Power grid oscillation parameter detection method and system fusing decomposition and dynamic mechanism network

Abstract

The application discloses a power grid oscillation parameter detection method and system fusing decomposition and dynamic mechanism network, and the power grid oscillation parameter detection method comprises the following steps: acquiring discrete synchronous measurement signals of a power grid; decomposing the synchronous measurement signals into high-frequency noise components and multiple oscillation components through intrinsic set integration; selecting a target oscillation component from the multiple oscillation components obtained through intrinsic set integration; interpolating and refining the target oscillation component to obtain a refined component; inputting the refined component into a pre-trained dynamic mechanism network to obtain a reconstructed oscillation waveform; and separating power grid oscillation parameters from the reconstructed oscillation waveform. The application aims to efficiently and accurately solve the oscillation parameter estimation problem under the condition of strong noise and multi-modal superposition in a modern power system, reduce the interference of measurement noise and adjacent modes on the target oscillation component, jointly constrain the reconstructed waveform from three dimensions of time domain, derivative domain and frequency domain, and realize the interpretable end-to-end estimation of amplitude, damping, frequency and phase.

Description

Technical Field

[0001] This invention relates to the field of power system oscillation analysis and parameter identification and detection technology, specifically to a method and system for detecting power grid oscillation parameters by integrating decomposition and dynamic mechanism networks. Background Technology

[0002] With the widespread integration of renewable energy, power electronic converters, and low-inertia devices, the dynamic characteristics of modern power systems have become increasingly complex. Especially in systems with a high proportion of renewable energy, the lack of coordination between different devices and control links makes the power grid more prone to underdamped oscillations, modal coupling, and time-varying interactions over a wide frequency range. These oscillations not only affect power quality but may also weaken the system's stability margin. Therefore, accurately obtaining parameters such as amplitude, damping, frequency, and phase of oscillation signals is crucial for power grid dynamic monitoring, stability assessment, and control design. Existing oscillation parameter estimation methods mainly fall into three categories. The first category is based on environmental data, typically utilizing the random fluctuations of synchronous phasors under normal operating conditions to extract modal information, such as autoregressive models, autoregressive moving average models, and subspace identification methods. These methods are suitable for online monitoring, but their estimation accuracy tends to decrease under low signal-to-noise ratios, similar modal frequencies, or highly non-stationary system dynamics. The second category is based on oscillation decay data, primarily addressing transient responses after disturbances, such as Prony analysis, matrix bundle methods, characteristic system realization algorithms, and Hilbert transforms. These methods directly estimate oscillation parameters by fitting exponentially decaying sinusoidal signals, but rely on clear decay segments and high-quality measurement data. The third category is data-driven methods, such as sparse signal processing, optimization identification, convolutional neural networks, and hybrid deep models. These can handle more complex nonlinear signals, but typically lack physical interpretability and require a high quantity and quality of training data. Additionally, Fourier analysis, wavelet transform, Hilbert transform, subspace identification, and model fitting are also commonly used in multimodal oscillation analysis. They perform well when the dominant mode is clear and noise is moderate, but have limitations when facing multimodal superposition and strong noise. Overall, current power grid oscillation parameter estimation still faces two challenges: first, multimodal superposition and measurement noise can simultaneously mask the target oscillation components in both the time and frequency domains; second, while many data-driven models have strong fitting capabilities, they do not explicitly incorporate oscillation dynamics, making it difficult to ensure that the identification results have clear physical meaning. Therefore, achieving efficient and accurate broadband oscillation parameter estimation under strong noise and multimodal coupling conditions remains a key problem to be solved in modern power systems. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and system for detecting power grid oscillation parameters by integrating decomposition and dynamic mechanism networks, which addresses the above-mentioned problems in the prior art. This invention aims to efficiently and accurately solve the problem of oscillation parameter estimation under strong noise and multi-mode superposition conditions in modern power systems, reduce the interference of measurement noise and adjacent modes on the target oscillation components, and jointly constrain and reconstruct the waveform from three dimensions: time domain, derivative domain and frequency domain, so as to achieve interpretable end-to-end estimation of amplitude, damping, frequency and phase.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for detecting power grid oscillation parameters that integrates decomposition and dynamic mechanism networks includes the following steps: S1, acquire discrete synchronous measurement signals of the power grid; S2, the synchronous measurement signal is decomposed into high-frequency noise components and multiple oscillation components through intrinsic integration, and the target oscillation component is selected from the multiple oscillation components obtained by intrinsic integration decomposition. S3, interpolate and refine the target oscillation component to obtain the refined component; S4, input the refined components into the pre-trained dynamic mechanism network to obtain the reconstructed oscillation waveform; S5, extract the grid oscillation parameters from the reconstructed oscillation waveform.

