Search

CN-122020405-A - Comprehensive evaluation method for inertia of power grid under multi-type disturbance

CN122020405ACN 122020405 ACN122020405 ACN 122020405ACN-122020405-A

Abstract

The invention relates to the technical field of power grid inertia evaluation, in particular to a comprehensive power grid inertia evaluation method under multi-type disturbance, which comprises the steps of periodically collecting power grid frequency and power data, identifying disturbance types and calculating initial inertia evaluation values; calculating the confidence coefficient of each moment through a unified quantization formula based on disturbance physical characteristics; and constructing a real-time sequence window by adopting a sliding window mechanism during online evaluation, performing out-of-limit calibration on the confidence level after model reasoning, and outputting a final evaluation result in a classification mode. The method realizes full-coverage adaptation of multiple types of disturbance, improves the accuracy, consistency and physical credibility of the evaluation result through a unified confidence coefficient system and multiple constraint modeling, can quickly respond to the change of the real-time running state of the power grid, and provides powerful support for the frequency stability control of the high-proportion new energy power grid.

Inventors

  • LI YARAN
  • ZHANG NINGYU
  • JIA YUQIAO
  • HUANG XUN

Assignees

  • 国网江苏省电力有限公司电力科学研究院

Dates

Publication Date
20260512
Application Date
20260128

Claims (16)

