Method and system for monitoring zero-current string of string type inverter

Abstract

The invention relates to the technical field of string type inverter monitoring, in particular to a string type inverter zero-current string monitoring method and system, and the method comprises the steps: obtaining the operation data of each string inverter; performing feature extraction on the operation data of each string inverter to obtain multiple pieces of feature data, including current features, voltage and power features, environment features and dynamic features; combining the string inverter feature data, inputting the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm, and solving a zero-current analysis result; and verifying the zero-current analysis result based on preset zero-current judgment logic, and outputting a final zero-current monitoring result. Through machine learning and multi-feature data fusion, real-time monitoring of the string inverter is realized, the fault detection accuracy and response speed are improved, the maintenance cost is reduced, and intelligent development of the photovoltaic industry is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring of string inverters, and in particular, to a method and system for monitoring zero-current strings of string inverters. Background Art

[0002] With the rapid development of renewable energy, string inverters, as an important part of photovoltaic power generation systems, the stability and reliability of their performance have become increasingly important. String inverters convert the direct current collected by solar cells into alternating current, and then deliver the power to the power grid. In order to improve the power generation efficiency, real-time monitoring of the operating status of string inverters has become a research hotspot. String inverters based on traditional monitoring methods often rely on manual inspections and empirical judgments, and often face problems such as lagging data acquisition and incomplete information. The monitoring means are often limited to regular inspections of current and voltage, and cannot reflect the dynamic performance of the inverter in real time. Not only are potential faults easily missed, but also there is a lack of timely response when a fault occurs, resulting in increased power generation losses and maintenance costs. In addition, the subjectivity of empirical judgment limits the accuracy of fault prediction and cannot meet the requirements of modern efficient and intelligent energy management. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for monitoring zero-current strings of string inverters, which realizes real-time monitoring of string inverters through machine learning and multi-feature data fusion, improves the accuracy of fault detection and response speed, reduces maintenance costs, and promotes the intelligent development of the photovoltaic industry.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for monitoring zero-current strings of string inverters, the steps of the method include:

[0006] Obtain the operating data of each string inverter;

[0007] Extract features from the operating data of each string inverter to obtain a plurality of feature data, including: current features, voltage and power features, environmental features, and dynamic features;

[0008] Combine the feature data of the string inverter, and input the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm to solve the zero-current analysis result;

[0009] Verify the zero-current analysis result based on a preset zero-current judgment logic, and output the final zero-current monitoring result.

[0010] Optionally, the obtaining of the operating data of each string inverter further includes a data processing process, specifically:

[0011] For the continuous missing values in the operation data of each string inverter, linear interpolation is used for filling. If the missing ratio of the data segment in the operation data of each string inverter is greater than the ratio threshold, this data column will be deleted;

[0012] Based on the Z-score module, outliers in the operation data of each string inverter are identified, and based on the outlier identification results, they are replaced by the adjacent average values of the outliers;

[0013] Through the ADF test module, the stationarity of the time series in the operation data of each string inverter is confirmed. If the stationarity of the time series does not reach the expected value, the operation data of each string inverter will be differenced to complete the data processing of the operation data of each string inverter.

[0014] Optionally, for the current characteristics, the feature extraction steps are as follows:

[0015] The current data in the operation data of each string inverter is segmented according to the time window;

[0016] Calculate the current mean, maximum value, minimum value and standard deviation of each time window;

[0017] Solve the current change rate to characterize the current characteristics of the operation data of each string inverter.

[0018] Optionally, for the voltage and power characteristics, the feature extraction steps are as follows:

[0019] The voltage and power data in the operation data of each string inverter are segmented according to the time window;

[0020] Calculate the voltage mean, maximum value, minimum value and standard deviation of each time window;

[0021] Solve the voltage change rate;

[0022] Based on the calculation results of the voltage change rate, solve the power mean, maximum value, minimum value, standard deviation and change rate. The voltage change rate and the power change rate characterize the voltage and power characteristics of the operation data of each string inverter.