[0005] Optionally, step S2 includes: S2.1, Multiple sets of zero-mean random disturbances are added to the synchronous measurement signal, and then multiple integrated disturbance signals are constructed through intrinsic integration decomposition. Each integrated disturbance signal includes multiple oscillation components with different dynamic scales. S2.2, average the oscillation components of each integrated disturbance signal at multiple different dynamic scales to obtain the average oscillation component corresponding to each integrated disturbance signal; S2.3 Calculate the dominant frequency and energy percentage of each average oscillation component, and select the average oscillation component whose dominant frequency falls within the preset target oscillation frequency band as the candidate target component. S2.4 Determine the target oscillation component based on the number of candidate target components: If the number of candidate target components is 1, then the candidate target component is directly used as the target oscillation component; if the number of candidate target components is greater than 1, then the candidate target component with the largest energy proportion is selected as the target oscillation component.

[0006] Optionally, the functional expression for interpolating and refining the target oscillation component to obtain the refined component in step S3 is: ; in, Let be the refined component at time t. The interpolation operator is a spline interpolation or cubic interpolation function. Let be the target oscillation component at time t.

[0007] Optionally, in step S4, the dynamic mechanism network is composed of a mechanistic encoder, a mechanistic block, an oscillating basis function layer, and a decoder connected sequentially. The mechanistic encoder is used to encode and map the refined input components into trainable parameters, including damping coefficients. γ Undamped natural frequency Sine basis functions Weighting coefficients and cosine basis functions Weighting coefficients The mechanism block is used based on the damping coefficient. γ Undamped natural frequency and the damped second-order oscillation relationship Calculate the damping angular frequency The oscillation basis function layer is used to determine the damping angular frequency. structure and Two orthogonal basis functions; the decoder is used to determine the trainable parameters output by the mechanistic encoder and the output of the oscillating basis function layer. and Two orthogonal basis functions generate the reconstructed oscillation waveform: ; in, Trainable parameters for dynamic mechanism-based networks The predicted reconstructed oscillation waveform at time t. These are the trainable parameters for dynamic mechanism networks. It is a natural constant. The damping coefficient is... sinusoidal basis functions The weighting coefficients, Cosine basis functions The weighting coefficients, The damping angular frequency, For time.

[0008] Optionally, in step S4, the loss function used by the dynamic mechanism network during training is the temporal reconstruction loss. First derivative consistency loss Second derivative consistency loss Frequency domain loss With envelope consistency loss The weighted sum of some or all of the losses is obtained.

[0009] Optionally, the temporal reconstruction loss The expression for the computation function is: ; The first derivative consistency loss The expression for the computation function is: ; The second derivative consistency loss The expression for the computation function is: ; The frequency domain loss The expression for the computation function is: ; The envelope consistency loss The expression for the computation function is: ; in, The loss is the temporal reconstruction loss, where N is the number of sampling points. For the reconstructed oscillation wave at the i-th sampling time, Let i be the target oscillation component at the i-th sampling time. The first derivative consistency loss, To reconstruct the first-order difference of the oscillation waveform at the i-th sampling time, Let the first-order difference of the target oscillation component be the i-th sampling time. The second derivative consistency loss, To reconstruct the second-order difference of the oscillation waveform at the i-th sampling time, Let be the second-order difference of the target oscillation component at the i-th sampling time. For frequency domain loss, To reconstruct the discrete Fourier transform of the oscillating waveform in a selected frequency band, To reconstruct the oscillation waveform, The discrete Fourier transform of the target oscillation component in a selected frequency band. For the target oscillation component, For envelope consistency loss, To reconstruct the instantaneous amplitude envelope of the oscillation waveform, The instantaneous amplitude envelope of the target oscillation component. This represents the i-th sampling time.

[0010] Optionally, in step S5, when separating the grid oscillation parameters from the reconstructed oscillation waveform, the separated grid oscillation parameters include amplitude. Damping coefficient Damping angular frequency With phase And there are , , It is the arctangent function in the four quadrants.

[0011] The present invention also provides a power grid oscillation parameter detection system based on a fusion decomposition and dynamic mechanism network, comprising a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the power grid oscillation parameter detection method based on the fusion decomposition and dynamic mechanism network.

[0012] The present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute, via a processor, the power grid oscillation parameter detection method of the fused decomposition and dynamic mechanism network.

[0013] The present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the power grid oscillation parameter detection method of the fused decomposition and dynamic mechanism network by a processor.

