A broadband oscillation risk early warning method, system and device

Abstract

This invention belongs to the field of power system technology and discloses a method, system, and device for broadband oscillation risk early warning. The method includes: preprocessing initial broadband measurement data to obtain a broadband analysis dataset; extracting and fusing the time-series and coupled features of the initial broadband measurement data to obtain comprehensive features; identifying anomalies in the comprehensive features to obtain an anomaly feature set; inputting the anomaly feature set into a broadband oscillation risk assessment model to obtain a model-driven risk assessment result; fusing the anomaly feature set and the model-driven risk assessment result to obtain a comprehensive risk assessment result and determine the broadband oscillation risk level; generating graded early warning information based on the broadband oscillation risk level, and optimizing the feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model. This invention can improve the accuracy of broadband oscillation risk identification, trend prediction, and early warning grading.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a broadband oscillation risk early warning method, system and equipment. Background Technology

[0002] With the widespread application of large-scale new energy (wind power, photovoltaic) grid connection and flexible DC transmission and other power electronic equipment, the new power system exhibits significant characteristics of low inertia, strong nonlinearity, and wide frequency domain oscillation. The characteristics of the power grid system have gradually evolved from being dominated by traditional electromechanical transients to being dominated by the interaction of electromechanical transients and electromagnetic transients, and will eventually develop towards being dominated by all electromagnetic transients. This has led to frequent occurrences of new wide frequency oscillation stability problems such as sub / supersynchronous oscillations and medium-to-high frequency oscillations, and has caused faults such as new energy grid disconnection and thermal power unit tripping.

[0003] Broadband oscillations cover the entire frequency band from 0.1Hz to 2500Hz and include various forms such as fundamental frequency fluctuations, harmonic exceedances, and interharmonic abrupt changes. The coupling between power electronic equipment and the power grid, as well as among power electronic equipment itself, can easily amplify the oscillation amplitude. If timely warnings and control measures are not implemented, it may lead to unexpected events such as grid disconnection of renewable energy units, damage to power electronic equipment, and system power angle instability. Currently, the response to broadband oscillations is mainly reactive, meaning that intervention only occurs after the oscillation has occurred, generated significant impact, and can be monitored, lacking proactive warning mechanisms and response strategies.

[0004] Current broadband oscillation risk early warning technologies are mainly divided into two categories: data-driven and model-driven. However, both have obvious limitations and are difficult to meet the early warning needs of new power systems.

[0005] Firstly, existing data-driven early warning methods mostly monitor only a single frequency band (such as subsynchronous or supersynchronous) or a single signal (such as fundamental or harmonic waves). Although broadband measurement data can achieve unified monitoring of fundamental, interharmonic, and harmonic waves within a 2500Hz range, this monitoring often only detects oscillations after they have occurred and become apparent, triggering an alarm. Furthermore, existing oscillation monitoring and identification technologies and oscillation feature mining methods lack specificity, easily overlooking key precursory features of oscillations, leading to false alarms or missed alarms. Therefore, given the inadequacy of using broadband measurement data alone for oscillation alarms, relying solely on data for early warning is even more difficult to guarantee. Thus, it is necessary to combine power grid models to improve and refine the alarm methods.

[0006] Secondly, current model-driven early warning methods mostly use fixed parameter models, which are difficult to adapt to the dynamic changes in the operating conditions of new power systems (such as fluctuations in the penetration rate of new energy sources and flexible DC power adjustment), resulting in large modeling errors and insufficient accuracy in risk quantification.

[0007] Third, current oscillation early warning systems lack a deep integration mechanism between models and data. Data-driven anomaly identification and model-driven risk extrapolation are disconnected, failing to fully leverage the advantages of both. This results in low early warning accuracy, delayed response, and difficulty in predicting the development trend of oscillation risks in advance.

[0008] In addition, existing early warning methods often lack closed-loop optimization mechanisms, and early warning parameters cannot be dynamically adjusted based on system operation feedback. After long-term operation, the accuracy of early warning will gradually decrease, making it difficult to adapt to the dynamic needs of long-term operation of new power systems.

[0009] Therefore, how to provide a broadband oscillation risk early warning method, system, and equipment is an urgent problem to be solved. Summary of the Invention

[0010] This invention provides a broadband oscillation risk early warning method, system, and device to solve the problems mentioned above in the prior art.

[0011] According to a first aspect of the present invention, a method for early warning of broadband oscillation risk is provided.

[0012] In one embodiment, the broadband oscillation risk early warning method includes:

[0013] The initial broadband measurement data of the preset monitoring nodes in the power system are preprocessed to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency.

[0014] Temporal and coupled features of the initial broadband measurement data are extracted from the broadband analysis dataset, and the temporal and coupled features are fused to obtain comprehensive features. Anomaly detection algorithms are used to identify anomalies in the comprehensive features to obtain an anomaly feature set.

[0015] A broadband oscillation risk assessment model was constructed and optimized. An abnormal feature set was input into the optimized broadband oscillation risk assessment model, and risk quantification and oscillation development trend extrapolation were performed in combination with power system model information to obtain model-driven risk assessment results.

[0016] An adaptive weighted fusion algorithm is used to fuse the abnormal feature set and the model-driven risk assessment results to obtain a comprehensive risk assessment result, and the broadband oscillation risk level is determined by combining it with the preset risk grading standard.

[0017] Based on the broadband oscillation risk level, a graded early warning information is generated, and broadband measurement data of the power system after the graded early warning information is generated is collected. The feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model are optimized.

[0018] In one embodiment, the preprocessing of the initial broadband measurement data of the preset monitoring nodes in the power system to obtain a broadband analysis dataset includes the following steps:

[0019] The amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves in the monitoring nodes are collected using a preset sampling frequency and wide frequency range.

[0020] Based on preset oscillation alarm values ​​and data processing algorithms, the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonic waves are subjected to anomaly removal, noise suppression, redundancy compression, and standardization to obtain a broadband analysis dataset. The data processing algorithms include a three-standard-deviation algorithm, a wavelet packet denoising algorithm, a redundancy compression algorithm, and a Z-score standardization algorithm.

[0021] In one embodiment, the extraction of temporal features and coupled correlation features from the initial broadband measurement data of the broadband analysis dataset, the fusion of the temporal features and coupled correlation features to obtain comprehensive features, and the use of an anomaly detection algorithm to identify anomalies in the comprehensive features to obtain an anomaly feature set including:

[0022] The broadband analysis dataset is input into a preset temporal feature extraction network, and the feature extraction layer is used to extract the temporal features corresponding to the fundamental, interharmonic, and harmonic frequencies in the initial broadband measurement data; wherein, the temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer;

[0023] The weighting coefficients of the interharmonic time series features are adjusted using an attention enhancement layer, and the ratio of the fundamental frequency to the interharmonic frequency is calculated to obtain the coupling correlation features.

[0024] Using a feature fusion layer, the time-series features and coupled correlation features corresponding to the fundamental wave, interharmonic wave, and harmonic wave are time-aligned, normalized, and vector-concatenated to obtain comprehensive features;

[0025] An anomaly detection algorithm using the isolated forest algorithm is used to detect comprehensive features, and combined with a preset anomaly judgment threshold, the detected comprehensive features are labeled to obtain anomaly types; wherein, the anomaly types include at least one of fundamental frequency fluctuation anomaly, harmonic exceedance anomaly, or interharmonic abrupt change anomaly.

[0026] An anomaly feature set is constructed by combining the anomaly type, the time of anomaly occurrence, and the corresponding key feature values.

