common connection endpoints of the power distribution network and each power consumption side in the zero-carbon park. The preprocessing module is used to extract the first preset timing features based on the collected three-phase voltage, three-phase current, power grid frequency and ambient temperature; The identification module is used to input the preset time-series features into a pre-trained identification model to obtain the power quality identification results at the current time and the power quality identification results at future time output by the identification model.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the power quality identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the power quality identification method as described in any one of claims 1 to 7.

Images