A broadband impedance analysis method based on network structure in DC distribution networks

Abstract

This invention discloses a broadband impedance analysis method based on the network structure of a DC distribution network, relating to the field of broadband impedance technology. The method includes: extracting joint frequency-time domain features based on time-frequency domain impedance characteristics using a lightweight convolutional neural network, and generating fault detection results using a multilayer perceptron classifier; synchronizing node detection results based on fault detection scores using a distributed network protocol, and employing a federated learning framework to weight and aggregate the extracted weight parameters of the joint frequency-time domain features to generate a grid health report containing a regional impedance deviation index. This invention synchronizes and aggregates data from each node using a distributed network protocol and a federated learning framework to generate a grid health report containing a regional impedance deviation index, a process that greatly enhances the collaborative working capability of grid systems in different geographical locations.

Description

Technical Field

[0001] This invention relates to the field of broadband impedance technology, and in particular to a broadband impedance analysis method based on the network structure in a DC distribution network. Background Technology

[0002] With the rapid development of new energy power generation, DC microgrids, and flexible DC transmission technologies, the complexity and dynamic characteristics of DC distribution networks are becoming increasingly significant. Their wideband impedance characteristics directly affect stability and power quality. Traditional impedance analysis methods are mainly based on frequency domain scanning or small-signal modeling. While these methods have a certain accuracy in offline analysis, they are difficult to adapt to the dynamic operating conditions of DC distribution networks.

[0003] In recent years, impedance analysis methods based on online monitoring have gradually become a research hotspot, such as using Fast Fourier Transform (FFT) for real-time impedance estimation. However, these methods typically focus only on a single frequency band or steady-state conditions, making it difficult to comprehensively capture impedance changes over a wide frequency range, especially transient issues such as high-frequency resonance and harmonic coupling. Furthermore, existing technologies largely rely on centralized computing architectures, which struggle to meet the real-time requirements of distributed DC distribution networks, and lack optimization mechanisms for multi-node collaborative analysis. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a broadband impedance analysis method based on the network structure in a DC distribution network, which solves the problem of insufficient accuracy in distributed collaborative analysis.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, this invention provides a broadband impedance analysis method based on the network structure of a DC distribution network. The method includes: real-time acquisition and preprocessing of voltage and current data, followed by time-frequency analysis using short-time Fourier transform to generate time-frequency domain impedance features; extraction of joint frequency-time domain features using a lightweight convolutional neural network based on the time-frequency domain impedance features, and generation of fault detection results using a multilayer perceptron classifier; data synchronization of node detection results using a distributed network protocol based on fault detection scores, and weighted aggregation of the extracted weight parameters of the joint frequency-time domain features using a federated learning framework to generate a grid health report including a regional impedance deviation index; feature space mapping and classification decision-making of the time-frequency domain impedance features and regional impedance deviation index using a support vector machine algorithm to generate a broadband impedance spectrum and hierarchical early warning information; and real-time adjustment of converter PWM control parameters using an adaptive control algorithm based on the broadband impedance spectrum and hierarchical early warning information, combined with the grid health report to generate equipment maintenance strategies.

[0008] As a preferred solution of the broadband impedance analysis method based on the network structure in the DC distribution network of the present invention, wherein: the preprocessing includes eliminating power frequency harmonics and random noise through a Gaussian filter and performing signal normalization processing.

[0009] As a preferred solution of the broadband impedance analysis method based on the network structure in the DC distribution network of the present invention, wherein: the time-frequency analysis is performed through the short-time Fourier transform to generate time-frequency domain impedance characteristics, and the specific steps are as follows.

[0010] Perform time-frequency decomposition on the preprocessed voltage and current data through the short-time Fourier transform to generate a time-frequency spectrum matrix;

[0011] Based on the time-frequency spectrum matrix, perform impedance characteristic conversion through the frequency-domain Ohm's law to generate time-frequency domain impedance characteristics.

[0012] As a preferred solution of the broadband impedance analysis method based on the network structure in the DC distribution network of the present invention, wherein: based on the time-frequency domain impedance characteristics, extract frequency-domain and time-domain joint characteristics through a lightweight convolutional neural network, and generate a fault detection result through a multi-layer perceptron classifier. The specific steps are as follows.