  1. 1. The comprehensive evaluation method for the inertia of the power grid under the multi-type disturbance is characterized by comprising the following steps of: s10, periodically collecting frequency data and power data of power grid operation, identifying and obtaining a disturbance type and calculating an initial inertia evaluation value, wherein the disturbance type comprises large disturbance/step disturbance, slope disturbance and noise-like disturbance; s20, calculating initial confidence coefficient of each acquisition moment by adopting a unified quantization formula based on the identified disturbance type and physical characteristics of the corresponding disturbance and combining with preset global parameters; s30, constructing a time sequence window and an LSTM-AE model, and based on sample data in the time sequence window, integrating confidence weighting constraint, time sequence smooth constraint and inertia fluctuation physical constraint, and performing offline training on the LSTM-AE model; S40, constructing a real-time sequence window by adopting a sliding window mechanism, judging whether sample data corresponding to large disturbance/step disturbance are contained in the real-time sequence window, if so, directly outputting an initial inertia evaluation value, a confidence coefficient of 1.0 and a corresponding disturbance type at the moment of the large disturbance/step disturbance, otherwise, inputting the sample data in the real-time sequence window into a trained LSTM-AE model for optimization processing, executing out-of-limit calibration on the optimized confidence coefficient by utilizing a confidence coefficient line-crossing processing rule, and outputting a final power grid inertia evaluation result according to the disturbance type.
  2. 2. The method according to claim 1, wherein in step S20, the global parameter is an offline statistical ramp disturbance maximum evaluation residual Noise-like disturbance maximum frequency stability index of offline statistics And offline statistics of noise-like disturbance maximum signal-to-noise ratio The acquisition and update modes are as follows: Acquiring global parameters, namely firstly acquiring historical disturbance sample data offline, respectively calculating values of physical characteristics corresponding to various disturbance types according to each disturbance type, carrying out statistical analysis on all values of the characteristics, and taking 99% quantiles of the values of the characteristics as corresponding global parameter values; And updating global parameters, namely when the running mode of the power grid is changed remarkably, collecting the latest historical disturbance sample data again, counting again to calculate 99% of the fractional number of each characteristic value, and updating the global parameter value.
  3. 3. The method for comprehensive evaluation of grid inertia under multiple types of disturbances according to claim 2, where in step S20, the confidence quantization formula of the ramp disturbance is: In the formula, In order for the ramp slope to be a rate of fluctuation, Standard deviation of slope sequence/mean of slope sequence; determining coefficients for a power-time fit; Evaluating a fitting residual error for inertia; The residual is estimated maximally for ramp disturbances of offline statistics.
  4. 4. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 2, wherein in the step S20, the confidence quantization formula of the noise-like disturbance is: In the formula, As a coefficient of power fluctuation, Standard deviation of power sequence/mean of power sequence; Is the frequency stationarity index, i.e. the sliding variance of the frequency sequence; the maximum frequency stability index of noise-like disturbance is counted offline; Signal to noise ratio of the power signal; The maximum signal to noise ratio is noise-like disturbance of offline statistics.
  5. 5. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein in step S20, the initial confidence corresponding to each disturbance is specifically: The confidence coefficient of the slope disturbance ranges from 0.75 to 1.0, the confidence coefficient of the noise-like disturbance ranges from 0.6 to 0.85, and the confidence coefficient of the large disturbance/step disturbance is fixed to 1.0.
  6. 6. The method according to claim 1, wherein in step S30, the LSTM-AE model includes an LSTM encoding layer, an AE feature extraction layer, and an LSTM decoding layer; The LSTM coding layer is provided with 2-3 layers, the dimension of the hidden layer is 32/64, and the LSTM coding layer is used for capturing short-term time sequence dependence at 12 moments; The AE feature extraction layer is used for converting the time sequence features after LSTM coding into global feature vectors and eliminating redundant noise; The LSTM decoding layer is used for reconstructing and outputting a two-dimensional sequence of 12 moments and ensuring the consistency of a time sequence structure.
  7. 7. The method for comprehensive evaluation of grid inertia under multiple disturbance types according to claim 1, wherein in step S30, input and output of the LSTM-AE model are set, specifically: Inputting a normalized time sequence window characteristic sequence, wherein the dimension is [12,3], and the normalized time sequence window characteristic sequence corresponds to an initial inertia evaluation value, disturbance type coding and initial confidence coefficient; And outputting a two-dimensional sequence with 12 moments, wherein the dimensions are [12,2], the first dimension is the corrected inertial value, and the second dimension is the optimized confidence level.
  8. 8. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein the coding modes of the disturbance type and the confidence coefficient are specifically as follows: the disturbance type coding adopts integer coding, the large disturbance/step disturbance coding is 2, the slope disturbance coding is 1, and the noise-like disturbance coding is 0; Confidence coefficient coding, namely floating point number coding of [0,1] interval is adopted, large disturbance/step disturbance fixed coding is 1.0, slope disturbance coding is floating point number of 0.75-1.0 interval, noise-like disturbance coding is floating point number of 0.6-0.85 interval, and coding value is calculated by a confidence coefficient quantization formula corresponding to disturbance type.
