Low frequency load shedding method and system based on zero-crossing point real-time frequency tracking

Abstract

The application discloses a low-frequency load shedding method and system based on zero-crossing point real-time frequency tracking, and specifically comprises the following steps: setting a dynamic time window according to the power frequency characteristics of a sensing signal, and eliminating abnormal zero-crossing points beyond the dynamic time window; matching a pre-stored harmonic feature library, establishing a mapping relationship between a harmonic mode and a phase shift, and adaptively compensating for the phase of pseudo zero-crossing points of the sensing signal; inputting the sensing signal, PMU phasor data and meteorological parameters into a three-dimensional convolutional neural network, calculating a cross-modal correlation matrix through a self-attention mechanism, combining an LSTM unit to predict a future frequency change trajectory, and outputting an optimal load shedding strategy; and based on the optimal load shedding strategy, combining forward acceleration criteria and reverse fault locking, and executing emergency load shedding according to load importance classification. The application effectively solves the technical bottleneck of traditional low-frequency load shedding devices, and significantly improves the performance and reliability of power system frequency stability control.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a low-frequency load shedding method and system based on real-time frequency tracking at zero crossing points. Background Technology

[0002] With the transformation of the global energy structure and the rapid development of new energy power generation technologies, the high proportion of new energy connected to the grid has become an inevitable trend in power system development. New energy power generation, such as wind power and photovoltaic power, is characterized by intermittency, volatility, and uncertainty. Its large-scale grid connection poses unprecedented challenges to the frequency stability of the power system. In cases of rapid fluctuations in new energy power output or sudden grid disconnection, the grid frequency may drop by milliseconds or even faster. If effective control measures are not taken in time, this could lead to system frequency collapse, resulting in widespread blackouts and seriously affecting the safe and stable operation of the power system.

[0003] Traditional low-frequency load shedding devices, as an important means of frequency stability control in power systems, play a crucial role in ensuring system security. However, facing the complex operating conditions of scenarios with a high proportion of renewable energy integration, traditional low-frequency load shedding devices have exposed many technical bottlenecks and limitations, specifically in the following aspects:

[0004] 1. Insufficient frequency measurement accuracy: Traditional low-frequency load shedding devices mostly use software algorithms based on fixed delay period measurement for frequency calculation. When the frequency changes rapidly, due to the limitations of the algorithm itself, it is easy to generate cumulative errors, resulting in deviations in the frequency calculation results, which in turn affects the accuracy of load shedding decisions.

[0005] 2. Sensitive to harmonic interference: When the harmonic content of the power grid is high (such as total harmonic distortion rate THD>8%), the zero-crossing detection algorithm of the traditional low-frequency load shedding device is susceptible to harmonic interference, and the misjudgment rate increases significantly (up to 15%), resulting in a frequency calculation error of more than ±0.2Hz, which seriously affects the reliability and effectiveness of the load shedding device.

[0006] 3. Lagging response speed: Traditional low-frequency load shedding devices typically require a time delay of 300-500ms from detecting a frequency exceeding the limit to finally executing the load shedding operation. This delay is obviously insufficient to meet the requirements for rapid response in the millisecond-level frequency drop scenario caused by new energy grid disconnection, and may cause the system frequency to drop to a dangerous level before the load shedding action.

[0007] 4. Single-criteria protection is prone to false tripping: Traditional low-frequency load shedding devices often rely solely on a single frequency criterion when determining whether load shedding is necessary. They do not fully consider the impact of factors such as voltage disturbances on frequency measurement, and are prone to false tripping under the illusion of frequency fluctuations caused by voltage disturbances, resulting in unnecessary load shedding and affecting the economy and reliability of the power system. Summary of the Invention

[0008] The purpose of this invention is to provide a low-frequency load shedding method and system based on real-time frequency tracking at zero crossing points, which effectively solves the technical bottlenecks of traditional low-frequency load shedding devices, significantly improves the performance and reliability of power system frequency stability control, and thus better adapts to the power system frequency stability control requirements under high-proportion renewable energy access scenarios, thereby solving at least one of the aforementioned prior art problems.

[0009] In a first aspect, the present invention provides a low-frequency load reduction method based on real-time frequency tracking at zero crossing points, the method specifically comprising:

[0010] The high-voltage signal from the power grid is converted into a low-voltage signal by a voltage transformer, a standard square wave signal is generated by a zero-crossing comparator circuit, and a hysteresis comparator circuit is constructed using a high-speed comparator to eliminate signal jitter and form a sensing signal.

[0011] Based on the power frequency characteristics of the sensor signal, a dynamic time window is set to eliminate abnormal zero crossings that exceed the dynamic time window.

[0012] The total harmonic distortion rate of the sensor signal is calculated in real time using sliding discrete Fourier transform. The activation of the dynamic weighted average algorithm is determined by comparing the total harmonic distortion rate with a preset distortion rate threshold.

[0013] By matching a pre-stored harmonic feature library and using a pre-trained deep generative adversarial network, a mapping relationship between harmonic modes and phase shifts is established, and adaptive phase compensation is performed on the pseudo-zero crossings of the sensing signal.

[0014] The sensor signals, PMU phasor data and meteorological parameters are input into a three-dimensional convolutional neural network. The cross-modal correlation matrix is ​​calculated through a self-attention mechanism. The LSTM unit is combined to predict the future frequency change trajectory and output the optimal load reduction strategy.

[0015] Based on the optimal load reduction strategy, combined with the positive acceleration criterion and the reverse fault blocking, emergency load reduction is performed in stages according to the importance of the load, and the load reduction amount is dynamically adjusted through the Nash equilibrium model to minimize load loss and frequency overshoot.

[0016] Secondly, the present invention provides a low-frequency load shedding system based on real-time frequency tracking at zero crossing points, the system specifically comprising:

[0017] The signal processing module is used to convert the high-voltage signal of the power grid into a low-voltage signal through a voltage transformer, generate a standard square wave signal through a zero-crossing comparator circuit, and use a high-speed comparator to construct a hysteresis comparator circuit to eliminate signal jitter and form a sensing signal.

[0018] The primary filtering module is used to set a dynamic time window based on the power frequency characteristics of the sensing signal and to eliminate abnormal zero crossings that exceed the dynamic time window.

[0019] The secondary verification module is used to calculate the total harmonic distortion rate of the sensor signal in real time using sliding discrete Fourier transform, and determines whether to activate the dynamic weighted average algorithm by comparing the total harmonic distortion rate with a preset distortion rate threshold.

[0020] The three-level compensation module is used to match the pre-stored harmonic feature library, and to establish the mapping relationship between harmonic modes and phase offsets using a pre-trained deep generative adversarial network, so as to perform adaptive phase compensation for the pseudo-zero crossings of the sensing signal.

[0021] The load reduction prediction module is used to input sensor signals, PMU phasor data and meteorological parameters into a three-dimensional convolutional neural network, calculate the cross-modal correlation matrix through a self-attention mechanism, and combine LSTM units to predict the future frequency change trajectory and output the optimal load reduction strategy.

[0022] The composite criterion module is used to perform emergency load shedding based on the optimal load shedding strategy, combining positive acceleration criteria and reverse fault blocking, and to dynamically adjust the load shedding amount through the Nash equilibrium model to minimize load loss and frequency overshoot.

[0023] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements the low-frequency load reduction method based on real-time frequency tracking of zero crossing as described in any of the above methods.

[0024] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the low-frequency load reduction method based on real-time frequency tracking of zero crossing as described in any of the above methods.

[0025] Compared with the prior art, the present invention has at least one of the following technical effects:

[0026] 1. This invention effectively solves the technical bottleneck of traditional low-frequency load shedding devices, significantly improves the performance and reliability of power system frequency stability control, and thus better adapts to the power system frequency stability control requirements under high-proportion renewable energy access scenarios.

[0027] 2. By tracking the zero-crossing frequency in real time, this invention effectively improves the response speed and accuracy of low-frequency load shedding, realizes intelligent load shedding strategy optimization under multi-modal data fusion, and significantly enhances the frequency stability of the power system.

[0028] 3. The dynamic time window setting and abnormal zero-crossing rejection technology of this invention effectively filters out measurement errors and noise interference, ensuring the accuracy and stability of frequency measurement and providing a reliable data foundation for subsequent processing.

