Adaptive Threshold Dynamic Determination Method and System for Multi-Source New Energy Data

Abstract

This invention provides an adaptive threshold dynamic outlier determination method and system for multi-source renewable energy data. It addresses the high false alarm and false negative rates of traditional static threshold detection under complex operating conditions. The method generates a steady-state token through a three-layer steady-state quality judgment; it filters candidate outlier signals in multi-channel consistency verification; it then analyzes their temporal persistence, attenuation characteristics, and spectral consistency to calculate the outlier stability index and assign a three-level rating; finally, it dynamically generates high, medium, and low three-level judgment thresholds by integrating real-time operating conditions and historical statistical features, achieving adaptive and accurate judgment. The system also features a feedback learning mechanism to optimize the pre-judgment logic and model parameters, continuously improving detection robustness.

Description

Technical Field

[0001] This invention relates to the field of new energy data processing and anomaly detection technology, specifically to an adaptive threshold outlier dynamic determination method and system for multi-source new energy data. Background Technology

[0002] With the large-scale grid connection of new energy sources, real-time monitoring and anomaly detection of their operational data are crucial for ensuring grid security and improving equipment operation and maintenance efficiency. Outlier detection is one of the key technologies for discovering potential equipment faults and data acquisition anomalies.

[0003] Currently, outlier identification in new energy data often relies on methods based on fixed thresholds or simple statistical models. However, the operation of new energy equipment is severely affected by external environmental factors such as weather, sunlight, and wind speed, resulting in intermittent and highly volatile power output. This makes fixed threshold methods unsuitable, easily generating a large number of false positives or false negatives. Furthermore, the reliability of identification based on a single data source is insufficient, failing to distinguish between genuine equipment anomalies and localized sensor malfunctions or transient interference.

[0004] Therefore, there is an urgent need for an outlier detection method that can adapt to the non-stationary and multi-disturbance characteristics of new energy data, comprehensively utilize multi-source information, and dynamically adjust the judgment strategy to improve the accuracy, reliability, and adaptability of the judgment. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of existing fixed threshold outlier detection methods in new energy application scenarios, such as poor adaptability, high false alarm rate, and inability to distinguish the source of anomalies. This invention provides an adaptive threshold dynamic determination method and system for multi-source new energy data, which can realize accurate, dynamic, and traceable identification and determination of outliers in the operation data of new energy equipment, thereby improving the reliability of anomaly detection.

[0006] The specific technical solution of this application is as follows:

[0007] According to one aspect of this application, an adaptive threshold outlier dynamic determination method for multi-source new energy data is provided, comprising:

[0008] For the current operating status of new energy equipment, it is determined in turn whether it is in a power stability period, whether there is a severe disturbance in the external environment, and whether the integrity and signal-to-noise ratio of the data acquisition channel meet the preset quality standards. Only when all three judgments pass, a steady-state measurable token is generated. If any judgment fails, a delayed re-examination mechanism is initiated, and the three judgments are re-executed after waiting for a preset sampling period.

[0009] After obtaining the steady-state measurable token, perform multi-objective consistency verification: select the target data channel and multiple associated auxiliary channels, construct the difference sequence between each auxiliary channel and the target channel, and verify each difference sequence; only when the verification confirms the existence of non-coordination deviation, output candidate outlier signals containing timestamps, channel identifiers, and deviation strength.

[0010] Based on the candidate outlier signals, we analyze their time duration, amplitude variation and spectral characteristics, calculate the OSI, and classify the candidate outlier signals into different stability levels based on the OSI.

[0011] Based on whether a steady-state measurable token is available, the stability level of the candidate outlier signal, and combined with the historical outlier distribution characteristics within the sliding time window, a judgment threshold corresponding to each stability level is dynamically generated.

[0012] The outlier determination results obtained based on the determination threshold are encapsulated into determination record units. The outlier patterns and feature information contained in the determination record units are then fed back to the three-layer judgment logic and multi-objective consistency verification rules to dynamically calibrate the generation conditions of steady-state measurable tokens and the selection strategy of auxiliary channels, and to optimize the model for dynamically generating thresholds.

[0013] As a further option of the method of the present invention, the step of sequentially performing three-level judgments on the current operating status of the new energy equipment includes:

[0014] First-level judgment: Collect data from the current moment and backtrack forward. Calculate the standard deviation of the power data sequence at each sampling point. with the mean The fluctuation coefficient is obtained. If the volatility coefficient Less than or equal to the preset power stability threshold If so, the power is determined to be stable;

[0015] Second-level judgment: Synchronously read the most recent... For each sampling point, calculate the maximum rate of change of the wind speed sequence and light intensity sequence, based on the formula... Calculate the comprehensive disturbance index ,in , These are the rated reference values ​​for ambient wind speed and light intensity. , The weighting coefficients for wind speed and light intensity variations in the environmental disturbance index. The maximum rate of change of wind speed. The maximum rate of change of illumination; if Less than or equal to the preset environmental disturbance threshold If so, it is determined that there is no drastic disturbance in the environment;

[0016] Third-level judgment: Check the target channel and preset key related channels in the most recent Packet reception rate per cycle Compared with the current signal-to-noise ratio If all inspected channels meet the requirements and If so, the channel quality is deemed acceptable;

[0017] A steady-state measurable token is generated only when all three levels of judgment pass.

