A power distribution network carrying capacity harmonic self-adaptive derating method and system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIJING UNIV
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-21
Smart Images
Figure CN122437014A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power distribution network technology, and specifically to an adaptive derating method and system for power distribution network carrying capacity harmonics. Background Technology
[0002] As a key link connecting the power grid and end users, the carrying capacity (i.e., open capacity) of the distribution substation directly determines the access capability of new loads such as distributed photovoltaics and charging piles.
[0003] In actual operation, the large-scale integration of power electronic equipment has led to increasingly serious harmonic pollution. The additional transformer losses, accelerated temperature rise, and malfunctions of protection devices caused by harmonic currents significantly impact the safe operation of distribution substations. Traditional load-bearing capacity assessments typically rely solely on fundamental current or simple harmonic coefficients for calculation, failing to reflect the differentiated thermal effects and safety risks posed by harmonics from different frequency bands and sources.
[0004] Existing harmonic mitigation and derating control methods mostly employ fixed threshold triggering. For example, when the Total Harmonic Distortion (THD) exceeds a preset threshold, the available capacity is reduced by a fixed percentage. This approach has two limitations: First, THD, as a global statistical indicator, cannot distinguish the different degrees of harm to equipment caused by different harmonic bands such as low-order, zero-sequence, and high-frequency harmonics, leading to excessive or insufficient derating. Second, the derating triggering and recovery mechanisms lack continuous sensing and adaptive response capabilities to the time-varying characteristics of harmonics, easily causing the control strategy to switch frequently near the threshold, affecting power supply continuity and equipment lifespan. Summary of the Invention
[0005] To address the aforementioned problems in the existing technology, this invention provides an adaptive derating method and system for harmonics in the carrying capacity of distribution networks. The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides an adaptive derating method for harmonics in the carrying capacity of a distribution network, comprising: The current waveform timing vector is obtained from multiple consecutive instantaneous current values of the distribution station area; The current waveform time-series vector is input into a pre-trained lightweight one-dimensional convolutional neural network to obtain the initial THD estimate in an end-to-end manner. Based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, the harmonic dominance type of the current harmonic energy is determined, and the initial THD estimate is weighted and compensated according to the harmonic dominance type to obtain the corrected THD evaluation value. Based on the continuously revised THD assessment values, the available capacity of the distribution radio station is dynamically adjusted through derating and gradual escalation control.
[0006] In one embodiment of the present invention, a current waveform timing vector is obtained based on multiple continuous instantaneous current values of a distribution substation, including: In the distribution station area, the instantaneous current values of K power frequency cycles are continuously collected at a preset sampling frequency to form the original current waveform sequence; The original current waveform sequence is subjected to DC bias removal processing to obtain a zero-mean current sequence; The zero-mean current sequence is linearly normalized to obtain the current waveform time sequence vector.
[0007] In one embodiment of the present invention, the lightweight one-dimensional convolutional neural network includes a feature extraction module, an attention module, and a regression output module connected in sequence; The current waveform time-series vector is input into a pre-trained lightweight one-dimensional convolutional neural network to obtain an initial THD estimate in an end-to-end manner, including: The current waveform timing vector is input to the feature extraction module to extract waveform distortion features and output a first feature map including multiple channels. The first feature map is input into the attention module to recalibrate the weights of each channel of the first feature map and output a second feature map after channel weighting. The second feature map is input to the regression output module to perform feature integration on the second feature map and output the initial estimate of THD.
[0008] In one embodiment of the present invention, based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, the dominant harmonic type of the current harmonic energy is determined, and the initial THD estimate is weighted and compensated according to the dominant type to obtain a corrected THD evaluation value, including: Perform a fast Fourier transform on the current waveform time-series vector to obtain a frequency domain complex sequence; Based on the modulus of the nth complex number in the frequency domain complex number sequence, the normalized amplitude of the nth harmonic is obtained, where 2≤n≤Nh, and Nh is the highest frequency of the harmonic analysis; The proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy are determined based on the normalized amplitudes of the low-order harmonic number, the zero-sequence harmonic number, and the high-frequency harmonic number, respectively, wherein the low-order harmonic number is less than the high-frequency harmonic number. Based on the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy, the dominant harmonic type of the current harmonic energy is determined. Based on the compensation weight corresponding to the dominant harmonic type, the initial THD estimate is weighted and compensated, and the corrected THD evaluation value is output.
[0009] In one embodiment of the present invention, the harmonic dominance type of the current harmonic energy is determined based on the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy, including: Determine the maximum and minimum harmonic energy percentages among the low-order harmonic energy percentage, the zero-sequence harmonic energy percentage, and the high-frequency harmonic energy percentage. Determine whether the difference between the maximum value of the harmonic energy proportion and the minimum value of the harmonic energy proportion is less than a preset threshold. If yes, determine that the harmonic dominance type is a mixed harmonic dominance type. If no, determine the harmonic dominance type based on the maximum value of the harmonic energy proportion.
[0010] In one embodiment of the present invention, based on the continuously corrected THD assessment values, the available capacity of the distribution radio area is dynamically adjusted through a control method of derating and gradual recovery, including: Obtain the current available capacity of the power distribution area; Determine if there exists a continuous N trig If all corrected THD estimates are greater than the harmonic distortion triggering threshold, then a bearing capacity derating state is triggered, and the most recent corrected THD estimate that triggered the bearing capacity derating state is taken as the THD estimate. cf ; Based on the THD estimate, THD cf The dynamic attenuation coefficient is determined based on the harmonic distortion trigger threshold, and the derated openable capacity is determined based on the dynamic attenuation coefficient and the current openable capacity, wherein the dynamic attenuation coefficient is used to map the degree of harmonic pollution to the reduction ratio of the openable capacity. Under the load-bearing capacity derating state, determine whether there are consecutive N... rec If all corrected THD estimates are less than or equal to the recovery threshold, then a gradual recovery state is triggered, wherein the harmonic distortion trigger threshold is greater than the recovery threshold. In the gradual recovery state, the current dynamic attenuation coefficient is updated based on the gradual increase step size to obtain the updated dynamic attenuation coefficient, and the capacity that can be opened after derating is calculated based on the updated dynamic attenuation coefficient until the updated dynamic attenuation coefficient is 1, at which point the capacity that can be opened is fully restored.