[0023] Optionally, for the environmental characteristics, the feature extraction steps are as follows:

[0024] The environmental data in the operation data of each string inverter are segmented according to the time window;

[0025] Calculate the temperature mean, maximum value, minimum value and standard deviation of each time window;

[0026] Calculate the humidity mean, maximum value, minimum value and standard deviation of each time window;

[0027] Calculate the mean, maximum, minimum, and standard deviation of the radiation intensity for each time window;

[0028] Solve for the temperature change rate, humidity change rate, and radiation intensity change rate to characterize the environmental characteristics of the operating data of each string inverter.

[0029] Optionally, for the dynamic characteristics, the feature extraction steps are as follows:

[0030] Record the mutation events of the current data in the operating data of each string inverter for each time window, including the mutation amplitude and the number of times;

[0031] Solve for the average rate of current change based on the mutation amplitude and the number of times of the current data to characterize the dynamic characteristics of the operating data of each string inverter.

[0032] Optionally, the zero-current string analysis model constructed based on the machine learning algorithm is specifically as follows:

[0033] Combine the current characteristics, voltage and power characteristics, environmental characteristics, and dynamic characteristics to form a multi-dimensional combined feature set, and perform format conversion on the multi-dimensional combined feature set;

[0034] Construct a zero-current string analysis model based on the VGG algorithm;

[0035] The zero-current string analysis model is provided with:

[0036] An input layer that inputs the multi-dimensional combined feature set after format conversion;

[0037] A first convolutional layer that applies 64 3x3 convolutional kernels to the multi-dimensional combined feature set, performs non-linear mapping through the ReLU activation function, and outputs a first feature map;

[0038] A first pooling layer that performs 2x2 max pooling operation on the first feature map;

[0039] A second convolutional layer that applies 128 3x3 convolutional kernels to the first feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs a second feature map;

[0040] A second pooling layer that performs 2x2 max pooling operation on the second feature map;

[0041] A third convolutional layer that applies 256 3x3 convolutional kernels to the second feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs a third feature map;

[0042] A third pooling layer that performs 2x2 max pooling operation on the third feature map;

[0043] The Flatten layer flattens the third feature map after the max pooling operation into a one-dimensional vector and outputs the flattened feature vector.

[0044] The fully connected layer processes the flattened feature vector and outputs the activation result of the fully connected layer.

[0045] The Dropout layer applies Dropout to the activation result of the fully connected layer.

[0046] The output layer sets a single neuron and limits the output result between 0 and 1 through the Sigmoid activation function, and outputs the zero current prediction probability.

[0047] Optionally, verifying the zero current analysis result based on a preset zero current judgment logic is specifically as follows:

[0048] Obtain the zero current prediction result output by the output layer, including: the zero current prediction probability of each input sample, where the sample is defined as the operation data of the string inverter at a specific time point.

[0049] Define the first current membership function to evaluate the relationship between the sample current value and the set zero current value, and output its membership value on the first current feature.

[0050] Define the second current membership function to evaluate the relationship of the sample current value within the normal operation range, and output its membership value on the second current feature.

[0051] Define the third current membership function to evaluate whether the sample current value exceeds the set current value, and output its membership value on the third current feature.

[0052] Calculate the comprehensive judgment score by combining the zero current prediction probability and the membership value.

[0053] When the comprehensive judgment score is greater than or equal to the set score threshold, it is determined that there is a zero current phenomenon.

[0054] When the comprehensive judgment score is lower than the set score threshold, it is determined that there is no zero current phenomenon.

[0055] Optionally, the calculating the comprehensive judgment score by combining the zero current prediction probability and the membership value is specifically as follows:

[0056] Calculate the first current membership of each sample.

[0057] Calculate the second current membership of each sample.

[0058] Calculate the third current membership of each sample.

[0059] Fuse the first current membership degree, the second current membership degree, and the third current membership degree according to the set weight coefficient to obtain the comprehensive membership degree;

[0060] Combine the zero-current prediction probability and the comprehensive membership degree to obtain the comprehensive judgment score.