[0018] As a further option of the method of the present invention, the startup delay re-examination mechanism includes:

[0019] If any level fails the check, check the current retry count. Is it less than the preset maximum number of retries? ;

[0020] like Then Increment by 1, and the system waits for the preset delay period. After one sampling interval, jump back to the first-level judgment and re-execute;

[0021] like If the current time period is marked as an unreliable interval, the current determination process will be terminated.

[0022] As a further option of the method of the present invention, the multi-target consistency verification includes:

[0023] Target channel determined based on steady-state measurable tokens. Based on the physical topology and electrical connections, select from predefined rules. These are associated auxiliary channels, forming an auxiliary channel set. ;

[0024] Set length as Time alignment window for each auxiliary channel Simultaneously acquire its connection with the target channel The observed sequence within the alignment window, and the difference sequence is calculated. ;

[0025] Calculate each difference sequence Standard deviation and target channel sequence With auxiliary channel sequence correlation coefficient With normalized root mean square error The calculation formula is: ;in, Represents the target channel observation sequence With auxiliary channel observation sequence Covariance. Represents the target channel observation sequence Standard deviation Indicates the auxiliary channel observation sequence Standard deviation Difference sequence In the window The average value within, , For target channel and auxiliary channels In the window The average of the observed values ​​within the range.

[0026] As a further option of the method of the present invention, the verification confirming the existence of non-coordination deviation includes:

[0027] For each auxiliary channel If both conditions are met and If the status of the auxiliary channel matches the target channel, it is determined that the channel is consistent with the target channel; otherwise, it is determined that they are inconsistent. The number of inconsistent auxiliary channels is counted. ;

[0028] Calculate the current observation value of the target channel Its short historical window average deviation ,in The standard deviation of the historical window;

[0029] If both conditions are met and If a global coordination deviation occurs, a candidate outlier signal is output.

[0030] As a further option of the method of the present invention, the OSI calculation formula is as follows: ;in, OSI value , , Preset weights; As a persistent factor, As the attenuation factor, This is the spectrum consistency factor.

[0031] As a further option of the method of the present invention, the step of classifying candidate outlier signals into different stability levels based on OSI includes:

[0032] Preset high stability level threshold With medium stability level threshold ,and ;

[0033] like Then, candidate outlier signals are classified into high stability levels;

[0034] like Then, the candidate outlier signals are classified into medium stability levels;

[0035] like Then, candidate outlier signals are classified as low stability levels.

[0036] As a further option of the method of the present invention, the dynamic generation of the judgment thresholds corresponding to each stability level includes:

[0037] Construct an input feature vector, which includes at least: the Boolean value of the steady-state measurable token, the one-hot encoding of the stability level, real-time operating context information, and historical outlier statistical features within the sliding time window;

[0038] The input feature vector is input into the threshold generation function. The function dynamically corrects the base threshold based on a predetermined adjustment rule and outputs three-level judgment thresholds corresponding to high, medium and low stability levels, respectively.

[0039] As a further option of the method of the present invention, the predetermined adjustment rule includes at least one of the following:

[0040] If a steady-state measurable token is not obtained, the basic thresholds at each level will be amplified by the first preset coefficient.

[0041] Based on the stability level, the basic threshold for each level is multiplied by a different level adjustment coefficient;

[0042] If historical outlier statistics indicate that recent outlier events are frequent and concentrated, then the basic thresholds at all levels will be amplified by the second preset coefficient.

[0043] If the real-time operating context indicates a low-load operating condition at night, then multiply the base thresholds at each level by a tightening factor of less than 1.

[0044] Another aspect of this application provides an adaptive threshold outlier dynamic determination system for multi-source new energy data. The system includes a processor and a memory. The processor is configured to execute instructions stored in the memory to achieve:

[0045] The steady-state quality judgment module is used to determine the current operating status of new energy equipment in sequence: whether it is in a power stable period, whether there is a severe disturbance in the external environment, and whether the integrity and signal-to-noise ratio of the data acquisition channel meet the preset quality standards. Only when all three judgments pass, a steady-state measurable token is generated. If any judgment fails, a delayed re-examination mechanism is initiated, and the three judgments are re-executed after waiting for a preset sampling period.

[0046] The multi-channel collaborative verification module is used to perform multi-target consistency verification after obtaining a steady-state measurable token: select the target data channel and multiple associated auxiliary channels, construct the difference sequence between each auxiliary channel and the target channel, and verify each difference sequence; only when the verification confirms the existence of non-coordination deviation, output a candidate outlier signal containing timestamp, channel identifier and deviation intensity.

[0047] The stability quantification and rating module is used to analyze the time persistence, amplitude variation and spectral characteristics of candidate outlier signals, calculate the outlier stability index, and classify candidate outlier signals into different stability levels based on the outlier stability index.