[0027] In one embodiment, the construction and optimization of the broadband oscillation risk assessment model, which involves inputting anomaly feature sets into the optimized model and combining power system model information for risk quantification and oscillation development trend deduction to obtain model-driven risk assessment results, includes the following steps:

[0028] A broadband oscillation risk assessment model is constructed based on an impedance-modal fusion model, and the model parameters of the broadband oscillation risk assessment model are optimized by combining a preset operating condition parameter mapping table; wherein, the model parameters include impedance correction coefficients and modal damping weights;

[0029] The abnormal feature set is input into the optimized broadband oscillation risk assessment model, and the sub-risk index of each abnormal feature in the abnormal feature set is calculated in combination with the power system model information;

[0030] A risk index sequence is constructed based on the sub-risk index and input into a pre-trained long short-term memory network for prediction. Based on the prediction results, the development trend of broadband oscillation risk is deduced, and the model-driven risk assessment results are obtained.

[0031] In one embodiment, the step of fusing the abnormal feature set and the model-driven risk assessment results using an adaptive weighted fusion algorithm to obtain a comprehensive risk assessment result, and determining the broadband oscillation risk level in conjunction with a preset risk grading standard, includes the following steps:

[0032] The model risk index is determined based on the model-driven risk assessment results, and the data risk index is determined based on the anomaly type, the time of anomaly occurrence, and the corresponding key feature values ​​in the anomaly feature set. The data-driven weight and model-driven weight are then determined using the data risk index and the model risk index.

[0033] Based on data-driven weights and model-driven weights, the data risk index and the model risk index are weighted and fused to obtain a comprehensive risk assessment result.

[0034] The comprehensive risk assessment results are compared with the preset risk classification thresholds to determine the broadband oscillation risk level.

[0035] In one embodiment, comparing the comprehensive risk assessment result with a preset risk grading threshold to determine the broadband oscillation risk level includes:

[0036] The preset risk classification thresholds include a first threshold, a second threshold, and a third threshold. When the comprehensive risk assessment result is less than the first threshold, it is determined to be at the first risk level. When the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined to be at the second risk level. When the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined to be at the third risk level. When the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined to be at the fourth risk level.

[0037] In one embodiment, the process of generating graded early warning information based on broadband oscillation risk levels, collecting broadband measurement data of the power system after the generation of graded early warning information, and optimizing the time-series feature extraction network and broadband oscillation risk assessment model using broadband measurement data includes:

[0038] Based on the broadband oscillation risk level, corresponding graded early warning information is generated. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention reminder information is generated; when the risk level is the third risk level, early warning reminder information and the first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning reminder information and the second prevention and control plan are generated.

[0039] Broadband measurement data collected after the generation of graded early warning information, the implementation of the first prevention and control plan, or the implementation of the second prevention and control plan are used to optimize the weight coefficients of the attention enhancement layer in the time-series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

[0040] In one embodiment, the expression for the comprehensive risk assessment result is:

[0041] ;

[0042] In the formula, Based on the comprehensive risk assessment results, For data risk index, For model risk index, Data-driven weights Use the model to drive the weights.

[0043] According to a second aspect of the present invention, a wideband oscillation risk early warning system is provided.

[0044] In one embodiment, the broadband oscillation risk early warning system includes:

[0045] A broadband full data and preprocessing module is used to preprocess the initial broadband measurement data of the preset monitoring nodes in the power system to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency.

[0046] The data-driven feature mining and risk assessment module is used to extract the temporal and coupled features of the initial broadband measurement data from the broadband analysis dataset, and to fuse the temporal and coupled features to obtain comprehensive features; anomaly detection algorithms are used to identify anomalies in the comprehensive features to obtain an anomaly feature set.

[0047] The model-driven risk assessment module is used to construct and optimize a broadband oscillation risk assessment model. It inputs the abnormal feature set into the optimized broadband oscillation risk assessment model and combines it with power system model information to quantify risk and extrapolate oscillation development trends, thereby obtaining model-driven risk assessment results.

[0048] The risk assessment module for data model fusion is used to fuse the abnormal feature set and model-driven risk assessment results using an adaptive weighted fusion algorithm to obtain a comprehensive risk assessment result, and to determine the broadband oscillation risk level in combination with the preset risk grading standard.

[0049] The early warning information output and closed-loop optimization module is used to generate graded early warning information based on the broadband oscillation risk level, and to collect broadband measurement data of the power system after the generation of graded early warning information, and to optimize the feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model.

[0050] In one embodiment, the broadband full data and preprocessing module includes: a data acquisition submodule and a data preprocessing submodule. The data acquisition submodule is used to acquire amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonics from the monitoring nodes using a preset sampling frequency and broadband range. The data preprocessing submodule is used to perform anomaly removal, noise suppression, redundancy compression, and standardization processing on the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonics based on preset oscillation alarm values ​​and data processing algorithms to obtain a broadband analysis dataset. The data processing algorithms include a three-standard-deviation algorithm, a wavelet packet denoising algorithm, a redundancy compression algorithm, and a Z-score standardization algorithm.

[0051] In one embodiment, the data-driven feature mining and risk assessment module includes: a feature extraction submodule, an attention enhancement submodule, a feature fusion submodule, an anomaly detection submodule, and an anomaly feature set construction submodule. The feature extraction submodule is used to input the broadband analysis dataset into a preset temporal feature extraction network, and uses the feature extraction layer to extract the temporal features corresponding to the fundamental frequency, interharmonic frequencies, and harmonic frequencies in the initial broadband measurement data. The temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer. The attention enhancement submodule is used to adjust the weight coefficients of the interharmonic temporal features using the attention enhancement layer and calculate the ratio of the fundamental frequency to the interharmonic frequencies to obtain... The system comprises the following modules: a coupling and correlation feature module; a feature fusion submodule, which uses a feature fusion layer to perform time alignment, normalization, and vector concatenation on the time-series features and coupling and correlation features corresponding to the fundamental wave, interharmonic waves, and harmonic waves to obtain comprehensive features; an anomaly detection submodule, which uses an anomaly detection algorithm based on the isolated forest algorithm to detect the comprehensive features and, in conjunction with a preset anomaly judgment threshold, marks the detected comprehensive features to obtain anomaly types; wherein, the anomaly types include at least one of fundamental wave fluctuation anomaly, harmonic exceedance anomaly, or interharmonic wave mutation anomaly; and an anomaly feature set construction submodule, which combines the anomaly type, the time of anomaly occurrence, and the corresponding key feature values ​​to construct an anomaly feature set. In one embodiment, the model-driven risk assessment module includes: a model building and optimization submodule, a risk index calculation submodule, and a trend extrapolation submodule. The model building and optimization submodule is used to construct a broadband oscillation risk assessment model based on an impedance-modal fusion model, and optimize the model parameters of the broadband oscillation risk assessment model by combining a preset operating condition parameter mapping table. The model parameters include impedance correction coefficients and modal damping weights. The risk index calculation submodule is used to input anomaly feature sets into the optimized broadband oscillation risk assessment model, and calculate a sub-risk index for each anomaly feature in the anomaly feature set by combining power system model information. The trend extrapolation submodule is used to construct a risk index sequence based on the sub-risk indices, input it into a pre-trained long short-term memory network for prediction, and extrapolate the development trend of broadband oscillation risk based on the prediction results to obtain the model-driven risk assessment result.

[0052] In one embodiment, the risk assessment module for data model fusion includes a weight determination submodule, a weighted fusion submodule, and a risk level determination submodule. The weight determination submodule determines the model risk index based on the model-driven risk assessment results, determines the data risk index based on the anomaly type, anomaly occurrence time, and corresponding key feature values ​​in the anomaly feature set, and determines the data-driven weights and model-driven weights using the data risk index and model risk index. The weighted fusion submodule performs weighted fusion of the data risk index and model risk index based on the data-driven weights and model-driven weights to obtain a comprehensive risk assessment result. The risk level determination submodule compares the comprehensive risk assessment result with a preset risk grading threshold to determine the broadband oscillation risk level.