[0014] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: The grid oscillation parameter detection method of the present invention, which integrates decomposition and dynamic mechanism network, includes acquiring discrete synchronous measurement signals of the power grid; decomposing the synchronous measurement signals into high-frequency noise components and multiple oscillation components through intrinsic ensemble decomposition; selecting a target oscillation component from the multiple oscillation components obtained from the intrinsic ensemble decomposition; interpolating and refining the target oscillation component to obtain a refined component; inputting the refined component into a pre-trained dynamic mechanism network to obtain a reconstructed oscillation waveform; and separating the grid oscillation parameters from the reconstructed oscillation waveform. The present invention aims to efficiently and accurately solve the problem of oscillation parameter estimation under strong noise and multi-mode superposition conditions in modern power systems, reduce the interference of measurement noise and adjacent modes on the target oscillation component, and jointly constrain the reconstructed waveform from three dimensions: time domain, derivative domain, and frequency domain, to achieve interpretable end-to-end estimation of amplitude, damping, frequency, and phase. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention.

[0016] Figure 2 The decomposition results of the intrinsic integration decomposition of the dual-mode signal in the embodiment of the present invention are shown, where (a) is the constructed mixed input, (b) is the dominant noise-related component, (c) is the 40 Hz oscillation component, and (d) is the 1.5 Hz oscillation component.

[0017] Figure 3 This is a schematic diagram illustrating the loss function values ​​of the dynamic mechanism network under different hyperparameters in an embodiment of the present invention, where (a) is the learning rate. = 0.01 and The result is 0.3, and (b) shows the optimal composite loss weight setting. =0.05、 = 0.01.

[0018] Figure 4 This is a schematic diagram of the fitting results of the dynamic mechanism network under the 1.5Hz low-frequency oscillation condition in an embodiment of the present invention.

[0019] Figure 5 This is a schematic diagram of the fitting results of the dynamic mechanism network under the 58Hz supersynchronous oscillation case in an embodiment of the present invention.

[0020] Figure 6 This is a schematic diagram of the oscillation parameter estimation error under different noise levels in an embodiment of the present invention, where (a) is a 1.55Hz low-frequency oscillation and (b) is a 32Hz subsynchronous oscillation. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] like Figure 1 As shown, the power grid oscillation parameter detection method integrating decomposition and dynamic mechanism networks in this embodiment includes the following steps: S1, acquire discrete synchronous measurement signals of the power grid; S2, the synchronous measurement signal is decomposed into high-frequency noise components and multiple oscillation components through intrinsic integration decomposition (IED), and the target oscillation component is selected from the multiple oscillation components obtained by intrinsic integration decomposition; S3, interpolate and refine the target oscillation component to obtain the refined component; S4, input the refined components into the pre-trained dynamic mechanism network to obtain the reconstructed oscillation waveform; S5, extract the grid oscillation parameters from the reconstructed oscillation waveform.

[0023] This embodiment of the power grid oscillation parameter detection method integrating decomposition and dynamic mechanism network can include two stages: The first stage employs a front-end decomposition strategy based on intrinsic ensemble decomposition to separate the measured signal into components with different dynamic scales. Subsequently, the target oscillation mode is refined through interpolation. The refined target component is processed by a physically guided dynamic mechanism network to estimate the amplitude, damping, frequency, and phase.

[0024] In this embodiment, when obtaining the discrete synchronization measurement signal of the power grid in step S1, the synchronization measurement signal in the time domain is... The discrete form is denoted as where t is time, This is the sampling point number.

[0025] In this embodiment, step S2 includes: S2.1, Multiple sets of zero-mean random disturbances are added to the synchronous measurement signal, and then multiple integrated disturbance signals are constructed through intrinsic integration decomposition. Each integrated disturbance signal includes multiple oscillation components with different dynamic scales. S2.2, average the oscillation components of each integrated disturbance signal at multiple different dynamic scales to obtain the average oscillation component corresponding to each integrated disturbance signal; S2.3 Calculate the dominant frequency and energy percentage of each average oscillation component, and select the average oscillation component whose dominant frequency falls within the preset target oscillation frequency band as the candidate target component. S2.4 Determine the target oscillation component based on the number of candidate target components: If the number of candidate target components is 1, then the candidate target component is directly used as the target oscillation component; if the number of candidate target components is greater than 1, then the candidate target component with the largest energy proportion is selected as the target oscillation component.

[0026] In step S2.1, intrinsic integration decomposition is used to construct the integrated perturbation signal. In each implementation, a baseline term is constructed based on local extrema, linear interpolation, and convex combinations. With rotation terms The oscillation components at different dynamic scales are extracted, and the integrated implementation is averaged to suppress artificial perturbations and retain intrinsic components. Then, the component corresponding to the target oscillation is selected. Intrinsic integration decomposition can decompose a signal into a low-frequency baseline and a series of oscillating residuals, which can reduce interference caused by non-stationary trends, adjacent modes, and noise.

[0027] The functional expression for constructing integrated perturbations from intrinsic integration decomposition is: Further decomposed into baseline terms With rotation terms , can be represented as: ； in, Capture slower trends It includes higher frequency fluctuations relative to this trend.