[0048] The dynamic threshold generation module is used to dynamically generate judgment thresholds corresponding to each stability level based on whether there is a steady-state measurable token, the stability level of the candidate outlier signal, and the historical outlier distribution characteristics within the sliding time window.

[0049] The feedback optimization and recording module is used to encapsulate the outlier judgment results obtained based on the judgment threshold into a judgment recording unit, and feed back the outlier patterns and feature information contained in the judgment recording unit to the steady-state quality judgment module and the multi-channel collaborative verification module to dynamically calibrate the generation conditions of the steady-state measurable token and the selection strategy of the auxiliary channel, and optimize the model of the dynamic threshold generation module.

[0050] The beneficial effects of this application are as follows:

[0051] First, this method sets multiple hard conditions, such as power fluctuation coefficient less than 2% of rated power, environmental disturbance index below the threshold, channel signal-to-noise ratio greater than 30dB and integrity rate higher than 98%, and introduces multi-channel consistency verification to effectively distinguish between real anomalies and local interference, thereby reducing false alarms and missed alarms from the source.

[0052] Second, the outlier stability index is divided into three levels: high, medium, and low. At the same time, it combines historical outlier statistics with real-time operating conditions, so that the judgment boundary can be flexibly adjusted according to the operating environment to adapt to the strong fluctuation characteristics of new energy data.

[0053] Third, the system automatically calibrates the pre-judgment threshold and outlier stability index weights based on the distribution of true positive and false negative samples, and dynamically updates the grading thresholds to maintain the proportion of samples at each level, so that the detection performance can be continuously improved as data accumulates. Attached Figure Description

[0054] Figure 1 A schematic diagram of the overall process of the adaptive threshold outlier dynamic determination method for multi-source new energy data;

[0055] Figure 2Detailed flowchart of S100 steps for adaptive threshold outlier dynamic determination method for multi-source new energy data;

[0056] Figure 3 Detailed flowchart of the S200 method for dynamic determination of outliers using adaptive thresholds for multi-source new energy data;

[0057] Figure 4 Detailed flowchart of the S300 method for dynamic determination of outliers using adaptive thresholds for multi-source new energy data;

[0058] Figure 5 Detailed flowchart of the S400 method for dynamic determination of outliers using adaptive thresholds for multi-source renewable energy data;

[0059] Figure 6 A detailed flowchart of the S500 method for dynamically determining outliers using adaptive thresholds for multi-source renewable energy data. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0061] Operational data from new energy power generation systems are characterized by multiple sources, high dimensionality, and time-varying nature. Outliers in the data may represent initial equipment failures, communication anomalies, or sudden environmental changes. Traditional static threshold detection methods are ill-suited to complex operating conditions, easily leading to false alarms and missed alarms. This invention proposes an outlier detection method that integrates steady-state quality assessment, multi-channel collaborative verification, and dynamic adaptive thresholding, aiming to improve detection accuracy and system robustness.

[0062] The theoretical foundation of this invention is based on the theory of multi-source data synergy, time series stability analysis, and adaptive learning theory of dynamic systems. By constructing a steady-state quality token mechanism, using multi-channel differential consistency verification to filter anomalies, combining the multi-dimensional characteristics of outlier signals for stability quantification, and dynamically generating a judgment threshold based on real-time context, an adaptive threshold dynamic judgment method for outlier points in multi-source new energy data is formed.

[0063] The derivation of the core theoretical formula is as follows:

[0064] Defined at discrete time points The collected number The observed values ​​of each data channel are ,in , The total number of channels. The target channel is denoted as... .

[0065] Steady-state measurability determination model:

[0066] steady-state tokens It is a Boolean variable, determined by three levels of conditional functions: ;in, The power state vector, This represents the environmental disturbance vector. This is the channel quality vector.

[0067] Power stability judgment Calculate the nearest Standard deviation of point power series , with threshold In comparison, among them, This represents the number of historical data points used to calculate power stability.

[0068] Environmental disturbance judgment Based on wind speed With irradiation Calculate the volatility index , with threshold In comparison, among them, , These are the rated reference values ​​for ambient wind speed and light intensity. , This refers to the weighting coefficients of wind speed and light intensity variation components in the environmental disturbance index.

[0069] Channel quality assessment Based on signal-to-noise ratio With completeness rate ,Require and .

[0070] Multi-channel consistency verification model: Selecting an auxiliary channel set Construct the target channel With auxiliary channels difference sequence Calculate the consistency measure, which is the correlation coefficient. With normalized root mean square error : ;in, Represents the target channel observation sequence With auxiliary channel observation sequence Covariance. Represents the target channel observation sequence Standard deviation Indicates the auxiliary channel observation sequence Standard deviation Difference sequence In the window The average value within, , For target channel and auxiliary channels In the window The average of the observed values ​​within the range.

[0071] Define consistent state If and only if and If the proportion of inconsistent channels exceeds the threshold. And the degree to which the instantaneous value of the target channel deviates from the historical average Then candidate outlier signals are generated. .

