Harmonic optimization method and system for photovoltaic power distribution network

Abstract

The invention belongs to the technical field of photovoltaic power distribution networks, and provides a harmonic optimization method and system for a photovoltaic power distribution network, and the method comprises the steps: collecting harmonic related information of the photovoltaic power distribution network in real time, integrating and generating a harmonic information sequence, judging whether harmonic optimization is needed or not, marking an optimization time period and analyzing a harmonic generation factor if the harmonic optimization is needed, and outputting a harmonic optimization result; processing and intercepting the harmonic information sequence to obtain a training data set comprising a precursor window matrix, constructing a harmonic precursor identification model and training by adopting the training data set, if the current moment is not in the optimization time period, identifying by adopting the trained harmonic precursor identification model, judging whether the optimization time period is triggered, and if the optimization time period is triggered, judging whether the current moment is not in the optimization time period; determining harmonic generation factors according to an identification result, selecting a harmonic optimization compensation mechanism based on different harmonic generation factors in an optimization time period, performing optimization compensation on the photovoltaic power distribution network, judging whether optimization compensation can be stopped or not in the optimization time period, and if yes, ending the current optimization time period.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power distribution networks, and in particular relates to a harmonic optimization method and system for photovoltaic power distribution networks. Background Art

[0002] As an important form of renewable energy access to the power grid, the scale and penetration of photovoltaic distribution networks are constantly increasing. However, the nonlinear characteristics of a large number of power electronic equipment in photovoltaic power stations and the dynamic nature of grid operation scenarios have led to increasingly prominent harmonic problems. Traditional distribution network harmonic control methods have exposed multiple technical bottlenecks when applied to photovoltaic scenarios. Existing technologies rely solely on the threshold of the total voltage distortion rate to determine whether optimization is needed. They fail to integrate scenario data such as light intensity, short-circuit capacity ratio, and load power factor, and lack a multi-parameter collaborative analysis mechanism. This leads to inaccurate identification of harmonic generation factors and inadequately targeted governance strategies. Traditional harmonic compensation equipment relies on passive triggering after the voltage distortion rate exceeds the standard, and is unable to capture precursory features before harmonics occur. In photovoltaic distribution networks, factors such as light intensity fluctuations and short-circuit capacity changes have obvious temporal correlations. For example, high-frequency fluctuations in light intensity in cloudy weather may indicate an impending harmonic exceedance. However, existing technologies have not established a correlation model between time series data and harmonic risks, resulting in delayed compensation actions, which can easily lead to equipment overload or transient grid shocks. On the other hand, different harmonic generation factors require differentiated management. Light fluctuations require real-time tracking of compensation current, insufficient short-circuit capacity requires dynamic adjustment of reactive power, and low load power factor requires filtering out specific frequency harmonics. However, existing technologies generally use a single compensation device or fixed parameter control strategy, and are unable to switch the compensation mechanism according to real-time inducements, resulting in low management efficiency and even the risk of resonance between the compensation equipment and the power grid. Existing methods usually use the compliance of the voltage distortion rate as the only termination condition, ignoring whether the harmonic generation factors have disappeared. For example, when the problem of insufficient short-circuit capacity ratio is not solved, stopping compensation only because the voltage distortion rate temporarily meets the standard may cause recurrence of harmonic problems, or continue to operate the compensation equipment after the inducement disappears, resulting in equipment loss and energy waste.

[0003] In response to the above problems, the present invention proposes a harmonic optimization method and system for a photovoltaic distribution network. Summary of the Invention

