A method for reducing the false alarm rate of photovoltaic series DC arc fault

Abstract

The present invention discloses a method for reducing the false alarm rate of photovoltaic series DC arcing faults, which relates to the field of artificial intelligence algorithms and photovoltaic fault diagnosis technology, including: S100, signal acquisition; S200, current signal processing; S300, cross-correlation calculation; S400, current characteristic vector calculation; S500, series DC arcing fault analysis. The present invention provides a method for reducing the false alarm rate of photovoltaic series DC arcing faults, which further prunes and subdivides the traditional Cart decision tree in combination with artificial intelligence technology, effectively reduces the false alarm rate of DC arc fault detection, reduces the maintenance cost of photovoltaic equipment, and reduces the operation and maintenance cost of photovoltaic equipment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence algorithms and photovoltaic fault diagnosis technology, and in particular to a method for reducing the false alarm rate of photovoltaic series direct current arcing faults. Background Art

[0002] In recent years, with the advancement of science and technology and the widespread application of new energy, photovoltaic power generation technology has developed rapidly and has become an important part of the global energy structure transformation. However, the problem of series DC arcing faults has followed, which seriously threatens the safety and reliability of photovoltaic systems. Series DC arcing is a common electrical fault phenomenon in photovoltaic systems. When there are reasons such as loose connectors, aging lines, or poor contact of plug wires, dielectric breakdown will occur, and series DC arcing will occur in the circuit. Series DC arcing will not only increase power loss, but may also cause safety hazards such as fire, posing a great threat to the safety of equipment and personnel. Series DC arcing does not cause huge fluctuations and obvious increases in line current, so it is more difficult to detect than parallel DC arcing, and it brings greater safety risks to photovoltaic systems. Therefore, the accurate detection of series DC arcing faults has become one of the core technologies to ensure the safe operation of photovoltaic systems.

[0003] Most of the existing series DC arc detection technologies rely on signal feature analysis, usually analyzing the collected line current signal to identify whether there is an arc. These methods can detect arc faults to a certain extent, but because the characteristics of the series DC arc signal are similar to those of other interference signals in the photovoltaic system, the false alarm rate is often high, affecting the accuracy and efficiency of the detection. Especially in complex working environments, the detection equipment is easily affected by external electromagnetic interference, environmental noise and other factors, which further increases the possibility of false alarms. The difficulty of series DC arc detection technology also lies in how to distinguish between electrical fluctuations in normal operation and series DC arc signals. Although the existing technology has made breakthroughs in high-precision detection, the control of false alarm rate is still a problem that the industry needs to solve urgently. The number of series plugs in photovoltaic power stations is very large, and photovoltaic power stations are usually built in remote areas. The misjudgment of the series DC arc fault detection unit will greatly increase the maintenance cost.

[0004] The patent "A series fault arc detection method based on VMD Hilbert marginal spectrum multi-feature fusion" (publication number CN116577612A) proposes a fault arc detection method based on variational mode decomposition (VMD) and Hilbert marginal spectrum analysis (HHT) multi-feature fusion. The method first extracts multiple intrinsic mode functions of different frequencies by performing variational mode decomposition on the collected current signal, and then performs marginal spectrum analysis on each modal component. For the selection of modal component marginal spectrum, a modal component selection method based on multi-feature fault discrimination is proposed to select feature quantities with high fault discrimination of each modal component marginal spectrum. In order to reduce feature redundancy, linear component analysis is used for feature fusion, and finally a neural network is introduced for classification and verification of detection accuracy. However, the current data extracted from the photovoltaic system in the patent is a huge collection, which increases the workload of the machine and cannot distinguish the valuable feature quantities well, which will lead to a high false alarm rate of series DC arc faults.

[0005] Therefore, technicians in this field are committed to developing a method to reduce the false alarm rate of photovoltaic series DC arcing faults. Summary of the invention

[0006] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is how to reduce the misjudgment of series DC arcing fault detection.

[0007] This application analyzes that although the existing technology has made breakthroughs in high-precision detection, the control of false alarm rate is still a problem that the industry needs to solve urgently. The number of series plugs in photovoltaic power stations is very large, and photovoltaic power stations are usually built in remote areas. Misjudgment of series DC arc fault detection will greatly increase maintenance costs.