[0053] In one embodiment, the risk level determination submodule includes a classification threshold submodule and a level judgment submodule. The classification threshold submodule is used to preset risk classification thresholds, including a first threshold, a second threshold, and a third threshold. The level judgment submodule is used to determine the risk level as follows: when the comprehensive risk assessment result is less than the first threshold, it is determined as a first risk level; when the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined as a second risk level; when the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined as a third risk level; and when the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined as a fourth risk level.

[0054] In one embodiment, the early warning information output and closed-loop optimization module includes an early warning information generation submodule and a feedback data acquisition and optimization submodule, wherein,

[0055] The early warning information generation submodule is used to generate corresponding graded early warning information based on the broadband oscillation risk level. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention prompt information is generated; when the risk level is the third risk level, early warning prompt information and the first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning prompt information and the second prevention and control plan are generated. The feedback data acquisition and optimization submodule is used to collect broadband measurement data after the graded early warning information is generated, after the first prevention and control plan is implemented, or after the second prevention and control plan is implemented, and to optimize the weight coefficients of the attention enhancement layer in the time series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

[0056] According to a third aspect of the present invention, a computer device is provided.

[0057] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0058] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0059] 1. Wideband Unified Monitoring: Breaking through the limitations of existing technologies that monitor single frequency bands and single signals, it enables unified real-time monitoring of one or more frequency bands of fundamental, interharmonic, and harmonic signals within the 2500Hz range. It fully captures the full-band characteristics of wideband oscillations, avoids missing key oscillation precursors, and reduces the false alarm rate.

[0060] 2. Flexible and adjustable model: Construct a broadband oscillation risk assessment model that can be flexibly adjusted. The model parameters can be dynamically adjusted according to the operating conditions of the new power system (new energy penetration rate, flexible DC capacity) to adapt to the time-varying characteristics of the system and solve the problems of large modeling error and poor adaptability of fixed parameter models.

[0061] 3. Data and Model Integration: An adaptive weighted fusion algorithm is adopted to achieve deep integration of data-driven anomaly identification and model-driven risk extrapolation. Weights can be dynamically allocated according to data quality and model accuracy to improve the accuracy of risk quantification and level determination.

[0062] 4. Closed-loop optimization capability: Construct a closed-loop early warning mechanism that includes data collection, mining, modeling, fusion, early warning, and optimization. By dynamically adjusting model parameters, fusion weights, and early warning thresholds through real-time feedback data, the early warning performance is continuously optimized, solving the problem that the accuracy of existing early warning methods decreases over time.

[0063] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0065] Figure 1 This is a flowchart illustrating a broadband oscillation risk early warning method according to an exemplary embodiment;

[0066] Figure 2 This is a schematic diagram illustrating the structure of a broadband oscillation risk early warning system according to an exemplary embodiment;

[0067] Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment;

[0068] Figure 4 This is a flowchart illustrating the time-series feature extraction network in a broadband oscillation risk early warning method according to an exemplary embodiment;

[0069] Figure 5 This is a schematic diagram of the processing flow of a broadband oscillation risk assessment model for a broadband oscillation risk early warning method according to an exemplary embodiment;

[0070] Figure 6 This is a schematic diagram illustrating the model data fusion risk assessment process of a broadband oscillation risk early warning method according to an exemplary embodiment;

[0071] Figure 7 This is a schematic diagram illustrating the closed-loop optimization process of a broadband oscillation risk early warning method according to an exemplary embodiment. Detailed Implementation

[0072] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0073] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0074] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0075] Figure 1 An embodiment of a broadband oscillation risk warning method of the present invention is shown.

[0076] In this optional embodiment, the broadband oscillation risk warning method includes:

[0077] S101. The initial broadband measurement data of the preset monitoring nodes in the power system are preprocessed to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency.

[0078] S102. Extract the temporal features and coupled correlation features of the initial broadband measurement data from the broadband analysis dataset, and fuse the temporal features and coupled correlation features to obtain comprehensive features; use anomaly detection algorithms to identify anomalies in the comprehensive features to obtain an anomaly feature set;

[0079] S103. Construct and optimize a broadband oscillation risk assessment model. Input the abnormal feature set into the optimized broadband oscillation risk assessment model, and combine it with power system model information to quantify the risk and extrapolate the oscillation development trend to obtain the model-driven risk assessment results.

[0080] S104. An adaptive weighted fusion algorithm is used to fuse the abnormal feature set and the model-driven risk assessment results to obtain a comprehensive risk assessment result, and the broadband oscillation risk level is determined in combination with the preset risk grading standard.

[0081] S105. Generate graded early warning information based on broadband oscillation risk level, and collect broadband measurement data of the power system after generating graded early warning information, and optimize the feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model.

[0082] In this optional embodiment, the preprocessing of the initial broadband measurement data of the preset monitoring nodes in the power system to obtain the broadband analysis dataset includes the following steps:

[0083] The amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves in the monitoring nodes are collected using a preset sampling frequency and wide frequency range.

[0084] Based on preset oscillation alarm values ​​and data processing algorithms, the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonic waves are subjected to anomaly removal, noise suppression, redundancy compression, and standardization to obtain a broadband analysis dataset. The data processing algorithms include a three-standard-deviation algorithm, a wavelet packet denoising algorithm, a redundancy compression algorithm, and a Z-score standardization algorithm.

[0085] In this optional embodiment, the step of extracting time-series features and coupled correlation features from the initial broadband measurement data of the broadband analysis dataset, fusing the time-series features and coupled correlation features to obtain comprehensive features, and using an anomaly detection algorithm to identify anomalies in the comprehensive features to obtain an anomaly feature set includes:

[0086] The broadband analysis dataset is input into a preset temporal feature extraction network, and the feature extraction layer is used to extract the temporal features corresponding to the fundamental, interharmonic, and harmonic frequencies in the initial broadband measurement data; wherein, the temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer;

[0087] The weighting coefficients of the interharmonic time series features are adjusted using an attention enhancement layer, and the ratio of the fundamental frequency to the interharmonic frequency is calculated to obtain the coupling correlation features.

[0088] Using a feature fusion layer, the time-series features and coupled correlation features corresponding to the fundamental wave, interharmonic wave, and harmonic wave are time-aligned, normalized, and vector-concatenated to obtain comprehensive features;

[0089] An anomaly detection algorithm using the isolated forest algorithm is used to detect comprehensive features, and combined with a preset anomaly judgment threshold, the detected comprehensive features are labeled to obtain anomaly types; wherein, the anomaly types include at least one of fundamental frequency fluctuation anomaly, harmonic exceedance anomaly, or interharmonic abrupt change anomaly.

[0090] An anomaly feature set is constructed by combining the anomaly type, the time of anomaly occurrence, and the corresponding key feature values.

[0091] In this optional embodiment, the construction and optimization of the broadband oscillation risk assessment model, which involves inputting the abnormal feature set into the optimized broadband oscillation risk assessment model and combining it with power system model information to quantify the risk and extrapolate the oscillation development trend, to obtain the model-driven risk assessment results, includes the following steps:

[0092] A broadband oscillation risk assessment model is constructed based on an impedance-modal fusion model, and the model parameters of the broadband oscillation risk assessment model are optimized by combining a preset operating condition parameter mapping table; wherein, the model parameters include impedance correction coefficients and modal damping weights;

[0093] The abnormal feature set is input into the optimized broadband oscillation risk assessment model, and the sub-risk index of each abnormal feature in the abnormal feature set is calculated in combination with the power system model information;

[0094] A risk index sequence is constructed based on the sub-risk index and input into a pre-trained long short-term memory network for prediction. Based on the prediction results, the development trend of broadband oscillation risk is deduced, and the model-driven risk assessment results are obtained.