[0028] Constructing baseline terms The function expression is: ; During the decomposition process, dynamic scaling constraints can be set based on the distance between adjacent extreme points to avoid over-decomposition or under-decomposition. Specifically, decomposition continues when the distance between extreme points in the current layer meets the preset scaling condition; otherwise, decomposition stops. Ultimately, the original signal can be represented as the superposition of multiple oscillating components and the residual baseline term, where the first... The final decomposition result of this implementation is: ; Where D is the preset maximum decomposition depth. For the d-th oscillation component, This represents the final residual baseline. The output expression for the final intrinsic ensemble decomposition is: .

[0029] The functional expression for interpolating and refining the target oscillation component to obtain the refined component in step S3 of this embodiment is as follows: ; in, Let be the refined component at time t. The interpolation operator is a spline interpolation or cubic interpolation function. Let be the target oscillation component at time t.

[0030] In step S4 of this embodiment, the dynamic mechanism network is composed of a mechanism encoder, a mechanism block, an oscillating basis function layer, and a decoder connected in sequence.

[0031] The mechanistic encoder is used to encode and map the refined components of the input into trainable parameters, including damping coefficients. γ Undamped natural frequency Sine basis functions Weighting coefficients and cosine basis functions Weighting coefficients The mechanism encoder maps the input signal to the coefficients of the ordinary differential equation. Since the control form is known in advance, the mechanism encoder does not search through a large equation library, but instead generates or refines a low-dimensional set of coefficients related to the damped oscillation model. The mechanism encoder can use the required encoder or network module as needed; for example, a one-dimensional convolutional neural network is used in this embodiment.

[0032] Mechanism blocks are used based on damping coefficients γ Undamped natural frequency and the damped second-order oscillation relationship Calculate the damping angular frequency The oscillation basis function layer is used to determine the damping angular frequency. structure and Two orthogonal basis functions.

[0033] The mechanistic block and the oscillating basis function layer are the core of the dynamic mechanism network. The basic idea is that the learned representation should evolve according to mechanistic laws, rather than arbitrary potential trajectories. In this embodiment, this role is undertaken by the oscillating block generated by damped second-order dynamics. Assume the refinement signal satisfies: ; For the underdamped case, the actual oscillation angular frequency is: ; The above equation defines the effective mechanistic block of the model. That is, the sinusoidal basis is not driven by physics-independent free frequency parameters, but by the frequencies inherent in the second-order oscillator itself. In this sense, the time evolution of the basis functions is determined by… The mechanism of encoding improves interpretability and helps maintain physically acceptable estimates.

[0034] The decoder is used to calculate the trainable parameters output by the mechanistic encoder and the output of the oscillating basis function layer. and Two orthogonal basis functions generate the reconstructed oscillation waveform: ; in, Trainable parameters for dynamic mechanism-based networks The predicted reconstructed oscillation waveform at time t. These are the trainable parameters for dynamic mechanism networks. It is a natural constant. The damping coefficient is... sinusoidal basis functions The weighting coefficients, Cosine basis functions The weighting coefficients, The damping angular frequency, Let the time be denoted by . To solve the above second-order dynamic equations, optimization is performed directly in a physically meaningful parameter space. Therefore, appropriate initialization is crucial for stable convergence. The initial frequency is given by . The dominant spectral peak is estimated, the initial damping value is empirically set, and the basis function coefficients B and C are obtained by least-squares fitting at temporary frequencies and damping values. These initial estimates place the optimization in a reasonable region of the parameter space and reduce sensitivity to noise and local minima.

[0035] To improve the robustness of parameter estimation under noise contamination and imperfect mode separation conditions, the loss function used by the dynamic mechanism network during training in step S4 is changed from the temporal reconstruction loss. First derivative consistency loss Second derivative consistency loss Frequency domain loss With envelope consistency loss The weighted sum is obtained as follows: ; in, For loss function, The weighting coefficients (greater than 0) can be set according to the initial magnitude of each loss term. Specific values ​​can be adjusted based on the noise level of the training samples, the sampling frequency, and the target oscillation frequency band. In this embodiment, the temporal reconstruction loss... The expression for the computation function is: ; This term constrains the direct consistency between the predicted mechanism response and the refined target oscillation in the original time domain.

[0036] The first derivative consistency loss The expression for the computation function is: ; This improves local slope consistency, helping the model capture short-term oscillation patterns.

[0037] The second derivative consistency loss The expression for the computation function is: ; By penalizing local curvature mismatches, this term helps estimate the sharpness and damping-related shape changes of the trajectory-maintaining oscillations.

[0038] The frequency domain loss The expression for the computation function is: ; This approach encourages the reconstruction of signals to retain the dominant spectral content of the target components and reduces the risk of obtaining visually plausible but frequency-characteristically incorrect time-domain fitting results.