[0004] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a harmonic optimization method for a photovoltaic distribution network, comprising: Collect harmonic information of the photovoltaic distribution network in real time, integrate it to generate a harmonic information sequence, and determine whether harmonic optimization is needed. If necessary, mark the optimization period and analyze the harmonic generation factors; The harmonic information sequence is processed and intercepted to obtain a training data set including a precursor window matrix. A harmonic precursor recognition model is constructed and trained using the training data set. If the current time is not in the optimization period, the trained harmonic precursor recognition model is used for recognition to determine whether the optimization period is triggered. If so, the harmonic generation factor is determined based on the recognition result. During the optimization period, the harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network. It is also determined in real time whether the optimization compensation can be stopped. If so, the current optimization period is ended. The harmonic information sequence is obtained as follows: Harmonic-related information includes the total voltage distortion rate of the photovoltaic distribution network and the corresponding scenario data. The scenario data includes light intensity, short-circuit capacity ratio, and load power factor. Set a historical collection period with the current moment as the end point. At each moment in the historical collection period, integrate the collected harmonic-related information of the photovoltaic distribution network into a harmonic-related information vector, and obtain a harmonic information sequence based on the time series integration. The marking method of the optimization period is as follows: Obtain harmonic-related information, including the total voltage distortion rate of the photovoltaic distribution network. Set a start flag value and assign its initial value to 0. Compare the current total voltage distortion rate with a preset total voltage distortion rate threshold. If the total voltage distortion rate is greater than the threshold, determine that the photovoltaic distribution network needs harmonic optimization. If it is determined that the photovoltaic distribution network needs to be optimized and the start flag value is 0, the current moment is marked as the optimization starting point, and the start flag value is set to 1. The period when the start flag value is continuously 1 is marked as the optimization period; The method of constructing the harmonic precursor identification model is as follows: The harmonic precursor recognition model constructed uses a CNN-GRU hybrid deep learning network architecture, including an input layer, a convolutional neural network part, a gated recurrent unit part, a fully connected layer, and an output layer. The input layer receives the precursor window matrix. The convolutional neural network part includes a convolution layer and a pooling layer. The unit structure of the gated recurrent unit part consists of two gate structures: an update gate and a reset gate. The fully connected layer uses the ReLU activation function to output a feature vector. The output layer uses the Softmax activation function to convert the feature vector output by the fully connected layer into the probability distribution of each harmonic generation factor and output the probability distribution vector of each harmonic generation factor. The training data set is obtained as follows: Acquire the harmonic information sequence and perform normalization processing to obtain a normalized harmonic information sequence; Extract each optimized moment within the historical acquisition period and set a buffer window. For the buffer window with the optimized moment as the end point, take the starting point of the buffer window as the end point in the normalized harmonic information sequence and intercept a fixed-length data sequence forward, which is defined as the precursor window sequence. Each precursor window sequence is organized into a matrix to obtain a precursor window matrix. For each precursor window matrix, the corresponding harmonic generation factors are marked for the precursor window matrix, and a training data set is obtained; Wherein, the method of determining the harmonic generation factor is: If the current time is not in the optimized period, the harmonic information sequence is obtained and normalized to obtain a normalized harmonic information sequence. The current window sequence is intercepted on the normalized harmonic information sequence, organized into a current window matrix and input into the harmonic precursor identification model for identification to obtain the probability distribution vector of each harmonic generation factor. Harmonic generation factors include light intensity fluctuation, insufficient short-circuit capacity ratio, and low load power factor. When the maximum probability value in the probability distribution vector of each harmonic generation factor is greater than a preset probability threshold, the optimization period is triggered, the end point of the buffer window starting from the current moment is marked as the optimization moment, and the start flag value is set to 1. The harmonic generation factor corresponding to the maximum probability value is used as the harmonic generation factor at the optimization moment; The harmonic optimization compensation mechanism includes: If the harmonic generation factor at the current optimization moment is light intensity fluctuation, the adaptive active power filter is activated to detect the harmonic current in the grid in real time, generate a compensation current of equal magnitude and opposite direction, and inject it into the photovoltaic distribution network to offset the harmonic current; If the harmonic generation factor at the current optimization moment is insufficient short-circuit capacity ratio, the static VAR compensator is put into operation to adjust its own reactive output, change the reactive power distribution of the power grid, and enhance the system's ability to resist harmonic disturbances; If the harmonic generation factor at the current optimization moment is due to a low load power factor, the hybrid filter bank is activated to extract the harmonic current, generating an equal and opposite offset current, which is injected into the photovoltaic distribution network to achieve optimal compensation for the harmonics. The method for determining whether to stop the optimization compensation is as follows: Obtain the stability mark value of the current optimization period. If it is greater than the preset stability value, determine that the total voltage distortion rate of the photovoltaic distribution network continues to meet the standard. Obtain the harmonic generation factor corresponding to the current moment and determine whether the harmonic generation factor has disappeared. If the total voltage distortion rate continues to meet the standard and the harmonic generation factors disappear, the start flag value will be reset to 0, and it is determined that the optimization compensation can be stopped; The stability flag value is obtained as follows: At the optimization starting point of the optimization period, the stability flag value of the current optimization period is set and its initial value is assigned to 0. During the optimization period, the total voltage distortion rate collected is the compensated total voltage distortion rate. If the total voltage distortion rate is less than or equal to the total voltage distortion rate threshold, the stability flag value performs a self-increment operation. Otherwise, the stability flag value is reset to 0.