[0072] Outlier stability index model:

[0073] For candidate signals, analyze their subsequent... point sequence ,in , Reference sequence The average value,

[0074] Calculate the outlier stability index : .

[0075] Persistence factor ,in To meet The number of points, This is the threshold used to determine whether the offset is significant.

[0076] Attenuation factor , The sensitivity coefficient is denoted as .

[0077] Spectral Consistency Factor , This is the coefficient of variation of the percentage of main frequency energy in the sub-window.

[0078] based on Three-level rating :high( ),middle( ),Low( ).

[0079] Dynamic adaptive threshold generation model:

[0080] Final determination threshold For dynamic functions: ;in For real-time operating context, This represents historical outlier statistical features. The model outputs three levels of thresholds: high, medium, and low. ,satisfy .

[0081] The above theoretical framework provides the mathematical and logical foundation for this invention.

[0082] The specific embodiments of the present invention will be described in detail below.

[0083] Example 1

[0084] Please see Figure 1 This illustrates an embodiment of the present invention providing an adaptive threshold outlier dynamic determination method for multi-source new energy data, the method comprising:

[0085] S100: Performs three-level judgments on the current operating status of new energy equipment: whether the equipment is in a power stable period, whether there is severe disturbance in the external environment, and whether the integrity of the data acquisition channel and the signal-to-noise ratio meet the preset quality standards, and generates a steady-state measurable token or marks an untrusted interval.

[0086] S200: After obtaining the steady-state measurable token, select the target data channel and its multiple associated auxiliary channels, construct and compare the difference sequence between the channels, perform multi-target consistency verification based on the difference sequence, and output candidate outlier signals or trigger re-evaluation.

[0087] S300: After receiving candidate outlier signals, analyze the time persistence, amplitude variation and spectral characteristics of the candidate outlier signals, calculate the outlier stability index, and perform high, medium and low stability ratings based on the index.

[0088] S400: By integrating steady-state measurable tokens, outlier stability indices and their stability ratings, and real-time operating context information, a dynamic adaptive threshold generation model is constructed to dynamically generate high, medium and low three-level judgment thresholds.

[0089] S500: Receives multi-level judgment thresholds and outlier judgment results, encapsulates them into traceable judgment record units, feeds back the information in the units to the pre-judgment logic and verification rules, dynamically calibrates relevant conditions and strategies, and performs incremental learning on the stability rating model.

[0090] The specific plan is as follows:

[0091] In an adaptive threshold outlier dynamic determination method for multi-source new energy data, S100 ensures that the input data has the basic quality for reliable outlier analysis through a three-level progressive judgment.

[0092] Please refer to Figure 2The diagram illustrates a flowchart of an exemplary adaptive threshold outlier dynamic determination method S100 for multi-source new energy data, the contents of which include:

[0093] S110: In this step, when the system starts or begins processing a new batch of data, it loads the preset judgment thresholds and operating parameters from the configuration library.

[0094] In one possible implementation of this step, the initialization operation includes:

[0095] Read power stability threshold Set to 2% of rated power. Read the environmental disturbance threshold. Read channel quality requirements: minimum signal-to-noise ratio Data packet integrity rate Set the delay re-examination waiting period. Each sampling interval, maximum number of retries .

[0096] S120: In this step, the active power sequence of the equipment within the most recent time window is obtained, and the fluctuation characteristics are calculated to determine the operational stability.

[0097] In one possible implementation of this step, the execution steps of the first-level judgment are as follows:

[0098] Collect the current time Backtracking Power data at each point Calculate the standard deviation of the sequence. with the mean Calculate the volatility coefficient. .Compare With threshold .like If the power is stable, a flag is set. Otherwise, set .

[0099] S130: In this step, environmental monitoring data is acquired, and the external environment is assessed to determine whether any drastic changes may occur that could cause abnormal fluctuations in the data.

[0100] In one possible implementation of this step, the execution steps of the second-level judgment are as follows:

[0101] Synchronous read of the latest Wind speed sequence at each point and light intensity sequence Calculate the maximum rate of change of wind speed. Maximum rate of change of illumination Similar calculation. Calculate the comprehensive disturbance index. ,in This is the rated reference value. (Comparison) With threshold .like If the environment is determined to be undisturbed, then a sign should be set up. Otherwise, set .

[0102] S140: In this step, assess the data acquisition reliability of the target channel and key associated channels, including communication integrity and signal quality.

[0103] In one possible implementation of this step, the execution steps of the third-level judgment are as follows:

[0104] For the target channel And the three preset key related channels, check the most recent The data packet reception logs for each period are used to calculate the reception rate for each channel. Estimate the current signal-to-noise ratio for each channel. If all channels satisfy and If the channel quality is deemed acceptable, a sign will be set. Otherwise, set .

[0105] S150: In this step, the results of the three-layer judgment are combined to determine whether to generate a token, initiate a re-examination, or mark an untrusted interval.

