Power grid inertia comprehensive evaluation method under multi-type disturbance
By adopting a comprehensive evaluation method for power grid inertia under multiple types of disturbances, the problem of insufficient evaluation accuracy of traditional methods under multiple types of disturbances is solved, realizing the accuracy and reliability of power grid inertia evaluation and providing support for power grid safety and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID JIANGSU ELECTRIC POWER CO LTD RESEARCH INSTITUTE
- Filing Date
- 2026-01-28
- Publication Date
- 2026-05-12
AI Technical Summary
Existing power grid inertia assessment methods lack adaptability under various types of disturbances, resulting in insufficient accuracy of assessment results. They also lack a unified confidence quantification system, ignore the temporal continuity of inertia, and fail to fully utilize time-series data, thus affecting the stability and reliability of assessment results.
A comprehensive evaluation method for power grid inertia under multiple types of disturbances is adopted. By periodically collecting frequency and power data, the disturbance type is identified and the initial inertia evaluation value is calculated. The confidence level is calculated by combining global parameters and a unified quantization formula. An LSTM-AE model is constructed for offline training. A sliding window mechanism is introduced for real-time evaluation. The confidence level weighting constraint, time series smoothing constraint and inertia fluctuation physical constraint are incorporated to achieve consistent evaluation optimization under multiple types of disturbances.
It enables accurate inertia assessment under various types of disturbances, improves the applicability and reliability of assessment results, provides support for the safe and stable operation of the power grid, ensures that the assessment results conform to the operating rules of the power grid, and has online real-time assessment capabilities.
Smart Images
Figure CN122020405A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power grid inertia assessment technology, and in particular to a comprehensive assessment method for power grid inertia under multiple types of disturbances. Background Technology
[0002] As the penetration rate of new energy sources in the power system continues to increase, the operation of the power grid is becoming increasingly complex, with diverse disturbance types, such as large / step disturbances, ramp disturbances, and noise-like disturbances. Traditional power grid inertia assessment methods are mostly designed for single disturbance types and lack adaptability to multiple disturbance types, resulting in insufficient accuracy of assessment results under complex disturbance scenarios. Furthermore, existing methods lack a unified confidence quantification system, making it impossible to compare the reliability of assessment results for different disturbance types horizontally. They also ignore the temporal continuity of inertia and physical fluctuation constraints, easily leading to abrupt changes in assessment values and inconsistencies with power grid operation patterns. In addition, the lack of an effective time-series data utilization mechanism during online assessment makes it difficult to fully explore the correlation between historical and real-time data, further affecting the stability and reliability of the assessment results.
[0003] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention
[0004] This invention provides a comprehensive evaluation method for power grid inertia under multiple types of disturbances, which can effectively solve the shortcomings of existing technologies in terms of multi-disturbance adaptation, confidence level uniformity, and timing consistency, and provide accurate inertia evaluation support for the safe and stable operation of the power grid.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A comprehensive assessment method for power grid inertia under multiple types of disturbances includes the following steps: S10. Periodically collect frequency and power data of the power grid operation, identify the disturbance type and calculate the initial inertia assessment value; wherein, the disturbance type includes large disturbance / step disturbance, ramp disturbance and noise-like disturbance; S20. Based on the identified disturbance types and the corresponding physical characteristics of the disturbances, and combined with the preset global parameters, the initial confidence level at each acquisition time is calculated using a unified quantization formula. S30. Construct a time series window and an LSTM-AE model. Based on the sample data in the time series window, incorporate confidence weighting constraints, time series smoothing constraints, and inertia fluctuation physical constraints to train the LSTM-AE model offline. S40. A real-time time series window is constructed using a sliding window mechanism. It is determined whether the real-time time series window contains sample data corresponding to large disturbances / step disturbances. If it does, the initial inertia assessment value, confidence level 1.0, and corresponding disturbance type at the time of the large disturbance / step disturbance are directly output. If it does not, the sample data in the real-time time series window are input into the trained LSTM-AE model for optimization. The confidence level is calibrated according to the confidence level over-limit handling rule. The final power grid inertia assessment result is output according to the disturbance type.
[0006] Furthermore, in step S20, the global parameter is the offline statistical maximum evaluation residual of the slope disturbance. Offline statistical maximum frequency stationarity index of noise disturbance Maximum signal-to-noise ratio of noise-like perturbations and offline statistics The specific methods for obtaining and updating it are as follows: Global parameter acquisition: First, collect historical disturbance sample data offline. For each disturbance type, calculate the value of the physical characteristic corresponding to each type of disturbance. After statistical analysis of all values of each characteristic, take the 99th percentile of each characteristic value as the corresponding global parameter value. Global parameter update: When the power grid operation mode changes significantly, the latest historical disturbance sample data is collected again, the 99th percentile of each feature value is recalculated, and the global parameter values are updated.
