Grid-connected detection method and system of energy storage converter

Abstract

The invention discloses a grid-connected detection method and system for an energy storage converter, and relates to the technical field of electrics, and the method comprises the steps: collecting the operation data of the energy storage converter through a multi-source sensor, and carrying out the preprocessing; fusing the virtual grid-connected scene data with the operation data of the energy storage converter to form an enhanced data pool, and performing feature extraction on the enhanced data pool to form a feature vector; based on the feature vectors, extracting the feature vectors of each time point through a sliding window method, and combining the feature vectors into a three-dimensional array; and based on the three-dimensional array, obtaining a grid-connected state through a grid-connected detection model. According to the method, the operation data of the energy storage converter are collected and preprocessed through the multi-source sensor, virtual grid-connected scene data are generated by utilizing the generative adversarial network, the adaptability of the model to complex working conditions is enhanced through rich data samples, and finally more accurate and more stable grid-connected state detection of the energy storage converter is realized.

Description

Technical Field

[0001] The present invention relates to the field of electricity, and particularly to a grid connection detection method and system for an energy storage converter. Background Art

[0002] With the growth of global energy demand and the development of renewable energy technologies, energy storage systems and their related technologies have become increasingly important. As a key component connecting energy storage devices to the power grid, the energy storage converter plays an important role in regulating power supply and demand, and improving the stability and reliability of the power grid. The grid connection detection technology of the energy storage converter is one of the important links to ensure its safe and reliable operation. In recent years, in order to improve the performance and adaptability of the energy storage converter, many studies have been dedicated to developing more accurate and efficient grid connection detection methods. Traditional grid connection detection methods mainly rely on fixed test conditions and simple mathematical models. Although these methods can meet the basic detection requirements to a certain extent, they often seem inadequate when faced with complex and changing actual application scenarios. Especially when dealing with the grid connection problems of multiple types of energy storage devices, the limitations of traditional methods in data acquisition, preprocessing, and analysis are particularly prominent.

[0003] Although certain progress has been made in the grid connection detection of energy storage converters in the prior art, there are still some deficiencies. First, most of the existing grid connection detection methods are modeled based on limited historical data, which makes it difficult for them to cope with various uncertain factors that occur during actual operation. Second, traditional methods lack sufficient flexibility in feature extraction and combination, and cannot fully exploit the potential information in the operation data of the energy storage converter. For example, when dealing with data collected by multi-source sensors, how to effectively fuse different types of data and extract useful feature vectors from them is a challenge faced by current technologies. In addition, the data sets used in the model training process of the prior art are relatively single, which limits the generalization ability and prediction accuracy of the model. Summary of the Invention

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

[0005] Therefore, the present invention effectively solves the problems of single data source and inflexible feature extraction existing in the prior art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a grid connection detection method for an energy storage converter, which includes collecting the operation data of the energy storage converter through a multi-source sensor and performing preprocessing; training a generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data; fusing the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and performing feature extraction on the enhanced data pool to combine them into a feature vector; based on the feature vector, using a sliding window method to extract the feature vector at each time point and combine them into a three-dimensional array; constructing a preliminary grid connection detection model based on a long short-term memory network, and training the preliminary grid connection detection model through the enhanced data pool to obtain a grid connection detection model; based on the three-dimensional array, obtaining the grid connection status through the grid connection detection model.

[0007] As a preferred embodiment of the grid connection detection method for the energy storage converter of the present invention, wherein: the step of collecting the operation data of the energy storage converter through a multi-source sensor and performing preprocessing is as follows, Collect the input and output power, grid frequency, internal and environmental temperature of the converter, voltages on the DC side and AC side, and currents on the DC side and AC side of the energy storage converter to form the operation data of the energy storage converter; Perform data cleaning on the operation data of the energy storage converter to check for missing values in the dataset, convert the timestamps of all operation data of the energy storage converter into a unified date and time format, and perform normalization processing.