[0061] A zero-current string monitoring system for string inverters, comprising:

[0062] A data acquisition unit that acquires the operating data of each string inverter;

[0063] A feature extraction unit that extracts features from the operating data of each string inverter to obtain a plurality of feature data, including: current features, voltage and power features, environmental features, and dynamic features;

[0064] A model calculation unit that combines the feature data of the string inverter, and inputs the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm to solve the zero-current analysis result;

[0065] A verification unit that verifies the zero-current analysis result based on a preset zero-current judgment logic and outputs the final zero-current monitoring result.

[0066] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0067] By integrating a variety of feature data and combining a zero-current string analysis model constructed based on a machine learning algorithm, the present invention can achieve accurate real-time monitoring and analysis of string inverters. The real-time collected current, voltage, power, and environmental data can comprehensively reflect the operating conditions of the inverter, not only improving the speed and accuracy of fault detection, but also greatly reducing the error risk brought by manual inspections. Using a preset zero-current judgment logic ensures that an alarm is quickly issued when a fault occurs, and maintenance is carried out in a timely manner, greatly improving the reliability and power generation efficiency of the system. Description of the Drawings

[0068] Figure 1 It is a schematic flow chart of a zero-current string monitoring method for string inverters provided by the present invention;

[0069] Figure 2 It is a schematic flow chart of the calculation of the comprehensive judgment score provided by the present invention;

[0070] Figure 3 It is a schematic principle diagram of a zero-current string monitoring system for string inverters provided by the present invention. Detailed Embodiments

[0071] To make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, what is described is a part, rather than all, of the present invention. Components of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0072] As Figure 1 shown, the present invention provides one of the embodiments: a zero-current string monitoring method for a string inverter. The steps of the method include:

[0073] Obtain the operation data of each string inverter;

[0074] Extract features from the operation data of each string inverter to obtain a plurality of feature data, including: current features, voltage and power features, environmental features and dynamic features;

[0075] Combine the feature data of the string inverter, and input the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm to solve the zero-current analysis result;

[0076] Verify the zero-current analysis result based on a preset zero-current judgment logic, and output the final zero-current monitoring result.

[0077] In this embodiment, in order to achieve efficient monitoring and fault detection of string inverters, especially in photovoltaic power generation systems. This embodiment uses a high-precision data acquisition system to obtain the operating data of each string inverter in real time, including multiple dimensions such as current, voltage, power, and environmental parameters. To ensure the accuracy and real-time nature of the data, high-frequency sensors are used to accurately monitor the current and voltage output by the inverter, and at the same time, the surrounding environmental conditions such as temperature, humidity, and light intensity are collected. In addition, the dynamic responses of the inverter, including states such as startup, shutdown, and fault recovery, will also be recorded incidentally to provide rich background information for subsequent analysis. After the data is obtained, feature extraction is performed on the operating data. Feature extraction includes multiple aspects: extracting relevant features from the current data, such as average current, peak current, and standard deviation, etc., which can reflect the performance of the inverter in normal and abnormal operating states. Analyzing the voltage and power data to extract features such as instantaneous power, average power, and power fluctuation, so as to evaluate the output efficiency and stability of the inverter. For environmental features, information such as temperature change and climate factors is extracted in combination with environmental data to study its impact on the performance of the inverter. The extraction of dynamic features focuses on the operating performance of the inverter under different working conditions, including its startup time, response speed, and fault recovery ability, etc., to evaluate its reliability. All the obtained feature data will be standardized and combined to form a unified data set. Data standardization is to eliminate the dimensional differences between different features so that they are convenient for comparison and analysis within the same range. The combined feature data will be used as input and transmitted to the zero-current string analysis model constructed based on machine learning algorithms. It can be understood that the zero-current string analysis model is trained using historical data, and during the training process, the model will continuously adjust parameters to improve the prediction ability for unknown data. In the process of data training and verification, techniques such as cross-validation are used to ensure that the model has good generalization ability, thereby improving the prediction accuracy and reliability. For the prediction results, the preset zero-current judgment logic will be applied to verify the results, and the final monitoring results will be output in the form of a report, including the operating status and judgment basis of each inverter, and providing strong data support for subsequent maintenance and operation.