[0008] This application analyzes the classic Cart decision tree algorithm, which uses the Gini coefficient as an indicator function to evaluate the quality of sample classification. Assume that for a binary classification problem, there are two types of samples, and let the first type of training sample set be S 0, the second type of training sample set is S 1. Training set for training binary decision tree S = S 0∪ S 1. Assumptions S There are a total of R training samples, S =[ x 1, x 2,…, x R ] T , each training sample hasD Features, x r ( d ) (1 ≤ d ≤ D , 1 ≤ r ≤ R ) indicates the r The first training sample d The value of the feature. Then, the set with class labels S The Gini coefficient can be calculated according to the following formula:

[0009]

[0010] in, Indicates the number of elements in a collection. p 0=| S 0| / | S | and p 1=| S 1| / | S | respectively represent the probability that the sample belongs to the first category and the second category, p 0+ p 1=1. The Gini coefficient essentially represents the S The probability that two samples randomly selected from the set do not belong to the same class. S The smaller the Gini coefficient, the smaller the set S The higher the purity.

[0011] The traditional two-classification decision tree algorithm requires that the two types of samples are relatively balanced and have the same probability of misclassification for the two categories. The Gini coefficient shows that the Cart decision tree algorithm evaluates the quality of fault and normal data classification equally. Due to the large number of photovoltaic equipment in large-scale photovoltaic power stations, the number of series plugs in photovoltaic power stations is very large, and photovoltaic power stations are usually built in remote areas. The false reporting of series DC arc faults will greatly increase the maintenance cost, and the power generation equipment hopes to reduce the false reporting failure rate as much as possible. Therefore, this application introduces the biased Gini coefficient in statistics, and preferentially reduces the biased Gini coefficient in the direction where the biased Gini coefficient decreases rapidly, that is, the biased Gini coefficient focuses more on segmenting faults. Based on a highly nonlinear model, AI further "prunes" and subdivides the traditional classification and regression tree (CART), calculates and iterates massive data, finds high-dimensional spatial feature laws, and effectively distinguishes feature signals with similar shapes, which can effectively reduce the false alarm rate in the detection process, thereby reducing the cost of photovoltaic equipment operation and maintenance.

[0012] Classification and regression tree (CART) is a type of decision tree algorithm. Like other decision tree algorithms, it consists of feature selection, tree generation and pruning. CART is widely used and used in tree ensemble models. For example, the base learners of ensemble algorithms such as GBDT and RF are all CART trees. When the dependent variable of the data set is a continuous value, the tree algorithm is a regression tree, and the mean of the leaf node observations can be used as the predicted value; when the dependent variable of the data set is a discrete value, the tree algorithm is a classification tree, which can solve the classification problem well.

[0013] In one embodiment of the present invention, a method for reducing the false alarm rate of photovoltaic series DC arc faults is provided, comprising:

[0014] S100, signal acquisition, collecting current signals under normal working conditions and DC arc fault conditions;

[0015] S200, current signal processing, windowing and smoothing the current signal, performing linear integral transformation, and obtaining a current sampling signal distribution matrix;

[0016] S300, cross-correlation calculation, using the Xcorr function to perform cross-correlation calculation on each column of the current sampling signal distribution matrix to obtain a cross-correlation matrix;

[0017] S400, current characteristic vector calculation, reducing the dimension and merging the current sampling signal distribution matrix and the cross-correlation matrix to obtain the current characteristic vector;

[0018] S500, series DC arcing fault analysis, input the current feature vector into the trained random forest model for analysis, if it is determined to be a DC arcing fault, it will be reported.

[0019] Optionally, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the linear integral transformation method includes short-time Fourier transform and fast Fourier transform.