[0011] In one embodiment of the present invention, based on the THD estimate, THD... cfThe dynamic attenuation coefficient is determined based on the harmonic distortion trigger threshold, and the derated available capacity is determined based on the dynamic attenuation coefficient and the current available capacity, including: Based on the THD estimate, THD cf The initial attenuation ratio is determined by the harmonic distortion triggering threshold, wherein the initial attenuation ratio varies with the estimated THD value. cf It decreases linearly with the increase; The larger of the initial attenuation ratio and the minimum attenuation coefficient is selected as the dynamic attenuation coefficient. Based on the dynamic attenuation coefficient and the current available capacity, the initial derated available capacity is obtained; Determine the relationship between the initial reduced open capacity and the minimum open capacity. If the initial reduced open capacity is greater than or equal to the minimum open capacity, then the initial reduced open capacity is taken as the reduced open capacity. If the initial reduced open capacity is less than the minimum open capacity, then the minimum open capacity is taken as the reduced open capacity.
[0012] In one embodiment of the present invention, updating the current dynamic decay coefficient based on the gradual increase step size to obtain the updated dynamic decay coefficient includes: Every Ts sampling cycles, the current dynamic attenuation coefficient is increased by one of the gradual increase steps to obtain the updated dynamic attenuation coefficient.
[0013] In one embodiment of the present invention, in the gradual recovery state, if N consecutive N events occur again before the current dynamic attenuation coefficient recovers to 1... trig If the corrected THD estimate is greater than the harmonic distortion trigger threshold, then the system exits the gradual recovery state and enters the load-bearing capacity derating state to perform derating calculations.
[0014] Secondly, the present invention provides an adaptive derating system for the carrying capacity harmonics of a distribution network, comprising: The timing vector generation module is used to obtain the current waveform timing vector based on multiple continuous instantaneous current values of the distribution substation. The THD initial estimate generation module is used to input the current waveform time-series vector into a pre-trained lightweight one-dimensional convolutional neural network to obtain the THD initial estimate in an end-to-end manner. The THD correction module is used to determine the harmonic dominance type of the current harmonic energy based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, and to perform weighted compensation on the initial THD estimate according to the harmonic dominance type to obtain the corrected THD evaluation value. The openable capacity adjustment module is used to dynamically adjust the openable capacity of the distribution radio station based on the continuously corrected THD assessment value, through derating and gradual increase control.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides an adaptive derating method for harmonic load capacity in distribution networks. It employs a lightweight one-dimensional convolutional neural network to directly estimate the end-to-end THD of the current waveform time-series vector, replacing the traditional method that relies on high-precision FFT hardware and complex calculations. This enables high-frequency, real-time harmonic monitoring on low-cost embedded terminals, reducing hardware and deployment costs. Furthermore, this invention introduces a harmonic type determination and differentiated weighted compensation method based on frequency domain analysis. It corrects the single THD index into an assessment value that more accurately reflects the different physical hazard levels of low-order, zero-sequence, and high-frequency harmonics, solving the problem of excessive or insufficient derating in traditional assessments. Based on continuously corrected THD estimates, this invention dynamically adjusts the available capacity of distribution substations through derating processing and a gradual escalation control method. This allows the available capacity of distribution substations to smoothly and adaptively adjust with the harmonic state, ensuring power supply safety when harmonics exceed limits and maximizing capacity utilization after harmonic improvement. This overcomes the shortcomings of traditional fixed threshold control methods, such as response lag and threshold oscillation, forming an intelligent, economical, and reliable harmonic load capacity management solution.
[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating an adaptive derating method for harmonic load-bearing capacity of a distribution network provided by the present invention. Figure 2 This is a schematic diagram of an adaptive derating system for the carrying capacity harmonics of a power distribution network provided by the present invention. Detailed Implementation
[0018] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0019] Example 1 Distribution substations typically lack expensive equipment such as high-precision harmonic analyzers. While traditional FFT (Fast Fourier Transform)-based THD calculation methods offer high accuracy, they require complete spectral analysis for each sampling period, consuming significant computational resources and demanding high sampling synchronization, making continuous, real-time monitoring (e.g., every 5 minutes) on low-cost smart terminals difficult. However, FFT still holds value in non-real-time, low-frequency harmonic energy distribution analysis—for example, performing an FFT every few periods or only when derating is triggered to identify dominant harmonic types (low-order / zero-sequence / high-frequency) without requiring high real-time performance. Therefore, a reasonable strategy is to use a lightweight neural network for continuous high-frequency THD estimation, while FFT serves as an auxiliary tool, providing low-frequency harmonic type information; the two work collaboratively. However, the lack of such a collaborative scheme of "rapid estimation + periodic classification" in the existing technology has led to the common problems of "harmonic state cannot be perceived in fine granularity, the degree of harm cannot be differentiated and assessed, and the derating decision cannot be adaptively closed-loop" in the load-bearing capacity regulation. There is an urgent need for a method that can efficiently estimate THD from the original current waveform, make differentiated corrections based on the characteristics of harmonic frequency band distribution, and realize dynamic closed-loop derating control.
[0020] Therefore, this invention provides an adaptive derating method for harmonics in the carrying capacity of distribution networks. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart illustrating an adaptive derating method for harmonics in the carrying capacity of a distribution network provided by this invention. The adaptive derating method provided by this invention includes: Step 1: Obtain the current waveform timing vector based on multiple continuous instantaneous current values of the distribution station area.
[0021] Specifically, a current transformer and sampling device are installed at the low-voltage side incoming line of the transformer in the distribution substation to continuously collect multiple instantaneous current values, forming a raw current waveform sequence with timestamps. This raw current waveform sequence is then preprocessed to obtain a standardized, dimensionless current waveform time sequence vector.