[0095] In this optional embodiment, the step of fusing the abnormal feature set and the model-driven risk assessment results using an adaptive weighted fusion algorithm to obtain a comprehensive risk assessment result, and determining the broadband oscillation risk level in conjunction with a preset risk grading standard, includes the following steps:

[0096] The model risk index is determined based on the model-driven risk assessment results. The data risk index is determined based on the anomaly type, the time of anomaly occurrence, and the corresponding key feature values ​​in the anomaly feature set. The data-driven weight and model-driven weight are determined using the data risk index and the model risk index.

[0097] Based on data-driven weights and model-driven weights, the data risk index and the model risk index are weighted and fused to obtain a comprehensive risk assessment result.

[0098] The comprehensive risk assessment results are compared with the preset risk classification thresholds to determine the broadband oscillation risk level.

[0099] In this optional embodiment, comparing the comprehensive risk assessment results with a preset risk grading threshold to determine the broadband oscillation risk level includes:

[0100] The preset risk classification thresholds include a first threshold, a second threshold, and a third threshold;

[0101] When the comprehensive risk assessment result is less than the first threshold, it is determined to be at the first risk level; when the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined to be at the second risk level; when the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined to be at the third risk level; when the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined to be at the fourth risk level.

[0102] In this optional embodiment, the step of generating graded early warning information based on broadband oscillation risk levels, collecting broadband measurement data of the power system after the generation of graded early warning information, and optimizing the time-series feature extraction network and broadband oscillation risk assessment model using broadband measurement data includes:

[0103] Based on the broadband oscillation risk level, corresponding graded early warning information is generated. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention reminder information is generated; when the risk level is the third risk level, early warning reminder information and the first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning reminder information and the second prevention and control plan are generated.

[0104] Broadband measurement data collected after the generation of graded early warning information, the implementation of the first prevention and control plan, or the implementation of the second prevention and control plan are used to optimize the weight coefficients of the attention enhancement layer in the time-series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

[0105] In this optional embodiment, the expression for the comprehensive risk assessment result is:

[0106] ;

[0107] In the formula, Based on the comprehensive risk assessment results, For data risk index, For model risk index, Data-driven weights Use the model to drive the weights.

[0108] Figure 2 An embodiment of a broadband oscillation risk early warning system of the present invention is shown.

[0109] In this optional embodiment, the broadband oscillation risk early warning system includes:

[0110] The broadband full data and preprocessing module 201 is used to preprocess the initial broadband measurement data of the preset monitoring nodes in the power system to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency.

[0111] The data-driven feature mining and risk assessment module 202 is connected to the broadband full data and preprocessing module. It receives the broadband analysis dataset and extracts the temporal features and coupled correlation features of the initial broadband measurement data from the broadband analysis dataset. It then fuses the temporal features and coupled correlation features to obtain comprehensive features. An anomaly detection algorithm is used to identify anomalies in the comprehensive features to obtain an anomaly feature set.

[0112] The model-driven risk assessment module 203 is connected to the data-driven feature mining and risk assessment module. It receives the abnormal feature set, which is used to construct and optimize the broadband oscillation risk assessment model. The abnormal feature set is input into the optimized broadband oscillation risk assessment model, and risk quantification and oscillation development trend deduction are performed in combination with power system model information to obtain the model-driven risk assessment result.

[0113] The risk assessment module 204 for data model fusion is connected to the data-driven feature mining and risk assessment module and the model-driven risk assessment module, respectively. It receives the data risk index and the model-driven risk assessment result, and uses an adaptive weighted fusion algorithm to fuse the abnormal feature set and the model-driven risk assessment result to obtain a comprehensive risk assessment result. It also determines the broadband oscillation risk level by combining the result with a preset risk grading standard.

[0114] The early warning information output and closed-loop optimization module 205 is connected to the risk assessment module that integrates the data model. It receives the comprehensive risk assessment result and the broadband oscillation risk level, and is used to generate graded early warning information based on the broadband oscillation risk level. It also collects broadband measurement data of the power system after the graded early warning information is generated, and optimizes the feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model.

[0115] The early warning information output and closed-loop optimization module 205 is also connected to the data-driven feature mining and risk assessment module 202 and the model-driven risk assessment module 203 respectively. It is used to optimize the feature extraction process in the data-driven feature mining and risk assessment module 202 and optimize the broadband oscillation risk assessment model in the model-driven risk assessment module 203 based on the broadband measurement data after generating hierarchical early warning information, so as to form a closed-loop early warning process of broadband measurement data acquisition, data-driven anomaly identification, model-driven risk assessment, data model fusion judgment, hierarchical early warning output and feedback optimization.

[0116] In this optional embodiment, the broadband full data and preprocessing module includes: a data acquisition submodule and a data preprocessing submodule. The data acquisition submodule is used to acquire amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonics from the monitoring nodes using a preset sampling frequency and broadband range. The data preprocessing submodule is used to perform anomaly removal, noise suppression, redundancy compression, and standardization processing on the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonics based on preset oscillation alarm values ​​and data processing algorithms to obtain a broadband analysis dataset. The data processing algorithms include a three-standard-deviation algorithm, a wavelet packet denoising algorithm, a redundancy compression algorithm, and a Z-score normalization algorithm.

[0117] In this optional embodiment, the data-driven feature mining and risk assessment module includes: a feature extraction submodule, an attention enhancement submodule, a feature fusion submodule, an anomaly detection submodule, and an anomaly feature set construction submodule. The feature extraction submodule is used to input the broadband analysis dataset into a preset temporal feature extraction network, and to extract the temporal features corresponding to the fundamental frequency, interharmonic frequencies, and harmonic frequencies from the initial broadband measurement data using the feature extraction layer. The temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer. The attention enhancement submodule is used to adjust the weight coefficients of the interharmonic temporal features using the attention enhancement layer and calculate the ratio of the fundamental frequency to the interharmonic frequencies to obtain coupled-related features. The feature fusion submodule is used to perform time alignment, normalization, and vector concatenation processing on the temporal features and coupled-related features corresponding to the fundamental frequency, interharmonic frequencies, and harmonic frequencies using the feature fusion layer to obtain comprehensive features.

[0118] The anomaly detection submodule is used to detect comprehensive features using the isolated forest algorithm and, in conjunction with a preset anomaly judgment threshold, to label the detected comprehensive features to obtain the anomaly type; wherein, the anomaly type includes at least one of fundamental wave fluctuation anomaly, harmonic exceedance anomaly, or interharmonic abrupt change anomaly; the anomaly feature set construction submodule is used to construct an anomaly feature set by combining the anomaly type, the time of anomaly occurrence, and the corresponding key feature values.

[0119] In this optional embodiment, the model-driven risk assessment module includes: a model building and optimization submodule, a risk index calculation submodule, and a trend extrapolation submodule. The model building and optimization submodule is used to construct a broadband oscillation risk assessment model based on an impedance-modal fusion model, and optimize the model parameters of the broadband oscillation risk assessment model by combining a preset operating condition parameter mapping table. The model parameters include impedance correction coefficients and modal damping weights. The risk index calculation submodule is used to input anomaly feature sets into the optimized broadband oscillation risk assessment model, and calculate a sub-risk index for each anomaly feature in the anomaly feature set by combining power system model information. The trend extrapolation submodule is used to construct a risk index sequence based on the sub-risk indices, input it into a pre-trained long short-term memory network for prediction, and extrapolate the development trend of broadband oscillation risk based on the prediction results to obtain the model-driven risk assessment result.