[0013] Based on the time-frequency domain impedance characteristic matrix, generate a multi-dimensional feature tensor with spatio-temporal correlation through sliding calculation with a three-dimensional convolution kernel;

[0014] Based on the multi-dimensional feature tensor, generate a frequency-domain and time-domain joint feature vector through a cross-channel attention fusion mechanism;

[0015] Based on the frequency-domain and time-domain joint feature vector, generate a fault detection result y through a multi-layer perceptron classifier;

[0016] Define a fault threshold A based on historical data;

[0017] When y < A, it is considered that the device is operating normally, and the current monitoring mode is maintained;

[0018] When y ≥ A, it is considered that the device is in an abnormal state, and the operation and maintenance personnel need to be notified immediately for repair.

[0019] As a preferred solution of the broadband impedance analysis method based on the network structure in the DC distribution network of the present invention, wherein: based on the fault detection result, synchronize node data through a distributed protocol, and use a federated learning framework to extract model parameters for the frequency-domain and time-domain joint characteristics, and generate a power grid health report including the regional impedance deviation index. The specific steps are as follows.

[0020] Through the distributed consistency protocol, perform time alignment and quality verification on the node detection data to generate a synchronized data set;

[0021] Based on the synchronous dataset, the neural network parameter update volume of each monitoring terminal is aggregated through a federated learning framework, and a weighted average algorithm is used to generate globally shared parameters.

[0022] Based on globally shared parameters, a power grid health report is generated through impedance deviation analysis and health score mapping.

[0023] As a preferred embodiment of the broadband impedance analysis method based on the network structure of a DC distribution network described in this invention, the method involves using a support vector machine algorithm to perform feature space mapping and classification decisions on time-frequency domain impedance characteristics and regional impedance deviation indices based on grid health reports, thereby generating a broadband impedance spectrum and hierarchical early warning information. The specific steps are as follows:

[0024] Based on the power grid health report, the time-frequency domain impedance characteristics and regional impedance deviation index are mapped to a high-dimensional space through kernel space mapping to generate a high-dimensional feature representation.

[0025] Based on high-dimensional feature representation, the optimal classification decision boundary is generated through a support vector machine optimization algorithm.

[0026] Based on the classification decision boundary, the spectrum is reconstructed through impedance feature clustering and abnormal region localization to generate a broadband impedance spectrum and hierarchical early warning information.

[0027] As a preferred embodiment of the broadband impedance analysis method based on the network structure of a DC distribution network described in this invention, the method involves adjusting the converter PWM control parameters in real time using an adaptive control algorithm based on the broadband impedance spectrum and hierarchical early warning information. The specific steps are as follows:

[0028] Based on broadband impedance maps and hierarchical early warning information, the characteristics of the dominant resonant frequency band are generated through Hilbert spectrum analysis.

[0029] Based on the characteristics of the dominant resonant frequency band, a PWM carrier frequency correction amount is generated through a PID controller.

[0030] Based on graded early warning information, a safe adjustment range for the PWM modulation ratio is generated through a dynamic limiting strategy.

[0031] As a preferred embodiment of the broadband impedance analysis method based on network structure in DC distribution networks described in this invention, the specific steps for generating equipment maintenance strategies in conjunction with power grid health reports are as follows:

[0032] Based on the PWM carrier frequency correction and the safe adjustment range of the PWM modulation ratio, the remaining service life assessment result of the DC support capacitor is generated by the lifetime prediction algorithm.

[0033] Based on the assessment results of remaining usage time, a preventive maintenance strategy is generated using a maintenance decision optimization method that incorporates current control parameters.

[0034] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the broadband impedance analysis method based on the network structure in a DC distribution network as described in the first aspect of the present invention.

[0035] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the broadband impedance analysis method based on the network structure in a DC distribution network as described in the first aspect of the present invention.

[0036] The beneficial effects of this invention are as follows: by synchronizing and aggregating data from each node through a distributed network protocol and a federated learning framework, a power grid health report including a regional impedance deviation index is generated. This process greatly enhances the collaborative working capability of power grid systems in different geographical locations. This method can not only accurately assess the overall health status of the power grid, but also dynamically generate a regional impedance deviation index, providing crucial information for subsequent maintenance decisions. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the broadband impedance analysis method based on the network structure in a DC distribution network in Example 1.

[0039] Figure 2 This is a diagram illustrating the time-frequency domain impedance characteristic generation process in Example 1.

[0040] Figure 3 This is a flowchart of the fault detection and decision-making process in Example 1.

[0041] Figure 4 This is a diagram generated for the adaptive control and maintenance strategy in Example 1. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Example 1, referring to Figures 1-4This embodiment provides a broadband impedance analysis method based on the network structure in a DC distribution network, including the following steps:

[0044] S1. Real-time acquisition of voltage and current data and preprocessing, while performing time-frequency analysis through short-time Fourier transform to generate time-frequency domain impedance characteristics;

[0045] Preprocessing includes eliminating power frequency harmonics and random noise using a Gaussian filter and performing signal normalization.