[0006] A harmonic optimization system for a photovoltaic distribution network includes the following modules: Acquisition and Analysis Module: This module collects harmonic information from the photovoltaic distribution network in real time, integrates it to generate a harmonic information sequence, and determines whether harmonic optimization is required. If so, it marks the optimization period and analyzes the factors that cause harmonics. Model training module: Process and intercept the harmonic information sequence to obtain a training data set including a precursor window matrix, and construct and use the training data set to train a harmonic precursor recognition model; Precursor recognition module: If the current time is not in the optimization period, the trained harmonic precursor recognition model is used to identify whether the optimization period is triggered. If triggered, the harmonic generation factor is determined based on the recognition result; Optimization compensation module: During the optimization period, the harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network. It is determined in real time whether the optimization compensation can be stopped. If so, the current optimization period is ended.

[0007] The beneficial effects of the present invention are as follows: 1. The present invention monitors power quality in multiple dimensions, with the total voltage distortion rate as the core, and accurately locates harmonic problems in combination with scene data. It comprehensively traces the harmonic causes of the photovoltaic distribution network, extracts harmonic precursor features using buffer windows and precursor window sequences, and combines deep learning models to predict risks in advance and improve recognition accuracy. Special compensation equipment is used for different causes, and parameters are dynamically adjusted to achieve precise management, effectively improve compensation efficiency and accuracy, and ensure the stable operation of the photovoltaic distribution network.

[0008] 2. The present invention determines whether to stop optimization based on the dual conditions of continuous compliance of voltage distortion rate and disappearance of harmonic inducement, avoiding the hysteresis or false stop of the traditional single time threshold termination method. By quantifying the voltage quality stability through self-addition and zeroing of the mark value, the compensation operation is stopped in time, reducing equipment loss and energy consumption, improving the economic efficiency of system operation, and taking into account the prediction lead time and accuracy, it provides an efficient, reliable and economical solution for harmonic optimization of photovoltaic distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The present invention will be further described below with reference to the accompanying drawings.

[0010] Figure 1 This is a flowchart of the steps of a harmonic optimization method for a photovoltaic distribution network according to an embodiment of the present invention; Figure 2 This is a module architecture diagram of a harmonic optimization system for a photovoltaic distribution network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0012] Example 1:

[0013] See also Figure 1 As shown, a harmonic optimization method for a photovoltaic distribution network according to an embodiment of the present invention includes the following steps: S1: Collect harmonic information of the photovoltaic distribution network in real time, integrate it to generate a harmonic information sequence, and determine whether harmonic optimization is needed. If necessary, mark the optimization period and analyze the harmonic generation factors; Real-time collection of harmonic-related information of the photovoltaic distribution network, including the total voltage distortion rate of the photovoltaic distribution network and corresponding scenario data. The scenario data also includes light intensity, short-circuit capacity ratio, and load power factor. The total voltage distortion rate is measured and calculated by the power quality monitor installed in the photovoltaic distribution network. The calculation formula is: ; In the calculation formula, t represents the current time, Indicates the effective value of the fundamental voltage, The hth harmonic voltage RMS value is represented by the Fast Fourier Transform (FFT) algorithm inside the power quality monitor, which decomposes the collected voltage signal into the fundamental wave and each harmonic signal. The fundamental voltage RMS value and each harmonic voltage RMS value are then calculated. h = 2, 3, ⋯, H, where H represents the highest harmonic order that can be monitored. It should be noted that the total voltage distortion rate is one of the important criteria for measuring power quality. If the total voltage distortion rate is too high, it may have an adverse impact on the normal operation of electrical equipment. Therefore, the total voltage distortion rate is the core basis for determining whether harmonic optimization is necessary. The light intensity S is measured by installing photosynthetic radiation sensors at the photovoltaic module array of the photovoltaic power station, and the short-circuit capacity ratio The load power factor is obtained and calculated through the technical data of the photovoltaic distribution network. The power is measured and calculated in real time by installing a power measurement device on the load side of the photovoltaic distribution network; Set a historical collection period with the current moment as the end point. At each moment in the historical collection period, integrate the collected harmonic-related information of the photovoltaic distribution network into a harmonic-related information vector, and obtain a harmonic information sequence based on the time series integration. Based on the real-time collected harmonic information of the photovoltaic distribution network, determine whether harmonic optimization is needed at the current moment. If so, determine the harmonic generation factors. Specifically, a startup flag value is set and its initial value is assigned to 0. According to relevant national standards and the specific operating requirements of the photovoltaic distribution network, a total voltage distortion rate threshold is set. The current total voltage distortion rate is compared with the total voltage distortion rate threshold. If the current total voltage distortion rate is greater than the total voltage distortion rate threshold, it is determined that the photovoltaic distribution network needs to be optimized for harmonics. If it is determined that the photovoltaic distribution network needs to be optimized and the start flag value is 0, the current moment is marked as the optimization starting point, and the start flag value is set to 1. The period when the start flag value is continuously 1 is marked as the optimization period; If the current moment is the optimization starting point, the specific factors causing harmonic generation are determined based on the scene data corresponding to the current moment. An analysis window is intercepted within the historical collection period with the current moment as the end point. The difference in light intensity between the end point and the starting point of the analysis window is obtained and marked as the light intensity fluctuation amplitude at the current moment. If the light intensity fluctuation amplitude at the current moment is greater than the preset fluctuation amplitude threshold, it is determined that the harmonic generation factor that needs to be optimized is light intensity fluctuation. A short-circuit capacity ratio critical value is set based on the actual situation of the photovoltaic distribution network. If the short-circuit capacity ratio of the photovoltaic distribution network at the current moment is less than the set short-circuit capacity ratio critical value, it indicates that the photovoltaic distribution network's support capacity for the photovoltaic power station is weak, and it is determined that the harmonic generation factor that needs to be optimized is insufficient short-circuit capacity ratio. Similarly, a load power factor standard value is set. If the load power factor of the photovoltaic distribution network at the current moment is less than the set load power factor standard value, it indicates that the proportion of nonlinear loads in the load is high, and it is determined that the harmonic generation factor that needs to be optimized is too low load power factor. It should be noted that this step is to monitor power quality in multiple dimensions, using the total voltage distortion rate as the core indicator. It combines scenario data to accurately locate harmonic problems and analyze parameters such as the sunlight fluctuation amplitude and the short-circuit capacity ratio critical value based on historical data windows. This improves the accuracy of determining harmonic generation factors, avoids misjudgments, and comprehensively covers the harmonic inducements unique to photovoltaic distribution networks, achieving multi-dimensional tracing of harmonic factors. S2: Process and intercept the harmonic information sequence to obtain a training data set including a precursor window matrix, construct a harmonic precursor recognition model and train it using the training data set. If the current time is not in the optimization period, use the trained harmonic precursor recognition model to identify whether the optimization period is triggered. If triggered, determine the harmonic generation factor based on the recognition result; Acquire a harmonic information sequence within a historical acquisition period, and normalize each component of the harmonic information vector to obtain a normalized harmonic information sequence; Extract each optimized moment within the historical acquisition period and set a buffer window. For the buffer window with the optimized moment as the end point, intercept a fixed-length data sequence from the start point of the buffer window in the normalized harmonic information sequence, which is defined as a precursor window sequence. Each precursor window sequence is organized into a matrix of dimension L×4, labeled as a precursor window matrix, where L represents the length of the precursor window sequence, that is, the number of moments in the precursor window sequence. It should be noted that the role of the buffer window is to optimize the response to the harmonics that will be generated in a timely manner after the precursory characteristics of the harmonic generation are subsequently identified, to prevent damage to the photovoltaic distribution network caused by untimely response. The buffer window duration setting cannot excessively lose precursory characteristics to prevent excessive deviations in preventive optimization and impact on the photovoltaic distribution network. The buffer window duration is the same for all optimization moments. For each precursor window matrix, the harmonic generation factor at the corresponding optimization moment is used as the harmonic generation factor of the precursor window matrix, and the corresponding harmonic generation factor is