[0008] As a preferred embodiment of the grid connection detection method for the energy storage converter of the present invention, wherein: the step of training a generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data is as follows, Initialize the generator and discriminator models of the generative adversarial network, fix the generator, and update the discriminator weights by inputting the historical operation data of the energy storage converter and the virtual data generated by the generator into the discriminator. Fix the discriminator, and calculate the generator loss and update the generator weights by inputting the virtual data into the discriminator; When the discriminator loss and the generator loss are stable and the quality of the generated data meets the standard, obtain the trained generative adversarial network model; Based on the operation data of the energy storage converter input into the trained generative adversarial network model, obtain the virtual grid connection scenario data.

[0009] As a preferred embodiment of the grid connection detection method for the energy storage converter of the present invention, wherein: the step of fusing the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and performing feature extraction on the enhanced data pool to combine them into a feature vector is as follows, Based on the amount and distribution of virtual grid-connection scenario data, analyze through a visualization tool, set the ratio of generated data to real data, mix the generated virtual data and real data according to the set ratio, and verify the quality of the mixed data set through statistical analysis methods to obtain an enhanced data pool; Based on the enhanced data pool, extract features through principal component analysis and time series feature extraction methods. Based on the extracted features, arrange them according to the time series window and splice them by dimension, and finally combine them into feature vectors.

[0010] As a preferred solution of the grid-connection detection method for the energy storage converter described in the present invention, wherein: based on the feature vectors, extract the feature vectors at each time point through a sliding window method and combine them into a three-dimensional array. The specific steps are as follows. Based on the feature vectors, adjust the time step of each sliding window through the sliding window size and the required number of samples , generate a window from to , and extract the feature vectors corresponding to each time point; Convert the number of samples, time step, and number of features into a three-dimensional array.

[0011] As a preferred solution of the grid-connection detection method for the energy storage converter described in the present invention, wherein: based on the long short-term memory network, construct a preliminary grid-connection detection model, and train the preliminary grid-connection detection model through the enhanced data pool to obtain a grid-connection detection model. The specific steps are as follows. Use multiple LSTM layers to capture the dynamic changes in the time series, add a fully connected layer after the LSTM layer, and use the Sigmoid activation function to compress the output to the normal range to construct a preliminary grid-connection detection model; Divide the enhanced data pool into a training set and a test set through random sampling; Train the preliminary grid-connection detection model through the training set, initialize the weight parameters of the preliminary grid-connection detection model, and judge the performance of the preliminary grid-connection detection model by calculating evaluation indicators to obtain a grid-connection detection model; Use the test set to predict the results through the grid-connection detection model, visualize the prediction results of the grid-connection detection model, and compare them with the operation data of the energy storage converter to verify the effect of the grid-connection detection model to obtain a grid-connection detection model.

[0012] As a preferred solution of the grid-connection detection method for the energy storage converter described in the present invention, wherein: based on the three-dimensional array, obtain the grid-connection status through the grid-connection detection model. The specific steps are as follows. Based on the three-dimensional array, obtain the grid-connection detection probability value through the grid-connection detection model ; Set the grid connection state threshold through statistical analysis based on the historical grid connection detection probability value ; Evaluate the grid connection state based on the grid connection detection probability value and the grid connection state threshold; When the current moment is in the grid connection state; When the current moment is not in the grid connection state.