[0046] It should be noted that the bandwidth of the Gaussian filter transfer function is set to a specific proportion of the fundamental frequency according to the power frequency harmonic characteristics. Discrete Fourier transforms are performed on the original voltage and current signals to obtain their frequency domain representations. Frequency domain filtering is achieved by performing point-by-point complex multiplication with the filter function in the frequency domain. Subsequently, random noise suppression is performed. A fixed-length sliding window is used to calculate the energy ratio of the filtered signal to the original signal. When this ratio is lower than a preset threshold, it is identified as a noise point and corrected using linear interpolation of adjacent sampling points. Finally, signal normalization is performed. The amplitude is normalized by calculating the effective values ​​of voltage and current within a complete power frequency cycle and dividing the denoised signal by its corresponding effective value. The entire preprocessing process ensures that the output signal simultaneously meets the requirements of power frequency harmonic suppression, random noise elimination, and amplitude normalization.

[0047] The preprocessed voltage and current data are decomposed into time-frequency components using short-time Fourier transform to generate a time-frequency spectrum matrix.

[0048] It should be noted that the Hanning window function is used to window the preprocessed voltage and current signals, with the window length set to an integer multiple covering the power frequency cycle. The windowed signal is then slidably divided along the time axis with a fixed overlap rate. A Discrete Fourier Transform is performed on the signal segment within each window to obtain the spectral distribution of each time period. All windowed spectra of the voltage and current signals are arranged and combined in chronological order to form the voltage time-frequency spectrum matrix and the current time-frequency spectrum matrix, respectively. The row vectors of the time-frequency spectrum matrix represent the frequency component distribution, the column vectors represent the time variation process, and the matrix elements contain the complex spectral values ​​at the corresponding time and frequency points. The voltage time-frequency spectrum matrix and the current time-frequency spectrum matrix have the same time and frequency resolution to ensure the accuracy of subsequent impedance characteristic conversion, and the final output is the time-frequency spectrum matrix.

[0049] Based on the time-frequency spectrum matrix, impedance characteristics are transformed using Ohm's law in the frequency domain to generate time-frequency domain impedance features.

[0050] It should be noted that the complex spectral values ​​at corresponding time and frequency points are extracted from the time-frequency spectrum matrix. Element-by-element complex division is then performed according to Ohm's law in the frequency domain to calculate the complex impedance values ​​at each time and frequency point. The real part of the complex impedance value represents the resistance component, and the imaginary part represents the reactance component. The calculation results are rearranged and combined according to the time-frequency structure of the original time-frequency spectrum matrix to form the impedance time-frequency spectrum matrix. The row vectors of the impedance time-frequency spectrum matrix represent the impedance variation characteristics with frequency, and the column vectors represent the impedance variation process with time. Each element in the impedance time-frequency spectrum matrix contains the complex impedance value at the corresponding time and frequency point, fully characterizing the impedance characteristics of the measured object in the time-frequency domain. The final generated impedance time-frequency spectrum matrix is ​​the time-frequency domain impedance characteristic.

[0051] S2. Based on the impedance characteristics in the time and frequency domains, the joint features of the frequency and time domains are extracted by a lightweight convolutional neural network, and the fault detection results are generated by a multilayer perceptron classifier.

[0052] Based on the time-frequency domain impedance feature matrix, a multidimensional feature tensor with spatiotemporal correlation is generated by sliding calculation using a three-dimensional convolution kernel;

[0053] It should be noted that the size of the convolution kernel is defined in each of the three dimensions of the time-frequency domain impedance feature matrix, where the first dimension corresponds to the frequency component, the second dimension to the time series, and the third dimension to the spatial distribution characteristics. The three-dimensional convolution kernel is slid along the frequency, time, and spatial dimensions of the time-frequency domain impedance feature matrix, performing dot-multiplication and accumulation operations at each sliding position to calculate the spatiotemporal coupling feature response of the local time-frequency region. A fixed stride is maintained during the kernel's sliding process to ensure dimensionality matching of the output spatiotemporal coupling feature tensor. Multiple sets of three-dimensional convolution kernels with different parameters are used in parallel computation to extract the spatiotemporal correlation characteristics of different modes in the time-frequency domain impedance feature matrix. The output feature maps of each set of convolution kernels are stacked and combined according to the channel dimension to form a multidimensional feature tensor with spatiotemporal correlation.