[0029] 4. The sliding discrete Fourier transform combined with the dynamic weighted average algorithm of this invention realizes real-time monitoring and adaptive adjustment of harmonic distortion rate, effectively suppresses the influence of harmonic interference on frequency measurement, and improves measurement accuracy.

[0030] 5. This invention achieves intelligent mapping between harmonic modes and phase offsets by applying deep generative adversarial networks and adaptive phase compensation for pseudo-zero crossings, further improving the accuracy and reliability of frequency measurement.

[0031] 6. This invention combines a three-dimensional convolutional neural network with an LSTM unit to achieve deep fusion of multimodal data and accurate prediction of future frequencies, providing a scientific basis for the formulation of optimal load reduction strategies and enhancing the intelligence and foresight of load reduction decisions.

[0032] 7. By combining the positive acceleration criterion with the reverse fault blocking and the application of the Nash equilibrium model, this invention achieves dynamic optimization of load reduction and minimization of load loss, effectively balancing system stability and economy.

[0033] 8. This invention constructs an objective function using a Nash equilibrium model, fully considering load weight and frequency stability requirements, and achieves the optimization of load reduction decisions, providing strong support for power system frequency stability control. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating a low-frequency load reduction method based on real-time frequency tracking at zero crossing points, provided in an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of the structure of a low-frequency load shedding system based on real-time frequency tracking at zero crossing points, provided in an embodiment of the present invention.

[0037] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0038] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0039] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0040] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0041] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0042] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0043] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0044] In this application embodiment, the entity executing the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 A flowchart illustrating the low-frequency load reduction method based on real-time frequency tracking at zero crossing points disclosed in the first embodiment of the present invention is shown below in detail:

[0045] S101 converts the high-voltage signal of the power grid into a low-voltage signal through a voltage transformer, generates a standard square wave signal through a zero-crossing comparator circuit, and uses a high-speed comparator to construct a hysteresis comparator circuit to eliminate signal jitter and form a sensing signal.

[0046] In this embodiment, a high-precision voltage transformer (such as type XX) with a rated transformation ratio of 10kV / 0.1kV is used to meet the conversion requirements of high-voltage signals (10kV) to low-voltage signals (0.1kV) from the power grid. The primary and secondary sides of the voltage transformer are insulated with epoxy resin, with a withstand voltage rating of ≥30kV to ensure the safety of the secondary side in the event of a grid-side fault. The transformation ratio error is ≤0.2%, and the phase angle difference is ≤10', ensuring that the low-voltage signal accurately reflects the amplitude and phase of the grid voltage.

[0047] The input of the high-speed comparator is connected to the low-voltage signal (0.1kV) on the secondary side of the voltage transformer. The signal amplitude is adjusted to the input range (±5V) of the high-speed comparator (e.g., LM319) using voltage divider resistors (R1=10kΩ, R2=1kΩ). The output of the high-speed comparator is connected to the external interrupt pin of a microcontroller (e.g., STM32F4 series) to capture zero-crossing events. When the input voltage changes from negative to positive, the comparator output transitions from low to high, triggering a microcontroller interrupt and recording a timestamp T1. When the input voltage changes from positive to negative, the comparator output transitions from high to low, triggering an interrupt and recording a timestamp T2. The time difference between two adjacent interrupts, ΔT=T2-T1, is used to calculate the current signal period T=2ΔT (assuming sine wave symmetry).

[0048] A positive feedback resistor (Rf=100kΩ) is added to the input of the high-speed comparator to create a hysteresis characteristic. The hysteresis voltage threshold is set to ±100mV to suppress glitches in the input signal (such as ±50mV noise). When the input voltage rises above the positive threshold (e.g., +100mV), the comparator outputs a high level; when the input voltage drops below the negative threshold (e.g., -100mV), the comparator outputs a low level. The hysteresis characteristic ensures that the comparator output remains stable when the input signal fluctuates near the threshold, eliminating signal jitter.

[0049] The microcontroller records zero-crossing timestamps at a sampling rate of 100kHz, forming the original sensor signal sequence {T1,T2,T3,…}. The signal period sequence {T1-T0,T2-T1,T3-T2,…} is calculated using the difference between adjacent timestamps and converted into a frequency sequence {f1,f2,f3,…}.

[0050] In this embodiment, the coordinated design of voltage transformer, zero-crossing comparator circuit and hysteresis comparator circuit realizes the generation of high-precision and anti-interference sensing signals, providing a reliable data foundation for subsequent frequency calculation and load reduction decision-making, and effectively solving the technical bottleneck of traditional low-frequency load reduction devices in high-proportion new energy scenarios.

[0051] S102, based on the power frequency characteristics of the sensing signal, sets a dynamic time window and eliminates abnormal zero-crossing points that exceed the dynamic time window.

[0052] In this embodiment, the power frequency (e.g., 50Hz or 60Hz) is the core characteristic of the power grid signal. However, the actual acquired signal often contains harmonics (e.g., 3rd or 5th harmonics) or noise, which may lead to false zero-crossings and affect the frequency measurement accuracy of traditional zero-crossing methods. In the prior art, fixed time windows or simple threshold methods cannot adapt to the dynamic changes in power frequency, and may misjudge or miss valid zero-crossings. Therefore, this invention ensures the robustness of frequency calculation by adjusting the time window range in real time and eliminating abnormal zero-crossings.

[0053] Specifically, the acquired signal is preliminarily analyzed using the sliding window method or the zero-crossing method to obtain the current power frequency range of the signal. For example, if the main frequency of the signal is 50Hz, the theoretical time interval between adjacent zero-crossing points is 20ms (1 / 50Hz). Considering that the actual power frequency may vary due to power grid fluctuations (e.g., 45Hz~55Hz), a center value and tolerance range for a dynamic time window are set: the center value is calculated based on the theoretical zero-crossing time interval according to the currently estimated power frequency; the tolerance range is set based on the width of the time window according to the power frequency fluctuation range, for example, a tolerance of ±10%. Assuming the currently estimated power frequency is 50Hz, the theoretical zero-crossing time interval is 20ms. The time window range is set to 18ms~22ms (i.e., ±10% tolerance). If a zero-crossing time interval is not within the range of 18ms~22ms, it is judged as an abnormal zero-crossing point and discarded.

[0054] The acquired signal is preprocessed using a bandpass filter (e.g., 40Hz~60Hz) to suppress high-frequency noise and harmonics. The zero-crossing times of the signal from negative to positive (or positive to negative) are recorded, forming a zero-crossing time series. After each zero-crossing is detected, the time window range is dynamically updated based on the current power frequency estimate. For example, if the power frequency estimate becomes 52Hz, the theoretical zero-crossing time interval becomes 19.23ms, and the time window range is updated to 17.31ms~21.15ms. The time intervals of adjacent zero-crossings are checked one by one; if a time interval is outside the current time window range, the corresponding zero-crossing is discarded. Only zero-crossings that pass the time window verification are retained, forming a valid zero-crossing time series. The signal frequency is calculated based on the valid zero-crossing time series. For example, if the valid zero-crossing time interval is 20ms, the signal frequency is 50Hz.

[0055] In this embodiment, the dynamic time window method effectively eliminates abnormal zero-crossing points by adjusting the time window range in real time, thereby improving the accuracy of frequency calculation, especially in scenarios with harmonic interference and frequency abrupt changes.

[0056] S103 uses sliding discrete Fourier transform to calculate the total harmonic distortion rate of the sensor signal in real time, and determines whether to activate the dynamic weighted average algorithm by comparing the total harmonic distortion rate with a preset distortion rate threshold.

[0057] In this embodiment, in power systems and industrial sensing signals, harmonic interference and signal fluctuations can increase the errors of traditional signal analysis methods (such as fixed-window Fourier transform). For example, the fixed-window method struggles to track dynamic changes in the signal in real time, leading to a significant deviation between the calculated THD (Total Harmonic Distortion) and the actual value. Noise or transient interference may generate false signal characteristics, affecting subsequent signal analysis (such as frequency calculation and waveform recognition). To address these issues, this invention improves the accuracy and robustness of signal processing by calculating THD in real time and adaptively adjusting the signal analysis strategy.