[0013] In a second aspect, the present invention provides a grid connection detection system for an energy storage converter, including grid connection detection of the energy storage converter, a virtual data module, a feature extraction module, a sliding window construction module, a preliminary grid connection detection module, and a grid connection detection module; a data acquisition module, which collects the operation data of the energy storage converter through multi-source sensors and performs preprocessing; a virtual data module, which trains a generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data; data fusion and feature extraction, which fuses the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and performs feature extraction on the enhanced data pool to combine into a feature vector; a sliding window construction module, which based on the feature vector, extracts the feature vector at each time point through the sliding window method and combines them into a three-dimensional array; a preliminary grid connection model module, which constructs a preliminary grid connection detection model based on a long short-term memory network and trains the preliminary grid connection detection model through the enhanced data pool to obtain a grid connection detection model; a prediction module, which based on the three-dimensional array, obtains the grid connection state through the grid connection detection model.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the grid connection detection method for the energy storage converter as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the grid connection detection method for the energy storage converter as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: By introducing the GAN technology, without relying on a large amount of real data, the high performance of the grid connection detection model is still ensured, the development cost is significantly reduced, the R & D cycle is shortened, and the flexibility and adaptability of the model in different grid environments are improved. Description of the Drawings

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of the grid connection detection method for the energy storage converter in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of the grid connection detection system for the energy storage converter in Embodiment 1.

[0020] Figure 3 It is a flowchart of the training of the generative adversarial network (GAN) in Embodiment 1.

[0021] Figure 4 It is a flowchart of the training of the long short-term memory network (LSTM) model in Embodiment 1. Specific Embodiments

[0022] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings in the specification.

[0023] Embodiment 1, referring to Figures 1 to 4 , is the first embodiment of the present invention. This embodiment provides a grid connection detection method for an energy storage converter, including the following steps: S1. Collect the operation data of the energy storage converter through multi-source sensors and perform preprocessing.

[0024] Collect the input and output power, grid frequency, internal and ambient temperature of the converter, voltages on the DC side and AC side, and currents on the DC side and AC side of the energy storage converter to form the operation data of the energy storage converter; It should be noted that the power sensor is used to measure the input and output power of the energy storage converter during the charging and discharging processes to ensure an accurate assessment of the energy conversion efficiency; the frequency sensor monitors the grid frequency to ensure the synchronization of the grid connection operation; the temperature sensor monitors the temperature inside and around the converter in real time to prevent overheating; the voltage and current sensors measure the voltages and currents on the DC side and AC side respectively to ensure the stability and safety of power transmission.

[0025] Perform data cleaning on the operation data of the energy storage converter, check for missing values in the dataset, convert the timestamps of all operation data of the energy storage converter to a unified date and time format, and perform normalization processing.

[0026] It should be noted that the specific expression is: ; wherein, is the normalized operation data of the energy storage converter, is the original operation data of the energy storage converter, is the minimum value in the operation data of the energy storage converter, is the maximum value in the operation data of the energy storage converter.

[0027] S2. Train the generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data.

[0028] Initialize the generator and discriminator models of the generative adversarial network, fix the generator, and update the discriminator weights by inputting the historical operation data of the energy storage converter and the virtual data generated by the generator into the discriminator. Fix the discriminator, calculate the generator loss by inputting the virtual data into the discriminator, and update the generator weights; It should be noted that the specific algorithm of GAN first defines and initializes the architectures of the generator and discriminator through deep neural networks, fixes the generator and updates the discriminator. Extract a batch of real data samples from the historical operation data of the energy storage converter, use the generator to generate a batch of virtual data samples based on random noise, input both the real data samples and the virtual data samples into the discriminator, and calculate the loss function of the discriminator (binary cross-entropy loss). Adjust the weight parameters of the discriminator according to the loss value. Then fix the discriminator and update the generator. Again, use the generator to generate a batch of virtual data samples based on new random noise. The generated virtual data samples are input into the discriminator in the current state. The discriminator gives the probability that these samples are considered real, calculate the loss function of the generator, and adjust the weight parameters of the generator according to the loss value. Continuously repeat these two steps to train the GAN model. When the quality of the generated data reaches the expected target, obtain the GAN model.

[0029] It should also be noted that in the construction process of the generator and discriminator, the generator receives a random noise vector (such as a 100-dimensional Gaussian noise) as input, learns the temporal dependence relationship through the LSTM layer, and is mapped to the target feature dimension (such as power, voltage, etc.) through the fully connected layer. The output layer uses the Tanh activation function to normalize the data to [-1, 1], align it with the preprocessed real data, and obtain the complete generator.