[0054] Based on the multidimensional feature tensor, a frequency-time joint feature vector is generated through a cross-channel attention fusion mechanism;

[0055] It should be noted that the global average pooling value of the feature maps of each channel is calculated along the channel dimension of the multidimensional feature tensor, serving as the benchmark for channel importance scoring. The pooling values ​​of each channel are input into an attention network consisting of two fully connected layers to generate channel attention weight vectors. After normalization, the attention weight vectors are multiplied by the original multidimensional feature tensor using channel weighting to highlight the sensitivity features of key channels. The weighted multidimensional feature tensor is then subjected to global average pooling along the spatial dimension, preserving the energy distribution features of the frequency and temporal dimensions. The pooling result is flattened into a one-dimensional vector, forming a joint feature vector that integrates frequency and temporal characteristics.

[0056] Based on the frequency-domain and time-domain joint feature vector, the fault detection result is generated by a multi-layer perceptron classifier, and the formula is as follows:

[0057]

[0058] where y is the fault detection score, σ is the Sigmoid activation function, β k is the contribution weight of the k-th sub-band, K is the total number of frequency band blocks, F (k) is the k-th time-frequency feature block, W k is the fault-sensitive filter for the k-th frequency band, k is the frequency band index, and b is the control score benchmark;

[0059] It should be noted that first, the frequency-domain and time-domain joint feature vector F is divided into k sub-feature blocks F(k) according to the frequency band, and each feature block is element-wise multiplied by the fault-sensitive filter W k for the corresponding frequency band to obtain the weighted feature block. The fault-sensitive filter W k is obtained through training with historical fault data and can highlight the characteristic response of the fault-sensitive frequency band. Then, the L2 norm is calculated for each weighted feature block to quantify the significance intensity of the frequency band feature. Then, the norm values of each frequency band are multiplied by the pre-trained contribution weight β k and accumulated, and the bias term b is added to obtain the linear combination value. Finally, the linear combination value is non-linearly mapped to the interval (0,1) through the Sigmoid function, and the fault detection score y is output. The closer the score is to 1, the higher the fault probability, and the closer it is to 0, the more normal the operating state. The entire calculation process effectively integrates multi-band feature information and obtains an intuitive probabilistic fault score through non-linear transformation.

[0060] Based on historical data, the fault threshold A is defined;

[0061] When y < A, it is considered that the device is operating normally, and the current monitoring mode is maintained;

[0062] When y ≥ A, it is considered that the device is in an abnormal state, and the operation and maintenance personnel need to be notified immediately for repair.

[0063] It should be noted that when the fault detection score y is less than the threshold A, for example, if the fault detection score is 0.35 and the threshold A is set to 0.5, the amplitude fluctuations of the time-frequency domain impedance characteristic matrix in each frequency band are within the historical normal range, and the values ​​of each channel of the multidimensional feature tensor output by the lightweight convolutional neural network remain stable, without triggering any control command adjustments or alarm notifications. Therefore, the DC distribution network equipment is determined to be in normal operating condition, and the current monitoring mode is maintained. When the fault detection score y is greater than or equal to the threshold A, for example, if the fault detection score is 0.65 and the threshold A is set to 0.5, the abnormal handling process is immediately executed. At this time, the amplitude of the time-frequency domain impedance characteristic matrix suddenly increases in a specific frequency band, and the values ​​of the key channels of the multidimensional feature tensor output by the lightweight convolutional neural network fluctuate abnormally. Therefore, the DC distribution network equipment is determined to be in an abnormal operating state.

[0064] S3. Based on fault detection scoring, data synchronization of node detection results is achieved through distributed network protocol, and a federated learning framework is used to weight and aggregate the weight parameters of frequency domain-time domain joint feature extraction to generate a power grid health report including regional impedance deviation index.

[0065] By using a distributed consensus protocol, the node detection data is time-aligned and quality-verified to generate a synchronized dataset;

[0066] It should be noted that the distributed consensus algorithm based on the Byzantine fault tolerance mechanism timestamps the fault detection results reported by each monitoring node to ensure that all node data has a unified time base; the integrity of node data is verified by a Merkle tree structure, calculating the hash value of each data block and comparing it with the root hash stored in the master node to remove abnormal data that fails hash verification; sliding window mean filtering is performed on the data that passes the verification to eliminate time-series jitter caused by network latency; a data quality weight matrix based on signal-to-noise ratio evaluation is established to downweight the detection results of nodes with low signal-to-noise ratio; finally, the multi-node data that has undergone time alignment and quality weighting are integrated in time series to form a synchronized dataset with consistency and reliability.