[0007] Further, in step S20, the confidence quantification formula for the slope disturbance is: In the formula, The slope fluctuation rate is the rate of change of the slope. =Standard deviation of the slope sequence / Mean of the slope sequence; The coefficients of determination for the power-time fit; To assess the fitting residuals for inertia; This represents the maximum evaluation residual for slope disturbances obtained through offline statistics.
[0008] Further, in step S20, the confidence quantification formula for the noise-like disturbance is: In the formula, For power fluctuation coefficient, =Standard deviation of the power series / Mean of the power series; It is a frequency stationarity index, namely the sliding variance of the frequency series; The maximum frequency stationarity index for noise disturbances obtained through offline statistics; The signal-to-noise ratio of the power signal; This represents the maximum signal-to-noise ratio for offline statistical noise-like disturbances.
[0009] Further, in step S20, the initial confidence level corresponding to each disturbance is specifically as follows: The confidence level of the slope disturbance ranges from 0.75 to 1.0, the confidence level of the noise-like disturbance ranges from 0.6 to 0.85, and the confidence level of the large disturbance / step disturbance is fixed at 1.0.
[0010] Furthermore, in step S30, the LSTM-AE model includes an LSTM encoding layer, an AE feature extraction layer, and an LSTM decoding layer; The LSTM encoding layer has 2-3 layers, and the hidden layer dimension is 32 / 64, which is used to capture short-term temporal dependencies at 12 time points; The AE feature extraction layer is used to convert the temporal features encoded by LSTM into global feature vectors and remove redundant noise; The LSTM decoding layer is used to reconstruct the two-dimensional sequence of the output at 12 time points, ensuring the consistency of the temporal structure.
[0011] Further, in step S30, the input and output of the LSTM-AE model are set, specifically as follows: The input is a normalized temporal window feature sequence with dimensions [12,3], corresponding to the initial inertia evaluation value, perturbation type encoding, and initial confidence level; The output is a two-dimensional sequence of 12 time points, with dimensions [12,2]. The first dimension is the corrected inertia value, and the second dimension is the optimized confidence.
[0012] Furthermore, the encoding method for the disturbance type and confidence level is specifically as follows: Disturbance type coding uses integer coding: large disturbances / step disturbances are coded as 2, ramp disturbances are coded as 1, and noise-like disturbances are coded as 0. The confidence level is encoded using floating-point numbers in the range [0,1]. Large disturbances / step disturbances are coded as 1.0, ramp disturbances are coded as floating-point numbers in the range of 0.75 to 1.0, and noise-like disturbances are coded as floating-point numbers in the range of 0.6 to 0.85. The encoded values are calculated using the confidence level quantification formula for the corresponding disturbance type.
[0013] Further, in step S30, the construction of the timing window specifically involves: Each time window contains sample data from 12 consecutive acquisition times; the sample data at a single time point includes the initial inertia assessment value, disturbance type code, and initial confidence level. Simultaneously label the disturbance type at each time point within the time series window to clearly identify the location of large disturbances / step disturbances.
[0014] Furthermore, in step S30, the temporal smoothing constraint is achieved by the sum of the absolute differences between the corrected inertia values at two adjacent time points in the loss function, which is used to force the inertia sequence to be smooth and continuous.
[0015] Furthermore, in step S30, the physical constraint on inertia fluctuation is achieved through the inertia fluctuation amplitude limit term within the window in the loss function, which is used to ensure that the output conforms to the physical law that the short-term fluctuation of the power grid inertia does not exceed 10%.
[0016] Furthermore, in step S30, the confidence weighting constraint is implemented through sample labeling and loss function-specific weighting constraints, and its output constraint has a higher priority than the time series smoothing constraint and the inertia fluctuation physical constraint. During model training, the model output for large perturbation / step perturbation points within the forced window is required to satisfy the following: the corrected inertia value is equal to its initial inertia evaluation value, and the optimized confidence level is 1.0.