[0022] In an optional embodiment, step 1 may include: Step 1.1: In the distribution station area, continuously collect the instantaneous current values of K power frequency cycles at a preset sampling frequency to form the original current waveform sequence.
[0023] Specifically, a preset sampling frequency is set to collect M current sampling points for each power frequency cycle, where M is a preset positive integer. A total of K power frequency cycles are collected continuously, so a total of K×M current sampling points are used as a current waveform sample. Thus, a current waveform sample can be generated at fixed time intervals and timestamped. Therefore, a current waveform sample is an original current waveform sequence.
[0024] For example, taking a 400kVA distribution substation as an example, a standard current transformer (CT, transformation ratio 1000A / 5A) and a smart fusion terminal (as a sampling device) are installed at the low-voltage side incoming line of the transformer. The number of sampling points per power frequency cycle is set to M=128 (power frequency 50Hz, corresponding to a sampling rate of 6400Hz), and the number of consecutive power frequency cycles collected is K=10. Therefore, the total number of sampling points for each current waveform sample is 1280 points, and the sampling interval (sample generation cycle) is set to 5 minutes. The acquisition process: The smart fusion terminal starts sampling every 5 minutes, continuously collecting current waveforms for 10 power frequency cycles, obtaining 1280 consecutive instantaneous current values as a single current waveform sample, and recording the start timestamp of this sample.
[0025] Step 1.2: Perform DC bias removal processing on the original current waveform sequence to obtain a zero-mean current sequence.
[0026] Specifically, in the actual current data collected, due to sensor errors or circuit characteristics, the waveform often does not fluctuate on the zero axis, but rather carries a certain DC component. Therefore, for the K×M current sampling points in the original current waveform sequence, their average value I is calculated. ave Subtract the average value I from each current sampling point in the original current waveform sequence. ave This yields a zero-mean current sequence. DC bias removal eliminates the DC component, ensuring waveform symmetry about the zero axis, thus preventing deviations caused by the DC component in subsequent processes.
[0027] Step 1.3: Perform linear normalization on the zero-mean current sequence to obtain the current waveform time sequence vector.
[0028] Specifically, the zero-mean current sequence is mapped to the [-1,1] interval through a linear transformation to obtain the current waveform time sequence vector.
[0029] Step 2: Input the current waveform time-series vector into a pre-trained lightweight one-dimensional convolutional neural network to obtain the initial THD estimate in an end-to-end manner.
[0030] Specifically, in this embodiment, based on the current waveform samples (current waveform time-series vectors of length K×M) output in step 1, a lightweight one-dimensional convolutional neural network (1D-CNN + attention) is constructed to achieve single-value estimation of total harmonic distortion (THD). The lightweight one-dimensional convolutional neural network extracts waveform distortion features through shallow convolutions and enhances the response of harmonic-sensitive frequency bands by combining a channel attention mechanism, completing end-to-end THD regression with low parameter count, thereby outputting an initial THD estimate.
[0031] Here, the tensor shape input to the lightweight one-dimensional convolutional neural network is (batch, L, 1), which represents the batch size, time sequence length, and number of channels (single channel).
[0032] Optionally, the lightweight one-dimensional convolutional neural network includes a feature extraction module, an attention module, and a regression output module connected in sequence.
[0033] In an optional embodiment, step 2 includes: Step 2.1: Input the current waveform timing vector into the feature extraction module to extract waveform distortion features and output the first feature map including multiple channels.
[0034] Specifically, in the feature extraction module, the current waveform time-series vector is first converted into a tensor of shape (batch, L, 1), and then input into a feature extraction network composed of multiple one-dimensional convolutional layers. Finally, the first feature map is output, which is a two-dimensional structure containing multiple channels, such as (batch, L / 8, 64). Each channel corresponds to a learned distortion feature, thereby realizing a distributed representation of the distortion features of the original waveform.
[0035] Furthermore, the feature extraction module includes an input layer, a normalization layer, convolutional block 1, convolutional block 2, and convolutional block 3. The normalization layer is used to normalize the input tensor along the temporal length dimension (L); convolutional blocks 1, 2, and 3 all contain Conv1D (convolution), ReLU (activation), and MaxPool (maximum pooling). Convolutional block 1 has a kernel size of 15, a stride of 1, a filter size of 16, a pooling window size of 2, a stride of 2, and padding='same'; convolutional block 2 has a kernel size of 7, a stride of 1, a filter size of 32, a pooling window size of 2, a stride of 2, and padding='same'; convolutional block 3 has a kernel size of 5, a stride of 1, a filter size of 64, a pooling window size of 2, a stride of 2, and padding='same'. After passing through convolution block 1, the tensor length becomes L / 2 and the number of channels becomes 16; after passing through convolution block 2, the tensor length becomes L / 4 and the number of channels becomes 32; after passing through convolution block 3, the tensor length becomes L / 8 and the number of channels becomes 64.
[0036] Step 2.2: Input the first feature map into the attention module to recalibrate the weights of each channel of the first feature map and output the second feature map after channel weighting.
[0037] Specifically, the first feature map is first fed into the attention module and compressed in the temporal dimension. Global average pooling is used to aggregate the entire time series information of each channel into a scalar, thus obtaining a vector representing the global saliency of each channel. Next, this vector is input into a fully connected network (including fully connected layers (16) and (64), with ReLU activation in between). This network learns the nonlinear dependencies between channels and outputs a weight vector with values ranging from 0 to 1 through the Sigmoid activation function, where each weight value corresponds to the importance score of a channel. Finally, the weight vector is multiplied with the original first feature map channel by channel to achieve channel weight recalibration, thereby outputting a second feature map with unchanged shape but reweighted information. This enhances the response of the feature channels related to the current harmonic distortion and suppresses the response of irrelevant or noisy channels.
[0038] Step 2.3: Input the second feature map into the regression output module to perform feature integration on the second feature map and output the initial THD estimate.
[0039] Specifically, the second feature map is input to the regression output module. The second feature map is compressed into a 64-dimensional feature vector through a global pooling layer (GlobalAvgPool1D). Then, the 64-dimensional feature vector is linearly transformed through a fully connected layer 1 (64 neurons) to extract a higher-level feature representation. After passing through the ReLU activation function and Dropout operation (Dropout rate of 0.3), it is linearly activated through the output layer (Dense+Linear) to output the initial THD estimate.