[0067] Based on the synchronous dataset, the neural network parameter update volume of each monitoring terminal is aggregated through a federated learning framework, and a weighted average algorithm is used to generate globally shared parameters.

[0068] It should be noted that the parameter update amount of the lightweight convolutional neural network is calculated based on the local data in the synchronous dataset, and the update amount is encrypted and transmitted to the aggregation server. The aggregation server calculates the aggregation weight according to the data quality and historical accuracy of each terminal, and uses an anomaly detection method based on Euclidean distance to remove abnormal parameter updates that deviate from the population distribution. The parameter update amounts that pass the verification are weighted and summed according to their corresponding weights to calculate the global parameter update amount. The global parameter update amount is added to the global shared parameters of the previous round to generate the global shared parameters of the new round.

[0069] Based on globally shared parameters, a power grid health report is generated through impedance deviation analysis and health score mapping.

[0070] It should be noted that the lightweight convolutional neural network of each monitoring terminal is initialized with globally shared parameters, and the time-frequency domain impedance characteristics in the synchronous dataset are recalculated to obtain a standardized impedance characteristic distribution. By comparing the Euclidean distance between the impedance characteristics of each node and the global reference impedance, the regional impedance deviation index is calculated to quantify the impedance matching degree between the local power grid and the overall network. A health score mapping function based on Mahalanobis distance is established to project the multidimensional impedance characteristics onto the scalar health score space. The lower the score, the healthier the power grid is. The regional impedance deviation index and health score results are integrated to generate a structured power grid health report.

[0071] S4. Based on the power grid health report, the support vector machine algorithm is used to perform feature space mapping and classification decision on the time-frequency domain impedance characteristics and regional impedance deviation index to generate a wideband impedance spectrum and hierarchical early warning information.

[0072] Based on the power grid health report, the time-frequency domain impedance characteristics and regional impedance deviation index are mapped to a high-dimensional space through kernel space mapping to generate a high-dimensional feature representation.

[0073] It should be noted that the original feature vectors are generated by nonlinearly transforming the time-frequency domain impedance features and regional impedance deviation index in the power grid health report using a radial basis function kernel. These original feature vectors are then projected onto a regenerative kernel Hilbert space. For each element in the time-frequency domain impedance feature matrix, its distance to the center of a preset kernel function is calculated, and a high-dimensional representation is obtained through a Gaussian kernel transformation. The regional impedance deviation index undergoes a polynomial kernel transformation to enhance its separability in the high-dimensional space. The transformed impedance features and deviation index features are then concatenated and combined in the kernel space to form a unified high-dimensional feature representation. This high-dimensional feature representation retains the spatiotemporal correlation characteristics of the original time-frequency domain impedance features while amplifying the discriminative features of the regional impedance deviation through kernel tricks. The final generated high-dimensional feature representation exhibits stronger linear separability than the original features and can be directly used in support vector machine optimization algorithms to solve for the optimal classification decision boundary.

[0074] Based on high-dimensional feature representation, the optimal classification decision boundary is generated through a support vector machine optimization algorithm.

[0075] It should be noted that sample points in the high-dimensional feature space are mapped to the regenerating kernel Hilbert space using a kernel trick. Based on the principle of minimizing structural risk, a separating hyperplane with the maximum geometric margin is constructed. This optimization problem is formulated as a quadratic programming problem with slack variables, and its objective function simultaneously considers maximizing the margin and minimizing the classification error. The similarity matrix between sample points is calculated using a Gaussian kernel function, establishing the corresponding Lagrange dual problem, which is then iteratively solved using a sequential minimum optimization algorithm. In each iteration, the sample pair that most severely violates the KKT conditions is selected to update the Lagrange multipliers, while the regularization parameter is dynamically adjusted to balance model complexity and classification accuracy. The optimization process continues until all sample points satisfy the KKT conditions. At this point, the decision boundary is uniquely determined by support vectors with non-zero Lagrange multipliers. The sample points corresponding to these support vectors lie on the maximum margin boundary, and their high-dimensional feature representation contains key information for distinguishing different operating conditions. The resulting nonlinear decision boundary can effectively separate the normal operating condition feature cluster from various fault mode feature clusters, constructing a classification region with the maximum tolerance space in the kernel space. This classifier is robust to outliers in the feature space and can adapt to dynamic changes in the operating conditions of the DC distribution network.