labeled for the precursor window matrix. All precursor window matrices in the historical acquisition period are labeled to obtain a training data set. The training dataset is used to train the harmonic precursor recognition model built on the CNN-GRU hybrid deep learning network; Specifically, a CNN-GRU hybrid deep learning network architecture was built for the harmonic precursor recognition model, including an input layer, a convolutional neural network (CNN) part, a gated recurrent unit (GRU) part, a fully connected layer, and an output layer; Among them, the input layer receives the precursor window matrix, the convolutional neural network part includes a convolution layer and a pooling layer, the convolution layer extracts features from the input precursor window matrix and generates a feature map, the pooling layer downsamples the feature map output by the convolution layer, reduces the data dimension, and reduces the amount of model calculation, the unit structure of the gated recurrent unit part consists of two gate structures, an update gate and a reset gate, which capture the temporal change trend and dependency of the precursor window matrix, the fully connected layer uses the ReLU activation function to further extract and integrate key features related to harmonic generation factors, and outputs a feature vector, the output layer uses the Softmax activation function to convert the feature vector output by the fully connected layer into the probability distribution of each harmonic generation factor, and outputs the probability distribution vector of each harmonic generation factor, the sum of each component of the probability distribution vector is 1, and the harmonic generation factors include light intensity fluctuation, insufficient short-circuit capacity ratio, and too low load power factor; A stratified sampling method was used to divide the training dataset into a training set and a validation set in a ratio of 7:3. Hyperparameters were set for the CNN-GRU hybrid deep learning network. The training set was input into the CNN-GRU hybrid deep learning network in batches. The cross-entropy loss function was used to calculate the loss value. The calculated loss value was backpropagated from the output layer to the input layer along the network structure through the backpropagation algorithm to calculate the parameter gradient. The Adam optimizer was used to update the parameters according to the calculated parameter gradient and the set learning rate, and the performance of the harmonic precursor recognition model was gradually optimized. The harmonic precursor recognition model was evaluated using the validation set data, and the recognition accuracy was calculated. An early stopping mechanism was introduced to prevent overfitting. When the recognition accuracy of the harmonic precursor recognition model in the validation set reached a preset threshold or the recognition accuracy no longer improved, the trained harmonic precursor recognition model was obtained. If the current moment is not in the optimization period, on the normalized harmonic information sequence, a data sequence of length L is intercepted forward from the current moment, which is defined as the current window sequence. The current window sequence slides with the change of the current moment. At the current moment, the current window sequence is organized into a matrix of dimension L×4, labeled as the current window matrix. The current window matrix is ​​input into the trained harmonic precursor recognition model for identification, and the probability distribution vector of each harmonic generation factor is obtained. Harmonic generation factors include light intensity fluctuations, insufficient short-circuit capacity ratio, and low load power factor. The maximum probability value in the probability distribution vector of each harmonic generation factor is compared with a preset probability threshold. If the maximum probability value is less than or equal to the preset probability threshold, it is determined that the precursory characteristics of harmonic generation cannot be identified in the current window sequence. The probability of the photovoltaic distribution network generating harmonics within the buffer window starting at the current time is low, and preventive optimization is not required, and the optimization period is not triggered. If the maximum probability value is greater than the preset probability threshold, the optimization period is triggered, the end point of the buffer window starting from the current moment is marked as the optimization moment, the start flag value is set to 1, and the harmonic generation factor corresponding to the maximum probability value is used as the harmonic generation factor at the optimization moment; It should be noted that the purpose of this step is to extract the timing characteristics before the occurrence of harmonics through the buffer window and precursor window sequence, predict harmonic risks in advance, shorten the response time, and combine the spatial feature extraction capability of the convolutional neural network with the timing dependency capture capability of the gated recurrent unit. This takes into account both the spatial characteristics and timing trends of the data, improves the recognition accuracy of harmonic precursors in complex scenarios, and balances the prediction lead time and accuracy without losing features, thus avoiding over-response or lag caused by preventive optimization. S3: During the optimization period, a harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network; If the current moment is the optimization moment, the harmonic generation factors at the optimization moment are obtained, a harmonic optimization compensation mechanism is selected for the photovoltaic distribution network based on different harmonic generation factors, and the selected harmonic optimization compensation mechanism is used to optimize and compensate for the harmonics during the optimization period; Specifically, if the harmonic generation factor at the current optimization moment is light intensity fluctuation, the adaptive active power filter (APF) is activated for harmonic compensation. During the optimization period, the adaptive active power filter detects the harmonic current in the power grid in real time, generates a compensation current of equal magnitude and opposite direction, and injects it into the photovoltaic distribution network, thereby offsetting the harmonic current. Compensation current command of adaptive active power filter The calculation formula is: ; Among them, the current moment is the tth moment in the optimization period, and the optimization starting point is the first moment in the optimization period. Indicates the difference between the total voltage distortion rate at the current moment and the total voltage distortion rate threshold. 