[0040] The initial THD estimate of the lightweight one-dimensional convolutional neural network output in this embodiment is a scalar value, representing the estimated total harmonic distortion rate corresponding to the input current waveform sample. It is typically mapped to [0%, 100%]. The higher the estimated THD value, the more severe the waveform distortion and the greater the degree of harmonic pollution. As shown in Table 1, Table 1 shows the specific structure of a lightweight one-dimensional convolutional neural network provided in this embodiment.
[0041] The lightweight one-dimensional convolutional neural network in this step is used to perform fast and lightweight THD estimation on the current waveform of each sampling period. The initial THD estimate output is denoted as THD. netestThe initial THD estimate will serve as the basis for subsequent derating judgments. It should be noted that the lightweight one-dimensional convolutional neural network in this embodiment does not output harmonic frequency band distribution information; the determination of harmonic types is completed in the next step through FFT analysis. The two form a complementary relationship of "rapid estimation + periodic classification," rather than redundant calculation.
[0042] Furthermore, this embodiment also provides a lightweight training method for a one-dimensional convolutional neural network. To address the difficulties in acquiring harmonic data from actual distribution radio stations and the high cost of acquiring high-precision labels, this embodiment employs a hybrid data construction method combining simulated and measured data to improve the model's generalization ability and engineering applicability, as detailed below: 1. Data source: (1) Simulation data construction: Establish a typical distribution area circuit model, including typical electrical equipment such as linear load, nonlinear load, photovoltaic inverter, frequency converter and switching power supply; by setting different harmonic source types, harmonic order, phase and amplitude ratio, the fundamental current is randomly generated, and harmonic components with different total harmonic distortion rates (THD range of 1%~50%) are superimposed to construct multi-condition current waveform data.
[0043] Each simulation outputs a current waveform sample (time points of length K×M), and its corresponding true THD value is calculated using frequency domain analysis methods, serving as a supervised learning label. Approximately 100,000 samples are generated in total for initial model training.
[0044] (2) Measured data calibration (optional implementation): Deploy sampling devices in actual distribution substations to collect real current waveform data, and use fast Fourier transform to calculate the corresponding THD value. Fine-tune or calibrate the model to correct the distribution deviation between the simulation data and the actual power grid, and improve the prediction accuracy of the model in the real operating environment.
[0045] (3) Data consistency processing: The same sampling frequency, data length and preprocessing process (including DC bias removal and normalization) are used for both simulation data and measured data to ensure the consistency of model input distribution, thereby improving the stability of model training and the actual deployment effect.
[0046] In addition, simulation data is used to cover a variety of extreme operating conditions and long-tailed harmonic distributions, while measured data is used to perform distribution calibration on the model. The combination of the two can effectively improve the robustness and generalization ability of the model.
[0047] 2. Data Augmentation: Based on the original data, in order to further improve the robustness and noise resistance of the model, at least one of the following data augmentation processes is performed on the current waveform samples to expand the data scale to approximately 300,000 samples.
[0048] (1) Amplitude scaling: The current waveform is linearly scaled by 0.9 to 1.1 times to simulate load fluctuations; (2) Time-domain cyclic shift: Periodically shift the waveform to simulate different initial phase conditions; (3) Noise injection: Gaussian white noise (signal-to-noise ratio of 30 dB) is superimposed to simulate the actual measurement noise environment.
[0049] The above enhancement methods improve the model's adaptability to different operating conditions and noise interference.
[0050] 3. Dataset partitioning: The training set accounts for 70% (210,000 samples), the validation set accounts for 15% (45,000 samples), and the test set accounts for 15% (45,000 samples).
[0051] 4. Training objective and loss function: The mean square error (MSE) is used as the loss function to minimize the error between the THD estimate output by the model and the true THD, so that the model can accurately regress the total harmonic distortion rate from the original current waveform.
[0052] 5. Optimizer and learning rate scheduling settings: Optimizer: Adam, initial learning rate lr=0.001, other parameters are default values (β1 =0.9, β2 =0.999).
[0053] Learning rate scheduling: Cosine annealing or ReduceLROnPlateau strategy is used. If the validation loss does not decrease for 5 consecutive epochs, the learning rate is multiplied by 0.5, with a minimum reduction to 1×10. -6 .
[0054] 6. Training process: The batch size is set to 128, and the maximum number of training epochs is 100. In each iteration, the model first calculates the loss between the predicted and actual values through forward propagation, then uses the backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters, and finally uses the Adam optimizer to update the model parameters based on these gradients to gradually reduce the loss and improve model performance. Furthermore, training stops and returns to the optimal state if the loss does not decrease for 10 consecutive epochs.
[0055] 7. Convergence judgment: During model training, the model is considered to have converged when the losses on both the training and validation sets tend to plateau and the loss on the validation set no longer decreases significantly. Simultaneously, a specific convergence quantification standard is set: the mean squared error (MSE) of the final validation set is less than 0.2 (where MSE is the square of the percentage error of THD, used to measure the deviation between predicted THD and true THD). After model training converges, the model is evaluated on the test set, requiring the model's mean absolute error (MAE) to be less than 2%, meaning the model's average absolute deviation from predicted THD must be less than 2 percentage points. This ensures the model has reliable prediction accuracy in real-world applications.
[0056] Step 3: Based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, determine the harmonic dominance type of the current harmonic energy, and perform weighted compensation on the initial THD estimate according to the harmonic dominance type to obtain the corrected THD evaluation value.
[0057] Specifically, firstly, the frequency domain complex sequence of the current waveform time sequence vector is obtained by frequency domain analysis, and then the amplitude of each harmonic can be calculated. Then, by comparing the proportion of harmonic energy in a specific frequency band to the total harmonic energy, the frequency band with the largest proportion is determined as the dominant harmonic type of the current harmonic energy. Finally, based on the determined dominant harmonic type, a preset weighting coefficient that reflects the difference in its physical hazards is selected to weight and compensate the initial THD estimate obtained in step 2, thereby obtaining a corrected THD assessment value that is more in line with the actual thermal effects and safety risks of the equipment.