[0076] Based on the classification decision boundary, the spectrum is reconstructed through impedance feature clustering and abnormal region localization to generate a broadband impedance spectrum and hierarchical early warning information.

[0077] It should be noted that in the high-dimensional feature space defined by the classification decision boundary, density clustering algorithm is used to automatically group impedance feature points and identify feature clusters with similar impedance characteristics. Based on the clustering results, the high-dimensional features are back-projected to the original time-frequency domain to reconstruct the impedance distribution pattern corresponding to each feature cluster, forming a broadband impedance spectrum reflecting the characteristics of different operating conditions. At the same time, the relative distance from each cluster center to the classification decision boundary is calculated to quantify the degree of anomaly, and the abnormal area is accurately located by combining the regional impedance deviation index. According to the anomaly degree threshold, the warning level is divided into three levels: Level 1 warning corresponds to significant abnormal clusters that directly cross the classification decision boundary, Level 2 warning corresponds to potential abnormal clusters near the boundary, and Level 3 warning targets the clusters to be observed that, although not exceeding the limit, have impedance characteristics that deviate from the typical distribution. The final output broadband impedance spectrum presents the impedance amplitude and phase characteristics of each frequency band in the form of a time-frequency matrix, and the graded warning information marks the abnormal frequency band range and suggested handling measures. Together, they constitute a complete power grid status assessment result.

[0078] S5. Based on wideband impedance spectrum and hierarchical early warning information, the converter PWM control parameters are adjusted in real time through adaptive control algorithm, and equipment maintenance strategies are generated in combination with power grid health report.

[0079] Based on broadband impedance maps and hierarchical early warning information, the characteristics of the dominant resonant frequency band are generated through Hilbert spectrum analysis.

[0080] It should be noted that for the frequency bands marked as warning states in the broadband impedance spectrum, the Hilbert-Huang transform method is used for time-frequency analysis. First, the impedance signal is decomposed into several intrinsic mode function components through empirical mode decomposition. Then, the instantaneous frequency and amplitude are calculated by performing a Hilbert transform on each component to construct the Hilbert energy spectrum. Frequency bands with energy concentration exceeding a preset threshold in the energy spectrum are identified as candidate resonant frequency bands. The quality factor and stability index of each candidate frequency band are calculated, and the dominant resonant frequency band that simultaneously meets the conditions of high Q value and low frequency fluctuation is selected. The final output characteristics of the dominant resonant frequency band include the center frequency, resonance intensity, quality factor, and attenuation coefficient.

[0081] Based on the characteristics of the dominant resonant frequency band, a PWM carrier frequency correction amount is generated through a PID controller.

[0082] It should be noted that the center frequency of the dominant resonant frequency band is compared with the current PWM carrier frequency, and the calculated frequency deviation is used as the input signal for the PID controller. The PID controller generates a carrier frequency adjustment based on the combined action of the proportional, integral, and derivative components. The proportional component responds to the real-time frequency deviation, the integral component eliminates steady-state error, and the derivative component suppresses frequency oscillation. After amplitude limiting, the controller output generates a PWM carrier frequency correction value, which is superimposed on the base carrier frequency to form a new carrier frequency setpoint. The amplitude and direction of the correction value are jointly determined by the intensity and stability indicators of the resonant frequency band characteristics, ensuring stable operation of the converter while suppressing resonance. Finally, the output PWM carrier frequency correction value directly acts on the converter control loop to achieve active avoidance of the resonant frequency point.

[0083] Based on graded early warning information, a safe adjustment range for the PWM modulation ratio is generated through a dynamic limiting strategy.

[0084] It should be noted that, based on the anomaly levels classified in the graded early warning information, and according to the limiting threshold boundaries, Level 1 early warning corresponds to the most stringent modulation ratio constraint range, Level 2 early warning adopts moderately relaxed boundary conditions, and Level 3 early warning maintains the standard operating limits; a constraint condition calculation model based on Lyapunov stability theory is established, and the maximum allowable upper limit of the modulation ratio is calculated by combining the current DC bus voltage fluctuation amplitude and AC side harmonic distortion rate; at the same time, considering the junction temperature change trend and heat dissipation conditions of power devices, the minimum lower limit of the modulation ratio is derived through a thermal stress algorithm; the upper and lower limit parameters are input into a hysteresis comparator for smoothing to eliminate boundary abrupt changes; the final generated PWM modulation ratio safe adjustment range ensures that the converter always operates in a safe and stable region under various early warning conditions.