[0106] In one possible implementation of this step, the comprehensive decision-making logic is as follows:

[0107] Calculate steady-state tokens .like Generate a steady-state measurable token, containing a timestamp and validity period, pass it to S200, and clear the retry counter. .like ,examine Is it true? If it is true, then... Add 1, system wait After a certain time, the process jumps back to S120 for reassessment. If the result is not found to be true, the current time period is marked as an unreliable interval, and the current assessment process is terminated.

[0108] In an adaptive threshold outlier dynamic determination method for multi-source new energy data, after ensuring data quality, S200 screens candidate signals that may represent global anomalies by comparing the consistency of multi-channel data.

[0109] Please refer to Figure 3 The diagram illustrates a flowchart of an exemplary adaptive threshold outlier dynamic determination method S200 for multi-source new energy data, the contents of which include:

[0110] S210: In this step, the target channel to be analyzed is determined based on the token information and operating mode, and the auxiliary channel group for comparison is selected according to predefined rules.

[0111] In one possible implementation of this step, the channel selection step is as follows:

[0112] Parse the steady-state token to obtain the device identifier and time. Load the list of monitoring target channels from the configuration library according to the current device operating mode, setting the first channel as the target channel by default. Based on the physical topology and electrical connections, select [the appropriate connection]. Most relevant Each channel serves as a set of auxiliary channels. .

[0113] S220: In this step, read the data of the target channel and each auxiliary channel within the alignment time window, and calculate the corresponding difference sequence.

[0114] In one possible implementation of this step, the steps for constructing the difference sequence are as follows:

[0115] Set alignment window , length is Each point. For each auxiliary channel Synchronous acquisition and exist Observation sequence within and Calculate the difference sequence. .

[0116] S230: In this step, the statistical characteristics of each difference sequence are analyzed to determine whether the synergistic relationship between the target channel and each auxiliary channel is disrupted, and a comprehensive evaluation is conducted.

[0117] In one possible implementation of this step, the multi-target consistency check is performed as follows:

[0118] For each Calculate its standard deviation ,as well as and correlation coefficient Define a consistent state. :like and ,but Its meaning is consistent, otherwise This refers to inconsistency. The number of inconsistent channels is counted. Calculate the current value of the target channel. Its short historical window average deviation ,in The standard deviation is the historical window value.

[0119] S240: In this step, based on the verification results and deviation, decide whether to output a candidate signal or trigger a pre-detection re-examination.

[0120] In one possible implementation of this step, the decision logic is as follows:

[0121] If both conditions are met and If a global coordination deviation occurs, a candidate outlier signal is generated. Output to S300. If only and This indicates a possible localized disturbance, immediately triggering the S100 reassessment process in reverse. If... If no significant outlier is found, the process is considered to wait for the next cycle.

[0122] In an adaptive threshold outlier dynamic determination method for multi-source new energy data, S300 performs in-depth time-frequency domain feature analysis on candidate outlier signals, quantifies their stability, and provides a basis for dynamic threshold generation.

[0123] Please refer to Figure 4 The diagram illustrates a flowchart of an exemplary adaptive threshold outlier dynamic determination method S300 for multi-source new energy data, the contents of which include:

[0124] S310: In this step, starting from the candidate signal timestamp, the observation window is expanded backward to obtain continuous data of the target channel for analysis of the evolution process.

[0125] In one possible implementation of this step, the extraction steps for the time-series tracing window data are as follows:

[0126] Read timestamps and channel signage Set the tracking window length. Point. Get from arrive observation sequence Get Sequence of the previous reference window Calculate its mean .

[0127] S320: In this step, the tracking window data is analyzed from three dimensions: persistence, attenuation characteristics, and spectral stability, and each factor is calculated.

[0128] In one possible implementation of this step, the calculation steps for each component factor are as follows:

[0129] Calculate the persistence factor Calculate the offset sequence , Statistical satisfaction Points ,in for The standard deviation of . Then .

[0130] Calculate the attenuation factor : Calculate the absolute difference sum of the offset sequences .calculate .

[0131] Calculate the spectral consistency factor :right Perform an FFT to calculate the energy percentage of the first three main frequencies. The sequence is divided into three overlapping sub-windows, and calculations are performed separately for each. We obtained three values. Calculate the coefficient of variation for these three values. .but .

[0132] S330: In this step, the factors are weighted and synthesized. And determine the stability level based on the threshold.

[0133] In one possible implementation of this step, the synthesis and rating are as follows:

[0134] Take weight .calculate Set grading thresholds. , .like Rating It means high. If Rating Meaning: Middle. If Rating It means low. (The following phrase appears to be a separate, unrelated sentence: "will") and It is added to the candidate signal information and transmitted to S400.

[0135] In an adaptive threshold outlier dynamic determination method for multi-source new energy data, S400 integrates all information and dynamically generates a determination threshold suitable for the current scenario to achieve adaptive judgment.

[0136] Please refer to Figure 5 The diagram illustrates a flowchart of an exemplary adaptive threshold outlier dynamic determination method S400 for multi-source new energy data, the contents of which include:

[0137] S410: In this step, steady-state tokens, stability ratings, real-time operating conditions, and historical outlier statistical features are collected and constructed into a feature vector.