[0058] Specifically, the continuously acquired sensor signal is segmented into segments of fixed length (e.g., N sampling points), each segment being called an "analysis window." After each analysis, the window slides forward one sampling point, forming a new analysis window, achieving real-time spectrum updates. A Fourier transform is performed on each analysis window to extract the amplitudes of the fundamental frequency (e.g., 50Hz) and each harmonic (e.g., 100Hz, 150Hz, etc.). Using the fundamental frequency amplitude as a benchmark, the relative proportions of each harmonic amplitude to the fundamental frequency amplitude are calculated, providing basic data for THD calculation. Based on the fundamental frequency amplitude and each harmonic amplitude, THD is calculated through the following logical steps: Calculate the sum of squares of all harmonic amplitudes; divide the sum of squares by the square of the fundamental frequency amplitude to obtain the dimensionless value of the distortion rate; convert the dimensionless value into a percentage form as the real-time output of THD.

[0059] A preset threshold for THD (e.g., 5%) is set based on the application scenario. For example, in a power system, 5% THD is generally considered an acceptable boundary value for power quality. The real-time calculated THD is compared with the preset threshold. If the THD exceeds the threshold, the signal is determined to have significant harmonic interference. When THD exceeds the threshold, a dynamic weighted averaging algorithm is activated to weight the subsequent signal analysis results (such as frequency and amplitude). Weights are assigned according to the reliability of the signal. For example, data from periods with lower THD are given higher weights, and data from periods with higher THD are given lower weights.

[0060] Maintain a fixed-length data buffer queue (e.g., data from the most recent 10 analysis windows) to store historical signal analysis results. Calculate the corresponding weight based on the THD value of each analysis window. For example, the weight can be inversely proportional to the THD value (the lower the THD, the higher the weight). Calculate a weighted average of the data in the buffer queue according to the weights. For example, if one window has a weight of 0.8 and another window has a weight of 0.2, the final result is the weighted sum of the data from both windows. Use the weighted average as the final analysis result for the current signal, which is then used for subsequent signal processing or decision-making.

[0061] In this embodiment, the accuracy and stability of signal processing are significantly improved by calculating THD in real time using SDFT and dynamically adjusting the signal analysis strategy, especially in harmonic interference and noise scenarios.

[0062] S104 matches a pre-stored harmonic feature library and uses a pre-trained deep generative adversarial network to establish a mapping relationship between harmonic modes and phase offsets, performing adaptive phase compensation for pseudo-zero crossings of the sensing signal.

[0063] In this embodiment, harmonic interference in sensor signal processing can cause false zero-crossings (i.e., zero-crossings at non-true zero-crossing locations), thus affecting the accuracy of key parameters such as frequency calculation and phase measurement. Traditional methods typically rely on fixed thresholds or empirical formulas for phase compensation, but these methods have the following drawbacks: harmonic modes differ significantly in different scenarios, making it difficult for fixed methods to cover all situations; noise or transient interference may generate false zero-crossings, leading to deviations in compensation results; and traditional methods have high computational complexity, making it difficult to meet real-time requirements. To address these issues, this invention achieves accurate correction of sensor signals by matching a harmonic feature library, training a GAN model, and adaptively compensating for phase shifts.

[0064] Specifically, sensor signal samples are collected from different scenarios (such as harmonic signals from power systems and output signals from industrial sensors). The samples must cover typical harmonic modes (such as the 3rd and 5th harmonics) and their phase offset ranges. Time-domain and frequency-domain analysis is performed on the sample signals to extract key features (such as harmonic amplitude, frequency, and phase). These features are then categorized and stored according to harmonic modes to form a pre-stored harmonic feature library. The feature library is periodically updated based on newly collected signal samples to ensure it covers the latest harmonic modes. Clustering algorithms (such as K-means) are used to merge features, reducing redundant data. A feature index is established to improve feature matching efficiency.

[0065] Harmonic patterns (input data) and their corresponding phase shifts (label data) are extracted from a pre-stored harmonic feature library to construct a training set. The data needs to be stratified and sampled according to dimensions such as harmonic order and amplitude range to ensure the representativeness of the training set. Data augmentation (e.g., adding noise, random phase shifts) is applied to the training set to improve the generalization ability of the GAN (Generative Adversarial Network). The GAN model includes a generator and a discriminator. The generator's input is harmonic pattern features (e.g., amplitude, frequency), and its output is the predicted phase shift. The discriminator's input is a combination of harmonic pattern features and phase shifts, and its output is a realism score (0-1) for the combined data. The generator and discriminator are trained alternately. The generator attempts to generate realistic phase shifts to deceive the discriminator, while the discriminator tries to distinguish between real and generated data. The training objective is to minimize the loss functions (e.g., cross-entropy loss) of the generator and discriminator. The GAN model is considered converged when the discriminator cannot distinguish between real and generated data.

[0066] Zero-crossing detection is performed on the real-time acquired sensor signals, and all zero-crossing positions are marked. Based on the harmonic mode characteristics of the signal, it is determined whether the zero-crossing is a pseudo-zero-crossing (e.g., by comparing it with a pre-stored feature library). Harmonic mode features (e.g., amplitude, frequency) of the current signal are extracted; the most similar feature pattern is retrieved from the pre-stored harmonic feature library. The similarity (e.g., cosine similarity) between the current feature and features in the feature library is calculated; if the similarity exceeds a preset threshold, a successful match is considered. The matched harmonic mode features are input into a trained GAN generator to predict the corresponding phase shift; the generator output is an adaptive phase compensation value. Based on the predicted phase shift, phase correction is performed on pseudo-zero-crossings; the corrected signal zero-crossing position is closer to the true zero point, improving the accuracy of signal analysis.

[0067] In this embodiment, by combining a pre-stored harmonic feature library with GAN, an adaptive mapping between harmonic modes and phase offset is achieved, effectively eliminating pseudo-zero crossing errors of the sensing signal and improving the accuracy and robustness of signal processing.

[0068] S105 inputs sensor signals, PMU phasor data and meteorological parameters into a three-dimensional convolutional neural network, calculates the cross-modal correlation matrix through a self-attention mechanism, and combines LSTM units to predict future frequency change trajectories, outputting the optimal load reduction strategy.

[0069] In this embodiment, power system frequency stability is a key indicator for ensuring the safe operation of the power grid. Traditional frequency prediction methods have the following shortcomings: sensor signals (such as current and voltage waveforms), PMU phasor data (such as amplitude and phase), and meteorological parameters (such as wind speed and light intensity) are usually analyzed independently, making it difficult to uncover the correlation between multimodal data; existing methods have limited ability to model the time-series dependence of frequency changes, resulting in insufficient prediction accuracy; load shedding strategies based on static thresholds cannot adapt to dynamically changing power grid operating states, easily leading to overload or underload. To solve the above problems, this invention achieves accurate prediction and stable control of power system frequency by fusing multimodal data, mining time-series correlations, and dynamically optimizing load shedding strategies.

[0070] Specifically, the high-precision clock synchronization function of the PMU is used to acquire synchronization phasor data (such as voltage amplitude and phase angle) of the power grid nodes. The phasor data is then processed using a sliding window (10-second window length, 1-second step) to generate a time-series feature sequence. The phasor data is converted into a two-dimensional feature map (such as an amplitude-phase heatmap) as another input data for the 3D-CNN. Meteorological parameters (such as wind speed, light intensity, and temperature) of the power grid coverage area are obtained from weather stations or weather forecasting platforms. These meteorological parameters are normalized to eliminate dimensional differences. The meteorological parameters are then mapped into a two-dimensional feature map (such as a wind speed-time curve) as the third input data for the 3D-CNN (three-dimensional convolutional neural network).

[0071] The input layer of the 3D-CNN receives the time-frequency map of the sensor signal, the PMU phasor feature map, and the meteorological parameter feature map, forming three-dimensional input data (time, space, and feature dimensions). The convolutional layers use three-dimensional convolutional kernels (e.g., 3×3×3) to extract the spatial-temporal features of the multimodal data. Multiple convolutional layers are set to gradually expand the receptive field and capture global features. The pooling layers use max pooling or average pooling to reduce the feature dimensionality and computational cost. A self-attention mechanism is applied to the output feature map of the 3D-CNN to calculate the correlation weight matrix between different modalities. The correlation weight matrix represents the contribution of each modality to frequency prediction. Based on the correlation weight matrix, the multimodal features are weighted and fused to generate a fused feature vector.