[0017] Further, in step S30, the training loss function of the LSTM-AE model is: In the formula, The window confidence score for the w-th window is the average of the initial confidence scores at 12 time points within a single time window. Let be the initial confidence level at time t in the w-th window; Let be the initial inertia evaluation value at time t in the w-th window; This is the corrected inertia value output by the model at the corresponding time point, i.e., the first dimension of the output sequence; The initial confidence level is calculated using the formula at time t of the w-th window; The optimized confidence level of the model output at the corresponding time point is the second dimension of the output sequence; This is the time series smoothing coefficient, with a value ranging from 0.1 to 0.3; As a temporal smoothing constraint, it forces the inertia sequence to be smooth and continuous by minimizing the sum of the absolute differences between the corrected inertia values at adjacent time points; The inertia fluctuation constraint coefficient is recommended to be between 1.0 and 2.0. As a physical constraint term for inertia fluctuation, a penalty loss is generated only when the inertia fluctuation amplitude within the window exceeds 10%, ensuring that the output conforms to the physical laws of the power grid; Optimize the coefficients for confidence level; and These are the maximum and minimum values of the inertia sequence after the w-th window correction, respectively, which represent the first dimension of the output.
[0018] Furthermore, in step S40, a real-time timing window is constructed using a sliding window mechanism, specifically as follows: During the system initialization phase, sample data from the first 11 acquisition times are collected, and the initial inertia evaluation value at each time is used as the output. After the sample data from the 12th time are collected, the first complete real-time timing window is constructed. At each acquisition cycle, new sample data for the current moment is added, the earliest sample data in the real-time time series window is removed, and the sample data of the first 11 acquisition moments in the real-time time series window are replaced with the inertia value corrected by the LSTM-AE model at the corresponding moment, the corresponding disturbance type code, and the optimized confidence, so as to maintain the sample data of 12 consecutive moments in the window.
[0019] Further, in step S40, the confidence level crossing processing rule is specifically as follows: For the optimized confidence level of the model output, it is determined whether it exceeds the preset range according to the type of perturbation. If it does, a limiting calibration is performed. The specific calibration logic is as follows: When the perturbation type is a ramp perturbation, if the post-optimized confidence of the model output is greater than 1.0, the post-optimized confidence will be calibrated to 1.0; if the post-optimized confidence of the model output is less than 0.75, the post-optimized confidence will be calibrated to 0.75. When the perturbation type is noise-like perturbation, if the post-optimized confidence of the model output is greater than 0.85, the post-optimized confidence will be calibrated to 0.85; if the post-optimized confidence of the model output is less than 0.6, the post-optimized confidence will be calibrated to 0.6. When the perturbation type is a large perturbation / step perturbation, the post-optimized confidence is fixed at 1.0, and no over-limit calibration is performed.
[0020] Furthermore, in step S40, the final power grid inertia assessment results are output according to the type of disturbance, specifically as follows: If the real-time time series window contains sample data at the moment of a large disturbance / step disturbance, the initial inertia assessment value, disturbance type, and confidence level of 1.0 at that moment of the large disturbance / step disturbance will be directly output. If the real-time time series window contains only sample data of ramp perturbation and / or noise-like perturbation, then the real-time time series window is input into the trained LSTM-AE model, and the model outputs a two-dimensional sequence of 12 time points, namely the corrected inertia value and the optimized confidence. The corrected inertia value and the optimized confidence after the limit-breaking process are extracted as the final output, and the corresponding perturbation type is also output.
[0021] The technical solution of this invention can achieve the following technical effects: The LSTM-AE model in this invention combines the ability of LSTM to capture time-series dependencies with the feature extraction and noise reduction advantages of AE, effectively handling the dynamic correlation problem of time-series data. By introducing confidence quantification and multiple constraint mechanisms, it can achieve consistent optimization of evaluation results under multiple types of disturbances. This addresses the shortcomings of existing technologies in multi-disturbance adaptation, confidence level unification, and time-series consistency, providing accurate inertia assessment support for the safe and stable operation of the power grid.
[0022] This invention provides comprehensive coverage adaptation for multiple types of disturbances. By designing evaluation logic and confidence quantification schemes for large / step disturbances, ramp disturbances, and noise-like disturbances respectively, it overcomes the limitations of traditional methods in adapting to a single disturbance and improves the applicability of evaluation in complex scenarios.
[0023] The unified confidence quantification system in this invention establishes a standardized confidence calculation model to achieve a horizontal comparison of the reliability of evaluation results for different types of disturbances, providing an intuitive reliability reference for power grid dispatching decisions.
[0024] This invention employs multiple constraints to ensure evaluation quality. By incorporating temporal smoothing constraints and inertia fluctuation physical constraints into the LSTM-AE model training, the evaluation results are forced to conform to the power grid operation rules, avoiding unreasonable situations such as sudden changes and overshoots. At the same time, the priority constraint of large disturbance points ensures the evaluation accuracy of high-reliability disturbance data.