[0030] The discriminator receives temporal data, extracts local features (such as voltage mutations, power fluctuations) through convolutional layers (such as 4 convolutional layers, with the number of channels 64 → 128 → 256 → 512), compresses it through global pooling and then inputs it into the fully connected layer, and outputs a Sigmoid probability value (0~1) to judge the authenticity of the data, obtaining the complete discriminator.

[0031] Fix the generator, update the discriminator with real data and generated data to optimize its discrimination ability; fix the discriminator and update the generator to generate more realistic data. Adopt the Wasserstein loss combined with gradient penalty (GP). The discriminator loss includes the data discrimination error and the gradient constraint term (to prevent gradient explosion). The generator loss minimizes the discriminator's confidence in fake data. Use the Adam optimizer (discriminator learning rate 5e-5, generator 1e-4). Balance the capabilities of both sides through alternating iterations (e.g., update the discriminator 5 times / generator 1 time), add label smoothing (e.g., set the real label to 0.9 and the generated label to 0.1) and mini-batch discrimination to enhance data diversity. Verify the distribution matching of the generated data and the real data through statistical metrics (mean, variance, autocorrelation coefficient) and downstream task tests (such as the LSTM grid connection detection model) to complete the training of the generator and the discriminator.

[0032] The discriminator loss and the generator loss are stable and the quality of the generated data meets the standard, obtaining a trained generative adversarial network model; Based on the operation data of the energy storage converter input into the trained generative adversarial network model, virtual grid connection scenario data is obtained.

[0033] It should be noted that the fully trained generator can learn the distribution characteristics of the real operation data of the energy storage converter and generate virtual data highly similar to the actual working conditions on this basis.

[0034] S3. Fuse the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and perform feature extraction on the enhanced data pool to combine into a feature vector.

[0035] Based on the amount and distribution of the virtual grid connection scenario data, analyze through a visualization tool, set the ratio of the generated data to the real data, mix the generated virtual data and the real data according to the set ratio, and verify the quality of the mixed data set through statistical analysis methods to obtain the enhanced data pool; It should be noted that based on the quantity and time distribution of virtual grid-connection scenario data: the dynamic characteristics of data on the time axis (such as power fluctuation period, voltage mutation frequency), statistical distribution: the statistical characteristics of key parameters (such as mean, variance, skewness, kurtosis), physical feature distribution: the distribution of core parameters strongly related to the grid-connection state (such as voltage harmonic distortion rate, frequency deviation, active power response time), and operating condition distribution: the data coverage ratio of different grid operating states (such as steady state, transient state, fault state). Use visualization tools to analyze the physical characteristics (such as voltage harmonic distortion rate, frequency deviation, power volatility), statistical characteristics (mean, variance, skewness, kurtosis), time series characteristics (autocorrelation coefficient, sliding window dynamic trend), and correlation characteristics (Pearson correlation coefficient, mutual information) of the energy storage converter grid-connection detection scenario to set the optimal mixing ratio of the generated data and the real data. Quantify the distribution difference between the virtual data and the real data in key features (such as voltage harmonic distortion rate, frequency deviation) based on Jensen-Shannon divergence (JSD). If JSD < 0.2, it is determined that the distributions are consistent, and the virtual data proportion is set to 60%; if JSD ∈ [0.2, 0.4), dynamically reduce the proportion according to the virtual proportion (such as 30% when JSD = 0.3); if JSD ≥ 0.4, it is necessary to regenerate the data, and the virtual data proportion is additionally reduced by 20%. Verify the quality of the mixed data through the K-S test and the correlation coefficient difference, and finally iteratively adjust to obtain the optimal mixing ratio. Mix the generated virtual data and the real data according to this ratio to form an enhanced data pool. Verify the quality of the data set by comparing the mean, variance, distribution form (such as histogram, kernel density estimation) of the virtual data and the real data, and the correlation between features (such as the correlation coefficient between power and voltage). If the differences in key statistical indicators between the virtual data and the real data are within the preset thresholds (such as mean error ≤ 5%, correlation coefficient deviation ≤ 0.1), and the diversity of real operating conditions (such as extreme temperature, grid frequency fluctuation and other scenarios) is covered, the data set quality is considered to meet the standard.