[0138] In one possible implementation of this step, the steps for constructing the input feature vector are as follows:

[0139] From the token (0 or 1). Rating from S300 Perform one-hot encoding: [1,0,0] [0,1,0] [0,0,1]. Read real-time operating conditions. Load rate Is it at night? (0 / 1), Weather Code Query outlier statistics for the past 24 hours. :Number of events Mean intensity Standard deviation Concatenate feature vectors .

[0140] S420: In this step, the feature vector is input into the threshold generation function to calculate the three-level threshold. .

[0141] In one possible implementation of this step, the dynamic threshold generation process is as follows:

[0142] The system has a basic threshold. Dynamic adjustments based on characteristics:

[0143] Based on token and rating adjustments: If The base value is magnified 1.5 times. (Based on rating) Multiply by coefficient: Multiply by 0.8, Multiply by 1.0, Multiply by 1.2.

[0144] Adjusted based on historical distribution: If and This indicates that outliers have been frequent and concentrated recently, and the threshold has been increased by 1.2 times to improve sensitivity.

[0145] Based on real-time operating condition adjustments: If and For low nighttime load, the threshold is tightened to 0.9 times.

[0146] Comprehensive calculation: ,in Corresponding rating coefficient, Corresponding to the adjustment coefficients mentioned above. Final satisfy .

[0147] S430: In this step, the deviation intensity of the candidate signal is compared with the dynamically generated three-level threshold to make a final judgment and give a confidence level.

[0148] In one possible implementation of this step, the final determination logic is as follows:

[0149] Obtain the deviation of the candidate signal Compare:

[0150] like It was identified as a high-confidence outlier.

[0151] like It was determined to be an outlier with medium confidence.

[0152] like It was identified as a low-confidence outlier.

[0153] like It was determined to be a non-outlier.

[0154] The final judgment result and the threshold used are output and passed to S500.

[0155] In an adaptive threshold outlier dynamic determination method for multi-source new energy data, the S500 achieves system learning and optimization, and continuously improves performance through feedback.

[0156] Please refer to Figure 6 The diagram illustrates a flowchart of an exemplary adaptive threshold outlier dynamic determination method S500 for multi-source new energy data, the contents of which include:

[0157] S510: In this step, the key data, results and context of a complete decision process are encapsulated into structured data units.

[0158] In one possible implementation of this step, the determination recording unit The encapsulation includes the following fields:

[0159] Meta-information: Unit ID, Device ID, Start and End Time.

[0160] Input and conditions: Original data identifier, steady-state token And the results of the three-level judgment.

[0161] Intermediate process: Auxiliary channel set Verification results Candidate signals .

[0162] Feature Analysis: Outlier Stability Index Component factors Stability rating .

[0163] Dynamic determination: generating threshold Final judgment result and confidence level.

[0164] Context: Real-time operating conditions Historical characteristics .

[0165] Verification label: Reserved field for subsequent entry of verification results via manual or independent methods.

[0166] S520: In this step, statistical analysis is performed on the judgment records to identify and adjust any potentially unreasonable thresholds in the previous three-layer judgment.

[0167] In one possible implementation of this step, the feedback calibration pre-logic steps are as follows:

[0168] Periodically collect judgment record units from past time periods. Filter out the verified TP samples and the false negative FN samples. Analyze the judgment status of these samples at S100:

[0169] For power stability assessment, the fluctuation coefficient of the FN sample is analyzed. Distribution. If a large number of FN Slightly higher than the current level Then it will automatically Increase by a small step. For environmental disturbance assessment, analyze the disturbance index of the FP sample. Distribution, similarly adjusted For channel quality assessment, the signal-to-noise ratio of the TP samples is analyzed. With reception rate Distribution, Adjustment and Adjusting direction is always about reducing verified errors.

[0170] S530: In this step, the accumulated records are used to evaluate and optimize the channel selection strategy and outlier stability index model parameters.

[0171] In one possible implementation of this step, the optimization process is as follows:

[0172] For a specific target channel Statistical analysis of each auxiliary channel in historical records Inconsistent results were given in the case of true positive TP ( The frequency of the state. Increase the selection priority of high-frequency channels. Decrease the priority of channels that frequently give consistent states in the case of false negatives (FN).

[0173] Collect a large number of verified tags Each contains Factors and Labels (TP is 1, others are 0). The optimal weights will be sought. Model as an optimization problem: minimize the predicted value and The cross-entropy loss between them. The new weights are solved using gradient descent. Based on the new weights... The distribution across all samples is recalculated and the grading threshold is updated. and To maintain the expected sample proportions at each level.

[0174] S540: In this step, the verified and effective adjustment and optimization results are updated to the system configuration library, completing one learning iteration.