[0058] In an optional embodiment, step 3 includes: Step 3.1: Perform a Fast Fourier Transform (FFT) on the current waveform time sequence vector to obtain a frequency domain complex sequence.
[0059] Specifically, the time-domain current waveform timing vector obtained in step 1 is subjected to a discrete fast Fourier transform (FFT). This transform does not need to be performed for every sampling period, but is performed periodically (e.g., once every 10 sampling periods) or triggered on demand (e.g., only when the load capacity derating state is triggered for the first time), thereby obtaining a frequency domain complex sequence, where each complex spectrum contains the amplitude and phase of each frequency component.
[0060] Step 3.2: Based on the modulus of the nth complex number in the frequency domain complex number sequence, obtain the normalized amplitude of the nth harmonic, where 2≤n≤Nh, and Nh is the highest frequency of the harmonic analysis.
[0061] Here, the normalized amplitude H of the nth harmonic (corresponding to frequency n*f0, where f0 is the power frequency) n Calculate using the following formula: in, The nth complex number in the frequency domain complex sequence The modulus of N is the number of sampling points.
[0062] In addition, to ensure calculation consistency and avoid aliasing, the highest frequency Nh in the harmonic analysis is determined by the sampling frequency f. s Determined by the power frequency f0, the formula for calculating the highest frequency number Nh in harmonic analysis is: With sampling frequency f s Taking a frequency of 6400 Hz and a power frequency of f0 = 50 Hz as an example, Nh = 64, meaning that the 2nd to 64th harmonics need to be analyzed.
[0063] Step 3.3: Determine the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy based on the normalized amplitudes of the low-order harmonic order, the zero-sequence harmonic order, and the high-frequency harmonic order, respectively, wherein the low-order harmonic order is less than the high-frequency harmonic order.
[0064] Specifically, to characterize the spectral distribution characteristics of different harmonic sources, three energy proportion features are defined: the energy proportion of low-order harmonics, the energy proportion of zero-sequence harmonics, and the energy proportion of high-frequency harmonics.
[0065] Here, the proportion of low-order harmonic energy is determined by the normalized amplitude of the low-order harmonic order, which is preferably n = 2 to 9. The formula for calculating the proportion of low-order harmonic energy is as follows: in, The percentage of low-order harmonic energy is given by e, which is a smoothing factor of 0.000001 to prevent the denominator from being 0.
[0066] Here, the proportion of zero-sequence harmonic energy is determined by the normalized amplitude of the zero-sequence harmonic order. The preferred zero-sequence harmonic order is n=3 or a multiple thereof. In this embodiment, to reduce the computational load, only the harmonic with order 3 is used. The formula for calculating the proportion of zero-sequence harmonic energy is as follows: in, This represents the proportion of zero-sequence harmonic energy.
[0067] Here, the proportion of high-frequency harmonic energy is determined by the normalized amplitude of the high-frequency harmonic order, preferably n≥13. The formula for calculating the proportion of high-frequency harmonic energy is as follows: in, This represents the proportion of high-frequency harmonic energy.
[0068] Step 3.4: Based on the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy, determine the dominant harmonic type of the current harmonic energy.
[0069] Step 3.41: Determine the maximum and minimum harmonic energy percentages among the low-order harmonic energy percentage, zero-sequence harmonic energy percentage, and high-frequency harmonic energy percentage.
[0070] Specifically, first, construct the feature vector: F=(Rlow, Rzero, Rhigh), define the criteria D1=argmax(F) and D2=argmin(F), the argmax function returns the index position of the maximum value and the argmin function returns the index position of the minimum value, take max(F) as the maximum value of the harmonic energy ratio and take min(F) as the minimum value of the harmonic energy ratio.
[0071] Step 3.42: Determine whether the difference between the maximum and minimum harmonic energy proportions is less than a preset threshold. If yes, determine that the harmonic dominance type is a mixed harmonic dominance type. If no, determine the harmonic dominance type based on the maximum harmonic energy proportion.
[0072] Specifically, the difference between the maximum and minimum harmonic energy percentages (i.e., max(F) - min(F)) is calculated. If max(F) - min(F) < δ, the harmonic dominance type is determined to be a mixed harmonic dominance type; otherwise, the harmonic corresponding to max(F) is taken as the harmonic dominance type. That is, the value of D1 is 0, 1, or 2, corresponding to low-frequency, zero-sequence, and high-frequency dominance, respectively. Therefore, if D1 = 0, the harmonic dominance type is low-order harmonic dominance; if D1 = 2, the harmonic dominance type is zero-sequence harmonic dominance; and if D1 = 2, the harmonic dominance type is high-frequency harmonic dominance. For example, δ = 0.1; this embodiment does not specifically limit this value.
[0073] Different power electronic devices have different harmonic generation mechanisms, resulting in a stable distribution of harmonic energy across different frequency bands. Low-order harmonic-dominated types originate from rectification and PWM modulation, commonly found in rectifiers, PWM inverters operating at low carrier ratios, and some photovoltaic inverters. Zero-sequence harmonic-dominated types originate from unbalanced loads, primarily three-phase unbalanced loads (such as single-phase charging piles and welding machines). The third harmonic of the zero-sequence order will superimpose current at the transformer neutral point, leading to neutral point overheating and metering errors. High-frequency harmonic-dominated types originate from high-frequency switching behavior, typically found in high-frequency switching power supplies, high-frequency inverters, and the high-switching frequency modes of some photovoltaic inverters. Although high-frequency harmonics typically have smaller amplitudes, they easily induce series resonance and communication interference. This embodiment achieves reliable classification without relying on complex models through a simple comparison of frequency band energy proportions.
[0074] Step 3.5: Based on the compensation weight corresponding to the dominant harmonic type, perform weighted compensation on the initial THD estimate and output the corrected THD evaluation value.