[0036] Based on the enhanced data pool, extract features through principal component analysis and time series feature extraction methods. Based on the extracted features, arrange them according to the time series window and splice them according to the dimension, and finally combine them into a feature vector.

[0037] It should be noted that the principal component analysis method is used to reduce the dimension of the operation data of the energy storage converter collected by multi-source sensors, and the principal components with the cumulative proportion of explained variance exceeding the preset cumulative explained variance proportion threshold (e.g., 95%) are extracted to form low-dimensional principal component features; the statistical features (including the mean, variance, and extreme values within the sliding window), trend features (including the linear regression slope and the change amount of moving average), periodic features (including the fundamental frequency extracted by Fourier transform and the period length calculated by the autocorrelation function), and volatility features (including the coefficient of variation and the range within the window) of each time point are extracted from the same data through the time series feature extraction method, and the principal component features and time series features of each time point are horizontally concatenated according to the feature dimension to form a multi-dimensional feature vector at this time point.

[0038] It should also be noted that under the guidance of domain knowledge, an empirical range value of the cumulative proportion of explained variance (usually between 80% - 99%) is selected, and the final value (such as 95%) is determined through actual tests (such as analyzing the influence of principal component features under different thresholds on the accuracy of subsequent health state recognition tasks) to obtain the preset cumulative explained variance proportion threshold.

[0039] S4. Based on the feature vector, through the sliding window method, the feature vector of each time point is extracted and combined into a three-dimensional array.

[0040] Based on the feature vector, the time step of each sliding window is adjusted by the sliding window size and the required number of samples , generating a window from to , and extracting the feature vector corresponding to each time point; It should be noted that the feature vector contains the PCA principal components (such as power-voltage coupling features) and time series features (mean, variance, trend, periodicity, etc.) of each time point. The time step of each sliding window is adjusted by setting the size of the sliding window and the required number of samples, generating a series of continuous windows from the start time point to the end time point. First, determine the length of the sliding window (e.g., 5 time steps), and then generate windows one by one on the time series according to the set time step (e.g., sliding 1 time step each time) to obtain the feature vector of each time point, which is used as the unit of the third dimension (number of features) in the three-dimensional array.

[0041] Convert the number of samples, time step, and number of features into a three-dimensional array.

[0042] It should be noted that first, the time step of each sample is determined (for example, each sample contains 5 consecutive time points), and then the number of features at each time point is determined (for example, there are 10 features in total such as the power change rate, voltage volatility, etc.). Next, these features are arranged in chronological order and concatenated by dimension to form a three-dimensional array with a shape of (number of samples, time step, number of features). For example, if there are 1000 samples, each sample contains 5 time steps, and each time step has 10 features, then the shape of the final generated three-dimensional array will be (1000, 5, 10). This process not only preserves the time series characteristics of the data but also provides a structured input format for subsequent deep learning models (such as the long short-term memory network LSTM), enabling the model to effectively capture and learn the dynamic change patterns in the time series.

[0043] It should also be noted that the data formula within the window is: ; where represents the data of the th window, is the step size (the distance of each slide), represents the sliding window size.

[0044] S5. Construct a preliminary grid connection detection model based on the long short-term memory network, and train the preliminary grid connection detection model through the enhanced data pool to obtain the grid connection detection model.