[0175] In one possible implementation of this step, the update step is as follows:

[0176] The adjustment suggestions obtained from the analysis of S520 and S530 are packaged into a policy update package. This update package is applied during periods of low system load. Simulation verification can be performed before application. After the update is completed, the system runs subsequent judgments with the new configuration. The version and timestamp of each update are recorded to ensure traceability.

[0177] Example 2

[0178] An adaptive threshold outlier dynamic determination system for multi-source renewable energy data, the system includes a processor and a memory, the processor being configured to execute instructions stored in the memory to achieve:

[0179] The steady-state quality judgment module is used to determine the current operating status of new energy equipment in sequence: whether it is in a power stable period, whether there is a severe disturbance in the external environment, and whether the integrity and signal-to-noise ratio of the data acquisition channel meet the preset quality standards. Only when all three judgments pass, a steady-state measurable token is generated. If any judgment fails, a delayed re-examination mechanism is initiated, and the three judgments are re-executed after waiting for a preset sampling period.

[0180] The multi-channel collaborative verification module is used to perform multi-target consistency verification after obtaining a steady-state measurable token: select the target data channel and multiple associated auxiliary channels, construct the difference sequence between each auxiliary channel and the target channel, and verify each difference sequence; only when the verification confirms the existence of non-coordination deviation, output a candidate outlier signal containing timestamp, channel identifier and deviation intensity.

[0181] The stability quantification and rating module is used to analyze the time persistence, amplitude variation and spectral characteristics of candidate outlier signals, calculate the outlier stability index, and classify candidate outlier signals into different stability levels based on the outlier stability index.

[0182] The dynamic threshold generation module is used to dynamically generate judgment thresholds corresponding to each stability level based on whether there is a steady-state measurable token, the stability level of the candidate outlier signal, and the historical outlier distribution characteristics within the sliding time window.

[0183] The feedback optimization and recording module is used to encapsulate the outlier judgment results obtained based on the judgment threshold into a judgment recording unit, and feed back the outlier patterns and feature information contained in the judgment recording unit to the steady-state quality judgment module and the multi-channel collaborative verification module to dynamically calibrate the generation conditions of the steady-state measurable token and the selection strategy of the auxiliary channel, and optimize the model of the dynamic threshold generation module.

[0184] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0185] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0186] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0187] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0188] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0189] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for dynamically determining outliers using adaptive thresholds in multi-source renewable energy data, characterized in that, The methods include: For the current operating status of new energy equipment, it is determined in turn whether it is in a power stability period, whether there is a severe disturbance in the external environment, and whether the integrity of the data acquisition channel and the signal-to-noise ratio meet the preset quality standards; a steady-state measurable token is generated only when all three judgments are passed. If any layer fails the judgment, the delayed re-examination mechanism is activated, and the three-layer judgment is re-executed after waiting for the preset sampling period; After obtaining the steady-state measurable token, perform multi-objective consistency verification: select the target data channel and multiple associated auxiliary channels, construct the difference sequence between each auxiliary channel and the target channel, and verify each difference sequence; only when the verification confirms the existence of non-coordination deviation, output candidate outlier signals containing timestamps, channel identifiers, and deviation strength. Based on the candidate outlier signals, we analyze their time duration, amplitude variation and spectral characteristics, calculate the OSI, and classify the candidate outlier signals into different stability levels based on the OSI. Based on whether a steady-state measurable token is available, the stability level of the candidate outlier signal, and combined with the historical outlier distribution characteristics within the sliding time window, a judgment threshold corresponding to each stability level is dynamically generated. The outlier determination results obtained based on the determination threshold are encapsulated into determination record units. The outlier patterns and feature information contained in the determination record units are then fed back to the three-layer judgment logic and multi-objective consistency verification rules to dynamically calibrate the generation conditions of steady-state measurable tokens and the selection strategy of auxiliary channels, and to optimize the model for dynamically generating thresholds.

2. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 1, characterized in that, The three-level judgment of the current operating status of new energy equipment includes: First-level judgment: Collect data from the current moment and backtrack forward. Calculate the standard deviation of the power data sequence at each sampling point. with the mean The fluctuation coefficient is obtained. If the volatility coefficient Less than or equal to the preset power stability threshold If so, the power is determined to be stable; Second-level judgment: Synchronously read the most recent... For each sampling point, calculate the maximum rate of change of the wind speed sequence and light intensity sequence, based on the formula... Calculate the comprehensive disturbance index ,in , These are the rated reference values ​​for ambient wind speed and light intensity. , The weighting coefficients for wind speed and light intensity variations in the environmental disturbance index. The maximum rate of change of wind speed. The maximum rate of change of illumination; if Less than or equal to the preset environmental disturbance threshold If so, it is determined that there is no drastic disturbance in the environment; Third-level judgment: Check the target channel and preset key related channels in the most recent Packet reception rate per cycle Compared with the current signal-to-noise ratio If all inspected channels meet the requirements and If so, the channel quality is deemed acceptable; A steady-state measurable token is generated only when all three levels of judgment pass.

3. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 2, characterized in that, The startup delay re-check mechanism includes: If any level fails the check, check the current retry count. Is it less than the preset maximum number of retries? ; like Then Increment by 1, and the system waits for the preset delay period. After one sampling interval, jump back to the first-level judgment and re-execute; like If the current time period is marked as an unreliable interval, the current determination process will be terminated.

4. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 1, characterized in that, The multi-objective consistency check includes: Target channel determined based on steady-state measurable tokens. Based on the physical topology and electrical connections, select from predefined rules. These are associated auxiliary channels, forming an auxiliary channel set. ; Set length as Time alignment window for each auxiliary channel Simultaneously acquire its connection with the target channel The observed sequence within the alignment window, and the difference sequence is calculated. ; Calculate each difference sequence Standard deviation and target channel sequence With auxiliary channel sequence correlation coefficient With normalized root mean square error The calculation formula is: ;in, Represents the target channel observation sequence With auxiliary channel observation sequence Covariance. Represents the target channel observation sequence Standard deviation Indicates the auxiliary channel observation sequence Standard deviation Difference sequence In the window The average value within, , For target channel and auxiliary channels In the window The average of the observed values ​​within the range.

5. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 4, characterized in that, The verification confirms the existence of non-coordination deviations, including: For each auxiliary channel If both conditions are met and If the status of the auxiliary channel matches the target channel, it is determined that the channel is consistent with the target channel; otherwise, it is determined that they are inconsistent. The number of inconsistent auxiliary channels is counted. ; Calculate the current observation value of the target channel Its short historical window average deviation ,in The standard deviation of the historical window; If both conditions are met and If a global coordination deviation occurs, a candidate outlier signal is output.

6. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 1, characterized in that, The OSI calculation formula is as follows: ;in, OSI value , , Preset weights; As a persistent factor, As the attenuation factor, This is the spectrum consistency factor.

7. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 6, characterized in that, The OSI-based classification of candidate outlier signals into different stability levels includes: Preset high stability level threshold With medium stability level threshold ,and ; like Then, candidate outlier signals are classified into high stability levels; like Then, the candidate outlier signals are classified into medium stability levels; like Then, candidate outlier signals are classified as low stability levels.

8. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 1, characterized in that, The dynamically generated judgment thresholds corresponding to each stability level include: Construct an input feature vector, which includes at least: the Boolean value of the steady-state measurable token, the one-hot encoding of the stability level, real-time operating context information, and historical outlier statistical features within the sliding time window; The input feature vector is input into the threshold generation function. The function dynamically corrects the base threshold based on a predetermined adjustment rule and outputs three-level judgment thresholds corresponding to high, medium and low stability levels, respectively.

9. The adaptive threshold outlier dynamic determination method for multi-source new energy data according to claim 8, characterized in that, The predetermined adjustment rules include at least one of the following: If a steady-state measurable token is not obtained, the basic thresholds at each level will be amplified by the first preset coefficient. Based on the stability level, the basic threshold for each level is multiplied by a different level adjustment coefficient; If historical outlier statistics indicate that recent outlier events are frequent and concentrated, then the basic thresholds at all levels will be amplified by the second preset coefficient. If the real-time operating context indicates a low-load operating condition at night, then multiply the base thresholds at each level by a tightening factor of less than 1.

10. An adaptive threshold outlier dynamic determination system for multi-source new energy data for performing the method as described in any one of claims 1 to 9, characterized in that, The system includes a processor and memory, the processor being configured to execute instructions stored in memory to achieve: The steady-state quality judgment module is used to determine the current operating status of new energy equipment in sequence: whether it is in a power stability period, whether there is a severe disturbance in the external environment, and whether the integrity and signal-to-noise ratio of the data acquisition channel meet the preset quality standards; a steady-state measurable token is generated only when all three judgments pass. If any layer fails the judgment, the delayed re-examination mechanism is activated, and the three-layer judgment is re-executed after waiting for the preset sampling period; The multi-channel collaborative verification module is used to perform multi-target consistency verification after obtaining a steady-state measurable token: select the target data channel and multiple associated auxiliary channels, construct the difference sequence between each auxiliary channel and the target channel, and verify each difference sequence; only when the verification confirms the existence of non-coordination deviation, output a candidate outlier signal containing timestamp, channel identifier and deviation intensity. The stability quantification and rating module is used to analyze the time persistence, amplitude variation and spectral characteristics of candidate outlier signals, calculate the outlier stability index, and classify candidate outlier signals into different stability levels based on the outlier stability index. The dynamic threshold generation module is used to dynamically generate judgment thresholds corresponding to each stability level based on whether there is a steady-state measurable token, the stability level of the candidate outlier signal, and the historical outlier distribution characteristics within the sliding time window. The feedback optimization and recording module is used to encapsulate the outlier judgment results obtained based on the judgment threshold into a judgment recording unit, and feed back the outlier patterns and feature information contained in the judgment recording unit to the steady-state quality judgment module and the multi-channel collaborative verification module to dynamically calibrate the generation conditions of the steady-state measurable token and the selection strategy of the auxiliary channel, and optimize the model of the dynamic threshold generation module.

Images