[0039] The envelope consistency loss The expression for the computation function is: ; The motivation for this is that, for damped oscillations, the signal envelope approximately follows an exponential decay law. Taking the logarithm transforms the multiplicative amplitude decay into a form that more closely resembles a linear time-varying phenomenon, making it easier to penalize decay rate bias during training. Therefore, It provides a direct constraint on amplitude attenuation, which is particularly helpful in improving the damping coefficient. The identification of the oscillations is as follows. In summary, the five loss terms impose complementary constraints on the reconstructed oscillations. The time-domain, first-order difference, and second-order difference terms control waveform fidelity and local geometric consistency; the spectral term maintains the dominant frequency structure. The envelope term enhances sensitivity to exponential decay behavior. This combination is particularly advantageous when the decomposition stage leaves residual noise or slight modal contamination in the input signal. The loss is the temporal reconstruction loss, where N is the number of sampling points. For the reconstructed oscillation wave at the i-th sampling time, Let i be the target oscillation component at the i-th sampling time. The first derivative consistency loss, To reconstruct the first-order difference of the oscillation waveform at the i-th sampling time, Let the first-order difference of the target oscillation component be the i-th sampling time. The second derivative consistency loss, To reconstruct the second-order difference of the oscillation waveform at the i-th sampling time, Let be the second-order difference of the target oscillation component at the i-th sampling time. For frequency domain loss, To reconstruct the discrete Fourier transform of the oscillating waveform in a selected frequency band, To reconstruct the oscillation waveform, The discrete Fourier transform of the target oscillation component in a selected frequency band. For the target oscillation component, For envelope consistency loss, To reconstruct the instantaneous amplitude envelope of the oscillation waveform, The instantaneous amplitude envelope of the target oscillation component. Let be the i-th sampling time. The functional expression for the first-order difference is: ; The functional expression for the second-order difference is: ; The functional expression for the instantaneous amplitude envelope of the reconstructed oscillation waveform is: ; The functional expression for the instantaneous amplitude envelope of the target oscillation component is: ; in, This is a small constant used for numerical stability.

[0040] In step S5 of this embodiment, when separating the power grid oscillation parameters from the reconstructed oscillation waveform, the separated power grid oscillation parameters include amplitude. Damping coefficient Damping angular frequency With phase And there are , , It is the arctangent function in the four quadrants.

[0041] To evaluate the performance of intrinsic integration decomposition in this embodiment, a mixed input was constructed consisting of two damped components, 1.5Hz and 40Hz, superimposed with 20dB additive white noise. The sampling rate was set to 1200Hz, the signal duration was approximately 2s, and other parameters, including damping coefficient, amplitude, and phase, were randomly selected. Figure 2 As shown in (a) of the diagram. Figure 2 In the example, (b) represents the dominant noise-related component, which includes two components: IED (Intrinsic Integration Decomposition) mode 1 and ITD (Intrinsic Time-Scaled Decomposition) mode 1. The intrinsic integration decomposition in this embodiment can clearly separate the input into three physically meaningful components: the high-frequency noise-related component, ... Figure 2 The 40Hz oscillation component shown in (c) is similar to... Figure 2 The 1.5Hz oscillation component is shown in (d) in the figure. For example... Figure 2 As shown in (c), the 40Hz oscillation component includes three components: IED mode 2, low frequency, and ITD mode 2. Figure 2 The 1.5Hz oscillation component shown in (d) includes three components: IED mode 3, low frequency, and ITD mode 3. Compared with commonly used intrinsic time-scale decomposition methods, the components extracted by intrinsic ensemble decomposition are closer to the true underlying components and can better preserve the oscillation modes in the original mixed signal. Especially under the conditions of coexistence of low-frequency and high-frequency components and the presence of noise, intrinsic ensemble decomposition has stronger decomposition stability and higher fidelity in tracking the true modal response.

[0042] To test the parameter sensitivity of the dynamic mechanism network proposed in this embodiment, a grid search experiment was conducted based on a composite weighted loss function. Four key parameters were selected as examples, including the learning rate. ,initial Weight and weight ,like Figure 3 As shown, (a) is the learning rate. =0.01 and The result is 0.3, and (b) shows the optimal composite loss weight setting. =0.05、 The result is =0.01. For example... Figure 3 As shown in (a), within the 0.1-100Hz test range, when the learning rate... =0.01 and When the learning rate is 0.3, the dynamic mechanism network proposed in this embodiment achieves optimal performance, obtaining the minimum target value of 0.1916. In contrast, reducing the learning rate to 0.003 or using a smaller initial value... This value will cause the target value to continue to increase. For example... Figure 3 As shown in (b), the optimal composite loss weights are set to =0.05、 =0.01, corresponding to a minimum target value of approximately 0.260. The results indicate that moderate spectral constraints are beneficial, while excessive increases... This will reduce performance, which shows that the effectiveness of the composite weighted loss function requires an appropriate balance between the auxiliary constraints and the dominant time-domain fitting objective.