[0025] The online real-time evaluation capability utilizes a sliding window mechanism to fully leverage the correlation information of time-series data. The model output combines the corrected inertia value with the optimized confidence level, ensuring both evaluation accuracy and rapid response to changes in the real-time operating status of the power grid, thus providing timely support for frequency stability control. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 This is a flowchart illustrating the comprehensive evaluation method for power grid inertia under multiple types of disturbances. Detailed Implementation
[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0030] like Figure 1 As shown, this application provides a comprehensive evaluation method for power grid inertia under multiple types of disturbances, including the following steps: S10. Periodically collect frequency and power data of the power grid operation, identify the disturbance type and calculate the initial inertia assessment value; wherein, the disturbance type includes large disturbance / step disturbance, ramp disturbance and noise-like disturbance; This invention provides comprehensive coverage adaptation for multiple types of disturbances. By designing evaluation logic and confidence quantification schemes for large / step disturbances, ramp disturbances, and noise-like disturbances respectively, it overcomes the limitations of traditional methods in adapting to a single disturbance and improves the applicability of evaluation in complex scenarios.
[0031] S20. Based on the identified disturbance types and their corresponding physical characteristics, and combined with preset global parameters, calculate the initial confidence level at each acquisition moment using a unified quantization formula; establish a standardized confidence evaluation system to provide feature support for model training. In step S20, the global parameter is the offline statistical maximum evaluation residual of the slope disturbance. Offline statistical maximum frequency stationarity index of noise disturbance Maximum signal-to-noise ratio of noise-like perturbations and offline statistics The specific methods for obtaining and updating it are as follows: Global parameter acquisition: First, collect historical disturbance sample data offline. For each disturbance type, calculate the value of the physical characteristic corresponding to each type of disturbance. After statistical analysis of all values of each characteristic, take the 99th percentile of each characteristic value as the corresponding global parameter value. Global parameter update: When the power grid operation mode changes significantly, the latest historical disturbance sample data is collected again, and the 99th percentile of each feature value is recalculated using the same method as above. The global parameter values are then updated to ensure that the parameters are accurately adapted to the current operating state of the power grid.
[0032] The global parameters provide key input for the calculation of the initial confidence level in step S20, supporting the accurate output of the initial confidence level at each acquisition time using a unified quantization formula. This initial confidence level, after model optimization and limit-breaking processing, ultimately forms and outputs the optimized confidence level. The stability and dynamic update characteristics of the global parameters ensure the robustness of model training and adaptability to different power grid operating states.
[0033] Specifically, the confidence quantification formula for the slope disturbance is as follows: In the formula, The slope fluctuation rate is the rate of change of the slope. =Standard deviation of the slope sequence / Mean of the slope sequence; The coefficients of determination for the power-time fit; To assess the fitting residuals for inertia; This represents the maximum evaluation residual for slope disturbances obtained through offline statistics.
[0034] After calculating the initial confidence level of the slope disturbance using the above formula, calibration is performed in conjunction with the confidence level exceeding the limit handling rule in step S4 to ensure that the optimized confidence level is within the range of 0.75 to 1.0.
[0035] Specifically, the confidence quantification formula for the noise-like disturbance is as follows: In the formula, For power fluctuation coefficient, =Standard deviation of the power series / Mean of the power series; It is a frequency stationarity index, namely the sliding variance of the frequency series; The maximum frequency stationarity index for noise disturbances obtained through offline statistics; The signal-to-noise ratio of the power signal; The maximum signal-to-noise ratio for noise-like perturbations obtained through offline statistics; After calculating the initial confidence level of the noise-like disturbance using the above formula, calibration is performed in conjunction with the confidence level exceeding the limit handling rule in step S4 to ensure that the optimized confidence level is within the range of 0.6 to 0.85.
[0036] In step S20, the initial confidence level calculated by the confidence quantification formula is specifically as follows: the confidence level of the ramp disturbance ranges from 0.75 to 1.0, the confidence level of the noise-like disturbance ranges from 0.6 to 0.85, and the confidence level of the large disturbance / step disturbance is fixed at 1.0.
[0037] The unified confidence quantification system in this invention establishes a standardized confidence calculation model to achieve a horizontal comparison of the reliability of evaluation results for different types of disturbances, providing an intuitive reliability reference for power grid dispatching decisions.