[0085] Furthermore, when establishing the constraint calculation model based on Lyapunov stability theory, a state-space expression is first constructed based on the equivalent circuit equations of the DC distribution network. The converter output voltage and current are used as state variables, and an energy function is constructed using Lyapunov functions to describe the dynamic characteristics. The Lyapunov matrix inequality is solved in the time domain to obtain sufficient conditions for asymptotic stability, which are directly transformed into constraints on the upper bound of the PWM modulation ratio. Combining the current DC bus voltage fluctuation amplitude, the voltage deviation is substituted into the Lyapunov stability criterion to derive the maximum allowable upper bound of the modulation ratio to maintain system stability. Simultaneously, based on the AC side harmonic distortion rate index, a second constraint correction is applied to the upper bound calculated by the Lyapunov stability theory using a predefined harmonic distortion rate-modulation ratio relationship curve. Finally, the upper bound parameter of the modulation ratio that simultaneously satisfies the Lyapunov stability theory and harmonic distortion rate constraints is output.

[0086] Based on the PWM carrier frequency correction and the safe adjustment range of the PWM modulation ratio, the remaining service life assessment result of the DC support capacitor is generated by the lifetime prediction algorithm.

[0087] It should be noted that the PWM carrier frequency correction is converted into harmonic current components through Fourier series expansion, and the effective value of the ripple current generated on the capacitor by each harmonic is calculated. The DC bus voltage fluctuation amplitude is determined according to the safe adjustment range of the PWM modulation ratio. Combined with the real-time monitored capacitor surface temperature data, a three-dimensional stress space of current, voltage, and temperature is established. By continuously collecting parameters such as the capacitor's equivalent series resistance, capacitance, and loss tangent, the parameter variation curves over time are plotted. The ripple current, voltage fluctuation, and temperature parameters under the current operating condition are input into the parameter variation curves and extrapolated to obtain two evaluation results: the lower limit of the remaining service life considering the maximum stress condition and the expected value of the remaining service life based on typical operating conditions.

[0088] Based on the assessment results of remaining usage time, a preventive maintenance strategy is generated using a maintenance decision optimization method that incorporates current control parameters.

[0089] It should be noted that the guaranteed and reference predicted values ​​in the remaining usage time assessment results are matched according to pre-set maintenance rules. When the predicted value is lower than a specific threshold, a corresponding maintenance suggestion is triggered. For the PWM carrier frequency correction, a carrier frequency adjustment scheme is generated based on the correlation between ripple current stress level and lifespan. For the safe adjustment range of the PWM modulation ratio, a modulation ratio limiting optimization suggestion is formed based on the correlation characteristics between voltage stress and lifespan. Combining the real-time monitoring trends of the capacitor's equivalent series resistance and capacitance, a preventative maintenance strategy including maintenance timing, priority, and specific measures is generated according to preset maintenance level standards. The final output maintenance strategy directly corresponds to the current control parameters, achieving optimal allocation of maintenance resources while ensuring reliable equipment operation.

[0090] This embodiment also provides a computer device applicable to the broadband impedance analysis method based on the network structure in a DC distribution network, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the broadband impedance analysis method based on the network structure in a DC distribution network as proposed in the above embodiment.

[0091] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0092] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the broadband impedance analysis method based on the network structure in a DC distribution network as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0093] In summary, this invention utilizes a distributed network protocol and a federated learning framework to synchronize and aggregate data from each node, generating a power grid health report that includes a regional impedance deviation index. This process significantly enhances the collaborative capabilities of power grid systems in different geographical locations. This method not only accurately assesses the overall health of the power grid but also dynamically generates a regional impedance deviation index, providing crucial information for subsequent maintenance decisions.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for broadband impedance analysis based on network structure in a direct current power distribution network, characterized by: The application relates to a wideband impedance spectrum and hierarchical early warning information generation method based on a power grid health report. Real-time acquisition and preprocessing of voltage and current data, time-frequency analysis through short-time Fourier transform, and generation of time-frequency domain impedance characteristics; Based on the time-frequency domain impedance characteristics, the frequency-time domain joint features are extracted through a lightweight convolutional neural network, and the fault detection results are generated through a multilayer perceptron classifier; Based on the fault detection score, the node detection results are synchronized through a distributed network protocol, and the frequency-time domain joint feature extraction weight parameters are weighted and aggregated through a federated learning framework to generate a power grid health report containing a regional impedance deviation index; Based on the power grid health report, the time-frequency domain impedance characteristics and the regional impedance deviation index are mapped and classified in the feature space through a support vector machine algorithm to generate a wideband impedance spectrum and hierarchical early warning information; Based on the wideband impedance spectrum and hierarchical early warning information, the converter PWM control parameters are adjusted in real time through an adaptive control algorithm, and the equipment maintenance strategy is generated in combination with the power grid health report.