、 Represent the proportional coefficient and integral coefficient respectively, Indicates the time between the start of the optimization period and the current time The points, represents the time variable within the optimization period; If the harmonic generation factor at the current optimization time is insufficient short-circuit capacity ratio, a static VAR compensator (SVC) is used to optimize harmonic compensation. The SVC adjusts its own reactive power output to change the reactive power distribution of the power grid, thereby increasing the short-circuit capacity ratio of the power grid and enhancing the system's ability to resist harmonic disturbances. Admittance compensation of static VAR compensator The calculation formula is: ; ; in, Indicates the reactive power to be compensated, which is the active power output by the photovoltaic inverter and target power factor Calculation shows that the photovoltaic inverter outputs active power The target power factor is obtained in real time through the monitoring device of the photovoltaic inverter itself. Parameters pre-set according to the photovoltaic distribution network operation requirements and equipment characteristics, Indicates the voltage at the common connection point, measured by the power quality monitor installed at the common connection point; If the harmonic generation factor at the current optimization moment is that the load power factor is too low, the hybrid filter bank (HBF) is activated for harmonic compensation. The harmonic extraction operator of the hybrid filter bank extracts the compensated harmonic current. The calculation formula is: ; ; in, It represents the load current waveform, which is acquired in real time by the power measurement device installed on the load side. FFT stands for Fast Fourier Transform, and f stands for The frequency is converted from the time domain to the frequency domain through fast Fourier transform. Represents the frequency domain filter function. When the frequency is the fundamental frequency, that is, When the function value is 0, the fundamental component is not extracted. When the function value is 1, the corresponding harmonic components are extracted and the harmonic current in the load current is separated. ; The hybrid filter bank is based on the calculated harmonic current , generating equal and opposite offset currents, which are injected into the photovoltaic distribution network to achieve optimal compensation for harmonics; The compensation effect is verified in real time. The total voltage distortion rate after compensation is collected in real time and compared with the total voltage distortion rate before compensation. The compensation efficiency is calculated to evaluate the optimization compensation effect. If the compensation efficiency reaches the efficiency threshold and the total voltage distortion rate after compensation is If the voltage is less than or equal to the total voltage distortion rate threshold, the current optimization compensation is determined to be valid. Otherwise, the optimization compensation parameter readjustment mechanism is triggered to adjust the optimization compensation parameters, including the proportional coefficient and integral coefficient of the adaptive active power filter, the admittance compensation amount of the static VAR compensator, and the harmonic extraction operator parameters of the hybrid filter bank. The optimization compensation operation is then performed again according to the adjusted optimization compensation parameters until the current optimization compensation is valid. It should be noted that the purpose of this step is to use dedicated compensation equipment for different harmonic inducements to improve compensation efficiency. The compensation parameters are dynamically adjusted based on the difference between the real-time detected voltage distortion rate and the threshold value, achieving adaptive optimization of the compensation strategy, achieving precise control of harmonic problems, and improving compensation accuracy in dynamic scenarios. S4: determining whether optimization compensation can be stopped during the optimization period, and if so, ending the current optimization period; During the optimization compensation process within the optimization period, determine whether the optimization compensation can be stopped; Specifically, during the optimization period, real-time information about the harmonics of the photovoltaic distribution network is obtained, including the total voltage distortion rate of the photovoltaic distribution network and corresponding scenario data. The scenario data includes light intensity, short-circuit capacity ratio, and load power factor. At the optimization starting point, a stability flag value is set and its initial value is 0. During the optimization period, the collected total voltage distortion rate is the compensated total voltage distortion rate. The total voltage distortion rate is compared with the total voltage distortion rate threshold. If the total voltage distortion rate at the current moment is less than or equal to the total voltage distortion rate threshold, the stability flag value is self-incremented. Otherwise, the stability flag value is reset to 0. If the stability mark value is greater than the preset stability value, it is determined that the total voltage distortion rate of the photovoltaic distribution network continues to meet the standard. Based on the harmonic generation factor corresponding to the current optimization moment, it is determined whether the harmonic generation factor has disappeared. Specifically, if the harmonic generation factor is light intensity fluctuation, the light intensity fluctuation amplitude at the current moment is calculated. If the light intensity fluctuation amplitude is less than or equal to a preset fluctuation amplitude threshold, it is determined that the light intensity fluctuation has stopped and the harmonic generation factor has disappeared. If the harmonic generation factor is insufficient short-circuit capacity ratio, and the short-circuit capacity ratio of the photovoltaic distribution network at the current moment is greater than or equal to the set short-circuit capacity ratio critical value, it is judged that the short-circuit capacity ratio has returned to normal and the harmonic generation factor has disappeared; If the harmonic generation factor is a low load power factor, and the load power factor of the photovoltaic distribution network at the current moment is greater than or equal to the load power factor standard value, it is judged that the load operating status has improved and the harmonic generation factor has disappeared; If the total voltage distortion rate continues to meet the standard and the harmonic generation factors disappear, the start flag value will be reset to 0, indicating that the harmonic optimization operation has been completed, ending the current optimization period, stopping all ongoing optimization and compensation operations, and returning the system to normal monitoring status to avoid unnecessary resource consumption and equipment operation losses; It should be noted that this step ensures the reliability of optimization termination based on the dual conditions of continuous compliance with voltage distortion standards and the disappearance of harmonic inducements. It promptly stops compensation operations, reduces equipment loss and energy consumption, and improves the economic efficiency of system operation. A dynamic termination mechanism is designed to combine power quality indicators with the elimination status of harmonic inducements, avoiding the lag or false shutdown of the traditional single time threshold termination method. Through the logic design of self-increment and zeroing of the flag value, the stability of voltage quality is quantified, realizing intelligent judgment of optimization termination. The technical solution of an embodiment of the present invention is: real-time collection of harmonic-related information of a photovoltaic distribution network, integration to generate a harmonic information sequence, and judgment on whether harmonic optimization is required; if necessary, marking the optimization period and analyzing harmonic generation factors; processing and intercepting the harmonic information sequence to obtain a training data set including a precursor window matrix; constructing a harmonic precursor recognition model and using the training data set for training; if the current moment is not in the optimization period, using the trained harmonic precursor recognition model for recognition, and judging whether the optimization period is triggered; if triggered, determining the harmonic generation factors according to the recognition results; within the optimization period, selecting a harmonic optimization compensation mechanism based on different harmonic generation factors, and optimizing and compensating the photovoltaic distribution network; judging whether the optimization compensation can be stopped within the optimization period; and if so, ending the current optimization period.