[0038] S30. Construct a time series window and an LSTM-AE model. Based on the sample data in the time series window, incorporate confidence weighting constraints, time series smoothing constraints, and inertia fluctuation physical constraints to train the LSTM-AE model offline; thus obtaining a model with inertia value correction and confidence optimization capabilities. The LSTM-AE model comprises an LSTM encoding layer, an AE feature extraction layer, and an LSTM decoding layer. The LSTM encoding layer has 2-3 layers with a hidden layer dimension of 32 / 64, used to capture short-term temporal dependencies at 12 time points. The AE feature extraction layer is used to convert the temporal features encoded by LSTM into global feature vectors and remove redundant noise. The LSTM decoding layer is used to reconstruct the two-dimensional sequence at 12 time points to ensure the consistency of the temporal structure.
[0039] Furthermore, the input and output of the LSTM-AE model are defined as follows: The input is a normalized temporal window feature sequence with dimensions [12,3], corresponding to the initial inertia evaluation value, perturbation type encoding, and initial confidence level; The output is a two-dimensional sequence of 12 time points, with dimensions [12,2]. The first dimension is the corrected inertia value, and the second dimension is the optimized confidence.
[0040] In this embodiment, the encoding method for the disturbance type and confidence level is specifically as follows: Disturbance type coding uses integer coding: large disturbances / step disturbances are coded as 2, ramp disturbances are coded as 1, and noise-like disturbances are coded as 0. The confidence level is encoded using floating-point numbers in the range [0,1]. Large disturbances / step disturbances are coded as 1.0, ramp disturbances are coded as floating-point numbers in the range of 0.75 to 1.0, and noise-like disturbances are coded as floating-point numbers in the range of 0.6 to 0.85. The encoded values are calculated using the confidence level quantification formula for the corresponding disturbance type.
[0041] In step S30, the construction of the timing window is specifically as follows: Each time window contains sample data from 12 consecutive acquisition times; the sample data at a single time point includes the initial inertia assessment value, disturbance type code, and initial confidence level. During the construction of the time series window, the disturbance type identifiers at each moment within the time series window are simultaneously marked to clarify the location of large disturbances / step disturbances.
[0042] In this embodiment, the time-series smoothing constraint is achieved by the sum of the absolute differences between the corrected inertia values at two adjacent time points in the loss function. This is used to force the inertia sequence to be smooth and continuous, so that the evaluation results are more in line with the dynamic continuity of power grid operation.
[0043] The physical constraint on inertia fluctuation is achieved through the inertia fluctuation amplitude limit term within the window in the loss function, which is used to ensure that the output conforms to the physical law that the short-term fluctuation of the power grid inertia does not exceed 10%.
[0044] The confidence weighting constraint is implemented through sample labeling and loss function-specific weighting constraints. Its output constraint has higher priority than the temporal smoothing constraint and the inertia fluctuation physical constraint. During model training, the model output of large perturbation / step perturbation points within the forced window is required to meet the following conditions: the corrected inertia value is equal to its initial inertia evaluation value, and the optimized confidence is 1.0.
[0045] The training loss function of the LSTM-AE model is: In the formula, The window confidence score for the w-th window is the average of the initial confidence scores at 12 time points within a single time window. Let be the initial confidence level at time t in the w-th window; Let be the initial inertia evaluation value at time t in the w-th window; This is the corrected inertia value output by the model at the corresponding time point, i.e., the first dimension of the output sequence; The initial confidence level is calculated using the formula at time t of the w-th window; The optimized confidence level of the model output at the corresponding time point is the second dimension of the output sequence; This is the time series smoothing coefficient, with a value ranging from 0.1 to 0.3; As a temporal smoothing constraint, it forces the inertia sequence to be smooth and continuous by minimizing the sum of the absolute differences between the corrected inertia values at adjacent time points; The inertia fluctuation constraint coefficient is recommended to be between 1.0 and 2.0. As a physical constraint term for inertia fluctuation, a penalty loss is generated only when the inertia fluctuation amplitude within the window exceeds 10%, ensuring that the output conforms to the physical laws of the power grid; Optimize the coefficients for confidence level; and These are the maximum and minimum values of the inertia sequence after the w-th window correction, respectively, which represent the first dimension of the output.
[0046] For large disturbances / step disturbances marked within the window, force and The constraint is achieved by setting the weight of the error term of this type of point to the maximum value, and its priority is higher than the time series smoothing constraint and the inertia fluctuation physical constraint.
[0047] This invention employs multiple constraints to ensure evaluation quality. By incorporating temporal smoothing constraints and inertia fluctuation physical constraints into the LSTM-AE model training, the evaluation results are forced to conform to the power grid operation rules, avoiding unreasonable situations such as sudden changes and overshoots. At the same time, the priority constraint of large disturbance points ensures the evaluation accuracy of high-reliability disturbance data.