[0045] Use multiple LSTM layers to capture the dynamic changes in the time series, add a fully connected layer after the LSTM layer, and use the Sigmoid activation function to compress the output to the normal range to construct the preliminary grid connection detection model; It should be noted that multiple LSTM layers are used to capture the time series dynamic changes in the operation data of the energy storage converter. Through these LSTM layers, complex patterns and dependencies within a long time span can be effectively learned and memorized. A fully connected layer is added after the LSTM layer, which is responsible for integrating the feature information from the LSTM layer and using the Sigmoid activation function to compress the output to the range between 0 and 1, thereby generating the grid connection status probability value at each time point. Specifically, the LSTM layer processes the time correlation of the input data through its unique gating mechanism to capture the change trends and features of the energy storage converter at different time points; while the fully connected layer further extracts and integrates these features, and converts the model output into an easily interpretable probability value through the Sigmoid function.

[0046] Divide the enhanced data pool into a training set and a test set by random sampling; It should be noted that usually 80% of the enhanced data pool is used for training, and 20% of the enhanced data pool is used for testing.

[0047] The preliminary grid connection detection model is trained using a training set, the weight parameters of the preliminary grid connection detection model are initialized, and the performance of the preliminary grid connection detection model is judged by calculating evaluation metrics to obtain the grid connection detection model. It should be noted that when initializing the weight parameters of the model, the weights of the model are usually initialized with a normal distribution. Samples in the training set are input into the preliminary grid connection detection model through forward propagation, and the output results are obtained through network calculations. Based on the difference between the prediction results of the preliminary grid connection detection model and the true labels, the loss value of the current batch is calculated using the selected loss function, and the backpropagation algorithm is executed according to the calculated loss value. The gradients of each parameter are calculated and the parameters of the preliminary grid connection detection model are updated using an optimizer to minimize the cross-entropy loss function as much as possible.

[0048] The specific expression should be noted as: ; Among them, is the true label, is the predicted probability of the preliminary grid connection detection model for the positive class, is the loss value.

[0049] The preliminary grid connection detection model is continuously iterated, evaluation metrics (such as accuracy, precision, recall, and F1 score) are calculated to monitor the performance of the model, and based on the above calculated evaluation metrics, the performance of the model is analyzed and optimized training is continuously carried out.

[0050] The specific expression should be noted as: ; Among them, is the number of samples that are actually positive and are correctly predicted as positive, is the number of samples that are actually negative and are correctly predicted as negative, is the number of samples that are actually negative but are wrongly predicted as positive, is the number of samples that are actually positive but are wrongly predicted as negative, is the accuracy.

[0051] The training of the grid connection detection model is an iterative process, and steps such as forward propagation, loss calculation, backpropagation, and parameter update need to be repeatedly executed in multiple loops until the loss value no longer significantly decreases to obtain the trained grid connection detection model.

[0052] The prediction results of the test set are used through the grid connection detection model, the prediction results of the grid connection detection model are visualized, and compared with the operation data of the energy storage converter to verify the effect of the grid connection detection model and obtain the grid connection detection model.

[0053] It should be noted that the test set is input into the trained grid connection detection model to generate the grid connection status prediction results at each time point. These prediction results are visualized, and by plotting the curve of the prediction probability value changing with time, the prediction performance of the grid connection detection model can be intuitively displayed.

[0054] S6. Based on the three-dimensional array, obtain the grid connection status through the grid connection detection model.

[0055] Based on the three-dimensional array, obtain the grid connection detection probability value through the grid connection detection model ; Set the grid connection status threshold according to the historical grid connection detection probability value through statistical analysis methods ; Evaluate the grid connection status based on the grid connection detection probability value and the grid connection status threshold; When the current moment is in the grid-connected state; When the current moment is not in the grid-connected state.