[0078] Furthermore, the obtaining of the operating data of each string inverter also includes a data processing process, specifically:

[0079] For the continuous missing values in the operating data of each string inverter, linear interpolation is used to fill them. If the missing proportion of the data segment in the operating data of each string inverter is greater than the proportion threshold, this data column will be deleted;

[0080] Based on the Z-score module, outliers in the operating data of each string inverter are identified, and based on the outlier identification results, they are replaced by the adjacent average values of the outliers;

[0081] The stationarity of the time series in the operating data of each string inverter is confirmed through the ADF test module. If the stationarity of the time series does not reach the expected value, the operating data of each string inverter is differenced to complete the data processing of the operating data of each string inverter.

[0082] In this embodiment, for the continuous missing values in the operating data of each string inverter, linear interpolation method will be applied to fill them. Specifically, by using adjacent known data points, the missing values are estimated based on the linear relationship, thus effectively retaining the continuity and trend of the data. Subsequently, this embodiment analyzes the outliers in the operating data of each string inverter based on the Z-score module. Z-score is a standardized index that measures the deviation between a data point and the mean of the data set. By setting a Z-score threshold, the data points that significantly deviate from the normal range can be effectively identified. To verify the stationarity of the time series of the operating data of each string inverter, the Augmented Dickey-Fuller (ADF) test module will be used to judge the stationarity of the time series data. If the test result shows that the time series does not reach the expected stationarity, the operating data will be differenced. Differencing can effectively eliminate trends and seasonality by calculating the difference between a data point and its previous point, making the time series more tend to be stationary.

[0083] In the specific implementation of this embodiment, for the current feature, the feature extraction steps are as follows:

[0084] The current data in the operating data of each string inverter is segmented according to a time window;

[0085] Calculate the mean, maximum, minimum and standard deviation of the current for each time window;

[0086] Solve the current change rate to characterize the current feature of the operating data of each string inverter.

[0087] In this embodiment, the current data in the operating data of each string inverter is segmented according to a time window, and the length of each time window is set to T w , and the current statistical features, including mean, maximum, minimum and standard deviation, are calculated for each time window. The specific calculation method is: the current mean I mean is where I i represents the i-th current data point within the time window, and N is the total number of current data points within the time window. The maximum value I max and the minimum value I min are respectively I max =max(I1, I2, …, I N ) and I min= min(I1, I2, …, I N ), the standard deviation I of the current std is The current change rate R is calculated by I the formula: characterizes the current characteristics of the string inverter operation data.

[0088] For the voltage and power characteristics, the feature extraction steps are as follows:

[0089] Divide the voltage and power data in the operation data of each string inverter according to a time window;

[0090] Calculate the voltage mean value, maximum value, minimum value and standard deviation of each time window;

[0091] Solve the voltage change rate;

[0092] Based on the calculation result of the voltage change rate, solve the power mean value, maximum value, minimum value, standard deviation and change rate. The voltage change rate and power change rate characterize the voltage and power characteristics of the operation data of each string inverter.

[0093] In this embodiment, the voltage and power data of each string inverter are divided according to the same time window T w for segmentation. Within each time window, the calculated characteristics of the voltage include the mean value, maximum value, minimum value and standard deviation. The voltage mean value V mean The formula for is The maximum value V of the voltage max and the minimum value V min are respectively V max = max(V1, V2, …, V N ) and V min = min(V1, V2, …, V N ), the standard deviation V of the voltage std is The formula for calculating the voltage change rate R V is: Based on the voltage change rate, the mean value, maximum value, minimum value and standard deviation of the power will be calculated. Among them, the power mean value P mean is The maximum value P of the power max and the minimum value P min are respectively P max = max(P1, P2, ..., P N ) and P min = min(P1, P2, ..., P N ), the standard deviation P of the power std is The power change rate RP The formula for the voltage change rate R V and the power change rate R P jointly characterize the voltage and power characteristics of the string inverter.