[0020] Further, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, step S200 includes:

[0021] S210, select a window function, select a rectangular window function, the formula is as follows:

[0022]

[0023] in F is the length of the window function, l represents discrete time points;

[0024] S220, signal segmentation, the time length is L The current signal is segmented at fixed time intervals through the window function. Each segment is a signal window, and the length of the segment is equal to the length of the window function. F , split into [ L / F ] signal windows, where [ ] is the rounding operation;

[0025] S230, short-time Fourier transform, performing short-time Fourier transform on each signal window to obtain a current sampling signal distribution matrix, the formula is as follows;

[0026]

[0027] in, f represents the sampling frequency, t Represents the time axis, t is the window length of the short-time Fourier transform, x ( t ) is the current signal to be analyzed; w(t) is the Hanning window function, w(τ-t) The Hanning window function representing the lag time t is used to weight the current signal, optimize the accuracy of spectrum analysis, reduce spectrum leakage and improve frequency resolution. j Is an imaginary unit.

[0028] Furthermore, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the fixed time interval ranges from 1 to 160 ms.

[0029] Preferably, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the fixed time interval is 4 ms.

[0030] Optionally, in the method for reducing the false alarm rate of photovoltaic series DC arc fault in any of the above embodiments, step S300 includes:

[0031] S310, select cross-correlation calculation data, select two columns of cross-correlation calculation data from the current sampling signal distribution matrix, p Column Data and q Column Data ;

[0032] S320, normalization processing, p Column Data and q Column Data Normalization is performed, and the normalization interval is [0, 1]. p Column Data For example, use the following formula:

[0033]

[0034] in It is p Column Data The k Data ， [1 ,n ], and Respectively represent the p The maximum and minimum values ​​of the column data, Express The normalized value;

[0035] S330, cross-correlation calculation, using the cross-correlation function to calculate the correlation data of the two selected columns and Perform cross-correlation calculations to obtain the cross-correlation matrix and cross-correlation function Defined as:

[0036]

[0037] in, Indicates the number of delays per unit time.

[0038] Optionally, in the method for reducing the false alarm rate of photovoltaic series DC arc fault in any of the above embodiments, step S400 includes:

[0039] S410, matrix dimensionality reduction processing, performing dimensionality reduction processing on the current sampling signal distribution matrix and the cross-correlation matrix to generate two one-dimensional matrices;

[0040] S420, matrix merging, merging two one-dimensional matrices to obtain a current eigenvector.

[0041] Optionally, in the method for reducing the false alarm rate of photovoltaic series DC arcing faults in any of the above embodiments, the random forest model uses a random forest model with a biased Gini coefficient, including multiple biased decision trees.

[0042] Furthermore, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the number of biased decision trees ranges from 1 to 100.

[0043] Preferably, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the number of biased decision trees is 15.

[0044] Furthermore, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the training process of the random forest model includes:

[0045] S510, training a random forest model, where the random forest model includes multiple biased decision trees, each biased decision tree is trained separately, the input is the current feature vector, and the output is 0 (indicating normal) or 1 (indicating a DC arcing fault);

[0046] S520: further classify the mixed leaf nodes, further classify the mixed leaf nodes, and put all the mixed leaf nodes into a set middle, , where the number of these mixed leaf nodes is , express The i Mixed leaf nodes;

[0047] S530: Determine the pruning order. To determine the pruning order of these mixed leaf nodes, for the set Each mixed leaf node in uses the pruning coefficient Prune mixed leaf nodes:

[0048]

[0049] Among them, S0 is the fault sample set, S1 is the normal sample set; use the Ei value to sort each leaf node from high to low;

[0050] S540, based on false alarm coefficient Pruning, where the pruning coefficient Ei is greater than The mixed leaf nodes are pruned off, [ ] is the rounding function;

[0051] S550, pruning is completed, and the mixed leaf nodes that have not been pruned continue to be split and classified, that is, S520-S540 are executed in a loop until the pruning of all mixed leaf nodes is completed;

[0052] S560, random forest model training is completed. When each biased decision tree is trained according to S520-S550, a random forest model can be obtained.