[0056] It should be noted that the grid connection detection probability value at each time point is obtained through the trained grid connection detection model. Specifically, the three-dimensional array is input into the grid connection detection model, and the model outputs the grid connection probability value at each time point, indicating the possibility of being in the grid-connected state at the current moment. Next, according to the historical grid connection detection probability value, a reasonable grid connection status threshold (such as 0.5) is set using statistical analysis methods (such as calculating the mean, standard deviation, or percentile). This threshold is used to distinguish between grid-connected and non-grid-connected states: when the grid connection detection probability value is greater than or equal to the threshold (i.e., P(t) >= threshold), it is determined that the current moment is in the grid-connected state; when the grid connection detection probability value is less than the threshold (i.e., P(t) < threshold), it is determined that the current moment is not in the grid-connected state. This process ensures the accuracy and consistency of the grid connection status judgment through systematic threshold setting and evaluation, enabling the grid connection detection model to reliably identify the grid connection status of the energy storage converter under complex and changing actual working conditions, thereby improving the stability and reliability of the overall system. Finally, this method provides a clear and operable decision-making basis, is applicable to the actual grid connection detection application of the energy storage converter, and significantly improves the credibility and practicality of the grid connection detection model through comprehensive evaluation and visualization analysis.

[0057] This embodiment also provides a grid connection detection system for an energy storage converter, including: a data acquisition module, a virtual data module, a feature extraction module, a sliding window construction module, a preliminary grid connection model module, and a prediction module; The data acquisition module collects the operation data of the energy storage converter through multi-source sensors and performs preprocessing; A virtual data module trains a generative adversarial network model based on the preprocessed historical operating data of an energy storage converter to obtain virtual grid connection scenario data; Data fusion and feature extraction: fuse the virtual grid connection scenario data with the operating data of the energy storage converter to form an enhanced data pool, and perform feature extraction on the enhanced data pool to combine into a feature vector; A sliding window construction module extracts the feature vectors at each time point based on the feature vector through the sliding window method and combines them into a three-dimensional array; A preliminary grid connection model module constructs a preliminary grid connection detection model based on a long short-term memory network, and trains the preliminary grid connection detection model through the enhanced data pool to obtain a grid connection detection model; A prediction module obtains the grid connection status based on the three-dimensional array through the grid connection detection model.

[0058] This embodiment also provides a computer device applicable to the grid connection detection method of an energy storage converter, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the grid connection detection method of the energy storage converter proposed in the above embodiment.

[0059] This computer device may be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0060] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the grid connection detection method of the energy storage converter as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0061] In summary, the present invention: by introducing GAN technology, without relying on a large amount of real data, still ensures the high performance of the grid connection detection model, significantly reduces the development cost, shortens the R & D cycle, and improves the flexibility and adaptability of the model in different grid environments.

[0062] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A grid connection detection method for an energy storage converter, characterized in that: including collecting the operation data of the energy storage converter through multi-source sensors and performing preprocessing; training a generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data; fusing the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and extracting features from the enhanced data pool and combining them into a feature vector; extracting the feature vector at each time point based on the feature vector through a sliding window method and combining them into a three-dimensional array; constructing a preliminary grid connection detection model based on a long short-term memory network, and training the preliminary grid connection detection model through the enhanced data pool to obtain a grid connection detection model; obtaining the grid connection status based on the three-dimensional array through the grid connection detection model.

2. The grid connection detection method of the energy storage converter according to claim 1, characterized in that: The step of collecting the operation data of the energy storage converter through multi-source sensors and performing preprocessing is as follows. Collecting the input and output power, grid frequency, internal and environmental temperature of the converter, voltages on the DC side and AC side, and currents on the DC side and AC side of the energy storage converter to form the operation data of the energy storage converter; Performing data cleaning on the operation data of the energy storage converter to check for missing values in the dataset, converting the timestamps of all operation data of the energy storage converter into a unified date and time format, and performing normalization processing.

3. The grid connection detection method of the energy storage converter according to claim 2, characterized in that: The step of training a generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data is as follows. Initializing the generator and discriminator models of the generative adversarial network, fixing the generator, and updating the discriminator weights by inputting the historical operation data of the energy storage converter and the virtual data generated by the generator into the discriminator. Fixing the discriminator, calculating the generator loss by inputting the virtual data into the discriminator, and updating the generator weights; When the discriminator loss and the generator loss are stable and the quality of the generated data meets the standard, a trained generative adversarial network model is obtained; Based on the operation data of the energy storage converter input into the trained generative adversarial network model, virtual grid connection scenario data is obtained.