[0094] For the described environmental characteristics, the feature extraction steps are as follows:

[0095] Segment the environmental data in the operation data of each string inverter according to a time window;

[0096] Calculate the temperature mean, maximum value, minimum value, and standard deviation for each time window;

[0097] Calculate the humidity mean, maximum value, minimum value, and standard deviation for each time window;

[0098] Calculate the radiation intensity mean, maximum value, minimum value, and standard deviation for each time window;

[0099] Solve for the temperature change rate, humidity change rate, and radiation intensity change rate to characterize the environmental characteristics of the operation data of each string inverter.

[0100] In this embodiment, segment the environmental data of each string inverter according to the time window T w Perform segmentation. Analyze the statistical characteristics of temperature, humidity, and radiation intensity for each time window. In this embodiment, taking temperature as an example, the temperature mean T mean is calculated as The maximum value of temperature T max and the minimum value of temperature T min are respectively T max =max(T1,T2,...,T N ) and T min =min(T1,T2,…,T N ), and the standard deviation of temperature T std is Similarly, the radiation intensity mean The maximum value of radiation intensity R max =max(R1,R2,...,R N ) and the minimum value of radiation intensity R min =min(R1,R2,...,R N ), and the standard deviation of radiation intensity is By calculating the fluctuations of temperature, humidity, and radiation intensity, the change rates are obtained: The formulas for the temperature change rate, humidity change rate, and radiation intensity change rate are respectively: Characterize the environmental characteristics of the operation data of each string inverter.

[0101] For the described dynamic characteristics, the feature extraction steps are as follows:

[0102] Record the mutation events of the current data in the operation data of each string inverter in each time window, including the mutation amplitude and the number of times;

[0103] Solve the average rate of current change according to the mutation amplitude and the number of times of the current data, so as to characterize the dynamic characteristics of the operation data of each string inverter.

[0104] In this embodiment, by recording the mutation events of the current data in the operation data of each string inverter in each time window, including the amplitude of the mutation and the number of occurrences. The mutation amplitude is calculated as the absolute value of the difference between two adjacent data points: ΔI i =|I i -I i-1 |, record the number of times n events that the mutation amplitude exceeds the set threshold ΔT. Its calculation formula is: Calculate the average rate of current change R dI : Characterize the dynamic characteristics of each string inverter.

[0105] In the specific application of this embodiment, the zero-current string analysis model constructed based on the machine learning algorithm is specifically:

[0106] Combine the current characteristics, voltage and power characteristics, environmental characteristics and dynamic characteristics to form a multi-dimensional combined feature set, and perform format conversion on the multi-dimensional combined feature set;

[0107] Construct a zero-current string analysis model based on the VGG algorithm;

[0108] The zero-current string analysis model is provided with:

[0109] An input layer, inputting the multi-dimensional combined feature set after format conversion;

[0110] The first convolutional layer applies 64 3x3 convolutional kernels to the multi-dimensional combined feature set, performs non-linear mapping through the ReLU activation function, and outputs the first feature map;

[0111] The first pooling layer performs 2x2 max pooling operation on the first feature map;

[0112] The second convolutional layer applies 128 3x3 convolutional kernels to the first feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs the second feature map;

[0113] The second pooling layer performs 2x2 max pooling operation on the second feature map;

[0114] The third convolutional layer applies 256 3x3 convolutional kernels to the second feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs the third feature map;

[0115] The third pooling layer performs a 2x2 max pooling operation on the third feature map;

[0116] The Flatten layer flattens the third feature map after the max pooling operation into a one-dimensional vector and outputs the flattened feature vector;

[0117] The fully connected layer processes the flattened feature vector and outputs the activation result of the fully connected layer;

[0118] The Dropout layer applies Dropout to the activation result of the fully connected layer;

[0119] The output layer sets a single neuron, restricts the output result between 0 and 1 through the Sigmoid activation function, and outputs the zero-current prediction probability.