  9. 9. The method for comprehensive evaluation of grid inertia under multiple types of disturbance according to claim 1, wherein in step S30, the time sequence window is constructed specifically as follows: Each time sequence window comprises sample data of 12 continuous acquisition moments, wherein the sample data of a single moment comprises an initial inertia evaluation value, a disturbance type code and an initial confidence coefficient; And synchronously marking disturbance type identifiers at all moments in a time sequence window, and determining the positions of large disturbance/step disturbance points.
  10. 10. The method for comprehensive evaluation of inertia of a power grid under multiple types of disturbances according to claim 1, wherein in step S30, the time-series smoothness constraint is implemented by a sum of absolute differences of corrected inertia values at two adjacent moments in a loss function, for forcing smooth succession of inertia sequences.
  11. 11. The method for comprehensive evaluation of inertia of a power grid under multiple types of disturbance according to claim 1, wherein in step S30, the physical constraint of fluctuation of inertia is implemented by a limit term of fluctuation amplitude of inertia in a window in a loss function, so as to ensure that the output conforms to a physical law that short-term fluctuation of inertia of the power grid does not exceed 10%.
  12. 12. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein in the step S30, the confidence weighting constraint is realized through sample labeling and loss function targeted weighting constraint, and the output constraint priority is higher than the time sequence smoothing constraint and the inertia fluctuation physical constraint; in the model training process, the model output of the large disturbance/step disturbance point in the forced window is satisfied, and the corrected inertia value is equal to the initial inertia evaluation value and the optimized confidence=1.0.
  13. 13. The method for comprehensive evaluation of grid inertia under multiple types of disturbances according to claim 1, where in step S30, the training loss function of the LSTM-AE model is: In the formula, The window confidence coefficient of the w window is the average value of initial confidence coefficient of 12 moments in a single time sequence window; is the first Initial confidence at time t of window; is the first Initial inertia evaluation values at the t-th moment of the window; outputting a corrected inertia value, namely a first dimension of an output sequence, for the corresponding moment model; the initial confidence coefficient calculated by the formula at the t moment of the w window; outputting the optimized confidence coefficient which is output by the corresponding time model, namely outputting a second dimension of the sequence; the value range is 0.1-0.3 for the time sequence smoothing coefficient; the inertia sequence is forced to be smooth and continuous by minimizing the sum of absolute differences of inertia values after correction at adjacent moments for a time sequence smooth constraint term; 1.0-2.0 is suggested for the inertia fluctuation constraint coefficient; the method is characterized in that the punishment loss is generated only when the inertia fluctuation range in the window exceeds 10% for the inertia fluctuation physical constraint item, so that the output is ensured to accord with the physical rule of the power grid; Optimizing coefficients for confidence; And And respectively outputting the maximum value and the minimum value of the inertia sequence after the w window correction, namely outputting the first dimension.
  14. 14. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein in the step S40, a sliding window mechanism is adopted to construct a real-time sequence window, specifically: In the system initialization stage, the sample data of the first 11 acquisition moments are accumulated, initial inertia evaluation values of all moments are adopted as output, and after the sample data of the 12 th moment are acquired, a first complete real-time sequence window is constructed; And (3) adding sample data at the current moment every one acquisition period, removing sample data at the earliest moment in the real-time sequence window, and replacing the sample data at the first 11 acquisition moments in the real-time sequence window with the inertia value corrected by the LSTM-AE model at the corresponding moment, the corresponding disturbance type coding and the optimized confidence coefficient so as to maintain the sample data at 12 continuous moments in the window.
  15. 15. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein in the step S40, the confidence line-crossing processing rule specifically includes: Judging whether the confidence coefficient exceeds a preset range according to the disturbance type aiming at the optimized confidence coefficient of the model output, and executing amplitude limiting calibration if the confidence coefficient exceeds the preset range, wherein specific calibration logic is as follows: when the disturbance type is slope disturbance, if the optimized confidence coefficient of the model output is greater than 1.0, calibrating the optimized confidence coefficient to be 1.0, and if the optimized confidence coefficient of the model output is less than 0.75, calibrating the optimized confidence coefficient to be 0.75; When the disturbance type is noise-like disturbance, if the optimized confidence coefficient output by the model is greater than 0.85, calibrating the optimized confidence coefficient to be 0.85, and if the optimized confidence coefficient output by the model is less than 0.6, calibrating the optimized confidence coefficient to be 0.6; When the disturbance type is large disturbance/step disturbance, the confidence coefficient after optimization is fixed to be 1.0, and out-of-limit calibration processing is not executed.
  16. 16. The method for comprehensively evaluating the inertia of the power grid under the multi-type disturbance according to claim 1, wherein in the step S40, the final power grid inertia evaluation result is output according to the disturbance type classification, specifically: If the real-time sequence window contains sample data of large disturbance/step disturbance moment, directly outputting initial inertia evaluation value, disturbance type and confidence coefficient of 1.0 of the large disturbance/step disturbance moment; If the real-time sequence window only contains sample data of slope disturbance and/or noise-like disturbance, inputting the real-time sequence window into a trained LSTM-AE model, outputting a 12-moment two-dimensional sequence by the model, namely, correcting the inertial value and optimizing the confidence coefficient, extracting the corrected inertial value at the current moment and optimizing the confidence coefficient after out-of-limit processing as final output, and outputting the corresponding disturbance type.