[0048] S40. A real-time time series window is constructed using a sliding window mechanism. It is determined whether the real-time time series window contains sample data corresponding to large disturbances / step disturbances. If it does, the initial inertia assessment value, confidence level 1.0, and corresponding disturbance type at the time of the large disturbance / step disturbance are directly output. If it does not, the sample data in the real-time time series window are input into the trained LSTM-AE model for optimization. The confidence level is calibrated according to the confidence level over-limit handling rule. The final power grid inertia assessment result is output according to the disturbance type.
[0049] In step S40, a real-time timing window is constructed using a sliding window mechanism, specifically as follows: During the system initialization phase, sample data from the first 11 acquisition times are collected, and the initial inertia evaluation value at each time is used as the output. After the sample data from the 12th time are collected, the first complete real-time timing window is constructed. At each acquisition cycle, new sample data for the current moment is added, the earliest sample data in the real-time time series window is removed, and the sample data of the first 11 acquisition moments in the real-time time series window are replaced with the inertia value corrected by the LSTM-AE model at the corresponding moment, the corresponding disturbance type code, and the optimized confidence, so as to maintain the sample data of 12 consecutive moments in the window.
[0050] In this scheme, the specific rules for handling confidence level exceeding the limit are as follows: For the optimized confidence level of the model output, it is determined whether it exceeds the preset range according to the type of perturbation. If it does, a limiting calibration is performed. The specific calibration logic is as follows: When the perturbation type is a ramp perturbation, if the post-optimized confidence of the model output is greater than 1.0, the post-optimized confidence will be calibrated to 1.0; if the post-optimized confidence of the model output is less than 0.75, the post-optimized confidence will be calibrated to 0.75. When the perturbation type is noise-like perturbation, if the post-optimized confidence of the model output is greater than 0.85, the post-optimized confidence will be calibrated to 0.85; if the post-optimized confidence of the model output is less than 0.6, the post-optimized confidence will be calibrated to 0.6. When the perturbation type is a large perturbation / step perturbation, the post-optimized confidence is fixed at 1.0, and no over-limit calibration is performed.
[0051] In step S40, the final power grid inertia assessment results are output according to the type of disturbance, specifically as follows: If the real-time time series window contains sample data at the moment of a large disturbance / step disturbance, the initial inertia assessment value, disturbance type, and confidence level of 1.0 at that moment of the large disturbance / step disturbance will be directly output. If the real-time time series window contains only sample data of ramp perturbation and / or noise-like perturbation, then the real-time time series window is input into the trained LSTM-AE model, and the model outputs a two-dimensional sequence of 12 time points, namely the corrected inertia value and the optimized confidence. The corrected inertia value and the optimized confidence after the limit-breaking process are extracted as the final output, and the corresponding perturbation type is also output.
[0052] Online real-time evaluation capability. Employing a sliding window mechanism to fully utilize the correlation information of time-series data, the model output combines corrected inertia values with optimized confidence levels, ensuring both evaluation accuracy and rapid response to real-time changes in the power grid's operating status, providing timely support for frequency stability control.
[0053] The LSTM-AE model in this invention combines the ability of LSTM to capture time-series dependencies with the feature extraction and noise reduction advantages of AE, and can effectively handle the dynamic correlation problem of time-series data. By introducing confidence quantification and multiple constraint mechanisms, the consistency optimization of evaluation results under multiple types of perturbations can be achieved.
[0054] In this invention, step S10 outputs the initial inertia assessment value and disturbance type; step S20 inputs and combines global parameters to output the initial confidence level; step S30 inputs the initial inertia, disturbance code, and initial confidence level to construct a time-series window and train the model; and step S40 inputs the real-time time-series window and the trained model to output the final evaluation result. This achieves full coverage adaptation for multiple types of disturbance scenarios. Through unified confidence quantification, multi-constraint modeling, and time-series window optimization, the accuracy, consistency, and reliability of power grid inertia assessment are improved.
[0055] This application provides a computer device including a processor and a memory. The memory stores a computer program executable by the processor, and when the computer program is executed by the processor, it performs the method described above.
[0056] This application also provides a storage medium storing a computer program, which is executed by a processor to perform the above-described method.
[0057] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic memory, flash memory, magnetic disk or optical disk.
[0058] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. "A plurality of" means two or more, unless otherwise explicitly specified.
[0059] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0060] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0061] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.
[0062] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.
[0063] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0064] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0065] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.