[0120] In this optional embodiment, the risk assessment module for data model fusion includes a weight determination submodule, a weighted fusion submodule, and a risk level determination submodule. The weight determination submodule is used to determine the model risk index based on the model-driven risk assessment results, determine the data risk index based on the anomaly type, anomaly occurrence time, and corresponding key feature values ​​in the anomaly feature set, and determine the data-driven weight and model-driven weight using the data risk index and model risk index. The weighted fusion submodule is used to perform weighted fusion of the data risk index and model risk index based on the data-driven weight and model-driven weight to obtain a comprehensive risk assessment result. The risk level determination submodule is used to compare the comprehensive risk assessment result with a preset risk classification threshold to determine the broadband oscillation risk level.

[0121] In this optional embodiment, the risk level determination submodule includes a classification threshold submodule and a level judgment submodule. The classification threshold submodule is used to preset risk classification thresholds, including a first threshold, a second threshold, and a third threshold. The level judgment submodule is used to determine the risk level as follows: when the comprehensive risk assessment result is less than the first threshold, it is determined as a first risk level; when the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined as a second risk level; when the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined as a third risk level; and when the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined as a fourth risk level.

[0122] In this optional embodiment, the early warning information output and closed-loop optimization module includes an early warning information generation submodule and a feedback data acquisition and optimization submodule. The early warning information generation submodule is used to generate corresponding graded early warning information based on the broadband oscillation risk level. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention prompt information is generated; when the risk level is the third risk level, early warning prompt information and a first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning prompt information and a second prevention and control plan are generated. The feedback data acquisition and optimization submodule is used to collect broadband measurement data after the graded early warning information is generated, after the first prevention and control plan is executed, or after the second prevention and control plan is executed, and to optimize the weight coefficients of the attention enhancement layer in the time-series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

[0123] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspective of architecture and principle, as follows:

[0124] Broadband measurement terminals equipped with the device of this invention are deployed at preset monitoring nodes (such as new energy power plants, substations, and flexible DC converter stations). The three-phase voltage and three-phase current signals of the monitoring nodes are collected according to the preset sampling frequency to obtain the amplitude, frequency, and phase data (i.e., initial broadband measurement data) of the fundamental, interharmonic, and harmonic waves within the preset broadband range. In this invention, the preset sampling frequency is not lower than 12.8kHz so that the acquisition results can cover broadband signals within the range of 2500Hz. In addition, in practical applications, the broadband measurement terminal can use a 1s time window for spectrum analysis, corresponding to a frequency resolution of 1Hz.

[0125] It should be noted that, in this embodiment, the initial broadband measurement data refers to the broadband measurement data collected by the preset monitoring node and used to construct the broadband analysis dataset before the generation of the graded early warning information; the broadband measurement data collected after the generation of the graded early warning information is used as feedback broadband measurement data to optimize the subsequent feature extraction process and the broadband oscillation risk assessment model.

[0126] Since some broadband oscillation signals fall under the category of interharmonic frequency data, their oscillation frequencies are not necessarily integer multiples of 1 Hz. Therefore, when obtaining interharmonic frequency data through spectral analysis, the target oscillation frequency component may be dispersed across multiple adjacent frequency points. For example, when the actual oscillation frequencies of the fundamental, interharmonic, and harmonic components are 17.8 Hz, at a frequency resolution of 1 Hz, their frequency energy may be dispersed across adjacent frequency points such as 17 Hz, 18 Hz, 19 Hz, and 16 Hz. Simultaneously, to reduce the impact of spectral leakage on the frequency identification results, interpolation fitting is performed on the adjacent frequency points corresponding to the target oscillation frequency component. Preferably, a bispectral interpolation method is used to correct the target oscillation frequency to obtain the corrected oscillation frequency data, i.e., the frequency data of the fundamental, interharmonic, and harmonic components.

[0127] After obtaining the oscillation alarm value, the amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves are subjected to anomaly removal, noise suppression, redundancy compression, and standardization. Specifically, a three-standard-deviation algorithm is used to remove abnormal data during the acquisition process (such as missing values ​​caused by sampling interruptions or extreme values ​​caused by sudden interference). A wavelet packet denoising algorithm is used to suppress broadband noise interference in the broadband measurement data. A redundancy compression algorithm is used to compress the denoised broadband measurement data. The redundancy compression algorithm can be implemented using a lightweight encoder, and the compression ratio can be set according to the data storage pressure and computing resources. In this embodiment, the compression ratio is set to 10:1. The Z-score standardization algorithm is used to map the amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves to a unified numerical scale. In this embodiment, the mapping interval is [-1,1], thereby eliminating the dimensional differences between different data types.

[0128] After the above processing steps, a broadband analysis dataset including feature identifiers, acquisition time, and measurement values ​​is generated.

[0129] like Figure 4 As shown, the broadband analysis dataset is input into a preset temporal feature extraction network (preferably a temporal feature extraction model based on a temporal convolutional network TCN in this embodiment), and the network includes a data input layer, a feature extraction layer, an attention enhancement layer, a feature fusion layer, and a recognition layer.

[0130] The feature extraction layer is used to extract the time-series features corresponding to the fundamental wave, interharmonic waves, and harmonic waves, namely, the fundamental wave amplitude variation feature, the fundamental wave frequency fluctuation feature, the harmonic amplitude variation feature, the harmonic order variation feature, the interharmonic amplitude variation feature, and the interharmonic frequency variation feature. Taking the fundamental wave frequency of 50Hz as an example, the weight of the interharmonic amplitude variation feature (i.e., the weight of the enhanced oscillation precursor feature) is adjusted through the attention enhancement layer.

[0131] It should be noted that the weighting of the interharmonic amplitude variation characteristics is a commonly used technique in this field. For example, if there are 10 frequency components, each frequency component is automatically set to a weight of 1 / 10=0.1. However, after an abnormal frequency appears, in order to improve its identification ability and recognition sensitivity, its weight can be increased, for example, to 0.6. Then the remaining 9 frequencies will only occupy 0.4 of attention.

[0132] In this embodiment, the interharmonic amplitude variation feature is amplified by the attention enhancement layer (i.e., the weight is 0.6), making it higher than other features.

[0133] In the feature fusion layer, the temporal features and coupling correlation features of the fundamental wave, interharmonic waves, and harmonic waves (i.e., the ratio of harmonic amplitude to interharmonic amplitude, the ratio of fundamental wave amplitude to interharmonic amplitude, and the ratio of fundamental wave frequency fluctuation value to harmonic amplitude) are fused to generate comprehensive features.

[0134] For example, in this embodiment, the temporal characteristics of the 50Hz fundamental amplitude and the coupled correlation characteristics of the 125Hz interharmonic amplitude ratio are fused, time-aligned and normalized, and vector splicing is performed to generate comprehensive features.