[0014] Example 2:

[0015] like Figure 2 As shown, a harmonic optimization system for a photovoltaic distribution network according to an embodiment of the present invention includes the following modules: Acquisition and Analysis Module: This module collects harmonic information from the photovoltaic distribution network in real time, integrates it to generate a harmonic information sequence, and determines whether harmonic optimization is required. If so, it marks the optimization period and analyzes the factors that cause harmonics. Model training module: Process and intercept the harmonic information sequence to obtain a training data set including a precursor window matrix, and construct and use the training data set to train a harmonic precursor recognition model; Precursor recognition module: If the current time is not in the optimization period, the trained harmonic precursor recognition model is used to identify whether the optimization period is triggered. If triggered, the harmonic generation factor is determined based on the recognition result; Optimization compensation module: During the optimization period, the harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network. It is determined in real time whether the optimization compensation can be stopped. If so, the current optimization period is ended.

[0016] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A harmonic optimization method for a photovoltaic distribution network, characterized by: include: Collect harmonic information of the photovoltaic distribution network in real time, integrate it to generate a harmonic information sequence, and determine whether harmonic optimization is needed. If necessary, mark the optimization period and analyze the harmonic generation factors; The harmonic information sequence is processed and intercepted to obtain a training data set including a precursor window matrix. A harmonic precursor recognition model is constructed and trained using the training data set. If the current time is not in the optimization period, the trained harmonic precursor recognition model is used for recognition to determine whether the optimization period is triggered. If so, the harmonic generation factor is determined based on the recognition result. During the optimization period, the harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network, and it is determined in real time whether the optimization compensation can be stopped. If so, the current optimization period is ended.

2. The method for harmonic optimization of a photovoltaic power distribution network according to claim 1, wherein: The method for obtaining the harmonic information sequence is as follows: Harmonic-related information includes the total voltage distortion rate of the photovoltaic distribution network and the corresponding scenario data. The scenario data includes light intensity, short-circuit capacity ratio, and load power factor. A historical collection period with the current moment as the end point is set. At each moment in the historical collection period, the harmonic-related information of the photovoltaic distribution network collected is integrated into a harmonic-related information vector, and a harmonic information sequence is obtained according to the time series integration.

3. The method for harmonic optimization of a photovoltaic power distribution network according to claim 2, wherein: The optimization period is marked as follows: Obtain harmonic-related information, including the total voltage distortion rate of the photovoltaic distribution network. Set a start flag value and assign its initial value to 0. Compare the current total voltage distortion rate with a preset total voltage distortion rate threshold. If the total voltage distortion rate is greater than the threshold, determine that the photovoltaic distribution network needs harmonic optimization. If it is determined that the photovoltaic distribution network needs to be optimized and the start flag value is 0, the current moment is marked as the optimization starting point, and the start flag value is set to 1. The period when the start flag value is continuously 1 is marked as the optimization period.

4. The method for harmonic optimization of a photovoltaic power distribution network according to claim 1, wherein: The method of constructing the harmonic precursor identification model is: The harmonic precursor recognition model constructed uses a CNN-GRU hybrid deep learning network architecture, including an input layer, a convolutional neural network part, a gated recurrent unit part, a fully connected layer, and an output layer. The input layer receives the precursor window matrix. The convolutional neural network part includes a convolution layer and a pooling layer. The unit structure of the gated recurrent unit part consists of two gate structures: an update gate and a reset gate. The fully connected layer uses the ReLU activation function to output a feature vector. The output layer uses the Softmax activation function to convert the feature vector output by the fully connected layer into the probability distribution of each harmonic generation factor and output the probability distribution vector of each harmonic generation factor.