Description

Comprehensive evaluation method for inertia of power grid under multi-type disturbance Technical Field The invention relates to the technical field of power grid inertia evaluation, in particular to a comprehensive power grid inertia evaluation method under multi-type disturbance. Background As the permeability of new energy in the power system is continuously improved, the running state of the power grid is increasingly complex, and the disturbance type presents various characteristics, such as large disturbance/step disturbance, slope disturbance, noise-like disturbance and the like. The traditional power grid inertia evaluation method is designed aiming at a single disturbance type, and the lack of the adaptive capacity to multiple disturbance types leads to insufficient precision of an evaluation result under a complex disturbance scene. Meanwhile, a unified confidence quantification system is not established in the conventional method, the credibility of the evaluation results of different disturbance types cannot be transversely compared, inertia time sequence continuity and physical fluctuation constraint are ignored, and the problems that evaluation values are suddenly changed and the operation rule of a power grid is not met are easily caused. In addition, an effective time sequence data utilization mechanism is lacked in the online evaluation process, and the correlation information of the historical data and the real-time data is difficult to fully mine, so that the stability and the reliability of an evaluation result are further influenced. The information disclosed in this background section is only for enhancement of understanding of the general background of the disclosure and is not to be taken as an admission or any form of suggestion that this information forms the prior art that is well known to a person skilled in the art. Disclosure of Invention The invention provides a comprehensive power grid inertia evaluation method under multiple types of disturbance, which can effectively solve the defects of the prior art in the aspects of multiple disturbance adaptation, unified confidence coefficient, time sequence consistency and the like and provides accurate inertia evaluation support for safe and stable operation of a power grid. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: the comprehensive evaluation method for the inertia of the power grid under the multi-type disturbance comprises the following steps: s10, periodically collecting frequency data and power data of power grid operation, identifying and obtaining a disturbance type and calculating an initial inertia evaluation value, wherein the disturbance type comprises large disturbance/step disturbance, slope disturbance and noise-like disturbance; s20, calculating initial confidence coefficient of each acquisition moment by adopting a unified quantization formula based on the identified disturbance type and physical characteristics of the corresponding disturbance and combining with preset global parameters; s30, constructing a time sequence window and an LSTM-AE model, and based on sample data in the time sequence window, integrating confidence weighting constraint, time sequence smooth constraint and inertia fluctuation physical constraint, and performing offline training on the LSTM-AE model; S40, constructing a real-time sequence window by adopting a sliding window mechanism, judging whether sample data corresponding to large disturbance/step disturbance are contained in the real-time sequence window, if so, directly outputting an initial inertia evaluation value, a confidence coefficient of 1.0 and a corresponding disturbance type at the moment of the large disturbance/step disturbance, otherwise, inputting the sample data in the real-time sequence window into a trained LSTM-AE model for optimization processing, executing out-of-limit calibration on the optimized confidence coefficient by utilizing a confidence coefficient line-crossing processing rule, and outputting a final power grid inertia evaluation result according to the disturbance type. Further, in step S20, the global parameter is an offline statistical ramp disturbance maximum evaluation residualNoise-like disturbance maximum frequency stability index of offline statisticsAnd offline statistics of noise-like disturbance maximum signal-to-noise ratioThe acquisition and update modes are as follows: Acquiring global parameters, namely firstly acquiring historical disturbance sample data offline, respectively calculating values of physical characteristics corresponding to various disturbance types according to each disturbance type, carrying out statistical analysis on all values of the characteristics, and taking 99% quantiles of the values of the characteristics as corresponding global parameter values; And updating global parameters, namely when the running mode of the power grid is changed remarkably, collecting t