[0135] The composite features are detected using an anomaly detection algorithm (preferably the Isolation Forest algorithm in this embodiment). For example, the number of trees in the Isolation Forest is set to 100, the subsampling size is 256, and an anomaly threshold (e.g., 0.8) is set using the percentile method on the training set (i.e., the anomaly scores of all samples in the training set are sorted in descending order, and the score corresponding to the 95th percentile is taken as the threshold). When running online, if the anomaly score of the composite feature in the Isolation Forest exceeds this threshold, it is determined to be a feature anomaly. Furthermore, the anomaly type is marked according to the magnitude of the deviation of the feature component from the normal range, and the anomaly type includes:

[0136] Abnormal fundamental frequency fluctuation (fundamental frequency amplitude change rate > 5% and frequency deviation > 0.1Hz). Abnormal harmonic exceedance (amplitude of a certain harmonic exceeds 3% of the fundamental frequency amplitude). Abnormal interharmonic abrupt change (interharmonic energy increases by more than 200% within 10ms). For example, if the 125Hz interharmonic amplitude exceeds the normal range by 15% and the fundamental frequency fluctuation exceeds ±0.5Hz, the abnormality type is marked as "abnormal interharmonic abrupt change and abnormal fundamental frequency fluctuation," and an abnormal feature set containing the time of occurrence, abnormality type, and corresponding key feature values ​​is output.

[0137] It's important to note that in Isolation Forest, the core of outlier score calculation is a normalized exponential function. The basic idea is that the fewer splits (shorter paths) a sample requires to be randomly isolated, the more likely it is to be an outlier, and the closer its score will be to 1. Typically, the average path length is calculated first, followed by normalization, and finally, the data value is obtained. A common value of 0.5 is usually set, and the calculation result is compared to this value to complete the outlier calculation.

[0138] like Figure 5 As shown, a broadband oscillation risk assessment model is constructed based on an impedance-mode fusion model. The model parameters are optimized using a pre-defined operating condition parameter mapping table. Specifically, the model has two adjustable parameters: an impedance correction coefficient. With modal damping weight For example, the current operating conditions of the power system are: a renewable energy penetration rate of 35% and a flexible DC capacity of 1000MW. Based on the pre-generated operating condition parameter mapping table, the above adjustable parameters are adjusted as follows: =0.92, =0.7.

[0139] It should be noted that the operating condition parameter mapping table is specifically a parameter configuration list or configuration table. For example, for a laptop, the table constructed from indicators such as CPU processing speed, memory size, and hard disk size is the operating condition parameter mapping table in this invention.

[0140] This adjustment allows the model to appropriately reduce the equivalent impedance estimate when the proportion of new energy sources is high, while giving higher attention weight to the modal damping ratio to reflect the oscillation sensitivity under weak grid conditions.

[0141] The abnormal feature set from the above steps is then input into the broadband oscillation risk assessment model with adjusted parameters. Simultaneously, the power system model information is read to calculate the sub-risk index for each abnormal feature in the abnormal feature set. The power system model information includes the topological impedance Zsys = 10 Ω (i.e., the equivalent impedance at the corresponding oscillation characteristic frequency) and the withstand threshold of key equipment (the amplitude of each harmonic must not exceed 5% of the rated voltage). The broadband oscillation risk assessment model calculates each abnormal feature individually, as follows:

[0142] The ratio of the characteristic amplitude to the impedance margin at the corresponding frequency is calculated, multiplied by the modal damping enhancement factor, and then normalized by sigmoid. The comprehensive calculation yields a model risk index of 0.58 at the current moment. An index greater than 0.5 indicates that the model has entered a medium-risk zone and requires enhanced monitoring.

[0143] A risk index sequence is constructed using a model risk index. For example, the model risk index of 120 sampling points (each sampling interval is 30 seconds) in a historical sampling step is used to construct a risk index sequence, which is then input into a pre-trained two-layer Long Short-Term Memory network (preferably using an LSTM model with 64 hidden units in this embodiment). The LSTM outputs the predicted risk index value for each sampling step within the next 10 minutes. The risk development rate is calculated based on the predicted risk index value, and the broadband oscillation risk is judged to be rising, stable, or falling based on the risk development rate. For example, if the current model risk index is 0.58, and it is predicted that the model risk index will rise to 0.65 after 10 minutes, an increase of 0.07 from the current value, the risk development rate is 0.007 / min. If this risk development rate is greater than a preset trend judgment threshold, and the risk index continues to increase within the prediction time range, then the broadband oscillation risk is judged to be rising.

[0144] For example: if the risk development rate over three consecutive cycles A value greater than 0.005 indicates an upward trend. In this case, 0.007 > 0.005, and the predicted value continues to increase unilaterally within the next 10 minutes. Therefore, the model-driven risk assessment result is judged to indicate an upward trend in oscillation risk, and this result triggers a yellow warning, prompting dispatchers to take pre-control measures (such as reducing the output of some new energy sources or adjusting the outer loop control parameters of the flexible DC transmission) within the next 8-10 minutes.

[0145] like Figure 6 As shown, at the current moment, the data risk index obtained through isolated forest detection ( =0.6) and the model risk index calculated by the broadband oscillation risk assessment model ( =0.58), and to balance the reliability of the above results, weights are assigned to them respectively (data-driven weights). =0.55 and model-driven weights =0.45), and the above weight values ​​can be flexibly adjusted. If the coverage of the field monitoring device is high, the weight of data-driven can be adjusted. For example, if the coverage of the line monitoring interval reaches 60%, the weight of data-driven can be adjusted to 0.6.

[0146] At this point, substituting the above parameters into the comprehensive risk assessment result expression for calculation, the assessment result is obtained. The calculation process is as follows:

[0147] =0.6×0.55+0.58×0.45=0.591;

[0148] The judgment is made based on the preset risk classification standards: no risk (first risk level), low risk (second risk level), medium risk (third risk level), and high risk (fourth risk level), combined with the preset thresholds of first threshold (0.2), second threshold (0.4) and third threshold (0.7).

[0149] For example, no risk (<0.2), low risk (0.2-0.4), medium risk (0.4-0.7), and high risk (≥0.7), according to... =0.591 indicates that it falls within the medium-risk range, therefore the current broadband oscillation risk level is judged to be medium risk.

[0150] According to the broadband oscillation risk level analysis, the 125Hz interharmonic amplitude exceeds the equipment's tolerance threshold most frequently (e.g., 18%), and the fundamental frequency fluctuation exceeds 0.08Hz / s, far exceeding the normal fluctuation range (less than 0.02Hz / s). Therefore, the core influencing factors are identified and output as: excessive 125Hz interharmonic amplitude and excessive fundamental frequency fluctuation. A comprehensive risk assessment result, the broadband oscillation risk level, and the core influencing factors are obtained and pushed to the external dispatch and early warning interface.

[0151] It should be noted that the oscillation characteristic quantities are based on the rate of change of the fundamental frequency under undisturbed or normal operation of the power grid and the typical interharmonic content threshold. Frequencies and abnormal characteristics whose interharmonic content exceeds the threshold are listed along with the oscillation risk level. Moreover, these thresholds can be preset according to the actual situation of the project or adjusted flexibly.

[0152] Differentiated early warning information is generated based on the risk level of broadband oscillation. For example, no risk outputs normal operation prompt information, low risk outputs attention prompt information, medium risk outputs early warning prompt information and provides preliminary prevention and control suggestions (i.e., the first prevention and control plan), and high risk outputs emergency early warning prompt information and priority prevention and control plan (i.e., the second prevention and control plan).

[0153] For example, according to The system automatically generates a warning message based on the medium risk level of 0.591. The message reads: "The current system has a medium risk of broadband oscillation. The core influencing factors are excessive amplitude of 125Hz interharmonic and excessive fluctuation of fundamental frequency. It is recommended to adjust the damping parameters of the wind power converter."