[0043] To verify the dynamic performance of the dynamic mechanism network proposed in this embodiment, tests were conducted for both low-frequency oscillation and synchronous oscillation scenarios. Figure 4 As shown, for the low-frequency case of 1.5 Hz under 20 dB noise, the method in this embodiment exhibits good time-domain reconstruction capability. The fitted signal is always close to the clean signal and is significantly less affected by random disturbances than the noisy observation. Especially in regions where the noisy signal exhibits significant fluctuations and local deviations, the proposed method can still accurately maintain the main oscillation shape, damping trend, and phase evolution of the clean waveform. This indicates that the dynamic mechanism network can reconstruct the target signal by incorporating damped oscillation dynamics into the fitting process. Therefore, the fitting result is not only smoother than the noisy signal but also more consistent with the physical characteristics of the real oscillation.

[0044] like Figure 5 As shown, the advantages of the method in this embodiment are more pronounced for the 58 Hz case under the same noise conditions, because high-frequency oscillations are generally more sensitive to noise contamination and local waveform distortion. Despite the greater challenge, the fitted signal remains highly consistent with the clean oscillation and significantly reduces the difference between noisy measurements and the true waveform. The reconstructed curve captures the rapid oscillation phase, amplitude changes, and damping behavior at t = 0 s with high fidelity, demonstrating that the proposed framework can effectively track the dynamics of rapid oscillations under noisy conditions. Therefore, the method in this embodiment achieves robust and accurate waveform recovery for both low-frequency and high-frequency oscillations.

[0045] In addition, to verify the performance of the method proposed in this embodiment under different noise levels and number of modes, this embodiment further tests two types of scenarios: low-frequency oscillation and subsynchronous oscillation. Figure 6 This diagram illustrates the estimation error of oscillation parameters under different noise levels in this embodiment, where (a) represents a 1.55Hz low-frequency oscillation and (b) represents a 32Hz subsynchronous oscillation. Figure 6As shown in (a), for the 1.55Hz low-frequency oscillation case, the intrinsic ensemble decomposition-dynamic mechanism network framework proposed in this embodiment exhibits good robustness under different signal-to-noise ratio (SNR) conditions: even at 20dB, the frequency and phase estimation errors are only 1.89×10⁻⁶, respectively. -3 and 1.27×10 -3 As the SNR increased from 20dB to 80dB, the amplitude error increased from 2.05×10⁻⁶. -2 Reduced to 9.30×10 -3 The phase error was further reduced to 10. -5 Quantity. In addition, such as... Figure 6 As shown in (b), for the 32Hz subsynchronous oscillation case, the framework proposed in this embodiment maintains reliable recognition performance at all tested SNR levels: the frequency estimation error is stable at approximately 2.2 × 10⁻⁶. -3 The damping ratio error decreased from 4.88 × 10 at 20 dB. -2 The value dropped to 2.86 × 10⁻⁶ dB at 80 dB. -2 The phase error is 1.50 × 10 -3 Decreased to below 4.00 × 10 -4 The above results verify that the intrinsic ensemble decomposition-dynamic mechanism network framework proposed in this embodiment possesses good noise resistance and stable parameter identification performance for both low-frequency and high-frequency oscillations. Furthermore, comparative experiments were conducted between the method of this embodiment and various existing methods such as the Hilbert transform method, matrix bundle method, ESPRIT method, FFT envelope method, and ARMA envelope method. The results are shown in Tables 1 and 2.

[0046]

[0047] As shown in Table 1, although traditional methods such as ESPRIT, FFT envelope, and matrix bundle can achieve reasonable frequency estimation, they cannot accurately recover the phase, with estimates of only about 0.92-0.96, far from the benchmark value of 2.496. The Hilbert method provides a phase estimate of 2.73, but its frequency and amplitude errors are still relatively large. Furthermore, matrix bundle, FFT envelope, and ARMA envelope significantly underestimate the amplitude, while ARMA envelope also produces a severely biased damping ratio estimate. Among all the compared methods, the proposed intrinsic ensemble decomposition-dynamic mechanism network framework achieves the most accurate and balanced parameter estimation. These results demonstrate that, under noisy low-frequency conditions, the intrinsic ensemble decomposition-dynamic mechanism network proposed in this embodiment has a significant advantage in maintaining the overall physical consistency of oscillation parameters, rather than simply matching one or two individual indices.