5. The method for harmonic optimization of a photovoltaic power distribution network according to claim 3, wherein: The training data set is obtained as follows: Acquire the harmonic information sequence and perform normalization processing to obtain a normalized harmonic information sequence; Extract each optimized moment within the historical acquisition period and set a buffer window. For the buffer window with the optimized moment as the end point, take the starting point of the buffer window as the end point in the normalized harmonic information sequence and intercept a fixed-length data sequence forward, which is defined as the precursor window sequence. Each precursor window sequence is organized into a matrix to obtain a precursor window matrix. For each precursor window matrix, the corresponding harmonic generation factors are marked for the precursor window matrix, and a training data set is obtained.

6. The method for harmonic optimization of a photovoltaic power distribution network according to claim 4, characterized in that: The method for determining the harmonic generation factor is: If the current time is not in the optimized period, the harmonic information sequence is obtained and normalized to obtain a normalized harmonic information sequence. The current window sequence is intercepted on the normalized harmonic information sequence, organized into a current window matrix and input into the harmonic precursor identification model for identification to obtain the probability distribution vector of each harmonic generation factor. Harmonic generation factors include light intensity fluctuation, insufficient short-circuit capacity ratio, and low load power factor. When the maximum probability value in the probability distribution vector of each harmonic generating factor is greater than the preset probability threshold, the optimization period is triggered, the end point of the buffer window starting from the current moment is marked as the optimization moment, and the start flag value is set to 1, and the harmonic generating factor corresponding to the maximum probability value is used as the harmonic generating factor at the optimization moment.

7. The method for harmonic optimization of a photovoltaic power distribution network according to claim 6, characterized in that: The harmonic optimization compensation mechanism includes: If the harmonic generation factor at the current optimization moment is light intensity fluctuation, the adaptive active power filter is activated to detect the harmonic current in the grid in real time, generate a compensation current of equal magnitude and opposite direction, and inject it into the photovoltaic distribution network to offset the harmonic current; If the harmonic generation factor at the current optimization moment is insufficient short-circuit capacity ratio, the static VAR compensator is put into operation to adjust its own reactive output, change the reactive power distribution of the power grid, and enhance the system's ability to resist harmonic disturbances; If the harmonic generation factor at the current optimization moment is too low a load power factor, the hybrid filter bank is activated to extract the harmonic current, generate an equal and opposite offset current, and inject it into the photovoltaic distribution network to achieve optimal compensation for the harmonics.

8. The method for harmonic optimization of a photovoltaic power distribution network according to claim 2, wherein: The method for determining whether to stop the optimization compensation is as follows: Obtain the stability mark value of the current optimization period. If it is greater than the preset stability value, determine that the total voltage distortion rate of the photovoltaic distribution network continues to meet the standard. Obtain the harmonic generation factor corresponding to the current moment and determine whether the harmonic generation factor has disappeared. If the total voltage distortion rate continues to meet the standard and the harmonic generation factors disappear, the start flag value will be reset to 0, and it is determined that the optimization compensation can be stopped.

9. The method for harmonic optimization of a photovoltaic power distribution network according to claim 8, characterized in that: The method for obtaining the stability flag value is as follows: At the optimization starting point of the optimization period, the stability flag value of the current optimization period is set and its initial value is assigned to 0. During the optimization period, the total voltage distortion rate collected is the compensated total voltage distortion rate. If the total voltage distortion rate is less than or equal to the total voltage distortion rate threshold, the stability flag value performs a self-increment operation. Otherwise, the stability flag value is reset to 0.

10. A harmonic optimization system for a photovoltaic distribution network, the system being used to implement the optimization method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Acquisition and Analysis Module: This module collects harmonic information from the photovoltaic distribution network in real time, integrates it to generate a harmonic information sequence, and determines whether harmonic optimization is required. If so, it marks the optimization period and analyzes the factors that cause harmonics. Model training module: Process and intercept the harmonic information sequence to obtain a training data set including a precursor window matrix, and construct and use the training data set to train a harmonic precursor recognition model; Precursor recognition module: If the current time is not in the optimization period, the trained harmonic precursor recognition model is used to identify whether the optimization period is triggered. If triggered, the harmonic generation factor is determined based on the recognition result; Optimization compensation module: During the optimization period, the harmonic optimization compensation mechanism is selected based on different harmonic generation factors to optimize the compensation of the photovoltaic distribution network. It is determined in real time whether the optimization compensation can be stopped. If so, the current optimization period is ended.

Images