[0154] This information is pushed to the human-machine interface of the dispatch center and recorded in the early warning log. For example... Figure 7 As shown, operators follow the prompts to implement control measures (adjusting the damping parameters of the wind power converter). After the adjustment is completed, they continue to collect three-phase voltage and three-phase current signals at a high-frequency sampling rate to obtain wideband measurement data. The new data is then synchronously fed back to the following stages:

[0155] 1. In the data-driven risk assessment step (i.e. step S102), the feature weights are dynamically adjusted based on the amplitude changes of the remaining interharmonic components in the broadband measurement data. Specifically, the attention weight of the 125Hz interharmonic feature is adjusted from the default value of 0.50 to 0.65 to enhance the monitoring sensitivity of this key frequency band.

[0156] 2. The model-driven risk assessment step (i.e., step S103) involves online identification of the system's equivalent impedance and damping characteristics based on broadband measurement data, and adjusting the impedance correction coefficient from the original value of 0.92 to 0.95 to more accurately reflect the system impedance characteristics after the implementation of control measures.

[0157] After the implementation of prevention and control measures, continuous monitoring is conducted. For example, one hour later, the abnormal feature set is recalculated and input into the adjusted broadband oscillation risk assessment model, resulting in a new risk index of 0.32. According to the grading standard (0.2-0.4 is low risk), when the current risk level drops to low risk, the system outputs a "pay attention prompt" instead of an early warning message. This completes the closed-loop optimization from "anomaly detection → risk warning → measure implementation → parameter feedback → risk downgrading".

[0158] Furthermore, optimizing the feature extraction process of the initial broadband measurement data does not involve changing the collected initial broadband measurement data itself. Instead, it utilizes the feedback broadband measurement data after generating hierarchical early warning information to adjust the weight coefficients of the attention enhancement layer in the temporal feature extraction network. This enhances the sensitivity of identifying the target oscillation frequency component when performing feature extraction on similar initial broadband measurement data in the future.

[0159] It should be noted that, in conjunction with the closed-loop process, the risk threshold level can be adjusted appropriately. For example, if warnings and controls for various risks occur frequently, the risk classification threshold or data-driven weight can be appropriately increased; if oscillation alarms occur less frequently, the anomaly judgment threshold or warning threshold can be appropriately lowered.

[0160] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0161] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0162] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0164] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A wideband oscillation risk early warning method, characterized in that, include: The initial broadband measurement data of the preset monitoring nodes in the power system are preprocessed to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency. Temporal and coupled features of the initial broadband measurement data are extracted from the broadband analysis dataset, and the temporal and coupled features are fused to obtain comprehensive features. Anomaly detection algorithms are used to identify anomalies in the comprehensive features to obtain an anomaly feature set. A broadband oscillation risk assessment model was constructed and optimized. An abnormal feature set was input into the optimized broadband oscillation risk assessment model, and risk quantification and oscillation development trend extrapolation were performed in combination with power system model information to obtain model-driven risk assessment results. An adaptive weighted fusion algorithm is used to fuse the abnormal feature set and the model-driven risk assessment results to obtain a comprehensive risk assessment result, and the broadband oscillation risk level is determined by combining it with the preset risk grading standard. Based on the broadband oscillation risk level, a graded early warning information is generated, and broadband measurement data of the power system after the graded early warning information is generated is collected. The feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model are optimized.

2. The method of claim 1, wherein the method further comprises: The process of preprocessing the initial broadband measurement data of the preset monitoring nodes in the power system to obtain the broadband analysis dataset includes the following steps: The amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves in the monitoring nodes are collected using a preset sampling frequency and wide frequency range. Based on preset oscillation alarm values ​​and data processing algorithms, the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonic waves are subjected to anomaly removal, noise suppression, redundancy compression, and standardization to obtain a broadband analysis dataset. The data processing algorithms include a three-standard-deviation algorithm, a wavelet packet denoising algorithm, a redundancy compression algorithm, and a Z-score standardization algorithm.

3. The method of claim 1, wherein the method further comprises: The process involves extracting time-series and coupled-correlation features from the initial broadband measurement data of the broadband analysis dataset, fusing these features to obtain comprehensive features, and then using an anomaly detection algorithm to identify anomalies in the comprehensive features, resulting in an anomaly feature set including: The broadband analysis dataset is input into a preset temporal feature extraction network, and the feature extraction layer is used to extract the temporal features corresponding to the fundamental, interharmonic, and harmonic frequencies in the initial broadband measurement data; wherein, the temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer; The weighting coefficients of the interharmonic time series features are adjusted using an attention enhancement layer, and the ratio of the fundamental frequency to the interharmonic frequency is calculated to obtain the coupling correlation features. Using a feature fusion layer, the time-series features and coupled correlation features corresponding to the fundamental wave, interharmonic wave, and harmonic wave are time-aligned, normalized, and vector-concatenated to obtain comprehensive features; An anomaly detection algorithm using the isolated forest algorithm is used to detect comprehensive features, and combined with a preset anomaly judgment threshold, the detected comprehensive features are labeled to obtain anomaly types; wherein, the anomaly types include at least one of fundamental frequency fluctuation anomaly, harmonic exceedance anomaly, or interharmonic abrupt change anomaly. An anomaly feature set is constructed by combining the anomaly type, the time of anomaly occurrence, and the corresponding key feature values.

4. The method of claim 1, wherein the method further comprises: The process of constructing and optimizing a broadband oscillation risk assessment model involves inputting anomaly feature sets into the optimized model and combining this with power system model information to quantify risk and extrapolate oscillation development trends, resulting in a model-driven risk assessment. This includes the following steps: A broadband oscillation risk assessment model is constructed based on an impedance-modal fusion model, and the model parameters of the broadband oscillation risk assessment model are optimized by combining a preset operating condition parameter mapping table; wherein, the model parameters include impedance correction coefficients and modal damping weights; The abnormal feature set is input into the optimized broadband oscillation risk assessment model, and the sub-risk index of each abnormal feature in the abnormal feature set is calculated in combination with the power system model information; A risk index sequence is constructed based on the sub-risk index and input into a pre-trained long short-term memory network for prediction. Based on the prediction results, the development trend of broadband oscillation risk is deduced, and the model-driven risk assessment results are obtained.

5. The broadband oscillation risk early warning method according to claim 1, characterized in that, The process of fusing the abnormal feature set and model-driven risk assessment results using an adaptive weighted fusion algorithm to obtain a comprehensive risk assessment result, and determining the broadband oscillation risk level by combining it with a preset risk grading standard, includes the following steps: The model risk index is determined based on the model-driven risk assessment results. The data risk index is determined based on the anomaly type, the time of anomaly occurrence, and the corresponding key feature values ​​in the anomaly feature set. The data-driven weight and model-driven weight are determined using the data risk index and the model risk index. Based on data-driven weights and model-driven weights, the data risk index and the model risk index are weighted and fused to obtain a comprehensive risk assessment result. The comprehensive risk assessment results are compared with the preset risk classification thresholds to determine the broadband oscillation risk level.

6. The broadband oscillation risk early warning method according to claim 5, characterized in that, The step of comparing the comprehensive risk assessment results with a preset risk grading threshold to determine the broadband oscillation risk level includes: The preset risk classification thresholds include a first threshold, a second threshold, and a third threshold; When the comprehensive risk assessment result is less than the first threshold, it is determined to be at the first risk level; When the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined to be the second risk level; When the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined to be the third risk level; When the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined to be the fourth risk level.