[0075] Here, the corrected THD assessment value is expressed as: in, This is the corrected THD assessment value. σ is the initial estimate of THD, and σ is the compensation weight.
[0076] Furthermore, for the low-order harmonic-dominated type: σ = k1, because the heat loss of motors and transformers caused by low-order harmonics is significant, the THD should be equivalently amplified to trigger a more severe derating; therefore, k1 = 1.5~1.8 is preferred. For the zero-sequence harmonic-dominated type: σ = k2, because the risk of neutral point overheating caused by zero-sequence harmonics is high, the THD needs to be moderately amplified; therefore, k2 = 1.2~1.4 is preferred. For the high-frequency harmonic-dominated type: σ = k3, because the skin effect of high-frequency harmonics leads to an increase in conductor resistance, but its energy is usually low and has little impact on core losses; the original THD can be maintained or slightly relaxed; therefore, k3 = 0.9~1.1 is preferred. For the mixed harmonic-dominated type: σ = 1.
[0077] Step 4: Based on the continuously corrected THD assessment value, dynamically adjust the available capacity of the distribution radio area through de-rated processing and gradual increase control.
[0078] Specifically, the control system will continuously monitor the corrected THD evaluation value obtained in step 3. When multiple corrected THD evaluation values meet the preset triggering conditions, the control system will immediately start the derating process, that is, derating according to the severity of the current harmonics. After derating, it will enter the lockout and monitoring state. If the harmonic situation is continuously improved afterward, the gradual recovery mechanism will be triggered to gradually increase the available capacity in a gentle step until it is fully recovered.
[0079] In an optional embodiment, step 4 includes: Step 4.1: Obtain the current available capacity of the distribution radio area.
[0080] Specifically, the current available capacity of the distribution transformer substation (denoted as C) is obtained from the power grid carrying capacity assessment system. available (Unit: kW). This value is usually given by the higher-level dispatching system or the distribution network energy management system based on factors such as the rated capacity of the transformer area, real-time load, and historical load rate.
[0081] Step 4.2: Determine if there exists a continuous N. trig If all corrected THD estimates are greater than the harmonic distortion triggering threshold, then a bearing capacity derating state is triggered, and the most recent corrected THD estimate that triggered the bearing capacity derating state is taken as the THD estimate. cf .
[0082] Specifically, set the number of consecutive judgments N. trig (N trig (A positive integer representing the number of consecutive samples) sets the harmonic distortion trigger threshold (THD). thr1 (Unit: %). When N consecutive trig The corrected THD estimate THD for 1 sample (each sample corresponds to a corrected THD estimate THDest, and the sample generation period is Ts). est All are greater than THD thr1 When the bearing capacity derating state is triggered, the corrected THD estimate of the most recent sample at the trigger time is taken. est As a THD estimate, THD cf For example, N trig =3, THD thr1 =5%, but this embodiment does not impose a specific limit on this.
[0083] Step 4.3: Based on the THD estimate, THD cf The dynamic attenuation coefficient is determined by the harmonic distortion trigger threshold, and the derated openable capacity is determined based on the dynamic attenuation coefficient and the current openable capacity. The dynamic attenuation coefficient is used to map the degree of harmonic pollution to the reduction ratio of the openable capacity.
[0084] Step 4.31: Based on the THD estimate, THD cf The initial attenuation ratio is determined by the harmonic distortion triggering threshold, where the initial attenuation ratio varies with the estimated THD value. cf It decreases linearly with the increase.
[0085] Here, the initial attenuation ratio is expressed as: in, The initial attenuation ratio is given, and sat is the saturation range (in %). This controls the linear decrease rate of α as THD increases. For example, sat = 20%. This embodiment does not specifically limit this.
[0086] Step 4.32: Select the larger of the initial attenuation ratio and the minimum attenuation coefficient as the dynamic attenuation coefficient.
[0087] Here, the dynamic attenuation coefficient is expressed as: in, The dynamic attenuation coefficient, This is the minimum value of the attenuation coefficient, i.e., the lower limit of the attenuation coefficient (0 < ≤1), to prevent excessive reduction, for example, =0.5, but this embodiment does not impose a specific limitation on this.
[0088] Step 4.33: Based on the dynamic attenuation coefficient and the current available capacity, obtain the initial derated available capacity.
[0089] Here, the initial capacity that can be opened up after the reduction is expressed as: in, The initial reduction in quota can be opened up.
[0090] Step 4.34: Determine the relationship between the initial derating capacity and the minimum derating capacity. If the initial derating capacity is greater than or equal to the minimum derating capacity, then the initial derating capacity is used as the derating capacity. If the initial derating capacity is less than the minimum derating capacity, then the minimum derating capacity is used as the derating capacity. This avoids insufficient power supply due to derating. For example, the minimum derating capacity C... min =5kW, but this embodiment does not specifically limit this.
[0091] Step 4.4: Under the load-bearing capacity derating condition, determine whether there are consecutive N... recIf all corrected THD estimates are less than or equal to the recovery threshold, a gradual recovery state is triggered, wherein the harmonic distortion trigger threshold is greater than the recovery threshold.
[0092] Specifically, after the bearing capacity derating state is activated, the system continues to monitor the corrected THD estimate, still acquiring the corrected THD estimate for new samples every Ts. When N consecutive rec All corrected THD estimates are less than or equal to the recovery threshold THD. thr2 At that time, a gradual recovery state is triggered. This embodiment introduces a hysteresis interval (THD). thr1 THD thr2 To avoid frequent switching between decrement and recovery, for example, N rec =5, THD thr2 =3%, but this embodiment does not impose a specific limit on this.
[0093] Step 4.5: In the gradual recovery state, update the current dynamic attenuation coefficient based on the gradual increase step size to obtain the updated dynamic attenuation coefficient, and calculate the available capacity after derating based on the updated dynamic attenuation coefficient until the updated dynamic attenuation coefficient is 1, then the available capacity is fully restored.