[0048]

[0049] As shown in Table 2, the proposed intrinsic ensemble decomposition-dynamic mechanism network method once again demonstrates superior overall performance for this higher frequency case. The estimated values ​​of 55.92 Hz, 0.82, 1.75, and 0.794 consistently approach the benchmark values, indicating that the proposed method maintains high estimation accuracy even for fast oscillation dynamics. While several benchmark methods, such as Hilbert, matrix bundle, ESPRIT, and FFT envelope, also provide acceptable frequency estimates, they exhibit greater deviations on one or more other parameters. For example, the matrix bundle significantly underestimates the amplitude, yielding 0.41, while the benchmark value is 0.82. The FFT envelope and ARMA envelope produce significantly biased damping ratio estimates, especially the ARMA envelope, which yields 0.406. Overall, by integrating signal separation based on intrinsic ensemble decomposition with mechanism learning based on dynamic mechanism networks, the proposed intrinsic ensemble decomposition-dynamic mechanism network method in this embodiment achieves better global consistency across all estimation parameters, highlighting its robustness, interpretability, and practical applicability for complex oscillation analysis.

[0050] In summary, the power grid oscillation parameter detection method integrating decomposition and dynamic mechanism networks in this embodiment includes acquiring the power grid synchronization measurement signal. Synchronous measurement signals were decomposed using intrinsic ensemble decomposition. By applying an integrated Gaussian perturbation and introducing a dynamic time-scale separation constraint, components at different dynamic scales are recursively separated, extracting the target oscillation component from multimodal and noisy measurement data. For the target oscillation component Interpolation refinement is performed to obtain the refined components. The data is then input into a two-stage method framework consisting of an intrinsic ensemble decomposition module and a dynamic mechanism network. The first step involves applying an integrated Gaussian perturbation to the measurement signal through intrinsic ensemble decomposition and introducing dynamic time-scale separation constraints. This recursively separates components at different dynamic scales, extracting the target oscillation component from the multimodal and noisy measurement data. Then, interpolation and refinement are performed to obtain the refined components. The second step involves constructing a dynamic mechanism network. This network explicitly embeds damped oscillation dynamics into the model structure through mechanism encoding, sine-cosine orthogonal oscillation basis functions, mechanism blocks constructed from damped second-order differential equations, and an analytical decoder. Finally, a composite weighted loss function consisting of time-domain reconstruction loss, first / second-order derivative consistency loss, frequency-domain loss, and logarithmic envelope consistency loss is constructed to jointly constrain the amplitude. A Damping coefficient γ angular frequency ω With phase Interpretable end-to-end estimation. This embodiment proposes a novel two-stage oscillation parameter estimation framework by integrating intrinsic ensemble decomposition and dynamic mechanism network (VIMN). First, the proposed front-end decomposition based on intrinsic ensemble decomposition effectively separates multimodal and noisy measurements into meaningful components with different dynamic scales. Mixed-signal decomposition tests show that multimodal oscillation components can be extracted from contaminated noisy measurements. Subsequently, the proposed dynamic mechanism network embeds damped oscillation dynamics into the model structure and employs a composite weighted loss function to improve robustness. Parameter selection results in the 0.1–100 Hz range show clear optimal points, with a minimum objective value of approximately 0.2, confirming a stable and efficient optimization process. Finally, by integrating intrinsic ensemble decomposition and dynamic mechanism network, the proposed framework achieves interpretable end-to-end parameter estimation without extensive training. Overall, experiments at different frequencies, multimodal conditions, and noise levels demonstrate that the proposed method achieves accurate and robust oscillation parameter identification even under strong noise interference.

[0051] Those skilled in the art will understand that the technical solutions provided by this invention can take the form of a method, a system, or a computer program product. Therefore, this invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. For example, this invention can provide a grid oscillation parameter detection system for a fused decomposition and dynamic mechanism network, including an interconnected microprocessor and a memory, the microprocessor being programmed or configured to execute the grid oscillation parameter detection method for the fused decomposition and dynamic mechanism network. This invention can provide a computer-readable storage medium storing a computer program or instructions programmed or configured to execute the grid oscillation parameter detection method for the fused decomposition and dynamic mechanism network via a processor. This invention can provide a computer program product including a computer program or instructions programmed or configured to execute the grid oscillation parameter detection method for the fused decomposition and dynamic mechanism network via a processor. Furthermore, this invention can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting power grid oscillation parameters by integrating decomposition and dynamic mechanism networks, characterized in that, Includes the following steps: S1, acquire discrete synchronous measurement signals of the power grid; S2, the synchronous measurement signal is decomposed into high-frequency noise components and multiple oscillation components through intrinsic integration, and the target oscillation component is selected from the multiple oscillation components obtained by intrinsic integration decomposition. S3, interpolate and refine the target oscillation component to obtain the refined component; S4, input the refined components into the pre-trained dynamic mechanism network to obtain the reconstructed oscillation waveform; The dynamic mechanism network consists of a mechanism encoder, a mechanism block, an oscillatory basis function layer, and a decoder connected sequentially. The mechanism encoder is used to encode and map the refined input components into trainable parameters, including damping coefficients. γ Undamped natural frequency Sine basis functions Weighting coefficients and cosine basis functions Weighting coefficients The mechanism block is used based on the damping coefficient. γ Undamped natural frequency and the damped second-order oscillation relationship Calculate the damping angular frequency ; The oscillation basis function layer is used to determine the damping angular frequency. structure and Two orthogonal basis functions; the decoder is used to determine the trainable parameters output by the mechanistic encoder and the output of the oscillating basis function layer. and Two orthogonal basis functions generate a reconstructed oscillatory waveform; S5, Separate the power grid oscillation parameters from the reconstructed oscillation waveform; Step S2 includes: S2.1, Multiple sets of zero-mean random disturbances are added to the synchronous measurement signal, and then multiple integrated disturbance signals are constructed through intrinsic integration decomposition. Each integrated disturbance signal includes multiple oscillation components with different dynamic scales. S2.2, average the oscillation components of each integrated disturbance signal at multiple different dynamic scales to obtain the average oscillation component corresponding to each integrated disturbance signal; S2.3 Calculate the dominant frequency and energy percentage of each average oscillation component, and select the average oscillation component whose dominant frequency falls within the preset target oscillation frequency band as the candidate target component. S2.4 Determine the target oscillation component based on the number of candidate target components: If the number of candidate target components is 1, then the candidate target component is directly used as the target oscillation component; if the number of candidate target components is greater than 1, then the candidate target component with the largest energy proportion is selected as the target oscillation component.