[0053] Further, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, step S510 includes:

[0054] S511, set the training sample set, for each biased decision tree, set is labeled with a class label(x) A set of training samples, class labels -1 and +1 represent two different classes.x express D dimensional current feature vector; assuming that the biased decision tree training is completed, the final set of all leaf nodes is the set of all training samples, expressed as follows:

[0055]

[0056] in, K Represents the number of leaf nodes in a biased decision tree. is the set of all leaf nodes, Indicates leaf nodes;

[0057] S512, introduce the biased Gini coefficient: as follows pre_Gini ( S ):

[0058]

[0059] in α is the preference coefficient, α >0,yes before _ Gini ( S ) to obtain the peak value of the biased Gini coefficient. p 0 value is p 0 ∗ =1 / ( α +1);

[0060] S513, training a biased decision tree. During the training process, in order to minimize the value of the biased Gini coefficient, when splitting a node, it is preferred to reduce the number of fault samples in the node, and the biased Gini coefficient is preferably reduced in the direction where the biased Gini coefficient decreases rapidly, that is, the biased Gini coefficient is more focused on segmenting faults; when α >1, and p 0∈[0, 1 / ( α +1)], α The larger is, the faster the biased Gini coefficient decreases, so p 0 reduces the loss function pre_Gini ( S )The benefit of reducing is higher, and the cost of the same misclassification is also higher;

[0061] S514, the biased decision tree training ends. When the criterion for stopping the biased decision tree training is met, the training ends.

[0062] Furthermore, in the method for reducing the false alarm rate of photovoltaic series DC arcing faults in the above embodiment, the criterion for stopping training of the biased decision tree is: when the internal node is a pure leaf node, that is, it only contains faulty samples or only contains normal samples, or the number of samples in the leaf node has reached the set minimum number.

[0063] Preferably, in the method for reducing the false alarm rate of photovoltaic series DC arc faults in the above embodiment, the minimum number of samples in the leaf node is set to 5.

[0064] Preferably, in the method for reducing the false alarm rate of photovoltaic series DC arc fault in the above embodiment, the false alarm coefficient Select 0.0001.

[0065] The present invention provides a method for reducing the false alarm rate of photovoltaic series DC arc faults based on biased Gini coefficient, and further prunes and subdivides the traditional Cart decision tree in combination with artificial intelligence technology, thereby effectively reducing the false alarm rate of DC arc fault detection, reducing the maintenance cost of photovoltaic equipment, and reducing the operation and maintenance cost of photovoltaic equipment.

[0066] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flow chart illustrating a method for reducing a false alarm rate of photovoltaic series DC arc faults according to an exemplary embodiment. DETAILED DESCRIPTION

[0068] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0069] In the drawings, components with the same structure are indicated by the same numerical reference numerals, and components with similar structures or functions are indicated by similar numerical reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the illustration clearer, the thickness of the components is schematically exaggerated in some places in the drawings.

[0070] This application analyzes the classic Cart decision tree algorithm, which uses the Gini coefficient as an indicator function to evaluate the quality of sample classification. Assume that for a binary classification problem, there are two types of samples, and let the first type of training sample set be S 0, the second type of training sample set is S 1. Training set for training binary decision treeS = S 0∪ S 1. Assumptions S There are a total of R training samples, S =[ x 1, x 2,…, x R ] T , each training sample has D Features, x r ( d ) (1 ≤ d ≤ D , 1 ≤ r ≤ R ) indicates the r The first training sample d The value of the feature. Then, the set with class labels S The Gini coefficient can be calculated according to the following formula:

[0071]

[0072] in, Indicates the number of elements in a collection. p 0=| S 0| / | S | and p 1=| S 1| / | S | respectively represent the probability that the sample belongs to the first category and the second category, p 0+ p 1=1. The Gini coefficient essentially expresses the S The probability that two samples randomly selected from the set do not belong to the same class. S The smaller the Gini coefficient, the smaller the set S The higher the purity.

[0073] The traditional two-classification decision tree algorithm requires that the two types of samples are relatively balanced and have the same probability of misclassification for the two categories. The Gini coefficient shows that the Cart decision tree algorithm evaluates the quality of fault and normal data classification equally. Due to the large number of photovoltaic equipment in large-scale photovoltaic power stations, the number of series plugs in photovoltaic power stations is very large, and photovoltaic power stations are usually built in remote areas. The false reporting of series DC arc faults will greatly increase the maintenance cost, and the power generation equipment hopes to reduce the false reporting failure rate as much as possible. Therefore, this application introduces the biased Gini coefficient in statistics, and preferentially reduces the biased Gini coefficient in the direction where the biased Gini coefficient decreases rapidly, that is, the biased Gini coefficient focuses more on segmenting faults. Based on a highly nonlinear model, AI further "prunes" and subdivides the traditional classification and regression tree (CART), calculates and iterates massive data, finds high-dimensional spatial feature laws, and effectively distinguishes feature signals with similar shapes, which can effectively reduce the false alarm rate in the detection process, thereby reducing the cost of photovoltaic equipment operation and maintenance.