[0094] Specifically, in the gradual recovery state, every Ts sampling periods, the current dynamic decay coefficient is increased by one gradual increase step size. Each increase of one gradual increase step size yields the updated dynamic decay coefficient. Each adjustment is as follows: ,in, This is the updated dynamic decay coefficient. The preset gradual increase step size (positive value), such as delta α =0.05, this embodiment does not specifically limit this value. Repeat this step until... = 1.0, fully restored available capacity.
[0095] Furthermore, in the gradual recovery state, if N consecutive events occur again before the current dynamic decay coefficient recovers to 1... trig If the corrected THD estimate is greater than the harmonic distortion trigger threshold, then the gradual recovery state is exited and the bearing capacity reduction state is entered, and the reduction calculation is performed in accordance with step 4.3.
[0096] This embodiment uses a 400kVA distribution substation as an example for illustration. Specific parameter settings are as follows: N trig The value is 3, THD thr1 It is 5%, Sat is 20%. It is 0.5, C min For 5 kW, THD thr2 3%, N recThe value is 5, delta α The value is 0.05. Based on the above parameters, the calculation of the bearing capacity derating state is as follows: If THD is 0.05, the bearing capacity derating state is as follows: cf = 5%, then α = max(0.5, 1 - (5-5) / 20) = 1.0, no reduction; if THD cf = 15%, then α = max(0.5, 1 - (15-5) / 20) = 0.5; if THD cf = 25% or higher, then α = max(0.5, 1 - (25-5) / 20) = 0.5, maintaining the lower limit. The recovery timing is: Normal state: α = 1.0, C = C available Triggering a reduction in THD: THD in 3 consecutive samples est Greater than 5%, based on the THD most recent at the trigger time. est Calculate α and perform de-rating; Recovery condition: THD of 5 consecutive samples est For values less than or equal to 3%, α increases by 0.05 every 5 minutes until it recovers to 1.0; mid-course deterioration: any three consecutive samples with THD... est If the threshold is >5%, immediately recalculate α (jump back to reduced threshold state).
[0097] This invention first employs a strategy of continuous multi-cycle sampling, DC bias removal, and normalization preprocessing to effectively suppress the influence of load fluctuations, noise interference, and DC components on the current waveform during the data acquisition stage. This enables the stable acquisition of harmonic distortion characteristic information under complex operating conditions, providing a high-quality input sequence for subsequent neural network regression.
[0098] Secondly, this invention is based on a lightweight one-dimensional convolutional neural network, which directly regresses the total harmonic distortion rate from the original current waveform time-series vector end-to-end. This transforms the traditional harmonic analysis method that relies on FFT into a lightweight, low-latency neural network estimation process, which significantly reduces the requirements for sampling synchronization and embedded terminal computing resources, and realizes real-time, low-cost perception of the degree of harmonic distortion.
[0099] Furthermore, this invention performs FFT frequency band energy analysis on the current waveform to extract the energy proportion characteristics of the low-order, zero-sequence, and high-frequency bands. Based on the dominant type, it performs equivalent weighted compensation on the THD output of the neural network, so that the corrected THD can reflect the degree of actual physical hazards of different harmonic sources to transformer heating, neutral point overcurrent, and high-frequency skin effect, thus overcoming the limitation of distortion in the assessment of a single THD index.
[0100] Finally, based on the corrected THD estimate, this invention designs a dynamic closed-loop derating control strategy that includes continuous trigger judgment, linear derating calculation, absolute lower limit protection, hysteresis recovery threshold, and gradual step recovery. This enables the distribution network's available capacity to adaptively adjust with changes in harmonic state, avoiding the response lag and frequent switching problems of traditional fixed threshold control methods. Thus, while ensuring the power supply safety of the distribution area, it maximizes the carrying capacity utilization efficiency under harmonic pollution conditions.
[0101] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of an adaptive derating system for distribution network carrying capacity harmonics provided by the present invention. Based on Embodiment 1, this embodiment of the present invention further provides an adaptive derating system for distribution network carrying capacity harmonics. This adaptive derating system is used to implement the adaptive derating method provided in Embodiment 1. The adaptive derating system provided in this embodiment includes: The timing vector generation module is used to obtain the current waveform timing vector based on multiple continuous instantaneous current values of the distribution substation. The THD initial estimate generation module is used to input the current waveform time series vector into a pre-trained lightweight one-dimensional convolutional neural network to obtain the THD initial estimate in an end-to-end manner. The THD correction module is used to determine the harmonic dominance type of the current harmonic energy based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, and to perform weighted compensation on the initial THD estimate according to the harmonic dominance type to obtain the corrected THD evaluation value. The openable capacity adjustment module is used to dynamically adjust the openable capacity of the distribution radio area based on continuously corrected THD assessment values, through derating and gradual increase control methods.
[0102] The implementation principle and technical effects of the adaptive derating system provided in this embodiment are similar to those of the adaptive derating method, and will not be repeated here.
[0103] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0104] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.
[0105] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, any modifications made without departing from the inventive concept should be considered within the scope of protection of the present invention.
Claims
1. An adaptive derating method for harmonic load capacity in a distribution network, characterized in that, include: The current waveform timing vector is obtained from multiple consecutive instantaneous current values of the distribution station area; The current waveform time-series vector is input into a pre-trained lightweight one-dimensional convolutional neural network to obtain the initial THD estimate in an end-to-end manner. Based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, the harmonic dominance type of the current harmonic energy is determined, and the initial THD estimate is weighted and compensated according to the harmonic dominance type to obtain the corrected THD evaluation value. Based on the continuously revised THD assessment values, the available capacity of the distribution radio station is dynamically adjusted through derating and gradual escalation control.
2. The adaptive derating method according to claim 1, characterized in that, The current waveform time sequence vector is obtained from multiple continuous instantaneous current values of the distribution substation, including: In the distribution station area, the instantaneous current values of K power frequency cycles are continuously collected at a preset sampling frequency to form the original current waveform sequence; The original current waveform sequence is subjected to DC bias removal processing to obtain a zero-mean current sequence; The zero-mean current sequence is linearly normalized to obtain the current waveform time sequence vector.