[0072] The input layer of the LSTM unit receives the fused feature vectors output by the 3D-CNN, forming a temporal input sequence. The temporal layer employs multiple LSTM units to capture the temporal dependence of frequency changes, using forget gates, input gates, and output gates to control the retention and updating of historical information. The output layer outputs the frequency change trajectory over a future period (e.g., the predicted frequency value within the next 10 minutes). The predicted frequency trajectory is compared with the actual frequency data to calculate the prediction error (e.g., mean squared error, mean absolute error). Based on the prediction error, the parameters of the LSTM unit (e.g., learning rate, number of hidden layer nodes) are dynamically adjusted to optimize prediction performance.

[0073] Define a load shedding strategy space, including parameters such as load shedding amount and priority for different nodes; generate a set of candidate load shedding strategies based on the power grid topology and load characteristics. Set evaluation indicators (such as frequency stability, load loss, and economic cost) to quantify the merits of each strategy. Use the frequency trajectory predicted by LSTM as a constraint to screen load shedding strategies that meet frequency stability requirements; employ multi-objective optimization algorithms (such as genetic algorithms and particle swarm optimization) to search for the optimal solution in the strategy space. Dynamically adjust the load shedding strategy based on real-time frequency changes to ensure power grid frequency stability.

[0074] In this embodiment, by fusing multimodal data, mining cross-modal correlations, and dynamically optimizing load reduction strategies, accurate prediction and stable control of power system frequency are achieved, significantly improving the grid's anti-interference capability and operating efficiency.

[0075] S106, based on the optimal load reduction strategy, combines positive acceleration criteria and reverse fault blocking, performs emergency load reduction according to the load importance level, and dynamically adjusts the load reduction amount through the Nash equilibrium model to minimize load loss and frequency overshoot.

[0076] In this embodiment, traditional emergency load shedding methods for power systems have the following shortcomings: load shedding strategies based on fixed thresholds cannot adapt to dynamically changing grid operating conditions, easily leading to overload or underload; they do not consider load priorities (such as industrial load, residential load, and critical infrastructure load), potentially causing unnecessary socio-economic losses; frequency fluctuations may occur significantly during load shedding, affecting grid stability; and the load shedding amount is usually statically set and cannot be dynamically adjusted according to real-time grid conditions. To solve these problems, this invention uses a positive acceleration criterion to trigger load shedding in advance and a reverse fault blocking mechanism to prevent misoperation. Combined with load importance classification and a Nash equilibrium model, it achieves dynamic optimization of the load shedding amount, minimizing load loss and frequency overshoot.

[0077] Specifically, the grid frequency change rate (df / dt) is monitored in real time. When the frequency change rate exceeds the positive acceleration threshold (e.g., df / dt > 0.5 Hz / s), an emergency load shedding process is triggered. By initiating the load shedding process early using the positive acceleration criterion, further frequency degradation is prevented, allowing time for subsequent load shedding operations. During load shedding, the grid status is monitored in real time. If a reverse fault is detected (e.g., short-circuit fault, switch malfunction), the load shedding operation is immediately blocked to prevent erroneous load shedding. The reverse fault blocking mechanism compares the current grid status with the state before the fault to determine if any abnormal changes exist.

[0078] Based on the socio-economic impact and outage tolerance of the loads, loads are divided into three categories: Category I loads: critical infrastructure (such as hospitals and transportation hubs) and important industrial loads; Category II loads: general industrial loads and commercial loads; Category III loads: residential loads and non-critical agricultural loads. Among them, Category I loads have the highest priority, and Category III loads have the lowest priority. During load shedding, lower priority loads are shelved first, while higher priority loads are retained. Load shedding is carried out in ascending order of load priority. After each round of load shedding, the grid frequency is reassessed. If the frequency has not returned to stability, the next round of load shedding continues.

[0079] Each load node in the power grid is defined as a game player, and each player chooses its load reduction amount based on its own interests (such as load loss and frequency stability). The game objective is to minimize the total loss of all players while meeting the power grid frequency stability requirements. Each player's strategy space is a set of selectable load reduction amounts (e.g., 0%, 10%, 20%). Through multiple rounds of game play, each player gradually adjusts its strategy until a Nash equilibrium is reached. During the load reduction process, the power grid frequency and load status are monitored in real time, and feedback information is input into the Nash equilibrium model. Based on the feedback information, the load reduction amount of each player is dynamically adjusted to ensure power grid frequency stability. When the strategies of all players no longer change, or the change is less than a set threshold, a Nash equilibrium is considered reached, and adjustments cease.

[0080] In some embodiments, step S102 above, which involves setting a dynamic time window based on the power frequency characteristics of the sensing signal and eliminating abnormal zero-crossing points that exceed the dynamic time window, specifically includes:

[0081] The initial time window is calculated based on the allowable range of the power frequency of the sensing signal, and the dynamic time window range is set in combination with the time margin used to compensate for measurement errors. The time margin is dynamically adjusted by the real-time frequency change rate.

[0082] Collect any two timestamps of adjacent zero crossings, calculate the time interval between the two timestamps, and if the time interval exceeds the range of the dynamic time window, mark the zero crossing as an outlier and remove it.

[0083] Historical periodic data is stored using a sliding window, and the weighted average period is calculated using an exponential decay weighting formula. The dynamic time window range is then adjusted based on the weighted average period.

[0084] If the continuous anomaly manifests as a rapid frequency drop, emergency control is triggered; otherwise, the zero-crossing sequence is reconstructed using weighted average period interpolation.

[0085] In this embodiment, the zero-crossing point of the sensor signal is used for frequency calculation and phase synchronization in power system monitoring. However, due to the influence of harmonics, noise, or transient faults, the zero-crossing point may deviate abnormally, leading to frequency measurement errors. Traditional fixed-time-window methods are difficult to adapt to rapid frequency fluctuations, and high-precision rejection needs to be achieved by dynamically adjusting the time window.

[0086] Specifically, based on power grid standards (e.g., 50Hz ± 0.5Hz), the power frequency range is determined to be 49.5Hz to 50.5Hz. 50Hz corresponds to a period of 20ms, 49.5Hz to approximately 20.2ms, and 50.5Hz to approximately 19.8ms. The initial time window is set to 19.8ms to 20.2ms to cover the power frequency fluctuation range.

[0087] To compensate for measurement errors, the initial time window is extended by 0.1 ms at both ends (e.g., 19.7 ms to 20.3 ms). The frequency change rate is monitored in real time (e.g., once per second). If the frequency change rate exceeds a threshold (e.g., ±0.1 Hz / s), the time margin is dynamically increased: when the frequency increases, the lower limit of the time window is shortened (e.g., a margin of 0.02 ms is added for every 0.1 Hz / s); when the frequency decreases, the upper limit of the time window is extended (e.g., a margin of 0.02 ms is added for every 0.1 Hz / s).

[0088] Record zero-crossing timestamps at a fixed sampling rate (e.g., 10kHz). For adjacent zero-crossing timestamps... and Calculate the time interval .like If the time exceeds the current dynamic time window range, it is marked as abnormal. For example, if the current time window is 19.8ms to 20.2ms, and... If the timer reaches 21ms, it is considered an anomaly and the device is removed. If there are three consecutive zero-crossing anomalies, the emergency control procedure is initiated.

[0089] Store the 50 most recent valid zero-crossing periods. Newer data has a higher weight (e.g., 0.9), while older data has a lower weight (e.g., 0.1). Calculate the weighted average period using an exponential decay mechanism. Assuming historical periods are 19.9ms, 20.0ms, and 20.1ms (with decreasing weights), the weighted average period is 20.0ms. Adjust the time window range (e.g., ±0.2ms) based on the weighted average period to form a new time window of 19.8ms to 20.2ms.

[0090] If the time intervals between five consecutive zero-crossing points show a decreasing trend (e.g., from 20ms to 18ms), it is determined to be a rapid frequency drop. An alarm is triggered, the current time window is frozen, and dynamic adjustment is paused. The upper-level system is notified to take protective measures (e.g., load shedding). For the removed abnormal zero-crossing points, interpolation is performed using the weighted average of the preceding and following valid points. For example, if the 10th point is abnormal, it is replaced by the linear interpolation of the 9th and 11th points.