[0120] In the above implementation, in this embodiment, the current characteristics, voltage and power characteristics, environmental characteristics, and dynamic characteristics are systematically combined to form a multi-dimensional combined feature set. After the feature set is prepared, a deep learning model dedicated to zero-current string analysis is constructed based on the VGG (Visual Geometry Group) algorithm. Through multi-level convolution and pooling operations, it can effectively extract the deep information in the input features. The input layer of the zero-current string analysis model directly receives the multi-dimensional combined feature set after format conversion to ensure the complete transmission of each feature. The first convolutional layer of the zero-current string analysis model applies multiple 3x3 convolutional kernels to the input features to extract the lower-level local information in the features and realizes non-linear mapping through the ReLU activation function. The output first feature map will be used to capture the extensive characteristics of the input data. The first pooling layer further reduces the dimension of the feature map through max-pooling operation on the feature map while retaining the relatively key information. In the second convolutional layer, more convolutional kernels are introduced to deepen the depth of the network to extract the complex features in the feature map. This layer also realizes non-linear mapping through the ReLU activation function and then passes through the second pooling layer to further compress the feature size and enhance the feature expression ability. The third convolutional layer further enhances the complexity of the zero-current string analysis model, applies more convolutional kernels and the same activation processing to ensure the extraction of higher-level and more abstract features. After completing the multi-level convolution and pooling operations, the feature map of the last layer is flattened into a one-dimensional vector through the Flatten layer, enabling the subsequent fully connected layer to process the feature data from different spatial structures. In the fully connected layer, these flattened feature vectors will be processed by weighted sum and bias to output the activation result, providing a construction basis for the prediction of the zero-current phenomenon. To prevent the zero-current string analysis model from overfitting during training, the zero-current string analysis model specifically introduces a Dropout layer to randomly inactivate the activation result of the fully connected layer, thereby improving the generalization ability of the model. The output layer of the zero-current string analysis model sets a unit neuron, and the final prediction result is restricted between 0 and 1 through the Sigmoid activation function to output the zero-current prediction probability for judging whether there is a zero-current situation in each inverter.

[0121] Specifically, the verification of the zero-current analysis result based on the preset zero-current judgment logic is as follows:

[0122] Obtain the zero-current prediction result output by the output layer, including: the zero-current prediction probability of each input sample, where the sample is defined as: the operation data of the string inverter at a specific time point;

[0123] Define the first current membership function to evaluate the relationship between the sample current value and the set zero-current value and output its membership value on the first current feature;

[0124] Define the second current membership function to evaluate the relationship of the sample current value within the normal operating range and output its membership value on the second current feature;

[0125] Define the third current membership function to evaluate whether the sample current value exceeds the set current value and output its membership value on the third current feature;

[0126] Calculate the comprehensive judgment score by combining the zero - current prediction probability and the membership value;

[0127] When the comprehensive judgment score is greater than or equal to the set score threshold, it is determined that there is a zero - current phenomenon;

[0128] When the comprehensive judgment score is lower than the set score threshold, it is determined that there is no zero - current phenomenon.

[0129] As Figure 2 shown, calculating the comprehensive judgment score by combining the zero - current prediction probability and the membership value is specifically as follows:

[0130] Calculate the first current membership of each sample;

[0131] Calculate the second current membership of each sample;

[0132] Calculate the third current membership of each sample;

[0133] Fuse the first current membership, the second current membership and the third current membership according to the set weight coefficient to obtain the comprehensive membership;

[0134] Combine the zero - current prediction probability and the comprehensive membership to obtain the comprehensive judgment score.