3. The adaptive derating method according to claim 1, characterized in that, The lightweight one-dimensional convolutional neural network includes a feature extraction module, an attention module, and a regression output module connected in sequence. The current waveform time-series vector is input into a pre-trained lightweight one-dimensional convolutional neural network to obtain an initial THD estimate in an end-to-end manner, including: The current waveform timing vector is input to the feature extraction module to extract waveform distortion features and output a first feature map including multiple channels. The first feature map is input into the attention module to recalibrate the weights of each channel of the first feature map and output a second feature map after channel weighting. The second feature map is input to the regression output module to perform feature integration on the second feature map and output the initial estimate of THD.
4. The adaptive derating method according to claim 1, characterized in that, Based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, the dominant harmonic type of the current harmonic energy is determined, and the initial THD estimate is weighted and compensated according to the dominant type to obtain a corrected THD evaluation value, including: Perform a fast Fourier transform on the current waveform time-series vector to obtain a frequency domain complex sequence; Based on the modulus of the nth complex number in the frequency domain complex number sequence, the normalized amplitude of the nth harmonic is obtained, where 2≤n≤Nh, and Nh is the highest frequency of the harmonic analysis; The proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy are determined based on the normalized amplitudes of the low-order harmonic number, the zero-sequence harmonic number, and the high-frequency harmonic number, respectively, wherein the low-order harmonic number is less than the high-frequency harmonic number. Based on the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy, the dominant harmonic type of the current harmonic energy is determined. Based on the compensation weight corresponding to the dominant harmonic type, the initial THD estimate is weighted and compensated, and the corrected THD evaluation value is output.
5. The adaptive derating method according to claim 4, characterized in that, Based on the proportion of low-order harmonic energy, the proportion of zero-sequence harmonic energy, and the proportion of high-frequency harmonic energy, the dominant harmonic type of the current harmonic energy is determined, including: Determine the maximum and minimum harmonic energy percentages among the low-order harmonic energy percentage, the zero-sequence harmonic energy percentage, and the high-frequency harmonic energy percentage. Determine whether the difference between the maximum value of the harmonic energy proportion and the minimum value of the harmonic energy proportion is less than a preset threshold. If yes, determine that the harmonic dominance type is a mixed harmonic dominance type. If no, determine the harmonic dominance type based on the maximum value of the harmonic energy proportion.
6. The adaptive derating method according to claim 1, characterized in that, Based on the continuously revised THD assessment values, the available capacity of the distribution radio area is dynamically adjusted through derating and gradual escalation control, including: Obtain the current available capacity of the power distribution area; Determine if there exists a continuous N trig If all corrected THD estimates are greater than the harmonic distortion triggering threshold, then a bearing capacity derating state is triggered, and the most recent corrected THD estimate that triggered the bearing capacity derating state is taken as the THD estimate. cf ; Based on the THD estimate, THD cf The dynamic attenuation coefficient is determined based on the harmonic distortion trigger threshold, and the derated openable capacity is determined based on the dynamic attenuation coefficient and the current openable capacity, wherein the dynamic attenuation coefficient is used to map the degree of harmonic pollution to the reduction ratio of the openable capacity. Under the load-bearing capacity derating state, determine whether there are consecutive N... rec If all corrected THD estimates are less than or equal to the recovery threshold, then a gradual recovery state is triggered, wherein the harmonic distortion trigger threshold is greater than the recovery threshold. In the gradual recovery state, the current dynamic attenuation coefficient is updated based on the gradual increase step size to obtain the updated dynamic attenuation coefficient, and the capacity that can be opened after derating is calculated based on the updated dynamic attenuation coefficient until the updated dynamic attenuation coefficient is 1, at which point the capacity that can be opened is fully restored.
7. The adaptive derating method according to claim 6, characterized in that, Based on the THD estimate, THD cf The dynamic attenuation coefficient is determined based on the harmonic distortion trigger threshold, and the derated available capacity is determined based on the dynamic attenuation coefficient and the current available capacity, including: Based on the THD estimate, THD cf The initial attenuation ratio is determined by the harmonic distortion triggering threshold, wherein the initial attenuation ratio varies with the estimated THD value. cf It decreases linearly with the increase; The larger of the initial attenuation ratio and the minimum attenuation coefficient is selected as the dynamic attenuation coefficient. Based on the dynamic attenuation coefficient and the current available capacity, the initial derated available capacity is obtained; Determine the relationship between the initial reduced open capacity and the minimum open capacity. If the initial reduced open capacity is greater than or equal to the minimum open capacity, then the initial reduced open capacity is taken as the reduced open capacity. If the initial reduced open capacity is less than the minimum open capacity, then the minimum open capacity is taken as the reduced open capacity.
8. The adaptive derating method according to claim 6, characterized in that, The updated dynamic decay coefficient is obtained by updating the current dynamic decay coefficient based on the gradual increase step size, including: Every Ts sampling cycles, the current dynamic attenuation coefficient is increased by one of the gradual increase steps to obtain the updated dynamic attenuation coefficient.
9. The adaptive derating method according to claim 6, characterized in that, In the described gradual recovery state, if N consecutive occurrences of [something] occur again before the current dynamic attenuation coefficient recovers to 1... trig If the corrected THD estimate is greater than the harmonic distortion trigger threshold, then the system exits the gradual recovery state and enters the load-bearing capacity derating state to perform derating calculations.
10. An adaptive derating system for the carrying capacity harmonics of a distribution network, characterized in that, include: The timing vector generation module is used to obtain the current waveform timing vector based on multiple continuous instantaneous current values of the distribution substation. The THD initial estimate generation module is used to input the current waveform time-series vector into a pre-trained lightweight one-dimensional convolutional neural network to obtain the THD initial estimate in an end-to-end manner. The THD correction module is used to determine the harmonic dominance type of the current harmonic energy based on the frequency domain complex sequence obtained by frequency domain analysis of the current waveform time sequence vector, and to perform weighted compensation on the initial THD estimate according to the harmonic dominance type to obtain the corrected THD evaluation value. The openable capacity adjustment module is used to dynamically adjust the openable capacity of the distribution radio station based on the continuously corrected THD assessment value, through derating and gradual increase control.