[0091] In this embodiment, the time window can be adjusted in real time according to frequency fluctuations, adapting to a power frequency range of ±0.5Hz. Weighted averaging of the period suppresses harmonic interference and improves zero-crossing detection accuracy. An emergency control mechanism can handle extreme conditions such as frequency drops, ensuring system safety.

[0092] In some embodiments, step S103 above, which involves calculating the total harmonic distortion (THD) rate of the sensing signal in real time using the sliding discrete Fourier transform and determining whether to activate the dynamic weighted average algorithm by comparing the THD rate with a preset distortion rate threshold, specifically includes:

[0093] The sensor signal is sampled according to the preset sampling frequency, and the k-th harmonic component is calculated by the recursive formula. The k-th harmonic component includes the fundamental component and the odd harmonic components.

[0094] Based on the k-th harmonic component, the total harmonic distortion rate of the signal in the current window is calculated using the sliding discrete Fourier transform.

[0095] The total harmonic distortion rate of the signal is compared with a preset distortion rate threshold. If the total harmonic distortion rate of the signal is greater than the preset distortion rate threshold, the weights are adaptively allocated according to the harmonic energy.

[0096] Based on the adaptively assigned weights, the fundamental frequency component is weighted and averaged to obtain the weighted fundamental frequency component. The zero-crossing sequence is reconstructed using the amplitude of the weighted fundamental frequency component, and the corrected power frequency is output.

[0097] In this embodiment, in power system monitoring, harmonic pollution of sensor signals leads to an increase in total harmonic distortion (THD), which in turn affects the accuracy of frequency calculation and phase synchronization. Traditional fixed threshold methods are difficult to adapt to dynamic changes in harmonics, and high-precision signal processing is achieved by monitoring THD in real time and adaptively adjusting filtering strategies.

[0098] Specifically, based on the power frequency signal frequency (e.g., 50Hz), the sampling frequency is set to an integer multiple of the power frequency (e.g., 10kHz, corresponding to 200 points / cycle). A circular buffer is used to store the latest N sampling points (e.g., N=256) to ensure real-time performance. The fundamental (1st harmonic) and odd harmonics (3rd, 5th, 7th, 9th, 11th, etc.) components are calculated point by point using a recursive algorithm (e.g., a simplified version of the Goertzel algorithm).

[0099] Select a window length related to the power frequency cycle. For each new sampling point, the window slides forward one point to achieve real-time updates. Calculate the sum of squares of all harmonic components (fundamental and odd harmonics) within the window, and calculate the THD based on the sum of squares. Set a THD threshold according to system requirements (e.g., 5%). If the current window THD > the threshold, activate the dynamic weighted average algorithm; otherwise, maintain the conventional filtering strategy. Assign weights to the fundamental component based on the energy proportion of each harmonic component, and weight the fundamental component accordingly.

[0100] Based on the phase information of the weighted fundamental component, the zero-crossing sequence is reconstructed. Then, based on the reconstructed zero-crossing sequence, the reciprocal of the time interval between adjacent zero-crossings is calculated to obtain the corrected power frequency.

[0101] If the THD exceeds the threshold for M consecutive windows (e.g., M=5), emergency control is triggered. The current filtering parameters are frozen, the backup filter is activated, and the upper-level system is notified to take protective measures (e.g., load shedding). The threshold and window length are dynamically optimized based on historical THD data; the harmonic model is updated periodically to improve the accuracy of weight allocation.

[0102] In this embodiment, the sliding window mechanism ensures that the THD calculation and the weighted average algorithm are updated synchronously. Dynamic weight allocation adapts to changes in harmonic energy, improving the accuracy of fundamental component extraction. An emergency control mechanism can cope with continuous harmonic pollution and ensure stable system operation.

[0103] In some embodiments, in step S104 above, the matching of the pre-stored harmonic feature library, the establishment of a mapping relationship between harmonic modes and phase shifts using a pre-trained deep generative adversarial network, and the adaptive phase compensation for the pseudo-zero crossings of the sensing signal specifically include:

[0104] By statistically analyzing the amplitude distribution of each odd harmonic component using historical fault data, collecting the initial phase angle and harmonic frequency of each odd harmonic component, establishing a pre-stored set of typical harmonic mode parameters, and forming a harmonic feature library.

[0105] A deep generative adversarial network is constructed, and the generator and recognizer of the deep generative adversarial network are trained using an adversarial loss function based on a harmonic feature library.

[0106] Calculate the current harmonic energy ratio. If the current harmonic energy ratio is greater than the preset energy ratio threshold, it is determined that there is a false zero crossing. The current harmonic parameters are input into the deep generative adversarial network, and the phase compensation amount is output.

[0107] In this embodiment, odd harmonics in a power system can cause phase shifts in the zero-crossing points of sensing signals, forming pseudo-zero-crossings and affecting the accuracy of frequency calculations. Traditional methods rely on fixed compensation parameters, which are difficult to adapt to dynamic changes in harmonic modes. This invention constructs a harmonic feature library and a deep generative adversarial network to achieve dynamic mapping between harmonic modes and phase shifts, thereby improving the accuracy of pseudo-zero-crossing compensation.

[0108] Specifically, harmonic data is collected from historical power grid fault records (such as transformer switching, capacitor bank connection, etc.). The amplitude distribution range of each odd harmonic component (3rd, 5th, 7th, etc.) is statistically analyzed to form an amplitude probability density map. For example, the amplitude distribution of the 3rd harmonic is 0.1%~10% of the fundamental amplitude; the amplitude distribution of the 5th harmonic is 0.05%~5% of the fundamental amplitude.

[0109] Record the initial phase angle (phase difference relative to the fundamental frequency) of each odd harmonic to form a phase angle statistics table. For example, the phase angle range of the 3rd harmonic is -30° to 30°, and the phase angle range of the 5th harmonic is -15° to 15°. Extract the deviation of the harmonic frequency from the power frequency (e.g., ±0.5Hz) to form a frequency offset distribution map.

[0110] Parameters such as amplitude, phase angle, and frequency are combined to form typical harmonic modes, creating a parameter set. For example, Mode 1: 3rd harmonic amplitude 5%, phase angle 10°, frequency offset 0.2Hz; Mode 2: 5th harmonic amplitude 2%, phase angle -5°, frequency offset 0.1Hz. The parameter set is stored as a database or file, forming a pre-stored harmonic feature library.

[0111] A deep generative adversarial network (GAN) is constructed, consisting of a generator and a discriminator. The generator's input layer receives harmonic parameters (amplitude, phase angle, and frequency) as input features. The hidden layers employ a multi-layer fully connected network (e.g., 3 layers, 128 neurons per layer) with ReLU activation. The output layer outputs phase compensation values ​​(e.g., -15° to 15°). The discriminator's input layer receives a vector combining harmonic parameters and phase compensation values. The hidden layers use a convolutional neural network (CNN) to extract features, followed by a fully connected layer for binary classification (real / fake). The output layer outputs probability values ​​(0 to 1).

[0112] The generator and discriminator are trained using an adversarial loss function. The generator is trained to the point that the discriminator cannot distinguish between the generated phase compensation and the real compensation, while the discriminator is trained to accurately distinguish between the real phase compensation and the fake compensation generated by the generator. The parameters in the harmonic feature library are randomly perturbed (e.g., amplitude ±5%, phase angle ±3°) to generate training samples. In the training samples, the real samples are the phase compensation corresponding to actual harmonic data, and the fake samples are the phase compensation generated by the generator.

[0113] The ratio of the current total harmonic energy to the fundamental energy is calculated. If the energy ratio exceeds a preset threshold (e.g., 10%), a false zero-crossing is identified. The amplitude, phase angle, and frequency of the current harmonics are input into a trained deep generative adversarial network (GAN). The generator outputs a corresponding phase compensation amount (e.g., -8°). The phase angle of the harmonic components is adjusted based on the compensation amount. Based on the adjusted harmonic components, the zero-crossing timestamp is recalculated to correct the false zero-crossing.

[0114] Newly acquired harmonic data and compensation results are periodically added to the training set for incremental training of the generative adversarial network. The weight parameters of the generator and recognizer are updated periodically to improve model adaptability. The frequency calculation errors before and after compensation are compared to evaluate the effectiveness of phase compensation. If the error after multiple consecutive compensations still exceeds the threshold, an anomaly alarm is triggered, notifying maintenance personnel.