[0074] Classification and regression tree (CART) is a type of decision tree algorithm. Like other decision tree algorithms, it consists of feature selection, tree generation and pruning. CART is widely used and used in tree ensemble models. For example, the base learners of ensemble algorithms such as GBDT and RF are CART trees. When the dependent variable of the data set is a continuous value, the tree algorithm is a regression tree, and the mean of the leaf node observations can be used as the predicted value; when the dependent variable of the data set is a discrete value, the tree algorithm is a classification tree, which can solve classification problems.

[0075] This application designs a method to reduce the false alarm rate of photovoltaic series DC arc faults, such as Figure 1 As shown, including:

[0076] S100, signal acquisition, collecting current signals under normal working conditions and DC arc fault conditions.

[0077] S200, current signal processing, windowing and smoothing the current signal, performing linear integral transformation, and obtaining a current sampling signal distribution matrix, wherein the linear integral transformation method adopts short-time Fourier transform; specifically, the following steps are included:

[0078] S210, select a window function, select a rectangular window function, the formula is as follows:

[0079]

[0080] in F is the length of the window function,l represents discrete time points;

[0081] S220, signal segmentation, the time length is L The current signal is segmented by the window function at a fixed time interval. The fixed time interval is 4ms. Each segment is a signal window, and the length of the segment is equal to the length of the window function. F , split into [ L / F ] signal windows, where [ ] is the rounding operation;

[0082] S230, short-time Fourier transform, performing short-time Fourier transform on each signal window to obtain a current sampling signal distribution matrix, the formula is as follows;

[0083]

[0084] in, f represents the sampling frequency, t Represents the time axis, t is the window length of the short-time Fourier transform, x ( t ) is the current signal to be analyzed; w(t) is the Hanning window function, w(τ-t) The Hanning window function representing the lag time t is used to weight the current signal, optimize the accuracy of spectrum analysis, reduce spectrum leakage and improve frequency resolution. j Is an imaginary unit.

[0085] S300, cross-correlation calculation, using the Xcorr function to perform cross-correlation calculation on each column of the current sampling signal distribution matrix to obtain a cross-correlation matrix; specifically including:

[0086] S310, select cross-correlation calculation data, select two columns of cross-correlation calculation data from the current sampling signal distribution matrix, p Column Data and q Column Data ;

[0087] S320, normalization processing, p Column Data and q Column Data Normalization is performed, and the normalization interval is [0, 1]. p Column Data For example, use the following formula:

[0088]

[0089] in It is p Column Data The k Data ， [1 ,n ], and Respectively represent the p The maximum and minimum values ​​of the column data, Express The normalized value;

[0090] S330, cross-correlation calculation, using the cross-correlation function to calculate the correlation data of the two selected columns and Perform cross-correlation calculations to obtain the cross-correlation matrix and cross-correlation function Defined as:

[0091]

[0092] in, Indicates the number of delays per unit time.

[0093] S400, current characteristic vector calculation, reducing the dimension and merging the current sampling signal distribution matrix and the cross-correlation matrix to obtain the current characteristic vector; specifically including:

[0094] S410, matrix dimensionality reduction processing, performing dimensionality reduction processing on the current sampling signal distribution matrix and the cross-correlation matrix to generate two one-dimensional matrices;

[0095] S420, matrix merging, merging two one-dimensional matrices to obtain a current eigenvector.