Claims
1. A comprehensive evaluation method for power grid inertia under multiple types of disturbances, characterized in that, Includes the following steps: S10. Periodically collect frequency and power data of the power grid operation, identify the disturbance type and calculate the initial inertia assessment value; wherein, the disturbance type includes large disturbance / step disturbance, ramp disturbance and noise-like disturbance; S20. Based on the identified disturbance types and the corresponding physical characteristics of the disturbances, and combined with the preset global parameters, the initial confidence level at each acquisition time is calculated using a unified quantization formula. S30. Construct a time series window and an LSTM-AE model. Based on the sample data in the time series window, incorporate confidence weighting constraints, time series smoothing constraints, and inertia fluctuation physical constraints to train the LSTM-AE model offline. S40. A real-time time series window is constructed using a sliding window mechanism. It is determined whether the real-time time series window contains sample data corresponding to large disturbances / step disturbances. If it does, the initial inertia assessment value, confidence level 1.0, and corresponding disturbance type at the time of the large disturbance / step disturbance are directly output. If it does not, the sample data in the real-time time series window are input into the trained LSTM-AE model for optimization. The confidence level is calibrated according to the confidence level over-limit handling rule. The final power grid inertia assessment result is output according to the disturbance type.
2. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S20, the global parameter is the offline statistical maximum evaluation residual of the slope disturbance. Offline statistical maximum frequency stationarity index of noise disturbance Maximum signal-to-noise ratio of noise-like perturbations and offline statistics The specific methods for obtaining and updating it are as follows: Global parameter acquisition: First, collect historical disturbance sample data offline. For each disturbance type, calculate the value of the physical characteristic corresponding to each type of disturbance. After statistical analysis of all values of each characteristic, take the 99th percentile of each characteristic value as the corresponding global parameter value. Global parameter update: When the power grid operation mode changes significantly, the latest historical disturbance sample data is collected again, the 99th percentile of each feature value is recalculated, and the global parameter values are updated.
3. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 2, characterized in that, In step S20, the confidence quantification formula for the slope disturbance is: In the formula, The slope fluctuation rate is the rate of change of the slope. =Standard deviation of the slope sequence / Mean of the slope sequence; The coefficients of determination for the power-time fit; To assess the fitting residuals for inertia; This represents the maximum evaluation residual for slope disturbances obtained through offline statistics.
4. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 2, characterized in that, In step S20, the confidence quantification formula for the noise-like disturbance is: In the formula, For power fluctuation coefficient, =Standard deviation of the power series / Mean of the power series; It is a frequency stationarity index, namely the sliding variance of the frequency series; The maximum frequency stationarity index for noise disturbances obtained through offline statistics; The signal-to-noise ratio of the power signal; This represents the maximum signal-to-noise ratio for offline statistical noise-like disturbances.
5. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S20, the initial confidence level corresponding to each disturbance is specifically as follows: The confidence level of the slope disturbance ranges from 0.75 to 1.0, the confidence level of the noise-like disturbance ranges from 0.6 to 0.85, and the confidence level of the large disturbance / step disturbance is fixed at 1.
0.
6. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the LSTM-AE model includes an LSTM encoding layer, an AE feature extraction layer, and an LSTM decoding layer; The LSTM encoding layer has 2-3 layers, and the hidden layer dimension is 32 / 64, which is used to capture short-term temporal dependencies at 12 time points; The AE feature extraction layer is used to convert the temporal features encoded by LSTM into global feature vectors and remove redundant noise; The LSTM decoding layer is used to reconstruct the two-dimensional sequence of the output at 12 time points, ensuring the consistency of the temporal structure.
7. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the input and output of the LSTM-AE model are set, specifically as follows: The input is a normalized temporal window feature sequence with dimensions [12,3], corresponding to the initial inertia evaluation value, perturbation type encoding, and initial confidence level; The output is a two-dimensional sequence of 12 time points, with dimensions [12,2]. The first dimension is the corrected inertia value, and the second dimension is the optimized confidence.
8. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, The encoding method for the disturbance type and confidence level is as follows: Disturbance type coding uses integer coding: large disturbances / step disturbances are coded as 2, ramp disturbances are coded as 1, and noise-like disturbances are coded as 0. The confidence level is encoded using floating-point numbers in the range [0,1]. Large disturbances / step disturbances are coded as 1.0, ramp disturbances are coded as floating-point numbers in the range of 0.75 to 1.0, and noise-like disturbances are coded as floating-point numbers in the range of 0.6 to 0.
85. The encoded values are calculated using the confidence level quantification formula for the corresponding disturbance type.
9. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the construction of the timing window is specifically as follows: Each time window contains sample data from 12 consecutive acquisition times; the sample data at a single time point includes the initial inertia assessment value, disturbance type code, and initial confidence level. Simultaneously label the disturbance type at each time point within the time series window to clearly identify the location of large disturbances / step disturbances.
10. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the temporal smoothing constraint is achieved by the sum of the absolute differences between the corrected inertia values at two adjacent time points in the loss function, which is used to force the inertia sequence to be smooth and continuous.
11. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the physical constraint on inertia fluctuation is achieved through the inertia fluctuation amplitude limit term within the window in the loss function, which is used to ensure that the output conforms to the physical law that the short-term fluctuation of the power grid inertia does not exceed 10%.
12. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the confidence weighting constraint is implemented through sample labeling and loss function-specific weighting constraint, and its output constraint has a higher priority than the time series smoothing constraint and the inertia fluctuation physical constraint. During model training, the model output for large perturbation / step perturbation points within the forced window is required to satisfy the following: the corrected inertia value is equal to its initial inertia evaluation value, and the optimized confidence level is 1.
0.
13. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S30, the training loss function of the LSTM-AE model is: In the formula, The window confidence score for the w-th window is the average of the initial confidence scores at 12 time points within a single time window. For the first The initial confidence level of each window at time t; For the first The initial inertia assessment value at time t of each window; This is the corrected inertia value output by the model at the corresponding time point, i.e., the first dimension of the output sequence; The initial confidence level is calculated using the formula at time t of the w-th window; The optimized confidence level of the model output at the corresponding time point is the second dimension of the output sequence; This is the time series smoothing coefficient, with a value ranging from 0.1 to 0.3; As a temporal smoothing constraint, it forces the inertia sequence to be smooth and continuous by minimizing the sum of the absolute differences between the corrected inertia values at adjacent time points; The inertia fluctuation constraint coefficient is recommended to be between 1.0 and 2.
0. As a physical constraint term for inertia fluctuation, a penalty loss is generated only when the inertia fluctuation amplitude within the window exceeds 10%, ensuring that the output conforms to the physical laws of the power grid; Optimize the coefficients for confidence level; and These are the maximum and minimum values of the inertia sequence after the w-th window correction, respectively, which represent the first dimension of the output.
14. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S40, a real-time timing window is constructed using a sliding window mechanism, specifically as follows: During the system initialization phase, sample data from the first 11 acquisition times are collected, and the initial inertia evaluation value at each time is used as the output. After the sample data from the 12th time are collected, the first complete real-time timing window is constructed. At each acquisition cycle, new sample data for the current moment is added, the earliest sample data in the real-time time series window is removed, and the sample data of the first 11 acquisition moments in the real-time time series window are replaced with the inertia value corrected by the LSTM-AE model at the corresponding moment, the corresponding disturbance type code, and the optimized confidence, so as to maintain the sample data of 12 consecutive moments in the window.
15. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S40, the confidence level crossing processing rule is specifically as follows: For the optimized confidence level of the model output, it is determined whether it exceeds the preset range according to the type of perturbation. If it does, a limiting calibration is performed. The specific calibration logic is as follows: When the perturbation type is a ramp perturbation, if the post-optimized confidence of the model output is greater than 1.0, the post-optimized confidence will be calibrated to 1.0; if the post-optimized confidence of the model output is less than 0.75, the post-optimized confidence will be calibrated to 0.
75. When the perturbation type is noise-like perturbation, if the post-optimized confidence of the model output is greater than 0.85, the post-optimized confidence will be calibrated to 0.85; if the post-optimized confidence of the model output is less than 0.6, the post-optimized confidence will be calibrated to 0.
6. When the perturbation type is a large perturbation / step perturbation, the post-optimized confidence is fixed at 1.0, and no over-limit calibration is performed.
16. The method for comprehensive evaluation of power grid inertia under multiple types of disturbances according to claim 1, characterized in that, In step S40, the final power grid inertia assessment results are output according to the type of disturbance, specifically as follows: If the real-time time series window contains sample data at the moment of a large disturbance / step disturbance, the initial inertia assessment value, disturbance type, and confidence level of 1.0 at that moment of the large disturbance / step disturbance will be directly output. If the real-time time series window contains only sample data of ramp perturbation and / or noise-like perturbation, then the real-time time series window is input into the trained LSTM-AE model, and the model outputs a two-dimensional sequence of 12 time points, namely the corrected inertia value and the optimized confidence. The corrected inertia value and the optimized confidence after the limit-breaking process are extracted as the final output, and the corresponding perturbation type is also output.