[0115] In this embodiment, the deep generative adversarial network (GAN) learns the mapping relationship between harmonic modes and phase shifts to achieve accurate compensation for pseudo-zero crossings. The online learning mechanism enables the system to adapt to dynamic changes in harmonic modes, improving long-term operational stability. Harmonic energy ratio calculation and phase compensation output are performed synchronously to ensure real-time compensation.

[0116] In some embodiments, step S105 above, which involves inputting the sensor signal, PMU phasor data, and meteorological parameters into a three-dimensional convolutional neural network, calculating the cross-modal correlation matrix through a self-attention mechanism, predicting the future frequency change trajectory using an LSTM unit, and outputting the optimal load reduction strategy, specifically includes:

[0117] Collect PMU phasor data and meteorological parameters, align the sensor signals, PMU phasor data and meteorological parameters according to time, and then stitch them together along the channel dimension to form a three-dimensional tensor.

[0118] A three-dimensional convolutional neural network is used to extract features from a three-dimensional tensor, and a high-order feature map is generated by the ReLU activation function;

[0119] Linear projection is performed on the high-order feature map to obtain the query matrix, key matrix and value matrix. Based on the query matrix, key matrix and value matrix, the cross-modal correlation matrix is ​​calculated and weighted fusion is performed to obtain the weighted fused features.

[0120] The weighted fusion features are input into the LSTM unit, the hidden state is updated through the state equation, and the frequency prediction value within the preset future time period is output.

[0121] In this embodiment, traditional power system frequency prediction methods often rely on single-mode data (such as PMU phasors), making it difficult to capture the impact of multi-source data (such as weather changes and equipment status) on frequency under complex operating conditions. This invention constructs a three-dimensional convolutional neural network with a self-attention mechanism to achieve cross-modal correlation analysis of multimodal data, and combines this with LSTM units to predict future frequency change trajectories, providing a basis for dynamic load shedding decisions.

[0122] Specifically, phasor data such as voltage amplitude, phase angle, and frequency of grid nodes are acquired through a synchronous phasor measurement unit (PMU) at a refresh rate of 50 frames per second. Real-time meteorological data, including temperature (range: -20℃~50℃), humidity (range: 10%~100%), and wind speed (range: 0m / s~30m / s), are obtained from weather stations. The sensor signals, PMU phasor data, and meteorological parameters are timestamped to ensure consistency across modal data in the time dimension. The three types of data are concatenated into a three-dimensional tensor along the channel dimension, with the structure [time step × number of channels × feature dimension]. The time step covers data from the past hour (e.g., 3600 time points, step size 1 second), the number of channels is 3 (sensor signals, PMU phasors, and meteorological parameters), and the feature dimension is the sum of the feature numbers of each modal data (e.g., 2-dimensional sensor signals: voltage, current; 6-dimensional PMU phasors: amplitude, phase angle, frequency, etc.; 3-dimensional meteorological parameters: temperature, humidity, wind speed).

[0123] A three-layer 3D convolutional architecture is employed, with a kernel size of 3×3×3, a stride of 1, "same" padding, and ReLU activation. Each convolutional layer is followed by a batch normalization layer to accelerate training convergence. Every two convolutional layers are followed by a max pooling layer with a 2×2×2 pooling window to reduce feature dimensionality. After convolution and pooling operations, a high-order feature map is generated, with a structure of [time step / compression ratio × number of channels / compression ratio × feature dimensionality / compression ratio]. For example, an input tensor [3600×3×11], after three convolutional layers, outputs a feature map [225×1×16] (compression ratio of 16).

[0124] Linear projection is performed on the high-order feature map to generate a query matrix (Q), a key matrix (K), and a value matrix (V), all with dimensions of [time step / compression ratio × feature dimension / compression ratio × number of attention heads]. For example, the feature map [225×1×16] is projected as Q, K, V [225×1×8] (with 8 attention heads).

[0125] Calculate the dot product of the query matrix and the key matrix to generate a cross-modal association matrix. The dimension of the association matrix is ​​[time step / compression ratio × time step / compression ratio × number of attention heads], representing the association strength of each modality feature at different time steps. Softmax normalize the association matrix to obtain the attention weights, and multiply them with the value matrix to generate a weighted fusion feature. The structure of the weighted fusion feature is [time step / compression ratio × feature dimension / compression ratio × number of attention heads].

[0126] Weighted fusion features are input into an LSTM unit with 64 hidden layer neurons and Tanh activation function. The output layer is a fully connected layer, outputting the predicted future frequency value. The hidden state is updated using the LSTM's state equation to capture time-series dependencies. The frequency change trajectory within a preset time period (e.g., 5 minutes, 10 minutes) is output, with a sampling frequency of 1 second. Based on the predicted frequency change trajectory and combined with grid safety constraints (e.g., a lower frequency limit of 49.5Hz), the optimal load reduction is calculated. The load reduction strategy is output, including the load reduction node, the load reduction amount, and the execution time. For example, node A: reduce load by 50MW, execute immediately; node B: reduce load by 30MW, execute after 30 seconds.

[0127] Newly collected multimodal data and prediction results are periodically added to the training set for incremental training of 3D-CNN and LSTM models. Model parameters are updated periodically to improve prediction accuracy. The mean squared error (MSE) and mean absolute error (MAE) are calculated by comparing the predicted frequency with the actual frequency. The effect of the load reduction strategy is simulated in a simulation environment to evaluate the frequency recovery speed and stability. If the prediction error exceeds the threshold multiple times consecutively, an alarm is triggered to notify maintenance personnel.

[0128] In this embodiment, a three-dimensional convolutional neural network and a self-attention mechanism are used to effectively capture cross-modal correlations between sensor signals, PMU phasors, and meteorological parameters. An LSTM unit combines historical data with real-time input to achieve accurate prediction of future frequency changes. Based on the prediction results, an optimal load shedding strategy is generated to improve grid frequency stability.

[0129] In some embodiments, step S106 above, which involves implementing emergency load shedding based on the optimal load reduction strategy, combining positive acceleration criteria and reverse fault blocking, and classifying loads according to their importance, while dynamically adjusting the load reduction amount through a Nash equilibrium model to minimize load loss and frequency overshoot, specifically includes:

[0130] The frequency change rate and voltage change rate are calculated in real time. When the frequency change rate is greater than the frequency change rate threshold and the voltage change rate is less than the voltage change rate threshold, a primary load reduction command is generated.

[0131] The positive sequence impedance mutation rate is calculated using PMU data. If the positive sequence impedance mutation rate is greater than the mutation rate threshold, the primary load reduction command is immediately blocked.

[0132] Based on a preset load weight matrix, an objective function that minimizes load loss and frequency overshoot is constructed using a Nash equilibrium model.

[0133] Based on the objective function and the initial load reduction command, the optimal load reduction amount for each load is obtained by solving the Lagrange multiplier method.

[0134] Based on the optimal load reduction, loads are tiered and removed starting with the highest-weighted loads.

[0135] In this embodiment, traditional power grid load shedding strategies often rely on fixed thresholds or single criteria, making it difficult to adapt to rapid frequency changes and dynamic load characteristics under complex operating conditions. This invention, by real-time monitoring of frequency and voltage change rates and positive sequence impedance mutation rates, combined with load importance classification and Nash equilibrium optimization, achieves dynamic adjustment of load shedding, thereby improving power grid frequency stability.

[0136] Specifically, the frequency change rate (Δf / Δt) and voltage change rate (ΔU / Δt) are calculated, and a sliding window method (window length of 1 second) is used to smooth data fluctuations. When the frequency change rate Δf / Δt > threshold A (e.g., 0.2Hz / s) and the voltage change rate ΔU / Δt < threshold B (e.g., -5V / s), a primary load shedding command is generated. For example, if Δf / Δt = 0.3Hz / s and ΔU / Δt = -3V / s, then primary load shedding is triggered.

[0137] Three-phase voltage and current data are acquired via PMU to calculate the positive sequence impedance (Z1 = U1 / I1). The positive sequence impedance mutation rate (ΔZ1 / Δt) is calculated using the first-order difference method. When the positive sequence impedance mutation rate ΔZ1 / Δt > threshold C (e.g., 10Ω / s), the primary load shedding command is immediately blocked. For example, if ΔZ1 / Δt = 15Ω / s, the primary load shedding is blocked to avoid accidental load shedding.