2. The method for detecting power grid oscillation parameters based on the fusion decomposition and dynamic mechanism network according to claim 1, characterized in that, The functional expression for the refined component obtained by interpolating and refining the target oscillation component in step S3 is as follows: ； in, Let be the refined component at time t. The interpolation operator is a spline interpolation or cubic interpolation function. Let be the target oscillation component at time t.

3. The method for detecting power grid oscillation parameters based on the fusion decomposition and dynamic mechanism network according to claim 1, characterized in that, The decoder is based on the trainable parameters output by the mechanistic encoder and the output of the oscillating basis function layer. and The functional expression for generating the reconstructed oscillatory waveform using two orthogonal basis functions is as follows: ； in, Trainable parameters for dynamic mechanism-based networks The predicted reconstructed oscillation waveform at time t. These are the trainable parameters for dynamic mechanism networks. It is a natural constant. The damping coefficient is... sinusoidal basis functions The weighting coefficients, Cosine basis functions The weighting coefficients, The damping angular frequency, For time.

4. The method for detecting power grid oscillation parameters using a fusion decomposition and dynamic mechanism network according to claim 1, characterized in that, In step S4, the loss function used by the dynamic mechanism network during training is the temporal reconstruction loss. First derivative consistency loss Second derivative consistency loss Frequency domain loss With envelope consistency loss The weighted sum of some or all of the losses is obtained.

5. The method for detecting power grid oscillation parameters using a fusion decomposition and dynamic mechanism network according to claim 4, characterized in that, The temporal reconstruction loss The expression for the computation function is: ； The first derivative consistency loss The expression for the computation function is: ； The second derivative consistency loss The expression for the computation function is: ； The frequency domain loss The expression for the computation function is: ； The envelope consistency loss The expression for the computation function is: ； in, The loss is the temporal reconstruction loss, where N is the number of sampling points. For the reconstructed oscillation wave at the i-th sampling time, Let i be the target oscillation component at the i-th sampling time. The first derivative consistency loss, To reconstruct the first-order difference of the oscillation waveform at the i-th sampling time, Let the first-order difference of the target oscillation component be the i-th sampling time. The second derivative consistency loss, To reconstruct the second-order difference of the oscillation waveform at the i-th sampling time, Let be the second-order difference of the target oscillation component at the i-th sampling time. For frequency domain loss, To reconstruct the discrete Fourier transform of the oscillating waveform in a selected frequency band, To reconstruct the oscillation waveform, The discrete Fourier transform of the target oscillation component in a selected frequency band. For the target oscillation component, For envelope consistency loss, To reconstruct the instantaneous amplitude envelope of the oscillation waveform, The instantaneous amplitude envelope of the target oscillation component. This represents the i-th sampling time.

6. The method for detecting power grid oscillation parameters using a fusion decomposition and dynamic mechanism network according to claim 1, characterized in that, In step S5, when separating the grid oscillation parameters from the reconstructed oscillation waveform, the separated grid oscillation parameters include amplitude. Damping coefficient Damping angular frequency With phase And there are , , It is the arctangent function in the four quadrants.

7. A power grid oscillation parameter detection system integrating decomposition and dynamic mechanism networks, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to execute the power grid oscillation parameter detection method of the fusion decomposition and dynamic mechanism network as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the power grid oscillation parameter detection method of the fusion decomposition and dynamic mechanism network as described in any one of claims 1 to 6.

9. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the power grid oscillation parameter detection method of the fusion decomposition and dynamic mechanism network as described in any one of claims 1 to 6.

Images