2. The method of claim 1, wherein the network structure is a DC power distribution network. The preprocessing includes elimination of power frequency harmonics and random noise through a Gaussian filter and signal normalization processing.

3. The method of claim 2, wherein the network structure is a DC power distribution network. The time-frequency analysis through short-time Fourier transform to generate the time-frequency domain impedance characteristics specifically includes the following steps, Time-frequency decomposition of the preprocessed voltage and current data through short-time Fourier transform to generate a time-frequency spectrum matrix; Based on the time-frequency spectrum matrix, impedance characteristics conversion is performed through the frequency domain Ohm's law to generate the time-frequency domain impedance characteristics.

4. The method of claim 3, wherein the network structure is a DC power distribution network. Based on the time-frequency domain impedance characteristics, the frequency-time domain joint features are extracted through a lightweight convolutional neural network, and the fault detection results are generated through a multilayer perceptron classifier, specifically including the following steps, Based on the time-frequency domain impedance characteristic matrix, a multi-dimensional feature tensor with space-time correlation is generated through sliding calculation of a three-dimensional convolution kernel; Based on the multi-dimensional feature tensor, a frequency-time domain joint feature vector is generated through a cross-channel attention fusion mechanism; Based on the frequency-time domain joint feature vector, a multilayer perceptron classifier is used to generate a fault detection result y; Based on historical data, a fault threshold A is defined; When y < A, it is considered that the equipment is normally operated, and the current monitoring mode is maintained; When y >= A, it is considered that the equipment is in an abnormal state, and the operation and maintenance personnel need to be immediately notified for repair.

5. The method of claim 4, wherein the network structure is a DC power distribution network. Based on the fault detection result, the node data is synchronized through a distributed protocol, and the frequency-time domain joint feature extraction model parameters are aggregated through a federated learning framework to generate a power grid health report containing a regional impedance deviation index, specifically including the following steps, Through a distributed consistency protocol, the node detection data is time-aligned and quality-checked to generate a synchronized data set; Based on the synchronized data set, the neural network parameter update amount of each monitoring terminal is aggregated through a federated learning framework, and a weighted average algorithm is used to generate global shared parameters; Based on the global shared parameters, a power grid health report is generated through impedance deviation analysis and health score mapping.

6. The method of claim 5, wherein the network structure is a DC power distribution network. Based on the power grid health report, the time-frequency domain impedance characteristics and the regional impedance deviation index are mapped and classified in the feature space through a support vector machine algorithm to generate a wideband impedance spectrum and hierarchical early warning information, specifically including the following steps, Based on the power grid health report, the time-frequency domain impedance characteristics and the regional impedance deviation index are mapped to a high-dimensional space through nuclear space mapping to generate a high-dimensional feature representation; Based on the high-dimensional feature representation, an optimal classification decision boundary is generated through a support vector machine optimization algorithm; Based on the classification decision boundary, a wideband impedance atlas and hierarchical warning information are generated through impedance feature clustering and abnormal region positioning for atlas reconstruction.

7. The method of claim 6, wherein the network structure is a DC power distribution network. Based on the wideband impedance atlas and hierarchical warning information, the converter PWM control parameters are adjusted in real time through an adaptive control algorithm, and the specific steps are as follows Based on the wideband impedance atlas and hierarchical warning information, the dominant resonance frequency band characteristics are generated through Hilbert spectrum analysis; Based on the dominant resonance frequency band characteristics, the PWM carrier frequency correction amount is generated through a PID regulator; Based on the hierarchical warning information, the PWM modulation ratio safety adjustment range is generated through a dynamic limiting strategy.

8. The method of claim 7, wherein the network structure is a DC power distribution network. The equipment maintenance strategy is generated in combination with the power grid health report, and the specific steps are as follows, Based on the PWM carrier frequency correction amount and the PWM modulation ratio safety adjustment range, the remaining use time length evaluation result of the DC support capacitor is generated through a life prediction algorithm; Based on the remaining use time length evaluation result, the preventive maintenance strategy of the current control parameter is generated through a maintenance decision optimization method. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the wideband impedance analysis method based on the network structure in the DC distribution network according to any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the wideband impedance analysis method based on the network structure in the DC distribution network according to any one of claims 1-8.

Images