[0135] In the above implementation, this embodiment obtains the zero - current prediction result output by the output layer, and this result specifically includes the zero - current prediction probability of each input sample. The sample is defined as the operating data of the string inverter at a specific time point, aiming to evaluate whether the inverter will have a zero - current phenomenon under specific conditions. The prediction result is calculated by the output layer of the deep - learning model and is usually presented in the form of probability, indicating the possibility of the sample having a zero - current phenomenon at this time point. This embodiment defines the first current membership function for the relationship between the sample current value and the set zero - current value. This function will be used to evaluate whether the sample current value is close to the set zero - current value and output its membership value on the first current feature: where I represents the sample current value, is the set zero - current value. The first current membership reflects the closeness of the sample current value to the zero - current. Define the second current membership function, aiming to evaluate whether the sample current value is within the normal operating range and output its membership value on the second current feature: Among them, I min and I max respectively represent the lower and upper limits of the normal operating range of the current. The second current membership degree helps to judge the normality of the sample current value. Define the third current membership degree function to evaluate whether the sample current value exceeds the set current threshold and output its membership degree value on the third current feature: Among them, I threshold is the set current threshold, indicating that the current above this value is likely to cause abnormal phenomena. Calculate the comprehensive judgment score by weighting the zero current prediction probability and each membership degree value. When the comprehensive judgment score is greater than or equal to the set score threshold, it is determined that there is a zero current phenomenon in this string inverter; on the contrary, if the comprehensive judgment score is lower than the set score threshold, it is determined that there is no zero current phenomenon.

[0136] As Figure 3 shown, a zero current string monitoring system for a string inverter includes:

[0137] A data acquisition unit that obtains the operating data of each string inverter;

[0138] A feature extraction unit that extracts features from the operating data of each string inverter to obtain multiple feature data, including: current feature, voltage and power feature, environmental feature, and dynamic feature;

[0139] A model calculation unit that combines the string inverter feature data and inputs the combined feature data into a zero current string analysis model constructed based on a machine learning algorithm to solve the zero current analysis result;

[0140] A verification unit that verifies the zero current analysis result based on a preset zero current judgment logic and outputs the final zero current monitoring result.

[0141] The above is only a preference of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A zero-current string monitoring method for a string inverter, characterized in that, The steps of the method include: Obtain the operation data of each string inverter; Extract features from the operation data of each string inverter to obtain multiple feature data, including: current features, voltage and power features, environmental features, and dynamic features; Combine the string inverter feature data, and input the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm to solve the zero-current analysis result; Verify the zero-current analysis result based on a preset zero-current judgment logic, and output the final zero-current monitoring result.

2. The zero-current string monitoring method for the string inverter according to claim 1, wherein The obtaining of the operation data of each string inverter further includes a data processing process, specifically: Fill in the continuous missing values in the operation data of each string inverter by linear interpolation. If the missing ratio of the data segment in the operation data of each string inverter is greater than the ratio threshold, delete this data column; Identify outliers in the operation data of each string inverter based on the Z-score module, and replace them with the adjacent average values of the outliers based on the outlier identification result; Confirm the stationarity of the time series in the operation data of each string inverter through the ADF test module. If the stationarity of the time series does not reach the expected value, perform differencing processing on the operation data of each string inverter to complete the data processing of the operation data of each string inverter.

3. The zero-current string monitoring method for the string inverter according to claim 1, characterized in that For the current features, the feature extraction steps are: Divide the current data in the operation data of each string inverter according to a time window; Calculate the current mean, maximum value, minimum value, and standard deviation of each time window; Solve the current change rate to represent the current features of the operation data of each string inverter.

4. The zero-current string monitoring method for the string inverter according to claim 3, characterized in that For the voltage and power features, the feature extraction steps are: Divide the voltage and power data in the operation data of each string inverter according to a time window; Calculate the voltage mean, maximum value, minimum value, and standard deviation of each time window; Solve the voltage change rate; Solve the power mean, maximum value, minimum value, standard deviation, and change rate based on the calculation result of the voltage change rate. The voltage change rate and the power change rate represent the voltage and power features of the operation data of each string inverter.

5. The zero-current string monitoring method for the string inverter according to claim 4, wherein For the environmental features, the feature extraction steps are: Divide the environmental data in the operation data of each string inverter according to a time window; Calculate the temperature mean, maximum value, minimum value, and standard deviation of each time window; Calculate the humidity mean, maximum value, minimum value, and standard deviation of each time window; Calculate the radiation intensity mean, maximum value, minimum value, and standard deviation of each time window; Solve the temperature change rate, humidity change rate, and radiation intensity change rate to represent the environmental features of the operation data of each string inverter.