[0138] Based on the load importance classification (e.g., primary, secondary, and tertiary loads), a preset weight matrix W is set. For example, primary load weight = 1.0 (highest priority); secondary load weight = 0.5; tertiary load weight = 0.2. An objective function is set, aiming to minimize the weighted sum of load loss (L) and frequency overshoot (Δf_max), with the constraint that the total load reduction does not exceed the grid's shearable load capacity. The total load reduction constraint L≤L_max is transformed into a Lagrangian function, and the Lagrangian function is optimized using an iterative algorithm (e.g., gradient descent) to solve for the optimal load reduction for each load. Based on the load weight matrix, load shedding is performed tiered, starting with high-weight loads: tertiary loads are shedding first; if the frequency remains unstable, some secondary loads are shedding; primary loads are only shedding in extreme cases (e.g., hospitals, data centers).

[0139] In this embodiment, the Nash equilibrium model and the Lagrange multiplier method are used to dynamically adjust the load shedding, minimizing load loss. The positive sequence impedance mutation rate criterion effectively avoids erroneous load shedding and improves strategy reliability. Load importance classification ensures power supply to critical loads and safeguards social stability.

[0140] Furthermore, the objective function for minimizing load loss and frequency overshoot, constructed based on a preset load weight matrix and using a Nash equilibrium model, specifically includes:

[0141] The load loss term is constructed based on the load priority weights in the preset load weight matrix, and the objective function is formed by combining the frequency overshoot integral term.

[0142] Set power balance constraints and frequency security constraints, and determine the solution range of the objective function based on the power balance constraints and frequency security constraints;

[0143] The weighting coefficients of the load loss term and the frequency overshoot integral term are adjusted according to the frequency change rate.

[0144] In this embodiment, traditional power grid load shedding strategies, when optimizing load loss and frequency overshoot, typically rely on fixed weights or a single objective function, making it difficult to adapt to rapid frequency changes and dynamic load characteristics under complex operating conditions. This invention achieves multi-objective collaborative optimization and improves power grid frequency stability through a preset load weight matrix, a Nash equilibrium model, and dynamic weight adjustment.

[0145] Specifically, the objective function satisfies

[0146]

[0147]

[0148]

[0149] in, Represents the objective function value. Indicates the amount of load reduction. Represents load loss items. This represents the frequency overshoot integral term. and These represent the weighting coefficients for the load loss term and the frequency overshoot integral term, respectively. This represents the priority weight of the i-th type of load, where n represents the number of loads. This represents the proportion of the i-th type of load that was removed. This represents the frequency security constraint, and t represents the optimization time window. This indicates the reference value for the system's rated frequency. and This represents the two endpoints of the optimization time window.

[0150] Specifically, the power balance constraint satisfies

[0151]

[0152] in, This represents the reference power for the i-th type of load. This indicates the total amount of load reduction required by the system.

[0153] Specifically, frequency security constraints are satisfied.

[0154]

[0155] The constraints ensure that none of the participants (load nodes) can unilaterally improve their own costs under a given strategy.

[0156] Furthermore, the dynamic adjustment formula for adjusting the weighting coefficients of the load loss term and the frequency overshoot integral term based on the frequency change rate satisfies...

[0157]

[0158] in, Here, f represents the rate of change of frequency, f represents the frequency, and k represents the sensitivity coefficient. This represents the hyperbolic tangent function.

[0159] By combining power balance constraints and frequency safety constraints, the feasible solution space of the objective function, i.e., the range of values ​​for load reduction, is determined. The frequency change rate is collected in real time using a power management unit (PMU), and a dynamically adjusted threshold is set. When the frequency change rate exceeds the threshold, the weight β of the frequency overshoot integral term is increased, and the weight α of the load loss term is decreased; conversely, α is increased, and β is decreased. Based on the adjusted weight coefficients, the objective function is recalculated, and the optimal load reduction is solved.

[0160] Reference Figure 2An embodiment of the present invention provides a low-frequency load shedding system 2 based on real-time frequency tracking of zero crossing points. The system 2 specifically includes:

[0161] The signal processing module 201 is used to convert the high voltage signal of the power grid into a low voltage signal through a voltage transformer, generate a standard square wave signal through a zero-crossing comparator circuit, and use a high-speed comparator to construct a hysteresis comparator circuit to eliminate signal jitter and form a sensing signal.

[0162] The primary filtering module 202 is used to set a dynamic time window based on the power frequency characteristics of the sensing signal and to eliminate abnormal zero crossings that exceed the dynamic time window.

[0163] The secondary verification module 203 is used to calculate the total harmonic distortion rate of the sensor signal in real time using sliding discrete Fourier transform, and to determine whether to activate the dynamic weighted average algorithm by comparing the total harmonic distortion rate with a preset distortion rate threshold.

[0164] The third-level compensation module 204 is used to match the pre-stored harmonic feature library, establish the mapping relationship between harmonic modes and phase offset using a pre-trained deep generative adversarial network, and perform adaptive phase compensation for the pseudo-zero crossings of the sensing signal.

[0165] The load reduction prediction module 205 is used to input sensor signals, PMU phasor data and meteorological parameters into a three-dimensional convolutional neural network, calculate the cross-modal correlation matrix through a self-attention mechanism, combine the LSTM unit to predict the future frequency change trajectory, and output the optimal load reduction strategy.

[0166] The composite criterion module 206 is used to perform emergency load reduction based on the optimal load reduction strategy, combining the positive acceleration criterion and the reverse fault blocking, according to the load importance classification, and dynamically adjust the load reduction amount through the Nash equilibrium model to minimize load loss and frequency overshoot.

[0167] It is understandable that, such as Figure 1 The content of the low-frequency load shedding method embodiment based on real-time frequency tracking based on zero crossing point shown is applicable to the low-frequency load shedding system embodiment based on real-time frequency tracking based on zero crossing point. The specific functions implemented by the low-frequency load shedding system embodiment based on real-time frequency tracking based on zero crossing point are the same as those shown in the figure. Figure 1 The low-frequency load reduction method based on real-time frequency tracking at zero crossing points shown is the same as the embodiment described above, and achieves the same beneficial effects. Figure 1 The beneficial effects achieved by the low-frequency load reduction method based on real-time frequency tracking at zero crossing points shown in the embodiment are also the same.

[0168] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0169] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0170] Reference Figure 3 The present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, it implements the low-frequency load reduction method based on real-time frequency tracking of zero crossing as described in any of the above methods.

[0171] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0172] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0173] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0174] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the low-frequency load reduction method based on real-time frequency tracking of zero crossings as described in any of the above methods.

[0175] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0176] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0177] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0178] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0179] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A low frequency load shedding method based on zero-crossing point real-time frequency tracking, characterized in that, The method specifically comprises: The power grid high-voltage signal is converted into a low-voltage signal through a voltage transformer, a standard square wave signal is generated through a zero-crossing comparison circuit, and a high-speed comparator is used to build a hysteresis comparison circuit to eliminate signal jitter, thereby forming a sensing signal; According to the power frequency characteristics of the sensing signal, a dynamic time window is set, and abnormal zero-crossing points exceeding the dynamic time window are removed; The signal total harmonic distortion rate of the sensing signal is calculated in real time by using a sliding discrete Fourier transform, and whether the dynamic weighted average algorithm is activated is determined by comparing the signal total harmonic distortion rate with a preset distortion rate threshold value; A pre-stored harmonic feature library is matched, a pre-trained deep generative adversarial network is used to establish a mapping relationship between the harmonic mode and the phase shift, and the adaptive phase compensation of the pseudo zero-crossing point of the sensing signal is performed; The sensing signal, PMU phasor data and meteorological parameters are input into a three-dimensional convolutional neural network, a cross-modal correlation matrix is calculated through a self-attention mechanism, a future frequency change trajectory is predicted by combining an LSTM unit, and an optimal load shedding strategy is output; Based on the optimal load shedding strategy, combined with the forward acceleration criterion and the reverse fault blocking, the emergency load shedding is performed according to the importance classification of the load, and the load shedding amount is dynamically adjusted through the Nash equilibrium model to minimize the load loss and the frequency overshoot; Based on the optimal load shedding strategy, combined with the forward acceleration criterion and the reverse fault blocking, the emergency load shedding is performed according to the importance classification of the load, and the load shedding amount is dynamically adjusted through the Nash equilibrium model to minimize the load loss and the frequency overshoot, specifically comprising: The frequency change rate and the voltage change rate are calculated in real time, and when the frequency change rate is greater than the frequency change rate threshold value and the voltage change rate is less than the voltage change rate threshold value, a primary load shedding instruction is generated; The positive sequence impedance mutation rate is calculated through the PMU data, and if the positive sequence impedance mutation rate is greater than the mutation rate threshold value, the primary load shedding instruction is immediately blocked; Based on the preset load weight matrix, a target function of minimizing the load loss and the frequency overshoot is constructed through the Nash equilibrium model; Based on the target function and the primary load shedding instruction, the optimal load shedding amount of each load is obtained by using the Lagrange multiplier method; Based on the optimal load shedding amount, the loads are classified and cut off from high-weight loads.