[0096] S500, series DC arc fault analysis, random forest model uses a random forest model with biased Gini coefficient. The random forest model includes multiple biased decision trees. The number of biased decision trees is selected as 15. The current feature vector is input into the trained random forest model for analysis. If it is judged to be a DC arc fault, it is reported; the training process of the random forest model includes:

[0097] S510, training a random forest model, where each biased decision tree is trained separately, the input is a current feature vector, and the output is 0 (indicating normal) or 1 (indicating a DC arc fault); including:

[0098] S511, set the training sample set, for each biased decision tree, set is labeled with a class label(x) A set of training samples, class labels -1 and +1 represent two different classes. x expressD dimensional current feature vector; assuming that the biased decision tree training is completed, the final set of all leaf nodes is the set of all training samples, expressed as follows:

[0099]

[0100] in, K Represents the number of leaf nodes in a biased decision tree. is the set of all leaf nodes, Indicates leaf nodes;

[0101] S512, introduce the biased Gini coefficient: as follows pre_Gini ( S ):

[0102] ;

[0103] in α is the preference coefficient, α >0,yes before _ Gini ( S ) to obtain the peak value of the biased Gini coefficient. p 0 value is p 0 ∗ =1 / ( α +1);

[0104] S513, training a biased decision tree. During the training process, in order to minimize the value of the biased Gini coefficient, when splitting a node, it is preferred to reduce the number of fault samples in the node, and the biased Gini coefficient is preferably reduced in the direction where the biased Gini coefficient decreases rapidly, that is, the biased Gini coefficient is more focused on segmenting faults; when α >1, and p 0∈[0, 1 / ( α +1)], α The larger is, the faster the biased Gini coefficient decreases, so p 0 reduces the loss function pre_Gini ( S )The benefit of reducing is higher, and the cost of the same misclassification is also higher;

[0105] S514, the biased decision tree training ends. When the criteria for stopping training of the biased decision tree are met, the training ends. The criteria for stopping training of the biased decision tree are: when the internal node is a pure leaf node, that is, it only contains faulty samples or only contains normal samples, or the number of samples in the leaf node has reached the set minimum number, and the set minimum number is set to 5.

[0106] S520: further classify the mixed leaf nodes, further classify the mixed leaf nodes, and put all the mixed leaf nodes into a set middle, , where the number of these mixed leaf nodes is , express The i Mixed leaf nodes;

[0107] S530: Determine the pruning order. To determine the pruning order of these mixed leaf nodes, for the set Each mixed leaf node in uses the pruning coefficient Prune mixed leaf nodes:

[0108]

[0109] Among them, S0 is the fault sample set, S1 is the normal sample set; use the Ei value to sort each leaf node from high to low;

[0110] S540, based on the false alarm coefficient Pruning, where the pruning coefficient Ei is greater than The mixed leaf nodes are pruned off, [ ] is the rounding function, and the false alarm coefficient Select 0.0001;

[0111] S550, pruning is completed, and the mixed leaf nodes that have not been pruned continue to be split and classified, that is, S520-S540 are executed in a loop until the pruning of all mixed leaf nodes is completed;

[0112] S560, random forest model training is completed. When each biased decision tree is trained according to S520-S550, a random forest model can be obtained.

[0113] The present application uses the traditional Gini coefficient and the random forest method with biased Gini coefficient disclosed in the present application to conduct experiments, and the results show that the false alarm rate is reduced from 0.27% to 0.01%.