6. The zero-current string monitoring method for the string inverter according to claim 5, characterized in that, For the dynamic features, the feature extraction steps are: Record the mutation events of the current data in the operation data of each string inverter in each time window, including the mutation amplitude and the number of times; Solve the average rate of current change based on the mutation amplitude and the number of times of the current data to represent the dynamic features of the operation data of each string inverter.

7. The zero-current string monitoring method for the string inverter according to any one of claims 3-6, characterized in that, The zero-current string analysis model constructed based on a machine learning algorithm is specifically: Combine current characteristics, voltage and power characteristics, environmental characteristics, and dynamic characteristics to form a multi-dimensional combined feature set, and perform format conversion on the multi-dimensional combined feature set; Build a zero-current string analysis model based on the VGG algorithm; The zero-current string analysis model is set with: An input layer that inputs the multi-dimensional combined feature set after format conversion; A first convolutional layer that applies 64 3x3 convolutional kernels to the multi-dimensional combined feature set, performs non-linear mapping through the ReLU activation function, and outputs a first feature map; A first pooling layer that performs 2x2 max pooling operation on the first feature map; A second convolutional layer that applies 128 3x3 convolutional kernels to the first feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs a second feature map; A second pooling layer that performs 2x2 max pooling operation on the second feature map; A third convolutional layer that applies 256 3x3 convolutional kernels to the second feature map after the max pooling operation, performs non-linear mapping through the ReLU activation function, and outputs a third feature map; A third pooling layer that performs 2x2 max pooling operation on the third feature map; A Flatten layer that flattens the third feature map after the max pooling operation into a one-dimensional vector and outputs the flattened feature vector; A fully connected layer that processes the flattened feature vector and outputs the activation result of the fully connected layer; A Dropout layer that applies Dropout to the activation result of the fully connected layer; An output layer that sets a unit neuron, restricts the output result between 0 and 1 through the Sigmoid activation function, and outputs the zero-current prediction probability.

8. The zero-current string monitoring method for the string inverter according to claim 7, characterized in that, Verify the zero-current analysis result based on the preset zero-current judgment logic, specifically: Obtain the zero-current prediction result output by the output layer, including: the zero-current prediction probability of each input sample, where the sample is defined as the operating data of the string inverter at a specific time point; Define a first current membership function to evaluate the relationship between the sample current value and the set zero-current value, and output its membership value on the first current feature; Define a second current membership function to evaluate the relationship of the sample current value within the normal operating range, and output its membership value on the second current feature; Define a third current membership function to evaluate whether the sample current value exceeds the set current value, and output its membership value on the third current feature; Calculate the comprehensive judgment score by combining the zero-current prediction probability and the membership value; When the comprehensive judgment score is greater than or equal to the set score threshold, it is determined that there is a zero-current phenomenon; When the comprehensive judgment score is lower than the set score threshold, it is determined that there is no zero-current phenomenon.

9. The zero-current string monitoring method for the string inverter according to claim 8, wherein The calculation of the comprehensive judgment score by combining the zero-current prediction probability and the membership value is specifically: Calculate the first current membership of each sample; Calculate the second current membership of each sample; Calculate the third current membership of each sample; Fuse the first current membership, the second current membership, and the third current membership according to the set weight coefficients to obtain the comprehensive membership; Combine the zero-current prediction probability and the comprehensive membership to obtain the comprehensive judgment score.

10. A zero-current string monitoring system for a string inverter, characterized in that, Including: A data acquisition unit that obtains the operating data of each string inverter; A feature extraction unit extracts features from the operation data of each string inverter to obtain multiple feature data, including: current features, voltage and power features, environmental features, and dynamic features; A model calculation unit combines the feature data of the string inverter and inputs the combined feature data into a zero-current string analysis model constructed based on a machine learning algorithm to solve the zero-current analysis result; A verification unit verifies the zero-current analysis result based on a preset zero-current judgment logic and outputs the final zero-current monitoring result.

Images