2. The method of claim 1, wherein, According to the power frequency characteristics of the sensing signal, a dynamic time window is set, and abnormal zero-crossing points exceeding the dynamic time window are removed, specifically comprising: An initial time window is calculated according to the power frequency allowable range of the sensing signal, and a dynamic time window range is set by combining a time margin for compensating measurement error, and the time margin is dynamically adjusted by the real-time frequency change rate; Any two time stamps of adjacent zero-crossing points are collected, the time interval of the two time stamps is calculated, and if the time interval exceeds the dynamic time window range, the zero-crossing point is marked as an abnormal value and removed; The historical period data is stored by using a sliding window, the weighted average period is calculated by using an exponential decay weight formula, and the dynamic time window range is adjusted according to the weighted average period; If the continuous abnormality is a rapid frequency drop, the emergency control is triggered, otherwise the zero-crossing point sequence is reconstructed by interpolation with the weighted average period.

3. The method of claim 1, wherein, The signal total harmonic distortion rate of the sensing signal is calculated in real time by using the sliding discrete Fourier transform, and whether to activate the dynamic weighted average algorithm is determined by comparing the signal total harmonic distortion rate with a preset distortion rate threshold, and specifically includes: The sensing signal is sampled according to a preset sampling frequency, and k-th harmonic components are calculated by using a recursive formula, wherein the k-th harmonic components include a fundamental component and odd harmonic components; Based on the k-th harmonic components, the signal total harmonic distortion rate of the current window is calculated by using the sliding discrete Fourier transform; The signal total harmonic distortion rate is compared with the preset distortion rate threshold, and if the signal total harmonic distortion rate is greater than the preset distortion rate threshold, the weights are adaptively distributed according to the harmonic energy; Based on the adaptively distributed weights, the fundamental component is weighted and averaged to obtain a weighted fundamental component, the zero-crossing point sequence is reconstructed by using the amplitude of the weighted fundamental component, and the corrected power frequency is output.

4. The method of claim 3, wherein, The matching pre-stored harmonic feature library is used to establish a mapping relationship between the harmonic mode and the phase shift by using a pre-trained deep generative adversarial network, and the adaptive phase compensation is performed on the pseudo zero-crossing point of the sensing signal, and specifically includes: The amplitude distribution of each odd harmonic component is counted by using historical fault data, the harmonic initial phase angle and the harmonic frequency of each odd harmonic component are collected, the pre-stored typical harmonic mode parameter set is established, and the harmonic feature library is formed; The deep generative adversarial network is constructed, the generator and the discriminator of the deep generative adversarial network are trained based on the harmonic feature library, and the adversarial loss function is used; The current harmonic energy ratio is calculated, and if the current harmonic energy ratio is greater than a preset energy ratio threshold, it is determined that there is a pseudo zero-crossing point, and the current harmonic parameters are input into the deep generative adversarial network to output the phase compensation amount.

5. The method of claim 1, wherein, The sensing signal, the PMU phasor data and the weather parameters are input into the three-dimensional convolutional neural network, the cross-modal correlation matrix is calculated by using the self-attention mechanism, the future frequency change trajectory is predicted by combining the LSTM unit, and the optimal load shedding strategy is output, and specifically includes: The PMU phasor data and the weather parameters are collected, the sensing signal, the PMU phasor data and the weather parameters are aligned in time, and then are spliced into a three-dimensional tensor along the channel dimension; The three-dimensional convolutional neural network is used to extract features of the three-dimensional tensor, and high-order feature maps are generated by using the ReLU activation function; The high-order feature maps are linearly projected to obtain a query matrix, a key matrix and a value matrix, the cross-modal correlation matrix is calculated based on the query matrix, the key matrix and the value matrix, and is weighted and fused to obtain a weighted fusion feature; The weighted fusion feature is input into the LSTM unit, the hidden state is updated by using the state equation, and the frequency prediction value in a preset future time period is output.

6. The method of claim 5, wherein, The preset load weight matrix is used to construct a target function for minimizing the load loss and the frequency overshoot by using the Nash equilibrium model, and specifically includes: A load loss term is constructed according to the load priority weight in the preset load weight matrix, and a frequency overshoot integral term is combined to form the target function; The power balance constraint and the frequency safety constraint are set, and the solution range of the target function is determined according to the power balance constraint and the frequency safety constraint; The weight coefficients of the load loss term and the frequency overshoot integral term are adjusted according to the frequency change rate.

7. A low frequency load shedding system based on zero-crossing point real-time frequency tracking, characterized by, The system specifically includes: The signal processing module is configured to convert a high-voltage signal of a power grid into a low-voltage signal through a voltage transformer, generate a standard square wave signal through a zero-crossing comparison circuit, and eliminate signal jitter by using a high-speed comparator to construct a hysteresis comparison circuit, thereby forming a sensing signal. The primary filtering module is configured to set a dynamic time window according to a power frequency characteristic of the sensing signal, and eliminate abnormal zero-crossing points that are outside the dynamic time window. The secondary verification module is configured to calculate a total harmonic distortion rate of the sensing signal in real time by using a sliding discrete Fourier transform, and determine whether to activate a dynamic weighted average algorithm by comparing the total harmonic distortion rate with a preset distortion rate threshold. The tertiary compensation module is configured to match a pre-stored harmonic feature library, establish a mapping relationship between a harmonic mode and a phase shift by using a pre-trained deep generative adversarial network, and perform adaptive phase compensation on pseudo zero-crossing points of the sensing signal. The load reduction prediction module is configured to input the sensing signal, PMU phasor data and meteorological parameters into a three-dimensional convolutional neural network, calculate a cross-modal correlation matrix by using a self-attention mechanism, predict a future frequency change trajectory by combining an LSTM unit, and output an optimal load reduction strategy. The composite criterion module is configured to execute emergency load reduction according to a load importance classification based on the optimal load reduction strategy, a forward acceleration criterion and a reverse fault lockout, and dynamically adjust a load reduction amount by using a Nash equilibrium model to minimize load loss and frequency overshoot. The composite criterion module is configured to execute emergency load reduction according to a load importance classification based on the optimal load reduction strategy, a forward acceleration criterion and a reverse fault lockout, and dynamically adjust a load reduction amount by using a Nash equilibrium model to minimize load loss and frequency overshoot, and specifically includes: calculating a frequency change rate and a voltage change rate in real time, and generating a primary load reduction instruction when the frequency change rate is greater than a frequency change rate threshold and the voltage change rate is less than a voltage change rate threshold; calculating a positive sequence impedance mutation rate based on PMU data, and immediately locking the primary load reduction instruction when the positive sequence impedance mutation rate is greater than a mutation rate threshold; constructing an objective function for minimizing load loss and frequency overshoot based on a preset load weight matrix by using a Nash equilibrium model; solving the optimal load reduction amount of each load based on the objective function and the primary load reduction instruction by using a Lagrange multiplier method; and classifying and cutting off each load from a high-weight load based on the optimal load reduction amount.

8. A computer device, comprising: The memory and the processor and the computer program stored in the memory, when the computer program is executed on the processor, realize the low-frequency load shedding method based on real-time frequency tracking of zero-crossing points as claimed in any one of claims 1 to 6. The computer program is stored on the processor, and when the computer program is executed on the processor, the low-frequency load shedding method based on real-time frequency tracking of zero-crossing points as claimed in any one of claims 1 to 6 is realized.

9. A computer-readable storage medium, characterized in that, ​

Images