[0114] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for reducing the false alarm rate of photovoltaic series DC arc faults, characterized in that: include: S100, signal acquisition, collecting current signals under normal working conditions and DC arc fault conditions; S200, current signal processing, windowing and smoothing the current signal, performing linear integral transformation, and obtaining a current sampling signal distribution matrix; S300, cross-correlation calculation, using the Xcorr function to perform cross-correlation calculation on each column of the current sampling signal distribution matrix to obtain a cross-correlation matrix; S400, current characteristic vector calculation, reducing the dimension and merging the current sampling signal distribution matrix and the cross-correlation matrix to obtain a current characteristic vector; S500, analyzing a series DC arcing fault, inputting the current feature vector into a trained random forest model for analysis, and reporting if it is determined to be a DC arcing fault; The training process of the random forest model includes: S510, training a biased decision tree, wherein the random forest model includes multiple biased decision trees, each biased decision tree is trained separately, the input is the current feature vector, and the output is 0 or 1; specifically including: S511, set the training sample set, for each biased decision tree, set is labeled with a class label(x) A set of training samples, class labels -1 and +1 represent two different classes. x express D dimensional current feature vector; assuming that the biased decision tree training is completed, the final set of all leaf nodes is the set of all training samples, expressed as follows: ； S512, introduce the biased Gini coefficient: as follows pre_Gini ( S ): ； S513, training a biased decision tree. During the training process, in order to minimize the value of the biased Gini coefficient, when splitting a node, it is preferred to reduce the number of fault samples in the node, and the biased Gini coefficient is preferably reduced in the direction where the biased Gini coefficient decreases rapidly, that is, the biased Gini coefficient is more focused on segmenting faults; when α >1, and p 0∈[0, 1 / ( α +1)], α The larger is, the faster the biased Gini coefficient decreases, so p 0 reduces the loss function pre_Gini ( S )The benefit of reducing is higher, and the cost of the same misclassification is also higher; S514, the biased decision tree training ends. When the criterion for stopping the biased decision tree training is met, the training ends. S520: further classify the mixed leaf nodes, further classify the mixed leaf nodes, and put all the mixed leaf nodes into a set middle, ; S530: Determine the pruning order. To determine the pruning order of the hybrid leaf nodes, for the set Each mixed leaf node in uses the pruning coefficient Prune the hybrid leaf nodes: ； S540, based on false alarm coefficient Pruning, where the pruning coefficient Ei is greater than The mixed leaf nodes are pruned off; S550, pruning is completed, and the mixed leaf nodes that have not been pruned continue to be split and classified, that is, S520-S540 are executed in a loop until the pruning of all mixed leaf nodes is completed; S560, the random forest model training is completed. When each biased decision tree is trained according to the steps S520 to S550, the random forest model can be obtained.

2. The method for reducing the false alarm rate of photovoltaic series DC arc faults according to claim 1, characterized in that: The linear integral transformation method includes short-time Fourier transform and fast Fourier transform.

3. The method for reducing the false alarm rate of photovoltaic series DC arc faults according to claim 1, characterized in that: The step S200 includes: S210, select a window function, select a rectangular window function, the formula is as follows: ； S220, signal segmentation, the time length is L The current signal is segmented at fixed time intervals through the window function, each segment is a signal window, and the length of the segment is equal to the length of the window function F , split into [ L / F ] signal windows; S230, short-time Fourier transform, performing short-time Fourier transform on each signal window to obtain the current sampling signal distribution matrix, the formula is as follows: 。 4. The method for reducing the false alarm rate of photovoltaic series DC arc faults as claimed in claim 3, characterized in that: The fixed time interval ranges from 1 to 160 ms.

5. The method for reducing the false alarm rate of photovoltaic series DC arc faults as claimed in claim 3, characterized in that: The step S300 includes: S310, selecting cross-correlation calculation data, selecting two columns of cross-correlation calculation data from the current sampling signal distribution matrix, p Column Data and q Column Data ; S320, normalization processing, p Column Data and the q Column Data Normalization is performed, and the normalization interval is [0, 1]. p Column Data For example, use the following formula: ； S330, cross-correlation calculation, using the cross-correlation function to calculate the correlation data of the two selected columns and Perform cross-correlation calculation to obtain the cross-correlation matrix, cross-correlation function Defined as: 。 6. The method for reducing the false alarm rate of photovoltaic series DC arc faults as claimed in claim 5, characterized in that: The step S400 includes: S410, matrix dimensionality reduction processing, performing dimensionality reduction processing on the current sampling signal distribution matrix and the cross-correlation matrix to generate two one-dimensional matrices; S420, matrix merging: merging the two one-dimensional matrices to obtain the current characteristic vector.

7. The method for reducing the false alarm rate of photovoltaic series DC arc faults as claimed in claim 6, characterized in that: The random forest model uses a random forest model with a biased Gini coefficient, including multiple biased decision trees.

8. The method for reducing the false alarm rate of photovoltaic series DC arc faults as claimed in claim 7, characterized in that: The criterion for stopping training of the biased decision tree is: when the internal node is a pure leaf node, that is, it only contains faulty samples or only contains normal samples, or the number of samples in the leaf node has reached a set minimum number.

Images