4. The grid connection detection method of the energy storage converter according to claim 3, characterized in that: The step of fusing the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and extracting features from the enhanced data pool and combining them into a feature vector is as follows. Based on the amount and distribution of the virtual grid connection scenario data, analyzing through a visualization tool, setting the ratio of the generated data to the real data, mixing the generated virtual data and the real data according to the set ratio, and verifying the quality of the mixed dataset through statistical analysis methods to obtain an enhanced data pool; Based on the enhanced data pool, extracting features through principal component analysis and time series feature extraction methods, arranging the extracted features according to the time series window, and splicing them according to the dimension, and finally combining them into a feature vector.

5. The grid connection detection method of the energy storage converter according to claim 4, characterized in that: The step of extracting the feature vector at each time point based on the feature vector through a sliding window method and combining them into a three-dimensional array is as follows. Based on the feature vectors, adjust the time step of each sliding window by the sliding window size and the required number of samples , generate a window from to , and extract the feature vectors corresponding to each time point; Converting the number of samples, time steps, and number of features into a three-dimensional array.

6. The grid connection detection method of the energy storage converter according to claim 5, characterized in that: The step of constructing a preliminary grid connection detection model based on a long short-term memory network and training the preliminary grid connection detection model through the enhanced data pool to obtain a grid connection detection model is as follows. Utilize multiple LSTM layers to capture the dynamic changes in the time series. Add a fully connected layer after the LSTM layers, and use the Sigmoid activation function to compress the output into the normal range to construct a preliminary grid connection detection model; Divide the enhanced data pool into a training set and a test set by random sampling; Train the preliminary grid connection detection model with the training set, initialize the weight parameters of the preliminary grid connection detection model, judge the performance of the preliminary grid connection detection model by calculating evaluation indicators, and obtain the grid connection detection model; Use the test set to predict the results through the grid connection detection model, visualize the prediction results of the grid connection detection model, and compare them with the operation data of the energy storage converter to verify the effect of the grid connection detection model and obtain the grid connection detection model.

7. The grid connection detection method of the energy storage converter according to claim 6, characterized in that: Based on the three-dimensional array, obtain the grid connection status through the grid connection detection model. The specific steps are as follows: Based on a three-dimensional array, obtain the grid connection detection probability value through a grid connection detection model ; Set the grid connection status threshold through statistical analysis based on the historical grid connection detection probability value ; Evaluate the grid connection status based on the grid connection probability value through the grid connection status threshold; When the current moment is in the grid-connected state; When the current moment is not in the grid-connected state.

8. A grid connection detection system for an energy storage converter, based on the grid connection detection method of the energy storage converter according to any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a virtual data module, a feature extraction module, a sliding window construction module, a preliminary grid connection model module, and a prediction module; The data acquisition module collects the operation data of the energy storage converter through multi-source sensors and performs preprocessing; The virtual data module trains the generative adversarial network model based on the preprocessed historical operation data of the energy storage converter to obtain virtual grid connection scenario data; Data fusion and feature extraction: fuse the virtual grid connection scenario data with the operation data of the energy storage converter to form an enhanced data pool, and perform feature extraction on the enhanced data pool to combine into a feature vector; The sliding window construction module extracts the feature vectors at each time point based on the feature vector through the sliding window method and combines them into a three-dimensional array; The preliminary grid connection model module constructs a preliminary grid connection detection model based on the long short-term memory network, and trains the preliminary grid connection detection model through the enhanced data pool to obtain the grid connection detection model; The prediction module obtains the grid connection status based on the three-dimensional array through the grid connection detection model.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the grid connection detection method of the energy storage converter according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the grid connection detection method of the energy storage converter according to any one of claims 1 to 7.

Images