7. The broadband oscillation risk early warning method according to claim 6, characterized in that, The process of generating graded early warning information based on broadband oscillation risk levels, collecting broadband measurement data of the power system after the generation of graded early warning information, and optimizing the time-series feature extraction network and broadband oscillation risk assessment model using broadband measurement data includes: Based on the broadband oscillation risk level, corresponding graded early warning information is generated. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention reminder information is generated; when the risk level is the third risk level, early warning reminder information and the first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning reminder information and the second prevention and control plan are generated. Broadband measurement data collected after the generation of graded early warning information, the implementation of the first prevention and control plan, or the implementation of the second prevention and control plan are used to optimize the weight coefficients of the attention enhancement layer in the time-series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

8. The broadband oscillation risk early warning method according to claim 5, characterized in that, The expression for the comprehensive risk assessment result is as follows: ； In the formula, Based on the comprehensive risk assessment results, For data risk index, For model risk index, Data-driven weights Use the model to drive the weights.

9. A broadband oscillation risk early warning system, characterized in that, include: A broadband full data and preprocessing module is used to preprocess the initial broadband measurement data of the preset monitoring nodes in the power system to obtain a broadband analysis dataset; wherein, the initial broadband measurement data includes fundamental frequency, interharmonic frequency and harmonic frequency. The data-driven feature mining and risk assessment module is used to extract the temporal and coupled features of the initial broadband measurement data from the broadband analysis dataset, and to fuse the temporal and coupled features to obtain comprehensive features; anomaly detection algorithms are used to identify anomalies in the comprehensive features to obtain an anomaly feature set. The model-driven risk assessment module is used to construct and optimize a broadband oscillation risk assessment model. It inputs the abnormal feature set into the optimized broadband oscillation risk assessment model and combines it with power system model information to quantify risk and extrapolate oscillation development trends, thereby obtaining model-driven risk assessment results. The risk assessment module for data model fusion is used to fuse the abnormal feature set and model-driven risk assessment results using an adaptive weighted fusion algorithm to obtain a comprehensive risk assessment result, and to determine the broadband oscillation risk level in combination with the preset risk grading standard. The early warning information output and closed-loop optimization module is used to generate graded early warning information based on the broadband oscillation risk level, and to collect broadband measurement data of the power system after the generation of graded early warning information, and to optimize the feature extraction process of the initial broadband measurement data and the broadband oscillation risk assessment model.

10. A broadband oscillation risk early warning system according to claim 9, characterized in that, The broadband full data and preprocessing module includes: a data acquisition submodule and a data preprocessing submodule, wherein... The data acquisition submodule is used to acquire amplitude, frequency, and phase data of the fundamental, interharmonic, and harmonic waves in the monitoring nodes using a preset sampling frequency and wide frequency range; The data preprocessing submodule is used to perform anomaly removal, noise suppression, redundancy compression, and standardization on the amplitude, frequency, and phase data of the fundamental wave, interharmonic waves, and harmonic waves based on preset oscillation alarm values ​​and data processing algorithms, to obtain a broadband analysis dataset. The data processing algorithms include the three-standard-deviation algorithm, wavelet packet denoising algorithm, redundancy compression algorithm, and Z-score normalization algorithm.

11. A broadband oscillation risk early warning system according to claim 9, characterized in that, The data-driven feature mining and risk assessment module includes: a feature extraction submodule, an attention enhancement submodule, a feature fusion submodule, an anomaly detection submodule, and an anomaly feature set construction submodule, wherein... The feature extraction submodule is used to input the broadband analysis dataset into a preset temporal feature extraction network, and use the feature extraction layer to extract the temporal features corresponding to the fundamental, interharmonic, and harmonic frequencies in the initial broadband measurement data; wherein, the temporal feature extraction network includes a feature extraction layer, an attention enhancement layer, and a feature fusion layer; The attention enhancement submodule is used to adjust the weight coefficients of the interharmonic timing features using the attention enhancement layer and calculate the ratio of the fundamental frequency to the interharmonic frequency to obtain the coupling correlation features. The feature fusion submodule is used to perform time alignment, normalization, and vector concatenation on the time-series features and coupled correlation features corresponding to the fundamental wave, interharmonic wave, and harmonic wave using the feature fusion layer to obtain comprehensive features. The anomaly detection submodule is used to detect the comprehensive features using the isolated forest algorithm and, in combination with a preset anomaly judgment threshold, to mark the detected comprehensive features to obtain the anomaly type; wherein, the anomaly type includes at least one of fundamental wave fluctuation anomaly, harmonic exceedance anomaly, or interharmonic abrupt change anomaly. The anomaly feature set construction submodule is used to construct an anomaly feature set by combining the anomaly type, the time of anomaly occurrence, and the corresponding key feature values.

12. A broadband oscillation risk early warning system according to claim 9, characterized in that, The model-driven risk assessment module includes: a model building and optimization submodule, a risk index calculation submodule, and a trend extrapolation submodule. The model construction and optimization submodule is used to construct a broadband oscillation risk assessment model based on the impedance-modal fusion model, and optimize the model parameters of the broadband oscillation risk assessment model by combining a preset operating condition parameter mapping table; wherein, the model parameters include impedance correction coefficients and modal damping weights; The risk index calculation submodule is used to input the abnormal feature set into the optimized broadband oscillation risk assessment model, and calculate the sub-risk index of each abnormal feature in the abnormal feature set in combination with the power system model information; The trend extrapolation submodule is used to construct a risk index sequence based on the sub-risk index and input it into a pre-trained long short-term memory network for prediction. Based on the prediction results, the development trend of broadband oscillation risk is extrapolated to obtain model-driven risk assessment results.

13. A broadband oscillation risk early warning system according to claim 9, characterized in that, The risk assessment module for data model fusion includes a weight determination submodule, a weighted fusion submodule, and a risk level determination submodule, wherein... The weight determination submodule is used to determine the model risk index based on the model-driven risk assessment results, determine the data risk index based on the anomaly type, anomaly occurrence time and corresponding key feature values ​​in the anomaly feature set, and determine the data-driven weights and model-driven weights using the data risk index and model risk index. The weighted fusion submodule is used to perform weighted fusion of data risk index and model risk index based on data-driven weights and model-driven weights to obtain a comprehensive risk assessment result. The risk level determination submodule is used to compare the comprehensive risk assessment results with the preset risk classification thresholds to determine the broadband oscillation risk level.

14. A broadband oscillation risk early warning system according to claim 13, characterized in that, The risk level determination submodule includes a grading threshold submodule and a level judgment submodule, wherein, The risk classification threshold submodule is used to preset risk classification thresholds, including a first threshold, a second threshold, and a third threshold. The risk level determination submodule is used to determine the risk level as the first risk level when the comprehensive risk assessment result is less than the first threshold. When the comprehensive risk assessment result is greater than or equal to the first threshold and less than the second threshold, it is determined to be the second risk level; When the comprehensive risk assessment result is greater than or equal to the second threshold and less than the third threshold, it is determined to be the third risk level; When the comprehensive risk assessment result is greater than or equal to the third threshold, it is determined to be the fourth risk level.

15. A broadband oscillation risk early warning system according to claim 14, characterized in that, The early warning information output and closed-loop optimization module includes an early warning information generation submodule and a feedback data acquisition and optimization submodule, wherein... The early warning information generation submodule is used to generate corresponding graded early warning information based on the broadband oscillation risk level. When the broadband oscillation risk level is the first risk level, normal operation information is generated; when the risk level is the second risk level, attention prompt information is generated; when the risk level is the third risk level, early warning prompt information and the first prevention and control plan are generated; when the risk level is the fourth risk level, emergency early warning prompt information and the second prevention and control plan are generated. The feedback data acquisition and optimization submodule is used to collect broadband measurement data after the generation of graded early warning information, the execution of the first prevention and control plan, or the execution of the second prevention and control plan, and to optimize the weight coefficients of the attention enhancement layer in the time series feature extraction network and the impedance correction coefficients of the broadband oscillation risk assessment model.

16. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the broadband oscillation risk warning method according to any one of claims 1 to 8.

Images